xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EHF3SITOX_RUEUZE INGEST_TIME 2014-10-03T21:32:27Z PACKAGE UFE0046133_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FINDING INTEGRATED SOLUTIONS TO WICKED ENVIRONMENTAL CHALLENGES: ECOSYSTEM BASED MANAGEMENT OF COTTONWOOD COMMUNITIES ON THE MISSOURI RIVER By KELLY A. BURKSCOPES A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNI VERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2014
2014 Kelly A. Burks Copes
3 To my amazing family (both immediate and extended) w ithout your support and faith, this would not have been possible
4 ACKNOWLEDGEMENTS A beginning is the time for taking the most delicate care that the balances are correct. Frank Herbert Dune (1965) Wicked problems have numerous intervention points, have consequences difficult to envision, and are surrounded by a dynamic uncertainty wrapped in a moving frontier of knowledge. Ioannis Petrus, 2009 I have come to believe that PhD project s are the quintessential examples of wicked pr oblem s. There are no right or wrong solutions, only compromises fraught with uncertainty For a perfectionist who face s crise s day in and day out as part of her other tasks as assigned job description, I found this experience to be both overwhelming, and yet somehow cathartic. I have literally felt myself changing as the project evolved physically, mentally, and yes, even spiritually. For those in the immediate vicinity, the ones who had to handle the brunt of my evolution and accommodating insanity I can only say two things: 1) it is finally over, and 2) I promise to behave myself from here on out. Your courage, understanding, and yes, your ability to disappear and leave me to my rantings without a disparaging word, was as humbling a s it was a we in spiring, and I hope to s omeday return the favor For those on the fringes, those who simply shook their heads smiled and offered to take the kids for a while I thank you for understanding and supporting my family who by the way had no idea I could get so crazy! Wicked problems, as this dissertation points out, are only tamed through massive stakeholder involvement and unfettered creative problem solving. I cannot begin to express the gratitude I feel toward every single team member on this project. This is a
5 labor of love of my part, but each of you had to tolerate being thrown into the synergistic vortex I realize now that the time and resources you dedicated to this effort were enormous, the exercises I put you through took you out of your comfort zones, and sometimes you had no idea where we were headed. I just want to thank each and every one of you. I hope I did you (and the cottonwoods) proud. Lastly, I had amazing mentors, and without their presence of mind and unfailing support, I would n ever have finished this massive undertaking. My bosses at the US Army Engineer Research and Development Center (Ms. Antisa Webb, Dr. Edmond Russo, and Dr. Todd Bridges) peeled me off the ceiling once or twice, made sure I had funding and time to both work and write and offered advice and critical review at every turn. I can not explain how remarkable it is to be surrounded by such amazing people with such wisdom and compassion. In addition to my coworkers, I want to thank my committee and more specifically my advisor, Dr. Greg ory Kiker. Talk about wherewithal. I think you all realized that my research proposal was almost beyond my capabilities, particularly since I had to hold down a full time job and raise two kids while attempting it. Yet, you encouraged me to tac kle this wicked problem, and your assistance every step of the way has made me a better researcher and a deeper thinker I thank you for having faith in me and teaching me how to cogitate and reflect rather than simply react. I dedicate this dissertation to my husband, David Copes, and my two children, Matthew Noah and Sarah Madison. You are the love of my life and the strength in every step I take. And with that, I close with a few final quotes that I found humorous and inspiring
6 The Missouri River was located in the United States at last report. It cuts corners, runs around at night, lunches on levees, and swallows islands and small villages for dessert. Its perpetual dissatisfaction with its bed is the greatest peculiarity of the Missouri. Time after t ime it has gotten out of its bed in the middle of the night with no apparent provocation, and has hunted a new bed, all littered with forests, cornfields, brick houses, railroad ties, and telegraph poles. Later it has suddenly taken a fancy to its old bed, which by this time has been filled with suburban architecture, and back it has gone with a whoop and a rush as if it had found something worthwhile. It makes farming as fascinating as gambling. You never know whether you are going to harvest corn or catfi sh. George Fitch, 1907 The thing the ecologically illiterate do not realize about an ecosystem is that it is a system. A system! A system maintains a certain fluid stability that can be destroyed by a misstep in just one niche. A system has order, a flow ing from point to point. If something dams the flow, order collapses. The untrained miss the collapse until too late. That i s why the highest function of ecology is the understanding of consequences. Frank Herbert Dune (1965) We now undertake a Socratic cascade of ever refining questions. Dr. John Nestler, National Conference on Ecosystem Restoration, 2011
7 TABLE OF CONTENTS Page ACKNOWLEDGEMENTS ............................................................................................... 4 LIST OF TABLES .......................................................................................................... 11 LIST OF FIGURES ........................................................................................................ 13 LIST OF ABBREVIATIONS ........................................................................................... 16 ABSTRACT ............................................................................................................... 18 CHAPTER 1 A N INTEGRATED DECISIONMAKING FRAMEWORK TO SUPPORT ECOSYSTEM BASED MANAGEMENT CHALLENGES IN WICKED SETTINGS .. 20 1.1 Introduction ....................................................................................................... 20 1.2 Problem Statement ........................................................................................... 22 1.3 Goals and Objectives ........................................................................................ 23 1.4 A Spiraling Framework for Ecosystem Planning ............................................... 24 1.5 Road Map ......................................................................................................... 26 2 UNCOVERING THE LINES OF EVIDENCE HIDDEN INSIDE WICKED PROBLEMS: USING CONCEPTUAL MODELS TO INFORM ECOSYSTEM BASED MANAGEMENT OF TH E MISSOURI RIVER COTTONWOODS .............. 31 2.1 Introduction ....................................................................................................... 31 2.2 Goals and Objectives ........................................................................................ 33 2.3 Literature Review .............................................................................................. 34 2.3.1 Ecosystem Based Management ............................................................ 34 2.3.2 Conceptual Modeling From Theory to Practice ................................... 37 2.3.3 Conceptual Modeling Mechanics ........................................................... 38 2.3.4 Spiral Modeling ...................................................................................... 41 2.4 Methods ............................................................................................................ 42 2.4.1 Study Area and Problem Space............................................................. 42 2.4.2 Transdisciplinary Teaming ..................................................................... 46 2.4.3 Spiraling Based Construction Strategy .................................................. 47 2.5 Results .............................................................................................................. 50 2.5.1 Drivers ................................................................................................... 50 2.5.2 Stressors ............................................................................................... 51 2.5.3 Valued Ecosystem Components ............................................................ 53 2.5.4 Ecological Indicators .............................................................................. 55 2.6 Discussion and Conclusions ............................................................................. 56
8 3 PRESCRIPTIVELY FORMALIZING THE USGS LAND CAPABILITY POTENTIAL INDEX (LCPI): A CASE STUDY ON THE MISSOURI RIVER ........... 68 3.1 Introduction ....................................................................................................... 68 3.2 Goals and Objectives ........................................................................................ 69 3.3 Review of Foundational Concepts .................................................................... 71 3.3.1 Prescriptive Formalization in the Presence of Data Gaps ...................... 71 3.3.2 Expert Kn owledgeIrrational and Indispensible ................................... 73 3.3.3 Elicitation Traps and Pitfalls ................................................................... 73 3.3.4 Ranking and Aggregation ...................................................................... 75 3.4 Methods ............................................................................................................ 76 3.4.1 Study Area and Problem Space............................................................. 76 3.4.2 Land Capability Potential Index (LCPI) for the Case Study Area ........... 7 7 3.4.3 Elicitation and Formalization .................................................................. 79 18.104.22.168 Preparation ............................................................................... 80 22.214.171.124 Elicitation and processing ......................................................... 82 126.96.36.199 Conversion ................................................................................ 84 3.5 Results .............................................................................................................. 85 3.5.1 Results of the Expert Elicitation Exercise .............................................. 85 188.8.131.52 Rankings of LCPI classes for older stands (> 25 yrs old) ......... 85 184.108.40.206 Rankings of LCPI classes for younger stands (225 yrs old) .... 86 220.127.116.11 Formalization of the rankings .................................................... 87 3.5.2 Results of the Overlay Analysis ............................................................. 88 3.6 Discussion and Conclusions ............................................................................. 90 4 SEEING THE FOREST FO R THE TREES: EXPLORING THE WICKED PROBLEM OF COTTONWOOD RECOVERY ON THE MISSOURI RIVER WITH SPIRAL BASED ECOSYSTEM RESPONSE MODELING ......................... 104 4.1 Introduction ..................................................................................................... 104 4.2 Goals and Objectives ...................................................................................... 106 4.3 Background and Literature Review ................................................................. 107 4.3.1 The Cottonwood Crisis ........................................................................ 107 4.3.2 Ecosystem Response Modeling ........................................................... 109 4.3.3 Spiral Modeling .................................................................................... 112 4.4 Methods .......................................................................................................... 114 4.4.1 Spiral #1: Problem Definition ............................................................... 116 18.104.22.168 Study area and model domain ................................................ 116 22.214.171.124 Study team and workshop format ........................................... 117 126.96.36.199 Modeling goals, objectives, and constraints ............................ 119 4.4.2 Spiral #2: Conceptualization ................................................................ 121 4.4.3 Spiral #3: Data Collection and Processing ........................................... 124 188.8.131.52 Cover type mapping ................................................................ 124 184.108.40.206 Field data collection ................................................................ 125 220.127.116.11 Spatial data geoprocessing ..................................................... 129 18.104.22.168 Hydrologic data collection and analysis .................................. 132 4.4.4 Spiral #4 Construction and Testing ...................................................... 133
9 22.214.171.124 Calibration ............................................................................... 134 126.96.36.199 Verification .............................................................................. 136 188.8.131.52 Validation using Independent Measures of Function (IMFs) ... 136 184.108.40.206 Holdout validation ................................................................... 138 220.127.116.11 Bald eagle validation ............................................................... 138 4.5 Results ............................................................................................................ 138 4.5.1 Final Response Indices (RIs) ............................................................... 138 4.5.2 Final Component Response Indices (CRIs) ......................................... 139 4.5.3 Final Ecosystem Response Index (ERI) .............................................. 140 4.6 Discussion ...................................................................................................... 141 4.7 Conclusions .................................................................................................... 144 5 COTTONWOOD RECOVERY INTEGRATED SITE IDENTIFICATION SYSTEM (CRISIS): A GIS BASED PATICIPATORY DECISION SUPPORT SYSTEM FOR THE MISSOURI RIVER ............................................................................... 167 5.1 Introduc tion ..................................................................................................... 167 5.2 Goals and Objectives ...................................................................................... 168 5.3 Background and Literature Review ................................................................. 170 5.3.1 Cottonwood Recovery on the Missouri River ....................................... 170 5.3.2 Sieve Mapping ..................................................................................... 172 5.3.3 Multi Criteria Decision Analysi s (MCDA) .............................................. 173 5.3.4 Spiral Framework ................................................................................. 174 5.4 Methods .......................................................................................................... 175 5.4.1 St udy Area ........................................................................................... 176 5.4.2 Study Team and Workshop Format ..................................................... 176 5.4.3 Rationale for Criteria Selection ............................................................ 177 5.4.4 Geoprocessing ..................................................................................... 180 5.4.5 Multicriteria Elicitation and Analysis ..................................................... 181 5.4.6 Sieve Mapping ..................................................................................... 182 5.5 Results ............................................................................................................ 183 5.5.1 Criteria Transformation and Standardization ....................................... 183 5.5.2 Expert Preferences .............................................................................. 183 5.5.3 Criteria Overlays and Evaluation ......................................................... 184 5.6 Discussion ...................................................................................................... 185 5.7 Conclusions .................................................................................................... 188 6 OVERCOMING THE LAW OF UNINTENDED CONSEQUENCES: GETTING SOME PERSPECTIVE AND PAUSING FOR REFLECTION ............................... 204 6.1 Concluding Remarks ....................................................................................... 204 6.2 Summary ........................................................................................................ 206 6.3 Path Forward .................................................................................................. 210 6.4 Future Research Opportunities ....................................................................... 212
10 APPENDIX A HISTORIC COVER TYPE MAPPING RESULTS .................................................. 216 B BOXAND WHISKER PLOTSRAW VARIABLE DATA ...................................... 220 C CALIBRATED RESPONSE INDEX (RI) CURVES ................................................ 222 D BOXAND WHISKER PLOTSTEST DATA SETS ............................................. 226 E VALIDATION DATA FOR RADIAL DIAGRAM ...................................................... 229 LIST OF REFERENCES ............................................................................................. 233 BIOGRAPHICAL SKETCH .......................................................................................... 247
11 LIST OF TABLES Table Page 2 1 A typology aggregating grouping conceptual models ......................................... 61 3 1 Numbers and affiliations of the transdiciplinary team members .......................... 96 3 2 Spearman rank correlations of site ranks among experts ( n = 10) for the aggregated older stands (> 25 yrs old). ............................................................ 99 3 3 Spearman rank correlations of site ranks among experts ( n = 10) for the aggregated younger stands (225 yrs old). ..................................................... 101 4 1 A summary of the key river reaches serving as the study domain for the cottonwood model. ........................................................................................... 149 4 2 Study team affiliations. ..................................................................................... 150 4 3 Forested stands in the study were discretized by age class ............................. 151 4 4 Description of the thirteen variables associated with the model ....................... 152 4 5 Component response index (CRI) algorithms for the model. ............................ 153 4 6 Descriptive statistics used to calibrate the individual variables in the ERI model. ............................................................................................................... 156 4 7 Results of the univariate analyses focused on the CRIs and ERIs ................... 159 4 8 Results of validation analyses focused on correlati ons and regression analyses of the CRI and ERI scores. ................................................................ 162 5 1 Transdiciplinary team members participating in the CRISIS exercise. ............. 192 5 2 Site suitability indicators and criteria descriptions for CRISIS .......................... 195 5 3 Spearman rank correlations of expert opinions of site selection criteria values ............................................................................................................. 199 5 4 Aggregated priorities based on expert elicitations and rank sum weighting. ..... 200 A 1 Existing cover types and aerial extents (in hectares) mapped i n the reference domain segments for the 20062008 time period. ............................................ 217 A 2 Post damming cover types and aerial extents (in hectares) mapped in the reference domain segments for the 19511958 period. .................................... 218
12 A 3 Pre damming cover types and aerial extents (in hectares) mapped in the reference domain segments for the 18921893 period. .................................... 219 E 1 Comparison of FQI values derived with the regression formula ....................... 230
13 LIST OF FIGURES Figure Page 1 1 The study domain spans the length of t he Missouri River Basin ........................ 27 1 2 The strategic spiral based decision support framework ...................................... 28 1 3 Under this structured decision making paradigm, experts and stakeholders are engaged throughout the ecosystem planning, ana lysis, and decision processes. .......................................................................................................... 29 1 4 Graphical road map for the dissertation. ............................................................. 30 2 1 Step 1 in the process call for the development of a conceptual model to uncover critical lines of evidence linking ecological endpoints to system drivers and stressors. ......................................................................................... 60 2 2 Study area for the Missouri River study. ............................................................. 62 2 3 The technical approach deployed to develop the conceptual model .................. 63 2 4 An overview of the conceptual model developed for the Missouri Rivers cottonwood modeling effort ................................................................................. 64 2 5 A closer look at the Driver Stressor relationships in the cottonwoods conceptual model. .............................................................................................. 65 2 6 Effects can be either directly measured or can be indicated by proxy variables ............................................................................................................. 66 2 7 In essence, t he cottonwood ecosystem responses have been conceptualized in terms of ecologically significant and important ecological indicators .............. 67 3 1 Step 2 in the process focuses on the mathematical for malization of ecosystem response indicators. ......................................................................... 93 3 2 The Segment 10 study area. .............................................................................. 94 3 3 Land Capability Potential Index (LCPI) classifications for Segment 10 .............. 95 3 4 An example of results derived from one of the five elicitations. .......................... 97 3 5 Comparison of expert opinions using intermediate elicitation results ................. 98 3 6 Intermediate elicitation results aggregated using weighted rank sums on an age class basis. ................................................................................................ 100
14 3 7 Final mathematical formalization of the LCPI categories based on expert opinion. ............................................................................................................. 101 3 8 Characterization of ecosystem response potentials ......................................... 102 3 9 Comparison of formalized LCPI scores at three sites in Segment 10 ............... 103 4 1 Step 3 in the process focuses on ecosystem response model construction ..... 146 4 2 Pre and Post damming trends in cottonwood community conditions on the Missouri River. .................................................................................................. 147 4 3 My spiral model development paradigm offers a unique opportunity for stakeholders to actively engage in the process ................................................ 148 4 4 Study area for the cottonwood ecosystem response model. ............................ 149 4 5 An example of the cover type mapping conducted for the study. ..................... 154 4 6 An example of the power analysis used to reduce the training data sets ......... 155 4 7 A comparison of CRI scores for the testing sites .............................................. 160 4 8 Predicted vs. actual scores generated in the holdout validation sites ............... 161 4 9 Comparison of central tendencies of the testing data sets ............................... 163 4 10 Predicted versus actual scores generated by the model for the dependent variable FQI ...................................................................................................... 164 4 11 Relative contribution of the three site types to overall ERIs scores in the testing sites located in Segment 10. ................................................................. 165 4 12 An overlay of the interpolated ERI scores and bald eagle nests located in Segment 10 ...................................................................................................... 166 5 1 Before the ecosystem response model can be deployed, intervention sites mu st be located. ............................................................................................... 189 5 2 Sieve mapping is a multistep recursive process ............................................... 190 5 3 The spiraling site selection process recursively addressed criteria selection, data processing, multicriteria evaluation and sieve mapping. ........................... 190 5 4 The Segment 10 study area ............................................................................. 191 5 5 Mitigation site suitability is dependent on hydrologic viability, ecologic integrity, and a measure of political necessity. ................................................. 193
15 5 6 CRISIS goal driven recovery site selection approach. ..................................... 194 5 7 Standardized criteria maps with normalization functions. ................................. 198 5 8 Ranks elicited from the stakeholders regarding their valu e preferences .......... 199 5 9 MCDA was used to combine team valueladen preferences ............................ 200 5 10 Central tendencies mapped for the ten crite ria ................................................. 201 5 11 Results of the sieve mapping application. ........................................................ 202 5 12 Original sites identified by experts overlayed on top of the wei ghted suitability map generated by CRISIS ............................................................................... 203 6 1 A word cloud depicting the major themes identified in the study. ..................... 215 B 1 Cent ral tendencies mapped for five of the thirteen variables used in the model. ............................................................................................................... 220 B 2 Box andwhisker plots for the system and segment level variables utilized in the model. ......................................................................................................... 221 C 1 Four of the 13 final RI curves developed for the cottonwood ERI model were calibrated using data collected from reference standard sites .......................... 222 C 2 Four of the 13 final RI curves developed for the cottonwood ERI model were calibrated using literature reviews and/or historical data.. ................................ 223 C 3 Two of the cottonwood models RI curves wer e developed using expert elicitation. .......................................................................................................... 224 C 4 The final RI curves for DISTPATCH (A) and PATCHSIZE (B) were calibrated using data at the reach level based on 1950s historical mapping ..................... 225 D 1 Comparison of the central tendencies of training site RIs ................................. 226 D 2 Comparison of the central tendencies of training site RIs ................................. 227 D 3 Comparison of the central tendencies of training site RIs ................................. 228
16 L IST O F ABBREVIATIONS 1 D One dimensional AHP Analytic Hierarchy Process (AHP) BiOp Biologi cal Opinion CMP Cottonwood Management Plan CRI Component Response Index CRISIS Cottonwood Recovery Integration Site Identification System CROC Cardinal Rank Ordering of Criteria DOER Dredging Operations and Environmental Research Program DPSIR Drivi ng force Pressure State Impact Response DSR Driving force State Response EBM Ecosystem based Management eDPSIR Enhanced D P SIR ERDC U.S. Army Engineer Research and Development Center ERI Ecosystem Response Index ESA Endangered Species Act ESH Emergen t Sandbar Habitat EWN Engineering With Nature Research Focus FQI Floristic Quality Index GIS Geographical Information System HEP Habitat Evaluation Procedures IMF Independent Measure of Function LCPI Land Capability Potential Index MCDA Multi Criter ia Decision Analysis MNRR Missouri National Recreational River NAVSYS Navigation Systems Research Program NRCS Natural Resources Conservation Service OWA Ordered Weight Average PSR Pressure State Response
17 RI Response Index RM River Mile S D SS Spatia l Decision Support System SSURGO NRCS Soil Survey Geographic Database USACE U.S. Army Corps of Engineers USEPA U.S. Environmental Protection Agency USFS U.S. Forest Service USFWS U.S. Fish and Wildlife Service USGS U.S. Geological Survey USNPS U.S. National Park Service VEC Valued Ecosystem Component
18 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy FINDING INTEG RATED SOLUTIONS TO WICKED ENVIRONMENTAL CHALLENGES: ECOSYSTEM BASED MANAGEMENT OF COTTONWOOD COMMUNITIES ON THE MISSOURI RIVER By Kelly A. Burks Copes May 2014 Chair: Gregory A. Kiker Major: Interdisciplinary Ecology A centurys worth of command and control river management decisions made by the U. S. Army Corps of Engineers (USACE) to reduce flooding, provide hydropower, increase navigation, and stimulate economic development along our nations waterways has resulted in numerous unintended system wide environmental consequences ranging from the degradation of water quality to the reduction of sediment transport and the loss of biodiversity. The latter has triggered Endangered Species Act (ESA) regulations forcing the USACE to take reasonable and prudent actions to recover threatened and endangered species imperiled by these flood control projects. With an ever growing sense of environmental awareness and appreciation, communities on these rivers have challenged the USACE to pursue holistic and sustainable solutions that adaptively restore ecosystems while maintaining flood protection for the floodplains inhabitants. The involvement of numerous stakeholders with disparate and often conflicting values and agendas generate a dynamic decision making envir onment riddled with critical knowledge gaps, teeming with uncertainty, and driven by high stakes negotiations perpetuated by a sense of institutional urgency to embrace quick fixes.
19 These highly uncertain, risky situations cannot be resolved solely with hard science or technical solutions. The USACE needs a transparent and prescriptive approach grounded in collaborative adaptive management to address these wicked problems. Here, a suite of decision support tools has been developed to assist the USACE in the ir endeavors. A n integrative framework is described herein that utilizes conceptual modeling to uncover critical lines of evidence that are then woven into a multimetric ecosystem response index to characterize benefits produced by proposed recovery plans. A series of GIS based participatory strategies are deployed to locate and prioritize potential recovery sites, and an expert based procedure is devised to forecast future conditions under a no action plan. A case study on the Missouri River addressing t he recovery of the plains cottonwood community ( Populus detltoides W. Bartram Marsh. S ubsp. m onilifera (Ait on) Eckenwalder ), is used to demonstrate the value of the spiraling strategy and the spinoff toolsets. USACE can now utilize the entire suite to adaptively co manage the system, transparently communicating the success of their recovery activities to the basins stakeholders and the public at large.
20 CHAPTER 1 AN INTEGRATED DECISION MAKING FRAMEWORK TO SUPPORT ECOSYSTEM BASED MANAGEMENT CHALLENGES IN WICKED SETT INGS 1.1 Introduction Over the past 100 years, the construction of dams, channels and levees on large, multi jurisdictional river systems across the country have produced enormous socioeconomic benefits (i.e., flood control, hydropower, navigation, water suppl ies for irrigation, recreation, etc.) at a significant cost to the environment (i.e., loss of sediment transport, degraded water quality, significant species loss) ( L ight et al., 2013 ; National Research Council, 2002 ; World Commission on Dams, 2000) The impacts to species in parti cular have triggered regulatory actions by the U.S. Fish and Wildlife Service (USFWS) under Section 7 of the Endangered Species Act of 1973 (16 U.S.C. 15311544) mandating reasonable and prudent actions to recover threatened and endangered species whose existence has been jeopardized by the construction and operation of these flood control projects. Unfortunately, these directives tend to narrowly focus on the species of concern, ignoring the systemic issues threatening the integrity and resilience of the system as a whole ( Benson, 2012 ) Recognizing the limitations of this species based approach, regulatory agencies are now transitioning to a more holistic Ecosystem Based M anagement (EBM) paradigm, one that promotes largesca le, long term recovery of ecosystem function and integrity from a systems perspective ( McLeod and Leslie, 2009; Ruhl, 2008) Moreover, the natural resource agencies responding to this regulatory paradigm shift are themselves embracing the EBM concept and expanding its definition to acknowledge that humans are key components of the ecosystem, and that healthy ecosystems are
21 essential to human well being ( National Research Council, 2005a ) As an example, the Department of Defense now defines EBM as: A goal driven approach to managing natural and cultural resources that supports present and future mission requirements; preserves ecosystem integrity; is at a scale compatible with natural processes; is cognizant of natures timeframes; recognizes social and economic viability within functioning ecosystems; is adaptable to complex and changing requirements; and is realized through effective partnerships among private, local, State, tribal, and Federal interests. Ecosystem based management is a process that c onsiders the environment as a complex system functioning as a whole, not as a collection of parts, and recognizes that people and their social and economic needs are a part of the whole.1This expanded concept thus calls for the management of ecosystems wi th the goal of maintaining the sustainable production of ecosystem goods and services ( Reed et al., 2013) to promote human well being by collaboratively engaging stakeholders in adaptive management to int egrate and balance ecological, social, and economic objectives ( White House Council on Environmental Quality, 2009, 2012) Unfortunately, recovering highly degraded ecosystems from and EBM perspective presents some unique challenges. The requi site involvement of numerous stakeholders with differing and oftentimes conflicting values, agendas, and mindsets generates a dynamic decisionmaking environment driven to high stakes negotiations. Moreover, there is often a sense of overriding urgency dri ving these activities that leads to both risky and inevitably controversial decision making. Inevitably, wicked problems ( Rittel and Webber, 1973 ) arise when the priorities of stakeholders shift or evolve as the processes of integration and reflection continually unveil new concerns altering the focus of the recovery efforts ( Light et al., 2013 ; Moser et al., 2012; Poff et al., 2003 ) 1 Department of Defense Instruction No. 4715.03 http://www.dodnaturalresources.net/files/DoDI_4715_03.p df p age 31 (Accessed Sept 2013).
22 Regrettably, these highly uncertain, risky situations cannot be resolved solely with hard science or technical solutions their complexity mandates prescriptive participatory techniques to fill knowledge gaps and promote transparency and confidence in the proposed solutions. 1.2 Problem Statement Across the country, the U. S. Army Corps of Engineers (USACE) is in the midst of planning broadscale ecosystem based mitigation studies to recover critical ecosystem impacted by the damming and regulation of flows on large river systems ( USACE, 2013a,b,and c, d). Like other natural resource agencies attempting to implement EBM in these types of situations, the agency is confronted with several overarching problems: 1. How to plan and implement effective recovery plans for thousands of miles of impacted ecosystems when shifting, oftentimes conflicting political, social, and ecological agendas influence the decision outcomes ? 2. How to measure success, when ecosystem integrity or system wholeness can never be restored with any degree of certainty? 3. How to efficiently integrate data, models and expert knowledge in a transparent manner that is both informative (risk based and scientifically driven), visually engaging (promoting rapid communication), yet adaptive (proactively responsive to uncertainty) in dynamic decision making environment over the long term? 4. H ow to effectively engage stakeholders and scientists alike in capturing the uncertainties arising from natural variability as well as those uncertainties stemming from the application of the locally derived knowledge, information and opinions to formulate sustainable and resilient solutions? The USACE planners and managers (and their sister agencies as well as their local and regional partners) require frameworks to effectively integrate data, models and expert knowledge into a decision making process in a manner that is both scientifically driven and visually engaging, yet adaptive in nature to respond to the dynamics of the
23 situations. In other words, an integrative framework and decision support tools are needed to link proposed interventions (i.e., easem ents, plantings, restoration of backwater sloughs, etc.) to ecosystem response, such that recovery plans can be formulated to address seemingly intractable technical, environmental and social problems inherent to these complex socio ecological conflicts. 1.3 G oals and Objective s The primary goal of this research is to develop and apply a spiraling decisionmaking framework ( Du Toit, 2005) that conforms to the principle tenets of EBM and support s regional recovery efforts by : promot ing reflection and recursive learning, merg ing applied ecology with deliberative insights garnered from prescr iptive methodologies ( Albar and Jetter, 2009; Riabecke et al., 2012) and adopt ing rigorous and defensible procedures to characterize ecosystem response to proposed recovery measures encouraging transparency and collaborative rationality ( Innes an d Booher, 2010 ) A case study on the Missouri River ( Figure 1 1 ) offers a demonstration platform to test the approach focused on generating a series of casespecific products targeting the basins declining cottonwood ecosystem ( Populus detltoides W. Bartram Marsh. S ubsp. m onilifera (Ait on) Eckenwalder ) The objective is to deploy the framework and: 1. C onceptually model an ecosystem and uncover lines of evidence that can be used to build an ecosystem response model to measure rec overy plan performance; 2. F ormalize lines of lines of evidence using scientifically defensible, transparent mechanisms based on professional judgment; 3. G enerate a multiscalar, multimetric ecosystem response model to assess recovery plan performance;
24 4. P roduce spatially explicit criteria and combine these using M ulti C riteria D ecision A nalysis (MCDA) techniques in order to identify, prioritize and select mitigation sites based on stakeholder values and preferences. The intent is to provide a responsive, defens ible, efficient, and operational approach that can be readily implemented within the constraints of agency policy and guidance ( USACE, 2000 ; 2003) with an eye towards cross walking the approach to other applications both inside and outside the USACE. 1.4 A Spiraling Framework for Ecosystem Planning The framework presented here is in essence an adaptation of the Guisan and Zimmermans ( 2000) model building process that has been modif ied to incorporate standard agency planning steps ( USACE, 2000, 2003 ) and altered from the traditional waterfall linear stepwise approach described in Royce (1970) to a spiraling strategy that calls for the continuous refinement of study tasks in a cyclical, recursi ve fashion ( Boehm, 1988; Boehm et al., 2012) ( Figure 1 2 ). Each step utilizes face to face meetings with the stakeholders to review progress and reflect on the previous decisions governing the outcomes, interjecting new information into the process to hone or refine the results in an incremental fashion. Unlike the traditional waterfall approach, spiraling allows new information to be incorporated into the process as it becomes available or is revealed (i.e., prototyping). As the figure conveys, t he spiraling framework serves as an operational roadmap guiding both the development of the ecosystem response model (s) ( S teps 1 3 ) and the application of the model (s) in assessments of proposed recovery opportunities ( Steps 4 6 ). To begin the planning process, the system is conceptualized (using influence diagrams) based on experimental data and experiential knowledge th at uncover key lines of evidence and measureable indicators of ecosystem integrity ( Step
25 1 ). Formalization of the indicators using literature, expert opinions and statistical analysi s of data is used to generate mathematical characterizations of the lines of evidence ( Step 2 ). The ecosystem response model is then derived using these formalized endpoints and in turn sub mitted for calibration, verification and validation procedures using reference based datasets ( Step 3). At the same time that the model s ar e under construction a site selection process can begin to identify, prioritize and select recovery sites to meet the study s goals and objectives. Once the models are complete and the sites have been selected for recovery the study team engages in futur e forecasting to describe conditions under both a No Action condition as well as the response of the ecosystem to the proposed recovery plans on each site ( Step 4). Alternatives are then assessed with the model and compared against costs to restore or re habilitate the sites ( Step 5 ), and a recommended plan is selected. The final step involves both the construction and monitoring of the sites success based on performance targets and response thresholds tied to the ecosystem response models parameters ( St ep 6). Adaptive co management activities direct the planners and managers to evaluate success (or failures) and learn from the feedbacks provided by the monitoring activities triggers the feedback spiral. Figure 1 3 suggests ho w important stakeholder feedback is to the process, illustrating injection point where their expertise and value laden preferences are integrated into the recursive and reflexive decisions Stakeholder feedback is initially used to capture and understand p roblem dimensions, to decide what data to collect and what not to collect, to choose what models to build, to interpret the results of any data collection to develop alternatives to forecast future conditions, to quantify results, and
26 to compile all the relevant information together to analyze and solve the problems. As a feedback loop, their judgment is in turn used to monitor the situation and trigger adaptive co management if exceedence thresholds are encountered over the life of the project. For the m ost part, the use of expert opinion is handled informally within the framework, oftentimes occurring behind the scenes At critical juncture s, however, opinions are formally elicited and aggregated using multicriteria analysis techniques that are fully doc umented, assuring transparency and instilling confidence in the study outcomes. 1.5 Road Map The objectives of this research have been intentionally aligned with the specific spirals of the decision support framework. As such, this d issertation has been divide d into chapters paralleling the application of the framework on the demonstration site Figure 1 4 Chapter 2 focuses on the development of conceptual models, while Chapter 3 focuses on the formalization of one of the line s of evidence contributing to the overall ecosystem response model which is itself presented in Chapter 4 Chapter 5 focuses on the selection and prioritization of site selection criteria and uses these to pinpoint particular sites in the study area that merit cons ideration for recovery initiatives. Chapter 6 summarizes the results, offering a discussion on the efficacy of the approach, highlighting research themes, describing next steps, and discussing future research opportunities.
27 Figure 1 1 The study domain spans the length of the Missouri River Basin located in the Midwest region of the U nited States and is specifically focused on the cottonwood community ( Populus detltoides W. Bartram Marsh. S ubsp. m onilifera (Ait on) Eckenwalder ) lining the banks of the Missouri River ( and its tributaries ) which serve as critical habitat for the recently delisted bald eagle ( Haliaeetus leucocehpalus ).
28 Figure 1 2 The strategic spiral based decision support framework developed herein to help planners and managers implement EBM strategies in the recover y of critical ecosystems using recursive and reflexive techniques to iteratively derive sustainable solutions.
29 Figure 1 3 Under this st ructured decision making paradigm, experts and stakeholders are engaged throughout the ecosystem planning, analysis, and decision processes. This strategy present s a deliberative environment where expert questions and responses direct and inform the proces s in every spiral
