<%BANNER%>

Burn Severity in a Central Florida Sand Pine Scrub Wilderness Area

Permanent Link: http://ufdc.ufl.edu/UFE0022868/00001

Material Information

Title: Burn Severity in a Central Florida Sand Pine Scrub Wilderness Area
Physical Description: 1 online resource (108 p.)
Language: english
Creator: Godwin, David
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: burn, clausa, fire, forest, juniper, management, national, ocala, pine, pinus, prairie, remote, sand, scrub, sensing, severity, wilderness
Forest Resources and Conservation -- Dissertations, Academic -- UF
Genre: Forest Resources and Conservation thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Sand pine scrub is a xeric, upland ecosystem found only in Florida and in isolated regions of coastal Alabama. Threatened by development and cleared for agriculture, sand pine scrub exists only in small pockets, a diminutive legacy of a once vast landscape. The Juniper Prairie Wilderness (JPW), a 56 square kilometer United States federally designated wilderness area in north central Florida represents one of the largest and most protected tracts of sand pine scrub in the state. In August of 2006, a prescribed fire escaped initial prescription and ultimately burned 44 square kilometers across the JPW. To study burn severity within the JPW sand pine scrub, 60 field sampling plots were established following the 2006 fire, within three sand pine stand classes (pole, seedling / sapling and hurricane damaged pole) and across four ocular assessed severity classes. Within each stand type and ocular severity class, five replicates were established to document burn severity using the Composite Burn Index (CBI) and to record stand level vegetative, topographic and fuel loading characteristics. Significant variations of burn severity were found within and among sand pine stand types with most high severity burned area in the pole class stands and the least amount of unburned area in the seedling / sapling stands. The variations of burn severity within the pole, damaged pole, and seedling / sapling stands were found to be inversely associated with sapling percent ground cover. Methods of mapping burn severity using remotely sensed imagery were also developed for the first time in this ecotype. A supervised classification of post burn Landsat imagery resulted in a burn severity map that achieved overall accuracy of 68%. These results provide new insight into the variations of fire effects within sand pine scrub ecosystems. Such understandings, coupled with new burn severity mapping methods, provide those tasked with managing sand pine scrub improved predictive and assessment tools for managing and perpetuating a unique ecotype.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by David Godwin.
Thesis: Thesis (M.S.)--University of Florida, 2008.
Local: Adviser: Kobziar, Leda Nikola.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2010-12-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022868:00001

Permanent Link: http://ufdc.ufl.edu/UFE0022868/00001

Material Information

Title: Burn Severity in a Central Florida Sand Pine Scrub Wilderness Area
Physical Description: 1 online resource (108 p.)
Language: english
Creator: Godwin, David
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: burn, clausa, fire, forest, juniper, management, national, ocala, pine, pinus, prairie, remote, sand, scrub, sensing, severity, wilderness
Forest Resources and Conservation -- Dissertations, Academic -- UF
Genre: Forest Resources and Conservation thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Sand pine scrub is a xeric, upland ecosystem found only in Florida and in isolated regions of coastal Alabama. Threatened by development and cleared for agriculture, sand pine scrub exists only in small pockets, a diminutive legacy of a once vast landscape. The Juniper Prairie Wilderness (JPW), a 56 square kilometer United States federally designated wilderness area in north central Florida represents one of the largest and most protected tracts of sand pine scrub in the state. In August of 2006, a prescribed fire escaped initial prescription and ultimately burned 44 square kilometers across the JPW. To study burn severity within the JPW sand pine scrub, 60 field sampling plots were established following the 2006 fire, within three sand pine stand classes (pole, seedling / sapling and hurricane damaged pole) and across four ocular assessed severity classes. Within each stand type and ocular severity class, five replicates were established to document burn severity using the Composite Burn Index (CBI) and to record stand level vegetative, topographic and fuel loading characteristics. Significant variations of burn severity were found within and among sand pine stand types with most high severity burned area in the pole class stands and the least amount of unburned area in the seedling / sapling stands. The variations of burn severity within the pole, damaged pole, and seedling / sapling stands were found to be inversely associated with sapling percent ground cover. Methods of mapping burn severity using remotely sensed imagery were also developed for the first time in this ecotype. A supervised classification of post burn Landsat imagery resulted in a burn severity map that achieved overall accuracy of 68%. These results provide new insight into the variations of fire effects within sand pine scrub ecosystems. Such understandings, coupled with new burn severity mapping methods, provide those tasked with managing sand pine scrub improved predictive and assessment tools for managing and perpetuating a unique ecotype.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by David Godwin.
Thesis: Thesis (M.S.)--University of Florida, 2008.
Local: Adviser: Kobziar, Leda Nikola.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2010-12-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022868:00001


This item has the following downloads:


Full Text

PAGE 1

1 BURN SEVERITY IN A CENTRAL FLORID A SAND PINE SCRUB WILDERNESS AREA By DAVID ROBERT GODWIN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2008

PAGE 2

2 2008 David Robert Godwin

PAGE 3

3 To my family and grandparents who have worked tirelessly to provide for my education and who instilled in me an interest in the natural world.

PAGE 4

4 ACKNOWLEDGMENTS This study would not have been possible without the encouragem ent, support, and guidance of my major professor (Dr. Leda Kobziar) and my supervisory committee members (Dr. Scot Smith and Dr. George Tanner). This research was funded by a grant from the University of Florida Institute of Food and Agricultural Sciences (IFAS) Innovation Fund entitled: Fire in The Juniper Prairie Wilderness : A Viable Management Tool? Data analysis assistance was graciously provided by Dr. Leda Kobziar and Meghan Brennan, IFAS Statistics Department. Field data were co llected through the tireless assistan ce of University of Florida Fire Science Lab technicians and students: Er in Maehr, Mia Requensens, Chris Kinslow and Cori Peters. Remote sensing software gui dance and meticulous digital aerial imagery delineations were provided by Zoltan Szantoi, w ith help from Dr. Alan Long and Dr. Leda Kobziar. Some spatial data were provided by the Ocala National Forest, USDA Forest Service, through the assistance of Mike Drayton and Ja net Hinchee. SPOT images were purchased through funds provided by Dr. Scott Smith. Lands at images were obtained from the MultiResolution Land Characteristics Consortium (MRLC). Finally, I thank my loving wife, brother, pa rents and grandparents for their patience, encouragement and dedication to my completion of this study.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ............................................................................................................... 4LIST OF TABLES ...........................................................................................................................7LIST OF FIGURES .......................................................................................................................10ABSTRACT ...................................................................................................................... .............12 CHAP TER 1 BURN SEVERITY MAPPING METHODS FOR USE IN SAND PINE SCRUB ............... 14Introduction .................................................................................................................. ...........14Background .................................................................................................................... .........15Study Area ..............................................................................................................................17Methods ..................................................................................................................................18Image Sources ................................................................................................................. 18Burn Severity Plots .......................................................................................................... 20Classifications ............................................................................................................... ...21Accuracy Assessment ...................................................................................................... 23Results .....................................................................................................................................24Image Source Accuracy ................................................................................................... 24Classification Accuracy ...................................................................................................24Class Accuracy ................................................................................................................25Discussion .................................................................................................................... ...........26Conclusions .............................................................................................................................302 STAND LEVEL CHARACTERISTICS AND B URN SEVERITY VARIATIONS FOLLOWING A CENTRAL FLORIDA WILDERNESS AREA FIRE ............................... 53Introduction .................................................................................................................. ...........53Study Area ..............................................................................................................................58Methods ..................................................................................................................................60Plot Description .............................................................................................................. .60Plot Level Burn Severity ................................................................................................. 61Remotely Sensed Burn Severity ...................................................................................... 61Plot Topographic and Vegetative Char acteristics and Fuel Loading .............................. 63Analysis ...........................................................................................................................63Results .....................................................................................................................................65Variations in Burn Severity Am ong Sand Pine Scrub Stand Types ................................ 65Variations in Topographic, Vegetative and Fuels Character istics Among Stand Types ......................................................................................................................... ...66Modeling Burn Severity Using Stand Characteristics .....................................................67

PAGE 6

6 Discussion .................................................................................................................... ...........69Conclusions .............................................................................................................................723 HIGH SEVERITY PRESCRIBED FIRE IN FLORIDA: SAND PINE SCRUB MANAGEMENT IN THE OCALA NATIONAL FOREST .................................................90Introduction .................................................................................................................. ...........90Sand Pine Scrub ......................................................................................................................92Sand Pine Scrub Management at the ONF .............................................................................95Challenges to Prescribed Fire in Sand Pine Scrub .................................................................. 97Experiments in Sand Pine Fire Management ........................................................................ 101Fire and the Future Management of Sand Pine Scrub .......................................................... 102Research Needs .....................................................................................................................103LIST OF REFERENCES .............................................................................................................104BIOGRAPHICAL SKETCH .......................................................................................................108

PAGE 7

7 LIST OF TABLES Table page 1-1 Descriptions of ocular severity classes used to quantify variation in burn severity in the Juniper Prairie W ilderness, Oc ala National Forest, Florida. ........................................311-2 Descriptions of Composite Burn Index ( CBI) burn severity indices used to quantify plot level burn severity in the Juniper Prairie Wilderness, Ocala National Forest, Florida. ...................................................................................................................... .........311-3 Burn severity classification accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florida, in cluding all severity indi ces and classifications. ...................... 311-4 Overall map accuracy best ten burn severi ty classifications for the Juniper Prairie Wilderness, Ocala National Forest, Florida, including all severity indices, image sources and classifications. ................................................................................................ 321-5 Mean map accuracy best ten burn severity classifications for the Juniper Prairie Wilderness, Ocala National Forest, Florida, including all severity indices, image sources and classifications. ................................................................................................ 321-6 Landsat Supervised burn severity cla ssification and ocular severity index error matrix for the Juniper Prairie Wilderne ss, Ocala National Forest, Florida. ...................... 331-7 Means of burn severity classification mean producers and users accuracies by severity class from all image sources a nd classifications; for the Juniper Prairie Wilderness, Ocala National Forest, Florida. ...................................................................... 331-8 Unburned class best ten burn severity classifications by users accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florid a, including all severity indices, image sources and classifications. ........................................................................ 341-9 Light severity class best ten burn severity classifications by users accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florid a, including all severity indices, image sources and classifications. ........................................................................ 341-10 Mid severity class best ten burn severity classifications by users accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florid a, including all severity indices, image sources and classifications. ........................................................................ 351-11 High severity class best te n burn severity classifications by users accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florid a, including all severity indices, image sources and classifications. ........................................................................ 351-12 Unburned class best ten burn severity cl assifications by producers accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florid a, including all severity indices, image sources and classifications. ........................................................................ 36

PAGE 8

8 1-13 Light severity class best ten burn severity classifications by producers accuracy for the Junip er Prairie Wilderness, Ocala National Forest, Florida, including all severity indices, image sources and classifications. ........................................................................ 361-14 Mid severity class best ten burn severity classifications by producers accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florida, including all severity indices, image sources and classifications. ........................................................................ 371-15 High severity class best ten burn severity classificati ons by producers accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florida, including all severity indices, image sources and classifications. ........................................................................ 371-16 Burn severity classification accuracy lis ted by burn severity index for the Juniper Prairie Wilderness, Ocala National Forest Florida, across all image sources. ................. 381-17 Means of burn severity classification mean producers and users accuracies by severity class from all image sources a nd classifications; for the Juniper Prairie Wilderness, Ocala National Forest, Florida. ...................................................................... 381-18 Best burn severity classification mean producers and users accuracies by severity class and classification for the Juniper Pr airie Wilderness, Ocala National Forest, Florida. ...................................................................................................................... .........382-1 Pinus clausa stand types sampled in the Junipe r Prairie Wilderness, Ocala National Forest, Florida. ...................................................................................................................742-2 Descriptions of ocular severity classes used to quantify variation in burn severity in the Juniper Prairie Wilderness, Oc ala National Forest, Florida. ........................................742-3 Descriptions of burn severi ty indices used to quantify pl ot level burn severity in the Juniper Prairie Wilderness, Ocal a National Forest, Florida. .............................................752-4 Descriptions and differences of sample plot variables among severity classes and stand types in the Juniper Prairie Wilderness of the Ocala Nationa l Forest, Florida. ....... 762-5 Differences of burn severity indices among ocular severity classes and stand types in the Juniper Prairie Wilderness of th e Ocala National Forest, Florida. ..............................772-6 Burn severity means and standard devi ations recorded in the Juniper Prairie Wilderness, Ocala National Forest, Florida. ...................................................................... 782-7 Spatial area and percent of the area for each burn severity class listed by sand pine scrub stand type within the Juniper Prairi e Wilderness of the Ocala National Forest, Florida. ...................................................................................................................... .........792-8 Spatial area and percent of the total area for each burn severity class in the Juniper Prairie Wilderness of the Ocal a National Forest, Florida. .................................................79

PAGE 9

9 2-9 Sample plot variable means and standard deviations for plot level variables selected for burn severity regression m odeling. .............................................................................. 802-10 Burn severity regression model results ba sed on data collected in the Juniper Prairie Wilderness, Ocala National Forest, Florida. ...................................................................... 812-11 Sample plot variable means and standard deviations for plot level variables not selected for burn severity regression modeling. ................................................................ 82

PAGE 10

10 LIST OF FIGURES Figure page 1-1 Location of Pinus clausa stands and sam ple plots in the Juniper Prairie Wilderness of the Ocala National Forest, Florida, USA. .......................................................................... 391-2 Digital Aerial classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocal a National Forest, Florida. .................................................401-3 SPOT 432 Unsupervised classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of th e Ocala National Forest, Florida. ..............................411-4 SPOT 421 Unsupervised classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of th e Ocala National Forest, Florida. ..............................421-5 Landsat 654 Unsupervised classificati on map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. .......................... 431-6 SPOT NDVI Unsupervised classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. .......................... 441-7 Landsat NDVI Unsupervised classificat ion map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. ....................451-8 Landsat NBR Unsupervised classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. .......................... 461-9 Landsat dNBR 11 Classes classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. .......................... 471-10 Landsat dNBR 17 Classes classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. .......................... 481-11 SPOT Supervised classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. .................................... 491-12 Landsat Supervised classi fication map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. .................................... 501-13 Distributions of burn severity classification mean map accuracies across all severity indices, by classification image source fo r the Juniper Prairie Wilderness, Ocala National Forest, Florida. ....................................................................................................511-14 Distributions of burn severity classification overall map accuracies across all severity indices, by classification image source fo r the Juniper Prairie Wilderness, Ocala National Forest, Florida. ....................................................................................................52

PAGE 11

11 2-1 Location of Pinus clausa stands and sam ple plots in the Juniper Prairie Wilderness of the Ocala National Forest, Florida, USA. .......................................................................... 842-2 Sample plots for recording burn severity, fuel loading, topographic and vegetative characteristics of sand pine scrub stands in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. ..........................................................................................852-3 Map of burn severity following a 2006 fire in the Juniper Prai rie Wilderness of the Ocala National Forest, Florida. ..........................................................................................862-4 Percent of area burned for each severity class by Pinus clausa stand type within the Juniper Prairie Wilderness, Ocal a National Forest, Florida. .............................................872-5 Multiple regression model output using plot level data and CBI burn severity values derived from a supervised image classification ................................................................. 882-6 Plot of correlation between plot Percent Sapling Cover and plot Total Percent Tree Cover recorded at the Juni per Prairie Wilderness, Ocal a National Forest, Florida ...........89

PAGE 12

12 Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science BURN SEVERITY IN A CENTRAL FLORID A SAND PINE SCRUB WILDERNESS AREA By David Robert Godwin December 2008 Chair: Leda Kobziar Major: Forest Resources and Conservation Sand pine scrub is a xeric, upland ecosystem found only in Florida and in isolated regions of coastal Alabama. Threatened by developmen t and cleared for agricu lture, sand pine scrub exists only in small pockets, a diminutive legacy of a once vast landscape. The Juniper Prairie Wilderness, a 56 square kilometer United States federally designated wilderness area in north central Florida represents one of the largest and most protected tracts of sand pine scrub in the state. In August of 2006, a prescribed fire escap ed initial prescription and ultimately burned 44 square kilometers across the JPW. To study bur n severity within the JPW sand pine scrub, 60 field sampling plots were established followi ng the 2006 fire, within three sand pine stand classes (pole, seedling / sapling and hurricane damaged pole) and across four ocular assessed severity classes. Within each stand type and ocular severity class, five replicates were established to document burn severity using the Co mposite Burn Index (CBI) and to record stand level vegetative, topographic and fu el loading characteri stics. Significant variations of burn severity were found within and among sand pine st and types with most high severity burned area in the pole class stands and th e least amount of unburned area in the seedling / sapling stands. The variations of burn severity w ithin the pole, damaged pole, and seedling / sapling stands were found to be inversely associated with sapling percent ground cove r. Methods of mapping burn

PAGE 13

13 severity using remotely sensed imagery were also developed for the first time in this ecotype. A supervised classification of post burn Landsat im agery resulted in a burn severity map that achieved overall accuracy of 68%. These results pr ovide new insight into the variations of fire effects within sand pine scrub ecosystems. Such understandings, coupled with new burn severity mapping methods, provide those tasked with managing sand pine scrub imp roved predictive and assessment tools for managing and perpetuating a unique ecotype.