30 Figure 1 4 Graphical road map for the dissertation.
31 CHAPTER 2 U NCOVERING THE LINES OF EVIDENCE HIDDEN INSIDE WICKED PROBLEMS: USING CONCEPTUAL MODELS TO INFORM ECOSYSTEM BASED MANAGEMENT OF THE MISSOURI RIVER COT TONWOODS 2.1 Introduction The construction and operation of dams, channels, and levees for flood control and navigation purposes on large, multi jurisdictional river systems has resulted in significant (but unintended) system wide conflicts in ecosystem goods and services provisioning at numerous spatial and temporal scales nationwide ( Light et al., 2013; Poff et al., 2003 ; World Commission on Dams, 2000) From the mouth of the Mississippi ( Moser et al., 2012 ) to the banks of the Missouri ( National Research Council, 2002) to the headwaters of the Columbia ( Watson, 2012) managers charged with the recovery of ecosystem integrity now face wicked socio ecological challenges ( Rittel and Webber, 1973) stemming from conflicts arising between competing interests, shifting goals and objectives, gaps in scientific understanding, and constraints on time and resources ( Light et al., 2013 ) In these situations, responsive and meaningful Ecosystem Based Management (EBM) strategies ( McLeod and Leslie, 2009 ) must plan for and embrace the involvement of numerous stakeholders, mediating their disparate, oftentimes conflicting mindsets, values, and agendas, in order to effectively forge a resilient and sustainable recovery plan. Across the country, the U.S. Army Corps of Engineers (USACE) is undertaking largescale, multi disciplinary studies to mitigate and recover critical ecosystems impacted by the regulation of l arge river systems ( USACE, 2013a,b,c,d ) These studies often face similar challenges h ow to mitigate for thousands of miles of impacts under shifting, oftentimes conflicting political, social, and ecological decisionmaking
32 paradigms and how to sufficiently estimate each recovery plans return on investment. Moreover, the USACE (and its stakeholders) must acknowledge and accept at the onset that ecosystem integrity or wholeness will likely never be restored with any degree of certainty as long as flood control remains the primary mission and desired ecosystem service output of the system. Regrettably, t hese highly uncertain, risky situations cannot be resolved solely with h ardscience solutions ( McIntosh et al., 2007 ) their complexity calls for a transparent and proactive post normative softer approach ( Funtowicz and Ravetz, 1992 ) grounded in creative problem solving ( Parnes, 1992 ) fast and frugal heuristics ( Gigerenzer, 2007) transformati ve design ( Sangiorgi, 2011 ) and a daptive co management ( Armitage et al., 2009; Cundill and Fabricius, 2009; Light et al., 2013 ) The problem becomes one of collaborative and transparent decision making, particularly when exploring lines of evidence that can clarify management goals and objectives. These lines of evidence serve as causal pathways characterizing and interpreting ecosystem response using information arising from a variety of sources or derived using various techniques (adapted from U.S. Environmental Protection Agency (USEPA), 2013 ). These pathways reveal economically important and socially relevant endpoints that direct managers towards performance metrics (i.e., ecolo gical indicators) that can be used to consistently estimate and report the anticipated consequences of recovery plan s with respect to predefined success criteria (adapted from Gregory et al., 2012) The USACE, its partners and collaborators require frameworks to effecti vely integrate data, models and expert knowledge into the decisionmaking process in a manner that is both informative (risk based and scientifically driven) and visually
33 engaging (promoting rapid communication), yet adaptive (proactively responsive to unc ertainty) in dynamic situations. Moreover, to capture the principle tenets of EBM, the solution must effectively and efficiently distill the ecosystems structure, function and dynamic processes down to a meaningful estimate of ecosystem state and response to management strategies to be useful. In other words, an approach is needed that uncovers the complicated lines of evidence tying management measures to ecosystem response such that recovery plans can be formulated to address seemingly intractable techni cal, environmental and social problems inherent to these complex socioecological conflicts. 2.2 Goals and Objectives In the last chapter I present ed my solution to this conundrum a structured ecosystem assessment approach that fuses soft and hard sciences in a meaningful manner to tame the wicked problem of EBM confounded by human dynamics I also promoted the idea of spiral modeling construction a process that encourages constant reflection, active learning and hypothesis driven monitoring an approach akin to adaptive management In this chapter I begin to demonstrat e the efficacy of the fusion approach by using the Missouri River study as a pilot test Here I focus strategically on the first step ( Figure 2 1 ) the development of a conceptual model characterizing the dynamic and complex plains cottonwood ecosystem ( Populus detltoides W. Bartram Marsh. S ubsp. m on ilifera (Ait on ) Eckenwalder ) lining the banks of this highly regulated river system My objective is to present a spiral ba sed conceptual modeling approach that untangles lines of evidence within a wicked problem. These conceptual models and the derived lines of evidence can be used to assess EBM effectiveness and aid in the
34 recovery of ecosystem integrity on large, regulated river systems. An example on the Missouri River demonstrates the value of this approach, producing a casespecific conceptual model that reveals critical lines of evidence and pinpoints key performance metrics that are now being used to compare and contras t the efficacy of proposed recovery plans for the watersheds plains cottonwood community. This chapter is divided into three primary sections. The first section focuses on EBM and highlights the role conceptual modeling plays in uncovering critical lines of evidence tying complex ecosystem processes to indicators of plan performance. In the methods section, the Missouri River case study is introduced, and both a spiraling model development process and a conceptual modeling template are presented to guide m anagers through the early stages of EBM. In the results section, a conceptual model for the Missouri River cottonwoods is presented and a discussion section describes how the models outputs (i.e., performance m etrics) will support not only the development of longterm recovery goals but inform the formulation and comparison of recovery options and establish adaptive comanagement thresholds that will trigger management responses in the future. To conclude, lessons learned, next steps, and future research opportunities are described. 2.3 Literature Review 2.3.1 Ecosystem B ased M anagement Ecosystem Based M anagement (EBM) provides a systemslevel methodology to deliver ecosystem goods and services to humans by means of natural capital conservation, preservation and res toration ( Gregory et al., 2013; Kareiva et al., 2011; McLeod and Leslie, 2009 ) In essence, an ecosystem has integrity when its dominant characteristics (i.e., composition, structure, function and processes) occur within its
35 natural range of variation (reference conditions) and is sustainable when it is resilient (i.e ., it can withstand and recover from most perturbations imposed by natural environmental dynamics or human disruptions) ( Dale and Beyeler, 2001; Society for Ecological Restoration International, 2004) The general return on investment for a society under an EBM paradigm is directly attributed to the improvement of an ecosystems integrity, measured in terms of ecosystem response to a variety of proposed changes via ecological indicators, and couched in terms of benefits (i.e., the ecological, sociological or economical gain provided through the production of an ecosystem good or service to promote human well being) (adapted from van O udenhover et al., 2012; Fisher et al., 2009 and references therein). The key to effective adaptive co management is the deployment of indicator based ecosystem response models that facilitate the monit oring of ecosystem status and its response to human interventions based on success criteria or performance measures tied to project goals and objects, and the establishment of triggers (i.e., ecological response thresholds) dictating a change in management activities ( Cundill et al., 2012 ; Cundill and Fabricius, 2009; Linkov et al., 2006 ) In EBM there is a continuum of management actions ranging from passive to active solutions that might produce a measureable and productive ecosystem response. Decision making in this context is facilitated by the identification of critical ecological indicators that best characterize and quantify ecosystem integrity and measure ecosystem response to proposed design alternatives ( Gentile et al. 2001 ; Kandziora et al., 2013; van Oudenhover et al., 2012) .Ideally, the establishment of EBM goals and objectives involves the close linkage between scientists and decision makers. Science
36 informs the process by characterizing the ecological conditions that are achieved under particular management scenarios, and decisions are made that reflect societal values including economics, politics, and culture. Because EBM is inherently adaptive (i.e., management activities are adjusted as necessary to achieve goals), the stakeholders must be routinely apprised of progress toward achieving the desired goals. Thus ecosystem response models serve to measure the ecosystems integrity and offer an avenue to guide monitoring and assessment to assure goal attainment (i.e., success). While specific USACE led ecosystem recovery efforts, monitoring programs, and landscapelevel assessments ultimately require sit e specific spatially explicit ecosystem response models to characterize impacts and benefits, system wide programs managed by the USACE are often initiated with more generalized conceptual modeling frameworks to facilitate communication amongst stakeholder s regarding ecosystem conditions and responses to proposed EBM activities. As a result, there are specific constraints that drive the development of ecological models in these large multi jurisdictional river studies administered by the USACE. For example, the USACE has extensive planning policies and guidance that must be closely aligned with conceptual modeling goals and objectives in order to sufficiently address the agencys missions and mandates ( USACE, 2000; 2003). As mentioned previously, the majority of USACE studies deploying EBM are large in scope and scale, and highly contentious, so any model developed to su pport and inform the process most avoid ambiguity (i.e., drivers, stressors, and particularly performance metrics must be understandable, direct, complete, concise and readily operational). Moreover, any modeling construct must be easy to adapt or modify i n the early reconnaissance phases in order to ensure holism
37 and encourage the inclusion rather than the exclusion of stakeholder perspectives. Finally, it is important to recognize that legal decisions (biological opinions, determinations of jeopardy, envi ronmental operating windows, wetland regulations, etc.) often limit the temporal and spatial scope of USACE decisions, and any characterization of the system must take these directives into account, yet afford the decision makers the freedom to consider a lternative solutions to address these concerns. 2.3.2 Conceptual M odel ing From Theory to Practice C onceptual model s are purposeful abstractions of reality designed to: 1) inform EBM by communicating context (timing, spatial scale, threats, etc.), 2) characteriz e hypothetical ecosystem response to management proposals, and 3) characterize the potential status of economically important and socially relevant endpoints ( Dennison et al., 2007; Fischenich, 2008 ) More importantly, conceptualization can inform and clarify EBM goals and objectives while offering a mechanism to establish performance criteria and flag critical thresholds that trigger adaptive responses ( Fischenich, 2008; Gregory et al., 2013 ; Henderson and O'Neil, 2007) In terms of E BM, conceptual models can serve as a venue to diagnose underlying EBM problems and highlight potential threats, offering a forum to present, communicate, and integrate transdisciplinary perspectives ( Fischenich, 2008) These t ransdisciplinary perspectives are drawn not only f rom academic researchers coming from different unrelated disciplines, but also include experiential perspectives garnered from nonacademic s in applied disciplines (e.g., natural resource managers, enduser groups and the general public ) ( Tress et al., 2005 ) The degree to which the se stakeholders are engaged in the process can vary depending on scope and magnitude
38 of the effort. Gregory et al. (2013) suggest a variety of options including: 1) models built by facilitators and presented to stakeholders for follow on modification and adaptation; 2) models built by a small group (including the facilitations) that are the n presented to the wider group of stakeholders as a stimulus for discussion; and 3) models built by the entire stakeholder community in a participatory manner. The old adage what gets measured, gets m anaged is particularly relevant in the context of EBM. Management activities are explicitly designed to illicit a particular ecosystem response, and conceptual models can be employed to both identify the stressor levels stimulating the response, and presc ribe indicators of ecosystem integrity to measure the status of an ecosystem along the continuum of responses ( Henderson and O'Neil, 2004 ) Their intent is to both make predictions about the relative efficacy of E BM options and to identify and explore the critical uncertainties that minimize their ability to accurately predict responses to interventions thereby facilitating their ability to choose the best course of action or the best investment strategy ( Rumpff et al., 2011) In the end, a sufficiently comprehensive conceptual model can serve as a precursor to more formal numerical modeling activities ( Gentile et al., 2001 ; Kandziora et al., 2013; Niemeijer and De Groot, 2008) 2.3.3 Conceptual Modeling Mechanics The process of conceptual modeling is in essence, a series of core steps, guiding principles, and structuring tools that generate a thought map of the modeling domain and problem space. In most instances conceptual models are presented as descriptive nar ratives and illustrated using influence diagrams (box andarrow flowcharts) that depict the causal relationships among natural forces and human activities that produce ecosystem responses ( Dennison et al., 2007; Gregory et al., 2013; Niemeijer and De
39 Groot, 2008) Reports by Fischenich (2008 ) and Henderson and O'Neil (2004 2007 ) offer direct guidance from USACE on developing conceptual models for planning studies, while papers from Davis et al. (2005 ); Gentile et al. (2001); Harwell et al. (1999) offer examples that have actually been applied by USACE and its partners, collaborat ors, and stakeholders to characterize recovery efforts in the Everglades, FL. Each of these models offered insight into primary components of a conceptual model (i.e., drivers, stressors, endpoints, etc.) and suggested using linear graphics to draw single lines of evidence between drivers and their endpoints. Alternatively, Niemeijer and De Groot (2008) offer a causal mapping approach using the Driving forcePressure State Impact Response (D PSIR) framework, but modified the approach to draw multiple line s of evidence routi ng through the modeling components in a network fashion. Although no two conceptual models are alike, their structure, composition, and function can be used to organize them into meaningful categories based on their similarities and differences ( Table 2 1 ). Conceptual modeling can be initiated by offering evidence as to ecosystem condition and expos ing causal pathways that establish Stressor Ecosystem Response relationships relevant to the studys goals and objectives. Stressors ( i.e., the physical, chemical, and biological changes driven by natural and humancaused forces altering ecosystem structure and/or function) ( Gentile et al., 2001 ; Gucciardo et al., 2004; Henderson and O'Neil, 2004) are then routed and aggregated using a Valued Ecosystem Components (VEC) matrix These VECs span the gamut of organizational hierarchy (i.e., species, populations, communities, system) capturing ecosystem condition from the stand, segment, and system temporal and spatial scales. Bo th
40 natural (i.e., biogeography, watershed physiography, biogeochemistry etc.) and anthropogenic Drivers (i.e., water resource management, land use management, human demographics, etc.) cause stress on the system altering VECs producing evidence of ecosystem condition ( Henderson and O'Neil, 2004, 2007 ) These stresses can be mediated by direct interventions (i.e., plantings, invasive species removal, etc.) or indirect actions or phenomena (i.e., succession, sediment deposition or scouring, etc.). Endpoints (i.e., loss of species diversity, declines in st ructural integrity, increased disturbance, etc.) are identified to provide insights into the state of the ecosystem ( Gucciardo et al., 2004 ) These endpoints represent a significant nexus between what soci ety values in terms of goods and services production and what is ecologically important to maintain ecosystem integrity and assure ecosystem resiliency and sustainability ( Gentile et al., 2001; Kandziora et al., 2013) Just as the stressors span the continuum of organizational hierarchy and spatial and temporal scales, endpoints can range across numerous ecosystem types (i.e., forests, lakes, marshes, etc.). Ecological indicators offer specific, measureable, discrete, but not necessarily independent performance metrics that can be used to quantify the condition or state of the endpoints, and have known or hypothes ized responses to stressors ( Gucciardo et al., 2004; Harwell et al., 1999 ; Kandziora et al., 2013) These metrics span intergenerational time scales, multidimensional spatial scales, and operate at varying degrees of granularity (i.e., the level of detail considered in a model or decision making process). Some e xamples of ecological indicators operating at the site (stand) level include species richness, herbaceous cover, and indicators of hydrologic regime. Some
41 examples at the regional (river segment) level include patch dynamics such as size, number, shape, gy ration and core to edge ratios. At the landscape (system) level, ecological indicators include characterizations of adjacent land use, connectivity or fragmentation such as nearest neighbor measurements and levels of habitat interspersion. These indicator s can be sorted by compositional (i.e., the variety of elements in the ecosystem), structural (i.e., the organizations or patterns within the ecosystem) or functional (i.e., processes in the ecosystem) categories to generate meaningful typologies that faci litate indicator screening and selection ( Scott et al., 2005 ) The process to select key performance metrics is a relatively lengthy undertaking that requires several iterations to stabilize the final portfolio. Although the final list of metrics is usually settled on as a matter of compromise, the transdisciplinary teams mandate must always be to select metrics that can register stressor intensity, frequency, duration, and distribution i.e., distinguish amongst proposed recovery plans with a high degree of confidence and rigor ( Harwell et al., 1999) 2.3.4 Spiral M odeling Jakeman et al. (2006) and Jrgensen and Bendoricchio (2001 ) offer general model development guidance promoting a modified waterfall strategy ( Royce, 1970 ) Their 10 step approach moves from concept through design, implementation, testing, and deployment with some iteration to assure that the goals and objectives are being addressed. Although the approach seems rational and logical, it assumes that all knowledge and data is readily available at the onset, and does not allow for much reflection, revision, or interaction with stakeholders in a collaborative environment. Alternatively, Bo ehm (1988) offers a spiral based approach that calls for the continuous
42 refinement of incremental versions of a conceptual models construct (i.e., prototyping) in a cyclical, recursive fashion ( Boehm, 1988; Boehm et al., 2012 ) Each cycle utilizes face to face meetings with the stakeholders to review the prototype and reflect on the previous decisions governing its architecture, interjecting new informat ion into the process to hone or refine the models structure and behavior in incremental fashion. Unlike the traditional waterfall model ( Royce, 1970) spiral modeling allows components or lines of evidence to be added to the model when they become available or are revealed. The approach is adaptive and responsive in nature and compatible with USACE policy and guidance regarding collab oration and adaptive co management strategies in EBM ( USACE, 2000 ; 2003 ). Notably, only one study by Bredeweg et al. (2008) has transparently deployed a spiral approach to build a conceptual model as a prelude to a more formal causal map that characterized the influence of human activities on basic features of a mountain stream in Bulgaria. In highly dynamic situations, they noted that the spiral approach afforded an opportunity to recursively reflect on the models goals and objectives and make c ourse corrections when necessary in response to shifting directives from multiple stakeholders a common occurrence in wicked socioecological conflicts. 2.4 Methods 2.4.1 Study A rea and P roblem S pace The Missouri River is the longest river in the United States, dr aining more than 1.3 million km2 ( 530,000 mi2), and has an extensive basin footprint that covers approximately one sixth of the continental United States ( Galat et al., 2005) ( Figure 2 2 ) Prior to its regulation, the Missouri River was known for its shifting channels, high turbidity, and periodic floods. These dynamic events maintained the
43 ecological health of the surrounding landscape (particularly the fringing riparian cottonwood forests) by providing moisture to sustain vegetation, fine sediment deposition, and nutrient transport that enhanced s oil fertility stimulating decomposition, seed deposition, and sandbar formation ( Dixon et al., 2012) In response to the Great Flood in 1927, and with the intent of stimulating economic growth and prosperity in the American Midwest at the height of the Great Depression (1920s 1940s), President Franklin D. Roosevelt signed into law the Flood Control Act of 1944 (P.L. 78 534). This leg islation authorized the construction of thousands of dams and levees across the United States, and the establishment of the Pick Sloan Missouri River Basin Program. As a result, over 50 dams and lakes were built on the Missouri six on the main stem alone. Today the system is managed for multiple purposes (i.e., ecosystem goods and services) including navigation, flood control, hydropower, public water supply, recreation, and fish and wildlife habitat ( Jacobson and Galat, 2008) The unintended consequences of these activities have left their mark on the landscape, and continue to this day. What was once a dynamic river system, notorious for its large floods, meandering across the floodplain transporting significant quantities of sediment downstream, has become a highly channelized system fixed in place by dikes and revetments regulated by dam releases. Today, the amplitude and frequency of the rivers natural peak flows have been sharply reduced ( Jacobson and Galat, 2008; National Research Council, 2002) As such, t he river no longer experiences natural spring/summer rises and the ecologically ben eficial low flows at other times of the year have been all but lost ( National Research Council, 2 002) Much of the rivers sediment load has been deposited in the large reservoirs behind the dams, resulting in sediment
44 imbalances and marked channel incision below the dams. The massive reservoirs have submerged significant stretches of riparian fores ts along the banks. Once these flood prone areas were protected from flooding, agricultural and urban development encroached on the floodplains ( Dixon et al., 2012; Johnson et al., 2012; Scott et al., 2012) Reduction of the ecologically beneficial flood pulse in combination with meander loss and land conversion has resulted in the loss or alteration of nearly three million acres of floodplain habitat system wide ( National Research Council, 2002) For th e disturbance focused ecosystems on the river that depend on pulsing river dynamics to create substrate for colonization, and flood flows to recharge soil moisture, transport sediment and disperse seed, such as t he cottonwooddominated riparian forests ( Populus deltoides) the regulation of the system has been particularly significant Recently released studies have shown that regeneration of cottonwoods (historically the most abundant and ecologically important s pecies on the systems extensive floodplain), has largely ceased ( Dixon et al., 2012; Johnson et al., 2012; Scott et al., 2012 ) a trend foretold by scientists in the early nineties ( Johnson, 1992 ) Cottonwood recruitment today occurs only sporadically in downstream reaches that receive overbank flooding as a result of extreme events such as the 1951, 1993 and 2011 floods ( National Research Council, 2002 ) The re maining cottonwood forests on the floodplain today are highly exposed, extremely s ensitive, and not likely to be sustained by the current regulated flows (i.e., their adaptive capacity is rapidly diminishing) ( Dixon et al., 2010; USACE, 2010) One of the primary factors influencing species survival in the basin today is loss of habitat ( National Research Council, 2002) To assure survival, wildlife species require
45 essential ecos ystem services including food, water, and sufficient shelter for both themselves and their offspringand wintering bald eagles ( Haliaeetus leucocehpalus ) on the Missouri River are no exception ( Guinn, 2004 and references therein, USACE, 2010) In 2000, a tipping point was reached and the U.S. Fish and Wildlife Service (USFWS) issued a Biological Opinion (BiOp) advising the USACE that their operation of the Missouri River system had severely altered the ecosystems resources, concluding that proposed USACE actions to operate and maintain the systems levees, channels and dams would likely jeopardize the continued existence of several species (i.e., least tern, piping plover, and pallid sturgeon) ( USACE, 2000). In 2003, the USFWS amended the BiOp to include reasonable and prudent measures to recover ecosystem integrity of cottonwood forested communities on the river to minimize the take on bald eagles owing to their threatened status under Sectio n 7 of the Endangered Species Act of 1973 (ESA) ( USACE, 2003). Even though the bald eagle was later delisted (ESA 71 FR 8238, February 16, 2006), the USACE has continued to follow the directives of the 2003 BiOp, recognizing that the preservation and establishment of cottonwood communities through plantings, discouragement of clearing, and easements will provide habitat for many species in cluding and in addition to the bald eagle. In 2010, the USACE developed a Cottonwood Management Plan (CMP) ( USACE, 2010 ) to address the reasonable and prudent measures mandated in the 2003 BiOp with the express intent ion of providing a single, comprehensive strategy to guide the efficient and effective preservation and rehabilitation of critical cottonwood community structure and function in the Missouri River Basin.
46 2.4.2 Transdisciplinary Teaming To support the development of the USACE cottonwood recovery plan, a team of regional and local experts was convened in 2006. The makeup of the team was intent ionally inclusive rather than exclusive, ranging from USACE planners and regulators, to ecologists and hydrologists from academia, to local natural resource managers, as well as private contractors responsible for the writing of the recovery plan itself. Over the course of the next seven years (2006 2013), 82 stakeholders spanning the range of academic (17), tribal (6), federal (41), state (11), local (1), private (3) and nongovernmental (3) interests were engaged in the process. Planners, regulators, ecol ogy professors, foresters, natural resource managers, and consultants all participated in the study. The majority of the team members had between 5 and 20 years of experience in the basin, and onefourth of the experts had well over 20 years of direct expertise on the river. Each year, a week long workshop was held onsite (in Vermillion, SD) to elicit expert knowledge with the intent of developing an ecosystem based response model to assess the efficacy of proposed recovery alternatives being formulated by the USACE. On average, 25 to 30 team members attended these workshops at any one time. D irected but open dialogues were used to consider tradeoffs, competing hypotheses, model character and potential utilization to guide the team through the process usin g probing techniques described in Meyer and B ooker (2001 ) The workshops served as a forum to present and review the data gathered thus far, and offered a unique opportunity to present and reflect on each modular prototype as it evolved. The objective was to combine what the experts thought shoul d be included in an assessment of ecosystem integrity with the realities of how endusers would actually be
47 making these decisions (based on the planning and management strategies described in the CMP ( USACE, 2010 ) with the intent of creating a methodology that was defensible, transparent, and efficient. In the months between each onsite meeting, several sub teams focused on systems ecology, field data collection, and hydrology, gathering data on the ecosystem and mapping the extant and historic coverage of cottonwood communities in the basin. To assure forward momentum, monthly subteam teleconfer ences and web meetings were used to delve deeper into the information generated from the workshops as well as review and integrate new data into the prototypes as it became available. 2.4.3 Spiraling B ased C onstruction Strategy After just one onsite workshop, i t became evident that the systems problems were messy (i.e., wicked) and that any number of ecological indicators could be used to assess ecosystem integrity to some extent although whether they would be effective in terms of distinguishing amongst pr oposed plans was unclear and remained questionable. Without some guiding construct or thought map to synthesize and simplify the modeling domain and problem space in the experts minds, any effort to construct a response model would likely fail. Budgetar y and resource (time and personnel) constraints precluded the convening of additional workshops to develop a conceptual map in a live forum with the team. As a result, the research team opted to develop a conceptual model separate from the expert group usi ng a strategy described in Gregory e t al. (2013) which was supported by both an extensive literature review and details stemming from notes generated in numerous study workshops and follow on teleconferences. A spiraling approach was used to uncover the experts lines of evidence and represent them
48 according ly in the causal map ( Figure 2 3 ). The process began with a problem defining activity where a straw man version of the conceptual model was developed based on the literature and the notes from the first meeting. Each year long spiral thereafter served as a reflexive developmental phase that engaged the team in a recursive critique of the emerging lines of evidence. At each subsequent meeting (both interim teleconferences and the yearly workshops) the latest vers ion of the prototype was presented based on the knowledge and data gathered up to that point, and the team was asked to offer constructive feedback to increase the level of the models representativeness, until the model eventual ly coalesc ed into a stabili zed construct. These dialogues were structured around the Interactive Group Methodology ( Meyer and Booker, 2001 ) where the research team served as the groups facilitators asking the team questions, clarifying context, minimizing biases, and exploring the various of lines of evidence. In these sessions, the ques tions for the team focused specifically examined their value based perceptions asking, What is important? and defining model behavior by asking, What are the consequences? Lines of evidence were uncovered when the participants were asked to: 1) describe the system in their own words, 2) compare the system to others in their experience, 3) associate evidence of the decline of ecosystem integrity with indicators of change, 4) analyze the degree to which the system had changed, 5) brainstorm interventions, and 6) give reasons why some interventions might be more productive than others. Each time an observable Effect was identified by the participants, the facilitators asked the group to link the ecosystems response to a Stressor or a group of Stressors
49 wh ich were then linked to system Drivers Then the facilitators asked the participants to identify potential Ecological Indicators (i.e., patch size, cottonwood domination, groundwater depth, etc.) that would capture the state of the ecosystem under various intervention scenarios. The research team then used simple heuristics to aggregate these indicators into VECs to clarify context and support future ecosystem response modeling efforts. For example, any indicator that addressed surface and groundwater condi tions were assigned to the Hydrology VEC ; any indicator that addressed species richness or diversity were assigned to the Biotic Integrity VEC. The VECs themselves were arrived at through both a literature review [predominantly based on the works of Davis et al. (2005); Fancy et al. (2009 ); Fischenich (2008); Gentile et al. (2001); Gucciardo et al. (2004); Harwell et al. (1999); Henderson and O'Neil (2004, 2007); Niemeijer and De Groot (2008) ] and discussions with the experts themselves. When the answers to these questions produced new data, when the sub teams generated their products (i.e., cover type mapping, field data results, well data, etc.), when knowledge of the systems operation evolved, when changes in program goals and objectives were refined, or when weather phenomena (i.e., floods and/or droughts) changed the state of the extant communities, lines of evidence were altered, removed, or added to reflect the revised problem space. For example, new imagery was acquired in 2008 capturing shifts in the r ivers position on the floodplain and changes in the location of emergent sandbar habitat as well as exposing shoreline (desirable settings for new cottonwood recruitment and establishment). As a result of this new information, the team decided to pursue l ines of evidence linking level of cottonwood recruitment to
50 the characterization of functional connectivity and shifts in population dynamics driven by surface/subsurface water and biogeography. The adaptive and recursive nature of the spiraling process w as meant to capture the experts evolving perceptions of the fundamental processes governing cottonwood ecosystem integrity, resiliency and sustainability on the river At the end of the study, t he final conceptual model, presented as a series of influence diagrams, w as given to the stakeholders for final proofing and adoption with the caveat that new information could be added to the model in the future if they so desired. 2.5 Results A tiered conceptual model was generated to illustrate the relationships betw een system wide drivers and stressors, highlighting the causal pathways between ecosystem responses and indicator of ecosystem integrity for the cottonwood communities found along the banks of the Missouri River ( Figure 2 4 ) 2.5.1 Driv ers Major anthropogenic drivers affecting the Missouri River cottonwood ecosystem include: 1) the regulation of water flows significantly altering the timing, frequency, duration and magnitude of flooding in the basin (i.e., Water Resources Management) 2) the o peration and maintenance of channels and reservoirs through bank stabilization and dredging activities (i.e., Sediment Management ), 3) urban growth (i.e., Human Demographics ) 4) land use conversions to accommodate urban expansion and agricultural operations ( i.e., Land Use and Management ), and 5) persistent changes in climate induced by humanrelated activities (i.e., increases in greenhouse gasses) resulting in changes in precipitation and temperature patterns as well as snowpack volume (i.e., Clima te ) ( Dixon et al., 2012; Johnson et al., 2012; Scott et al., 2012 )
51 Alternatively, forces above and beyond these humaninduced driver affect change in the basin including: 1) naturally occurring cyclical climatic changes in precipitation and temperature operating at the seasonal, annual, and decadal scales (i.e., Clima te), 2) the systems chemical, physical, geological, and biological character (i.e., Biogeochemistry 3) a natural disturbance regime generated by the meandering of the river (where it is remains unregulated), the pulsing of flood waters, and the presence/ absence of subsurface aquifers, 4) the fluvial geomorphology of the system including the channel forms and processes, and interactions among channel, floodplain, network, and catchment (i.e., Watershed Physiography ) and 5) the naturally shifting habitat m osaic in the basins geographic space over time (i.e., Biogeography ) ( Jacobson and Galat, 2008; Jacobson et al., 2011) 2.5.2 Stressors Ecosystem dysfunction was associated with a series of eight categories of disturbances that ranged from alterations to the systems physical and chemical processes to changes in biological communities that included: 1. Human encroachment between 1930 and 1955 flood control measures (i.e., dams and levees) offered protection to counties immediately adjacent to the river and its tributaries, encouraging rapid population growth ( http://www.census.gov/prod/www/decennial.html ) and human expansion into the region. 2. Land use conversions with flood protection, and increased economic incentives to farm the floodplain, the basin has experienced a significant increase in land use conversion from pasture and forest to agricultural production (an increase in relative coverage of agricultural croplands 26 percent to 52 percent between 1893 and 2006 within the study area) ( Dixon et al., 2010 ) 3. Hydrologic alterations the damming of the system has substantially altered the annual hydrograph below the reservoirs and has reduced the intraannual flow variability by alt ering the timing and decreasing the magnitude of spring pulses while increasing summer low flows ( Jacobson and Galat, 2008)
52 4. Sediment transport alterations m uch of the rivers sediment load has been deposited in the reservoirs behind the dams resulting in sediment imbalances marked channel incision below the dams and a cessation of flood deposi ted point bars formations downriver ( National Research Council, 2002 ) 5. Habitat alterations t he damming of the river s ubmerged large stretches of cottonwood forests lining the banks of the river upstream of the dams, while the reduction in sediment transport and the deposition behind these dams has reduced the amount of alluvium available for cottonwoods to colonize new floodplain surfaces below the dams ( Jacobson and Galat, 2006; Johnson et al., 2012) 6. Water quality alterations reductions in turbidity as well as discharge/runoff from the adjacent agricultural lands have reduced water quality system wide ( Jacobson and Galat, 2006 ) 7. Geomorphic alterations clearing and snagging, along with bank stabilization features have narrowed and focused the thalwag such that geomorphology of the system has become somewhat independent of flow, and as such, channel forming processes generating exposed alluvium have all but ceased ( Jacobson and Galat, 2006) 8. Shifts in populations dynamics increases in invasive understory species including Spanish brome ( Bromus inermis Leyss), reed canary grass ( Phalaris arundinacea L.), Canada thistle ( Cirsium arvense L. Scop. ) and Russian olive ( Elaeagnus angustifolia), and compositional shifts to latesuccessional tree species such as green ash ( Fraxinus pennsylvanica var. Lanceolata (Borkh.) Sarg.), box elder ( Acer negundo L.), American elm ( Ulmus Americana L.), eastern red cedar ( Juniperus virginiana ) and bur oak ( Quercus macrocarp Michx.) have been observed throughout the systems vegetative communities ( Dixon et al., 2012; Johnson et al., 2012; Scott et al., 2012 ) As these stressors indicate, the dri ving forces thus far have led to negative shifts in system dynamics (i.e., stressors), and as such, this conceptualization leans toward an impact assessment or negative characterization of the system. Again, it is important to note that the model is a neutral entity interventions targeting the recovery of ecosystem integrity at the level of the stressors would alter the characterization of these categories and elicit a positive response reversing these perturbations. Furthermore, the model was int entionally designed to emulate a network rather than a linear construct to capture system interdependencies and reinforcements.
53 As such, the stressors are not independent, nor is there a linear oneto one relationship between the drivers and the stressors themselves ( Figure 2 5 ). 2.5.3 Valued Ecosystem C omponents Six VEC categories were adopted to organize follow on quantitative model parameterization. These six were specifically selected to identify causal relationships between ecosyst em responses and stressors, but were also used to delimit the Missouri River Basin modeling domain. The six VECs included: 1. Hydrology focuses on the characterization of floodplain hydrology, hydraulics, landsurface elevations, soil permeability (or saturat ed hydraulic conductivity), and depths of groundwater. The quality and quantity of ground and surface water input into cottonwood stands is almost entirely determined by the condition of the surrounding landscape ( Jacobson et al., 2008; Jacobson et al., 2007 ) Therefore, the integrity of these communities is partly determined by processes operating in the surrounding landscape, especially in the local watershed. Different types of land use can alter surface runoff and vary the recharge of local aquifers. Land use also influences the introduction of excess nutrients, pollutants, or sediments. The cottonwood systems on the M issouri have been substantially impacted by the development of both groundwater and surface water for irrigation ( Johnson et al., 2012) 2. Soils target the availability and distribution of point bars characterized by coarse textured, well aerated mineral soils within the 2 to10 year highwater levels on floodplains or located where th e water table decline does not exceed the physical capacity of root growth ( Lewis et al., 2003 ) Note that cottonwood s eed dispersal coincides with late spring flows when water tables are high, fresh alluvium has been deposited, and competition has been minimized ( Johnson, 1992) 3. Structure is defined as the physiognomy or architecture of the community with respect to density, horizontal/vert ical stratification, and frequency distributions of species populations, and the sizes and life forms of the organisms that comprise the cottonwooddominated communities ( Franklin et al., 2002) but also alludes to the temporal or ageclass distributions of the cottonwooddominated riparian zones ( Dixon et al., 2012; Johnson et al., 2012 ) Alteration of natural hydrological processes (by dams, diversions, ditches, roads, etc.) and abiotic resource consumption through groundwater pumping have considerably altered the presettlement condition of the area ( Dixon et al., 2012 ) Riparian f orests respond to these hydrologic changes by shifting from cottonwood dominated systems to more m esophytic species typical of adjacent uplands and/or encroaching into the stream channel. When periodic flooding is eliminated by water management, disturbances that foster recruitment and regeneration are eliminated, and these ecosystems
54 dynamically shift to communities dominated by disturbance intolerant, l ate successional species 4. Biotic I ntegrity focuses on the capability of a n ecosystem to support and maintain a balanced, integrated, adaptive community of organisms having a species composition, diversi ty, and functional organization comparable to that of the natural habitat of the region. Ecosystem response to disturbance (natural or otherwise) depends greatly on biotic integrity. Response is tied directly to the functional characteristics of organisms present in the ecosystem and the distribution and abundance of those organisms over space and time ( Johnson, 1992; Scott et al., 2005) As such, biodiversity (a m easure of biotic integrity that captures the variety and variability among living organisms and the ecological complexes in which they occur) encompasses much more than just species richness (the number of species) or species diversity (a measure of both t he number of species and their relative abundances). Rather, biodiversity considers not only the number of species present, but also the inherent complexity of the system. It operates in a variety of contexts, integrating concepts of dominant species, keys tone species, ecological engineers, and interactions among species (e.g., shading of cottonwood seedlings by woody species). 5. Spatial I ntegrity directs attention to both the cottonwood communities place in the landscape, as well as identifying key process es that shape the ecosystem. The structural arrangement or patterns of vegetative patches and corridors in combination with the matrix in which they exist, determines the flow of energy and materials through an ecosystem, capturing dynamic changes to landscape function and form over time ( Forman, 1995) The structures and patterns evidenced today were produced by the flows of yesterday (refer to references in Formann, 1995 and Turner et al. 2001 ). Furthermore, a linkage or feedback mechanism exists between structure and functionstructure regulates fl ows and movements. Over time, these processes alter the mosaic, much like turning a kaleidoscope to produce different patterns ( Forman, 1995, page 135). 6. Disturbance deals with humaninduced disturbances occurring in areas adjacent to the cottonwood ecosy stem on the river. Land use and urban/agricultural development activities in the Missouri River basin have been encroaching on the cottonwood riparian ecosystem at the watershed scale for the past 50 years ( Dixon et al., 2012 ; Johnson et al., 2012; National Research Council, 2002; Scott et al., 2012) These ecosystems do not exist in a steady state; they are dynamic and possess characteristic compositions structure s and functions that have adapted to natural disturbances over long periods of time. At the landscape level, natural disturbances destroy patches of cottonwood forests and re initiate plant succession. Human activities (both onsite and offsite) that deviate from these patterns affect individual species (and through biotic interactions many other species and ecological processes) by direct exploitation, habitat elimination, and modification of ecological processes Agricultural practices greatly affect hydrologic patterns and t he clearing of forests generally decreases interception of rainfall by natural plant cover and reduces soil infiltration resulting in increased overland flow, channel incision,
55 floodplain isolation, and head ward erosion of stream channels Human activities, such as land clearing and erosion, can cause the loss of nutrient s (e.g., phosphorus), disrupt natural cycling of nutrients, and limit the systems productivity At the same time, agriculture and industry can discharge excessive amounts of nutrients (e.g., nitrogen) into the ecosystem and drastically change their trophi c structure, and degrade w ater quality. A oneto one relationship was used to associate ecosystem responses and VECs so that follow on ecosystem response model parameterization could be simplified to some extent ( Figure 2 6 ). The VECs, when considered as a whole, fit together like a puzzle to capture the essence of the ecosystems character. 2.5.4 Ecological I ndicators T he VECs aggregated the ecosystem responses in a meaningful manner to better capture the functionality of the community in the face of change based on expert opinion, scientific literature, and statistical evaluation ( Figure 2 7 ) The selected indicators span the gamut of spatial and temporal scales capturing ecosystem structure, composition and f unction measured at varying degrees of granularity. Indicators such as adjacent land use, habitat interspersion, Land Capability Potential Indices (LCPIs) ( Jacobson et al., 2008; Jacobson et al., 2007) and levels of cottonwood domination and recruitment operate at the system scale and return measures of system wide ecosystem integrity. Groundwater depths, patch sizes and distance between patches are unique to each segment, and therein offer measures of segment level functionality. At the stand or site level, specific biodiversity indicators (i.e., herbaceous and shrub canopy cover, endemism, native species richness, and w etland indicator scores) serve to capture smaller scaled ecosystem responses and conditions.
56 2.6 Discussion and Conclusions No conceptual model is perfect nor is it ever finished. I n its infancy it is a n abstraction of reality that requires critical thinking sensible construction, and rational thought in the face of multi dimensional choices characterized by uncertain science, conflicting objectives, and difficult tradeoffs. In its maturity, a conceptual model remains a construct designed to adapt as new knowledge and shifting paradigms alter EBM goals and objectives. In its dotage, a conceptual model can stand as a useful snapshot of system understanding that can evolve towards new purposes and directions. The purpose of this study was to present a spiral based conceptual modeling approach that uncover ed critical lines of evidence to inform and support EBM. The approach was founded on the exploration of values ( What is important? ) and model behavior ( What are the consequences? ) linked with current scientific understanding of the fundamental ecological processes governing ecosystem response. The approach was designed to bring together a wide range of readily available data, to assemble these into essential ecosystem components based on logic that is scientific ally informed, and to synthesize the assessment of system conditions across space and time into meaningful indicators The Missouri River application resulted in an inclusive conceptual model grounded in the fundamentals supported by the latest science, a nd attuned to the priorities of the studys stakeholders Iterative and recursive reflection supported group learning and increased both the teams understanding of the ecosystem as well as the manner in which the model will be utilized to support the USAC E recovery efforts. Through the application of the spiral development process, both anthropogenic and natural drivers (i.e., land use management, climate, etc.) were linked to stressors (i.e.,
57 human encroachment, hydrologic alterations, etc.) that were then characterized by observable ecosystem effects (i.e., loss of species richness, habitat fragmentation, reductions in age class distributions, etc.). More importantly, these effects were tied to VECs (i.e., Biotic Integrity, Structure, Disturbance, etc.) by uncovering lines of evidence linking responses to measureable ecosystem endpoints (i.e., adjacent land use, cottonwood recruitment, habitat interspersion, etc.) that can now be derived to measure the performance of proposed recovery efforts in the future. All told, nine drivers, eight stressors, six VECs, 17 effects, and 13 multi scalar metrics were connected through clear lines of evidence to visualize ecosystem integrity of the cottonwood ecosystem on the Missouri River. More importantly, the USACE, its partners, and its stakeholders can not only use these metrics to compare and contrast potential interventions on the river, they can use these lines of evidence to develop monitoring plans and establish thresholds that will trigger adaptive management res ponses well into the future. The collaborative and transparent creation of meaningful conceptual models to inform ecosystem recovery efforts led by natural resource agencies offers some unique challenges to the field. Within the Missouri River example, budgetary constraints limited the extent to which the expert team could be assembled to develop and review proposed conceptual model components, lines of evidence, and ecosystem response metrics. At times, decisions by the development team to add or remove potential pathways required three to six months for review, alteration, and acceptance by the wider expert group. Increasing the frequency and number of workshops could have improved efficiencies inside the spiral framework to reduce these issues. Another pr ocess challenge emerged from the institutional budgeting process concerning
58 ecosystem monitoring activities. Pressure from participating institutions to quickly identify and ratify performance metrics to clarify budget allocations for field monitoring meant that operative parts of the conceptual model were still being formulated at the same time that the ecosystem monitoring was being developed and deployed. This resulted in additional costs and delays as the field team collected data on performance metrics that were later left unused when the causal pathways were eliminated and the models were finetuned. In the future, it would be more productive to develop the conceptual model in advance of any ecological sampling or modeling to generate a more direct an d informed route to effective EBM. Finally, the large number of stakeholders involved in the study made the spiral process unwieldy at times. In the future, splitting the team into two or three working groups (either on a random basis or through a more sys tematic process focused on their disciplines) could improve efficiency and untangle the lines of evidence more quickly. Undoubtedly, the cottonwood community conceptual model presented here is selective in its aspects, and by no means captures the entire extent and complexity of the ecosystem as it exists today, but the stakeholders have labeled it "good enough" to proceed with recovery design formulation often the best result one can expect when attempting to untangle these critical lines of evidence. Eff ective conceptual models offer an organized, inclusive, and practical strategy to characterize complex ecosystem attributes and communicate these relationships to a wide audience. Their intent must be to enable participants to clarify their predicament, co nverge on potentially actionable responses and agree to paths forward that will resolve the issues at hand. Conceptual models need to strike an appropriate balance
59 between over simplification and over sophistication, providing enough structure that decisi on makers can take responsibility for the consequences of their actions in a rational, transparent, and scientifically defensible manner with sufficient confidence to make the necessary decisions.
60 Figure 2 1 Step 1 in the process call for the development of a conceptual model to uncover critical lines of evidence linking ecological endpoints to system drivers and stressors. Step 1: Conceptual Modeling Laboratory and Field Experiments Description Data from Literature and ExpertsModel Goals and Objectives Performance Measures Model Performance Step 3: Calibration Ecosystem Response Models Step 2: Mathematical Formalization Step 4: Forecasting Step 5: Alternative Evaluation Quality of The Fit Model Verification Sampling Design Reference Datasets Evaluation Datasets Fitted ValuesResponse Thresholds Adaptive Co Management Step 6: Construction and Monitoring Statistical Literature, Existing Models, Expert Contributions Predicted ValuesModel Validation Study Goals and Objectives Performance Hypotheses Site Selection via GIS Based Decision Support System
61 Table 2 1 A typology aggregating grouping conceptual models based on st ructure, composition and function. Note that presentation of these models is oftentimes a combination of narration, influence diagrams (box arrow flowcharts), and tabular matrices Model t ype Structure Composition Function Relevant e xamples Causal Chains a nd Networks Circular Diagrams 1) Pressure State Response (PSR) 2) Driving force State Response (DSR) 3) Driving force Pressure State Impact Response (DPSIR) 4) Enhanced DPSIR (eDPSIR) Capture key relationships between factors in society and the environment Gregory et al. (2013) ; Niemeijer and De Groot (2008) State a ndTransition Circular Diagrams 1) State variables (vegetative condition) and transitional pathways Present possible changes in state variables and depict successional pathways Johnson (1992 ) ; Rumpff et al. (2011) Effects Driven Hierarchical Diagrams 1) Dri ver Stressor Essential Ecosystem Characteristic Endpoint Measure 2) Driver Stressor Essential Ecosystem Characteristics Endpoints Retrospectively analyze effects to determine the causes for the current conditions and use scenario analysis to evaluate future options Harwell et al. (1999) ; Gentile et al. (2001) ; Henderson and O'Neil (2004 ) ; Henderson and O'Neil (2007 ) Indicator Production Hierarchical and/or Circular 1) Dr iver Stressor Ecosystem Effect Attributes (Vital Signs) Measures (Ecological Indicators) 2) ProcessFunctionService Benefits Value tied to Impact ResponseDriver Pressure State Identify relevant ecological indicators that can be used to depict or evaluate ec osystem conditions Gucciardo et al. (2004) ; Scott et al. (2005 ) ; Fancy et al. (2009) ; Kandziora et al. (2013)
62 Figure 2 2 Study area for the Missouri River study. Priority segments (highlighted in red) and reference segments (highlighted in orange) served as the express focus of the CMP ( USACE, 2010) and were mandated under the conditions of the USFWS Biological Opinion (BiOp) of 2000, amended 2002 ( USFWS 2000; 2003)
63 Figure 2 3 The technical approach deployed to develop the conceptual model for the Missouri River cottonwood ecosystem was a spiraling process that recursively addressed model prototype production as each line of evidence was uncovered [adapted from Du Toit (2005 ) ].