PAGE 14

14 CHAPTER 1 BURN SEVERITY MAPPING METHODS FOR USE IN S AND PINE SCRUB Introduction Sand pine scrub is a d iminishing stand-replacement fire type ecosystem endemic to the state of Florida and small regions of coastal Alabama. The Ocala National Forest (ONF) is home to the largest remaining tract of sand pine ( Pinus clausa (Chapman ex Engelm.) Vasey ex Sarg.) scrub in the world. The United States Forest Se rvice (USFS) is tasked with managing the sand pine scrub of the ONF for multiple uses, includi ng the perpetuation of the ecotype and providing habitat for threatened and endangered species. Fire is common in the ONF, with burning occurring as both wildfires and carefully planne d prescribed fires for management purposes. Post-fire mapping is a tool that the managers of the ONF use to determine annual prescribed burning plans and as well as to a ssess the results of prescribed fires and wildfires. Current fire mapping methods used by the ONF are designed onl y to quantify the spatial extent of burned and unburned areas. Burned area mapping methods such as this fail to prov ide any quantitative or qualitative information describing the differential im pacts of fire on the lands cape. Burn severity can be delineated, quantified and mapped using remotely sensed imagery and software, as demonstrated after many large fires in other forested ecosystems (Wulder and Franklin, 2007). Knowledge of burn severity can be influen tial in post-fire management decisions as demonstrated by burn severity mapping programs in the western United States that are utilized by inter-agency burned area emerge ncy response (BAER) teams fo llowing wildfires. Methods of mapping burn severity in sand pine scrub woul d allow ONF managers to make more informed post burn management decisions. For example, bur n severity maps following fire in sand pine scrub could be used to assess the landscape for areas of vegetation sufficiently impacted by fire as to provide suitable habitat conditions for threatened faunal species.

PAGE 15

15 This study investigates methods of mapping sand pine scrub burn severity within the ONF. To study this topic, the following re search questions are addressed: Which sand pine scrub burn severity mapping method is the most accurate? Which burn severity mapping method is best suited for adoption by the ONF staff for management of the JPW? To address these questions, three image sources were selected to investigate burn severity mapping: digital aerial photographs SPOT 4 satellite images, and La ndsat TM satellite images. These sources were chosen as they represent a variety of traditional and contemporary methods of burn severity mapping, image acquisition costs, resolutions and user te chnical requirements. To answer the first study questi on, accuracy of the burn severity classifications was assessed by comparison to quantitative burn seve rity values recorded in field sample plots. It was believed that the digital aerial photographs, classified using heads-up-digitizing (HUD) would provide the most accurate burn severity map for two reasons. First, of all the image sources tested, the digital aerial photographs were captu red closest to the actual date of the burn, presumably before any post-fire vegetation re-growth. Second, the di gital aerial photographs were classified using HUD which was the most user intensive classi fication method employed in this study and is thought to be more adept at delineating burn sever ity patterns. To answer the second question, a discussion of accuracy, mapping cost, equipment, and software knowledge requirements were considered in terms of manageme nt applications for the JPW. Background One of the prim ary benefits of using remo tely sensed imagery for burn mapping is the ability to detect landscape level variations in vegetation changes due to fire (Clark and Bobbe, 2007). Traditional fire mapping methods have involved management personnel physically walking the burned area perimeter with a GPS unit or flying over the burned area and manually

PAGE 16

16 sketching the fire perimeter. These methods can be both costly and time consuming and may not be feasible on very large fire s or in rough terrain (Henry, 2008) In addition, such on-the-ground methods offer little quan titative data describing the variations and effects of heterogeneous fire behavior and impacts. Remote sensing offers the capability of ma pping fires using multispectral images: images that record electromagnetic spectrum wavelengt hs reflected beyond the visible spectrum. Groundcover reflects differing amounts of electromagnetic radiation (Wulder and Franklin, 2007). In sand pine scrub, variations in reflectance values have been used to differentiate burned and unburned vegetation and to deli neate burned areas at landscap e scales (Henry, 2008). In other forest types, variations in reflectance values within burned areas have been used to quantify the severity of burns (Clark and Bobbe, 2007). This study seeks to adopt similar techniques for application in burn severity mapping within sand pine scrub. To our knowledge there exists no previous published works in burn severity mapping in sand pine scrub. There are a number of common burn severity classification technique s that can be used when working with multi-spectral satellite based imagery. These techniques include (the): Kauth-Thomas transformation (Tasseled Cap), pr inciple component analysis (PCA), neural networks, normalized differential vegetation index (NDVI), normalized burn ratio and delta normalized burn ratio (NBR/dNBR), and the delta normalized differential vegetation index (dNDVI) (Henry, 2008; Clark and Bobbe, 2007). Most classification techniques utilize differences in near infrared (NIR) reflectance values to identify and quantify changes in vegetation and soils following a fire. Hea lthy green vegetation tends to reflect NIR electromagnetic radiation (EMR) and absorb red band EMR. Following a fire, changes in vegetation structure and productiv ity often result in lower NIR reflectivity and higher red band

PAGE 17

17 reflectivity (Jensen, 2005). Additionally, increased soil exposure increases reflectivity in the middle infrared (MID-IR) bands. Combinations a nd variations of these band differentiations are the basis for the NBR and NDVI classification techniques. Analysis using multi-temporal datasets are the basis for the dNBR and dNDVI t echniques, which look at temporal changes in reflectance values from images captured pr e and post-burn (Henry, 2008; Clark and Bobbe, 2007; Jensen, 2005). NBR and dNBR are two of the most popular burn severity classification methods employed for mapping using multispectra l imagery. Map accuracies of 50-80% have been achieved in four level (unburned, low, mode rate, high) classification schemes (Clark and Bobbe, 2007; Epting et al. 2005; Cocke et al. 2005). Study Area This study takes place within the Juni per Prairie W ilderness (JPW), a 56 km2 United States federally designated wilderness area in the ONF in north central Florida (Figure 1-1). The JPW represents a portion of one of the largest and most protected tracts of sand pine scrub in the state (Greenburg, 1996). Federally desi gnated as a wilderness area in 1984, the JPW falls under the resource management protection of the United States 1964 Wilderness Act (The Wilderness Act, 1964). An endemic natural community, sand pine sc rub is found only in Florida and in a small coastal region in southern Alabama (Myers, 1985). Threatened by development and cleared for agriculture, sand pine scrub now exists only in small pockets, a diminutive legacy of a once vast landscape that stretched along the central highlands of the Flor ida peninsula (Myers, 1985). Within the JPW, sand pine scrub is the dominant natural community type. Sand pine scrub is typically dominated by an even -aged monoculture overstory of sand pine. Sand pine scrub regions of inland peninsular Fl orida tend to be comprised of stands of fire-dependent, bradysporous or serotinous Ocala sand pine (P. clausa var. clausa D.B. Ward) while regions of

PAGE 18

18 coastal sand pine scrub tend to be comprised of stands of fire independent Choctahatchee sand pine ( P. clausa var. immuginata D.B. Ward). The understory and midstory are generally dominated by scrubby sclerophyllic oaks, ( Quercus. chapmanii Sarg. Q. myrtifolia Willd Q. geminate Small), saw palmetto ( Serenoa repens (Bart.) Small), scrub palmetto ( Sabal etonia Swingle ex Nash), tree lyonia ( Lyonia ferruginea (Walt.) Nutt.), red bay ( Persea humilis Nash) and Florida rosemary ( Ceratiola ericoides Michx.) (Outcalt and Greenberg, 1998; Greenberg, 1996; Menges et al. 1993; Myers, 1985; Veno, 1976). In late July of 2006, the ONF staff initiated pres cribed fire operations within the JPW with the intent of burning a number of tall-grass prairies. Within several days the prescribed fires escaped the initial prescriptions and were reclas sified as a wildfire. The resulting wildfire continued into August of 2006 and ultimately burned an 44 km2 across the JPW. The fire burned a portion of nearly all of the vegetative commun ities represented in the JPW and across multiple sand pine stands. The burning of such large tr acts of sand pine scr ub, under wilderness area management restrictions, presented a unique opport unity to study burn severity within a variety of sand pine scrub stand types. Methods Image Sources Our study used three rem otely sensed image s ources to derive a total of 13 different burn severity classifications. Those classifications were assessed for accuracy based on agreement with burn severity plots es tablished within the JPW. Satellite Pour L observation de la Terre (SPOT 4) SPOT 4 multi-spectral (MS) and panchrom atic (PAN) scenes captured using the HRVIR2 instrument on September 25, 2006, were obtained from SPOT Image Corp. Inc. at a pre-processing level 2A. The MS scenes contai ned three visible spectrum bands ranging from

PAGE 19

19 0.50-0.89 m and one shortwave infrared (SWIR) band ranging from 1.58 1.75 m. The study area exists across the two image scenes (which were mosaicked and subsetted). The MS and PAN images were re-sampled to 10 m resoluti on using the ERDAS Imagine High Pass Filter (HPF) Merge tool. This proce ss uses the 10 m spatial resolution PAN image to resample the 20 m MS image to 10 m resolution. The re-sampled images were geometrically corrected using the nearest-neighbor function using 1 m spatial resolution dig ital ortho quarter-quads (DOQQs) (Salt Springs / 4314 and Juniper Springs / 4214) from the Florida Department of Environmental Protection (FL DEP) Land Boundary Information System (LABINS). Landsat 5 TM A post-burn, Landsat 5 TM scene (path 16, row 40) captured on November 19, 2006 using the Thematic Mapper (TM) instrument wa s obtained from the Multi-Resolution Land Characteristics Consortium (MRLC). A preburn Landsat 5 TM scene (path 16, row 40) captured September 13, 2005, was also obtained from the MRLC. Both scenes were recorded at 30 m spatial resolution and contained six ba nds, ranging from 0.45-2.35 m. Landsat band six, thermal (10.40-12.5 m), was omitted in MRLC preprocessing and band seven (2.08-2.35 m) subsequently referred to as band six. Both s cenes were preprocessed (level 10) by the MRLC, which included radiometric and geometric correc tions and were packaged with MRLC derived REFL (at satellite reflectance) and NBR (norma lized burn ratio) scenes. These images were nearest-neighbor geometrically corrected using the Salt Springs and Juniper Springs DOQQs acquired from LABINS. Digital Aerial Photograph Mosaic In the days following the end of the JPW fire a light, single-engine, fixed-wing airplane flew multiple paths over the JPW, with an ONF staff member ta king a series of 228 over-lapping

PAGE 20

20 digital photographs using a consumer grade, handhel d, digital camera. It should be noted that the mosaic images were acquired by the ONF prior to th e initiation of this proj ect. Details regarding the flight plans, weather and atmospheric c onditions at the time of the flights were not documented. Each image was printed, given an ID number and manually overlaid as a mosaic of the burned area. This mosaic of printed images was replicated using the original digital files to produce a digital aerial photograph mosaic of the JPW. The mosaic was geometrically corrected using the Salt Springs and Juniper Springs DOQQs using ESRI ArcGIS 9.2. Burn Severity Plots To quantify the accu racy of the burn severity cl assifications, a series of burn severity plots was established within burned and unburned sand pine scrub stands, in the winter and spring following the 2006 JPW burn. Plot data were coll ected following an adaptation of the Key and Benson FIREMON: Landscape Assessment sampling approach (Key and Benson, 2006). This method uses the Composite Burn Index (CBI) to sample and quantify burn severity at the plot level (Key and Benson, 2006). The CBI was deve loped as a consistent and adaptable rapid assessment tool for collecting field data and also as a method for the analysis of remotely sensed burn severity maps derived using the NBR (Key and Benson, 2006). Plots were established in stands visually identified at f our levels of burn severity: unbur ned, light severity, mid severity and high severity (Table 1-1). Within each burn severity type, 15 plots were established, giving a total of 60 field plots for the entire study area (Figure 1-1). Plots were randomly established with a minimum of 100 m from any edges (inclu ding edges between ocular severity levels and other vegetation or stand types) a nd with a minimum of 100 m separa tion between plots. At each 10 x 10 m plot, burn severity was visually assessed across seven vertical layers of vegetative strata (substrate, herbaceous, tall shrubs, understory, intermediate trees, big trees, overstory) as

PAGE 21

21 described in Landscape Assessment (Key and Be nson, 2006)(Table 1-2). The seven CBI strata values combine to give a total, overall plot severity value referred to as CBI Total. These eight quantified, field derived severities, along with the ocular severities used to locate the plots, were used individually to assess the accuracies of classifications. Classifications The digital aerial m osaic (F igure 1-2) was delineated into four burn severity classes (unburned, light severity, mid se verity, high severity) by hand using HUD in ERDAS Imagine 9.1. The delineated classes were identified with the assistance of fire professionals experienced in traditional burn mapping techniques, wherein burn severity level perimeters are manually delineated and traced on aerial photographs. The SPOT unsupervised classifications (Figure 1-3 and Figure 1-4) were created using a hybrid unsupervised classificati on method in ERDAS Imagine 9.1 (Erdas, Inc., Atlanta, GA, USA). The SPOT 432 and 431 unsupervised cla ssifications were de rived using RBG band combinations: 4-3-2 and 4-3-1. Both SPOT ba nd combinations were classified in ERDAS Imagine 9.1 using an iso-data unsupervised clas sification running a minimum of 30 iterations and resulting in 20 output classes. The resulting classes were merged using HUD into four burn severity classes (unburned, light sever ity, mid severity, high severity). The Landsat unsupervised classification (Fi gure 1-5) used the 2006 post-fire Landsat image (with RGB band combination: 6-5-4) in the ERDAS Imagine 9.1 iso-data unsupervised classification tool (running a minimum of 30 ite rations and producing 20 output classes). The resulting 20 classes were HUD merged into four burn severity classes (unburned, light severity, mid severity, high severity). The SPOT NDVI (Figure 1-6) image was derive d using the RSI ENVI 4.3(ITT Industries Inc., Boulder, CO, USA) NDVI tool. The re sulting NDVI image was classified in ERDAS

PAGE 22

22 Imagine using the iso-data unsupervised classifi cation tool, running a minimum of 30 iterations and resulting in 20 output classes. The 20 outpu t classes were HUD merged into four severity classes (unburned, light severit y, mid severity, high severity). The Landsat NDVI hybrid unsupervised classifi cation (Figure 1-7) was derived from the Landsat 2006 image using the NDVI calculator f unction in RSI ENVI 4.3. The resulting NDVI image was classified through a hybrid unsupervised method using the ERDAS Imagine tool (running a minimum of 30 iterations and producing 20 output classes). The 20 output classes were merged into four burn severity classes (unburned, light severi ty, mid severity, high severity) using HUD. The Landsat scenes provided by the MRLC pr ogram come prepackaged with NBR derived images (Key and Benson, 2006) (Figure 1-8). These images were classified using a hybrid unsupervised classification technique in ER DAS Imagine 9.1. An unsupervised iso-data classification running a minimum of 30 itera tions and producing 20 output classes was performed. The resulting cla sses were merged using HUD into four burn severity classes (unburned, light severity, mid severity, high severity). The Landsat dNBR image is a NBR difference calculation created by subtracting an unburned NBR image from a burned NBR image (Key and Benson, 2006). This change detection image calculation was run in RSI ENVI 4.3 using the Landsat 2006 NBR image as the post-fire values, and the Landsat 2005 NBR image as the pre-fire values. The change detection output was run twice, producing one image with 11 change output classes and another image with 17 output change classes (Figure 19 a nd Figure 1-10). The change classes from each image subtraction were HUD merged into four bu rn severity classes ( unburned, light severity, mid severity, high severity).

PAGE 23

23 The SPOT Supervised classifica tion was derived from HPF resa mpled SPOT 4 images that were classified using a minimum-distance supervised classification usi ng RSI ENVI 4.3 (Figure 1-11). Three burn severity training sample plots from each severity level, selected at random, were used in the classification. These points were used to derive a four class burn severity classification (unburned, light seve rity, mid severity, high severity ). The training sample points used in the SPOT supervised classification were us ed in the Landsat supervised classifications as well. The training samples chosen for the supe rvised classifications were not used in the accuracy analysis of the supervised image. The Landsat Supervised classification postfire image was derived through a minimumdistance supervised classification using RSI ENVI 4.3(Figure 1-12). Three burn severity training sample plots were chosen at random to represent each severity level. The Landsat supervised classification used the same random training sample plots used in the SPOT supervised classification. The training sample s were not used in the accuracy analysis of the supervised image. The supervised classification resulted in a four severity leve l image (unburned, light severity, mid severity and hi gh severity) classification. Accuracy Assessment Classificatio n accuracy analysis was conducted for each image using an error matrix with the category level accuracy statistics: overall accuracy, mean accuracy, individual class users accuracy, and individual class producers accuracy (Jensen, 2005). The producers accuracy is a measure of how well the classes were assigned during the classifi cation process. The users accuracy is a probability measure of the correctness of the classes on the map. For select classifications, kappa (KHAT) statistic was calculat ed, a measure that indi cates the classification departure from a random pixel assignment (Liu et al. 2007; Cocke et al. 2005; Congalton and Green, 1999). Errors of classification accuracy were determined by the misclassification of

PAGE 24

24 pixels to severity classes not re flected by values recorded in the CBI and ocular severity sample plots. Classification accuracy an alysis was conducted for each cl assification using each of the nine severity indices individually. For select classifications, correlations between classification severity values and severity index values were explored using statistics software JMP 7.0.1 (SAS Institute Inc., Cary, NC, USA). Results Image Source Accuracy The classifications derived from Landsat imagery had the highest mean overall map accuracy (49.6%) and mean map accuracy (49.9 %) (across all classifications and severity indices) (Table 1-3). The digital aerial photog raph mosaic image source had the lowest mean map accuracy (35.8%) and the lowest overall map accuracy (40.2%) (Table 1-3). Of the ten most accurate classifications by overall map accur acy, across all severity types, eight of the classifications were based on Landsat imagery (Table 1-4). Of the ten most accurate classifications by mean map accuracy, across all seve rity types, nine were derived from Landsat imagery (Table 1-5). A histogram analysis of both overall map accuracy and mean map accuracy indicated that the Landsat and SPOT de rived classifications consistently had higher mean accuracy and overall accuracy than the Di gital Aerial classifica tion (Figure 1-13 and Figure 1-14). Classification Accuracy The m ost accurate classification, across all se verity indices and all image sources, as measured by the overall map and mean map accurac y, was the Landsat supervised classification (Table 1-4 and Table 1-5). Using the ocular severity index values for accuracy assessment, the Landsat supervised classificati on had an overall accuracy and m ean map accuracy of 68.8%. The kappa (KHAT) statistic was 0.58, indicating th at the classification was significantly better

PAGE 25

25 than a random pixel assignment (Congalton and Green, 1999) (Table 1-6). The Landsat NBR unsupervised classification was the second most accurate classification measured by both the mean map accuracy and overall map accuracy (65 %) (Table 1-4 and Table 1-5). By comparison, the digital aerial HUD classificat ion was not in any of the top ten most accurate classification tables (Table 1-4 and Table 1-5). The most accu rate Digital Aerial classification by mean map accuracy was the CBI Herbs assessment (45.7%) a nd the most accurate classification by overall map accuracy was the CBI Tall Shrubs assessment (51.7%). Class Accuracy Differences in burn severity class accuracy were found (Table 1-7). Across all im age classifications and severity index assessmen ts, the unburned class ha d the highest accuracy, indicating that the classi fications were most adept in deli neating unburned sand pine scrub from burned sand pine scrub. Mean accuracies for the other severity cla sses varied with respect to the accuracy test (producers accuracy or users accu racy). In terms of us ers accuracy, across all severity indices and classifications, the ranking of accuracies from highest to lowest was: unburned (89.9%), light severity (37.2%), mid se verity (35.6%), high severity (24.1%) (Table 17). In terms of mean produce rs accuracy, across all severity indices and classifications, the ranking of accuracies from highest to lowest was as follows: unburned (74.20%), mid severity (56.5%), high severity (37.3%), low severity (22.6%) (Table 1-7). The top ten highest classification accuracies by severi ty class, listed by producers accu racy and users accuracy are listed in (Tables 1-8:15). For the Landsat Supervised classification, which was the most accurate classification overall, the producers and users accuracies fo r each severity (as assessed using the ocular assessment) were: producers: unburned (100%), mid severity (75%), high severity (75%), light

PAGE 26

26 severity (25%) and users: light severity (100%), unburned (92.3%), high severity (69.2%), mid severity (47.4%) (Table 1-6). Discussion Which Sand Pine Scrub Burn Severity Mapping Method is the Mos t Accurate? The Landsat imagery, when classified using a supervised classifi cation method, provided the most accurate classification of all methods assessed in this st udy (68% mean map accuracy). These results are within the ra nge of accuracies (50-80%) achie ved in other studies conducted within different ecosystems that developed four le vel burn severity classifications using remotely sensed imagery (Clark and Bobbe, 2007; Epting et al. 2005; Cocke et al. 2005). The hypothesis, that the digital aerial photographs classified using HUD, would provide the most accurate burn severity map, was not supported by the results of this study. The results demonstrated that Landsat images, even using other classification methods, consistently provided the most accurate image source for mapping burn severity in sand pine scrub. The four band SPOT multi-spectral imagery proved to be less accurate than the Landsat imagery, despite having higher spatial resolution (10 m as compared to 30 m La ndsat). The digital aerial photograph HUD classification, which had the highest spatial resolution (although somewhat variable due to the flight patterns) and lowest spectr al resolution, had the lowest accuracy of all the classifications. The results were surprising given that HUD of digital imagery (s imilar in resolutions to the digital aerial mosaic) is a commonly applied and highly regarded method for delineating land use and land cover patterns (LULC) in Florida ecosystems (S Smith, pers. comm ., 2008). The Florida LULC classifications produced using th e HUD method have demonstrated to be more accurate that automated cla ssification methods (supervised or unsupervised) (S Smith, pers. comm., 2008). It is important to note however that these HUD LULC classifications are typically performed on non-multispectral imagery and are often delineations of patterns that are

PAGE 27

27 not as fragmented or spatially variable as the burn severity patterns encountered during this study. Such spatial variability may be underlyi ng the lower accuracy of the HUD method when delineating burn severity patterns. The digital aerial photograph mosaic may have been limited by coverage gaps (holes) within the imagery. The gaps were the result of variations in the aircra ft flight path and the amount of overlap between the images. Com pounding the irregular im age arrangement, were atmospheric distortions caused by clouds and smoke present during the days of the digital aerial acquisition flights. Gaps, distor tions and cloud shadows may have led to the misclassification of some severity classes during the HUD process. Methods of adjusting image contrast may have increased the delineations betw een classes in the digital ae rial photograph; however these contrast adjustments may have also led to exaggerated severity classifications. It is speculated that when mapping burn sever ity in regions with less heterogeneity and texture among classes, the digital aerial HUD classification method might prove more competitive in terms of accuracy, against the Landsat supervised and unsupervised methods. The burn severity classifications of the JPW show extremely textured and complex burn severity patterns: potentially the resu lt of burning during fluctuating weather conditions and active suppression techniques that invo lved back-firing. These complex variations may be better differentiated and delineated thr ough automated computer classifi cations than through the eyes of a technician conducting a HUD cla ssification. It is also possibl e that temporal resolution may have affected the conclusions of this study. Th e effects of fire on ve getation are temporally dependent and change as the time since fire increases. The Landsat post-fire imagery was captured in November, the latest acquisition of al l the image sources used. This late acquisition was actually the closest, temporally, to the da tes during which the field severity plots were

PAGE 28

28 established. It is possible that the accuracies obtained by the Landsat imagery were in part a reflection of that temporal image / plot synchrony. Using the CBI sample plots as a method of cla ssification assessment raises questions of the efficacy of the CBI to assessing burn severity in sand pine scrub vegetation. The CBI was developed in the western US within forested vegetation types different in composition and structure from what is found in sand pine scrub stands (Key and Benson, 2006). As burn severity is a function of the mu ltitudinous effects of fire on vegeta tion, it follows that methods of assessing burn severity should be tied to sp ecific vegetation assemb lages (Key and Benson, 2006). This is not to say that the CBI does poorly in assessing burn severity within sand pine scrub; rather, more specific measurements of ve getative burn severity might be possible with burn severity assessment systems developed with specific vegetation assemblages in mind. In other words, the study of burn severity might benefit from the development of regional burn severity indices, all developed and tied to specific vegetation types. Which Burn Severity Mapping Method is Best Suited for Adoption by the ONF Staff fo r Management of the JPW? The Landsat imagery, classified using a supe rvised method, is best suited for adoption by the ONF for management of the JPW. This burn severity mapping method was not only the most accurate, but it also one of th e easiest and least costly. The Landsat images used in this study were obtained at no financial cost and are available to managers and the public for free through th e MRLC program. The MRLC program does not catalogue all Landsat imagery; in fact it contains only a relatively small collection of imagery as compared to the entire Landsat library. Ho wever according to the United States Geological Survey (USGS), managers of the Landsat programs, beginning February 2009 all Landsat imagery will be available, at no-charge, to th e public as the USGS discontinues their purchasing

PAGE 29

29 and sales program. Low scene cost, wide availab ility, and high classification accuracy, indicate that future burn severity mapping programs in sand pine scrub will likely be utilizing Landsat imagery. The supervised classification was one of th e most basic image classification methods attempted in this study. Supervised classificatio ns are an elementary remote sensing technique, familiar to most remote sensing or GIS technici ans. For those not familiar with the technique, there are many resources and published works ut ilizing Landsat imagery for burn severity mapping and many software applica tions have preinstalled tools for Landsat image manipulation (Key and Benson, 2006; Cocke et al. 2005). One potential drawback is that this classifi cation technique does re quire training samples to derive the classification, while the unsupervised classification techniqu es did not require any field samples. The training samples used in this study however utilized a very basic ocular severity assessment that allowed for rapid determinat ion of plot severity (T able 1-1). The ocular severity assessment required only a few minutes to determine plot severity while the CBI based indices required a minimum of 1 hour per plot The ocular index was the most expedient severity index to record and required the least training for field personnel. In addition, the collection of training samples beyond those used in the classification, offers the ability to assess final classification accuracy. The results of this study support the remote sensing image source recommendations discussed in FIREMON: Landscape Assessment which advised the use of Landsat imagery for burn severity mapping (Key and Benson, 2006). This study also reinforces a common recommendation among experienced remote sensing practitioners, who work with imagery for natural resources a nd environmental studies: spect ral resolution is often more advantageous than

PAGE 30

30 spatial resolution. It is believed that the higher spectral resolution of the Landsat images led to their subsequent classification accuracy. Conclusions For those tas ked with restoring or maintaini ng certain ecosystems, the ability to monitor and assess the effects of prescribed fire and wi ldfire can be influential in making post fire management decisions. Knowledge of the subseque nt ecosystem effects of fire can improve the capability of managers to apply fire to achie ve desired ecosystem outcomes. Mapping burn severity is one way to monitor and assess the eff ects of fire at landscape scales. This study has demonstrated that a plot based ocular assessment of burn severity, when used as part of a supervised classification of a post-burn Landsat TM image, can produce a four-level burn severity map in sand pine scrub with accuracy results similar to burn severity maps used in other regions and ecotypes (Cocke et al 2005; White et al. 1996). This is the first known study to use remote sensing to map burn severity pattern s in sand pine scrub. Further assessment of supervised classifications of Landsat imagery for burn severity mapping in other regions of burned sand pine scrub is recommended, with partic ular considerations for accuracy in detecting variations in burn severity at di fferent spatial and temporal scales It is hoped that current and future studies in sand pine scrub burn severity mapping will provide valuable, tested tools for the improved management of the JPW and the ONF.

PAGE 31

31 Table 1-1. Descriptions of ocular severity classes used to quantify variation in burn severity in the Juniper Prairie Wilderness, Ocala National Forest, Florida. Severity Class Description Unburned No sign of fire within plot. Light Severity Mixture of burned and unburned areas within plot. Some leafy vegetation scorch. Mid Severity Burned across entire plot, partial consumption of leafy vegetation. High Extensive leafy vegetation mortality and consumption Notes: Burn severity classes were visually estimated. Burn severity was established as a sampling class in an effort to achieve a sampling of all burn severities within the JPW. Table 1-2. Descriptions of Composite Burn Inde x (CBI) burn severity indices used to quantify plot level burn severity in the Juniper Prairie Wilderness, Ocala National Forest, Florida. Index Description CBI Substrate Duff, litter and non-living surface material. CBI Herbs Grasses, forbs, small shr ubs and small trees <1m in height. CBI Tall Shrubs Shrubs and trees 1-5m in height. CBI Understory Additive value comprised of the burn severity values of the Substrate, Herbs and Tall Shrubs indexes. CBI Intermediate Trees Pole trees 8-20m in height not considered tall shrubs or big trees. CBI Big Trees Mature, domina nt and co-dominant trees. CBI Overstory Additive value comprised of the burn severity values of the Intermediate Trees and Big Trees indexes. CBI Total Additive value comprised of all CBI strata burn severity values. Notes: Descriptions from Key and Benson FIREMON: Landscape Assessment (Key and Benson, 2006) Table 1-3. Burn severity classification accura cy for the Juniper Prairie Wilderness, Ocala National Forest, Florida, including all severity in dices and classifications. Image Source Mean Map Accuracy (%) Overall Map Accuracy (%) Landsat 49.59 49.98 SPOT 47.80 49.34 Digital Aerial 35.27 40.19 Notes: Columns are averages of all classifica tion results of mean map accuracy and overall map accuracy. Accuracy was assessed by agreement of classification derived severity values with field quantified severity index values.