64 Figure 2 4 An ov erview of the conceptual model developed for the Missouri Rivers cottonwood modeling effort both natural and anthropogenic drivers cause stress on the system that affect valued ecosystem components (i.e., hydrology, soils, structure, etc.) whose alterations lead to conditional changes in the ecosystem. Endpoints, in this instance, indicators of cottonwood ecosystem integrity can be measured at the system, segment and stand scales spanning intergenerational timeframes.
65 Figure 2 5 A closer look at the Driver Stressor relationships in the cottonwoods conceptual model. The drivers are those process, structures and regimes that control or cause (i.e., force) changes in the ecosystem (i.e., stressors). As noted in the table, the forces generated by these drivers are not homogenous. For example, biochemical changes cause habitat alteration, water quality degradation, and shifts in population dynamics, but do not have notable affects on human encroachment.
66 Figure 2 6 Effects can be either directly measured or can be indicated by proxy variables (via endpoints). Here, Valued Ecosystem Components (VECs) have been organized into a structure commensurate with an index based community modeling paradigm usi ng an aggregation strategy (i.e., through a Hydrography, Biota, and Spatial component modeling architecture).
67 Figure 2 7 In essence, the cottonwood ecosystem responses have been conceptualized in terms of ecologically signi ficant and important ecological indicators that can be measured or quantified to characterize the structure, composition and function of the ecosystem thereby capturing ecosystem integrity in a meaningful manner. Here the individual variables are cross wal ked with the critical effects highlighted in the conceptual modeling exercise. These effects have been sorted and colorized to better identify their association with the VECs (i.e ., Hydrology, Soils, Structure etc.).
68 CHAPTER 3 PRESCRIPTIVELY FORMALIZING THE USGS LAND CAPABILITY POTENTIAL INDEX (LCPI): A CASE STUDY ON THE MISSOURI RIVER 3.1 Introduction Wicked problems ( Rittel and Webber, 1973) like the U.S. Army Corps of Engineers (USACE) attempt to recov er cottonwood ecosystems ( Populus detltoides W. Bartram Marsh. S ubsp. m onilifera (Ait on) Eckenwalder ) on the Missouri River ( USACE, 2010) are confounded by issues of complexity, uncertainty, and disparate, oftentimes conflicting agendas, mindsets and value laden decisions Ecosystem Based Management (EBM) ( McLeod and Leslie, 2009) in the face of these challenges often becomes a matter of experimentation over multiple years, where resolution is sometimes attained only through political compromise. In these highly dynamic, value driven conflicts, carefully prepared and highly detailed technical analyses (i.e., hard science) cannot charact erize all the ecosystem responses nor remove all the uncertainties ( McIntosh et al., 2007 ) and decision makers must resort to fillingin the data gaps with softer information (i.e., expert opinion and/or professional judgment) in a post normative ( Funtowicz and Ravetz, 1992) and intuitive fashion ( Albar and Jetter, 2009) An integrated ecosystem based approach is needed to fuse the qualitative elements of science (i.e., professional opinions and expertise) with more quantifiable analyses to fill these gaps. The key to success in these situations will be quality control providing a deliberative environment where experts can offer experience based opinions in a collaborative, yet disciplined manner ( Pahl Wostl, 2007)
69 3.2 Goals a nd Objectives In Chapter 1 I discussed my solution to this quandary an integrated ecosystem based approach designed to fuse the softer side of science (i.e., qualitative professional opinions and expertise) with harder more quantifiable analyses Chapt er 2, I presented my spiral based conceptual modeling strategy and used a case study on the Missouri River to demonstrate the approach on the recovery of the rivers threatened cottonwood ecosystem. One outcome of this conceptual modeling effort was a dis crete list of measureable lines of evidence (i.e., ecosystem response indicators) that c an be used in an EBM plan to assess successful intervention (i.e., alter ations of ecosystem structures and processes aligned with shifting social and policy drivers) under an adaptive co management paradigm ( Cundill and Fabricius, 2009; Linkov et al., 2009) Weaving these individual lines of evidence into a meaningful characterization of ecosystem integrity requires creative problem solving and critical thinking at both the strategic ( future oriented ) and tactical ( near term) scales The first step is to mathematically formalize or transform the conceptual models ecosystem response indicators (i.e., patch size, shrub canopy cover, cottonwood domination, groundwater depths, etc.) into numerically based indices normalized on a standard scale (0 1) ( Figure 3 1 ). The key is to use a combination of both qualitative and quantitative approaches in a scientifically defensible, transparent manner to ensure confidence in the model outcomes. In these instances, expert elicitation techniques can be used to transparently bridge the gaps where current data is limited. The objective of this chapter is to present a prescriptive approach that can be used to formalize key lines of evidence generated by conceptual models of dynamic ecosystems using expert elicitation of profess ional judgment in situations where
70 technical data is limiting or absent. An example on the Missouri River demonstrates the value of this prescriptive approach, producing a casespecific quantitative indicator of ecosystem response for the rivers cottonwood ecosystems. The intent is to demonstrate a meaningful interpretation of hydrologic / soil characteristics and an indication of site suitability (i.e., niche provisioning) for re establishing and/ or preserving cottonwood ecosystem integrity (sustainable and resilient functions, structures, and processes) on the Missouri River As I mentioned in the previous chapters, t he degree to which a studys stakeholders (defined as a group of entities with a vested interest in the studys outcomes that includes subjec t matter experts) can be engaged in the development of performance metrics often varies depending on the scope and magnitude of the effort. I also referred to Gregory et al. (2013 ) who suggest ed that a variety of participatory options exist including: 1) the development of metrics by facil itators that are then subsequently presented to the stakeholders for follow on modification and adaptation; 2) the development of m etrics by a core group that are then presented to full the full stakeholder audience to stimulate discussion and arrive at a consensus; and 3) the participatory development of metrics by the entire stakeholder community. In this exercise I opted to use the second approach, devising a hydrologic performance indicator for cottonwood forest integrity using the input from a small c ore group of subject matter experts My goal was to combine what the local scientific experts thought should be included in an ecosystem integrity performance metric with the realities of how endusers would actually use this metric to mak e decisions, wi th the intention of creating an indicator that was defensible, transparent, efficient and
71 operational (i.e., could be readily implemented within the constraints of the USACE planning and decision making process) ( USACE, 2000 ; 2003 ). This chapter has been divided into four primary sections. The first section describes the nature of the problem (i.e., the concern that data gaps prevent the direct formalization of hydrologic indicators in highly degraded watersheds) and discusses a solution the use of expert knowledge to fill these gaps. A brief literature review of key concepts (i.e., formalization, expert knowledge, and the recent trend towards prescriptive decision theory) is presented followed by a description of strategies and techniques commonly used to elicit these expert opinions. In addition, the section describes potential pitfalls and traps (cognitive and motiv ational biases) that should be acknowledged and neutralized in order to extract this knowledge in an objective manner. In the methods section a case study on the Missouri River is used to demonstrate the use of one such elicitation method (an independent r anking exercise) to extract judgments from a core stakeholder group. The results of the metric formalization exercise are then presented and compared to both a retrospective analysis of predam (1890s) conditions on the river, as well as newly acquired im agery showing ecosystem response to the 2011 recordsetting floods on the Missouri. The final section concludes with a short discussion of the lessons learned, next steps, and future application opportunities using this newly formalized performance metric. 3.3 Review of Foundational Concepts 3.3.1 Prescriptive Formalization in the Presence of Data Gaps Prescriptive formalization, by definition, is the process of strategically converting a complex entity into a simpler, more straightforward representation (i.e., categ ories or numbers) whose properties are more easily understood and communicated to decision
72 makers and utilized in M ulti C riteria D ecision A nalysis (MCDA) and modeling ( Abbott, 1997; Riabecke et al., 2012; te Brmmelstroet, 2010) Typically, this conversion is based on a spatially and temporally specific s tandard of reference. Reference conditions, and more specifically reference sites, function as physical representations of the ecosystem s range of character whose attributes are both observable and measureable ( Society for Ecological Restoration Internationa l, 2004 ) These standards of reference make it possible to establish a likely range of variability for particular measures of ecosystem integrity, facilitating the development of relational indices for ecosystem response models The sites themselves can serve as templates for rehabilitation designs and specifications, as well as offer benchmarks or performance targets to measure the progress of recovery efforts and stimulate adaptive management responses ( Miller et al., 2012) Unfortunately, in highly degraded settings, locating any, much less enough, reference sites with undisturbed hydrologic regimes that can be used to statistically as sure a meaningful formalization of hydrologic indicators is difficult. The challenge is to fill this void with transdisciplinary and collaborative rationality ( Innes and Booher, 2010 ) using faceto face participatory strategies to synthesize implicit and explicit knowledge and expertise in an effort to generate informed inferences in the absence of actual reference data. In other words, ask ing experts to fill these gaps by assembling, summarizing, organizing, interpreting and reconciling their experiences using heuristics in a rational, transparent, and collaborative manner to produce a relevant, sufficient, and defensible solution.
73 3.3.2 Expert K nowledge Irrational and Indispensible In years past, planners and managers have been known to reluctantly and somewhat apologetically resort to professional judgment when either gaps in their understanding of complex ecosystem functions or short timelines and constrained budgets pushed them in that direction. The literature surrounding the elicitation and use of professional judgment in these instances is particularly large and expansive, distributed across the disciplines of psychology, ecology, medicine, engineering, statistics, computer science, and decision analytics This review limits its focus specifically to applications from the ecological modeling and ecosystem based planning and management arenas. Numerous studies now openly discuss the ubiquitous and unavoidable use of expert opinion in these fields ( Ferretti and Pomarico, 2013; Keeley et al., 2012 ; Kloprogge et al., 2011; Leskinen et al., 2003; Oliver, 2002; Price et al., 2012; Reza et al., 2013; Rger et al., 2005 ; Yamada et al., 2003) Moreover, both Innes and Booher (2010) and Gregory et al. (2012 ) review the increas ing use of participatory processes to construct models, while Kreuger et al. ( 2012) ; Kareiva et al. (2011) and Tress et al. (2005) discuss the mar ked increase in range of expertise now being employed in these types of studies. Riabecke et al. (2012 ) and ( Albar and Jetter, 2009 ) strongly advocate prescriptive approaches that aim to obtain data in a structured and systematic way, emphasizing human participation and awareness of descriptiv e realities while emphasizing cost efficiencies and pragmatism. 3.3.3 Elicitation Traps and Pitfalls The collective works of Meyer and Booker (2001) Gregory et al. (2012 ) Innes and Booher (2010) and Malczewski (1999) offer indepth guidance to eliciting opinions/judgments and debiasing results. The challenge is to acknowle dge and
74 embrace the uncertainty surrounding the use of expert opinions in a manner that strengthens rather than weakens the process ( Sear et al., 2008 ) A well structured expert elicitation strategy using a transparent, repeatable and systematic methodology followed up with careful documentation of assumption, procedures, and results can be quality controlled by peer reviewers, legitimizing the outcomes of any fast and frugal elicitation. A formal elicitation is designed like an experimental physicist plans and implements an experiment controlling the environment and determ ining the initial conditions. Jane M. Booker, October 2011 Of course, a defensible estimation of ecosystem response based on expert judgment depends on many things including: 1) the nature of the problem being addressed, 2) the quality and/or reliability of th e experts, 3) the context in which the judgments will be utilized, and most importantly, the 4) elicitation methodology ( Gregory et al., 2012; Kreuger et al., 2012; Meyer and Booker, 2001 ) According to Meyer and Booker (2001) three basic strategies exist to elicit these judgments and opinions: 1) individual interviews ; 2) interactive group workshops; or 3) isolated extractions. In the first instance, each expert is interviewed individually by the modeler, usually in a faceto face setting, where the interview can be conducted to produce the formalized metric. In the second case, the modeler conducts a works hop inviting a group of experts to participate and come to consensus on the formalization of the metric. The third option calls for the isolation of the experts their expertise is extracted either through a survey or a voting instrument (specifically a ranking or assignment of value to the metrics qualitative categories).
75 As with any analytical (hard data) assessment, there are advantages and disadvantages to each strategy. As the table indicates, biases (the tendenc ies or preferences made towards particu lar perspectives, ideologies or result s) can interfere with the experts abilit ies to be impartial, unprejudiced, or objective under any and all of these methods. These experiences become worrisome when the experts opinions are not voiced accurately, when their answers do not follow logical rules, or when their responses are either misinterpreted or misrepresented. Several types of motivational and cognitive biases are of particular concern in the formalization of performance metrics in this study including those categorized as Wishful Thinking, Impression Management Group Think ing, Anchoring, Inconsistency and Availability biases ( Kahneman et al., 1982; Koehler and Harvey, 2007; Meyer and Booker, 2001) Practitioners intent on using elicitation strategies to extract expert opinion to formalize performance metrics should prescriptively exploit the experts decision making heuristics, guarding against the biases that could potentially skew the data and impact the credibility of the modeling outcomes ( Meyer and Booker, 2001) 3.3.4 Ranking and Aggregation Using transdisciplinary teams to interpret categorical data inevitably generates multiple answers to singular questions of recovery priorities and/or plan performances. Unless a mutual consensus is envisaged and achieved through group exercises, ranking and aggregation must be deployed to collapse the ordinal (categorical) answers into a single coherent and therefore useful indicator of response in these situations. Many methods exist for ranking and aggregating criteria, and their various attributes and procedures can have an impact on their applicability in practice. Riabecke et al. (2012 ) offer an extensive review of the most common MCDA methods employed in prescriptive
76 set tings, and discuss these from the perspective of reasonably applicable weight elicitation [s] (page 2). They note that the elicitation of weights can be a cognitively demanding task, subject to biases, and heavily dependent on the method of valuation. They offer three overarching categories of MCDA approaches including: 1) rank order, 2) semantic categories, and 3) intervals established using upper and lower bounds, and indicate that the former approach is appropriate when handling ordinal data. Rank order ing is a process where modelers : 1) ask experts to rank the different ordinal criteria, 2) translate these responses into proxy cardinal weights consistent with the supplied rankings using either rank sum weightings, reciprocal weightings, and centroid wei ghtings strategies, and 3) then normalize the results generating ratios based on the principle of maximizing the expected values ( Malczewski, 1999 ; Riabecke et al., 2012) Empirical studies conducted by Stillwell et al. (1981) demonstrated that rank sum weighting performs well in most situations, and Riabecke et al. (2012) claim the approach is less cognitively demanding and more advantageous when engaging large, disparate groups (e.g., transdisciplinary teams). 3.4 Methods A prescriptive elicitation strategy was devised to formalize lines of evi dence arising from conceptual modeling activities where data gaps exist and expert opinions serve as the best data available. The following sections describe the problem context and detail the methods used to devise a casespecific example garnered from results of Chapter 2 3.4.1 Study Area and Problem Space The study area corresponds to one of the six priority segments on the Missouri River designated in the BiOp ( USFWS 2000; 2003), and referred to as Segment 10,
77 which begins at River Mile (RM) 811.1 (Gavins Point Dam) and ends at RM 753 (Ponca State Park) in northeast Nebrask a, USA (855 km2) (Figure 3 2 ). As the figure shows, Segment 10 is the downstream section of the Missouri National Recreational River (MNRR) referred to commonly as the 59 mile Segment which is administered by the National Park S ervice ( USACE, 2010) Segment 10 runs predominantly west to east, from Yankton, South Dakota t o the bottom of Ponca State Park in northeast Nebraska, and has a wide valley (over 10 miles or 16 km in places), bounded by high bluffs to the north and south in Nebraska and South Dakota. Segment 10 is often considered one of the more natural or least a ltered sections of the lower Missouri b ecause it occurs below the farthest downstream dam (Gavins Point) o n the regulated system and is an unchannelized reach with some physical features characterist ic of the preregulation river ( Dixon et al., 2010) In contrast to inter reservoir segments on the Missouri, there are no reservoir s downstream from Gavins Point Dam and therefore no slackwater effects on the segment are evidenced. On the other hand, s ignificant channel degradation (measured via changes in water surface elevation) has been recorded since 1962 along the entire length of the segment along with declines in bed water surface elevations (at flows of 30,000 cfs) exceeding two meters over the entire length of the segment ( Dixon et al., 2010) 3.4.2 Land Capability Potential Index (LC PI) for the Case Study Area The LCPI is a categorical index developed by the U S G eological Survey (USGS) ( Jacobson et al., 2008; Jacobson et al., 2007) for the Lower Missouri River floodplain that integrates wetness and retention classes to systematically characterize land capability potenti al along the riverine corridor. Constructed through the systematic intersection of readily available hydrology, landsurface elevations, hydraulics, and soil
78 drainage class datasets in a Geographic Information System (GIS) environment, the resultant LCPI classification system can serve as a relatively coarsescale proxy to guide land management decisions on a systems scale in its current qualitative configuration ( Jacobson et al., 2008; Jacobson et al. 2007) Within the LCPI, wetness classes are created by intersecting high resolution land surface datasets with water surface elevations derived by estimates of discharges under nine flood frequency intervals (i.e., the 1 2 5 10, 25 50 100 250, and 500year flood recurrences assuming current river operations) using a one dimensional ( 1 D ) unsteady hydraulic model ( USACE, 2004). The water retention classes in the LCPI are based on Natural Resources Conservation Service (NRCS) Soil Survey Geographic Database (SSURGO) soil drainage classes that conceptually integrat e saturated hydraulic conductivity of th e soil s with underlying geologic materials, and to some extent contain information related to surface topography The final LCPI characterization is generated by overlaying the retention and wetness classes on the landscape to reveal all possible unique combinations of categories in a gradient ranging from strongly retentive/frequently flood areas to nonretentive/rarely flooded areas. The wetness classes in the LCPI characterize flood ing frequency and indirectly capture the degree of flood pulsing in the s ystem, while the retention classes in the LCPI indirectly describe channel complexity (via roughness, morphology, and slope variation), sediment transport/deposition and surface elevations. As the ecological response of the cottonwood community is tied dir ectly to the hydrologic pulse and the resultant diversification of the channel the LCPI can be considered a measure of both
79 landscapelevel connectivity as well as stream migration evidenced by spatial complexity. Between 2007 2009 Jacobson et al. (2008) developed the LCPI for Segment 10 on the Missouri River. Essentially, a GIS based derived intersection of water surface altitudes, landsurface altitudes, and soil character, LCPI provided a multi criteria characterization of the valley bottom habitat. Four classes of flooding regimes were combined with four classes of soil types, to generate 16 classes that wer e then mapped on the river ( Figure 3 3 ) Of course the mapped LCPI outputs still required a formalization or translation to make them meaningful in terms of cottonwood forest functional capacity. In essence, an injection of soft h ard science was neededwhile the LCPI offered a qualitative characterization of the setting, expert elicitation and professional judgment was used to tie the LCPI values to a measure of site potential or suitability hydrologically speaking for the cottonwo od forest community 3.4.3 Elicitation and Formalization The goal of the exercise was to use expert knowledge, combined with experiences and judgments, to generate an estimate of ecosystem response and/or consequence that could be used to inform recovery efforts The objective was to characterize, to the degree possible, each experts beliefs about the relationship between niche provisioning (which equates to recovery potential and habitat suitability) and LCPI classifications of land cover along the Missouri Riv er in the Segment 10 study area. The key was to utilize participatory strategies and expert elicitation techniques to assure the legitimacy, clarity, sincerity, and accuracy of each participants opinions to build consensus and reduce bias ensuring a degree of quality control. Essentially, the elicitation of expert knowledge and the formalization of the qualitative LCPI scores using
80 this knowledge could be undertaken in three basic steps: 1) preparation which included the selection of experts, framing of the question, and selection of an elicitation m ethod; 2) e licitation and processing of the results; and 3) conversion of the LCPI scores to a formalized scale with mapped values 18.104.22.168 Preparation In 2006, a transdisciplinary ( Fry et al., 2007; Tress et al., 2005 ) team of subject matter experts was convened to make explici t the published and unpublished cottonwood ecosystem details regarding life history characteristics and spatial distributions on the river based on their own personal experiences and wisdom. The definition of transdisciplinary expertise purposely extended well beyond the conventional domains of academia ( Kreuger et al., 2012 ) to include participants having specialist knowledge acquired through education, trai ning, and/or experience. The intent was to actively engage both regional academic researchers from different and unrelated disciplines as well as nonacademics including natural resource managers and representatives from organizations, governmental entities, and experienced members of the public who operate in a professional capacity (i.e., contractors) Note that this extended peer tactic deliberately imposed a level playingfield on the process applied experiences garnered from nonacademics were consid ered as useful and meaningful as theoretical knowledge derived from regionally or even world renown academics. Descriptions of the ten participants associations are presented in Table 3 1 Their expertise in cottonwood ecosystem s ranged from 4 to 42 years ( = 21 years). Four of the ten participants were world renown scientists with more than 200 peer reviewed papers attributed to their names. Although three of the participants were affiliated with the USACE, these participant s were either associated with different USACE Districts, or
81 as was the case with the two participants from the USACE Omaha District, the participants came from differing business lines (namely Planning and Regulatory). It should also be noted that Dr. Jaco bson, the author of the LCPI methodology, abstained from participation in the elicitation exercise so as to remove any appearances of bias with regards to anchoring and/or wishful thinking. Prior to polling, each expert received a copy of the USGS report detailing the specifics of the LCPI protocol ( Jacobson et al., 2007) and Dr. Jacobson briefed the team on the procedure to assure they understood and could accurately interpret the 16 L CPI classes, thereby reduc ing the linguistic uncertaint ies inherent to the use of these types of categorical criteria ( Burgman, 2005; Burgman et al., 2005; Elith et al., 2002) and reducing Availability bias. Furthermore, the elicitation was conducted in a blind manner although the USGS scientists described the LCPI methodology to the team, they were not allowed to present the findings from their analysis. This debiasing action removed any possibility of the experts developing an anchoring bias of expected outcomes tied to cover type distributions or recovery site posit ions on the landscape (e.g., the arbitrary and capricious assignment of high scores to preferred recovery sites, i.e., the Impression Management bias). Next, a preliminary polling question was presented to the experts: Given the 16 LCPI categories which have been derived based on a combination of wetland and retention classes, in your opinion, how strongly are cottonwood communities associated with each LCPI category? Inevitably, this oversimplification of the problem statement was immediately discredited by the experts. After some discussion, the experts came to the consensus that the formalization of the LCPI could not be conducted at the community level, but rather at
82 the assemblage level based on stand age. In other words, in the absence of human dis turbance (specifically agricultural activities), the experts agreed that the life history characteristics of the cottonwood species and their phreatophytic nature would dictate their spatial distribution on the river. As such, the team decided that the for malization must be made on an age class by age class basis (i.e., decomposition). Five age classes were defined by the experts to channel the effort: Stands >114 years old (Old Growth), Stands between 50114 years old (Mature), Stands between 2550 years o ld (Intermediate), Stands between 1025 years old (Poles), and Stands between 210 years old (Saplings). With these distinctions in hand, the study question was amended and the experts were then asked: Given the 16 LCPI categories (which have been derived based on a combination of wetland and retention classes), use your years of experience and onthe ground knowledge of the system to rank the individual LCPI categories based on their ability to provide optimal niches for each of the cottonwood age classes. 22.214.171.124 Elicitation and processing Once the question was finalized, an elicitation strategy was debated. Based on the strengths and weaknesses of the strategies described in Meyer and Booker (2001 ) and recognizing that there were limited funds and timing constraints, an isolated polling approach was s elected. A voting instrument (essentially a series of spreadsheets) was developed that was emailed out to the participants and populated at their convenience. On an age class by age class basis, the experts were asked to rank the qualitative LCPI categor ies on a scale of 1 to 16 in terms niche provisioning for the cottonwood age
83 class (where 16 represented the optimal niche and 1 represented the worst fit) in the spreadsheet The experts were discouraged from halving or double assigning ranks in oth er words, unique scores were to be assigned to each of the 16 LCPI categories within each age class. When completed, the experts returned their file back to the research team for compilation. Confidentiality was preserved throughout the process, and follow on phone interviews between the research team and the individual participants were conducted to assure quality control and debias the results when inconsistencies attributed to both the Anchoring and Inconsistency biases were observed. The isolated nature of the exercise was designed to minimize the Group Think and Wishful Thinking bias es which can be evidenced in the summary of the voting results ( Figure 3 4 ) T he shifting trends in the participants responses indicat ed a degree of agreement, but not full consensus. Demographics have been masked in this table (to assure confidentiality) but it is interesting to note that experts A D and H are renown researchers (academics) in the voting pool, and even they had varying judgment s as to the relationship between site potential and LCPI classes (columns lined in blue). The responses from the experts were combined using weighted rank sum s ( Malczewski, 1999 ) for each age class : i= j+ 1 k+ 1 ( 3 1 ) where wi is the normalized weight for the jth criterion, n is the number of criteria under consideration (i.e., 16 LCPI categories), ( k = 1,2,,n ), and rj is the rank position of the sum of all weights, t hat is k+ 1 ( Figure 3 5 ). Based on the trends in these intermediate outcomes (note the increasing green squares or higher rankings moving upwards from left to right in the table), the experts
84 decided that the age clas ses should be aggregated into two overarching categories: Older communities ( greater than 25 years old) and Younger communities (between 225 years old). The weighted LCPI response scores were combined using summing to form the two categories A Pearsons ( ) combination with univariate statistics to evaluate the results using MS Excel. For purposes of this study, it was assumed that correlations ( or < 0.2 0 indicated weak relationships, whil e correlation values between 0.20 and 0.50 indicated moderate relationships, and correlations > 0.5 indicated strong relationships. Statistical significance was defined at p = < 0.05 for all tests. Comparisons were made between the Academics subgroup and the Managers subgroup. Spatial analyses were conducted in ArcMap 10.1 using ArcToolbox features, ET Geowizards, and XTools. All GIS derived data was exported to MS Excel for statistical analyses. 126.96.36.199 Conversion The sum med results were transformed to a 0 to 1 scale by dividing by the total sums, and normalized using a percent of maximum approach to generate formalized index values for the sixteen LCPI categories No additional weighting scheme was deployed, i.e., all votes were considered meaningful and valuable and contributed equally to the final outcome regardless of participant backgrounds and associations. These scores were then joined with the original LCPI mapping product and an analysis of spatial distribution commenced using ArcMap 10.1. In a related effort, Dixon et al., 2012 developed land use land cover maps for Segment 10 by interpreting and digitizing 2006 and 1950s imagery This coverage was overlayed with the formalized LCPI l ayer to explore and interpret the conditions of the potential sites being considered for
85 recovery of cottonwood communities along this critical reach. An inverse distance weighting scheme was performed on the formalized LCPI indices and a g lobal Morans I analyses was deployed to assess spatial autocorrelation. 3.5 Results 3.5.1 Results of the Expert Elicitation Exercise 188.8.131.52 Rankings of LCPI classes for older stands (> 25 yrs old) Rank correlations and univariate analyses were used to compare and contrast the individual expert preferences and the overall weighted rankings of the LCPI classes for the Older stands (> 25 yrs old). Estimates from the individual 10 experts were moderately correlated with the weighted average rank sums for the Older stands (Pearsons = 0.9590.076 ) and the mean of these correlations ( = 0.521) was slightly higher than the median ( P10 = 0.418), indicating a slight positive skew. Variations amongst participants were evident, and the average Spearmans rank correlation coefficient rank sums method) was 0.51 ( s = 0.37) ( Table 3 2 ). Four of the ten participants (two of which were academics) were strongly correlated with the weighted av erage rank sums for the Older stands ( > 0.88) with a high degree of statistical significance ( p <0.001). Stronger correlations were found with the weighted average sum ranks when the participants scores were aggregated into subgroups (Academics vs. Managers) ( = 0.93 and 0.97 respectively), and a moderately high correlation was seen between the two subgroups ( = 0.86) suggesting a high degree of agreement amongst the subgroups at a significant level ( p <0.001). Based on the expert responses, the infrequently flooded ( I ) and moderately flooded ( M ) areas in the LCPI classification were the most likely to support older
86 (> 25 year) stands of cottonwoods, and within those categories, the mostly likely categories to offer niches were moderately retentive ( IM received the highest rank = 16; MM received the second highest rank = 15) ( Figure 3 6 ). On the other hand, areas characterized at the extreme ends of the LCPI classification were voted least likely to support older stands ( FS received the lowest rank = 1; RN received the second lowest rank = 2). 184.108.40.206 Rankings of LCPI classes for younger stands (2 25 yrs old) Similar rank correlations and univariate analyses were used to compare and contrast the individual expert preferences and the overall weighted rankings of the LCPI classes for the Younger stands (2 25 yrs old). Individual estimates from 10 experts were more strongly correlated with the weighted average rank sums for the Younger stands 0.96) and the mean of these correlations ( = 0.73) was slightly lower than the median ( P10 = 0. 81), indicating a slight negative skew. Variations amongst participants aggregated rank (combining all the experts using the weighted rank sums method) was 0.73 ( s = 0.27) ( Table 3 3 ). This time, four of the ten participants (including the same two academics) were very strongly correlated with the weighted average rank sums for the Younger stands ( > 0.91) and four more participants were strongly correlated with the weighted average rank sums ( > 0.740.87). Again, when aggregated into subgroups (Academics vs. Managers) even stronger correlations with the weighted average sum ranks were noted ( = 0.91 and 0.97 r espectively), and a moderately high correlation between the two
87 subgroups ( = 0.81) again, suggesting a high degree of agreement amongst the subgroups at a significant level ( p <0.001). Notably, the experts indicated that the younger stands (225 year o lds) were more likely to be supported by areas receiving increasing levels of flood waters (higher scores trended more toward the MM and MP LCPI categories, scoring 16 and 15 respectively), although too much flooding (i.e., the FS category) was considered less than optimal (ranks = 6) ( Figure 3 5 ). In the case of the younger stands, areas rarely receiving floodwaters (i.e., those at the extreme end of the LCPI classification) were voted least likely to support younger stands ( RN, R P RM and RS received the lowest rankings of 1, 2, 3, and 4 respectively). 220.127.116.11 Formalization of the rankings Formalization of these rankings (based on the weighted scores) reinforced these trends while transforming their values to a 0 to 1 scale more commonly utilized in standard habitat suitability evaluations ( USFWS, 1980) ( Figure 3 7 ). As the figure suggests, the experts assigned higher response scores ( > 0.75) to wetter, more retentive conditions for the younger stands (light green bars) with the exception of the most severe instance (i.e., FS ) where in all likelihood frequently flooded, strongly retentive areas reduce suitability of the areas. In t he older stands, the experts assigned higher response scores ( > 0.75) to the LCPI categories that displayed moderate wetness scales (fluctuating slightly with retentiveness) (refer to the dark green bars in the figure). In the younger stands, optimal niche provisioning (score = 1.0) was assigned to only one of the 16 LCPI classes (i.e., MM), while two of the 16 LCPI classes (i.e., MM and IM ) in the older stands (> 25 years old) were assigned optimal scores (1.0).
88 3.5.2 Results of the Overlay Analysis Substituting the categorical LCPI data with the formalized index scores and remapping the Segment 10 reach using these new values allowed us to identify areas of highest potential for recover efforts based on this single hydrologic indicator ( Figure 3 8 ). A g lobal Morans I analyses using inverse distance weighting w as performed on the formalized LCPI index scores using their position on the landscape to assess spatial autocorrelation. As one might expect, a strong correlation was found the m ajority of the high scoring areas (>0.75) were located immediately surrounding the river and its associated tributaries (note the greener areas on the map) ( z score > 2.58, p<0.001 using either age class scale). In these areas, existing older cottonwood st ands (> 25 years old) scored between 0.3 to 1.0 on the LCPI scale and the mean ( = 0.81) was slightly lower than the median ( P11678 = 0.85), indicating a slight negative skew where only 27 percent of the forests scored higher than a 0.75. Existing younger cottonwood stands (225 years old) along the river scored between 0.2 to 0.9 and the mean ( = 0.62) was again slightly lower than the median ( P2270 = 0.65), indicating a slight negative skew as well, where 45 percent of these communities scored greater than 0.75. More notably, the highest LCPI scores on the landscape were not f ound in forested areas, but were co located with farmland and pastoral settings (51 percent under the older stands LCPI scale; 11 percent under the younger stands LCPI scale) suggesting that many areas under cultivation/production today hold enormous potential for cottonwood forest recovery in the future. Interestingly, 54 percent of the existing cottonwood seedling sites (stands less than 2 years of age) were aligned with high scoring LCPI areas (using the younger stand LCPI scale) suggesting that the LCPI indicator could be a good gauge of future cottonwood recruitment and emergence.
89 Assuming that higher scores (greater than 0.75) should be correlated with existing cottonwood communities if these Missouri River system is fully functional, categorical data were used to score each cover type polygon on the landscape based on whether the LCPI scores indicated presence/absence of cottonwood communities and whether the observed results mirrored these expectations. Only a slight, positive yet significant correlat ion was found between the highest LCPI scores (greater than 0.75) and the presence of cottonwood communities in the current landscape ( = 0.04, p <0.001). Twenty six percent of the time, older forest stands were found where high LCPI scores indicated they should be present. Seventy four percent of the time, high LCPI scores indicated that the cottonwoods should be present, but some other cover type was found instead. Again, these results suggest that many areas could serve as potential targets for cottonw ood forest re establishment, but for one reason or another (possibly competing interests with farming and ranching) there are no forests present on these sites today. Since the LCPI curves were calibrated using expert elicitation rather than field data, an d because the LCPI values capture potential ecosystem response rather than realized suitability, a comparison of the LCPI scores and the preregulation land use land cover of Segment 10 was undertaken to identify agreement or dissonance between the pas t conditions and the present expectations. Comparing the 1950s land use land coverage and the contemporary map of the formalized LCPI results o nly 31 percent of the older cottonwood communities and 27 percent of the younger communities mapped in the 1950s coverage were colocated with the high LCPI scores ( greater than 0.75) of today Even more recently, the Missouri River experienced recordsetting floods in
90 November of 2011 that realigned the thalwag, altering the distribution of younger forested communi ties throughout the system. A comparison of preand post flood imagery along with the LCPI results in Segment 10 anecdotally suggests that the formalized LCPI could prove useful in targeting new sites to reestablish cottonwood communities ( Figure 3 9 ). 3.6 Discussion and Conclusions The rapid and extensive degradation of critical habitats on highly regulated large river systems across the country increasingly calls for the strategic intervention of planners and managers to maintain and recover ecosystem structure and function before critical tipping points are reached and communities are no longer sustainable. In these urgent and politically charged situations carefully prepared technical information (i.e., hard data) is not always readily available and decision makers are regularly faced with a repeatedly unpalatable decision: do nothing and suffer the consequences of uncertainty paralysis ( Chapin et al., 2010; McElhany et al., 2010) or choose to fill in the gaps with the best available, albeit softer and more qualitative, information, and then choose to adaptively manage the consequences thereafter. Oftentimes, as is the case on the Missouri River, this qualitative ecosystem based knowledge is implicit, warehoused in unpublished materials and gray literature, or locked within the minds of academics and professionals who have gathered knowledge and wisdom throughout their careers based on experiences and lessons learned. In these situations, it is reasonable, nay incumbent upon regional planners and managers, to actively engage these experts and acquire this knowledge and wisdom in an unbiased manner using prescriptive expert elicitation methodologies.
91 The purpose of this article was to present a prescriptive approach to reliably formalize relevant indicators of ecosystem integrity to a degree that sufficiently answers the problem at hand in situations where knowledge gaps and time and/or budgetary constraints limit a managers ability to otherwise quantify ecosystem response. The approach called on experts to fill these gaps and managers to strategically assemble, summarize, organize, interpret, and reconcile their experiences using heuristics in a rational, transparent, and collaborative manner to produce a relevant, sufficient, and defensible solution. The Missouri River case study demonstrated the efficacy of this approach, generating a quantitative indicator of ecosystem response (niche provisioning) for the rivers cottonwood ecosystems based on the experts interpretation of the LCPIs landscape classification system. Although the formalization approach has been successful thus far, it is important to note that this study was limited in scope in that only one of the thirteen priority segments identified in the BiOp ( USFWS 2000 ; 2003 ) have been assessed to date. Although the approach is believed to be generally applicable to other segments in the basin, a change in transdisciplinary team membership (to engage natural resource managers from local environs) could affect the results as a function of the number of new participants and their level of expertise. Notably, an LCPI analysis has already been conducted on Segment 13 [running from the mouth of the Platte River to Kansas City, MO; River Mile (RM) 595.5 to 365.5] ( Jacobson et al., 2007) An obvious next step would be to transfer the prescriptive protocol to this location and tes t its utility in the new setting. Conducting this follow on analysis (as well as ones on the remaining segments as LCPI analyses become available) would strengthen the conclusions of the current
92 study. Any applications of this approach in the future should follow the same elicitation and aggregations protocols to assure consistency. Sensitivity analyses should be undertaken to account for variations in the results if new information is acquired. In addition, shifting from the somewhat low tech approach desc ribed here to a more automated toolset (i.e., Survey Monkey, TurningPoint Technologies, etc.) should be considered to improve the experience and reduce the error rates in the polling activities. Moreover, there are a variety of MCDA approaches (i.e., Order ed Weighting, Analytic Hierarchy Process, Swing Weighting, etc) that should be explored as well (refer to Linkov et al., 2009 and Kiker et al., 2005 for general of these options). The prescriptive f ormalization approach presented herein offers a transparent and meaningful mechanism to characterize (both qualitatively and quantitatively) ecosystem integrity and promote EBM. In wicked situations where budgetary and time constraints limit the amount of hard data that can be collected, and/or the information is limited or nonexistent, regional planners and managers on the Missouri River can confidently use this formalized version of the LCPI to: 1) screen and select potential recovery sites, 2) assess ba seline conditions on these sites, 3) evaluate and compare proposed recovery plans for the sites, and 4) support adaptive comanagement for the Missouri River cottonwoods ecosystem .
93 Figure 3 1 Step 2 in the process focuses on the mathematical formalization of ecosystem response indicators using a combination of both of expert elicitation, hard data, literature review and numerical modeling. Performance Measures Model Performance Step 3: Calibration Ecosystem Response Models Step 2: Mathematical Formalization Step 4: Forecasting Step 5: Alternative Evaluation Quality of The Fit Model Verification Sampling Design Reference Datasets Evaluation Datasets Fitted ValuesResponse Thresholds Adaptive Co Management Step 6: Construction and Monitoring Statistical Literature, Existing Models, Expert Contributions Predicted ValuesModel Validation Study Goals and Objectives Performance Hypotheses Site Selection via GIS Based Decision Support System Step 1: Conceptual Modeling Laboratory and Field Experiments Description Data from Literature and ExpertsModel Goals and Objectives
94 Figure 3 2 The Segment 10 study area.
95 Figure 3 3 Land Capability Potential Index (LCPI) classifications for Segment 10 (RM 811.1 to RM 753.0Gavins Point Dam to Ponca State Park NE) ( adapted from Jacobson et al., 2008)
96 Table 3 1 Numbers and affiliations of the t ransdiciplinary team members participating in the LCPI formalization exercise. Category Participant Association Academia 3 Benedictine College University of South Dakota South Dakota State University Federal 4 USACE Omaha District USACE Omaha District USACE Kansas City District U.S. Geological Survey (USGS) State 2 N orth Dakota Forest Service South Dakota Dep artment of Agriculture, Div ision of Resource Conservation Private 1 EA Engineering
97 Figure 3 4 An example of results derived from one of the five elicitations t his example focused on the linkages between LCPI scores and the O ld G rowth age class. Exper ts highlighted in blue were affiliated with the Academics subgroup.