PAGE 32

32 Table 1-4. Overall map accuracy best ten burn seve rity classifications for the Juniper Prairie Wilderness, Ocala National Forest, Florida, including all severity indices, image sources and classifications. Classification Method Severity Type Image Source Overall Map Accuracy (%) Landsat Supervised Ocular Landsat 68.75 Landsat NBR Unsupervised Ocular Landsat 65.00 Landsat dNBR 11 Classes CBI Total Landsat 63.33 Landsat Supervised CBI Total Landsat 62.50 Landsat Supervised CBI Tall Shrubs Landsat 62.50 SPOT Supervised CBI Tall Shrubs SPOT 60.42 Landsat NDVI Unsupervised Ocular Landsat 60.00 Landsat dNBR 11 Classes CBI Understory Landsat 60.00 Landsat Supervised CBI Herbs Landsat 58.33 SPOT 421 Unsupervised Ocular SPOT 58.33 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values. Table 1-5. Mean map accuracy best ten burn seve rity classifications for the Juniper Prairie Wilderness, Ocala National Forest, Florida, including all severity indices, image sources and classifications. Classification Method Severity Type Image Source Mean Accuracy (%) Landsat Supervised Ocular Landsat 68.75 Landsat NBR Unsupervised Ocular Landsat 65.00 Landsat Supervised CBI Total Landsat 64.30 Landsat Supervised CBI Tall Shrubs Landsat 62.09 Landsat NBR Unsupervised CBI Understory Landsat 60.24 Landsat Supervised CBI Understory Landsat 60.16 Landsat NDVI Unsupervised Ocular Landsat 60.00 SPOT Supervised CBI Tall Shrubs SPOT 59.63 Landsat NBR Unsupervised CBI Total Landsat 59.56 Landsat Supervised CBI Herbs Landsat 59.16 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values.

PAGE 33

33 Table 1-6. Landsat Supervised burn severity classification and ocul ar severity index error matrix for the Juniper Prairie Wilderness, Ocala National Forest, Florida. Landsat Supervised (Classification Values) OCULAR SEV (Reference Values) Severity Unburned Light Mid High Total Producer's Accuracy Error of Omission Unburned 12 0 0 0 12 100.00% 0.00% Light 1 3 7 1 12 25.00% 75.00% Mid 0 0 9 3 12 75.00% 25.00% High 0 0 3 9 12 75.00% 25.00% Total 13 3 19 13 User's Accuracy 92.31% 100.00% 47.37% 69.23% Kappa (KHAT) = 0.58 Error of Omission 7.69% 0.00% 52.63% 30.77% Notes: Values indicate number of plot assignments to each se verity class. Accuracy was assessed by agreement of classification derived se verity values with field quantified severity index values. Table 1-7. Means of burn sever ity classification mean produce rs and users accuracies by severity class from all image sources a nd classifications; for the Juniper Prairie Wilderness, Ocala National Forest, Florida. Severity Mean Producer's Accuracy (%) Mean User's Accuracy (%) Unburned 74.20 89.92 Light Severity 22.56 37.21 Mid Severity 56.46 35.56 High Severity 37.33 24.02 Notes: Accuracy was assessed by agreement of cl assification derived severi ty values with field quantified severity index values.

PAGE 34

34 Table 1-8. Unburned class best ten burn severity classificati ons by users accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florid a, including all severity indices, image sources and classifications. Classification Severity Type Imag e Source User's Accuracy (%) Landsat Supervised CBI Overstory Landsat 100.00 Landsat Supervised CBI Intermediate Trees Landsat 100.00 Landsat NDVI Unsupervised CBI Overstory Landsat 100.00 Landsat NDVI Unsupervised CBI Intermediate Trees Landsat 100.00 Landsat NDVI Unsupervised CBI Total Landsat 100.00 Landsat NDVI Unsupervised CBI Tall Shrubs Landsat 100.00 Landsat NDVI Unsupervised CBI Understory Landsat 100.00 Landsat NDVI Unsupervised CBI Substrate Landsat 100.00 Landsat NDVI Unsupervised Ocular Landsat 100.00 Landsat NDVI Unsupervised CBI Herbs Landsat 100.00 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values. Table 1-9. Light severity class best ten burn severity classifications by users accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florid a, including all severity indices, image sources and classifications. Classification Severity Type Image Source User's Accuracy (%) Landsat NDVI Unsupervised Ocular Landsat 100.00 Landsat Supervised Ocular Landsat 100.00 Landsat NBR Unsupervised CBI Understory Landsat 100.00 Landsat NBR Unsupervised Ocular Landsat 100.00 Landsat NBR Unsupervised CBI Total Landsat 75.00 Landsat NBR Unsupervised CBI Substrate Landsat 75.00 Landsat NBR Unsupervised CBI Tall Shrubs Landsat 75.00 Landsat NBR Unsupervised CBI Herbs Landsat 75.00 Landsat 654 Unsupervised Ocular Landsat 75.00 Landsat NDVI Unsupervised CBI Total Landsat 66.67 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values.

PAGE 35

35 Table 1-10. Mid severity class be st ten burn severity classificati ons by users accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florid a, including all severity indices, image sources and classifications. Classification Severity Type Image Source User's Accuracy (%) Landsat NDVI Unsupervised CBI Herbs Landsat 60.00 SPOT 421 Unsupervised CBI Total SPOT 58.33 Landsat dNBR 11 Classes CBI Total Landsat 57.69 SPOT NDVI Unsupervised CBI Understory SPOT 57.14 SPOT NDVI Unsupervised CBI Total SPOT 57.14 SPOT NDVI Unsupervised CBI Herbs SPOT 57.14 Landsat NDVI Unsupervised CBI Total Landsat 55.00 Landsat NDVI Unsupervised CBI Understory Landsat 55.00 Landsat NBR Unsupervised Ocular Landsat 54.55 Landsat NBR Unsupervised CBI Total Landsat 54.55 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values. Table 1-11. High severity class best ten burn severity classifications by users accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florid a, including all severity indices, image sources and classifications. Classification Severity Type Image Source User's Accuracy (%) SPOT Supervised CBI Tall Shrubs SPOT 75.00 Landsat Supervised CBI Tall Shrubs Landsat 75.00 SPOT Supervised Ocular SPOT 75.00 Landsat Supervised Ocular Landsat 69.23 SPOT 432 Unsupervised Ocular SPOT 66.67 Landsat dNBR 11 Classes Ocular Landsat 57.14 SPOT 432 Unsupervised CBI Tall Shrubs SPOT 55.56 SPOT 432 Unsupervised CBI Overstory SPOT 55.56 SPOT 432 Unsupervised CBI Intermediate Trees SPOT 55.56 Landsat NBR Unsupervised Ocular Landsat 50.00 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values.

PAGE 36

36 Table 1-12. Unburned class best te n burn severity classifications by producers accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florid a, including all severity indices, image sources and classifications. Classification Severity Type Image Source Producer's Accuracy (%) Landsat Supervised Ocular Landsat 100.00 Landsat NBR Unsupervised Ocular Landsat 100.00 Landsat NDVI Unsupervised Ocular Landsat 100.00 SPOT 421 Unsupervised Ocular SPOT 100.00 SPOT Supervised Ocular SPOT 100.00 SPOT 432 Unsupervised Ocular SPOT 100.00 Landsat 654 Unsupervised Ocular Landsat 100.00 Landsat dNBR 17 Classes Ocular Landsat 100.00 SPOT NDVI Unsupervised Ocular SPOT 100.00 Landsat dNBR 17 Classes Ocular Landsat 100.00 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values. Table 1-13. Light severity class best ten burn severity classifications by pr oducers accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florida, including all severity indices, image sources and classifications. Classification Method Severity Type Image Source Producer's Accuracy (%) SPOT NDVI Unsupervised CBI Big Trees SPOT 100.00 Landsat NDVI Unsupervised CBI Big Trees Landsat 100.00 Digital Aerial CBI Big Trees Aerial 100.00 SPOT 432 Unsupervised CBI Int Trees SPOT 66.67 SPOT 432 Unsupervised CBI Overstory SPOT 57.14 SPOT 421 Unsupervised CBI Int Trees SPOT 50.00 SPOT 421 Unsupervised CBI Overstory SPOT 50.00 SPOT 432 Unsupervised CBI Substrate SPOT 45.00 Digital Aerial CBI Herbs Aerial 41.18 SPOT 432 Unsupervised CBI Herbs SPOT 40.00 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values.

PAGE 37

37 Table 1-14. Mid severity class be st ten burn severity classifica tions by producers accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florida, including all severity indices, image sources and classifications. Classification Severity Type Image Source Producer's Accuracy (%) SPOT NDVI Unsupervised CBI Big Trees SPOT 100.00 Landsat NDVI Unsupervised CBI Big Trees Landsat 100.00 SPOT 421 Unsupervised CBI Big Trees SPOT 100.00 Landsat 654 Unsupervised CBI Big Trees Landsat 100.00 Landsat NBR Unsupervised CBI Big Trees Landsat 100.00 Landsat Supervised CBI Big Trees Landsat 100.00 Landsat 654 Unsupervised CBI Overstory Landsat 87.50 Landsat 654 Unsupervised CBI Int Trees Landsat 87.50 SPOT 421 Unsupervised Ocular SPOT 80.00 Landsat dNBR 11 Classes CBI Total Landsat 75.00 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values. Table 1-15. High severity class best ten burn severity classifications by producers accuracy for the Juniper Prairie Wilderness, Ocala National Forest, Florida, including all severity indices, image sources and classifications. Classification Severity Type Image Source Producer's Accuracy (%) Landsat NBR Unsupervised CBI Total Landsat 100.00 Landsat NBR Unsupervised CBI Understory Landsat 100.00 Landsat NBR Unsupervised CBI Substrate Landsat 100.00 Landsat NBR Unsupervised Ocular Landsat 93.33 Landsat NBR Unsupervised CBI Tall Shrubs Landsat 92.31 Landsat NBR Unsupervised CBI Overstory Landsat 88.89 Landsat NBR Unsupervised CBI Int Trees Landsat 88.89 Landsat NBR Unsupervised CBI Herbs Landsat 87.50 Landsat Supervised CBI Overstory Landsat 85.71 Landsat Supervised CBI Int Trees Landsat 85.71 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values.

PAGE 38

38 Table 1-16. Burn severity clas sification accuracy listed by burn se verity index for the Juniper Prairie Wilderness, Ocala National Fore st, Florida, across all image sources. Burn Severity Index Mean Map Accu racy (%) Overall Map Accuracy (%) CBI Total 53.10 55.68 Ocular 54.89 54.89 CBI Big Trees 27.50 28.56 CBI Herbs 50.89 53.94 CBI Intermediate Trees 42.28 43.41 CBI Overstory 50.54 43.71 CBI Understory 52.30 53.07 CBI Tall Shrubs 53.45 54.20 CBI Substrate 51.87 52.35 Notes: Columns are averages of all classifica tion results of mean map accuracy and overall map accuracy. Accuracy was assessed by agreement of classification derived severity values with field quantified severity index values. Table 1-17. Means of burn sever ity classification mean produce rs and users accuracies by severity class from all image sources a nd classifications; for the Juniper Prairie Wilderness, Ocala National Forest, Florida. Severity Mean Producer's Accuracy (%) Mean User's Accuracy (%) Unburned 74.20 89.92 Light Severity 22.56 37.21 Mid Severity 56.46 35.56 High Severity 37.33 24.02 Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values. Table 1-18. Best burn severity cl assification mean producers and users accuracies by severity class and classification for the Juniper Pr airie Wilderness, Ocala National Forest, Florida. Severity Accuracy Type (%) Classification Unburned Producer's 76.84 Landsat dNBR 17 Class User's 100 Landsat NDVI Unsupervised Light Producer's 37.86 SPOT 432 Unsupervised User's 61.11 Landsat NBR Unsupervised Mid Producer's 56.25 SPOT Supervised User's 40.74 SPOT NDVI Unsupervised High Producer's 83.44 Landsat NBR Unsupervised User's 47.22 SPOT Supervised Notes: Accuracy was assessed by agreement of cl assification derived sever ity values with field quantified severity index values.

PAGE 39

39 Figure 1-1. Location of Pinus clausa stands and sample plots in the Juniper Prairie Wilderness of the Ocala National Fo rest, Florida, USA.

PAGE 40

40 Figure 1-2. Digital Aerial classi fication map of burn severity fo llowing a 2006 fire in the Juniper Prairie Wilderness of the Ocala National Fore st, Florida. Map derived from a series of 228 overlapping digital aerial images captured September 2006. Image classified into 4 severity classes using HUD.

PAGE 41

41 Figure 1-3. SPOT 432 Unsupervised classification map of burn se verity following a 2006 fire in the Juniper Prairie Wilderness of the Ocal a National Forest, Florida. Map derived from a pair of post-burn SPOT 4 scen es (MS 4619-291 06-09-25) and (PAN 4619292 06-09-25) captured September 25, 2006. Imag e classified using an unsupervised classification of bands 4-3-2 with an initial output of 20 classes which were merged using HUD.

PAGE 42

42 Figure 1-4. SPOT 421 Unsupervised classification map of burn se verity following a 2006 fire in the Juniper Prairie Wilderness of the Ocal a National Forest, Florida. Map derived from a pair of post-burn SPOT 4 scen es (MS 4619-291 06-09-25) and (PAN 4619292 06-09-25) captured September 25, 2006. Imag e classified using an unsupervised classification of bands 4-2-1 with an initial output of 20 classes which were merged using HUD.

PAGE 43

43 Figure 1-5. Landsat 654 Unsuperv ised classification map of bur n severity following a 2006 fire in the Juniper Prairie Wilderness of the O cala National Forest, Florida. Map derived from a single post-burn Landsat 5 TM scen e (path/row: 16/40) captured November 19, 2006. Image classified using an unsupervis ed classification of bands 6-5-4 with an initial output of 20 classes which were merged using HUD.

PAGE 44

44 Figure 1-6. SPOT NDVI Unsuperv ised classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the O cala National Forest, Florida. Map derived from a pair of post-burn SPOT 4 scen es (MS 4619-291 06-09-25) and (PAN 4619292 06-09-25) captured September 25, 2006. Image classified using NDVI band math with an unsupervised classification initial output of 20 classes which were merged using HUD.

PAGE 45

45 Figure 1-7. Landsat NDVI Unsupe rvised classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. Map derived from a single post-burn Landsat 5 TM scene (path/row: 16/40) captured November 19, 2006. Image classified using NDVI band math with an unsupervised classification initial outp ut of 20 classes which were merged using HUD.

PAGE 46

46 Figure 1-8. Landsat NBR Unsuperv ised classification map of bur n severity following a 2006 fire in the Juniper Prairie Wilderness of the O cala National Forest, Florida. Map derived from a single post-burn Landsat 5 TM scen e (path/row: 16/40) captured November 19, 2006. Image classified using NBR band math with an unsupervised classification initial output of 20 classes which were merged using HUD.

PAGE 47

47 Figure 1-9. Landsat dNBR 11 Cla sses classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the O cala National Forest, Florida. Map derived from a pair of Landsat 5 TM scenes (p ath/row: 16/40) captured November 19, 2006 (post burn) and September 5, 2005 (pre-bur n). Image classified using dNBR band math with an unsupervised classification initial output of 11 classes which were merged using HUD.

PAGE 48

48 Figure 1-10. Landsat dNBR 17 Cla sses classification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the O cala National Forest, Florida. Map derived from a pair of Landsat 5 TM scenes (p ath/row: 16/40) captured November 19, 2006 (post burn) and September 5, 2005 (pre-bur n). Image classified using dNBR band math with an unsupervised classification initial output of 17 classes which were merged using HUD.

PAGE 49

49 Figure 1-11. SPOT Supervised cl assification map of burn severity following a 2006 fire in the Juniper Prairie Wilderness of the Ocala Nati onal Forest, Florida. Map derived from a multi-spectral SPOT 4 scene (4619-291 06-09-25) and a panchromatic scene (4619292 06-09-25) captured September 25, 2006. Image classified using a minimum distance supervised technique.

PAGE 50

50 Figure 1-12. Landsat Supervised classification map of burn severi ty following a 2006 fire in the Juniper Prairie Wilderness of the Ocala Nati onal Forest, Florida. Map derived from a Landsat 5 TM scene (path/row: 16/4 0) captured November 19, 2006. Image classified using a minimum di stance supervised technique.