98 Figure 3 5 Comparison of expert opinions using i ntermediate elicitation results aggregated with weighted rank sums based on an age class basis. Higher ranks indicate best fit to the LCPI category.
99 Table 3 2 Spearman rank correlations of site ranks among experts ( n = 10) for the aggregated older stands (> 25 yrs old). Experts A + B C D + E F G H + I J A + B 0.47 C 0.44 0.51 D + 0.77* 0.10 0.51 E 0.76* 0.27 0.33 0.93* F 0.05 0.81* 0.96 0.32 0.05 G 0.20 0.51 0.89* 0.26 0.09 0.93 H + 0.27 0.15 0.29 0.19 0.07 0.23 0.11 I 0.09 0.37 0.67 0.09 0.22 0.60 0.89 0.00 J 0.85* 0.23 0.47 0.90* 0.95* 0.12 0.22 0.21 0.09 Aggregated Rank 0.90* 0.36 0.44 0.88* 0.92* 0.02 0.20 0.30 0.14 0.96* +These experts belonged to the Academics subgroup. S tatistically significa nt at p > 0.001.
100 Figure 3 6 Intermediate elicitation results aggregated using weighted rank sums on an age class basis. Higher ranks indicate best fit to the LCPI category.
101 Table 3 3 Spearman r ank correlations of site ranks among experts ( n = 10) for the aggregated younger stands (225 yrs old). Experts A + B C D + E F G H + I J A + B 0.82* C 0.89* 0.96* D + 0.95* 0.92* 0.94* E 0.09 0.53 0.40 0.25 F 0.61 0.86* 0.78* 0.75* 0.75* G 0.75 0.71 0.84* 0.74* 0.18 0.53 H + 0.40 0.32 0.44 0.34 0.32 0.44 0.48 I 0.13 0.09 0.07 0.03 0.31 0.21 0.57 0.13 J 0.61 0.66 0.60 0.68 0.56 0.74 0.27 0. 48 0.46 Aggregated Rank 0.84* 0.96* 0.96* 0.92* 0.56 0.90* 0.73* 0.53 0.10 0.76* +These experts belonged to the Academics subgroup. Only these correlations were statistically significant at p > 0.001. Figure 3 7 Fina l mathematical formalization of the LCPI categories based on expert opinion.
102 Figure 3 8 Characterization of ecosystem response potentials for : A ) older age classes and B ) younger age classes on the Missouri River based on the formalized LCPI scores generated in this exercise.
103 Figure 3 9 Comparison of formalized LCPI scores at three sites in Segment 10 using the : A) Young age class formalized LCPI scores compared to the B) pre flood (2008) and C ) post flood (November 2011) imagery (imagery courtesy of the USACE Omaha District).
104 CHAPTER 4 SEEING THE FOREST FOR THE TREES: EXPLORING TH E WICKED PROBLEM OF COTTONWOOD RECOVER Y ON THE MISSOURI RIVER WITH SPIRAL BASED ECOSYSTEM RESPONSE MODELING 4.1 Introduction T he wicked problems associated with the damming and regulation of large rivers are considered to be overwhelmingly complex but at the same time fundamentally simple ( World Commission on Dams, 2000, page xxvii ) On the surface, the situation is fairly straightforward dams dredging, channel straightening, bank stabilization and levee construction all provide clear and tangible ecosystem goods and services including flood control, irrigation, hydropower, navigation and recreation directly contributing to human well being ( National Research Council, 2005a; van Oudenhover et al., 2012 ) Underneath however, t hese socioeconomic benefits are attained at the expense of a broad range of unintended system wide environmental consequences e.g., water quality degradation, sediment transport reduction, and significant losses of biodiversity ( National Research Council, 2002 ) From the salmon in the Columbia River, to the least terns on the Missouri River, to the pallid sturgeon in the Mississippi River, across the nation the U. S. Army Corps of Engineers (USACE) is now engaged in numerous large scale, multi disciplinary recovery efforts to address more than fifty years of unintended environmental impacts arising from command and control river management practices ( Holling and Meffe, 1996; National Resea rch Council, 2002, 2004, 2005b) With an increasing sense of environmental awareness and appreciation, stakeholders on these rivers have challenged the USACE to pursue adaptive solutions that restore ecosystem integrity while maintaining flood protection for the floodplains inhabitants. Sustainable water
105 resource management in these contexts is not a trivial task There is sense of overriding urgency driving decisions that lead to both risky and inevitably controversial solutions. Moreover, multi stakeholder priorities tend to shift or evolve as the process of integration and reflection continually unveil new concer ns altering the focus of these socio ecological conflicts the epitome of wicked problems as described by Rittel and Webber (1973 ) In these situations, the USACE and its partners regularly turn to ecological modeling to measure benefits and discriminate amongst proposed recovery plans ( Tarboton et al., 2004; USACE, 2002 2004a ) These efforts are typically driven by reasonable and prudent measures identified in Biological Opinions i ssued by the U. S. Fish and Wildlife Service (USFWS) in response to Endangered Species Act (ESA) considerations. These directives tend to narrowly focus the USACE on the recovery of the parts (i.e., the species of concern) at the expense of the whole ( i.e., the functionality, integrity, and resilience of the system) even though the threats to the species are almost always systemic (Benson, 2012, page 4). Consequently, ecological models to support these initiatives tend to lean toward individual faunal based performance metri cs that often overlook or simplify system complexity and disregard the multidimensionality of these dynamic socio ecological settings ( Lester and Fairweather, 2011; Rogers, 2003) Decision makers require performancebased metrics to evaluate and assess proposed recovery plans based not on the species of concern, but rather on the ecosystems upon which they depend. Movement in this direction will allow the USACE and its partners to better understand and promote resilience while addressing the
106 complexity associated with the challenges of biodiversity loss ( Benson, 2012) Multidimensional, multimetric response models that capture ecosystem structure, composition and function will support the regional planners and managers in their efforts to adopt a more holistic, Ecosystem Based Management (EBM) approac h ( McLeod and Leslie, 2009 ) focused on sustainability through adaptive comanagement ( Armitage et al., 2009; Cundill and Fabricius, 2009; Light et al., 2013 ) in the face of shifting social and political regimes with an eye on improving adaptive capacity ( Benson, 2012) 4.2 Goals and Objectives Earlier I offered an approach to work w ithin wicked problem s a structured decision making paradigm that integrates the opinions and expertise of professionals with quantifiable data in a recursive, reflect ive and adaptive manner Figure 4 1 In Chapter 2 a discrete list of performance metrics (i.e., ecosystem response indicators) was produced to characterize the Missouri Rivers riparian forest community One of these variables (i.e., the Land Capability Potential Index or LCPI) was formalized in Chapter 3 The remai ning 12 variables must now be formalized (i.e., mathematically transformed) and assembled into a multivariate ecosystem response model for use in the assessment of proposed recovery plans under an adaptive management program The objective of this chapter is to present a spiral based modeling approach that generates multiscalar, multimetric ecosystem response models to support transparent decision making in these wicked situations. An example on the Missouri River demonstrates the value of this strategy, pr oducing a casespecific index model targeting the basins declining cottonwood riparian community. The intent was to create a performance metric that was defensible, efficient and operational (i.e., could be readily implemented within the constraints of t he federal planning and decision making
107 process) ( USACE, 2000 ; 2003 ). The challenge was to develop a performancebased metric that could transparently communicate the rationale for making hard decisions in the cottonwood recovery efforts to not only the stakeholders, but to the public at large. This chapter describes the evolution of an ecosystem response model to addr ess the challenges of the Missouri Rivers wicked problem, and discusses how the final product can be employed to assess proposed ecological interventions on the river under an adaptive management paradigm. To begin, the ecosystem setting is characterized and the historical context leading up to the current crisis is described. A review of the foundational science behind ecosystem response modeling is then offered, and the attributes of performance metrics that best address the wicked situation are provided. In the methods section, the basic spiral modeling steps are outlined and the model calibration strategies are then described. In the results section, the verification and validation activities are presented to demonstrate the models efficacy, and a disc ussion of the lessons learned, next steps, and future application opportunities using this new performance metric is offered in the conclusions. 4.3 Background and Literature Review 4.3.1 The Cottonwood Crisis At the end of the Great Depression, the USACE was direct ed by Congress in the Flood Control Act of 1944 to construct six dams on the Missouri River and stabilize the channel with more than 1,200 km of levees to reduce severe flooding and stimulate economic development across the region. Unfortunately, the socio economic benefits of these projects (i.e., flood control, irrigation, hydropower, navigation and recreation ) were attained at the expense of many unintended environmental consequences ( National Research Council, 2002 ) There is a rich, extensive body of scientific research on the
108 Missouri River describing the impacts of regulation on the system. Bot h Jacobson et al. (2011) and Dixon et al. (2012 ) offer indepth descriptions of the preand post system hydrologic conditions, recounting the loss of the bi annual flood pulse, the shortening of the river itself by roughly 200 miles, and the massive reconfiguration of the channel bed resulting in losses of sandbars, side channels and backwaters on a massive scale. Reduction of these ecologically beneficial flood pulses, in combination with meander loss and land use conversions, has resulted in the loss or alteration of nearly three million acres of floodplain habitat system wide ( National Research Council, 2002 ) A decade later, Dixon et al. (2012) confirm these estimates and show that more than 70 percent of the riparian habitat on the Missouri has functionally declined since the damming of the rivers. The regulation of the system has been particularly devastating f or disturbance loving species on the river like the plains cottonwoods ( Populus detltoides W. Bartram Marsh. S ubsp. m onilifera (Ait on) Eckenwalder ) that depend on pulsing river dynamics to create substrate for colonization, and flood flows to recharge soil moisture, transport sediment and disperse seed. Rood and Mahoney (1990) and Scott et al. (1997 ) offer full descriptions of the plains cottonwood life hi story characteristics and landscape patterns associated with cottonwood community recruitment and establishment with regards to hydrological regime and geomorphology. In essence, t hey point out that the unique phreatophytic character of the cottonwood community suggests a clear relationship between life history traits of the species and the Missouri Rivers dynamic geofluvial processes now in abstentia. Moreover, Johnson et al. (2012) Dixon et al. (2012) and Scott et al. (2012 ) note three disturbing trends highlighting the cottonwood community crisis on the Missouri River:
109 1. T here has been significant land use conversio n from natural communities to agricultural croplands in the last 50 years (refer to panel (a) in Figure 4 2 ); 2. A dynamic shift has occurred with the reduction of flooding and pulsing within the riparian zone the hydrophytic community dominated by phraetophytes such as plains cottonwood, sandbar willow ( Salix interior ), and peachleaved willow ( Salix amygdaloides Anders.) have converted successionally to a mesic community dominated by more mesophytic species including silver maple ( Acer saccharinum ), box elder ( Acer negundo), hackberry ( Celtis occidentalis ), black walnut ( Juglans nigra), green ash ( Fraxinus pennsylvanica), American elm ( Ulmus americana), and red cedar ( Juniperus virginiana) (refer to panel (b) in Figure 4 2 ); and 3. A dramatic decline in cottonwood recruitment and the aging of existing cottonwood stands suggests a depauperization of the latesuccessional forest community and the floodplain landscape as a whole (refer to panel (c) in Figure 4 2 ). In response to this crisis, and in compliance with the U.S. Fish and Wildlife Biological Opinion (BiOp) ( USFWS 2000 ; 2003) the USACE has developed a Cottonwood Management Plan (CMP) (USACE, 2010) with the intent of providing a single, comprehensive ecosystem based strategy to guide cottonwood community recovery efforts on the river. The plan includes potential interventions (i.e., alterations of ecosystem structures a nd processes aligned with shifting social and policy drivers) ( Hobbs et al., 2011) targeting the improvement restor ation, and expansion of sustainable cottonwood communities basinwide Numerous management options have been proposed, and the program managers are now faced with the unenviable task of determining which solutions are feasible, equitable, and likely to result in the resilient and sustainable recovery of the ecosystem. 4.3.2 Ecosystem Response Modeling In these situations, the federal agencies and their regional partners often turn to ecological models to characterize ecosystem response and discriminate amongst proposed plans. In the past, mitigation and conservation decisions in this context were informed by simplistic habitat suitability assessments tied to char ismatic wildlife species
110 populations (e.g., birds, fish, mammals), e.g., habitat suitability indices [reviewed in Dijak and Rittenhouse ( 2009) ; Larson et al. ( 2009) ; and first originating in USFWS ( 1980) ]; ecological niche models [reviewed in Hirzel and Le Lay ( 2008 ) ] ; and resource selection functions [reviewed in Boyce et al., (2002) ]. Virtually all attempts to use these models in support of ecosystem level management decisions have been heavily criticized by the community of practice because their central tenet assumes that the responses of a single species can serve as a surrogate for an ecosystems state without regard for scale, causal relationships, or system interdependencies ( Lester and Fairweather, 2011; Rogers, 2003) More recently, researchers have turned to more holistic mechanisms to characterize ecosystem integrity, e.g., wetland functional capacity indices ( Smith et al., 1995) ecological integrity indices ( Andreasen et al., 2001; Reza and Abdullaha, 2011) and weight of evidence approaches ( Ferretti and Pomarico, 2013; Linkov et al., 2009 ) Unfortunately, a review of the relevant literature has shown that these models are based largely upon literature reviews and untested expert knowledge with little or no quantifiable data to calibrate their internal mechanisms, or strong evidence to verify and validate their estimates of ecosystem response. One option is to take a broader ecosyst em perspective and assess recovery plans at multiple temporal and spatial scales ( Poiani et al., 2000) building models based on the integration of hard data and soft knowledge (extracted from subject matter experts) that recognize ecosystem integrity must be captured using a full complement of physical and biological components. One approach, referred to as ecosystem response modeling is a relatively new practice in the ecosystem restoration community, but one that shows significant promise in meeting this c hallenge ( Lester
111 and Fairweat h er, 2011, page 2690). These models, considered abstractions of ecosystem integrity characterizations, are designed to predict change in ecosystem state at multiple spatial and temporal scales as a consequence of one or more external stimuli [adapted from Marsh and Cuddy (2010 ) ]. In the case of EBM, proposed action plans are considered stimuli altering ecosystem structure, composition and/or function, and the ecosystem of concern is conceptualized using driver stressor effects causal mapping to identify key relationships and interdependencies amongst valued ecosystem components (e.g., biotic integrity, hydrology, soils, spatial integrity, disturbance, etc.) ( Burks Copes and Kiker, 2014 (submitted) b ) Mechanisticallyspeaking, these valued ecosystem components are decomposed into individual performance metrics (aka state variables or endpoints) whose values can be related to reference conditions through response index curves that are normalized on a 0 to 1 scale, where 1 represents optimal conditions (emulating reference standards) and 0 indicates a wholly dysfunctional state. Variables are aggregated in an algorithm on a component by component basis. These component stores are then aggregated to generate an overall composite index (again scaled from 0 to 1). The conditions of a site are characterized on a cover type basis using this procedure, but only cover ty pes associated with the model are assessed (i.e., all other land classes are assumed to be nonfunctional; score = 0). In an application, these scores are then relatively weighted by aerial extent of contributing cover types to generate a characterization o f integrity for the site overall. The decomposition process can and should enhance a decision makers ability to perceive functional relationships and interdependencies, offering insight into the range
112 of possible consequences that could result from presc riptive interventions from a systems perspective, while simultaneously improving and facilitating transparency. On the other hand, too much unstructured information garnered from experts and managers in these decomposition exercises can paralyze the decisi on making process ( Rogers 2003) In ecosystem response modeling, the true challenge is to avoid becoming mired in the complex array of potential performance metrics, and tease out the more critical indicators of change that can truly be affected or addressed with proposed management measures, balancing complexity with representativeness, to ensure that the models are both intuitive and easy to use ( Lester and Fairweather, 2011; Marsh and Cuddy, 2010) Several heuristics (i.e., rules of thumb) exist in the community of practice offering model developers guidance on deciding how many variables to include in a model ( Guisan and Zimmerman, 2000; Harre ll et al., 1996 ) 4.3.3 Spiral Modeling Model building transforms initially vague and general ideas into clearer and more formalized representations of the ecosystem ( Bredeweg et al., 2008) Jakeman et al. (2006) and Jrgensen and Bendoricchio (2001) offer extensive guidance on developing models (in general) in a disciplined and credible manner The former calls for a codification of best management practices, urging modelers to be more creative in their examination of modeling options and more rigorous in their model testing. The latter offers detailed definitions and procedures to calibrate, v erify, and validate models. Both offer a 10step model development procedure that begins with problem definition, moves through conceptualization and data collection/analyses, and ends with model construction and testing. Although both suggest an iterative process, they portray these in a sequential modified waterfall construct ( Royce, 1970 ) ignoring or avoiding
113 activities that woul d engage stakeholders in the participatory or collaboratively rational ( Innes and Booher, 2010) process. Alternatively, from the softwar e development and information technology communities of practice, Boehm (1988) offers a proof of concept approach that spirals through iterative productions of a models version to reduce risks of f ailure and improve model representativeness. The process starts with a problem defining activity and the development of a straw man version of the product (i.e., the model). Each spiral thereafter serves as a reflexive developmental phase that engages the customers (aka the stakeholders) in a recursive critique of the emerging prototype. Under this paradigm, not only is the model finetuned with customer feedback spiral by spiral, the developers encourages the customer to revisit their goals, objectives and constraints to reduce the risk of failure. Interestingly, each subsequent spiral is triggered by the planning of the next phase of development, encouraging forward momentum toward a final conclusion. However, the process allows for routine maintenance on a continuing basis (i.e., synonymous with the concepts of monitoring and adaptive management under the EBM paradigm) to assure success. Few studies have attempted to implement spiral based paradigms into EBM initiatives ( Bredeweg et al., 2008; Du Toit, 2005) Bredeweg et al. (2008) establishes a causal mapping framework to support conceptual modeling with regards to sustainable development for the NatureNet Redtime1Du Toit (2005 project, while the ) application has established a new legal framework to water resources management in South Africa. 1 http://staff.science.uva.nl/~bredeweg/pdf/NNR/D4.2.2.pdf (Accessed Sep t 2013).
114 No study to date has merged the spiral based modeling concepts of Boehm (1988) with the concepts of creative problem solving envisioned by Steiner (2009) to develop a multivariate, multi scalar ecosystem response model that can help decision makers determine which solutions will likely result in the resilient and sustainable recovery of an ecosystem. With the guidance offered by Jakeman et al. (2006) and ( Jrgensen and Bendoricchio, 2001) in mind, and with the intention of constructing a model for the cottonwoods on the Missouri similar to those generated by Lester and Fairweather (2011 ) and Marsh and Cuddy (2010) a spiral based, participatory approach was devised to produce a performancebased tool to assist the USACE and its regional partners in directing recovery efforts under the CMP ( USACE, 2010 ). 4.4 Methods In this study a series of facilitated and interactive expert driven workshops were used to establish an index based ecosystem response model characterizing ecosystem integrity of the Missouri River cottonwood forests. The goal was to combine what the experts agreed should be included in an ecosystem response model with the realities of how endusers would a ctually be using this metric to mak e decisions The model development process was designed as an exercise in reflexive learning in context a term coined by Du Toit ( 2005, page 229) to describe an interactive group approach that encourages stakeholders to identify problems, deliberate, propose solutions and respond to contextual changes in recursive reflection cycles (centered around i nformation presented at each workshop/web meeting). A structured four phase iterative construction spiral was devised with intermittent in person and remote meetings to promote transparency and encourage creative problem solving i.e., the participatory env ironment encouraged the stakeholders to inject
115 reflection and experience into the model architecture thereby improving model performance ( Figure 4 3 ). The first phase, referred to as the Problem Definition Spiral focused on deli neating the study domain, verbalizing the problem and study constraints, setting goals and objectives, convening a team of subject matter experts (both from the academic and the natural resource management arenas), and planning for the next spiral. Phase 2, referred to as the Conceptualization Spiral focused on generating a conceptual model of the problem space (i.e., identifying key drivers, stressors, and effects) that led to the identification of critical ecosystem components and meaningful endpoints to characterize component state. The third phase, referred to as the Data Collection and Analysis Spiral concentrated on data mining and data transformation processes in support of the final phase, referred to as the Construction and Testing Spiral At cri tical junctures all along the spiral paths, the stakeholder team was engaged through onsite workshops or via remote web meetings/telecoms to review the state of the model, reflecting on decisions made in the last meetings, as well as learning and adapting the model to address new challenges as they surfaced. Constant and structured team interactions promoted trust amongst the participants, and led to increased confidence in the final model construct. Iterative and recursive reflection supports group learning and increases both the participants understanding of the ecosystem as well as the manner in which the model will be utilized to support the recovery efforts. Constant feedback to the modeler(s) increased modeling competence and improved articulation of ecosystem response. The spiral was left openended in
116 the future, the team will be responsible for collaboratively monitoring the models use in the recovery program and incorporating their feedback with regards to utility and reliability will support long term adaptive management of the model itself. The following sections briefly describe the activities performed in each spiral of the Missouri Rivers cottonwood model development. 4.4.1 Spiral #1: Problem Definition To define the problem, the models operational domain was delineated and a multidisciplinary team of subject matter experts was convened to define the problem space and set the goals and objectives for the models itself. The key was to include just enough information to make the model sufficient a nd not too much information to render the model unwieldy. 18.104.22.168 Study area and model d omain The Missouri River is the longest river in the United States draining more than 1.3 million km2 ( 530,000 mi2), and cover ing approximately onesixth of the continental United States ( Galat et al., 2005) ( Figure 4 4 ). The study area is limited to eight reaches in the basin, and approximately 1,497.2 river miles in length ( Table 4 1 ). Six of these reaches (Segments 4, 6, 8, 9, 10, and 13) have been identified as high and moderate prio rity sites for bald eagle compliance with the 2000 BiOp (amended in 2003) ( USFWS, 2000; 2003). Segment 2 has been included to capture the full complement of inter reservoir reaches in the basin, while Segment 0 has been included to provide a reference reach for the study that can be co nsidered less impacted under the current and historic flow regulations. Six dams were constructed on the Missouri River by the USACE in the mid 1950s: 1) Fort Peck Dam at the top of Segment 2 (RM 1771.3), 2) Garrison Dam at the top of Segment 4(RM 1390), 3 ) Oahe Dam at the top of Segment 6
117 (RM 1072.3), 4) Big Bend Dam at the bottom of Segment 6 (RM 987.4), 5) Fort Randall Dam at the top of Segment 8 (RM 880), and 6) Gavins Point Dam at the top of Segment 10 (RM 811.1). Character, size and areal extent all p lay a role in distinguishing amongst the eight study reaches. The upper segment (i.e., Segment 0) is the only freeflowing segment in the project It is the third longest segment of concern, but smallest in areal extent mapped. Segments 2 through 10 are inter reservoir systems with dams controlling their flows to a great extent. Segment 13 is a fully channelized system located at the bottom of the study area, and considered both the longest and largest reach based on areal extent. However, Segment 10 is the widest segment included in the study. Segment 10 (highlighted in dark blue in the figure) has been chosen for purposes of model testing in this study because it has a hybrid hydrogeomorphological setting as it presents physical characteristics of both re gulated and non regulated rivers. In other words, its location just below Gavins Point Dam, and it its lack of channelization, offers a preregulated hydrology (i.e., no slackwaters). However, s ignificant channel degradation and declines in bed water surfa ce elevations have led to significant changes in geomorphological conditions since the construction of this dam in 1957 ( Dixon et al., 2010 ) 22.214.171.124 Study team and workshop f ormat A transdiciplinary team was convened to develop the model, engaging both academic researchers from different unrelated disciplines as well as nonacademics such as land managers and other stakeholders in the process. From the onset, it was agreed that practical experience garnered from professionals was as desirable as theoretical knowledge derived from academia, and that interactive group methodologies
118 would be used to extract knowledge and expertise from the team via onsite workshops and follow on conference calls and web meetings. Over the course of seven years, 82 stakeholders spanning the gamut of academic, tribal, federal, state, local, private and nongovernmental arenas were engaged from across the region to participate in the process ( Table 4 2 ). In general, participant expertise spanned the gamut of biological sciences ( sixty two percent ), hydrology ( fifteen percent ), geology ( eight percent ), planning and management ( e ight percent ), and engineering ( seven percent ). Their job descriptions were characterized as planners, regulators, natural resource managers, professors, geographers, and consultants. Their level of familiarity with the system ranged from those with less t han five years working on the system ( thirty eight percent ) to individuals with well over twenty years of experience on the system (twenty three percent ). The majority of the participants had between five and twenty years of hands on expertise ( forty eight percent ) in this setting. Sixty nine percent of the participants had no formal training in habitat model building, but eight percent had received some formal training (Habitat Evaluation Procedures (HEP) training offered by the USFWS), and twenty three percent indicated they had received on the job training. All told, six intensive week long onsite workshops were held to conduct the group elicitation. Each workshop was regularly attended by 20 30 participants, and two separate note takers were used to produce both informal notes and official minutes to document the decisions in each meeting. Each workshop began with a review of the decisions in the previous meetings and a discussion of the effects these decisions had on the model development thus far. The agenda then turned to the presentation of any
119 new data collected, and consensus building activities were used to incorporate this new information into the development and fine tuning of the final model. Some elicitations were formalized (i.e., via elicitation spreadsheets or automated voting activities using TurningPoint technologies), but in general, the majority of these exercises were conducted in a group interactive manner that encouraged open discussion and debate resulting in a verbal group consensus In between the workshops, monthly two hour teleconferences were used to facilitate data transfer between the individual study teams (i.e., planning, sampling, mapping, alternative formulation, and final CMP documentation) and formulate paths forward to c lose any knowledge gaps. These decisions and activities were recursive results and decisions were revisited at least twice in follow on meetings and teleconferences to facilitate course corrections and assure that in the end, the models behavior conformed to the stakeholders perceptions of ecosystem integrity on the river. 126.96.36.199 Modeling goals, objectives, and constraints A series of decisions shaped and focused the model development process. The goal of the modeling effort was fairly straightforward: Develop an ecosystem response model that the USACE (and its partners) can use to either evaluate recovery options or deploy within an EBM framework to monitor conditions and trigger adaptive co management activities when ecosystem state conditions f all outside ex pected trajectories. A multivariate, community level index based model was envisioned, but it was clear that a single construct would likely be insensitive to the age class variations in structure, composition and function exhibited across the cottonwood c ommunity. A bimodal algorithm was envisionedone tailored to younger forested stands, and another geared toward older forested stands ( Table 4 3 ).
120 Model calibration followed a referencebased paradigm ( Bailey et al., 1998; Miller et al., 2012 ; Rheinhardt et al., 2007; Society for Ecological Restoration International, 2004) and the documented successional shift from hydrophytic to mesophytic species occurring on the river ( Dixon et al., 2012; Johnson et al., 2012 ) offered a unique opportunity to improve the models representativeness by using the distinctions between these two systems to sensitize the models response curves. In other words, the model was designed to capture the functionality of both the hydrophytic and mesophytic settings, but optimal response scores were associated with the characteristics of a hydrophytic rather than the mesophytic community. In the absence of adequate referencelevel data, structured expert elicitation techniques were deployed to fill the data gaps with local expertise and knowledge using strategies and techniques detail ed in ( Gregory et al., 2012; Kreuger et al., 2012; Meyer and Booker, 2001 ) After in depth discussions with the USACE, it was determined that the model should be operational at the site level on the river, capable of discriminating amongst plans proposed to alter conditions on a clearly delineated footprint. The cottonwood community however, is not limited to the arbitrary boundary of a siteecological integrity is tied to processes occurring at the stand, segment, reach and system wide scales. To capture this dynamic range of influence, ecosystem structure, composi tion and function was characterized at multiple spatial and temporal scales and varying degrees of granularity. Assuming that granularity, scale, and the models sampling paradigm would be inextricably linked, a hierarchical heuristic was used to establish sampling protocols that measured performance metrics in each of the three levels (stand, segment and system). This allowed the multi dimensionality of the ecosystem to
121 be captured and integrity to be measured at multiple scales (i.e., local, regional and landscape). 4.4.2 Spiral #2: Conceptualization A causal mapbased conceptual model was generated by Burks Copes and Kiker (2014 (submitted) b ) to illustrate the relationships between system wide drivers and stressors, uncovering the lines of evidence tying drivers and stressors to ecosystem responses generating a series of performance measures of ecosystem integrity for the cottonwood communities found along the banks of the Missouri River Nine drivers ( five anthropogenic and four natural), and eight stressors were identified. These were linked to effects on the ecosystem through t he delineation of valued ecosystem components that can be collectively described as hydrography, biotic integrity, and spatial integrity. Combining these individual indicators into a characterization of ecosystem integrity using a weight of evidence approach ( Linkov et al., 2009) called for creative problem solving and critical thinking at both the strategic (futureoriented) and tactical (near term) scales ( Burks Copes and Kiker, 2014) One workshop and several follow on teleconferences p roduced the proposed model construct: ERI = ( ) + + ( 4 1) where the Ecosystem Response Index (ERI) was equal to the compensatory combination of the three components Biotic Integrity (BIOTA), Hydrography (HYDRO), and Spatial Integrity (SPATIAL), but the BIOTA component was considered to be three times more important in capturing ecosystem integrity than the other components. This hypothesis would need to be tested, but the experts agreed to move for ward with this initial model construct.
122 A second outcome of the conceptualization was a discrete list of thirteen measureable lines of evidence or ecosystem response indicators that could be used to assess the state of these three components under various intervention plans ( Table 4 4 ). Transformation of these indicators into useable constructs would become a multi step task. Each indicator would need to be systematically measured or characterized qualitatively and then mathematica lly formalized (i.e., transformed into a mathematically based interpretation using either statistical analyses or professional judgment into ecosystem response curves using a standardized scale ranging from 0 to 1 in a prescriptive, rational, and scientifi cally defensible manner It is important to recognize that the model was designed to operate at the site level, in accordance with standard USACE planning protocols which call for the formulation and comparison of alternative designs focused on site speci fic recovery activities in the study domain ( USACE 2000, 2003). The cottonwood community however, is not limited to the arbitrary boundary of a site, nor is the ecological integrity of the community tied to only the processes occurring within a locations footprint. Rather, model variables were intentionally included to capture ecosystem structure, composition and func tion operating at multiple spatial and temporal scales, measured at varying degrees of granularity. As Table 4 4 indicates, some variables were calibrated with data collected in the present setting (i.e., CANHERB, CANSHRUB, CVALUE, RICHNATIVE, and WIS ), while others were developed using historic settings (19=890s and 1950s) capturing the physical setting of the system before and just after impoundment (i.e., ADJLANDUSE, INTERSPERS, DISTPATCH, and PATCHSIZE).
123 Spatially speaking, the multi scalar nature of the cottonwood community dictated a hierarchical nesting of metrics spanning a breadth of scales from the stand to the system level. Measurements taken at the stand scale were aggregated to generate response curves characterizing t he condition at the landscape scale. Thus, a single response curve was developed for some variables based on data gathered across the system at the stand level to characterize ecosystem integrity (e.g., CANHERB, CANSHRUB, CVALUE, RICHNATIVE, and WIS ). Five variables (i.e., ADJLANDUSE, INTERSPERS, LCPI, PROPCTW and RECRUIT ) were sampled at the system level, but again aggregated to generate a single response curve that would then applied at varying scales (i.e., Site vs. Region vs. Landscape). The remaining v ariables (i.e., DISTPATCH, PATCHSIZE, and DEPTHGW ) were finely tuned (i.e., data was gathered on a segment basis and applied at either the site or regional scales) to distinguish amongst reaches in the system based on their unique hydrologic setting and context (e.g., free flowing vs. inter reservoir vs. unchannelized vs. channelized). Moreover, some variables were designed to assess only the conditions of certain types of patches in the site (e.g., CANHERB is assessed only in younger stands; DEPTHGW is on ly applied to older stands, etc.), while some variables were specifically designed to look beyond the site boundary and take into account the broader landscape level processes affecting the community (e.g., ADJLANDUSE PATCHSIZE, etc.). Thus, the model is operating at multiple scales and degrees of spatial and temporal resolution with varying levels of granularity to capture the multi scalar complexity of these ecosystems within the defined domain. Upon the conclusion of the workshops and
124 teleconferences, t he components of the conceptual model were incrementally operationalized and refined into an implementable model construct ( Table 4 5 ). 4.4.3 Spiral #3: Data Collection and Processing The following sections detail the spatiotemporal sca les and granularities associated with the thirteen model variables, describing the data collection and statistical processing strategies deployed to generate response curves for each indicator. 188.8.131.52 Cover type m apping In keeping with traditional HEP application protocols ( USFWS 19 80a c), and with the intent of conforming to traditional USACE planning paradigms ( USACE 2000 ) the priority segments were mapped using distinguishable vegetative class characteristics. This process, referred to as cover typing distinguishes between vegetative types (e.g., forest, shrub lands, wet/dry meadows, etc.) as well as distinctive hydrology and soils characteristics associated with these settings, and clearly delineates these distinctions on a map. As reported in Dixon et al. (2012 ) Scott et al. (2012 ) and Johnson et al. (2012 ) the mapping team cover typed the segments at three separate time increments, namely 20062008, 1951 1958, and 18921893. Heads up digitizing was deployed to cover type the system viewing images at a 1:10,000 scale for finely resolved polygons (i.e., forests and other patches), and 1:24,000 for larger, relatively simple polygons (i.e., agricultural croplands). Note that forested communities were decomposed by age class into two distinct community types: Older stands ( > 25 years of age) and Younger stands (225 years of age) and characterized as either dominated by cottonwoods or dominated by mesophytic riparian species. Roughly 479,805 ha ( 1 185, 631 ac) were mapped and delineated into seventeen discrete land use
125 classifications across the eight reaches in the study domain. Figure 4 5 offers an example of the mapping completed for each segment in the study. Channel straightening, expansion of agricultural croplands and the fragmentation of the cottonwood communities along the river are clearly evident in this example. Appendix A contains tables detailing the particulars of the mapping efforts. The cottonwood model is designed to assess conditions in cottonwooddominated and noncottonwood riparian (mesophytic) forest and shrubland stands older than 2 years of age (refer highlighted rows in the appendix tables to identify applicable cover types associated with the model). 184.108.40.206 Field data collection Between 2007 and 2009, regional data collection teams from the University of South Dakota, South Dakota State University, Benedictine College, and the USGS surveyed 327 forested stands in the eight study reaches to both groundtruth the cover type mapping and characterize the ecosystems state based on vegetative measurements ( Dixon et al., 2010) The intent was to establish three distinct data sets: 1) a training data set of reference standard conditions exhibiting the conditional range of fully functioning cottonwood comm unities; 2) a verification data set of noncottonwood and highly degraded cottonwood sites that could be utilized to verify accurate model response; and 3) a validation data set of holdout sites that could be used to independently validate model response. When possible, 3 0 cottonwood stands were sampled per segment Stands were defined as homogenous in character (i.e., defined by both dominant plant species and stand age). Within each segment two stands were sampled from each of the five age classes (i.e., > 100 yrs, 50100 yrs, 2550 yrs, 10 25 yrs, and 2 10 yrs). Of the 327 sites, 216 of the sites were categorized as cottonwood
126 dominated reference standard conditions. Of these, 136 sites were classified as Older stands (> 25 years of age) and 80 were clas sified as Younger stands (between 2 and 25 years of age). Thirty six sites were randomly extracted from the 216 sites and set aside for use as testing sets (i.e., holdout validation). At the opposite end of the spectrum, sub optimal stands (i.e., noncot tonwood stands and/or highly disturbed cottonwood dominated stands) were explicitly sampled for model verification purposes. One hundred one sites (71 noncottonwood and 30 highly disturbed cottonwood stands) were added to the sampling efforts (69 older/26 younger) to round out the data collection. In each of these sample sites, overstory composition and structure was measured using the point centered quarter method, fixed radius circular plots, and/or complete plot census methods. On sites sampled using the point centered quarter method, forty points were sampled per stand, with four trees per point (160 total per stand). At each point, the area was divided into four 90 degree quadrants, relative to the transect bearing and a line perpendicular to it. Withi n each of these quadrants, the nearest live tree was located with a trunk diameter at breast he ight (dbh) identified it to species, measured the dbh to the nearest centimeter, and measured the distance from the point to the center of the tree trunk to the nearest 0.1 meters or finer. For trees with multiple trunks, all stems that equaled or exceeded 10 cm dbh were measured and recorded. If the nearest tree in a quadrant is dead, the species (if known), dbh, and distance from point were recorded, and then the nearest live tree within the quadrant was located. In cases where no live tree could be located within a reasonable distance in the quadrant (e.g., > 35 m), the quadrant was recorded as
127 open. Distances were measured using an electronic measuring device (Sonin multi measure), optical rangefinder, or measuring tapes. For sites with open q uadrants, a correction factor was applied to estimates of stem density, using the simple correction suggested by DahdouhGuebas and Koedam (2006) T ree density at younger sites (typically characterized by high densities with stem diameters < 10 cm) were sampled using 12 fixed radius (15 m) circular plots instead of or in addition to the point centered quarter sampling. Within each circular plot, the number of stems were identified, tallied and stem diameter s for all trees ( were recorded. Shrub/sapling composition, density, and cover were measured using belt sampling and line intercept methods, while herbaceous species composition and cover were estimated using 1m2 quadrats. Twelve points were sam pled per stand for shrubs, while 24 points were sampled per stand for herbaceous layer plants. These points were either on completely separate transects from those used in the overstory sampling, or were offset to avoid trampling the herbaceous vegetation. These were generally arranged on four transects, as with the trees, with six herb points and three shrub points per transect. Plants occupying the shrub layer (shrubs and tree saplings > l m tall < 10 cm dbh) were sampled using a 2 m x 10 m linestrip met hod ( Lindsey, 1955) beginning at the point and running along the transect bearing. Woody stem density (number per ha) in the shrub layer was estimated by counting all individual shrubs, saplings, and woody vines found within the sampling strip (one meter to either side of the 10meter transect). Numbers were tallied for each species. Percent cover was estimated by recording cover by shrubs (or saplings and woody vines) that intercepted the centerline vertical plane of the plot above one meter off the ground. The total distance along the
128 10meter tape length with overhead shrub cover by each species was noted, and the contributions of individual species were summed to generate an estimate of total cover In 2008, the data recording procedure was modified to enable quantification of overlapping coverage, allowing estimation of total shrub cover (without inflated estimates from overlap) on each plot. Plants in the herbaceous layer (herbs and woody seedlings < 1 m tall) were sampled using a 1 m2 sampling frame (quadrat) beginning at or centered on the sampling point. Th e herbaceous quadrat was sampled prior to the shrub sampling top avoid trampling issues All species of nonwoody vascular plants and woody seedlings were noted and recorded and their percent cover within the 1m2 quadrat was estimated to the nearest five percent Species with trace occurrence were recorded as one percent cover. The field efforts produced: 1) stand level and complete plant (vascular plant) species lists 2) frequency and percent cover of each species in the herbaceous layer ; 3) frequency, percent cover, and density of each species in the shrub layer; and 4) frequency, density, basal area (m2/ha) and importance value (sum of per cent relative frequency, density, and basal area, with a maximum value of 300) of each tree species. For complete census plots for trees, there was no way to calculate relative frequency separately from relative density. Hence, for those sites (predominant ly in S egment 0), an importance value using relative basal area plus two times the relative density for each species was computed ( Dixon et al., 2010) An estimation of the wetland affinity and overall quality of the vegetation in each stand b y assigning published wetland indicator values ( Reed, 1988) and Coefficients of Conservatism (C values) ( Swink and
129 Wilhelm, 1994; Taft et al., 1997 ; The Northern Great Plains Floristic Quality Assessment Panel, 2001) to plant species. Response curve calibration for field collected variables (i.e., CANHERB CANSHRUB CVALUE, and RICHNATIVE ) was based on central tendencies of the data and breakpoints were derived using a variety of methods based on the results of various descriptive statistics (quartiles, means, medians, standard errors and standard deviations) as well as bootstrapping of the medians. Herbaceous and shrub canopy variables were only assessed for the younger stands per model architectural constraints, and wetland indicator curves were derived for both younger and older stands independently based on expert opinion. Curve calibration for the final field collected variable ( WIS ) was based on expert consensus. 220.127.116.11 Spatial data g eoprocessing Between 2011 and 2012, a series of geoprocessing protocols were devised to derive and calibrate spatially explicit landscape scale variables for the model using the baseline (20062008) and hist oric (1893 and 1950) vegetative mapping presented in Dixon et al. (2012 ) Scott et al. (2012) and Johnson et al. (2012) All shapefiles were re projected to conform to a standard geor eference (i.e., NAD_1983_UTM_Zone_14N, Transverse Mercator GCS_North_American_1983). Metric units were used for all distance measurements, and a minimum mapping unit of one acre was established based on the vector based data produced in the cover type mapping exercise. A cell size of 10 m2 (onethird of the minimum mapping unit) was established for all grid calculations to assure overlapping coverage in the raster based analyses. At the system level, two variables (i.e., ADJLANDUSE and INTERSPERS) were cal ibrated using historic (1893/1950s) conditions produced in the cover type mapping
130 exercise. Adjacent Land Use ( ADJLANDUSE ) considered the setting from a vector based landscape perspective, focusing on land use activities occurring within a 2 km buffer around the riparian zone lining the river (and associated tributaries) and calculating the relative level of disturbance within this buffer based on weighted sums of the areas contributing to four primary land use classes on each segment. The response curve wa s calibrated based on expert opinion, assigning incrementally declining values to each category exhibiting ever increasing levels of disturbance (i.e., Natural = 1.0, Pasture = 0.75, Farmed = 0.50, and Urban = 0.00). Relative Interspersion of Habitats ( INT ERSPERS) focused on a raster based analysis using Neighborhood Statistics tools in ArcGIS to calculate the variety of habitat types occurring within a 3m x 3m floating neighborhood window on a segment by segment basis. Response score values were assigned based on historic (1890s/1950s) values, assuming that increasing variety correlated with increasing ecosystem integrity (i.e., 1 = 0.0, 2 = 0.2, 3 = 0.4, 4 = 0.6, 5 = 0.8, and 6 = 1.0). Outputs of the variety calculations were exported to MS Excel, and relatively weighted by contributions to the segment. Relative weight was equal to n/N where n was the cell count per variety score and N was the total number of grid cells in the segment Two additional variables, namely the Level of Cottonwood Domination ( PR OPCTW ) and the Level of Cottonwood Recruitment ( RECRUIT ) were assessed at the system level. For application purposes, this variable was measured at the regional scale, and relevant patches were defined as polygons greater than 1 ac (minimum mapping unit) whose cover types were associated with the model (i.e., all age classes of shrubs and forests including both cottonwooddominated and mesic dominated
131 stands). The regional footprint was defined as the site plus any additional overlap of intersected polygons extending outside the site boundary. Cottonwood domination was measured by calculating the relative proportion of the forest and shrub communities dominated by cottonwood species on or in the vicinity of the site. The level of cottonwood dominance for each forest/shrub polygon was calculated relatively by dividing the number of acres dominated by cottonwoods by the total acreage of all forested communities (cottonwood and mesic combined). Once the proportion was derived for each polygon, the median of the regional footprints polygons was calculated. On the other hand, cottonwood recruitment looked exclusively at the number of cottonwood acres across the system associated with the younger age classes (2 25 years old) and compared this value with the total number of forest and shrubland acres across the system to determine how much recruitment was occurring in the system Unfortunately, the degraded setting precluded the use of field data to calibrate these two variables. As an alternative, a study conducted by Johnson in 1992 offered suggested breakpoints for the curves. The study indicated that prior to damming, 47% of the Missouri River f loodplain forest was once dominated by young (pioneer) age classes (< 40 years of age) and that over 90% of these forests were dominated by cottonwoods. A positive linear response was assumed for both the recruitment and domination variables in the model, with the assumption that the lines would change trajectory at each of the respective data points (0.47 for RECRUIT and 0.90 for PROPCTW ). At the segment level, two variables (i.e., PATCHSIZE and DISTPATCH ) were calibrated using 1950s mapping generated in t he cover typing exercise. Patch Size
132 ( PATCHSIZE) focused exclusively on the median size of cottonwood dominated forest and shrubland polygons (merged) and calibrated on a segment basis to capture the irregularities of the system (i.e., impounded vs. freef lowing remnants vs. channelized settings). Distance to Nearest Patch ( DISTPATCH ) was calculated in meters using the ET GeoWizards Closest Feature Distance tool. Central tendencies (i.e., means of bootstrapped medians, quartiles, and standard deviations of the bootstrapped values) were used to calibrate each segments curve for both variables, with minimum and maximum values establishing extremal breakpoints. 18.104.22.168 Hydrologic data collection and a nalysis Outside of the field collection and spatial analysis of cov er type mapping, b ackground research was conducted to determine available ground water level data ( DEPTHGW ) for the study area. T he South Dakota Water Rights Program and the Nebraska Lewis and Clark Natural Resources District were mined to generate w ater l evels from wells in the study area. Average ground water elevations over the last 10 years (19972007) were computed for each well in the study area. These values were plotted on a USGS 1:24,000scale topographic map. The data points were contoured by hand. The contours were digitized into ArcMap and a grid of ground water elevations was generated using the ArcGIS Spatial Analyst extension. A grid of ground surface elevations in the study area was acquired from the USGS National Elevation Dataset. The eleva tions in this grid were based on 1:24,000scale source data at a resolution of 10and 30meter cell sizes. Accuracy of the ground surface elevations was one half of the source data contour intervals. Using the ArcGIS Spatial Analyst extension, the ground water elevation grid was subtracted from the ground surface elevation grid. The result yielded a new grid which represented the depth to ground water. Again, central
133 tendencies (i.e., means of bootstrapped medians, quartiles, and standard deviations of the bootstrapped values) were used to calibrate segment curves, with minimum and maximum values establishing extremal breakpoints. Data gaps prevented the direct formalization of the LCPI using field measurements, so expert elicitation was utilized to calibr ate the response curve for this variable. Burks Copes and Kiker (2014 (submitted) a ) offer a detailed description of the formalization process and produce a normalized (01 scale) relationship between ecosystem integrity and LCPI for use herein. 4.4.4 Spiral #4 Const ruction and Testing Model construction and testing stepwise, iterative process that involved the development and evaluati on of: 1) individual Response I ndex (RI) curves, 2) algorithms to combine the RIs into meaningful characterizations of essential ecosys tem components (i.e., hydrography, biotic integrity, and spatial integrity arising from the conceptual modeling) referred to as Component Response Indices (CRIs), and 3) a composite Ecosystem Response Index (ERI) that aggregates the components in a manner that best captures ecosystem integrity (i.e., structure, function, and composition) and measures response to proposed recovery plans. Each step along this path employed techniques to calibrate, verify and validate the models curves, component algorithms, and the final model construct using both expert input and statistical analyses in a recursive fashion. Data entry, error checking, descriptive statistics, regressions and correlations were completed using MS Excel and Sigma Plot 10. The majority of the fie ld data processing and analysis was conducted i n the Statistical Analysis System software (SAS, version 9.1) by the field team ( Dixon et al., 2010) Spatial analyses were conducted in ArcMap 10.1 using ArcToolbox features,
134 ET Geowizards, and XTools. All GIS derived data was exported to MS Excel for statistical analyses. 22.214.171.124 Calibration For purposes of this study calibration focused on the estimation and/or adjustment of the models variables (and algorithms) using training data sets to improve the agreement between the models outputs (based on training data sets) and a set of testing data. To establish the training data set s, a series of systematic power analyses were conducted on all field data collected system wide (Segments 0 13). The results of the initial power analysis were conclusive five out of the six datasets demonstrated marked improvement in power with the om ission of outlier data associated with the upper and lower ends of the basin (i.e., Segments 0 and 13 respectively).This was not surprising given the fact that Segment 0 has a markedly different geomorphological context from the inter reservoir segments (occupying a narrow, post glacial valley with steep cliffs on both sides that limits the extent of cottonwood forest establishment along the banks of the river) and Segment 13 has been significantly modified by channelization and levees (resulting in small, highly degraded relictual stands of cottonwoods). Figure 4 6 offers an example of the refining power analysis used to establish the limits of the training sets. In the figure, the variable CVALUE (i.e., Mean Conservatism Value) was measured and the cumulative variance was analyzed across the eight priority reaches (Segments 0, 2, 4, 6, 8, 9, 10, and 13), the r2 value indicated that 82% of the variation was explained by the sampling set. When Segment 0s data was removed from the analysis the r2 value rose to 88%, and the r2 value rose again to 91% when both Segment 0 and Segment 13s data were removed. Based on this evidence (and the results of the analysis on the other variables in the model), the
135 training set and holdout validation sets were constrained to data collected in Segments 2 10) (Training data set n = 157; 103 older/54 younger stands; Holdout data set n = 27; 16 older/11 younger stands). Future efforts will need to be undertaken to fully calibrate the models field data variables for use in Segments 0 and 13. Response curves were then calibrated using the training data sets. Variables tied to field data collections taken in the 20062008 timeframe were calibrated using only the training data gathered at undisturbed cottonwooddominated stands existing under the current regulated setting. Variables tied to geospatial data analyses were calibrated using historic mapping (1950+) made at these same undisturbed locations. In these instances, the dams and levees were in place, but floodplain development (particularly agricultural conversions) had not yet occurred. Two variables were calibrated using literature ( PROPCTW and RECRUIT ) and two other variables were calibrated using expert opinion (namely LCPI and ADJLANDUSE ). Univari ate statistics (i.e., means, medians, variance, etc.) were generated for each variable on a cover type basis based on the scales and granularities described in Table 4 4 Central tendencies and variability was examined using box a ndwhisker plots. Recognizing that statistically derived means have a tendency to be affected by their outliers, calibrations in this study were focused on medians which have been shown to be a more reliable estimate of the center of skewed distributions ( Quinn and Keough, 2002) B ootstrapping of the training data was used to generate 1,000 replications of the site medians and univariate statistics were performed on the bootstrap results to generate distributions, quartiles, st andard deviations, and means of medians. This data was then used to train the curves (i.e., establish breakpoints along
136 a continuum of function to dysfunction). Over the course of several years (20072009), the team was engaged in a recursive and reflexi ve learning process to train the curves (i.e., reality checks) based on their experiential knowledge. As the curves stabilized, the team was once again engaged to develop the aggregated model components for biotic integrity, hydrography, and spatial integrity. This effort was again recursive and reflexive as each new curve matured, the team reviewed its contribution to the whole and placed the variables into positions that seemed to make sense and offered the greatest characterization of ecosystem response based on their background knowledge and expertise. Component (CRI) testing (described below) was employed to measure the efficacy of these constructs. 126.96.36.199 Verification Verification was based on whether the model was able to distinguish between the disturbed and undisturbed communities on the river. By definition, the testing data set included a combination of cottonwood dominated communities the more degraded mesophytic dominated or woody riparian stands and a handful of disturbed cottonwood stands all located within Segments 210. Univariate statistics were performed on the various model outputs to generate descriptive characteristics of each site and box and whisker plots were developed to compare outcomes. Student t tests were run on the model outputs to test for adequacy (i.e., ability to distinguish between site conditions. 188.8.131.52 Validation u sing Independent Measures of Function (IMFs) For purposes of this study validation refer red to the multi level independent collection and comparison of data to evaluate model performance on a site by site basis. The first level of validation concentrated on the comparison of model outputs to
137 the response of two independent vegetative datasets that were collected on the sites at the same time the original data was coll ected. Referred to collectively as independent measures of function (IMFs), invasive species coverage and a Floristic Quality Index (FQI) ( Swink and Wilhelm, 1994) were shown to effectively distinguish between cottonwood dominated sites and noncottonwood sites based on univariate statistical analyses and Student tests ( ( 190 )= 3 28, < 0 05; ( 190 )= 1 38, < 0 20 respectively). Correlation and regression analyses were used to demonstrate the strength and direction of the relationships between the ranked IMFs and both the individual variable response indices system wide and composite model scores (i.e., ERIs) on the pilot segment (Segment 10). Perfect correlations [Pearson, Spearman, and Kendall Tau correlation coefficient 1.0) are rarely if ever found in actual modeling development activities ( Urdan, 2010 ) Generally speaking, correlation coefficients stay between 0.70 and +0.70 ( Quinn and Keough, 2002; Urdan, 2010 ) In their meta analysis of peer reviewed ecological papers published between 1996 and 2000, Mller and Jennions (2002 ) found that on average, correlations above +0.60 (or below 0.60) indicated strong relationships between variables. Alternatively, Cohen (1988 ) suggests a more conservative heuristic < 0.2 0 indicated weak en 0.20 and 0.50 indicated moderate relationships, > 0.5 indicated strong relationships. Statistical significance was defined at p = < 0.05 for all tests.
138 184.108.40.206 Holdout v alidation Twenty seven sites were randomly extracted from the original data sets in Se gments 210 for use in holdout validation exercise. Response indices for these holdouts were calculated and compared with the scores of the IMFs under two scenarios: 1) the holdout sites ranked on their own, and 2) the holdout sites were re embedded into t he full reference data set and reranked. Composite ecosystem response indices (combinations of variable response indices using the weighted component model formulation) were calculated for the pilot reach (Segment 10) and results were verified in a simila r manner comparing model results amongst the three stand types (undisturbed, disturbed, and noncottonwood sites) and between the training and testing sites. 220.127.116.11 Bald eagle v alidation A final validation of the ERIs on the pilot segment (Segment 10) was made using nonvegetative data, specifically spatially explicit bald eagle ( Haliaeetus leucocephalus ) nesting data provided by the USACE and USFWS. The ERIs were krig g ed across the pilot segment and the Global Morans I statistic ( Chang, 2006; Moran, 1948) was computed in a GIS using ArcGISs Spatial Analysis tools The intent was to determine whether the nests were spatially autocorrelated with the higher ERI scores, testing whether the models results reflected ecosystem integrity with respect to megafaunal species utilization. 4.5 Results 4.5.1 Final Response Indices (RI s) Calibration, verification and validation of the models RI curves was accomplished using a combination of expert opinion, literature reviews and statistical
139 analyses with the goal of developing a tool that could distinguish between functioning and non f unctioning ecosystems, and with the intent of devising a mechanism that could compare and contrast proposed recovery plans to reestablish cottonwoods on the Missouri River. Table 4 6 presents the range of univariate statistics us ed to first calibrate the individual response curves using the training data set, and Appendix B offers box and whisker plots illustrating the central tendencies of the variables in these settings. Appendix C presents the final calibrated RI curves themsel ves. The verification results using the testing data sets (undisturbed vs. disturbed, vs. mesophytic sites) showed some distinctions between RIs across site types, but these findings were not statistically significant (refer to Appendix D ). Correlation and regressions analysis in the validation tests demonstrated weakly positive but significant results (p<0.05) when comparing the RI scores to IMFs, suggesting a progression towards a satisfactory fit of the model, but indicating that no variable can be used as a standalone indicator of ecosystem integrity of the system. 4.5.2 Final Component Response Indices (CRIs) Component indices were calculated for the training and testing sites (including holdouts) and univariate statistics were generated on a segment basi s for the stud y area ( Table 4 7 ). Verification and validation of the CRI algorithms (i.e., Biotic Integrity Hydrography and Spatial Integrity ) using testing data sets produced mixed results as well ( Figu re 4 7 ). As the graphs indicate the means and medians of the three types of sites across the three components were distinctive, but the verification analyses showed only minor measureable differences, and the majority of these were not statistically si gnificant. Validations tests again showed some improvement in the models construct, but only one of the three CRIs showed significant and strong positive correlations and
140 regression response ( BIOTA ) throughout the validation exercises ( Figure 4 8 ). The majority of the validation correlations and regressions were positively responsive (although only moderately so) and the majority were significant (p<0.05) ( Table 4 8 ) indicating again that the models behavior was improving, but that that no single component could be used independently to measure ecosystem response of proposed plans with any degree of confidence. 4.5.3 Final Ecosystem Response Index (ERI) Calibration, verification and validation of the final com posite ERI model was accomplished through a combination of comparative analyses using the testing site data, the IMFs, the embedded holdout data sets, and a set of independent bald eagle nesting data provided by the USACE and USFWS. Comparative analyses of the three site types [i.e., cottonwooddominated ( CW ) vs. mesophytic dominated ( RIP ) vs. disturbed sites ( DistCW )] in the pilot Segment (Segment 10) demonstrated the models ability to satisfactorily distinguish between the three settings with a high degr ee of confidence ( CW RIP t(45) = 2.91, p <0.05); CW DistCW t(34) = 3.63, p <0.05 ) (Figure 4 9 ). Validation using regression analysis on Segment 10s testing sites (including holdouts) showed that 78% of the variance in the FQI IMF and 29% of the variance in the Exotics IMF could be explained by the model ( r2 = 0.78 and 0.29 respectively), and these results were statistically significant ( p <0.001) ( Table 4 8 ). Correlations of this same data indicated the re lationships were positive, strong, and statistically significant (ERI t(57) = 14.74, p <0.001; ERI t(57) = 3.8, p <0.001). A regression formula was written specifically for the ERI FQI analysis to determine the average margi n of error (9.1%) in the prediction of FQI scores using the ERIs on the 57
141 test sites (including holdouts) ( Appendix E ). These values are plotted in a radial diagram demonstrating the fit of the model at a 95% production level and an 0 .05 ( Figure 4 10 ). Based on the overall ERI results in Segment 10, twenty two percent of the sites had ERIs less than 0. 40 ( Figure 4 11 ). Seventy seven percent of the low scores were associated with the n oncottonwood and disturbed site types Sixty two percent of cottonwood dominated sites received scores > 0.6. As a final independent validation of the model, the ERI scores were interpolated across the areal extent of Segment 10, and known bald eagle ( Haliaeetus leucocephalus ) nests were mapped onto this surface. The Global Moran's I statistic was computed on these intersections to determine whether the nest location and high ERI scores were spatially autocorrelated ( Figure 4 12). The Morans index for the analysis was 0.613, and the expected index was 0.278. The variance was 0.015, and the z scores was 5.214 (p<0.00). This was a statically significant indication of spatial autocorrelation between the location of the nests and the scores produced by the ERI model. 4.6 Discussion The guiding principles and overarching tenets of EBM focus on the enhancement of ecosystem integrity from a systems perspective, managing the system as a whole rather than focusing on charismatic species of concern to promote resiliency and sustainability ( Dale and Beyeler, 2001; Parrish et al., 2003; Society for Ecological Restoration International, 2004 ) Under this paradigm, managers and decision makers on the Missouri River are continually challenged to assess the consequenc es of long term, system wide anthropogenic disturbances to ecosystem structure,
142 composition, and function at the landscape scale and reverse the process through thoughtful interventions. These holistic, often multi agency efforts to maintain accountability for regional ecosystem integrity while tracking the loss and compensation of key services and function, present a unique and challenging environment in which innovative approaches to ecosystem evaluation are essential to program success. To meet this cha llenge, a prescriptive spiral based modeling approach was developed that emphasized transdisciplinary teaming, active learning, and the integration of hard science with subject matter expertise. Using this spiraling approach, a multimetric ecosystem response model was developed for the Missouri Rivers cottonwood community that characterizes ecosystem conditions and assesses response to proposed recovery initiatives. The model was calibrated using a combination of multiscalar data gathered through literatur e, expert elicitations, field data collections, and geoprocessing of both extant and historical mapping of the system. Verification and validation of the models individual variable and component indices clearly indicated that no single indicator could be used as a standalone measure of system response, and that a composite index that captured not only biotic integrity, but also hydrography and spatial context was necessary to model the system in a holistic fashion. Holdout validation, independent measures of function, and an assessment of bald eagle nest data offered strong evidence that the model is relevant, reliable, and sufficient to evaluate prospective recovery plans for the cottonwoods on the river. Despite the demonstrated strengths and robustness of the model presented herein, there are a number of key limitations to the spiral modeling process in general, and the cottonwood ERI model in particular that should be acknowledged and explored
143 in future applications. First, the spiraling process was a lengthy undertaking that required participants to commit significant time and energy to the process over the course of six years. Moreover, the large number of stakeholders involved in the process made it unwieldy at times. In the future, limiting the part icipant list or breaking the team into working subgroups might improve efficiency. With regards to the model itself, it is important to recognize that it has not been fully calibrated and tested in all eight priority segments. Ecosystem recovery on the Mi ssouri is a slow process budgetary limitations forced the team to prioritize its initial efforts. The gathering of data and the modeling of hydrologic conditions on the system was a time consuming undertaking that is only now coming to fruition. Groundwater monitoring and the development of LCPI coverage across the basin are currently underway, but the results on all priority segments simply were not ready for inclusion in this study. Moreover, the results of this pilot application indicate that these hydrographic proxies are the weakest indicators of ecosystem functionality when compared against independent measures of function. An alternative strategy should be explored, possibly developing a component index that utilizes high fidelity hydraulic modeling s uch as HEC EFM2 2 to better characterize ecosystem response with regards to hydrology and geomorphology. In addition, remember that the power analysis demonstrated that the field data collected in Segments 0 and 13 were not representative of the population ( due in large part to their unique topographic characteristics and altered hydrologic setting). Segment 0 is not a priority under the BiOp ( USACE 2000, 2003), so in all likelihood recovery efforts will not be made in that segment, but the use of this model to assess any proposed http: //www.hec.usace.army.mil/software/hec efm/ (Accessed Dec 2013).
144 recovery efforts in Segment 13 will require additional study and refinement of the model to capture the unique setting in this lower reach. 4.7 Conclusions The ERI model presented here offers a reliable and holistic means of capturing ecosystem condition and measuring ecosystem response to recovery pl ans proposed for cottonwood forests within the Missouri River basin. T he results support the conceptual premise underlying the models construct and demonstrate its representative capabilities. S cientific studies characterizing the state of the cottonwood communities in the Missouri River Basin suggest an overall decline in ecosystem integrity (i.e., health, biodiversity, stability, sustainability, naturalness, etc.) ( D ixon et al., 2010; Jacobson et al., 2011; Johnson et al., 2012; National Research C ouncil, 2002; Scott et al., 2012 ) a finding the model has now irrefutably quantified in the Segment 10 pilot assessment. Furthermore, the results indicate an opportunity to redress impacts through reactiv e, active or proactive intervention ( Hobbs et al., 2011) There is great potential to reactively improve hydrography in the basin and reduce invasive species encroachment, actively manage (and potentially rehabilitate) cottonwood communities across the system, while pro actively planning for the constraints imposed by climate change implementing appropriate and sustainable activities targeting these sub functional communities. In essence, this cottonwood model is a systems construct it draws in a wide range of readily available multidimensional data, assembles these into essential ecosystem components based on logic that is scientifically informed, and synthesizes the assessment of system conditions across space and time into a single output, affording the end user an oppo rtunity to holistically evaluate conditions on the river using a transparently communicated ecosystem based
145 approach. With this model, regional planners and managers in the Missouri River basin can now reliably estimate the performance of their proposed recovery plans and adaptively co manage the system while transparently communicating their solutions to their stakeholders and the public at large.
146 Figure 4 1 Step 3 in the process focuses on ecosystem response model construction (calibration, verification and validation). Performance Measures Model Performance Step 3: Calibration Ecosystem Response Models Step 2: Mathematical Formalization Step 4: Forecasting Step 5: Alternative Evaluation Quality of The Fit Model Verification Sampling Design Reference Datasets Evaluation Datasets Fitted ValuesResponse Thresholds Adaptive Co Management Step 6: Construction and Monitoring Statistical Literature, Existing Models, Expert Contributions Predicted ValuesModel Validation Study Goals and Objectives Performance Hypotheses Site Selection via GIS Based Decision Support System Step 1: Conceptual Modeling Laboratory and Field Experiments Description Data from Literature and ExpertsModel Goals and Objectives
147 Figure 4 2 Pre and Post damming t rends in cottonwood community conditions on the Missouri River A) Historical conversions of land use since the 1890s ; B) Changes in relativ e distribution of cottonwood (hydrophytic) versus noncottonwood (mesophytic) forests on the Missouri River pre and post damming; C) Declines in recruitment for cottonwoods on the Missouri preand post damming.
148 Figure 4 3 My spiral model development paradigm offers a unique opportunity for stakeholders to actively engage in the process through reflexive team meetings that promote active learning, increasing knowledge and promoting trust and confidence in the model while hon ing the skills and competence of the modelers themselves [adapted in part from Boehm (1988); Du Toit (2005); Steiner (2009) ].
149 Figure 4 4 Study area for the cottonwood ecosystem response model. Today, the river is commonly referred to as a river of thirds ( National Research Council, 2002) while onethird of the river is still freeflowing (highlighted in blue), the second thir d of the system has been transformed into a lakes system above the six dams (highlighted in pink), and the final third of the system has been lined with flood control levees to keep floodwaters within the channel (highlighted in grey) resulting in a hydr ologic setting that is controlled by dam releases effectively disconnecting the floodplain terraces from the river. Table 4 1 A summary of the key river reaches serving as the study domain for the cottonwood model. No. Name River m iles Length (km) Area (ha) 0 Wild and Scenic River Reach 2073.4 to 1917 251.7 13,226 2 Peck Reservoir to Lake Sakakawea Headwaters 1771.3 to 1543.3 366.9 107,438 4 Garrison Dam to Lake Oahe Headwaters 1390 to 1286 167.4 43,761 6 Oahe Dam to Big Be nd Dam 1072.3 to 987.4 136.6 29,408 8 Fort Randall Dam to Running Water Bridge 880 to 841 62.8 14,475 9 Running Water Bridge to Confluence of the Niobrara 841 to 811.1 48.1 14,571 10 Gavins Point Dam to Ponca, NE 811.1 to 753.0 93.5 85,483 13 Platte Ri ver mouth to Kansas City, MO 595.5 to 365.5 370.2 171,440
150 Table 4 2 Study team affiliations. Category Participant c ount Associations Academia 17 University of South Dakota Benedictine College University of Nebraska at Omaha South Dakota State University Tribal 6 Cheyenne River Sioux Tribe Lower Brule Sioux Tribe, Fish, Wildlife and Recreati o n Omaha Tribe Pine Ridge Agency, Fish & Wildlife Rosebud Sioux Tribe/Sinte Gleska University Winnebago Tribe of Nebraska Federal 41 U.S. Army Corps of Engineers (USACE) U.S. Geological Survey (USGS) U.S. National Park Service (USNPS) U.S. Fish and Wildlife Service (USFWS) U.S. Forest Service (USFS) U. S. Environmental Protection Agency Region s 7 & 8 (USEPA) Natural Resources Conservation Service (NRCS) State 11 Iowa Department of Natural Resources Kansas Department of Wildlife and Parks Region 2 Office Missouri Department of Conservation Nebraska Forest Service Nebraska Gam e and Parks Commission North Dakota Forest Service South Dakota Division Resource Conservation South Dakota Department of Game, Fish, and Parks Local 1 Lewis and Clark Natural Resources District NGOs 3 Northern Prairies Land Trust Iza ak Walton League of America The Nature Conservancy Private 3 EA Engineering
151 Table 4 3 Forested stands in the study were discretized by age class (five classes were associated with the model but aggregated into two mod es for model construction. Older stands were defined as forests greater than 25 years of age vs. y ounger stands ranged between 2 and 25 years in age Seedlings were not modeled here. No. Age c lass Description Model a ssociation 1 >114 years old Old Growth Older 2 50 114 years old Mature Older 3 25 50 years old Intermediate Older 4 10 25 years old Pole s Younger 5 2 10 years old Sapling s Younger 6 0 years old Seedling s -
152 Table 4 4 Description of the thirteen variables ass ociated with the model, and a comparison of their scale and temporal context (i.e., calibration source dates and granularity for application). Variable Variable d escription Scale Details Herbaceous Canopy Cover (CANHERB) The percent canopy cover of plants in the herbaceous layer (herbs and woody se edlings < 1 m tall) in the plot Stand Field Data 20062008 Shrub Canopy Cover (CANSHRUB) The percent canopy cover of plants occupying the shrub layer (shrubs and tree saplings > l m tall < 10 cm dbh) in the plot Mean Coefficient of Conservatism (CVALUE) The average Coefficient of Conservatism (Mean C) scores were calculated using the following formula: = ( 1 + 2 + 3 + ) / Native Species Richness (RICHNATIVE) The total numbe r of native plant species encountered in the stand. Wetland Indicator Score (WIS) Unweighted average WIS scores (average of all WIS score types encountered at a site). = ( 1 + 2 + 3 + ) / Expert Opinion & Field Data 2006 2008 Depth to Groundwater (DEPTHGW) The elevation of ground water levels (m) averaged over the last 10 years taken from all available well data in Segment 10 (8 wells in total) Segment Well Data 1996 2006 Distance to Nearest Patch (DISTPATCH) The median distance between polygon edges of the nearest model relevant patches found on or in the near vicinity of the site GIS 1950s Patch Size (PATCHSIZE) The median size of model relevant patches found on or in the near vicinity of the site. Relative Interspersion of Habitats (INTERSPERS) The direct measure of the variety of niches (i.e., defined as individual cover types grid cells) found within a 3m x 3m (30m2) floating neighborhood window. = ( ) Sy stem GIS 18931950s Land Capability Potential Index (LCPI) A categorical index developed by the USGS ( Jacobson et al., 2008 ; Jacobson et al., 2007) GIS & Expert Opinion Level of Cottonwood Domination (PROPCTW) The relative proportion of the forest and shrub communities dominated by cottonwood species on or in the vicinity of the site Field Data Johnson (1992 ) Level of Recruitment ( R ECRUIT) The relative proportion of the forest and shr ub communities that were classified as either saplings (210 years old) or poles (10 25 years old) on or in the vicinity of the site Adjacent Land Use (ADJLANDUSE) The relative amount of human disturbance on the landscape is defined by the proportion of adjacent land use practices (weighted by levels of disturbance intensity) contributing to the character of the landscape matrix immediately surrounding the riverine core (2 km buffer) Expert Opinion & GIS 18931950s
153 Table 4 5 Component response i ndex (CRI) algorithms for the model Note that the older and younger models differ only in their characterization of Hydrography (HYDRO), but note that each models ecosystem response curves will be calibrated using data only from t heir respective age classes. Variables codes correspond to those in Table 4 4 Model c omponents Age class distinctions Younger s tands (2 25 yrs old) Older s tands (>25 yrs old) Hydrography (HYDRO) + Biotic Integrity (BIOTA) ( + + ) + ( + ) + + Spatial Integrity (SPATIAL) ( + ) + + +
154 Figure 4 5 An example of the cover type mappin g conducted for the study. Here, Segment 10 ( RM 811.1 to RM 753.0Gavins Point Dam to Ponca, NE ) demonstrates the significant conversion of natural communities along the river to agricultural croplands and other land uses and vegetative coverages (i.e., sh ifts from hydrophytic forested communities) between preand post impoundment timeframes (1890s 2006).
155 Figure 4 6 An example of the power analysis used to reduce the training data sets (variable displayed = CVALUE)
156 Table 4 6 Descriptive statistics used to calibrate the individual variables in the ERI model. Variables are sorted by scale and granularity (S td = Stand, Seg = Segment, and Sys = System). Variable c ode Unit Calibration source Scale T emporal calibration s ource Segment(s) Mean Standard d eviation Minimum Maximum Count First q uartile Third quartile Median Bootstrapped mean of m edian (n = 1,000) Standard error of bootstrap Standard deviation of b ootstrap CANHERB 1 Percent Field Data Std 2006 2008 2 10 36 19 4 86 43 22.5 47.0 29 30 0.12 3.72 CANSHRUB 1 Percent Field Data Std 2006 2008 2 10 10 13 0 59 43 0.0 16.5 3 5 0.07 2.31 CVALUE Index Scaled 1 5 Field Data Std 20062008 2 10 3 0.6 1 4 130 2.3 3.1 3 3 0.00 0.06 RICHNATIVE Count Field Data Std 2006 2008 2 10 25 9.1 8 54 130 18.0 30.8 24 24 0.05 1.43 WIS OLDER Category Scaled 1 5 Expert Opinion Std 20062008 2 10 2 0.3 2 3 87 2.3 2.5 2 2 0.00 0.04 WIS YOUNGER Category Scaled 1 5 Expert Opinion Std 20062008 2 10 3 0.4 2 3 43 2.4 3.0 3 3 0.00 0.10 1These variables were calibrated with data from younger stands (2-25 yrs old).
157 Table 4 6 Continued Variable c ode Unit Calibration source Scale Temporal calibration s ource Segment(s) Mean Standard d eviation Minimum Maximum Count First q uartile Third quartile Median Bootstrapped mean of m edian (n = 1,000) Standard error of bootstrap Standard deviation of b ootstrap DEPTHGW Meters Well Data Seg 1996 2008 10 3 4.3 0 69 12,331 3.0 4.0 3 3 0.00 0.13 DISTPATCH Meters GIS Seg 1950 0 13 5.8 2 24 122 8.3 17.5 13 13 0.02 0.57 2 19 3.9 8 24 69 16.0 22.0 19 19 0.02 0.62 4 15 6.0 1 24 77 10.9 20.3 16 16 0.03 0.99 6 13 6.6 1 23 42 7.7 19.0 14 14 0.06 1.96 8 11 8.1 1 24 21 2.1 17.1 12 12 0.12 3.90 9 16 3.2 12 20 6 13.5 17.7 17 16 0.07 2.35 10 19 3.8 7 24 34 16.6 21.9 19 19 0.03 0.84 13 17 5.9 1 24 161 13.3 21.0 19 18 0.02 0.79 PATCHSIZE Acres GIS Seg 1950 0 67 41.4 32 202 54 38.1 81.4 49 50 0.13 3.97 2 352 708.3 32 6,952 204 55.6 349.5 112 114 0.39 12.21 4 563 747.0 32 3,164 77 63.6 831.4 171 175 1.47 46.40 6 309 324.5 33 1,453 57 64.2 423.0 217 209 2.01 63.60 8 213 247.8 32 1,130 39 58.7 281.4 119 124 0.73 23.01 9 274 2 57.7 42 1,014 22 78.1 428.6 172 192 2.37 75.06 10 329 454.0 31 1,928 72 51.4 380.6 108 109 0.49 15.37 13 357 698.1 31 4,660 200 50.1 302.8 77 82 0.33 10.45
158 Table 4 6 Continued Variable c ode Unit Calibration sour ce Scale Temporal calibration s ource Segment(s) Mean Standard d eviation Minimum Maximum Count First q uartile Third quartile Median Bootstrapped mean of m edian (n = 1,000) Standard error of bootstrap Standard deviation of b ootstrap INTERSPERS Index Scale d 0 -1 GIS Sys 1893 0 13 0.502 0.001 0.500 0.503 8 0.501 0.502 0.501 0.502 0.000 0.000 1950 0 13 0.253 0.004 0.250 0.263 8 0.251 0.252 0.251 0.251 0.000 0.000 2006 2008 0 13 0.203 0.004 0.200 0.212 8 0.201 0.203 0.202 0.202 0.000 0.000 LCPI OLDER2 Category Scaled 1 -16 Expert Opinion Sys 2006 10 0.81 0.2 0.30 1.00 11,678 0.75 0.95 0.85 0.85 0.00 0.00 LCPI YOUNGER 2 Category Scaled 1 16 Expert Opinion Sys 2006 10 0.62 0.2 0.20 1.00 2,270 0.45 0.75 0.65 0.65 0 .00 0.00 PROPCTW 3 Percent Lit. Sys 1893 0 13 1.000 0.000 1.00 1.00 859 --1.00 ---1950 0 -13 0.940 0.225 0.00 1.00 3,287 --1.00 ---2006 2008 0 -13 0.836 0.352 0.00 1.00 3,918 --1.00 ---RECRUIT 3 Percent Lit. Sys 1893 0 13 0.963 0.186 0.00 1.00 859 --1.00 ---1950 0 13 0.276 0.414 0.00 1.00 3,287 --0.00 ---2006 2008 0 13 0.225 0.401 0.00 1.00 3,918 --0.00 ---ADJLANDUSE 4 Categor y Scaled 0 -1 Expert Opinion Sys 1893 0 13 0.851 0.074 0.710 0.948 8 0.830 0.910 0.836 0.849 0.001 0.025 1950 0 13 0.728 0.053 0.635 0.789 8 0.688 0.762 0.753 0.740 0.001 0.025 2006 2008 0 13 0.724 0.121 0.585 0.975 8 0.664 0.744 0.721 0.714 0.001 0.033 2Data shown here were not used to calibrate the variable -an expert elicitation exercise was utilized instead. 3Data shown here were not used to calibrate the variablepeer -reviewed literature was utilized instead (Johnson, 1992). 4Data shown here were not used to calibrate the variableexpert consensus was utilized to establish l and use categories.
159 Table 4 7 Results of the univariate analyses focused on the CRIs and ERIs calculated on a segment by segment basis for the training and testing sites (including holdouts). BIOTA SPATIAL HYDRO ERI Segme nt Site t ype Sample Size Mean Standard d eviation Median Mean Standard d eviation Median Mean Standard d eviation Median Mean Standard d eviation Median 2 CW 30 0.50 0.13 0.51 0.47 0.13 0.43 ------2 RIP 15 0.68 0.13 0.71 0.39 0.06 0.41 -----2 DistCW -------------4 CW 35 0.62 0.20 0.68 0.45 0.13 0.42 ------4 RIP 6 0.53 0.21 0.55 0.54 0.10 0.55 ------4 DistCW 10 0.64 0.17 0.69 0.40 0.08 0.42 ------6 CW 17 0.47 0.10 0.48 0.37 0.14 0.40 ------6 RIP 11 0.32 0.08 0.34 0.44 0.12 0.41 ------6 DistCW 4 0.36 0.03 0.37 0.39 0.21 0.45 ------8 CW 34 0.59 0.16 0.61 0.46 0.12 0.47 ------8 RIP 12 0.44 0.19 0.49 0.47 0.19 0.54 ------8 DistCW 4 0.46 0.19 0.37 0.38 0.14 0.40 ------9 CW 7 0.54 0.12 0.53 0.54 0.16 0.54 ------9 RIP -------------9 DistCW 1 0.47 0.47 0.47 0.25 0.25 0.25 ------10 CW 34 0.65 0.15 0.71 0.45 0.14 0.45 0.70 0.17 0.75 0.62 0.08 0.64 10 RIP 18 0.55 0.19 0.59 0.41 0.14 0.42 0.71 0.14 0.75 0.55 0.10 0.57 10 DistCW 7 0.44 0.17 0.40 0.43 0.16 0.38 0.68 0.14 0.73 0.49 0.09 0.46
160 Figure 4 7 A comparison of CRI scores for the testing sites (ex cluding holdouts) discretized by site type ( CW = cottonwooddominated sites; RIP = mesophytic riparian sites; DistCW = disturbed cottonwood sites). Red lines indicate means and black lines inside the boxes indicate medians ( actual value shown to the right of the graphs)
161 Figure 4 8 Predicted vs. actual scores generated in the holdout validation sites (alone) by the BIOTA CRI scores for the dependent variable FQI A ctual FQI scores are shown as triangles and predicted FQI scores (predicted based on the relationship between the CRI and FQI ) are shown as circles. The colors are coordinated one triangle + one circle => the sites actual vs. predicted FQI value based on the CRI algorithm.