PAGE 51

51 Landsat Mean Map Accuracy Distribution SPOT Mean Map Accuracy Distribution Digital Aerial Mean Ma p Accuracy Distribution Figure 1-13. Distributions of burn severity classi fication mean map accuracies across all severity indices, by classification image source fo r the Juniper Prairie Wilderness, Ocala National Forest, Florida. Accuracy distri butions show higher classification accuracy trends in the Landsat and SPOT classifications. 0 10 20 Count 0 10 20 30 40 50 60 70 80 90 100 Mean Accuracy (%) 0 5 10 15 20Count 0 10 20 30 40 50 60 70 80 90 100 Mean Accuracy (%) 0 1 2 3 4 5Count 0 10 20 30 40 50 60 70 80 90 100 Mean Accuracy (%)

PAGE 52

52 Landsat Overall Accuracy Distribution SPOT Overall Accuracy Distribution Digital Aerial Overall Accuracy Distribution Figure 1-14. Distributions of burn severity cl assification overall map accuracies across all severity indices, by classification image source for the Juniper Prairie Wilderness, Ocala National Forest, Florida. Accuracy distributions show higher classification accuracy trends in the Landsat and SPOT classifications. 0 10 20 30Count 0 10 20 30 40 50 60 70 80 90 100 Overall Accuracy (%) 0 5 10 15 20 25Count 0 10 20 30 40 50 60 70 80 90 100 Overall Accuracy (%) 0 1 2 3Count 0 10 20 30 40 50 60 70 80 90 100 Overall Accuracy (%)

PAGE 53

53 CHAPTER 2 STAND LEVEL CHARACTERISTICS AND B URN SEVERITY VARIATIONS FOLLOWING A CENTRAL FLORIDA WILDERNESS AREA FIRE Introduction W ildfires and prescribed fires in sand pine ( Pinus clausa (Chapman ex Engelm.) Vasey ex Sarg.) scrub have historica lly resulted in extensive sand pine mortality and are often simplistically described as homogeneous disturbance events. A recent publication even proclaimed Once fires get into the sand pine stands, however, they always are crown fires (Fonda, 2000). As a threatened and narrowly ende mic ecotype native to the Southeastern US, informed management of sand pine scrub is critic al for the perpetuation of the system. Sand pine itself is an early-successional a nd relatively short-lived species, which will be replaced by xeric hardwood hammock species if fire is suppr essed (Veno, 1976). Ma ngers are therefore challenged to prescribe the appropriate fire regime for sand pine scrub: fires must burn hot enough to release serotinous cones and open growing space, but not so hot that the overstory seed sources are destroyed and sprouting species outcompete sand pine se edlings. Knowledge of the spatial heterogeneity of fire effects, along with an understanding of what drives this heterogeneity, is fundamental info rmation for the emulation of natural fire regimes to perpetuate sand pine scrub. Understanding burn severity and maintaining na tural fire regimes in sand pine scrub is important for species richness, wildlife habitat, and the future existence of this ecotype. Following disturbance, the divers ity and richness of sand pine scrub herbaceous species have been found to be highest (Gree nberg et al. 1995). Prolonged fi re suppression or high fire frequency in sand pine scrub is believed to lead to transitions to different community composition and or structure, possibly resulting in the loss of endemic or rare species (Menges et al. 1998; Myers, 1985; Veno, 1976; Laessle 1968). Veno found that following 40 years of fire

PAGE 54

54 suppression, over a study period of 20 years, scrub Shannon-Weaver diversity declined 13.8% while basal area increased 683%, indicating extr eme community structure changes (Veno, 1976). In long term studies of fire suppressed sandhill an d sand pine scrub, successi onal shifts to a xeric oak hammock have been observed as shade-tolerant oaks invade previously pine-dominated sites (Abrahamson and Abrahamson, 1996; Myers, 1985; Veno, 1976; Webber, 1933). Ocala sand pine ( P. clausa var clausa ) is an obligate seeding species wh ich releases seeds from serotinous cones when sufficiently heated by fire similar to jack pine ( Pinus banksiana Lamb.), some lodgepole pine ( P. contorta Dougl. ex Loud.), and giant sequoia ( Sequoiadendron giganteum (Lindl.) Buchh.) (Custer and T horsen, 1996; Myers, 1986). Without fire, Ocala sand pine only releases limited seed. Under natural stand condit ions sand pine mortality increases dramatically after 40 years (Ross, 1970). A study of old growth sand pine st ands found that the average age of old growth trees was only 55 years likely due to root rot fungi, a common cause of sand pine mortality (Outcalt, 1997). Fire return intervals in the ra nge of 5-15 years are likely to result in a natural community shift to l ongleaf pine or slash pine dom inated sandhill or flatwoods depending on nearby seed sources and topographic location (Myers, 1985). This species shift due to frequent fires increases the mortality of vulnerable sand pine seedlings, while producing favorable seedling habitat conditions for fire resistent species such as longleaf pine ( Pinus palustris P. Mill.) and slash pine (Pinus elliottii Engelm.), (Fonda, 2001; Veno, 1976). The fire-resisting nature and the extreme differences betw een scrub and its neighboring sandhills have been noted by naturalists and researchers since the la te 1800s (Myers, 1985; Nash, 1895). The ability to stop fires coming from adjacent natural communities is in part due to the discontinuous ground cover and compact, shallo w litter layer (Outcalt, 2003). In addition, the patchy understory vegetation lacks the flamma ble fine fuels needed to carry surface fires,

PAGE 55

55 typical of neighboring sandhills, into the scrub. Additionally, the flammability characteristics of sand pines are different from other Florida pine speci es. Sand pine has been classified as a fire evader and exists in a vegetative community classified as fire-resi lient (Rowe, 1984; Fonda, 2000). Fire evaders are species th at do not have the thick protectiv e bark to shield them from damage caused by fire, but rather depend on regene rative mechanisms to perpetuate their species following disturbances (Rowe, 1983 ). Fire-resilient communities are generally comprised of fire evader species that are typically killed by infrequent intense fires. The heat from such fires open the serotinous cones of the fire evaders to rel ease seed (Fonda, 2000). As a fire evader, sand pine initial needle combustion is very slow. In one study, the duration of the visible flaming period for sand pine needles was almost the shorte st duration of the species tested, followed only by jack pine (Fonda 2000). Eventually however, sand pine scrub does burn, following long firefree periods that result in sufficient fuel loads, which coincide with extended drought conditions. This typically occurs in the late spring months from March to May, befo re peninsular Florida summer weather patterns br ing consistent rain. Previous studies have shown that neighboring sandhill natural communities burn with spatial variability; recording wide variations in fire temperatur es across a study area (Outcalt and Greenburg, 1996). Many studies of fire ecology in sand pine scr ub have investigated species level vegetative responses to fire or fire suppression (Myers 1985; Abrahamson, 1984; Carrington, 1999; Hawkes and Meng es, 1996). Other studies have quantified small-scale spatial variations in fire intensity as recorded by temperature sensors (Abrahamson and Abrahamson, 1996; Outcalt and Greenburg, 1996). However, no an alysis of first orde r disturbance effects (burn severity), or landscape-sc ale processes in sand pine scrub has been conducted. Studies to date which document variations in fire intens ity give little towards understanding the impacts of

PAGE 56

56 fire and what underlying processes drive spatial va riation in such effects. A landscape scale assessment of burn severity is also needed to ve rify the perception that sand pine scrub fires burn with homogenous behavior, and therefore imp acts, across the landscape. Planning for the impacts of future fire use in sand pine scrub de pends on an understanding of the degree of fire effects variability and the biotic and e nvironmental drivers of such variability. Previous studies of fire in sand pine scrub would suggest that burn severity variations exist within and among sand pine scrub stands. Ab rahamson and Abrahamson (1996) conducted a 7.2 ha low intensity winter test burn of sand pine scrub in February of 1986 that resulted in spatial variations of temperature measurements acro ss 16 sampling locations. Not surprisingly, the authors described spatial variations in the severity of the fire specifically noting that 45% of mature sand pine were killed. They reported th at the fire produced a mosaic of burned and unburned patches across the 7.2 ha study area as i ndicated by temperature sensitive paints and changes in litter depth (Abrahamson and Abraha mson, 1996). Little attention was given to the variations in first order fi re effects, as the study was focused on long term community compositional changes due to fire (Abrahamson and Abrahamson, 1996). Similarly, in a May 1996 stand-replacing 12.2 ha prescrib ed fire, variations in fire temperature and intensity were quantified through analysis of aluminum tag th ermal degradation (Outcalt and Greenburg 1996). Three dimensional analyses of these relative temper atures demonstrated spatial variations in fire intensity within and among the six 10 x 50 m plot s established within sand pine stands in the Ocala National Forest (ONF) of north central Florid a. The authors suggested that such variations may have been associated with edge effects, t opographic slope and fuel loading although further correlation analysis was limited due to the small sample size, limited pre-fire data and lack of quantified observations (Outcalt and Greenburg 1996).

PAGE 57

57 This study investigates the relationships between burn severity and vegetative and topographic characteristics across a sand pine scrub landscape of the ONF. Variations in sand pine scrub burn severity at the landscape leve l have been documented and mapped in Chapter I of this Thesis (Figure 1-12). To our knowle dge there are no published works that quantify landscape and vegetative characteristics and their c ontributions and relationships to burn severity in sand pine scrub. In spite of this research gap, prescribed fire management programs are planned and conducted annually acro ss tens of thousands of hectar es in sand pine scrub regions across Florida. The research site is in one of the ONF de signated wilderness areas (Wilderness Act, 1964) containing three sand pine scrub age / type clas ses: seedling/sapling, mature pole class, and hurricane damaged pole class. Each of these stand types was burned in a 2006 prescribed fireturned-wildfire administered by O cala National Forest personnel. It is hypothesized that burn severity will di ffer significantly between the damaged pole stand type and the pole and seedling / sa pling stand types, with higher severity more closely related to anticipated higher fuel loading in damaged stands. It is hypothesized that among the pole and seedli ng / sapling stand types, variations in burn severity will be closely related to variations in stand density. To test these hypotheses, this study establishe d field plots within the burned and unburned (serving as controls) areas of the three stand types. St and topographic and vegetative characteristics along with fuel loading and quantitat ive burn severity values were collected at the plot level. Differences among pl ot characteristics and burn severi ty were assessed for the three stand types. To investigate th e contributions of fuel loadi ng and other plot topographic and vegetative characteristics to burn severity, regr ession models were developed to predict burn

PAGE 58

58 severity and identify the plot characteristics most significantly related to both ground-based and remote sensing assessments of burn severity. Study Area This study takes place at the Junipe r Prairie W ilderness (JPW), a 56 km2 United States federally designated wilderness area in the Ocala National Forest in north central Florida. The Ocala National Forest (ONF) was created in 1908 as the Florida National Forest by decree of President Theodore Roosevelt. The original 81,779 ha expanse was unclaimed, undeveloped land that was approximately 98% sand pine scrub, encompassing a region known as the Great Scrub or Big Scrub (Sekerak and Hinchee, 2001). In the early 20 th century the ONF sand pine stands were not logged for timber, as opera tions elsewhere in the st ate extracted the stronger and more valuable longleaf pine and cypress ( Taxodium L.C. Rich.). The Big Scrub was used by the residents of the area to graze cattle, pigs a nd as a hunting and camping area. Early managers of the ONF noted the peculiar fire ecology that shaped the sand pine scrub (Sekerak and Hinchee, 2001). The Big Scrub Fire of 1935 dem onstrated the stand replacement nature of the scrub fire regime on a vast sc ale: burning a record 14,000 ha in 4 hours across a swath of 58 kilometers (Sekerak and Hinchee, 2001; Myer s and Ewel, 1990). The JPW represents a portion of one of the largest and best pr otected tracts of sand pine scrub in the state (Greenburg, 1996). Federally designated as a wilder ness area in 1984, the JPW falls under the resource management protection of the United States 1964 Wilderness Act. The Act states that a designated Wilderness Area is to be protected and manage d so as to preserve its natural conditions (Wilderness Act of 1964). In an effort to furthe r preserve the character of the area, the Act dictates that there shall be no temporary road, no use of motor vehicles, motorized equipment or motorboats, no landing of aircraft, no other form of mechanical transport, and no structure or installation within the desi gnated Wilderness Area (Wilderness Act of 1964). These access

PAGE 59

59 restrictions apply both to users of the area and to the land manageme nt authority, in this case, the ONF. An endemic natural community, sand pine sc rub is found only in Florida and in a small coastal region in southern Alabama (Myers, 1985 ). Occupying a geographic region comprised of Miocene-Pleistocene marine and aeolian depositio ns, sand pine scrub exists on dry, well-drained sandy soils with low nutrient leve ls (Carrington, 1999; Myers, 1985) Predominantly, this region is an ancient coastal dune running along the central Brooksville Ridge of peninsular Florida stretching from St. Johns County to Dade County (Outcalt, 2003). Within the JPW, sand pine scrub is the dominant natural community type. Sand pine scrub is typically dominated by an even-aged monoculture overstory of sand pine. Sand pine scrub regions of inland peninsular Florida tend to be comprised of stands of fire-dependent, bradysporous or serotinous Ocala sand pine ( P. clausa var. clausa D.B. Ward) while regions of coastal sand pine scrub tend to be comprised of stands of fire inde pendent Choctahatchee sand pine ( P. clausa var. immuginata D.B. Ward). The understory and midstory ar e generally dominated by scrubby sclerophyllic oaks, ( Quercus. chapmanii Sarg ., Q. myrtifolia Willd, Q. geminate Small), saw palmetto ( Serenoa repens (Bart.) Small), scrub palmetto ( Sabal etonia Swingle ex Nash), tree lyonia ( Lyonia ferruginea (Walt.) Nutt.), red bay ( Persea humilis Nash) and rosemary ( Ceratiola ericoides Michx.) (Outcalt and Greenb erg, 1998; Greenberg, 1996; Menges et al. 1993; Myers, 1985; Veno, 1976). Relying on natu ral ignitions, sand pine scr ub typically burns every 30-60 years, much less frequently than the neighboring slash pine flatwoods or longleaf pine sandhills (Fonda, 2000; Menges et al. 1998; Rich ardson et al. 1988; Myers, 1985;). In late July of 2006, the ONF staff initiated pr escribed fire operations on the JPW with the intent of burning a number of prai ries. Within several days the prescribed fires escaped the

PAGE 60

60 initial prescriptions and were reclassified as a wildfire. The resulting wildfire continued into August of 2006 and ultimately burned an unprecedented 44 km2 across the JPW. The fire burned a portion of nearly all of the vegetative commun ities represented in the JPW and across multiple sand pine stands. The burning of such large tr acts of sand pine scr ub, under wilderness area management restrictions, presented a unique opport unity to study burn severity within a variety of sand pine scrub stand types. Methods Plot Description To quantify burn severity at the stand level and record plot to pographic and vegetative characteristics, sem i-permanent sample plots were established in the winter and spring (January May 2007) following the 2006 JPW burn (Figures 2-1:2). Plots were based on a stratified sampling design as part of a mu lti-faceted study of burn severity and ecosystem response within the JPW. Plots were stratified by sand pi ne stand type-class as documented by the ONF: seedling/sapling (SS), pole (PP) and damaged pole (DP) (Table 2-1). These stands were identified using GIS stand histor y layers provided by the ONF. This stratification was done to provide comparisons of burn severity and vege tative / topographic charac teristics between the three major stand types found within the JPW. The seedling /sapling class (SS) is a USFS stand type description for sand pine st ands as delineated by individual tree diameter at breast height (DBH). SS stands are defined as stands where greater than 50% of the stocked trees have a DBH < 12.7 cm. The pole class stands (PP) are simila rly defined as containi ng 50% or more stocked trees with DBH > 12.7 cm. The damaged pole sta nds (DP) comprise pol e class trees (DBH > 12.7 cm) that were significantly structurally a ltered during the year 2004 hurricane season (June November); during which four separate hurrica nes impacted the central Florida region. The