162 Table 4 8 Results of validation analyses focused on correlations and regression analyses of the CRI and ERI scores and the independent measures of function ( Exotics and FQI ) for full suite of testing sites ( including holdouts) Dependent v ariable Independent v ariable df t -Test Spearmans Pearson's Kendall's p -value Significant? R R2 Adjusted R2 p -value Significant? Exotics BIOTA 243 11.39 0.59 0.55 0.43 p<0.001 Yes 0.57 0.33 0.32 p<0.001 Yes SPATIAL 243 1.57 0.1 0.11 0.07 p<0.20 0 No 0.15 0.02 0.02 p <0.02 0 Yes HYDRO 57 0.61 0. 08 0.14 0.06 p>0.20 0 No 0.07 0 0.01 p>0.63 0 No FQI BIOTA 243 30.43 0.89 0.85 0.71 p<0.001 Yes 0.86 0.75 0.75 p<0.001 Yes SPATIAL 243 2.2 0.14 0.2 0.09 p<0.05 0 Yes 0.13 0.02 0.01 p<0.05 0 Yes HYDRO 57 0.84 0.11 0.2 0.08 p>0.20 0 No 0.07 0 0.01 p>0.62 0 No Exotics ERI 57 3.8 0.45 0.04 0.33 p<0.001 Yes 0.54 0.3 0.28 p<0.001 Yes FQI ERI 57 14.74 0.89 0.86 0.71 p<0.001 Yes 0.87 0.75 0.75 p<0.001 Yes
163 Figure 4 9 Comparison of central tendencies of the testing data sets revealed measureable and significant differences between site types using the ERI model ( CW = cottonwooddominated sites ; RIP = mesophytic riparian sites ; DistCW = d isturbed sites ) Red lines indicate data means and bl ack lines inside the boxes indicate data medians.
164 Figure 4 10. Predicted v ersus actual scores generated by the model for the dependent variable FQI using the entire testing data set (including holdouts) The actual values are plotted in blue for each site, and the predicted values are plotted in solid red. The dotted and hashed lines indicate the upper and lower bounds of the analysis [ ] Holdout validation sites are highlighted in grey.
165 Fi gure 4 11. Relative contribution of the three site types to overall ERIs scores in the testing sites located in Segment 10. Note that the disturbed ( DistCW ) and mesophytic riparian ( RIP ) sites have lower ERI scores (<0.54) than the cottonwooddominated sites
166 Figure 4 12. An overlay of the interpolated ERI scores and bald eagle nests located in Segment 10 (shown as a blue boundary) The results of the Global Morans I analysis are presented in the upper right hand corner. The calculated index was 0.613 suggesting the nests are clustered around the high scores generated by the ERI model.
167 CHAPTER 5 C OTTONWOOD RECOVERY INTEGRATED SITE IDENTIFICATION SYSTEM (CRISIS): A GIS BASED PATICIPATORY DECISION SUPPORT SY STEM FOR THE MISSOUR I RIVER 5.1 Introduction Over the last century, the traditional command and control w ater resource management paradigms regulating large rivers across the country have produced valuable ecosystem goods and services for the nation (i.e., f lood protection, hydroelectric power, water storage for irrigation, etc.) but these gains have come at a significant ecological cost (i.e., the system wide loss of biodiversity and a significant decline ecosystem structure and function on a landscape scal e) ( Holling and Meffe, 1996; Light et al., 2013; Poff et al., 20 03; World Commission on Dams, 2000) Mandates and directives to mitigate and recover these failing ecosystems have shifted toward more holistic ecosystem based paradigm sfocused on the sustainable production of benefits that promote human well being, attained through the strategic recovery of ecosystem integrity (i.e., function, complexity, and adaptive capacity) on a watershed scale ( Beattie, 1996; White House Council on Environmental Quality, 2009, 2012) Across the country, n atural resource managers now face a significant challenge where to act first? (Orsi et al., 2011 page 337). With limited budgets, and a desire to make the best use of these limited resources planners and managers must strategically position mitigation sites on the landscape in a manner that maximizes benefits following Ecosystem Based Management ( EBM ) doctrine, balanc ing the interests of multiple stakeholders in a transparent, fair, and productive manner ( McLeod and Leslie, 2009) In reviewing various mitigation projects, efforts tend to fail for two
168 reasons: 1) poor or inappropriate structural designs and, 2) the improper location of the site relative to other landscape features ( van Lonkhuyzen et al., 2004) In other words, watershed context matters, and thus a s cientifically robust lan dscape scale approach is needed to mitigate the risks and ensure a reasonable return on investment. In order to improve ecosystem resiliency assuring long term sustainability the task must be conducted from a systems perspective ( Gharajedaghi, 2011) focusing on the recover y of the system as a whole rather than addressing the individual requirements of a few charismatic species of concern ( Benson, 2012 ) Moreover, the solution must be both spatially cognizant and politically sensitive taking into account surrounding land use activities, future trends in land use conversions, hydrologic regimes and variability, biological dynamics, and socioeconomic factors that combine to determine the overall productivity or suitability of potential mitigation sites ( National Research Council, 2001) 5.2 Goals and Objectives Working from the same structured decision making paradigm described in previous chapters, an approach must now be devised to position mitigation sites on the landscape to maximize return on investment ( Figure 5 1 ). This step often happens simultaneously with the es tablishment of ecosystem response model s, is often influenced by this activity, and must be tied to the same goals and objectives driving the conceptualization of the system in order to generate useful results. The goal therefore of this chapter is to pres ent a recursive and reflective (spiral based) approach that generates spatially explicit criteria using a Geographic Information System (GIS) and Multi Criteria Decision Analysis (MCDA) in a participatory manner. The approach described here can be applied at a landscape or watershed scale to identify, prioritize and select mitigation sites based on stakeholder values and preferences. A case study
169 on the Missouri River demonstrates the efficacy of this approach, generating suitability maps that reveal hot s pots or areas that balance the perceived opportunities and constraints governing the watersheds recovery plans for the plains cottonwood community ( Populus detltoides W. Bartram Marsh. S ubsp. m onilifera (Ait on ) Eckenwalder ) Coined, the Cottonwood Recovery Integrated Site Identification System (CRISIS) the final product integrates spatial analysis with the power of MCDA to identify, map and sieve spatially explicit siting criteria using a participatory GIS based approach. This chapter describes the development and evolution of CRISIS and discusses how the approach can be used to inform EBM activities and aid in the recovery of ecosystem integrity on large, regulated river systems in general, and on the Missouri in particular. The chapter is divided into three primary sections. The first section offers background describing the problems impacting the cottonwood ecosystem on the Missouri River, and offers a review of useful techniques (i.e., GIS based participatory approaches and sieve mapping) that can be used to strike a balance between competing interests, positioning mitigation sites on the landscape in fair and productive manner. In the methods section, a spiraling (recursive and reflexive) approach is presented that can guide managers through the identification, selection, and prioritization of suitability criteria for site positioning purposes. In the results section, CRISIS products are presented (i.e., criteria transformations, weighting matrices, and hotspot maps), and a discussion of possible ut ilization strategies is offered to describe how CRISIS can assist the regional planners and managers in the recovery of the
170 rivers cottonwood ecosystem. To conclude, lessons learned and future research opportunities are described. 5.3 Background and Literatur e Review 5.3.1 Cottonwood Recovery on the Missouri River In the mid 1900s, the USACE constructed six dams on the mainstem of the Missouri River for the purposes of flood control, navigation, hydroelectric power generation, and water storage for irrigation in an attempt to bring stability and prosperity to the region in the midst of the Great Depression. The ecological ramifications of these activities continue today ( National Research Council, 2002 ) : Nearly 3 million acres of riverine and riparian ecosystems have been altered through landuse changes, inundation, channelization, and levee building; The amplitude and frequency of the rivers natural peak flows have been dramatically reduced and with it sediment transport has been dramatically reduced; Cottonwood forest reproduction (historically the most abundant and ecologically significant species on the rivers extensive floodplain) has largely ceased ; and Several species (the least tern, piping plover, and pallid sturgeon) have been placed on the federal Endangered Species List. In 2000, in response to a proposal by the USACE to undertake additional f lood control actions, the U.S. Fish and Wildlife Service issued a Biological Opinion (BiOp) directing the USACE to conduct collaborative, longterm planning efforts to restore critical ecosystem functions, mitigate for habitat losses, and recover least ter n, piping plover, and pallid sturgeon populations, while seeking to enhance social, economic, and cultural values for future generations along the Missouri River ( USFWS, 2000 ) In 2003, the USFWS amended the BiOp to include reasonable and prudent measures to recover ecosystem integrity of cottonwood forested communities on the river to minimize the
171 take on bald eagles owing to their threatened status under Section 7 of the Endangered Species Act of 1973 (ESA) ( USACE 2003 ). In 2010, the USACE developed a Cottonwood Management Plan (CM P) ( USACE, 2010 ) to address the reasonable and prudent measures mandated in the 2003 BiOp with the express intent ion of providing a single, comprehensive strategy to guide the efficient and effective recovery of critical cottonwood communit ies in the Missouri River Basin. The CMP specifically called for the development and utilization of standardiz ed, multiscalar assessment procedures that promoted collaboration and offered scientifically defensible and adaptive recovery solutions The plan focused on three initial steps: 1. Map the extent of existing and historic cottonwood communities on the river; 2. Develop a multimetric ecosystem response model to evaluate proposed recovery plans, and 3. Explore and engage stakeholders in the development and implementation of long term collaborative monitoring strategies to support adaptive comanagement of the cottonw ood ecosystem basinwide. Once these tasks were well underway, a small committee of local experts from the CMP study team were convened and asked to identify locations on the river that could serve as pilot sites for the initial recovery efforts. These sit es were then presented to the entire CMP study team in a workshop held in Vermillion, SD in the winter of 2008. Questions regarding how the experts located the sites (i.e., what criteria they used to narrow their focus what decisionmaking heuristics they used to screen potential sites, and even how they drew the boundaries around the sites ) were voiced. Although they did not necessarily disagree with the experts choices, the CMP team decided that a more structured approach was needed one that captured the same logic and rationale
172 the experts were using to identify sites but one that operated in a transparent, repeatable fashion, integrating the inclusion of multiple criteria and numerous, opposing stakeholders value preferences. 5.3.2 Sieve Mapping In these situations, many managers have turned to S patial D ecision Support S ystems ( S DSS) to sort through the available data, characteriz e the system and communicate relevant information in a visually engaging medium to inform the decisionmaking process ( Ferretti and Pomarico, 2013; Malczewski, 2006 ; van Lonkhuyzen et al., 2004; Wang et al., 2012) More specifically, a commonly used GIS based processing technique referred to as sieve mapping ( Keeble, 19 52) allows participants to assess the value of an areas contribution (land availability, ecosystem integrity, land use conflicts, etc.) toward attainment of overall goals and objectives fulfilling the project purpose (e.g., locating recovery sites for c ottonwoods). Using this technique, e ach constraint or opportunity (i.e., criterion) is mapped as a rasterized sieve, and the area of concern is passed through the sieves systematically in a definitive sequence to reveal areas suitable for the intended use ( Figure 5 2 ) Throughout the process, stakeholders are encouraged to participate in the analysis selecting indicators, scoring their outputs (i.e., transforming each layer to a standardized scale ), and weighting their cont ributions to the overall suitability map based on their valueladen preferences. The specific term value laden refers to the notion that the experiences, perceptions and motivations of all those involved in the process can: 1) shape the way the studys q uestions are framed; 2) affect the selection, interpretation, and rejection of data; and 3) influence the methodologies deployed to generate the final answer ( Kloprogge et al.,
173 2011) The challenge is to conduct the sieve mapping in a structured, scientifically defensible manner to ensure transparency and instill confidence in the outcome. 5.3.3 Multi C riteria Decision Analysis (MCDA) Using participatory approaches to develop and preferentially value suitability criteria to position mitigation sites on the landscape inevitably generates multiple answers to singular questions of recovery priorities. One solution is to deploy MCDA to collapse these multiple value preferences into a single coherent measure of recovery potential. MCDA consists of a series of techniques (i.e., weighted summation, concordance, analysis, etc.) that facilitate the scoring, ranking, or weighting of decisionmaking criteria based on participants value based preferences ( Higgs, 2006; Kiker et al., 2005 ) These techniques originated over four decades ago in the fields of mathematics and operations research and have been well developed and documented particularly with respect to their use in combination with GIS based spatial decision support systems ( Malczewski, 1999, 2004, 2006 ) Riabecke et al. (2012) offer an extensive review of the most common rank based MCDA procedures used to generate the singular answers to multiple elicitation responses including: 1) rank order, 2) semantic categories, and 3) intervals established using upper and lower bounds, and indicate that the former approach is appropriate when handling ordinal (categorical) data. Rank ordering is a process wherein experts are first asked to value the different ordinal criteria based on their value laden preferences These responses are then translated into proxy cardinal weights consistent with the supplied rankings using various weighting strategies (i.e., rank sum, reciprocal, or centroid). As a final step in the process, the result of the rank sum weighting are nor malized to generat e ratios based on the principle of maximizing the expected values
174 ( Malczewski, 1999 ; Riabecke et al., 2012 ) Notably, empirical studies conducted by Stillwell et al. (1981 ) demonstrated that rank sum weightings perform well in most situations, and Riabecke et al. (2012 ) suggested that this approach was less cognitively demanding and more advantageous when engaging large, disparate groups of stakeholders. 5.3.4 Spiral Framework When attempting to elicit criteria from multiple stakeholders, and merging these into a single vision of recovery potential on a landscape scale using sieve mapping, it can be useful to adopt a spiraling approach one that encourages the continuous refinement of incremental versions of the criteria in a cyclical, recursive fashion that adjusts to shifting pr iorities, evolving knowledge, and novel and intermittent data emergence ( Boehm, 1988; Boehm et al., 2012) Under this paradigm, each cycle of the spiral u tilizes face to face meetings with the team to develop and review existing criteria and reflect on the previous decisions governing their inclusion interjecting new information into the process to hone or refine their definition and behavior in the overal l suitability assessment in an incremental fashion. Unlike the traditional waterfall model ( Royce, 1970) spiral approaches allow cr iteria to be added to the list when they become available or are revealed as new indicators of recovery potential In spiral modeling, not only are the criteria fine tuned with feedback spiral by spiral, the team is asked to revisit their goals, objectives and constraints fine tuning the overall suitability analysis P lanning of the next spiral at the end of each cycle encourages forward momentum toward a final conclusion, but it is acknowledged from the onset that the final product is a working version t hat can be adjusted to meet future needs when
175 necessary This working classification is synonymous with the concepts of monitoring and adaptive management under the EBM paradigm. 5.4 M ethods A collaborative spiraling approach was used to develop and demonstrat e CRISIS on Segment 10 ( Figure 5 3 ) The process was intended to be an exercise in reflexive learning in context a term coined by Du Toit ( 2005, page 229) to describe an interactive group exercise that encourages the stakeholders to identify problems, deliberate, propose solutions and respond to contextual changes in recursive reflection cycles (centered around information presented at each workshop/web meeting) and organized within a spiral framework ( Boehm, 1988; Du Toit, 2005) The first spiral focused on verbalizing the problem and study constraints, setting goals and objectives, convening a team of 15 subject matter experts (both from the academic and the natural resource management arenas). The second spi ral focused on generating site selection criteria and both acquiring and geoprocessing data to support the spatial expression of each criterion. The third spiral concentrated on eliciting value preferences from the team regarding the importance of each cri terion in determining site suitability. The final spiral utilized GIS based sieve mapping to screen the criteria and generate heat maps (i.e., maps indicating hotspots or locations where suitability was maximized based on the weighted criteria. Each spiral took on average six to twelve months to complete beginning in 2008 and concluding in 2011. At critical junctures all along the spiral paths, the team was engaged through onsite workshops or remote web meetings /telecoms to review the state of the criteria production, reflecting on decisions made in the last meetings as well as learning and adapting the suitability map to address new challenges as they arose.
176 5.4.1 Study Area The study area, designated in the BiOp as Segment 10 ( USFWS, 2003 ) begins at River Mile (RM) 811.1 at Gavins Point Dam near Yankton, South Dakota, and extends downstream to RM 753 at Ponca State Park in northeast Nebraska, U SA (855 km2) (Figure 5 4 ). As the figure shows, Segment 10 is the downstream section of the Missouri National Recreational River (MNRR) referred to commonly as the 59mile Segment and administered by the U.S. National Park Servi ce ( USACE, 2010) Segment 10 is bounded by high bluffs to the north in Nebraska and to the sou th in South Dakota. Its valley floor is exceptionally wide extending over 16 km (10 miles) in some locations. Although seventeen separate land use land cover classes have been identified and mapped in the reach totaling some 85,482 hectares, seventy seven percent of these are under cultivation (65,726 ha) and only 7% of the area is currently classified as cottonwood forest and shrub lands (6,377 ha). More importantly, several studies have documented a decline in these stands and a shift toward more mesophyt ic riparian conditions in the next generation ( Dixon et al., 2012; Johnson et al., 2012; Scott et al., 2012 ) 5.4.2 Study Team and Workshop Format Fifteen members of the cottonwood management team volunteered to participate in the development and application of CRISIS ( Table 5 1 ). In general, their backgrounds spanned the gamut of disciplines ranging from biological sciences, to hydrology, geology, planning and management, and engineering. Their job descriptions were characterized loosely as planners, regulators, natural resource managers, professor s, geographers, and consultants. Their level of familiarity with the system ranged from
177 those with less than 5 years working on the system to individuals with well over 20 years of experience on the river. T hree intensive week long onsite workshops were u sed to create, calibrate and apply CRISIS for the study area. Each workshop began with a review of the decisions in the previous meetings and a discussion of the effects these decisions had on the CRISIS development thus far. The agenda then turned to the presentation of any new data collected, and consensus building activities were used to incorporate this new information into the development and fine tuning of the tool Two separate note takers were used to produce both informal notes and official minutes to document the decisions in each meeting. Value preferences for the MCDA were obtained through f ormalized elicitation (i.e., via elicitation spreadsheets ) but selection of criteria and standardization of these factors was conducted in a group interactiv e manner that encouraged open discussion and debate resulting in a verbal group consensus. In between the workshops, monthly 2 hour teleconferences were used to facilitate data transfer and formulate paths forward to close any knowledge gaps. These decisions and activities were recursive results and decisions were revisited at least twice in follow on meetings and teleconferences to facilitate course corrections and assure that in the end, the suitability maps conformed to the teams perceptions of cottonwood recovery priorities on the river 5.4.3 Rationale for Criteria Selection In the first workshop, the team questioned the original experts who had been initially tasked with developing the original list of mitigation sites, decomposing and exploring the logic they used to solve the problem. Major themes were revealed when the experts were asked to: 1) describe the site selection problem in their own words, 2)
178 compare their approach to others in their experience, 3) associate evidence of site suitability with ec osystem integrity and indicators of sustainability and resilience, 4) analyze the degree to which their criteria were measureable and meaningful, 5) brainstorm additional criteria, and 6) give reasons why some criteria might be more productive than others. When these answers produced useful information, criteria were altered, removed, or added to reflect the revised problem space. For example, the team initially considered using site size as a criterion under the assumption that some minimum size indicator could be used to focus recovery efforts on only large sites that would provide significant returns on investment. However, after much debate about what threshold would be used to standardize this criterion, the team ultimately came to the conclusion that size was not a meaningful indicator of suitability even small sites offered productive contributions to the overall recovery of the cottonwood communities on the river. In other words, the adaptive and recursive nature of the spiraling process captured t he experts ever evolving perceptions of the fundamental processes governing site suitability for the recovery plans in a reflexive manner Based on th is discussion, the team decided that positioning mitigation sites in the watershed based on three major t hematic foci would yield resilient, responsive, yet tolerable solutions that would promote watershedlevel thinking (i .e., sustainable offsets from a systems perspective ) ( Figure 5 5 ): H ydrologic V iability defined as the feasibi lity, practicality, and capability of a mitigation site to promote cottonwood recruitment and establishment based on hydrologic regime, these factors would promote ecosystem resiliency while balancing competing ecosystem goods and service interests (i.e., flood protection and water supplies for irrigation) ;
179 E cologic I ntegrity described as the ability of an ecosystem to support and maintain a cottonwood community comparable to those of reference standard conditions in the region. A system has ecological i ntegrity when its dominant ecological characteristics (e.g., elements of composition, structure, function, and ecological processes) occur within their natural ranges of variation and can withstand and recover from most perturbations imposed by natural env ironmental dynamics or human disruptions; and Political Necessitypractically speaking, any recovery plans targeting cottonwoods on the river under the CMP ( USACE, 2010 ) would need to be sensitive to other ongoing efforts mandated under the BiOp (e.g., piping plover, least tern, and pallid sturgeon) ( USFWS, 2000 ) as well as the interests of the local communities The team then identified t en separate criteria to capture the essence of each theme ( Figure 5 6 ). Some criteria were linked to desirable states (positive correlations) while others were used as exclusionary constraints (negative correlations). As the figure denotes, a simple rulebased heuristic was deployed to organize the ten criteria around the three themes : 1. Criteria that assessed mitigation site potential based on the depth of the groundwater table, the hydrologic regime, and the avoidance of highly erodible areas were associated with Hydrologic Viability 2. Criteria that assessed the potential f or sites to naturally recruit and establish cottonwoods (i.e., near existing sites and existing seed sources in particular) and reduce fragmentation by increasing connectivity but near existing hot spots and/or areas that were threatened by urban conver sions so in need of protection, were associated with Ecologic Integrity and 3. Criteria that assessed the practical potential of mitigation sites to provide complimentary solutions (i.e., options that avoided nesting piping plovers and least terns, but supp lemented the efforts to restore backwater habitats for pallid sturgeon habitat) and were owned by landowners inside the Missouri National Recreational River ( MNRR) boundary were associated with Political Necessity Table 5 2 offer s rationales supporting the inclusion of each criteri on in the assessment and presents initial data sources for each layer
180 5.4.4 Geoprocessing GIS based transformations of the spatially explicit data characterizing the criteria were standardized through the pr ocess of reclassification to generate raster based maps that captured the conditions of the watershed using expert opinion obtained through directed dialogues on the last days of the first workshop. All shapefiles were re projected to conform to a standard georeference (i.e., NAD 1983 UTM Zone 14N, Transverse Mercator GCS North American 1983). Metric units were used for all distance measurements, and a minimum mapping unit of one acre was established based on the original vector based data. A cell size of 10 m2 (onethird of the minimum mapping unit) was established for all grid calculations to assure overlapping coverage in the raster based analyses. Geoprocessing of the individual raw data under pining each criterion followed a relatively straightforward tact. In most instances, vector data was acquired from the source and rasterized using a 10m2 cell grid. Map overlays, buffering, distance mapping, reclassifications and spatial querying were used to transform the data, with the ultimate goal of standardiz ing the results to present findings on a 1 to 5 scale. This process was repeated in various combinations for each criterion note that the sources for each criteria varied, and as such, the preand post geoprocessing varied in relation to the situation. Exclusionary masks were used where necessary to constrain the analysis (i.e., in/out data for the MNRR boundary, exclusion zones for the tern and plover nesting sites, etc.). ESRI ArcMap 10.1 was used to perform the majority of the vector based geoprocessi ng, and the Spatial Analyst toolbox was used for raster processing. All GIS derived data was exported to MS Excel using XTools, and descriptive statistics,
181 regressions, and correlations were completed using MS Excel Central tendencies were plotted using S igma Plot 10. 5.4.5 Multi c riteria Elicitation and Analysis In the second workshop, the team was presented with a series of maps (one per criterion) detailing the results of the geoprocessing and standardization of thematic coverages. This allowed the participants to familiarize themselves once again with the criteria, and offered a last chance to debate and discuss the merits of each criterion in a mediated manner. Once the discussion closed, a formalized voting instrument was handed out polling the participants on their preferences with regards to criteria importance. The following paraphrased question was posed to the participants to initiate the elicitation process : U se your expertise to rank the criteria presented here based on what you feel should be the i r contribution to an overall site suitability index characterizing the potential for proposed sites to support the cottonwood recovery initiatives detailed in the Cottonwood Management Plan. The participants were asked to rank the criteria on a scale of 10 to 1 in terms mitigation site suitability (where 10 represented the most important factor and 1 represented the least important factor ) in the spreadsheet The participants were discouraged from halving or double assigning ranks in other words, uniq ue scores were to be assigned to each of the 10 criteria. When completed, the participants turned in their polls for review and compilation. Confidentiality was preserved throughout the process, and follow on phone interviews were conducted to assure quali ty control and to debias the results when inconsistencies attributed to both the Anchoring and Inconsistency biases were observed ( Meyer and Booker, 2001) Spearmans rank correlations were calculated to compare the results across experts. Statistical significance was defined at
182 p = < 0.05 for all tests. In addi tion, experts were clustered by affiliation (i.e., academia, federal, state, local, etc.) and results were analyzed for trends. The responses from the experts were then combined using weighted rank sum s ( Malczewski, 1999 ) : i= j+ 1 k+ 1 ( 5 1 ) where wi is the normalized weight for the j th criterion, n is the number of criteria under consideration (i.e., 10), ( k = 1,2,,n ), and rj is the rank position of the sum of all weights, that is k+ 1 Descriptive statistics and box andwhisker plots were generated on a criteriaby criteria basis using MS Excel to search for trends. Central tendencies were plotted using Sigma Plot 10. 5.4.6 Sieve Mapping The ten rasterized criteria maps (standardized on a scale 1 to 5) were combined in a GIS and the ESRI Spatial Analyst Raster Calculator tool was used to apply weights and sieve the various criteria using the algorithm : = ( 5 2 ) where Si is the suitability index score for the ith 10m2 grid cell which is equal to the sum of the values of the ith 10m2 grid cells criteria ( ci j where j = 1, 2, 3, . .10) multiplied by their normalized weights ( wj) to generate a final composite map highlighting the areas considered m ost suitable based on these inputs. A final reclassification was utilized to normalize the outputs on a scale of 1 to 5 using natural breaks in the data. The process was repeated for comparative purposes using a nonweighted assumption ( i.e., all criteria contribute equally to the final suitability assessment ) to demonstrate the differences in results if expert opinions were not taken into account.
183 The resultant hotspot maps were compared using the ESRI Spatial Analyst Cut Fill tool The weighted hotspot map was overlayed with the original site selections for comparative purposes. All data was exported to MS Excel for descriptive statistics and comparative analys e s. 5.5 Results 5.5.1 Criteria Transformation and Standardization Geoprocessing (map overlays, spati al querying, reclassifications, buffering and distance mapping, etc.) were used to transform raw, vector based data into standardized (raster based) presentation of the criteria using a standard scale of 1 to 5 where 1 represented the least suitable cond ition and 5 represented the most suitable condition ( Figure 5 7 ). The relationships between the data and the standardized scale are presented in bar charts at the bottom right hand corner of each map. The criteria have been sort ed and presented based on their association with the overarching themes, namely Hydrologic Viability Ecologic Integrity and Political Necessity 5.5.2 Expert Preferences The value of each criterion contributed toward and overall site potential or suitability s core was derived through the formalized expert elicitation process (i.e., blind balloting) in the second workshop. The results of this polling are presented in ( Figure 5 8 ) Demographics have been masked in this table (to assure c onfidentiality) but it is interesting to note that experts C D, and N were members of the original site selection committee, and even they had varying judgment s as to the value of criteria in identifying potential mitigation sites for the study (columns lined in blue). T he variety of participant responses indicat ed some degree of agreement when assessed using Spearmans rank correlations ( Table 5 3 ) F ive of the fifteen participants were strongly
184 correlated with the weighted aver age rank sums ( > 0. 73) and these findings were shown to be statistically significant at the p > 0.05 level. A comparison of responses was made by aggregating participants by affiliation (i.e., academia, federal, state, local, etc.) ( Figure 5 9 ). Interestingly, these varied only slightly, but are particularly evident with respect to the views of the Local and Private participants who tended to devalue criteria the other groups preferred (refer to Criteria #1 Groundwater Depth) and value higher criteria regarding more localized issues (i.e., Criteria #9 Urban Conversion). Weightings were aggregated using the rank sum weighting approach ( Table 5 4 ) and central tendencies were plotted using box andwhisk er graphs ( Figure 5 10). Criteria #1 ( Suitable Groundwater Depths ) was determined to be the most valued factor considered, while Criteria #3 ( Avoiding Tern and Plover Sites ) was determined to be the least valued factor, but the variation and overlap of central tendencies in the responses was extensive as indicated in the box and whisker plots. 5.5.3 Criteria Overlays and Evaluation The sieve mapping exercise used the ten preferentially weighted criteria to sieve the study area and pinpoi nt hotspots or areas where opportunities constrained by exclusions offered the highest potential (i.e., suitability) to establish mitigation sites for the recovery efforts. A comparison of the weighted versus unweighted sieving is presented in Figure 5 11. Hotspots are indicated by the darkest green colorations in the map (panels A and B). Localized effects of weightings are evident in the popout boxes on the right hand side of the figure. In this case, groundwater depth, period ic inundation, and connectivity play a much heavier role in determining site suitability under the weighted scenario.
185 The differences between weighted and unweighted results are highlighted in the lower panel (C). Areas shown here in green are elevated in status under the weighted scenario, and areas in red are now considered less suitable for acquisition and recovery interventions. Generally speaking, the introduction of stakeholder valuepreferences offered additional focus an increase in hectares of high suitability (5s) and low suitability ( 1 s) were experienced (refer to the pie charts in the A and B panels). The weighted criteria map was overlayed with the original committees mitigation site footprints for comparative purposes ( Figure 5 12 ) Notably, the majority of the original sites had suitability scores ranging from 3 to 5 indicating a moderate to high level of suitability or potential to meet recovery needs. However, as the figure indicates, the site boundaries for these areas could be redrawn to better capture the full potential of the areas (expanding or contracting the footprints to maximizing suitability). Of the 14 highest scoring polygons identified by CRISIS only four were contained within the boundaries of the original set of sites, and all of those were found in only one site (Site #7 in the figure). More importantly, CRISIS was able to identify more than 50 additional sites greater than 10 hectares in size with suitability scores in excess of 4 (moderately high suitability) that could be provide additional mitigation opportunities. 5.6 Discussion Striking an equitable balance between flood protection and the recovery of sustainable ecosystems is not merely a quest for a suitable tool, but also a search for a suit able means of communication (Carsjens and Ligtenberg 2007, page 82). Capturing and integrating conflicting stakeholder concerns using a GIS based SDSS affords planners and managers a unique opportunity to transparently consider multiple site
186 selection criteria, simultaneously balancing disparate measures of ecosystem int egrity while integrating conflicting stakeholder preferences within a visually engaging medium. The purpose of this study was to present a spiral based conceptual modeling approach that generates spatially explicit criteria using a GIS and MCDA in a parti cipatory manner to identify, prioritize and select mitigation sites based on stakeholders values and preferences in a transparent, repeatable and inclusive fashion. The approach was founded on the exploration of values ( What is important?) and return on investment ( Wh ere should we act ? ) linked with current scientific understanding of ecosystem functionality and the fundamental institutional processes governing EBM The approach was designed to bring together a wide range of readily available data, to assemb le these into essential themes based on logic that is scientifically informed, and to synthesize the assessment of system potential across space and time using meaningful indicators of suitability. The CRISIS results on Segment 10 indicate a high potential to mitigate cottonwood losses, and the hotspot maps point directly to locations where managers can act. The expert elicitation results presented herein clearly indicate a disagreement amongst experts with regards to what is important when positioning site s on the landscape, but the CRISIS approach offered a reasonable means to overcome these conflicts and generate a compromise. Despite the demonstrated strengths of the CRISIS approach, there are a number of key limitations to the spiral modeling process i n general, and to the application of CRISIS in particular that should be acknowledged and explored in future research initiatives. First, the process took significantly longer (approximately 3 months) than the committee deliberations that generated the fir st cut of mitigation sites for the study area
187 (approximately 1 month). The participants were required to make significant time and resource investments in the process over the course of four years a commitment they were likely unaware of when they volunteered to participate in the project. Better upfront management of expectations will likely improve the process. Moreover, the elicitation process was unwieldy and sometimes confusing to the participants. There are a variety of commercial off the shelf s oftware packages that can be used to survey respondents in a more interactive and automated manner (i.e., Survey Monkey, TurningPoint Technologies, etc.) that would improve the experience and reduce the error rates in the polling activities. Moreover, ther e are a variety of more robust weight of evidence approaches (Ordered Weighting, Analytic Hierarchy Process, Swing Weighting, etc) to conduct MCDA that could be explored to improve robustness of the elicitation and aggregation exercises (refer to Linkov et al., 2009 and Kiker et al., 2005 for general of these options). With regards to the application itself, the criteria selected were never intended to be comprehensive, and given the dynamic setting ( both ecologically speaking and politically speaking), it is highly l ikely that shifting agendas and changes in the environmental setting (e.g., the 2011 flooding event) will alter the perceived values of the criteria or negate their inclusion altogether. M oreover, one criteria in particular, that of Landowner Willingness (Criteria # 2) was assigned on an ad hoc basis by two of the teams experts who work closely with the local residents on a day to day basis. A more inclusive approach could be devised, poss ibly by surveying the local residents, to formally capture their willingness to participate in the recovery program.
188 5.7 Conclusions The spiral based CRISIS approach presented herein encourages the placement of mitigation sites in locations of greatest ecological, hydrological, and political benefits, increasing the probability for long term sustainability something desperately needed in the EBM arena on the Missouri River. This approach allows for better positioning of mitigation sites on the landscape impro ving each site s ability to compensate for lost ecosystem services and functions alike, while integrating watershed management with other ongoing land use planning efforts ensuring the recovery program is consistent with local, regional, and/or state water quality and floodplain management plans. The final result is a rigorous, inclusive, defensible and transparent system that can assist stakeholders in the strategic formulation of options fashioning a common understanding amongst participants and identif ying relevant information that encourages the exploration of innovative solutions to difficult, oftentimes contentious problems along the way
189 Figure 5 1 Before the ecosystem response model can be deployed, intervention sit es must be located. Site identification, prioritization and selection can be conducted based of stakeholder preferences and GIS based sieve mapping. Performance Measures Model Performance Step 3: Calibration Ecosystem Response Models Step 2: Mathematical Formalization Step 4: Forecasting Step 5: Alternative Evaluation Quality of The Fit Model Verification Sampling Design Reference Datasets Evaluation Datasets Fitted ValuesResponse Thresholds Adaptive Co Management Step 6: Construction and Monitoring Statistical Literature, Existing Models, Expert Contributions Predicted ValuesModel Validation Study Goals and Objectives Performance Hypotheses Site Selection via GIS Based Decision Support System Step 1: Conceptual Modeling Laboratory and Field Experiments Description Data from Literature and ExpertsModel Goals and Objectives
190 Figure 5 2 Sieve mapping is a multistep recursive process that focuses on t he selection, derivation, and reclassification of exert derived suitability criteria. When overlaid in a GIS, the process sieves the conditions and determines plausible solutions based on study goals, objectives, constraints, and opportunities. Figure 5 3 The spiraling site selection process recursively addressed criteria selection, data processing, multicriteria evaluation and sieve mapping.
191 Figure 5 4 The Segment 10 study area begins and Gavins Point Dam (RM 811.1) and follows the river downstream to Ponca State Park (RM 753).
1 92 Table 5 1 Transdiciplinary team members participating in the CRISIS exercise. Category Participant Count Association Academia 4 Benedicti ne College University of South Dakota South Dakota State University University of South Dakota Federal 5 U.S. Geological Survey USACE Omaha District USACE Kansas City District U.S. Geological Survey National Park Service ( Missouri National Rec reation River ) State 2 North Dakota Forest Service South Dakota Department of Agriculture, Division of Resource Conservation Local 2 Izaak Walton League of America Missouri River Futures/Natural Resources Conservation Service Non Government Organization 1 The Nature Conservancy Private 1 EA Engineering
193 Figure 5 5 Mitigation site suitability is dependent on hydrologic viability, ecologic integrity, and a measure of political necessity.
194 Figure 5 6 CRISIS goal driven recovery site selection approach characterizes opportunities and constraints based on ten criteria that center around three major themes: hydrologic viability, ecologic integrity and political necessity
195 Table 5 2 Site suitability indicators and criteria descriptions for CRISIS Suitability c riteria Logic for i nclusion Reclassification s and d erivations Criteria #1: Have Suitable Groundwater Depths Cottonwoods are p hraetophytes dependent upon water for survival. Depth to groundwater has been shown to be a good indicator of functioning and sustainable cottonwood ecosyst ems in the Missouri River Basin. Average groundwater elevations over the last 10 years (1997-2007) w ere computed for each well in the study area, a gradient map was developed by Dr. Tim Cowman (University of South Dakota/ Missouri River Institute ) using readily available data 1 = >3 m eters 2 = 2 -3 meters 3 = 1 -2 m eters 4 = 0 -1 meters 5 = < 0 m eters Criteria #2: Be Inside the Missouri National Recreational River (MNRR) Boundary O wned by "Willing" Land Owners Locating sites within the Missouri Natural Recreational River boundary that are owned by landowners interested in working with the Federal govern ment to restore cottonwoods in the basin is assumed to be a priority given current funding opportunities in the region. Ms. Theresa Smydra (Missouri River Futures/Natural Resources Conservation Service) and Dr. Tim Cowman (University of South Dakota/Missouri River Institute) developed a map indicating their experiential knowledge of landowner willingness to participate in cottonwood recovery efforts and combined this with MNRR boundary mapping to produce the original dataset. 0 = Sites are Outside the MNRR Boundary or in the River 1 = Sites are Inside the MNRR Boundary, but Landowners are Unwilling to Participate 3 = Sites are Inside the MNRR Boundary, but the Landowner Participation Status is Unknown 5 = Sites are Inside the MNRR Boundary, and Landowners ar e Willing to Participate Criteria #3: Avoid Tern and Plover Sites Direct competition for resources (in this case land) among Federally Threatened and Endangered species under the Missouri River BiOp must b e avoided. Both Existing nest sites and active recovery sites for piping plovers and least terns were identified and buffered in 100foot increments. Any locations within the first 600 feet of these areas were considered off limits. A preference to cottonwood recovery sites was given to locations along the banks if they were outside the avoidance zones. Ms. Kelly Crane and Mr. Tim Fleeger (USACE Omaha District) provided emergent sandbar habitat maps indicating exclusion zones for terns and plovers on the river. 0 = Sites within the 600 foot Buffer 1 = Sites are 600-700 feet 2 = Sites are 700-800 feet 3 = Sites are 800-900 feet 4 = Sites are 900-1000 feet 5 = Sites are >1000 feet Criteria #4: Be Near Potential Backwaters It is desirable to select sites that overlap with ongoing backwater recov ery initiatives for sturgeon recovery to optimize mobilization and planning costs/efforts, and restore cottonwood riparian ecosystems to their full functionality. Dr. Mark Dixon (University of South Dakota) generated current cover type maps for the segment indicating backwater areas for the project. 1 = >400 meters 2 = 300-400 meters 3 = 200-300 meters 4 = 100-200 meters 5 = < 100 meters
196 Table 5 2 Continued Suitability c riteria Logic for i nclusion Reclassification s and d erivatio ns Criteria #5: Be Adjacent to Existing Young Cottonwood Stands Young stands indicate areas where accretion is occurring, a condition that is favorable to the establishment of cottonwood stands as well. Dr. Dixons cover type maps were also used to identify current stands of young cottonwoods on the river. 1 = 0 100 meters 2 = 100-300 meters 3 = 300-600 meters 4 = 600-1000 meters 5 = > 1000 meters Criteria #6: Be Subject to Periodic Inundation Cottonwoods required periodic inundation in order to establish. Flow regulation and channelization substantially changed the Missouri Rivers historic hydrologic and geomorphic regimes and the natural variability in flows along many rivers has been modified by water management activities. Dr. Robert Jacobson (USG S) provided flood frequency mapping for the study area. 1 = Rarely Flooded 2 = Infrequently Flooded 3 = Frequently Flooded 4 = Moderately Flooded Criteria #7: Avoid High Erosion Areas Sustainability is the key to successful recovery initiatives, and as such, locating areas with low erosion potential suggests resilience in the face of large episodic flooding events. Dr. Tim Cowman (University of South Dakota/Missouri River Institute) conducted a time -series imagery analyses of river shoreline (1993-2003) to locate areas experiencing high erosion rates and mapped these for the study. 1 = Areas likely to erod e in the next 20 years 2 = Areas likely to erode in the next 40 years 3 = Areas lik ely to erode in the next 60 years 4 = Areas likely to erode in the next 80 years 5 = Areas not likely to erode Criteria #8: Provide Connectivity Connectivity of the landscape mosaic is absolutely necessary for species to survive. Disturbances periodically make portions of the landscape uninhabit able. Corridors fulfill an escape function b y permitting species to flee disturbance. Corridors also aid in re-colonization of the recovering site by plants and animals. Habitat patches that are isolated from similar habitat patches by great distances or inhospitable terrain are likely to have fewer species than less isolated patches. Dr. Dixons cover type maps were used to map corridors and gaps in forested coverage on the river. 1 = Offers Low Connectivity 2 = Offers Moderately Low Connectivity 3 = Offers Moderate Connectivity 4 = Offers Moderately High Connectivity 5 = Offers High Connectivity
197 Table 5 2 Continued Suitability c riteria Logic for i nclusion Reclassification s and d erivations Criteria #9: Be At Risk to Urban Conversion In an effort to capture the potential land use conversion trends in the reach over the course of the next 100 years (2006 2110), the research team devised a series of spatially -explicit heuristics (rules -based decisions) to identify critical changes in coverages, wit h the intent of developing a series of trend maps on a target -year basis to better illustrate these changes. Drivers of change included urban sprawl, erosion, agricultural conversions, protected lands, and cottonwood succession. Recovery initiatives would regard these areas as "threatened" and would therefore target these areas for protection and recovery. Census data was used to project future growth of urban centers in the study area. 1 = High Risk 2 = Moderately High Risk 3 = Moderate Risk 4 = Moderately Low Risk 5 = Low Risk Criteria #10: Be Near Existing Seed Sources There is a higher likelihood that heavier seed fall on recovery sites will be heightened if they are located adjacent to existing seed sources Dr. Dixons cover type maps were also us ed to identify current sources of cottonwood seed on the river. 1 = 0 100 meters 2 = 100-300 meters 3 = 300-600 meters 4 = 600-1000 meters 5 = > 1000 meters
198 Figure 5 7 Standardized criteria maps with normalization functio ns.
199 Figure 5 8 Ranks elicited from the stakeholders regarding their value preferences for criteria to identify and prioritize potential cottonwood recovery sites in the study area (Segment 10). Criteria receiving high scor es indicate strongest preferences. Participants highlighted in blue were part of the original site selection team. Table 5 3 Spearman rank correlations of expert opinions of site selection criteria values ( n = 1 5 ; df = 8 ). Expe rts A B C D E F G H I J K L M N O A B 0.09 C 0.08 0.32 D 0.58 0.30 0.36 E 0.52 0.08 0.58 0.39 F 0.72 0.12 0.18 0 .16 0.68 G 0.13 0.68 0.30 0.04 0.04 0.09 H 0.09 0.33 0.61 0.06 0.10 0.07 0.57 I 0.22 0.44 0.03 0.35 0.12 0.04 0.54 0.36 J 0.09 0.02 0.66 0.22 0.52 0.51 0.08 0.16 0.27 K 0.11 0.08 0.18 0.26 0.60 0.34 0.18 0.17 0.74 0.30 L 0.48 0.60 0.29 0.56 0.35 0.18 0.57 0.43 0.30 0.15 0.04 M 0.54 0.20 0.65 0.48 0.87 0.71 0.08 0.13 0.15 0.77 0.50 0.38 N 0.12 0.20 0.18 0.14 0.23 0.42 0.36 0.01 0.47 0.48 0.24 0.17 0.30 O 0.74 0.42 0.22 0.36 0.58 0.72 0.41 0.25 0.18 0.29 0.29 0.68 0.60 0.30 Sum Rank 0.77* 0.30 0.35 0.49 0 .73* 0.85* 0.17 -0.05 0.13 0.56 0.31 0.50 0.83* 0.49 0.90* Experts belonged to the original group tas ked with intuitively locating recovery sites in Segment 10. Correlations were statistically significant at p > 0.05.
200 Figure 5 9 MCDA was used to combine team valueladen preferences with regards to the value of the ten cri teria into the suitability assessment of potential mitigation sites in the watershed. As the figure illustrates, the stakeholder values differed slightly among groups formed on the basis of affiliation. Table 5 4 Aggregated pri orities based on expert elicitations and rank sum weighting. Ranks have been enumerated and colorized ( 10 = most valued through 1 = least valued) (RSW = Rank Sum Weight Normalized). Criteria d escription Mean Rank RSW Criteria #1: Have Suitable Groundwat er Depths 0.83 10 0.18 Criteria #2: Be Inside the MNRR and Owned by "Willing" Land Owners 0.31 2 0.04 Criteria #3: Avoid Tern and Plover Sites 0.29 1 0.02 Criteria #4: Be Near Potential Backwaters 0.56 6 0.11 Criteria #5: Be Adjacent to Existing Young Cottonwood Stands 0.55 5 0.09 Criteria #6: Be Subject to Periodic Inundation 0.66 9 0.16 Criteria #7: Avoid High Erosion Areas 0.54 3 0.05 Criteria #8: Provide Connectivity 0.61 8 0.14 Criteria #9: Be At Risk to Urban Conversion 0.60 7 0.13 Criteria # 10: Be Near Existing Seed Sources 0.55 5 0.09
201 Figure 5 10. Central tendencies mapped for the ten criteria based on expert opinions of importance in locating and selecting mitigation sites in the watershed. The boundary of t he boxes represents the 25th and 75th percentiles, and the black line inside represent medians, while the red lines indicate data means. Whiskers (error bars) to the left & right of each box indicate the 90th and 10th percentiles and black dots indicate outliers. Preferred Suitability Value 0 2 4 6 8 10 12 Criteria #1: Have Suitable Groundwater Depths Criteria #2: Be Inside MNRR and "Willing" Land Owners Criteria #3: Avoid Tern and Plover Sites Criteria #4: Be Near Potential Backwaters Criteria #5: Be Adjacent to Existing Young Cottonwood Stands Criteria #6: Be Subject to Periodic Inundation Criteria #7: Avoid High Erosion Areas Criteria #8: Provide Connectivity Criteria #9: Be At Risk to Urban Conversion Criteria #10: Be Near Existing Seed Sources
202 Figure 5 11. Results of the sieve mapping application. Areas of low suitability are indicated in red and transform to a red amber green coding scheme that indicates high suitability in the darkest green areas. In Panel A, the results are displayed under a nonweighted scenario (as if all the criteria were considered equally important). Panel B displays the results of the weighted values applied to these same criteria. Panel C shows the difference between the two maps.
203 Figure 5 12. Original sites identified by experts overlayed on top of the weighted suitability map generated by CRISIS
204 CHAPTER 6 OVERCOMING THE LAW OF UNINTENDED CONSEQUENCES : GETTING SOME PERSPEC TIVE AND PAUSING FOR REFLECTION 6.1 Concluding Remarks In retrospect, it was inevitable that the damming and channelization of the nations largest rivers produced cascading environmental impacts that generated wicked socio ecological conflicts between competing interests vying for the use of these valuable water resources T hese situations are after all textbook examples of the law of unintended consequences. Moreover, it is not surprising that after sixty plus years of environmental regulation mandating the mitigation of the se impacts on a species by species basis f rustration with the situation has led to an institutional paradigm shift toward more holistic systemsbased thinking. After all, as Benson (2012) so eloquently point s out, by the time a species has reached a federal threatened and/or endangered stat us the causes are no longer l ocalized but rat h er systemic, and recovery efforts must be directed toward a higher systemslevel solution Realistically speaking, a return to preregulated cond itions in these situations is highly unlikely F lood control and regulation have stimulated the economy, enabled floo dplain development and perpetuated dependenc ies on riverderived hydropower, navigation, and water supply for irrigation purposes leading to a condition Holling and Meffe (1996 ) refer red to as the pathology of command and control natural resource management Fortunately c ontemporary environmental awareness and appreciation has grown in these region s, and communities now clamor for a balance d solution one that maintains flood protection while improving ecosystem conditions at a watershe d (or higher) level Strategically speaking, recovery efforts must now focus on thoughtful watershedlevel interventions ( Hobbs et al., 2011) that move the ecosystem toward a
205 more desirable state by concentrating on improvements in ecosystem structure and function. Thi s re covery of ecosystem integrity simultaneously support s the production of a full suite of ecosystem goods and services t hat promote human health and well being while increasing biodiversity and habitat suitability for the species of concern. This balance is of course fundamentally simple to conceive, and overwhelming difficult to achieve. EBM offer s a viable solution to the challenge asking the questions What is important? and What are the consequences of our actions? from a prescriptive perspective ( Albar and Jetter, 2009; Riabecke et al., 2012) Unfortunately, natural resource managers often lack the tools and techniques to effectively and efficiently implement EBM strategies E cosystem complexity and data paucity preclude the exclusive use of hard science to characterize ecosystem response to proposed recovery efforts In the absence of carefully prepared technical analys e s, managers must oftentimes resort to filling in the gaps with the best available, albeit softer and more qualitative, experiential data which introduces additional uncertainty undermining confidence in the outcomes. The involvement of numerous stakeholders with conflicting values and agendas generates a constantly shifting decision making environment perpetuat ed by high stakes negotiations that are undermined by a sense of institutional urgency to compromise and quickly resolve the issues at hand with minimal conflict What the resource agencies need is an integrat ive decisionmaking framework that promotes transparent and adaptive planning in the face of multi dimensional choices characterized by uncertain science, conflicting objectives and difficult tradeoffs They also need decision support tools to a lign proposed interventions (i.e., easements,
206 plantings, restoration of backwater sloughs, etc.) with ecosystem response, such that recovery plans can be formulated to address seemingly i ntractable technical, environmental and social problems inherent to these wicked socio ecological conflicts. 6.2 Summary This research produced a recursive and reflexive spiraling approach to decisionmaking that integrates the iterative steps of the USACE planning paradigm with the principle operating tene ts of EBM and adaptive co management stressing the importance of actively engaging experts in the process. Moreover, I stress the importance of inclusion rather than exclusion in the expert pool expert practi tioners (i.e., planners and managers who apply scientific knowledge in their professions, but who typically do not conduct research and publish their knowledge in the peer reviewed literature) ( Perera et al., 2012) should be involved throughout the process in all the deliberations. T he degree to which these academic experts and practitioners (r eferred to collectively as stakeholders ) are or should be engaged in the process varie s depending on the scope and magnitude of the effort as well as their availability and interest in participating in the efforts I demonstrated three separate strategies to engage these experts ranging from a passive approach where I observed their decisionmaking heuristics and mapped these in a conceptual diagram to uncover their reasoning (lines of evidence) for selecting particular ecosystem response indicators to a highly dynamic workshop approach where I actively facilitated decision making using nominal group techniques that ultimately led to a consensus on the architecture of the ecosystem response model itself As a third strategy, I deployed an isolated extraction technique (anonymous polling and MCDA aggregation) to bot h elicit opinions on the relationships
207 between categorical hydrologic data and cottonwood suitability and determine the ir value laden preferences regarding the i mportance of particular criteria in identifying and measuring recovery potential for sites sca ttered across the basin. Anecdotally speaking, I found that the face to face interactions with the extended expert community (aka stakeholders ) promoted trust amongst the group and increased their confidence in the final study outcomes The i terative and recursive reflections ( impressed upon the group as a function of the constant spiraling processes ) encouraged group learning and increased both the team s understanding of the ecosystem as well as the manner in which the products w ould be used to support the Missouri River recovery efforts. Each step along the frameworks path engaged the stakeholders in: 1) the t houghtful integration of hard science and soft values 2) the q uantitative assessment of consequences 3) the c haracterization of the relative i m portance placed on th os e consequences and 4) the transparent examination of options. The individual studies presented in Chapters 2 4 confirm th e usefulness of the framework, and the power of the spiraling technique. To summarize: 1. Chapter 2 offered a l ook into the conceptual modeling process and described the evolution of a driver stressor model for the demonstration project that uncovered critical lines of evidence tying ecosystem response to measureable endpoints to capture recovery plan performance. Over the course of seven years, a transdisciplinary team of 82 stakeholders was engaged in the process to develop the final model construct. No less than nine drivers, eight stressors, and six valued ecosystem components were linked to one another in an i nfluence diagram to generate a series of 13 individual indicators of ecosystem response. 2. Chapter 3 moved deeper into the framework focusing on the formalization of one of the qualitative ecosystem response indicators, namely the USGS Land Capability Pote ntial Index (LCPI) a hydrology based indicator of ecosystem condition derived by overlaying flood frequency data over soil retention capacity in a GIS. This time, ten experts were engaged to convert the 16 qualitative categorical classes in the LCPI to sta ndardized (quantitative) ecosystem response scores (on a scale of
208 0 to 1). An anonymous polling instrument was used to elicit scores, and an MCDA technique (i.e., w eighted average rank sums) w as used to aggregate the responses into final ecosystem response curve. A global Morans I analysis based on overlays of the scores and the existing cottonwood communities confirmed the fit of the curve, but also discovered areas where recovery efforts might be focused. 3. Chapter 4 discussed the use of an interactive group technique to bring transdisciplinary experts together to devise a multimetric response model using the indicators identified in the conceptual modeling exercise above. Over the cours e of seven years, the team met and formalized the remainder of the ec osystem response indicators using a mixture of field data collection activities, GIS based processing, and statistical analyses. An Ecosystem Response Index (ERI) was produced that combined the various indicators into a meaningful construct centered on thr ee essential ecosystem components: biotic integrity, hydrography, and spatial integrity. The index was calibrated verified and validated using a pilot study application on Segment 10 of the Missouri River Results indicated a good fit for all curves, and the composite index as a whole, clearly showing that the model was capable of distinguishing between areas of low and high ecosystem function. In addition, independent validation using presence/absence data for bald eagles demonstrated the efficacy of th e model in determining suitability for one of the many charismatic species of concern on the river. 4. Chapter 5 turned to the question of where to act? employing a volunteer team of experts to identify, calibrate and apply a series of site selection criter ia to the pilot area in an attempt to pinpoint hotspots or areas of concern that c ould serve as targets for recovery actions. The approach, referred to as CRISIS ( Cottonwood Recovery Integrated Site Identification System ) is a GIS based spatial decision support system that deploys sieve mapping to screen potential site candidates on a landscape scale. The criteria in CRISIS were generated through interactive group meetings with the experts, and an anonymous balloting instrument was again used to determine their preferences regarding the importance of each criteri on in determining the suitability of the sites. Rank sum weighting (i.e., MCDA) was again used to aggregate the scores, and a weighted overlay produced the final target map. A comparison was made between originally proposed sites and the CRISIS map to measure the potential of these sites and pinpoint missed opportunities that should be investigated in the recovery efforts. To synthesize the study as whole in a more visual manner a word cloud (tag cloud) was constructed here to capture th e numerous themes (wickedness, competing interests, complexity, etc.) and proffered solutions ( EBM response modeling, expert elicitation, multicriteria analysis, formalized site selection, etc.) detailed within th is manuscripts text ( Figure 6 1 ). In essence, w ord clouds are simple visual
209 representations of key word importance based on the frequency of their occurrence in a report (or on a website) ( Bongshin et al., 2010 ) In this case, a list of 61 words were generated that best described the project, and then a frequency analysis was performed o n the text to assign weight or value to importance of these themes. Not surprisingly, the most frequent words in this dissertation were Recovery, Ecosystem integrity and the Spiral approach T hese final closing remarks are offered w ith this visualization in mind. The l ong term, management of large river systems requires critical thinking, thoughtful interventions, and collaborative decision making to ensure sustainability and promote ecosystem resiliency. The reflexive orientation of the spiral framework presented herein encourages collaboration by posing questions identifying problems, encouraging deliberation, exploring solutions, and responding to contextual changes in an ongoing series of action and reflection cycles. The process emulates adaptive management, promoting a learning by doing paradigm, integrating what was learned into the next generation of actions and responses The spiraling framework promotes a longterm vi sion of planning and management. After all, the Missouri River did not degrade overnight it has been in decline for over 70 years. It will take at least that long to recover the systems structure and function. To that end, multigenerational solutions must be devised, implemented, and adaptive ly managed if it is to be returned to a fully functioning and sustainable ecosystem providing the full complement of ecosystem goods and services to the region and the nation. Th e spiral process, with its openended configuration offers a dynamic and adaptive decisionmaking framework to support EBM activities far into the future.
210 6.3 Path Forward The cottonwood recovery efforts on the Missouri River are in their infancy the publication of the Cottonwood Management Plan (CMP) i n 2010 ( USACE, 2010 ) officially kicked off the recovery process but the actual design, implementation and adaptive co management of the system remains on the planning event horizon. Just recently, the ERI model developed in Chapter 4 was submitted to the USACE National Planning Center of Expertise (EcoPCX) for review and model certification1The entire process will need to be repeated for each of the priority segments named in the BiOp ( i.e., Segments 4, 6, 8, 9, and 13) ( US FWS 2003 ) It is important to note that the model has not been fully calibrated for these reaches hydrologic data (LCPI and ground water depths) will need to be derived for all five and Segment 13 will require additional data collection, calibration, verification and validation to ensure will hopefully be received early in the spring of 2014. Once the model has been certified, the USACE has plans to select recovery sites on Segment 10 (using CRISIS Chapter 5 ) develop alternative designs forecast conditions for the action and no action plans ( Step 4 in the framework), and compare and contrast the alternatives usin g the newly certified ERI model ( Step 5 in the framework) Once the recommended plans are chosen (based on cost effectiveness and efficiency), they will be constructed, and adaptive management response triggers will be established based on the ERI model parameters to guide monitoring and longterm management. At every step along the path, the transdisciplinary team will need to be reengaged to inform and offer support to the decisionmaking process to ensure transparency and instill confidence in the outcomes 1 http://el.erdc.usace.army.mil/ecocx/model.html (A ccessed Sept 2013).
211 representativeness in that unique (i.e., fully channelized) setting. CRISIS has only been calibrated and applied on Segment 10 thus far, and all data supporting that S DSS will need to be acquired, processed, and sieved (including a new elicitation to address differing priorities in the new reaches) Additionally, a prototype model has been developed using ESRIs Model Builder extension to process and sieve the maps, but the model needs refining and possibly updating to make it more user friendly and transferable to the end users (i.e., the stakeholders) As an aside, the expert elicitations thus far have been relatively simplistic in execution (i.e., hard copy polling instruments and interactive nominal group activities) Offthe shelf commercial software packages exist (e.g., Survey Monkey or TurningPoint Technologies) that could be deployed to automate the processes. In the next steps of the application, particularly the future forecasting step, these tools could be used in a dynamic workshop setting to project ecosystem response to proposed actions using the ERI model variables as assessment foci. Moreover, more robust MCDA strategies [Cardinal Rank Ordering of Criteria (CROC), Ordered Weight Average (OWA) and/or possi bly the Analytic Hierarchy Process (AHP)] could be explored as substitutions for the Rank Sum Weights methodology (possibly using Expert Choice or Decision Criterium Plus to automate the activities). An additional opportunity exists to validate the cottonwood ERI model with song bird survey data currently being collected by students under the direction of Dr. Mark Dixon at the University of South Dakota (pers. comm.). Preliminary data has already been made available to the cottonwood team, but the results are mixed owing primarily to the small samples sizes collected thus far. Other faunal data collections (mammals,
212 amphibians, reptiles, etc.) would be useful as well to advance the validation of the ERI model. 6.4 Future Research Opportunities Along the lines of new ideas and potential new research opportunities, personal communications with Ms. Kelly Cr ane and Mr. Tim Fleeger [USACE members on the Missouri Rivers multi agency Emergent Sandbar Habitat (ESH) team focused on least tern and piping plover recovery requirements] indicate that a GIS based sieve mapping site selection prototype has been developed to support the ESH Program. As the least terns and piping plovers avoid forested areas because of predator issues, and with the intention of avoiding areas of great potential for ESH recovery, it would be useful to add an additional exclusionary layer to CRISIS generated by the ESH prototype, to widen the avoidance zones for cottonwood recovery site positioning. Uncertainty clearly pervades the decisionmaking processes underlying any systemslevel ecosystem recovery effort. Determining the appropriate level of uncertainty analysis in these studies will depend on the types, sources and magnitudes of uncertainty as well as the decision context. By incorporating uncertainty into the findings it is possible to recast options in the context of decision risk (i.e., the distribution of risks, the stakeholders and decision makers risk aversion, and potential consequences of the decision). An analysis of uncertainty and risks was beyond the scope of this dissertation, but these issues merit further exploration. For example, an analysis of expert confidence in polling exercises could reveal a level of uncertainty in the responses. On the more technical si d e, a dditional SDSSrelated research could explore an analysis of uncertainty using t echniques like the Dempster Sha fer belief intervals approach as presented in Hunsaker et al. (2001) or the shadow map p ing
213 approach as described in Berry (2004 ) Doing so would allow one to establish spatial portraits (Eastman, 2001, page 389) of the value of information mapped, providing clarity while instilling transparency. With regards to the spiral framew ork itself, efforts are underway to transfer and apply the process to other studies in the USACE.2 3 The approach is being used to develop different types of ecosystem response models (i.e., bottomland hardwoods, coastal and freshwater wetlands prairies, lakes, rivers, etc.) Moreover, the approach is being used to generate non monetized ecosystem goods and services production functions in support of the post Superstorm Sandy initiative focused on reducing flood damage potential while restoring ecosystem int egrity to the North Atlantic coastline using naturebased (green infrastructure) features ( USACE, 2013a ) Opportunities exist to expand the use of the approach nationwide in support of ecosystem goods and service accountability for the USACE navigation mission under two funded work unit s I lead from the Dredging Operations and Environmental Research Program (DOER) ,4 and the Engineering With Nature (E WN) Navigation Systems Research Program (NAVSYS) .5The social sciences literature provides some guidance and advice on strategies for selecting expert, extracting information, and consolidating the knowledge into meaningful results, however, more research is needed to integrate this body of knowledge into the fields of landscape ecology and ecosystem based management. For 2 http://acwc.sdp.sirsi.net/client/search/asset/1010440 [ TR 12 21 (Middle Rio Grande HEP); Accessed Nov 2013) 3 http://acwc.sdp.sirsi.net/client/search/asset/1028169 [TR 1315 (Clear Creek HEP); Accessed Nov 2013] 4 http://el.erdc.usace.army.mil/do ts/doer/doer.html (Accessed Nov 2013) 5 http://el.erdc.usace.army.mil/ewn/Resources.html (Accessed Nov 2013)
214 example, while there are some general guidelines offering guidance on sample size ( Kreuger et al., 2012 ) in EBM studies (particularly those associated with longterm management of large regulated rivers), it seems likely that the complexity and potential controversies surrounding the recover y planning issues merits a broader and larger group of collaborators. But how much is enough, and what are the tradeoffs of using more (or fewer) experts? Moreover, practical limitations (funding, expert exhaustion, etc.) oftentimes limit the manager s res ource pool. Research and follow on guidance is needed on when the use of too few (or too many) experts threatens the validity of a studys conclusions. Further, studies that explore the implications of using transdisciplinary teams (of varied expertise and knowledge) should be undertaken to reveal the implications such decisions have on the effectiveness and reliability of the outcomes. These more inclusive approaches might provides results with a broader perspective, or a larger sample size, but do they necessarily guarantee a more reliable answer or does the uncertainty increase? More research on understanding the implications of incorrect parameterization of ecosystem response models is clearly warranted. Development and testing of automated tools that incorporate problem synthesis, hypothesis building, and model parameterization would allow users to test for and correct prototypes as they emerge in the spirals. Development of tools or strategies that better match spatial and temporal concerns to the proposed questions is the quest for the holy grail of landscape ecology. Bringing the spatiotemporal experiences and knowledge of experts to bear on the problem, offer a novel line of exploration for future studies.
215 Figure 6 1 A w ord cloud depicting the m ajor themes identified in the study Following standard word cloud design principles, the f requency in which the themes re occur in the study are reflected in the font size displayed.
216 APPENDIX A HISTORIC COVER TYPE MAPPING RESULTS
217 Tab le A 1 Existing c over types and aerial extents (in hectares) mapped in the reference domain segments for the 2006 2008 time period. Code Description Segment a reas (ha) 0 2 4 6 8 9 10 13 AGCROPLAND Farms and Croplands Grain, Cotton, Groves, and Orchards 1,926 44,212 10,771 140 4,322 12 65,726 120,815 BARREN Barren Areas 1 10 2 0 0 0 3 64 BCKWATR Flow through Channels and Backwaters 23 611 248 126 28 11 232 781 CTWFOREST* Cottonwood Dominant Forests (>25 yrs) 1,790 21,750 10,673 744 2,201 150 5,088 15,429 CTWSEED Cottonwood Seedlings (< 2 years old; < 1m in Height) 1,129 636 295 272 987 3,571 246 0 CTWSHRUB* Cottonwood Dominant Shrublands (2 25 yrs) 189 2,555 786 40 274 110 1,043 4,469 HERBWET Herbaceous Wetlands (Catta ils, Wet Meadows, etc.) 13 28 601 271 386 0 49 360 INSANDBAR In Channel Sandbars (Emergent Sandbar Habitat) 0 61 115 0 29 75 395 0 LACUSTRINE Open Water Lakes, Ponds, Borrow Pits 4 969 0 6 99 82 70 226 PASTURE Pastures, Haylands, Grazing Lands 2,834 24, 469 6,682 2,570 1,480 62 803 12,811 RIPFOREST* Non Cottonwood Riparian (Mesophytic) Forests (>25 yrs) 390 2,127 2,730 261 282 12 351 53 RIPSEED Non Cottonwood Riparian (Mesophytic) Seedlings (< 2 years old; < 1 meter in Height) 67 83 16 0 8 1 0 0 RIPSH RUB* Non Cottonwood Riparian (Mesophytic) Shrublands (2 25 yrs) 85 492 652 36 79 2 79 64 RIVER Main Channels, Tributaries, Backwater Streams 4,530 8,557 7,359 23,712 4,056 10,424 5,637 8,753 SANDBAR Non ESH Sandbars 97 36 66 13 17 9 26 1 UPLFOREST Uplan d Forests/Woodland on Bluffs 14 53 91 153 51 10 1,764 201 URBAN Residential, Commercial, Industrial and Transportation Avenues 136 790 2,673 1,065 175 43 3,970 7,413 Totals 13,228 107,439 43,760 29,409 14,474 14,574 85,482 171,440 *The cottonw ood model is designed to assess conditions in cottonwood-dominated and non-cottonwood riparian (mesophytic) forest and shrubland stands older than 2 years of age.
218 Table A 2 Post damming c over types and aerial extents (in hectares) mapped in the reference domain segments for the 19511958 period. Code Description Segment a reas (ha) 0 2 4 6 8 9 10 13 AGCROPLAND Farms and Croplands Grain, Cotton, Groves, and Orchards 2,033 41,007 13,900 5,437 5,503 2,511 60,794 117,255 BARRE N Barren Areas 0 0 0 0 0 0 0 57 BCKWATR Flow through Channels and Backwaters 58 617 139 132 11 6 193 1,100 CTWFOREST* Cottonwood Dominant Forests (>25 yrs) 1,310 14,999 14,220 5,510 2,378 1,206 7,334 19,488 CTWSEED Cottonwood Seedlings (< 2 years old; < 1m in Height) 293 0 0 0 0 0 0 0 CTWSHRUB* Cottonwood Dominant Shrublands (2 25 yrs) 641 15,613 4,122 2,057 1,250 998 2,762 11,831 HERBWET Herbaceous Wetlands (Cattails, Wet Meadows, etc.) 1 578 51 0 0 0 0 324 INSANDBAR In Channel Sandbars (Emergent Sa ndbar Habitat) 190 1,861 1,800 1,272 588 170 947 96 LACUSTRINE Open Water Lakes, Ponds, Borrow Pits 0 9 0 0 0 0 0 0 PASTURE Pastures, Haylands, Grazing Lands 2,970 20,558 1,798 6,829 278 321 1,852 4,100 RIPFOREST* Non Cottonwood Riparian (Mesophytic) Fo rests (>25 yrs) 104 0 0 0 0 0 40 0 RIPSEED Non Cottonwood Riparian (Mesophytic) Seedlings (< 2 years old; < 1 meter in Height) 28 0 0 0 0 0 0 0 RIPSHRUB* Non Cottonwood Riparian (Mesophytic) Shrublands (2-25 yrs) 374 0 0 0 0 0 11 0 RIVER Main Channels, Tributaries, Backwater Streams 4,590 8,913 6,108 7,042 3,758 5,153 6,139 10,903 SANDBAR Non ESH Sandbars 510 2,209 1,026 399 251 49 1,334 237 UPLFOREST Upland Forests/Woodland on Bluffs 0 307 103 108 162 30 2,341 1,813 URBAN Residential, Commercial, In dustrial and Transportation Avenues 127 496 495 624 296 53 1,739 4,239 Totals 13,229 107,167 43,762 29,410 14,475 10,497 85,486 171,443 *The cottonwood model is designed to assess conditions in cottonwood-dominated and non-cottonwood riparian ( mesophytic) forest and shrubland stands older than 2 years of age.
219 Table A 3 Pre damming c over types and aerial extents (in hectares) mapped in the reference domain segments for the 18921 893 period. Code Description Segment a reas (ha) 0 2 4 6 8 9 10 13 AGCROPLAND Farms and Croplands Grain, Cotton, Groves, and Orchards 111 11 84 155 100 431 27,447 97,004 BARREN Barren Areas 0 0 0 0 0 0 0 0 BCKWATR Flow through Channels and Backwaters 0 0 0 0 0 0 0 0 CTWFOREST* Cottonwoo d Dominant Forests (>25 yrs) 1,523 43,727 17,385 2,931 2,709 3,254 11,783 20,558 CTWSEED Cottonwood Seedlings (< 2 years old; < 1m in Height) 0 0 0 0 0 0 0 0 CTWSHRUB* Cottonwood Dominant Shrublands (2 25 yrs) 590 6,591 7,441 1,539 416 2,877 4,900 14,278 HERBWET Herbaceous Wetlands (Cattails, Wet Meadows, etc.) 12 731 186 6 0 122 588 292 INSANDBAR In Channel Sandbars (Emergent Sandbar Habitat) 420 1,744 2,353 3,113 1,159 1,003 2,199 8,403 LACUSTRINE Open Water Lakes, Ponds, Borrow Pits 1 124 181 0 0 1 44 2,049 PASTURE Pastures, Haylands, Grazing Lands 4,488 42,135 8,261 13,701 6,856 3,384 30,020 1,953 RIPFOREST* Non Cottonwood Riparian (Mesophytic) Forests (>25 yrs) 0 0 0 0 0 0 0 0 RIPSEED Non Cottonwood Riparian (Mesophytic) Seedlings (< 2 years o ld; < 1 meter in Height) 0 0 0 0 0 0 0 0 RIPSHRUB* Non Cottonwood Riparian (Mesophytic) Shrublands (2 25 yrs) 0 0 0 0 0 0 0 0 RIVER Main Channels, Tributaries, Backwater Streams 4,460 9,202 5,246 5,640 2,625 1,921 4,040 15,010 SANDBAR Non ESH Sandbars 9 89 3,175 2,516 1,835 544 1,425 3,116 8,619 UPLFOREST Upland Forests/Woodland on Bluffs 0 0 0 0 0 153 839 0 URBAN Residential, Commercial, Industrial and Transportation Avenues 51 0 110 489 66 1 510 3,276 Totals 12,645 107,440 43,763 29,409 14, 475 14,572 85,486 171,442 *The cottonwood model is designed to assess conditions in cottonwood-dominated and non-cottonwood riparian (mesophytic) forest and shrubland stands older than 2 years of age.
220 APPENDIX B BOXAND WHISKER PLOTSRAW VARIABLE DATA Figure B 1 Central tendencies mapped for five of the thirteen variables used in the model. The boundary of the boxes represent s the 25th and 75th percentiles, and the black line inside represent medians, while the red lines indicate data means. Whiskers (error bars) to the left & right of each box indicate the 90th and 10th percentiles and b lack dots indicate outliers.
221 Figure B 2 Box andwhisker plots for the system and segment level variables utilized in the model. Boxes capture the 25th to 75th percentiles and whiskers capture the 10the to 90th percentiles. Black lines inside the boxes highlight the medians, while red lines highlight the means of the data and black dots outside the whiskers indicate outl iers.
222 APPENDIX C CALIBRATED RESPONSE INDEX (RI) CURVES Figure C 1 Four of the 13 final RI curves developed for the cottonwood ERI model were calibrated using data collected from reference standard sites (i.e., cottonwooddominated st ands on Segments 210). CANHERB (A) CANSHUB (B) were used exclusively to assess Younger stands (2 25 years in age). CVALUE (C) and RICHNATIVE (D) across all age classes
223 Figure C 2 Four of the 13 final RI curves developed fo r the cottonwood ERI model were calibrated using literature reviews and/or historical data. PROPCTW (A) and RECRUIT (B) were developed based on trends found in Johnson (1992) DEPTHGW (C) was calibrated using well data collected exclusively in Segment 10 and applies only to Older cottonwood stands (> 25 yrs) being as sessed b y the model. INTERSPERS (D) was calibrated at a systems scale and is used in both the Older and Younger model versions.
224 Figure C 3 Two of the cottonwood models RI curves were developed using expert elicitation A) As a firs t step in calibrating the ADJLANDUSE indicator, experts classified the response value of various land covers categories B) Once all polygons within 2 km of the study areas core were assigned a RI score based on this graph, relative area was used to devel op a composite score for the site using a linear curve calibrated using 1950 historical mapping data. C) The WIS indicator was derived from expert opinion generating individual scores for Younger (2 25 yrs) vs. Older stands (> 25 yrs).
225 Figure C 4 The f inal RI curves for DISTPATCH (A) and PATCHSIZE (B) were calibrated using data at the reach level based on 1950s historical mapping (post damming, but prior to the acceleration of floodplain development in the basin). The LCPI (C ) indicator was calibrated using and expert elicitation methodology described in Chapter 3
226 APPENDIX D BOXAND WHISKER PLOTSTEST DATA SETS Figure D 1 Comparison of the central tendencies of training site RI s separated by type (i.e ., DistCW = disturbed cottonwooddominated, RIP = mesophytic dominated, and CW = cottonwooddominated). FQI (A) and INVEXOTIC (B) served as IMFs for the validation analysis. Inside the boxes, red lines indicate means and black lines indicate medians. CANHE RB (C) and CANSHRUB (D) relate only to younger aged test sites (225 yrs in age). CVALUE (E) and RICHNATIVE (F) relate to all test sites (spanning all age classes).
227 Figure D 2 Comparison of the central tendencies of training site RIs separated by type (i.e., DistCW = disturbed cottonwooddominated, RIP = mesophytic dominated, and CW = cottonwooddominated). Inside the boxes, red lines indicate means and black lines indicate medians. DEPTHGW (A) relates solely to Segment 10 test sites. The remaining panels (B E) have been calibrated for the entire suite of test sites.
228 Figure D 3 Comparison of the central tendencies of training site RI s separated by type (i.e., DistCW = disturbed cottonwooddo minated, RIP = mesophytic dominated, and CW = cottonwooddominated). Inside the boxes, red lines indicate means and black lines indicate medians. LCPI distributions (A B ) relate exclusively to Segment 10 test sites. The remaining panel s (C F) characterize scores across all test sites in Segments 210.