PAGE 61

61 DP stands were delineated and mapped by th e ONF staff following the 2004 season. These delineated stands were visually significantly distinct from undamaged stands. Plot Level Burn Severity Plot data were collected utilizing an adaptation of the Key and Benson FIREMON: Landscape Assessm ent sampling approach (Key and Benson, 2006). This method uses the Composite Burn Index (CBI) to sample and quan tify burn severity at the plot level (Key and Benson, 2006). CBI was developed as a consistent adaptable, rapid assessment system for collecting field data for the analysis of remote ly sensed burn severity mapping products using the Normalized Burn Ratio (NBR) scale (Key and Benson, 2006). Within each type-class (SS, PP, DP) plots were established in areas visually iden tified at four levels of burn severity: unburned, light severity, mid severity and hi gh severity. These ocular burn seve rity classes (Tab le 2-2) were delineated visually in the field. The ocular severity classes were established to ensure that the very distinct visual levels of burn severi ty were represented w ithin the study design. Within each burn severity type, for each stand type-class, five systematically-located plots were established, giving a total of 60 field plots for the entire study area. Plots were established a minimum of 100 m from any edge s (including edges between severi ty levels and type-classes) and with a minimum of 100 m separation between plots. At each 10 x 10 m plot, burn severity was assesse d visually in terms of vegetative impacts due to fire across seven vertical layers of vegetative strata (substrate, herbaceous, tall shrubs, understory, intermediate trees, big trees, oversto ry) according the CBI methods (Table 2-3)(Key and Benson, 2006). Remotely Sensed Burn Severity As no specific index has been utilized to qua ntify burn severity variations in sand pine scrub before, it follows that this study should inve stigate drivers using m ore than one index. In

PAGE 62

62 addition to the CBI method, a burn severity index derived from a supervised classification of a post-burn Landsat TM scene was used (Figure 2-3 and Table 2-3). In a related study the author conducted of multiple burn severity classi fication mapping methods following the JPW 2006 fire, the Landsat post-burn supervised clas sification method demons trated the highest classification accuracy. The post-burn, Landsat 5 TM scene (LT50160402006323EDC00) (path 16, row 40) captured on November 19, 2006 using the Thematic Mapper (TM) instrument was obtained from the Multi-Resolution Land Charac teristics Consortium (MRLC). The scene was recorded at 30 m spatial resolu tion and contained six bands spect ral information ranging from 0.45-2.35 m wavelength. Landsat band 6, thermal (10.40-12.5 m), was omitted in MRLC preprocessing and band 7 (2.08-2.35 m) referred to as band 6. The scene was preprocessed (level 10) by the MRLC, which included ra diometric and geometric corrections and was packaged with MRLC derived REFL (at satellite reflectance) a nd NBR (normalized burn ratio) scenes. These images were nearest-neighbor ge ometrically corrected us ing digital orthographic quarter quads (DOQQs). The supervised clas sification of the postburn image was derived through a minimum-distance method using RSI E NVI 4.3(ITT Industries Inc., Boulder, CO, USA). To train the classification, the ocular severity values recorded during the field sampling portion of this study were used as ground control points (GCP). One GCP was chosen at random to represent each stand class and severity level (total of 12 selected plots). The supervised classification resulted in a four severity cla ss image (unburned, light severity, mid severity and high severity). The burn severity values that were assigned to each plot in the classification were then used as dependent variable burn severity va lues (Supervised Classifi cation) in the regression modeling.

PAGE 63

63 Plot Topographic and Vegetative Ch aracteristics and Fuel Loading In addition to burn severity data collected at the post-fire sample plots m entioned previously, additional plot ch aracteristics identified as poten tial contributing independent variables were recorded or measured (Table 2-4) Field recorded or measured variables were collected at each plot following the plot sa mpling guidelines established for the FIREMON database system (Key and Benson, 2006). Additi onal variables were re corded from spatial datasets provided by the USFS ONF staff. The study initially planned to investigate time since previous burning (wildfire or prescribed fire) as additional inde pendent variables, as these data were also contained within GIS layers provided by the ONF. However, after esta blishing the plots, it wa s observed that no plots were contained within the boundaries of previously documented wildfires. Weather at time of burning was also considered as a potential independent variable; however data documenting the extent of the fire were insufficient for ma tching with recorded weather observations. Analysis Data analy sis was conducted using JMP 7.0.1 (S AS Institute Inc., Cary, NC, USA). Oneway analysis of variance (ANOVA) and frequency histograms of burn severity metrics (CBI Total, CBI Overstory, CBI Tall Sh rubs, CBI Herbs, CBI Substrate, CBI Intermediate Trees, CBI Big Trees, CBI Understory, Supervised Classifica tion) were used to identify burn severity differences among stand classes and among ocular se verity classes. Sign ificance of severity differences were determined using one-way ANOVA (p=0.05). Nested model ANOVA tests were run using JMP 7.0.1 to detect effects of burn severity level and stand type on field re corded plot characteristics. Where significant differences (p<0.05) were indicated in ANOVA, Tukeys HS D test was used to identify and explore differences.

PAGE 64

64 Differences in spatial area burned by the severi ty classes, as determined by the classified Landsat image (Figure 2-3), were explored using ESRI ArcGIS 9.2. The area burned per severity class was determined for the entire JPW and also within the three stand types. A step-wise multiple regression procedure was used to develop burn severity predictor models for each stand type and each burn severity me tric. Burn severity metric values for each stand type were used as response variables, whil e plot characteristics we re used as independent predictor variables in the model regressions. Initial data explorations reduced the numb er of independent variables for regression modeling using a principal components analysis (P CA) that identified key independent variables based on component eigen value loading. A hybrid Kaiser-Guttman / scree plot approach was used to investigate the first three components for va riable loadings. In dependent variables identified through this method were used in a step-wise regression procedure. The resulting models were not as effective based on overall R2 and RMSE values as the models derived from the qualitative reduction of independent vari ables. The PCA based approach was thus abandoned in favor of the qualitative variable reduction method. The results of the PCA approach are not presented here. For the multiple regression procedure, a qualitative reduction of independent variables was conducted, identifying ten key potential predictor variables: aspect, elevation, percent slope, percent tree cover, percen t sapling cover, stand height, canopy fuel height, sapling per hectare, trees per hectare and total pl ot fuel load (Table 2-4). These independent variables were checked fo r significant correlati ons using a pair-wise correlations test (p<0.05) in JMP. Pairs of terms that were significantly correlated were skipped

PAGE 65

65 in the step-wise regression procedure to avoid potential model misinterpretations due to multicollinearity. Using the ten independent pred ictor variables, least-square-means step-wise regression models were run in JMP. The burn severity metric s were used as response variables: CBI Total, CBI Substrate, CBI Herbs, CBI Tall Shrubs, CBI Understory, CBI Overstory and Supervised. The CBI Big Trees and CBI Intermediate Trees bur n severity metrics were not included in the regression modeling due to insuffici ent coverage in their datasets across differing stand types. The step-wise procedure utilized a forward a dditive method, wherein model components were added based on their cumulative co ntributions to overall model R2 and RSME values. In most cases model components were only added if thei r indicated contributions were significant (Ftest). In all cases of the procedure, the resulting models utili zed either one or two predictive terms, but no more. Following the model building, stepwise selected predictor variables were rechecked for pair-wise correlations. Where model components were significantly correlated, the stepwise procedure was conducted again to select components that were not significantly correlated. Results Variations in Burn Severity Am ong Sand Pine Scrub Stand Types Mean burn severity did differ significantly am ong the severity cla sses and stand types (Table 2-5:6). However where significant differe nces were indicated am ong stand types (CBI Intermediate Trees, CBI Overstory) the differences were likely an artifa ct of the CBI metric used, as opposed to the actual variations in ve getative severities. The CBI Intermediate Trees and CBI Overstory severity metrics were targeted at big tree s and overstory components not represented in the SS stand type. For this r eason, the CBI Intermediate Trees metric was not utilized in any of the stand t ype regression modeling and the CBI Overstory was not used in the

PAGE 66

66 SS stand type regression modeling. Similarly, wh ile no significant differe nces were indicated, the CBI Big Trees metric was also not used in th e regression modeling due to a lack of big trees within any of the stands sampled. When di sregarding the results of those metrics with incomplete data, there were no significant diffe rences among the stand types from the remaining CBI and ocular severity metrics. When aggrega ting the individual CBI strata scores for the CBI Total, the resulting mean CBI burn severities we re also not significantly different. Mean burn severity was also not found to be significantly different among the stand types derived from the Landsat supervised classification image. Burn severity did differ however, among the st and types in terms of area burned by each severity class (Table 2-7 and Fi gure 2-4). The DP stand, which wa s expected to have the highest burn severity, actually had the leas t percent area burned as high severity. Across the entire JPW, the amount of area burned by each burn severity class also varied (Table 2-8). It is interesting that, for a fire regime often referred to as high severity, only 17.4% of the JPW burned as the high severity class according to the Landsat classification. Within the stand types, among the ocular se verity classes (unburne d, light severity, mid severity and high severity) signifi cant differences were found for a ll stand types (Tables 2-5:6). Tukeys HSD tests demonstrated that not all of the ocular severity cla sses were significantly different from within each stand type. In many cases there were no significant differences between the mid severity class and the high severity cl ass or the low severity class and the mid severity class. In all of the stand types and se verity indexes there were significant differences of the burn severity means between the unbur ned and the high severity classes. Variations in Topographic, Vegetative and Fuels Characteristics Among Stand Types The ANOVA test results from th e ten individual stand characteristics selected for use in the regression modeling demonstrated some si gnificant differences among both stand types and

PAGE 67

67 ocular severity classes within th e stand types (Table 2-4). Of th e ten stand characteristics: seven were found to be significantly different among the stand types (total tree cover, sapling cover, stand height, canopy fuel height, sapl ings per ha., total trees per ha ., and total fuel loading) five were found to be significantly different among ocular severity classes (ele vation, total tree cover, sapling cover, canopy fuel height, and saplings per ha.) and only two did not demonstrate any significant differences at any leve l (aspect and slope) (Table 2-4). Variations in elevation within the ocular se verity classes were surprising. The DP stand type recorded the lowest elevation stand class (light severity, 13.6 m) while the PP stand type had the highest elevation stand class (unburned, 30.8 m). Explorat ory analysis revealed for the PP stand type, plot elevation was significantl y inversely correlated with severity class (R2: 0.7017, p=0.0006). Significant correla tions were not observed for the SS or PP stand types. Total tree cover and sapling cover were both significantly different among stand types and within stand types among the ocular severity cl asses. Among stand types, the SS stands had significantly greater total tree cover than the DP and PP stand types. Significant differences were found among the stand types in terms of the total fuel load per plot. Surprisingly, the DP stand type did not have the highest to tal fuel loading as expected, rather the SS stand type recorded highest total fu el loading. The SS stand type also recorded the highest 1000 hr. fuel loadings ( both sound and rotten) (Table 2-9). Modeling Burn Severity Using Stand Characteristics The step-wise leas t squares regression analysis developed multiple burn severity models for each stand type (Table 2-10 and Figure 2-5). Models were created for the following severity indices (by stand type): CBI Substrate, CBI Herbs, CBI Tall Shrubs, CBI Understory, CBI Overstory, CBI Total, and Supervised. The CBI Ov erstory severity index was not used to model

PAGE 68

68 the SS stand type burn severity because the CBI Overstory index did not record an overstory component within the SS stand type plots. Model R2 values for all stand types and severity indices ranged from: R2 = 0.55 (CBI Tall Shrubs (DP)) to R2 = 0.77 (CBI Total (PP)). The PP sta nd type CBI Total burn severity model had the highest overall R2 value (R2 = 0.77) (across all burn severi ty indices and stand types). The CBI Total (PP) burn severity model also ha d the lowest overall error (RMSE = 0.52) (across all burn severity indices and st and types). In terms of the mean model error (increasing), the same order followed: SS (mean RMSE = 0.61), PP (mean RMSE = 0.62) and DP (mean RMSE = 0.71). The SS stand type models, across all se verity indices, had the highest mean R2 values (mean R2 = 0.71) followed by the PP stand type models (mean R2 = 0.68) and lastly the DP stand type models (mean R2 = 0.60). These models provide evidence of the trends that exist between burn severity and plot characteristics at the stand level. Out of th e ten plot characteristic s chosen for regression modeling, only six ultimately were chosen for their significant contributions to the models (tree cover, sapling cover, elevation, aspect, stand height, and canopy fuel height). Of those six selected characteristics, at leas t one of the three variables: to tal tree cover, sapling cover and elevation, were found in 19 of the 20 models. St and level characteristics not present in the models are those that did not c ontribute significantly. Interestingly, total fuel load was not a significant addition to any of the models even in the DP stand type. In all of the models where th ey were selected as paramete rs, both sapling cover and total tree cover had negative model coe fficients. This indicates that those plot characteristics have negative correlations with burn severity. The elev ation parameter was not as consistent. For the Seedling / Sapling stand type, elevation had a negative coefficient while the PP and DP stand

PAGE 69

69 type models had a positive elevation coefficient. The models developed for the SS and PP stand types were consistent within each stand type with regard to selected model terms and R2 / RMSE values. The DP stand type however was less consis tent. DP models had the greatest mean error, the lowest mean R2 and the least consistency of parame ters. Despite the variability and perception of greater uncertainl y in the DP models, the sapli ng cover parameter was still a component of six of the seven DP models, all with negative coeffici ents. These data suggest that sapling cover and total tree cove r may play influential roles in determining the post fire burn severity. Discussion This study developed 20 burn severity m odels based on seven different burn severity indices and three stand types. Based on the R2 values and model error, it is evident that linear relationships exist between plot level variables and burn severity. The results of this study have revealed some interesting aspects about the initial hypothesis. It was predicted th at significant differences of mean burn severity would exist among the sand pine stand types sampled within th e JPW, and that the damaged pole class would have higher severity on average. This was not found to be the case based on the field-sampled severity data. Burn severity data collected at plots across four oc ular severity classes from three stand types demonstrated no signi ficant differences between the mean severities between stand types. However, differences were found among the three stand types in term s of the spatial area burned by each severity class as indicated by the Landsat classification. In addition, differences were found in the amount of area burned by each severity class, acro ss the entire JPW. Significant differences were also found among the four ocular severity classes within the stands. This indicates that severity vari ations do in fact exist within bu rned sand pine scrub, regardless of the pre-burn stand conditions or recent distur bance history. These results suggest that the

PAGE 70

70 notions of fire burning within sand pine scrub as a homogenous high severity disturbance event may need to be seriously questioned. The results of this study do not support the second part of th e hypothesis, that burn severity within the DP stand type would be higher as a result of increased fuel loading. This is because burn severity was less correlated with overall fuel loading and more corre lated with percentages of sapling cover. Likewise, the third hypothesis, that burn severity within the SS and PP stand types was influenced by stand density, was not supported. The results of the regression modeling indicated that burn severity was co rrelated primarily to variations in tree cover, sapling cover and elevation with some models indicating correlati ons with canopy fuel height, aspect and stand height. Interestingly, no model i ndicated that total fuel load or trees per ha added significant contributions to predicting burn severity. It is possible that fuels were still significant contributors to burn severity in the JPW, however by means other than total loading. Models indicated that higher sapling cover (p rimarily in PP stand types) and total tree cover (primarily in SS stand type s) were inversely related to bur n severity. The trend of the models to select for total tree cover for the SS st and type and sapling cover for the PP stand type suggests that there may be someth ing about, or influenced by, sap ling cover that is driving burn stand level severity patterns. For the SS sta nd type, sapling cover wa s significantly linearly correlated with total tree cover (R2 = 0.89 RSME = 0.095) (Figure 2-6) suggesting that the terms were potentially interchangeable in the SS stand type models. Similar studies of burn severity in different ecotypes have found burn severity variations influenced by tree cover. The study of the y ear 2000 Jasper fire, wh ich burned 34,000 ha of predominately mature ponderosa pine (Pinus jeffreyi Grev. & Balf.) forests in South Dakota, USA, found increases in burn severity linked to higher total tree cover (L entile et al. 2006).