229 APPENDIX E VALIDATION DATA FOR RADIAL DIAGRAM
230 Table E 1 Comparison of FQI values derived with the regression formula ( Predicted FQI = 9.808 + 45.119 x ERI ) and the actual FQI values using the calibrated ERI model. Site n ame Site t ype Age c lass ERI Actual FQI Predicted FQI t value Std. error of p rediction Margin of e rror Lower b ound Upper b ound Interval w idth (p redicteda ctual) % a bove or Below a ctual Meridian Bridge CW Old Growth 0.62 1 7.12 18.35 2.01 4.64 9.31 9.04 27.66 18.62 1.23 7% Gerald Koster CW Old Growth 0.61 18.61 17.69 2.01 4.60 9.23 8.46 26.92 18.45 0.92 5% Airport CW Old Growth 0.66 19.28 20.12 2.01 4.75 9.53 10.59 29.66 19.07 0.85 4% Frost CW Old Growth 0.74 21.95 23.7 7 2.01 4.98 10.00 13.77 33.77 20.00 1.82 8% Vermillon River (Halverson) CW Old Growth 0.64 19.02 19.20 2.01 4.69 9.42 9.78 28.62 18.83 0.18 1% Weeks CW Old Growth 0.65 20.68 19.39 2.01 4.70 9.44 9.95 28.83 18.88 1.29 6% Mt. Marty (South) CW Mature 0.5 4 11.88 14.41 2.01 4.40 8.82 5.59 23.23 17.64 2.53 21% Gurney Farms CW Mature 0.68 24.46 21.03 2.01 4.81 9.65 11.38 30.68 19.30 3.43 14% Clay County CW Mature 0.63 21.09 18.81 2.01 4.67 9.37 9.44 28.18 18.73 2.28 11% Caton/Ray CW Mature 0.70 23.43 2 1.88 2.01 4.86 9.76 12.13 31.64 19.51 1.55 7% Gunderson Island CW Mature 0.67 16.95 20.19 2.01 4.75 9.54 10.65 29.73 19.08 3.23 19% Schmidt #2 CW Intermediate 0.66 19.94 19.83 2.01 4.73 9.50 10.33 29.33 18.99 0.11 1% Sister Island #1 CW Intermediate 0.46 11.93 11.01 2.01 4.19 8.41 2.60 19.41 16.81 0.92 8% Schmidt #3 CW Intermediate 0.69 20.23 21.25 2.01 4.82 9.68 11.57 30.92 19.35 1.01 5% Sippel CW Intermediate 0.64 17.91 19.17 2.01 4.69 9.41 9.76 28.58 18.83 1.26 7% Blickle CW Intermediate 0.71 26.91 22.01 2.01 4.87 9.77 12.24 31.78 19.55 4.90 18% Gunderson Backwater CW Intermediate 0.52 10.43 13.42 2.01 4.33 8.70 4.72 22.12 17.40 2.99 29% Elk Point CW Intermediate 0.75 18.41 24.12 2.01 5.01 10.04 14.08 34.17 20.09 5.71 31% Bacon CW Interme diate 0.66 21.14 20.14 2.01 4.75 9.54 10.61 29.68 19.07 1.00 5% Bacon Island CW Intermediate 0.58 13.42 16.47 2.01 4.52 9.07 7.39 25.54 18.15 3.04 23% Clay County North CW Old Growth 0.60 14.83 17.19 2.01 4.57 9.16 8.03 26.36 18.33 2.36 16% James Rive r Island CW Mature 0.67 17.90 20.60 2.01 4.78 9.59 11.00 30.19 19.19 2.70 15% Smith CW Young 0.78 19.52 25.53 2.01 5.10 10.23 15.30 35.76 20.45 6.01 31% Coulson DCW DistCW Old Growth 0.44 9.62 9.80 2.01 4.12 8.26 1.54 18.06 16.52 0.19 2% Armstrong/Iverson DistCW Old Growth 0.49 14.94 12.26 2.01 4.26 8.56 3.70 20.81 17.11 2.69 18% Schmidt #1 DistCW Mature 0.66 20.85 20.08 2.01 4.75 9.53 10.55 29.60 19.05 0.77 4% Clay County Park Campground DistCW Mature 0.39 8.30 7.89 2.01 4.00 8.03 0.15 15.92 16.07 0.41 5% Chief White Crane DistCW Intermediate 0.55 14.73 14.99 2.01 4.43 8.89 6.10 23.88 17.78 0.26 2%
231 Table E 1 Continued Site n ame Site t ype Age c lass ERI Actual FQI Predicted FQI t value Std. error of p r ediction Margin of e rror Lower bound Upper b ound Interval w idth (p redicteda ctual) % a bove or Below a ctual Weeldryer RIP Old Growth 0.61 16.82 17.62 2.01 4.59 9.22 8.40 26.83 18.44 0.80 5% Mt. Marty (Oak woods) RIP Old Growth 0.54 13.29 14.41 2.01 4.40 8.82 5.59 23.23 17.64 1.12 8% James River Hwy 50 RIP Old Growth 0.59 15.35 16.82 2.01 4.54 9.12 7.70 25.93 18.24 1.47 10% Bancroft RIP Old Growth 0.52 13.87 13.56 2.01 4.34 8.72 4.85 22.28 17.43 0.31 2% Myron grove RIP Old Growth 0.71 23.83 22.12 2.0 1 4.88 9.79 12.33 31.91 19.58 1.71 7% Gun Club RIP Old Growth 0.60 18.77 17.32 2.01 4.58 9.18 8.14 26.50 18.36 1.44 8% Gregg/Taggart RIP Old Growth 0.63 18.72 18.47 2.01 4.65 9.32 9.15 27.80 18.65 0.25 1% Wynot River Farms RIP Mature 0.66 26.42 20 .11 2.01 4.75 9.53 10.58 29.64 19.06 6.31 24% Coulson NCW RIP Intermediate 0.56 12.98 15.25 2.01 4.45 8.92 6.33 24.18 17.85 2.28 18% James River Island Intermediate RIP Intermediate 0.49 9.81 12.42 2.01 4.27 8.58 3.84 21.00 17.15 2.61 27% Heine RIP In termediate 0.64 19.59 19.03 2.01 4.68 9.40 9.64 28.43 18.79 0.56 3% Ketter (Bow Creek) RIP Intermediate 0.69 20.93 21.18 2.01 4.82 9.67 11.51 30.85 19.34 0.25 1% Ponca Island Pole RIP Intermediate 0.50 11.39 12.94 2.01 4.31 8.64 4.30 21.58 17.28 1.54 1 4% Yankton Island (Sister Island #2) CW Pole 0.48 10.84 11.63 2.01 4.23 8.48 3.15 20.11 16.96 0.78 7% Jepsen CW Pole 0.59 15.37 16.60 2.01 4.53 9.09 7.51 25.69 18.18 1.22 8% Goat Island #2 CW Pole 0.55 17.69 15.18 2.01 4.44 8.91 6.27 24.1 0 17.83 2.51 14% Bolton CW Pole 0.62 18.47 18.31 2.01 4.64 9.31 9.01 27.62 18.61 0.16 1% Finnegan/Hanson CW Pole 0.68 19.54 20.63 2.01 4.78 9.60 11.03 30.22 19.19 1.09 6% Coulson CW Sapling 0.52 14.21 13.59 2.01 4.34 8.72 4.87 22.30 17.44 0.63 4% East Yankton Island CW Sapling 0.56 10.36 15.47 2.01 4.46 8.95 6.52 24.43 17.90 5.11 49% Goat Island CW Sapling 0.58 16.06 16.48 2.01 4.52 9.08 7.40 25.55 18.15 0.42 3% Ponca CW Sapling 0.70 20.48 21.79 2.01 4.86 9.74 12.04 31.53 19.49 1.30 6% Dringman CW Pole 0.49 16.12 12.49 2.01 4.28 8.58 3.90 21.07 17.17 3.63 23% Blickle #2 CW Sapling 0.55 15.21 14.95 2.01 4.43 8.89 6.06 23.84 17.77 0.26 2% Merkwan DistCW Pole 0.42 7.41 9.14 2.01 4.08 8.18 0.96 17.32 16.36 1.73 23% Anderson Dis tCW Sapling 0.46 8.29 10.77 2.01 4.17 8.38 2.39 19.15 16.75 2.48 30%
232 Table E 1 Continued Site n ame Site t ype Age c lass ERI Actual FQI Predicted FQI t value Std. error of p rediction Margin of e rror Lower bound Upper b ound Inte rval w idth (p redicteda ctual) % a bove or Below a ctual James River Island Pole RIP Pole 0.34 8.72 5.69 2.01 3.88 7.78 2.09 13.47 15.56 3.03 35% Smith NCW RIP Pole 0.46 7.99 11.02 2.01 4.19 8.41 2.62 19.43 16.81 3.04 38% Goat Island North RIP Sapling 0.41 13.18 8.80 2.01 4.06 8.14 0.66 16.94 16.28 4.38 33% Bohan RIP Sapling 0.60 19.84 17.28 2.01 4.57 9.18 8.11 26.46 18.35 2.55 13% Ponca Island Sapling RIP Sapling 0.40 8.94 8.42 2.01 4.03 8.10 0.32 16.51 16.19 0.52 6%
233 LIST OF REFERENCES Abbott, A., 1997. Seven types of ambiguity. Theory and Soc. 26, 357391. Albar, F.M., Jetter, A.J., 2009. Heuristics in Decision Making, PICMET 2009 Conference Proceedings: Technology Management in the Age of Fundamental Change. IEEE, http://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=5230479, (Accessed July 2013), pp. 578584. Andreasen, J.K., O'Neill, R.V., Noss, R., Slosser, N.C., 2001. Considerations for the development of a terrestrial index of ecological integrity. Ecol. Indic. 1, 2135. Armitage, D.R., DavidsonHunt, I.J., McConney, P., Plummer, R., Berkes, F., Diduck, A.P., Pinkerton, E.W., Arthur, R.I., Doubleday, N.C., Wollenberg, E.K., 2009. Adaptive co management for social ecological complexity. Front. Ecol. Environ. 7, 95 102. Bailey, R.C., Norris, R.H., Reynoldson, a.T.B., 1998. Bioassessment of freshwater ecosystems using the reference condition approach: Comparing predicted and actual benthic inve rtebrate communities in Yukon streams. Freshwat. Biol. 39, 765774. Beattie, M., 1996. An ecosystem approach to fish and wildlife conservation. Ecol. Appl. 6, 696699. Benson, M.H., 2012. Intelligent Tinkering: the Endangered Species Act and Resilience. Ec ol. Soc. 17, 28. Berry, J.K., 2004. Use of "shadow maps" to understand overlay errors, GeoWorld, pp. 1819. Boehm, B., 1988. A spiral model of software development and enhancement. IEEE Computer 22, 6172. Boehm, B., Koolmanojwong, S., Lane, J.A., Turner, R., 2012. Principles for successful systems engineering. Proc. Compt. Sci. 8. Bongshin, L., Henry Riche, N., Karlson, A.K., Carpendale, S., 2010. IEEE Trans. Visual. Comp. Graph. 16, 11821189. Boyce, M.S., Vernier, P.R., Nielsen, S.E., Schmiegelow, F.K.A., 2002. Evaluating resource selection functions. Ecol. Model. 157, 281300. Bredeweg, B., Salles, P., Bouwer, A., Lierm, J., Nuttle, T., Cioaca, E., Nakova, E., Noble, R., Caldas, A.L.R., Uzunov, Y., Varadinova, E., Zitek, A., 2008. Towards a structured approach to building qualitative reasoning models and simulations. Ecol. Inform. 3, 1 12.
234 Burgman, M., 2005. Risks and decisions for conservation and environmental management, Ecology, Biodiversity and Conservation Series. Cambridge University Press, UK, p. 488. Burgman, M.A., Lindenmayer, D.B., Elith, J., 2005. Managing Landscapes for Conservation under Uncertainty. Ecol. 86, 20072017. Burks Copes, K.A., Kiker, G.A., 2014 (submitted) a. Formalizing the USGS Land Capability Potential Index (LCPI) the Fast and Frugal Way: A Case Study on the Missouri River, USA. Environ. Model. Software. Burks Copes, K.A., Kiker, G.A., 2014 (submitted) b. Uncovering Lines of Evidence Hidden within Wicked Problems: Using Conceptual Models to Inform Ecosystem based Management of the Missouri River Cottonwoods. Environ. Sys. Decisions. Carsjens, G.J., Ligtenberg, A., 2007. A GIS based support tool for sustainable spatial planning in metropolitan areas. Landscape Urban Plann. 80, 7283. Chang, K., 2006. Introduction to Geographic I nformation Systems, Third Edition. McGraw Hill, Boston, MA, 431 p. Chapin, F.S., Carpernter, S.R., Kofinas, G.P., Folke, C., Abel, N., Clark, W.C., Olsson, P., Stafford Smith, D.M., Walker, B., Young, O.R., Berkes, F., Biggs, R., Grove, J.M., Naylor, R.L., Pinkerton, E., Steffen, W., Swanson, F.J., 2010. Ecosystem stewardship: sustainability strategies for a rapidly changing planet. Trends in Ecol. Evol. 25, 241249. Cohen, J., 1988. Statistical power analysis for the behavioral sciences. Second Edition. Er lbaum, Hillsdale, NJ. Cundill, G., Cumming, G.S., Biggs, D., Fabricius, C., 2012. Soft systems thinking and social learning for adaptive management. Conserv. Biol. 26, 1320. Cundill, G., Fabricius, C., 2009. Monitoring in adaptive co management: Toward a learning based approach. Environ. Manage. 90, 3205 3211. DahdouhGuebas, F., Koedam, N., 2006. Empirical estimate of the reliability of the use of the point centered quarter method (PCQM): Solutions to ambiguous field situations and description of the PCQM + protocol. For. Ecol. Manage. 228, 18. Dale, V.H., Beyeler, S.C., 2001. Challenges in the development and use of ecological indicators. Ecol. Indicators 1, 3 10. Davis, S.M., Gaiser, E.E., Loftus, W.G., Huffman, A.E., 2005. Southern marl prairies concept ual ecological model. Wetlands 25, 821831. Dennison, W.C., Lookingbill, T.R., Carruthers, T.J.B., Hawkey, J.M., Carter, S.L., 2007. An eyeopening approach to developing and communicating integrated environmental assessments. Front. Ecol. Environ. 5, 307314.
235 Dijak, W.D., Rittenhouse, C.D., 2009. Development and application of habitat suitability models to large landscapes, in: Millspaugh, J.J., Thompson, I., F. R., (Eds.), Models for Planning for Wildlife Conservation in Large Landscapes. Elsevier, London, England, pp. 367389. Dixon, M.D., Johnson, W.C., Scott, M.L., Bowen, D., 2010. Missouri River Cottonwood Study: Final Report. Prepared for the U.S. Army Corps of Engineers, Omaha and Kansas City Districts. 59 pp. Dixon, M.D., Johnson, W.C., Scott, M.L., Bowen, D., Rabbe, L.A., 2012. Dynamics of plains cottonwood (Populus deltoides) forests and historical landscape change along unchannelized segments of the Missouri River, USA. Environ. Manage. 49, 9901008. Du Toit, D.R., 2005. Preparing people for integrated catchment management: a proposed learning alliance for the implementation of a new legal framework for water management in South Africa; reflexive learning in context, in: Smits, S., Fonseca, C., Pels, J. (Eds.), Symposium on Learning Alliances for Scaling Up Innovative Approaches in the Water and Sanitation Sector, 79 June 2005, IRC International Water and Sanitation Centre, Delft, The Netherlands, pp. 229241. Eastman, R., 2001. Uncertainty management in GIS: decision support tools for effective use of spatial data resources, in: Hunsaker, C.T., Goodchild, M.F., Friedl, M.A., Case, T.J. (Eds.), Spatial Uncertainty in Ecology. Springer, New York, NY, pp. 379390. Elith, J., Burgman, M.A., Regan, H., 2002. Mapping epistemic uncertainties and vague concepts in predictions of species distribution. Ecol. Model. 157. Fancy, S.G., Gross, J.E., Carter, S.L., 2009. Monitoring the condition of natural resources in US national parks. Environ. Monit. and Assess. 151, 161174. Ferretti, V., Pomarico, S., 2013. Ecological land suitability analysis through spatial indicators: An application of the Analytic Network Process technique and Ordered Weighted Average approach. Ecol. Indicators 34, 507 519. Fischenich, J.C., 2008. The Application of Conceptual Models to E cosystem Restoration. ERDC/EBA TN 081. U. S. Army Engineer Research and Development Center, Vicksburg, MS. Fisher, B., Turner, R.K., Morling, P., 2009. Defining and classifying ecosystem services for decision making. Ecol. Econ. 68, 643653. Forman, R.T.T ., 1995. Land Mosaics: The Ecology of Landscapes and Regions. Cambridge University Press, Cambridge, United Kingdom. Franklin, J.F., Spies, T.A., Van Pelt, R., Carey, A.B., Thornburgh, D.A., Berg, D.R., Lindenmayer, D.B., Harmon, M.E., Keeton, W.S., Shaw, D.C., Bible, K., Chen, J., 2002. Disturbances and structural development of natural forest ecosystems with silvicultural implications, using Douglas fir forests as an example. For. Ecol. Manage. 155, 399 423.
236 Fry, G., Tress, B., Tress, G., 2007. Integrativ e landscape research: Facts and challenges, in: Wu, J., Hobbs, R.J.E. (Eds.), Key Topics in Landscape Ecology. Cambridge University Press, Cambridge, UK, pp. 246 207. Funtowicz, S.O., Ravetz, J.R., 1992. Risk management as a postnormal science. Risk Anal. 12, 95 97. Galat, D.L., C. R. Berry Jr., Peters, E.J., White., R.G., 2005. Missouri River Basin, in: Benke, A.C., Cushing, C.E. (Eds.), Rivers of North America. Elsevier, Oxford, pp. 427480. Gentile, J.H., Harwell, M.A., Jr., W.C., Harwell, C.C., DeAngeli s, D., Davis, S., Ogden, J.C., Lirman, D., 2001. Ecological conceptual modes: A framework and case study on ecosystem management for South Florida sustainability. Sci. Total Environ. 274, 231253. Gharajedaghi, J., 2011. Systems Thinking: Managing Chaos and Complexity: A Platform for Designing Business Architecture, 3rd ed. Elsevier, Burlington, MA. Gigerenzer, G., 2007. Fast and frugal heuristics: The tools of bounded rationality in: Koehler, D.J., Harvey, N. (Eds.), Blackwell Handbook of Judgment & Decisi on Making. Blackwell Publishing, Malden, MA, pp. 6288. Gregory, A., Atkins, J., Burdon, D., Elliott, M., 2013. A problem structuring method for ecosystem based management: The DPSIR modelling process. Eur. J. Oper. Res. 227, 558569. Gregory, R., Failing, L., Harstone, M., Long, G., McDaniels, T., Ohlson, D., 2012. Structured Decision Making: A Practical Guide to Environmental Management Choices. Wiley Blackwell, A John Wiley & Sons, Ltd., Publication, Oxford, UK. Gucciardo, S., Route, B., Elias, J., 2004. Conceptual models for longterm ecological monitoring in the Great Lakes Network. Great Lakes Technical Report: GLKN/2004/'04. National Park Service, Great Lakes Inventory Monitoring Network, Ashland, WI, p. 101 pp. Guinn, J.E., 2004. Bald Eagle Nest Site Selection and Productivity Related to Habitat and Human Presence in Minnesota. Dissertation, Biological Sciences Department, North Dakota State University of Agriculture and Applied Science, Fargo, ND. Guisan, A., Zimmerman, N.E., 2000. Predictive habitat distribution models in ecology. Ecol. Model. 135, 147 186. Harrell, F.E., Lee, K.L., Mark, D.B., 1996. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring reducing errors. Stat. Med. 15.
237 Harwell M.A., Meyers, V., Young, T., Bartuska, A., Gassman, N., Gentile, J.H., Harwell, C.H., Appelbaum, S., Barko, J., Causey, B., Johnson, C., Mclean, A., Smola, R., Templet, P., Tosini, S., 1999. A framework for an ecosystem integrity report card. Bioscience 49, 543556. Henderson, J.E., O'Neil, L.J., 2004. Conceptual models to support environmental planning and operations. ERDC/TN SMART 049. U.S. Army Engineer Research and Development Center, Vicksburg, MS. Henderson, J.E., O'Neil, L.J., 2007. Template for C onceptual Models Construction: Model Components and Application of the Template. ERDC TN SWWRP077. U.S. Army Engineer Research and Development Center, Vicksburg, MS. Higgs, G., 2006. Integrating multi criteria techniques with geographical informations systems in waste facility location to enhance public particpation. Waste Manage. Res. 24, 105 117. Hirzel, A.H., Le Lay, G., 2008. Habitat suitabiity modeling and nich theory. J. Appl. Ecol. 45, 1372 1381. Hobbs, R.J., Hallett, L.M., Ehrlich, P.R., Mooney, H .A., 2011. Intervention ecology: applying ecological science in the twenty first century. Bioscience 61, 442450. Holling, C.S., Meffe, G.K., 1996. Command and Control and the Pathology of Natural Resource Management. Con. Bio. 10, 329 337. Hunsaker, C.T., Goodchild, M.F., Friedl, M.A., Case, T.J., 2001. Spatial Uncertainty in Ecology. Springer, New York, NY. Innes, J.E., Booher, D.E., 2010. Planning with Complexity: An Introduction to Collaborative Rationality for Public Policy. Routledge, London and New Y ork, NY. Jacobson, R.B., Chojnacki, K., Huhnmann, B., 2008. LandCapability Potential Index, Missouri River, Ponca, Nebraska to Gavins Point Dam. USGS CERC: 2008 LCPI Completion Report, 6 pp. Jacobson, R.B., Chojnacki, K.A., Reuter, J.M., 2007. Land Capabi lity Potential Index (LCPI) for the Lower Missouri River Valley. Scientific Investigations Report 20075256, Developed by the U. S. Geological Survey in cooperation with the U. S. Fish and Wildlife Services, Nebraska Game and Parks Commission, and The Natu re Conservancy ( http://publs.usgs.gov/sir/2007/5256/) Jacobson, R.B., Galat, D.L., 2006. Flow and form in rehabilitation of largeriver ecosystems: An example from the Lower Missouri River. Geomorph. 77, 249 269. Jacobson, R.B., Galat, D.L., 2008. Design of a naturalized flow regime an example from the Lower Missouri River, USA. Ecohydrology 1, 81 104.
238 Jacobson, R.B., Janke, T.P., Skold, J.J., 2011. Hydrologic and geomorphic considerations in restorati on of river floodplain connectivity in a highly altered river system, Lower Missouri Rivers, USA. Wetlands Ecol. Manage. 19, 295316. Jakeman, A.J., Letcher, R.A., Norton, J.P., 2006. Ten iterative steps in development and evaluation of environmental model s. Environ. Model. Software 21, 602614. Johnson, W.C., 1992. Dams and riparian forests: case study from the upper Missouri River. Rivers 3, 229242. Johnson, W.C., Dixon, M.D., Scott, M.L., Rabbe, L., Larson, G., Volke, M., Werner, B., 2012. Forty years of vegetation changes on the Missouri River floodplain. Bioscience 62, 123 135. Jrgensen, S.E., Bendoricchio, G., 2001. Fundamentals of Ecological Modelling. Elsevier Science Ltd., Kidlington, Oxford. Kahneman, D., Slovic, P., Tversky, A., 1982. Judgment U nder Uncertainty: Heruristics and Biases. Cambridge University Press, New York, NY, p. 555. Kandziora, M., Burkhard, B., Mller, F., 2013. Interaction of ecosystem propoerties, ecosystem integrity and ecosystem service indicators a theoretical matrix exercise. Ecol. Indic. 28, 5478. Kareiva, P., Tallis, H., Ricketts, T.H., Daily, G.C., Polasky, S., 2011. Natural Capital: Theory and Practice of Mapping Ecosystem Services. Oxford University Press, Oxford. Keeble, L.B., 1952. Principles and practice of town and country planning. Estates Gazzette, London, p. 594. Keeley, N.B., Macleod, C.K., Forrest, B.M., 2012. Combining best professional judgement and quantile regression splines to improve characterisation of macrofaunal responses to enrichment. Ecol. Indic ators 12, 154166. Kiker, G.A., Bridges, T.S., Varhese, A., Seager, T.P., Linkov, I., 2005. Application of multicriteria decision analysis in environmental decision making. Integ. Environ. Assess. Manage. 1, 95 108. Kloprogge, P., van der Sluijs, J.P., Pet ersen, A.C., 2011. A method for the analysis of assumptions in model based environmental assessments. Environ. Model. Software 26, 289301. Koehler, D.J., Harvey, N., 2007. Blackwell Handbook of Judgment & Decision Making. Blackwell Publishing, Malden, MA, p. 664. Kreuger, T., Page, T., Hubacek, K., Smith, L., Hiscock, K., 2012. The role of expert opinion in environmental modelling. Environ. Model. Software 36, 418.
239 Larson, M.A., Millspaugh, J.J., Thompson, F.R., 2009. A review of methos for quantifying wi llife habitat in large landscapes, in: Millspaugh, J.J., Thompson, F.R. (Eds.), Models for Planning for Wildlife Conservation in Large Landscapes. Elsevier, London, England, pp. 225 250. Leskinen, P., Kangas, J., Pasanen, A., 2003. Assessing ecological val ues with dependent explanatory variables in multi criteria forest ecosystem management. Ecol. Model. 170, 112. Lester, R.E., Fairweather, P.G., 2011. Ecosystem states: creating a dataerived, ecosystem scale ecological response model that is explicit in s pace and time. Ecol. Model. 222, 26902703. Lewis, L., Clark, L., Krapf, R., Manning, M., Staats, J., Subirge, T., Townsend, L., Ypsilantis, B., 2003. Riparian area management: Riparian wetland soils. Technical Reference 1737 19. Bureau of Land Management, Denver, CO. BLM/ST/ST 03/001+1737, p. 109. Light, S., Medema, W., Adamowski, J., 2013. Exploring collaborative adaptive management of water resources. J. Sustainable Dev. 6, 31 46. Lindsey, A.A., 1955. Testing the linestrip method against full tallies in diverse forest types. Ecol. 36, 485949. Linkov, I., Loney, D., Cormier, S., Satterstrom, F.K., Bridges, T.S., 2009. Weight of evidence evaluation in enviromental assessment: review of qualitative and quantitative approaches. Sci. Total Environ. 407, 51995205. Linkov, I., Satterstrom, F.K., Kiker, G.A., Bridges, T.S., Benjamin, S.L., Belluck, D.A., 2006. From optimization to adaptation: Shifting paradigms in enviromental management and their application to remedial decisions. Integ. Environ. Assess. Manage. 2, 9298. Malczewski, J., 1999. GIS and Multicriteria Decision Analysis. John Wiley & Sons, Inc., New York, NY. Malczewski, J., 2004. GIS based landuse suitability analysis: a critical overview. Prog. Plan. 62, 365. Malczewski, J., 2006. GIS based mul ticriteria decision analysis: a survey of the literature. Int. J. Geogr. Inf. Sci. 20, 703 726. Marsh, N., Cuddy, S., 2010. Ecological modelling to support natural resource management, in: Saintilan, N., Overton, I. (Eds.), Ecosystem Response Modelling in the Murray Darling Basin. CSIRO Publishing, Collingwood VIC, Australia, pp. 3 14. McElhany, P., Steel, E.A., Avery, K., Yoder, N., Busack, C., Thompson, B., 2010. Dealing with uncertainty in ecosystem models: lessons from a complex salmon model. Ecol. Appl 20, 465 482.
240 McIntosh, B.S., Seaton, R.A.F., Jeffrey, P., 2007. Tools to think with? Towards understanding the use of computer based support tools in policy relevant research. Environ. Model. Software 22, 640648. McLeod, K.L., Leslie, H.M., 2009. Why ec osystem based management?, in: McLeod, K.L., Leslie, H.M. (Eds.), Ecosystem based Management for the Oceans. Island Press, Washington, DC, pp. 3 12. Meyer, M.A., Booker, J.M., 2001. Eliciting and Analyzing Expert Judgment: A Practical Guide. American Stati stical Association and Society for Industrial and Applied Mathematics, Philadelphia, PA. Miller, S.J., Pruitt, B.A., Theiling, C.H., Fischenich, J.C., Komlos, S.B., 2012. Reference concepts in ecosystem restoration and environmental benefits analysis (EBA) : Principles and practices. ERDC TN EMRRP EBA12. U.S. Army Engineer Research and Development Center (ERDC) Technical Notes Collection, Vicksburg, MS, http://cw environment.usace.army.mil/eba/EMR RP (Accessed June 2013). Mller, A.P., Jennions, M.D., 2002. How much variance can be explained by ecologists and evolutionary biologists? Oecol. Aquat. 132, 492500. Moran, P., 1948. The interpretation of statistical maps. J. R. Stat. Soc. B 10 B 10, 243251. Moser, S.C., Williams, S.J., Boesch, D.F., 2012. Wicked Challeges at Land's End: Managing Coastal Vulnerability Under Cliamte Change. Annu. Rev. Environ. Resourc. 37, 5178. National Research Council, 2001. Compensating for Wetland Losses Under the C lean Water Act. National Academy Press, Washington, D.C.. National Research Council, 2002. The Missouri River Ecosystem: Exploring the Prospects for Recovery. National Academy Press, Washington, D.C. National Research Council, 2004. Managing the Columbia River: Instream Flows, Water Withdrawals, and Salmon Survival. National Academy Press, Washington, D.C. National Research Council, 2005a. Valuing ecosystem services: Toward better environmental decision making. National Academies Press, Washington, DC. N ational Research Council, 2005b. Water Resources Planning for the Upper Mississippi River and Illinois Waterway. National Academies Press, Washingtomn, DC. Niemeijer, D., De Groot, R.S., 2008. Framing environmental indicators: moving from causal chains to causal networks. Environ. Dev. Sustain. 10, 89106.
241 Oliver, I., 2002. An expert panel based approach to the assessment of vegetation condition within the context of biodiversity conservation Stage 1: the identification of condition indicators. Ecol. Indicators 2, 223237. Orsi, F., Geneletti, D., Newton, A.C., 2011. Towards a common set of criteria and indicators to identify forest restoration priorities: an expert panel based approach. Ecol. Indic. 11, 337347. Pahl Wostl, C., 2007. The implications of com plexity for integrated resources management. Environ. Model. Software 22, 561569. Parnes, S.J., 1992. Source Book for Creative Problem Solving. Creative Eduction Foundation Press, Buffalo, NY. Parrish, J.D., Braun, D.P., Unnasch, R.S., 2003. Are we conser ving what we say we are? Measuring ecological integrity within protected areas. Bioscience 53, 851. Perera, A.H., Ashton Drew, C., Johnson, C.J., 2012. Experts, expert knowledge, and their roles in landscape ecological applications, in: Perera, A.H., Asht on Drew, C., Johnson, C.J. (Eds.), Expert Knowledge and Its Application in Landscape Ecology. Springer, New York, NY, pp. 1 10. Poff, N.L., Allan, D., Palmer, M.A., Hart, D.D., Richter, B.D., Arthington, A.H., Rogers, K.H., Meyer, J.L., Stanford, J.A., 200 3. River flows and water wars: emerging science for environmental decision making. Front. Ecol. Environ. 1, 298306. Poiani, K.A., Richter, B.D., Anderson, M.G., Richter, H.E., 2000. Biodiversity conservation at multiple scales: functional sites, landscapes, and networks. Bioscience 50, 133 146. Price, J., Sibernagel, J., Miller, N., Swaty, R., White, M., Nixon, K., 2012. Eliciting expert knowledge to inform landscape modeling of conservation scenarios. Ecol. Model. 229, 7687. Quinn, G.P., Keough, M.J., 20 02. Experimental Design and Data Analysis for Biologists. Cambridge University Press, New York, NY, 537 pp. Reed, D., Martin, L., Cushing, J.A., 2013. Using Information on Ecosystem Goods and Services in Corps Planning: An Examination on Authorities, Polic ies, Guidance, and Practices. in: Engineers, U.S.A.C.o. (Ed.). Institute of Water Resources, Ft. Belvoir, VA. Reed, P.B., 1988. National list of plant species that occur in wetlands: national summary. Reza, M.I.H., Abdullaha, S.A., 2011. Regional index of ecological integrity: a need for sustainable management of natural resources. Ecol. Indic. 11, 220229.
242 Reza, M.I.H., Abdullaha, S.A., Norc, S.B.M., Ismalid, M.H., 2013. Integrating GIS and expert judgment in a multi criteria analysis tomap and develop a habitat suitability index: A case study of largemammals on the Malayan Peninsula. Ecol. Indicators 34, 149158. Rheinhardt, R.D., Brinson, M.M., Christian, R.R., Miller, K.H., Meyer, G.F., 2007. A referencebased framework for evaluating the ecological condition of stream networks in small watersheds. Wetlands 27, 524542. Riabecke, M., Danielson, M., Ekenberg, L., 2012. Stateof the art prescriptive criteria weight elicitation. Adv. Decis. Sci. 2012, 24 pages. Rittel, H.W., Webber, M.M., 1973. Dilemmas in a general theory of planning. Policy Sciences 4, 155 169. Rogers, K.H., 2003. Adopting a heterogeneity paradigm: implications for biodiversity management in protected areas, in: du Toit, J., Biggs, H. (Eds.), The Kruger Experience: Ecology and Management of Savanna Heterogeneity. Island Press, Washington, DC, pp. 41 58. Rood, S.B., Mahoney, J.M., 1990. Collapse of riparian poplar forests downstream from dams in western prairies: Probable causes and prospects for mitigation. Environ. Manage. 14, 451464. Royce, W.W., 1970. Managing the development of large software systems (Waterfall Model), http://www.cs.umd.edu/class/spring2003/cmsc838p/Process/waterfall.pdf (Accessed Au g 2013). Rger, N., Schlter, M., Matthies, M., 2005. A fuzzy habitat suitability index for Populus euphratica in the Northern Amudarya delta (Uzbekistan). Ecol. Model. 184, 313328. Ruhl, J.B., 2008. Adaptive Management for natural resoruces inevitable, impossile, or both? Rocky Mtn. Law Instit. 54, .01. 06. Rumpff, L., Duncan, D.H., Vesk, P.A., Keith, D.A., Wintle, B.A., 2011. State and transition modelling for adaptive management of native woodlands. Biol. Conserv. 144, 12241236. Sangi orgi, D., 2011. Transformative services and transformation design. Int. J. Design 5, 2940. Scott, M.L., Auble, G.T., Dixon, M.D., Rabbe., L., 2012. Long term cottonwood forest dynamics on the upper Missouri River, USA. River Res. Appl. Early version avail able online at: http://onlinelibrary.wiley.com/doi/10.1002/rra.2588/full (Accessed January 2013), DOI: 10.1002/rra.2588. Scott, M.L., Auble, G.T., Friedman, J.M., 1997. Flood Dependenc y of Cottonwood Establishment Along the Missouri River, Montana, USA. Ecol. Appl. 7, 677 690.
243 Scott, M.L., Brasher, A.M.D., Reynolds, E.W., Caires, A., Miller., M.E., 2005. The structure and functioning of riparian and aquatic ecosystems of the Colorado Pl ateau conceptual models to inform monitoring. U. S. Gelological Survey (USGS). Available online at: http://science.nature.nps.gov/im/monitor/docs/ScottM_etal_2005_riparian _models.pdf (Accessed January 2013). Sear, D.A., Wheaton, J.M., Darby, S.E., 2008. Uncertain restoration of gravel bed rivers and the role of geomorphology, in: Habersack, H., Piegay, H., Rinaldi, M.E. (Eds.), Gravel bed Rivers: From Process Understanding to River Restoration. Elsevier, pp. 739760. Smith, R.D., Amman, A., Bartoldus, C., Brinson, M.M., 1995. An approach for assessing wetland functions based on hydrogeomorphic classification, reference wetlands, and functional indices. WRP DE 9, http://el.erdc.usace.army.mil/elpubs/pdf/wrpde9.pdf U.S. Army Engineer Waterways Experiment S tation, Vicksburg, MS. Society for Ecological Restoration International, 2004. The Society of Ecological Restoration International Primer on Ecological Rehabilitation, Version 2. Available online at: http://www.ser.org/content/ecological_rehabilitation_primer.asp (Accessed January 2013). Steiner, G., 2009. The concept of open creativity: collaborative creative problem solving for innovation generation a systems approach. J. Bus Manage. 15, 533. Stillwell, W.G., Seaver, D.A., Edwards, W., 1981. A comparison of approximation techniques in multiattribute utility decision making. Organ. Behav. Hum. Perform. 28, 6277. Swink, F.A., Wilhelm, G.S., 1994. Plants of the Chicago region. Fourth Edition. ed. Indiana Academy of Sciences, Indianapolis. Taft, J., Wilhelm, G., Ladd, D., Masters, L., 1997. Floristic quality assessment for vegetation in Illinois. A method for assessing vegetation integrity. Eriginia 15, 995. Tarboton, K.C., Ir izarry Ortiz, M.M., Loucks, D.P., Davis, S.M., Obeysekera, J.T., 2004. Habitat Suitability Indices for Evaluating Water Management Alternatives. Office of Modeling Technical Report, South Florida Water Management District, West Palm Beach, FL, p. 170. te B rmmelstroet, M.C.G., 2010. Making planning support systems matter : improving the use of planning support systems for integrated land use and transport strategy making. Phd Dissertation, ISBN: 97890 90251301, UvA DARE, the institutional repository of the University of Amsterdam (UvA) http://dare.uva.nl/document/164157.
244 The Northern Great Plains Floristic Quality Assessment Panel, 2001. Coefficient s of conservatism for the vascular flora of the Dakotas and adjacent grasslands. U.S. Geological Survey, Biological Resources Division, Information and Technology Report USGS/BRD/ITR 20010001. 32 pp., http://www.npwrc.usgs.gov/resource/plants/fqa/index.htm (Accessed August 2013). Tress, G., Tress, B., Fry, G., 2005. Clarifying integrative research concepts in landscape ecology. Landscape Ecol. 20, 479493. Turner, M.G., Gardener, R.H., ONeill, R.V., 2001. Landscape Ecology in Theory and Practice: Pattern and Process. Springer Verlag, New York, New York. U. S. Fish and Wildlife (USFWS), 1980. Standards for the development of habitat suitability index models, Washington, DC. U. S. Fish an d Wildlife (USFWS), 2000. Biological Opinion on the Operation of the Missouri River Main Stem Reservoir System, Operation and Maintenance of the Missouri River Bank Stabilization and Navigation Project, and Operation of the Kansas River Reservoir System, D enver, CO and Fort Snelling, MN. U. S. Fish and Wildlife (USFWS), 2003. Amendment to the 2000 Biological Opinion on the Operation of the Missouri River Main Stem Reservoir System, Operation and maintenance of the Missouri River Bank Stabilization and Navig ation Project, and Operation of the Kansas River Reservoir System. Denver, CO and Fort Snelling, MN. U.S. Army Corps of Engineers (USACE), 2000. Planning Guidance Notebook. Engineer Regulation 11052 100, Washington, DC. U.S. Army Corps of Engineers (USACE ), 2002. Final Lower Snake River Juvenile Salmon Migration Feasibility Report and Environmental Impact Statement, Appendix L: Lower Snake River Mitigation History and Status. USACE Walla Walla District, Walla Walla, WA, p. 86. U.S. Army Corps of Engineers (USACE), 2003. Planning Civil Work Projects Under the Environmental Operating Principles, Engineering Circular 1105 2 404, Washington, DC. U.S. Army Corps of Engineers (USACE), 2004a. Final Upper Mississippi River Illinois Waterway (UMR IWW) System Navig ation Integrated Feasibility Report and Programmatic Environmental Impact Statement, Appendix ENV B: Site Specific Habitat Assessment Environmental Report. USACE Mississippi Valley Division, Vicksburg, MS, p. 261. U.S. Army Corps of Engineers (USACE), 2004b. Upper Mississippi River System Flow Frequency Study, Final Report. U. S. Army Corps of Engineers Rock Island District, ( http://www.mvr.usace .army.mil/Missions/FloodRiskManagement/UpperMississippiFlow FrequencyStudy.aspx Accessed February 2013).
245 U.S. Army Corps of Engineers (USACE), 2010. Proposed Implementation of a Cottonwood Management Plan Along Six Priority Segments of the Missouri River. U.S. Army Corps of Engineers Omaha District, Paper 40, ( http://digitalcommons.unl.edu/usarmyceomaha/40, Accessed March 2012). U.S. Army Corps of Engineers (USACE), 2013a. Coastal Risk Reduction and Resilience: Using the Full Array of Measures. U.S. Army Corps of Engineers Civil Works Directorate, Washington, DC. U.S. Army Corps of Engineers (USACE), 2013b. The Comprehensive Everglades Restoration Program (CERP) website, http://www.evergladesplan.org/ (Accessed August 2013). U.S. Army Corps of Engineers (USACE), 2013c. Great Lakes Restoration Initiative website, http://greatlakesrestoration.us/in dex.html (Accessed August 2013). U.S. Army Corps of Engineers (USACE), 2013d. Missouri River Recovery Program website, http://moriverrecovery.usace.army.mil/mrrp/f?p=136:1:0::NO (Ac cessed August 2013). U.S. Army Corps of Engineers (USACE), 2013e. Salmon Recovery Caucus, website: http://www.salmonrecovery.gov/Home.aspx (Accessed August 2013). U.S. Environmental Protection Agency ( USEPA), 2013. Waste and Cleanup Risk Assessment Glossary website, http://www.epa.gov/oswer/riskassessment/glossary.htm#l (Accessed August 2013). Urdan, T.C., 2010. Statistics in Pla in English. 3rd Edition, Rutledge Taylor & Francis Group, New York, NY, 211 p. van Lonkhuyzen, R.A., Lagory, K.E., Kuiper, J.A., 2004. Modeling the suitability of potential wetland sites with a geographic information system. Environ. Manage. 33, 368 375. van Oudenhover, A.P.E., Petz, K., Alkemade, R., Hein, L., de Groot, R.S., 2012. Framework for systematic indicator selection to assess effects of land management on ecosystem services. Ecol. Indicators 21, 110122. Wang, S., Huang, S., Budd, W., 2012. Integrated ecosystem model for simulating land use allocation. Ecol. Model. 227, 4655. Watson, J.E., 2012. A river loved: facilitating cooperative negotiation of transboundary water resource management in the columbia river basin through documentary film., Wat er Resources Policy & Management. Oregon State University, p. 66. White House Council on Environmental Quality, 2009. Interim Report of the Interagency Ocean Policy Task Force, Washington, DC, p. 38.
246 White House Council on Environmental Quality, 2012. Fina l Report of the Interagency Ocean Policy Task Force, Washington, DC, p. 96. World Commission on Dams, 2000. Dams and Development: A New Framework for Decision Making. Earthscan Publications Ltd., London, N1 9JN, UK. Yamada, K., Elith, J., McCarthy, M., Zer ger, A., 2003. Eliciting and integrating expert knowledge for wildlife habitat modelling. Ecol. Model. 165, 251 264.
247 BIOGRAPHICAL SKETCH Ms. Kelly Burks Copes is a research ecologist working in the Environmental Laboratory at the U.S. Army Engineer Research and Development Center (ERDC) in Vicksburg, MS. She has been with the U.S. Army Corps of Engineers for over 18 years and intends to continue on with the labs into foreseeable future. She is project manager for a series of groundbreaking studies includ ing a study of sea level rise and coastal storm impacts on naval installations worldwide, and a study of potential ecosystem goods and service performance metrics for post Superstorm Sandy recovery efforts on the North Atlantic coast. She is leading a new initiative for the U.S. Navy to conduct a global vulnerability assessment of all naval installations worldwide with respect to sea level rise and coastal storm hazards. She is the principle investigator on two additional studies tying ecosystem goods and s ervices to navigation and operational missions throughout the USACE. She is also the principle investigator on a series of studies utilizing the structured decision making paradigm described herein on studies throughout the country including Galveston, TX, Chicago, IL, Albuquerque, NM, Phoenix, AZ Tucson AZ, Brownsville, TX, and now on the Missouri River. Ms. Burks Copes obtained her Ph.D. from the University of Florida in Multidisciplinary Ecology in 2014, a M.S. from the New Mexico State University in E cology in 1993, and a B.S. from the University of New Mexico in Biology in 1991. She was the recipient of the prestigious ERDC R&D Award four years running (20072011) for her ecosystem restoration and climate change studies, and hopes to continue to contr ibute to the ERDCs pioneering research initiatives into the foreseeable future.