PAGE 71

71 Significant differences exist betw een ponderosa pine forests and sand pine scrub, with ponderosa pine much more similar in struct ure and fire type to long leaf pi ne forests. Changes in sapling cover may be indicative of changes in the stru cture of pre-burn fuels. Landscape level fuels structure and arrangement were not documented in this study and have been shown to influence fire severity in other stand replacement burn ecosystems. A study of the fire history of Yellowstone National Park, Wyoming, USA, dete rmined that young stands of lodgepole pine ( Pinus contorta Dougl. ex Loud.) were not prone to high seve rity crown fire due to a lack of fine fuels on the ground, while mature stands of lodgepol e pine were prone to such fires (facilitated by increased ground fuel loads and ladder fuels) (Romme, 1982). It is possible that a similar cycle of fuel loading and stand structure is drivin g the burn severity patterns in sand pine scrub. As a stand replacement system, it follows that sand pine would be evolutionarily predisposed not to burn at least until the stands are sexually mature (5-10 years of age). The other primary stand level variable select ed for significant contribution to the burn severity models was elevation. This was a surp rise, as casual observation s in the field did not suggest elevation as a significant driver of bur n severity. The results indicate otherwise, however, based on the parameter coefficients and the order of addition to the model, elevation was not as significant as sapling cover or total tree cover. In the PP stands, decreases in elevation were indicative of higher burn severity. This contrasts with a 16 year landscape study of Alaskan boreal forest burn severity patterns, wherein elevation was positively correlated with burn severity (Epting and Verbyla, 2005). Great differences exist howev er between the Alaskan study landscapes topography and the JPW topogra phy. The Alaskan study area ranged from 3001500 m above sea level (a.s.l.), while the JPW ranged from 6.5-45 m a.s.l. Elevation changes in the Alaskan study were associated wi th variations in vegetation assemblages and

PAGE 72

72 moisture availability. The limited range of elevation within the JPW does not necessarily indicate that variations of sp ecies composition and microclimate due to elevation are less likely to be found, however further study of the role of elevation in sand pine scrub burn severity would be needed to draw broader conclusions. It should be noted that the relationships between the plot variables used in the models and burn severity do not necessarily indicate or infer burn severity causation. Determining the cause of plot level burn severity is beyond the scope of this study. This st udy suggests the need to study sand pine scrub and other stand replacement fire types with. Quan tifying stand structure and fuel arrangement as well as topographic variation (even in locations with relatively low topographic variations compared to western states) may be ke y in furthering understanding of complex stand replacement systems. Conclusions This study has dem onstrated that within stand variations of first or der fire effects (burn severity) exist following high intensity fire in sand pi ne scrub. Variations of fire severity within sand pine scrub stand types are influenced by stand characterist ics and elevation. While this study provides insight into processes influencing burn severity, it also reiterates the need for further study, as the exact mechanisms associated with sapling cover and elevation that drive stand level severity are not yet understood. This study has demonstrated that burn severi ty in sand pine scrub can be regression modeled (0.55 R2 0.77) based on plot level vegetative an d topographic characteristics. These models are the first known attempt to develop predictive methods for understanding burn severity in sand pine scrub. The capabilities of these models should not be over extended, as they were developed following a single fire event that burned over uncontrolled weather and fuels conditions.

PAGE 73

73 Given the threatened nature of sand pine sc rub and the imperiled st ate of some of the species endemic to it, any contribution that research can provide towards its improved management is significant. The models devel oped through this study hopefully contribute to the groundwork for continuing sand pine scrub burn seve rity research. Research which will provide for those who are tasked with managing scr ub, tested and validate d, quantitative tools for improved land management.

PAGE 74

74 Table 2-1. Pinus clausa stand types sampled in the Junipe r Prairie Wilderness, Ocala National Forest, Florida. Stand Type Description Notes Seedling / Sapling (SS) dbh < 12.7 cm Young, dense, largely homogenous stands Pole (PP) dbh > 12.7 cm Pole class stands are the largest / oldest trees within the burned area Damaged Pole (DP) dbh > 12.7 cm Pole class stands damaged during the 2004 hurricane season. These stands were mapped and identified by the USFS staff. Notes: Mature sand pine stands were not re presented within the burned area and were not sampled, although some stands exist within the JP W. Sparse pole stands were found within the burned area of the JPW, however not enough stands were burned to support the sampling method. Stands were delineated and mappe d by the ONF staff under USFS standards. Table 2-2. Descriptions of ocular severity classes used to quantify variation in burn severity in the Juniper Prairie Wilderness, Ocala National Forest, Florida. Severity Class Description Unburned No sign of fire within plot. Light Severity Mixture of burned and unburned areas within plot. Some leafy vegetation scorch. Mid Severity Burn evidence across entire plot, partial consumption of leafy vegetation. High Severity Extensive leafy vegetation mortality and consumption Notes: Burn severity classes were visually estimated. Burn severity was established as a sampling class in an effort to achieve a sampling of all burn severities within the JPW.

PAGE 75

75 Table 2-3. Descriptions of burn severity indices us ed to quantify plot level burn severity in the Juniper Prairie Wilderness, O cala National Forest, Florida. Index Description CBI Substrate Duff, litter and non-living surface material. CBI Herbs Grasses, forbs, small shr ubs and small trees <1m in height. CBI Tall Shrubs Shrubs and trees 1-5m in height. CBI Understory Additive value comprised of the burn severity values of the Substrate, Herbs and Tall Shrubs indexes. CBI Intermediate Trees Pole trees 8-20m in he ight not considered tall shrubs or big trees. CBI Big Trees Mature, domina nt and co-dominant trees. CBI Overstory Additive value comprised of the burn severity values of the Intermediate Trees and Big Trees indexes. CBI Total Additive value comprised of all CBI strata burn severity values. Supervised Classification Landsat post-burn image classification derived burn severity values. Notes: Descriptions from Key and Benson FIREMON: Landscape Assessment (Key and Benson, 2006)

PAGE 76

76 Table 2-4. Descriptions and di fferences of sample plot variables among severity classes and stand types in the Juniper Prairie Wilderne ss of the Ocala Nationa l Forest, Florida. Variable Range of Va lues Description ANOVA Significant Differences (p<0.05) Regression Variables Aspect 0 345 Recorded at plot (degree) NA Elevation 6.5 43 m Recorded using Garmin handheld GPS at plot (m) ** Slope 0 12% Measured using handheld clinometers (degree) NA Total Tree Cover 10 90% Ocular Estimation (%) *, ** Sapling Cover 0 90% Ocular Estimation (%) *, ** Stand Height 1.8 25 m Measured using Haglof vertex hypsometer (m) Canopy Fuel Height 0 15 m Measured using Haglof vertex hypsometer (m) *, ** Saplings per ha 10 4040 Recorded following FIREMON protocol (trees per hectare) *, ** Total Trees per ha 450 1927 Recorded following FIREMON protocol (trees per hectare) Total Fuel Load 0 42 kg/m2 Total fuel load (1-1000 hr) recorded following FIREMON protocol (kg/m2) Variables Not Selected for Regression Plot Seedling Cover Ocular Estimation (%) NA Medium Tree Cover Ocular Estimation (%) NA Pole Tree Cover Ocular Estimation (%) Total Shrub Cover Ocular Estimation (%) NA Low Shrub Cover Ocular Estimation (%) NA Medium Shrub Cover Ocular Estimation (%) NA Tall Shrub Cover Ocular Estimation (%) NA Graminoid Cover Ocular Estimation (%) *,** Forb Cover Ocular Estimation (%) ** Moss / Lichen Cover Ocular Estimation (%) ** Bare Soil Cover Ocular Estimation (%) ** Rock Cover Ocular Estimation (%) NA Woody Ground Cover Ocular Estimation (%) *,** Char Cover Ocular Estimation (%) *,** Basal Vegetation Cover Ocular Estimation (%) *,** Litter / Duff Cover Ocular Estimation (%) *,** Canopy Cover Ocular Estimation (%) ** Seedlings per ha Recorded following FIRE MON protocol (trees per hectare) Snags per ha Recorded following FIREMON protocol (trees per hectare) *,** Duff / Litter Depth Measured following FIREMON protocol and planar intercept method (cm) ** 1 hr Fuels Fuel load (1 hr) record ed following FIREMON protocol (kg/m2) *,** 10 hr Fuels Fuel load (10 hr) recorded following FIREMON protocol (kg/m2) *,** 100 hr Fuels Fuel load (100 hr) recorded following FIREMON protocol (kg/m2) *,** 1-100 hr Fuels Total fuel load (1-100 hr) recorded following FIREMON protocol (kg/m2) *,** 1000 hr (Sound) Fuels Fuel load (Sound 1000 hr) recorded following FIREMON protocol (kg/m2) NA 1000 hr (Rotten) Fuels Fuel load (Rotten 1000 hr ) recorded following FIREMON protocol (kg/m2) 1000 hr (Total) Fuels Fuel load (Total 1000 hr) recorded following FIREMON protocol (kg/m2) NA Notes: NA = No significant differences. = Di fferences found among stand types. ** = Differences found within stand types, among se verity classes. Listed by variables selected for burn severity regression modeling and thos e variables not select ed for burn severity regression modeling.

PAGE 77

77 Table 2-5. Differences of burn severity indices am ong ocular severity classes and stand types in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. Severity Index ANOVA Significant Differences CBI Substrate ** CBI Herbs ** CBI Tall Shrubs ** CBI Understory ** CBI Intermediate Trees *,** CBI Big Trees NA CBI Overstory *,** CBI Total ** Supervised Classification ** Notes: NA = No significant differences. = Diffe rences found among stand types. ** = Differences found within stand types, among severity classes

PAGE 78

78 Table 2-6. Burn severity means and standa rd deviations recorded in the Juniper Pr airie Wilderness, Ocala National Forest, Flor ida. Severity Index Damaged Pole Pole Seedling / Sapling High Severity Mid Severity Low Severity Unburned High Severity Mid Severity Low Severity Unburned High Severity Mid Severity Low Severity Unburned CBI Substrate 1.25 ( 0.97) 1.10 ( 0.91) 1.25 ( 1.07) 2.4 ( 0.55) 1.6 ( 0.55) 1.0 ( 0) 0 2.2 ( 0.45) 1.4 ( 0.55) 0.80 ( 0.45) 0 2.4 ( 0.89) 1.8 ( 0.45) 0.8 ( 0.45) 0 CBI Herbs 1.45 ( 1.1) 1.20 ( 1.01) 1.30 ( 1.03) 2.8 ( 0.45) 2.0 ( 0) 1.0 ( 0) 0 2.4 ( 0.55) 1.6 ( 0.55) 0.80 ( 0.45) 0 2.4 ( 0.55) 2.0 ( 0) 0.8 ( 0.45) 0 CBI Tall Shrubs 1.55 ( 1.15) 1.35 ( 1.14) 1.35 ( 1.09) 3.0 ( 0) 2.0 ( 0) 1.2 ( 0.45) 0 3.0 ( 0) 1.4 ( 0.55) 1.0 ( 0) 0 2.2 ( 0.84) 2.2 ( 0.45) 1.0 ( 0.71) 0 CBI Understory 1.35 ( 0.99) 1.15 ( 0.87) 1.30 ( 1.08) 2.4 ( 0.55) 2.0 ( 0) 1.0 ( 0) 0 2.2 ( 0.45) 1.4 ( 0.55) 1.0 ( 0) 0 2.4 ( 0.89) 2.0 ( 0) 0.8 ( 0.45) 0 CBI Intermediate Trees 1.36 ( 1.16) 0.81 ( 1.13) 0.17 ( 0.42) 1.41 ( 1.07) 1.58 ( 1.44) 1.08 ( 1.48) 1.36 ( 0.88) 2.16 ( 1.21) 1.11 ( 0.79) 0 0 NA NA NA 0.68 ( 0.64) CBI Big Trees 0.14 ( 0.54) 0.005 ( 0.02) NA 0.19 ( 0.37) 0 0 0.46 ( 1.03) NA 0.02 ( 0.04) NA NA NA NA NA NA CBI Overstory 1.10 ( 1.21) 1.35 ( 1.27) NA 2.4 ( 1.34) 0.8 ( 1.1) 1.2 ( 0.45) 0 3.0 ( 0) 2.0 ( 0) 0.4 ( 0.55) 0 NA NA NA NA CBI Total 1.40 ( 1.05) 1.25 (1.02) 1.30 ( 1.08) 2.6 ( 0.55) 2.0 ( 0) 1.0 ( 0) 0 2.4 ( 0.55) 1.8 ( 0.45) 0.8 (0.45) 0 2.4 ( 0.89) 2.0 (0) 0.8 ( 0.45) 0 Supervised Classification 1.80 ( 1.15) 1.60 ( 1.23) 1.60 ( 1.14) 2.8 ( 0.45) 2.2 ( 0.45) 2.2 (0.45) 0 3 ( 0) 2.2 ( 0.45) 1.2 ( 0.84) 0 2.6 ( 0.55) 2.2 ( 0.45) 1.6 ( 0.89) 0 Notes: NA = No values recorded Listed by stand type and ocular severity class for each severity index.

PAGE 79

79 Table 2-7. Spatial area and percent of the area for each burn severity class listed by sand pine scrub stand type within the Juniper Prairi e Wilderness of the Ocala National Forest, Florida. Seedling / Sapling Pole Damaged Pole Severity Class Ha. Percent of Stand Type Ha. Percent of Stand Type Ha. Percent of Stand Type Unburned 202.75 13.81% 109.28 24.17% 324.34 21.39% Light 346.59 23.61%22.16 4.90% 349.14 23.03% Mid 512.08 34.88% 138.36 30.61% 541.57 35.72% High 406.63 27.70% 182.29 40.32% 300.96 19.85% Total 1468.04 100.00%452.09100.00% 1516.01 100.00% Notes: Area burned determined by a supervised classification of a pos t burn Landsat image. Table 2-8. Spatial area and per cent of the total area for each burn severity class in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. Severity Class Ha.Percent of Total Unburned 2088.79 35.28% Light 1427.76 24.11% Mid 1376.35 23.25% High 1027.77 17.36% Total 5920.67 100.00% Total Burned Area 3831.87 64.72% Total Area of the JPW 5920.67 Notes: Area burned determined by a supervised classification of a pos t burn Landsat image.

PAGE 80

80 Table 2-9. Sample plot variable means and standard deviations for plot level variables selected for bur n severity regression m odeling. Damaged Pole Pole Seedling / Sapling Variable High Severity Mid Severity Light Severity Unburned High Severity Mid Severity Light Severity Unburned High Severity Mid Severity Light Severity Unburned Aspect () 147 ( 104) 139 ( 117) 156 ( 88) 76 ( 61) 191 ( 58) 164 ( 147) 155 ( 115) 137 ( 105) 184 ( 135) 49 ( 75) 187 ( 119) 170 ( 85) 143 ( 56.30) 138 ( 118) 174 ( 108) Elevation (m) 18.09 ( 6.88) 21.09 ( 7.99) 18.63 (4 .03) 19.96 ( 9.32) 16.62 ( 4.96) 13.66 ( 5.05) 22.12 ( 5.97) 14.72 ( 6.12) 19.10 ( 4.56) 19.76 ( 3.00) 30.80 ( 7.89) 19.04 ( 2.85) 18.12 ( 2.67) 20.72 ( 6.35) 16.66 ( 3.30) Slope (%) 3.67 ( 3.06) 2.67 ( 2.79) 2.77 ( 2.73) 3.80 (3.09) 3.40 ( 4.45) 3.20 ( 2.56) 4.30 ( 2.77) 3.20 ( 1.79) 2.20 ( 2.75) 2.30 ( 4.88) 3.0 ( 1.22) 2.4 ( 2.19) 1.5 ( 0.5) 1.9 ( 2.13) 5.3 ( 3.83) Total Tree Cover (%) 37.0 ( 16.89) 39.0 ( 12.09) 57.5 ( 22.27) 24.0 ( 13.42) 34.0 ( 5.48) 32.0 ( 4.47) 58.0 ( 17.89) 30.0 ( 12.25) 32.0 ( 10.95) 48.0 (4.47) 46.0 ( 8.94) 34.0 ( 13.42) 44.0 ( 11.94) 70.0 (4.47) 82 ( 8.37) Sapling Cover (%) 15.78 ( 16.65) 14.25 ( 10.47) 41.15 ( 27.49) 3.90 ( 3.58) 7.20 ( 3.83) 14 ( 5.48) 38 ( 19.24) 4.40 ( 3.13) 8.60 ( 7.13) 20.0 ( 4.47) 24.0 ( 5.48) 16.0 ( 5.48) 21 ( 16.36) 53.6 ( 13.13) 74.0 ( 16.73) Stand Height (m) 11.24 ( 7.18) 15.05 ( 3.10) 3.54 ( 0.93) 13.32 ( 6.36) 8.46 ( 7.33) 12.80 ( 8.17) 10.38 ( 8.07) 13.20 ( 1.89) 13.80 ( 1.79) 15.80 ( 4.27) 17.42 ( 2.59) 3.28 ( 1.16) 3.24 (0.54) 4.0 ( 0.94) 3.64 ( 1.06) Canopy Fuel Height (m) 3.57 ( 5.45) 5.87 (4.86) 0.53 ( 1.41) 9.60 ( 5.77) 0.50 ( 0.87) 4.20 ( 5.76) 0 7.0 ( 4.47) 8.0 ( 3.16) 0.20 ( 0.45) 8.30 ( 5.19) 0.24 ( 0.33) 0.29 ( 0.40) 1.4 ( 2.81) 0.22 ( 0.13) Saplings / ha 77.50 ( 57.02) 66 ( 36.91) 745.90 ( 1207.48) 66.0 (27.02) 82.0 ( 47.64) 48.0 ( 16.43) 114.0 (96.10) 28.0 (19.24) 82.0 ( 32.71) 72.0 (41.47) 82.0 (28.64) 230.0 ( 60.42) 2249.60 ( 1770.91) 268.0 ( 46.58) 236.0 ( 118.45) Trees / ha 6664.0 ( 2540.27) 9249.5 ( 3449.55) 5090.0 ( 2866.65) 8066.0 ( 638.93) 5012.0 ( 1612.38) 5078.0 ( 737.61) 8500.0 ( 3724.68) 9828.0 ( 3434.22) 9042.0 ( 2927.76) 7430.0 ( 1027.13) 10698.0 ( 5279.05) 4610.0 ( 3595.16) 4140.0 ( 3024.33) 6388.0 ( 510.66) 5222.0 ( 3627.34) Total Fuel Load (kg/ha) 3.14 ( 3.16) 0.615 ( 0.46) 6.32 ( 10.07) 1.94 ( 0.67) 1.76 ( 1.38) 2.56 ( 1.44) 6.3 ( 0.67) 0.52 ( 0.45) 0.32 (0.40) 1.02 ( 0.51) 0.60 ( 0.24) 8.14 ( 10.70) 11.42 ( 17.33) 2.84 (1.13) 2.9 (0.87) Notes: Recorded in the Juniper Prairie Wilderness, Ocala Nationa l Forest, Florida and listed by st and type and ocular severity class for each variable.

PAGE 81

81 Table 2-10. Burn severity regression model results based on data collected in the Ju niper Prairie Wilderness, Ocala National Fo rest, Florida. Severity Index Seedling / Sapling Pole Damaged Pole R2 RMSE Variables Variable Coefficients Intercept R2 RMSE Variables Variable Coefficients Intercept R2 RMSE Variable Variable Coefficients Intercept CBI Substrate 0.7 6 0.55 Tree Cover -4.2244 2.404 0.69 0.53 Sapling Cover -5.4873 2.6916 0.60 0.65 Sapling Cover -4.1373 2.2551 Elevation 0.0684 Elevation -0.0383 Aspect -0.8626 CBI Herbs 0.7 6 0.53 Sapling Cover -3.2852 1.4977 0.6 0.67 Sapling Cover -5.8242 2.7926 0.56 0.75 Sapling Cover -4.9348 2.2284 Elevation 0.0619 Elevation -0.0361 ** CBI Tall Shrubs 0.6 6 0.67 Tree Cover -3.9495 2.093 0.71 0.65 Sapling Cover -6.4863 3.4378 0.55 0.79 Sapling Cover -5.0971 2.3541 Elevation 0.0819 Elevation -0.0551 ** CBI Understory 0.7 3 0.60 Tree Cover -4.1795 2.5348 0.65 0.54 Sapling Cover -4.4989 2.7507 0.60 0.65 Sapling Cover -4.5826 4.3054 Elevation 0.0626 Elevation -0.0454 ** CBI Overstory 0.65 0.80 Sapling Cover -6.7672 3.6062 0.64 0.74 Canopy Fuel Height 0.1780 0.4635 Elevation -0.0612 ** CBI Total 0.7 3 0.60 Tree Cover -4.1795 2.5348 0.77 0.52 Sapling Cover -7.4108 3.6237 0.63 0.67 Sapling Cover -4.7464 2.4462 Elevation 0.0626 Stand Height -0.0875 Aspect -0.7280 Supervised 0.6 5 0.71 Sapling Cover -3.1485 0.6726 0.74 0.67 Sapling Cover -6.2493 4.0623 0.65 0.72 Sapling Cover -5.2387 3.3409 Elevation 0.1192 Elevation -0.0745 Elevatio n -0.0395 Notes: Model results are listed by stand type and ocular severity class for each burn severity index modeled. The CBI Big Tre es and CBI Intermediate Trees burn severity metrics were not used in the regression modeling due to a lack of intermediate and big tre es found in the plots. (*)The CBI Overstory burn severity metric was not used in the regression modeling of the SS stand type due to a lack of overstory in the SS plots. (**) Additional explanator y variables were not added to the regression model if the additio nal R2 and RMSE values were not significant for F

PAGE 82

82 Table 2-11. Sample plot variable means and standard deviations for plot level variables not selected for burn severity regress ion modeling. Variable Damaged Pole Pole Seedling / Sapling High Severity Mid Severity Light Severity Unburned High Severity Mid Se verity Light Severity Unburne d High Severity Mid Severit y Light Severity Unburned Seedling Cover (%) 17.0 ( 7.32) 20.0 ( 12.56) 17.4 ( 12.95) 16.0 (.4) 20.0 ( 7.0) 14.0 ( 5.5) 18.0 ( 4.47) 22.0 ( 10.95) 8.0 ( 10.95) 30.0 ( 12.25) 20.0 ( 7.0) 15.20 ( 13.8) 20.60 ( 17.93) 20.0 ( 14.58) 14.0 ( 5.8) Medium Tree Cover (%) 0.3 ( 0.92) 0.5 ( 2.2) 0 0 0.6 ( 1.34) 0 0.6 ( 1.34) 0 0 2.0 ( 4.47) 0 0 0 0 0 Pole Tree Cover (%) 3.4 ( 4.87) 12.2 ( 9.58) 1.5 ( 6.7) 1.9 ( 1.52) 4.6 ( 8.7) 5.8 ( 3.83) 1.3 ( 1.56) 7.2 ( 3.83) 10.6 ( 6.06) 17.2 ( 16.87) 14.0 ( 5.48) 0 0 0 6.0 ( 13.42) Total Shrub Cover (%) 13.9 ( 9.28) 13.25 ( 9.28) 16.65 ( 10.67) 10 ( 0) 12.6 (.33) 14.0 ( 5.48) 19.20 ( 16.39) 15.20 ( 11.86) 9.20 ( 6.98) 10.60 ( 6.07) 18.0 ( 10.95) 24.0 ( 15.17) 16.0 ( 8.94) 8.6 ( 3.13) 18.0 ( 8.37) Low Shrub Cover (%) 10.8 ( 5.55) 9.2 ( 7.18) 13.7 ( 7.14) 10 ( 0) 12.6 ( 7.33) 12.0 ( 4.47) 8.6 ( 7.73) 13.2 ( 11.69) 7.2 (.8) 5.8 ( 3.83) 10.6 ( 6.06) 16.0 ( 5.48) 16.0 ( 8.94) 6.7 ( 4.60) 16.0 ( 5.48) Medium Shrub Cover (%) 1.7 ( 2.45) 3.3 ( 4.17) 1.4 ( 3.07) 1.2 ( 1.64) 0.6 ( 1.34) 1.8 ( 1.64) 3.2 ( 4.08) 0.7 ( 1.30) 2.6 ( 4.33) 3.8 ( 3.70) 6.0 ( 5.48) 0 0.2 ( 0.27) 2.6 ( 4.33) 2.8 ( 4.19) Tall Shrub Cover (%) 1.7 ( 4.88) 1.3 ( 3.1) 1.0 ( 4.47) 0 0 2.0 ( 4.47) 4.7 ( 8.64) 0.6 ( 1.34) 0 2.0 ( 4.47) 2.7 ( 4.27) 0 0 0 4.0 ( 8.94) Graminoid Cover (%) 0.6 ( 0.81) 1.4 ( 1.37) 0.6 ( 1.04) 0.5 ( 0) 0.4 ( 0.22) 0.9 ( 1.19) 0.9 ( 1.19) 1.8 ( 1.64) 1.5 ( 1.37) 1.8 ( 1.64) 0.4 ( 0.22) 0.4 ( 0.22) 2.0 ( 1.37) 0.1 ( 0.22) 0.1 ( 0.22) Forb Cover (%) 0.07 ( 0.18) 0.17 ( 0.24) 0.2 ( 0.25) 0.3 ( 0.27) 0 0 0 0.2 ( 0.27) 0.2 ( 0.27) 0.1 ( 0.22) 0.2 ( 0.27) 0.3 ( 0.27) 0.5 ( 0) 0 0 Moss / Lichen Cover (%) 7.4 ( 13.62) 3.0 ( 5.34) 1.1 ( 2.33) 0 0.2 ( 0.27)) 3.4 ( 3.89) 26.0 ( 16.73) 0 0.1 ( 0.22) 3.4 ( 3.89) 8.7 ( 7.59) 0 0.1 ( 0.22) 1.5 ( 1.37) 2.9 ( 4.11) Soil Cover (%) 7.1 ( 8.32) 5.0 ( 6.56) 7.7 ( 7.85) 14.0 ( 5.48) 10.7 ( 11.58) 3.9 ( 3.58) 0 14.0 ( 5.48) 3.3 ( 4.0) 2.9 ( 4.11) 0 13.2 ( 11.69) 12.0 ( 4.47) 2.5 ( 1.11) 3.3 ( 4.0) Woody Ground Cover (%) 6.7 ( 9.01) 3.4 ( 4.86) 6.7 ( 7.76) 0 2.8 ( 4.19) 3.9 ( 3.58) 20 ( 7.07) 0.2 ( 0.27) 1.5 ( 1.37) 9.2 ( 6.98) 2.5 ( 1.12) 2.7 ( 4.27) 2.7 ( 4.27) 5.3 ( 4.41) 16.0 ( 8.94) Char Ground Cover (%) 29.5 ( 27.73) 16.9 ( 22.17) 27.7 ( 29.14) 64.0 ( 20.74) 34.0 ( 18.15) 18.6 ( 12.03) 0 52.0 ( 10.95) 9.2 ( 6.98) 6.4 ( 7.60) 0 66 ( 18.16) 40 ( 7.07) 4.8 ( 4.85) 0.1 ( 0.22) Basal Vegetation Cover (%) 11.5 ( 3.66) 13.4 ( 2.29) 23.5 ( 15.65) 10 ( 0) 10 ( 0) 12 ( 4.47) 14 ( 5.48) 12 ( 4.47 16 ( 5.48) 20 ( 7.07) 5.8 ( 3.83) 14 ( 11.40) 18 ( 8.37) 34 ( 23.02) 28 ( 10.95)

PAGE 83

83 Table 2-11. Continued. Variable Damaged Pole Pole Seedling / Sapling High Severity Mid Severity Light Severity Unburned High Severity Mid Severity Light Severity Unburned High Severity Mid Severity Light Severity Unburned Litter / Duff Cover (%) 44.67 ( 33.24) 62.0 ( 26.48) 51.2 ( 34.05) 14.6 ( 14.52) 20.1 ( 18.57) 54.0 ( 11.40) 90.0 ( 0) 26.0 ( 18.16) 62.0 ( 16.43) 70.0 ( 7.07) 90.0 ( 0) 22.7 ( 28.86) 22.0 ( 13.04) 72.0 ( 13.04) 88.0 ( 4.47) Canopy Cover (%) 13.72 ( 14.30) 19.95 ( 13.70) 23.0 ( 30.10) 3.30 ( 3.99) 4.6 ( 5.08) 19.0 ( 18.6) 28.0 ( 8.37) 12.6 ( 7.33) 7.2 ( 3.83) 30.0 ( 15.81) 30.0 ( 7.07) 0 0 32.0 ( 27.75) 60.0 ( 20.0) Seedlings per ha 6560 ( 2505.87) 9070 ( 3436.81) 4342 ( 2888.81) 8000.0 ( 651.92( 4920.0 ( 1588.08) 4980.0 ( 746.32) 8340.0 ( 3645.27) 9800.0 ( 3426.37) 8960.0 ( 2940.75) 7220.0 ( 1044.99) 10300.0 ( 5270.19) 4380.0 ( 3607.91) 1890.4 ( 1295.63) 6120.0 ( 491.93) 4980.0 ( 3567.49) Snags per ha 122.0 ( 49.37) 166.5 ( 128.77) 15.5 ( 25.01) 132.0 ( 54.50) 160.0 ( 49.48) 126.0 ( 16.73) 70.0 ( 23.45) 292 ( 47.64) 260 ( 101.49) 74 ( 65.03) 40 ( 25.49) 10.0 ( 12.25) 0 6.0 ( 8.94) 46.0 ( 33.61) Duff / Litter Depth (cm) 3.9 ( 1.04) 3.5 ( 1.60) 3.4 ( 2.31) 1.1 ( 0.38) 3.6 ( 0.63) 4.9 ( 0.71) 6.1 ( 0.32) 1.5 ( 0.38) 3.8 ( 0.65) 5.4 ( 1.28) 3.6 ( 0.49) 2.4 ( 1.80) 1.4 ( 0.54) 4.2 ( 1.12) 5.6 ( 2.70) 1hr Fuel (kg/ha) 0.04 ( 0.048) 0.03 ( 0.037) 0.02 ( 0.024) 0.01 ( 0) 0.02 ( 0.01) 0.02 ( 0.01) 0.12 ( 0.03) 0.008 ( 0) 0.016 ( 0.01) 0.024 ( 0.01) 0.094 ( 0.02) 0.002 ( 0) 0.006 ( 0.01) 0.02 ( 0.01) 0.05 (0.03) 10hr Fuel (kgh/ha) 0.15 ( 0.17) 0.14 ( 0.15) 0.08 ( 0.09) 0.04 ( 0.03) 0.06 ( 0.02) 0.11 ( 0.12) 0.4 ( 0.13) 0.02 ( 0.01) 0.06 ( 0.03) 0.16 ( 0.17) 0.32 ( 0.09) 0.02 ( 0.02) 0.06 ( 0.04) 0.07 ( 0.06) 0.17 ( 0.13) 100hr Fuel (kg/ha) 0.34 ( 0.29) 0.14 ( 0.13) 0.24 ( 0.23) 0.21 ( 0.15) 0.19 ( 0.08) 0.33 ( 0.36) 0.60 ( 0.32) 0.13 ( 0.13) 0.08 ( 0.09) 0.22 ( 0.18) 0.11 ( 0.07) 0.12 ( 0.05) 0.17 ( 0.14) 0.19 ( 0.20) 0.49 ( 0.29) 1-100hr Fuel (kg/ha) 0.53 ( 0.46) 0.31 ( 0.22) 0.34 ( 0.31) 0.26 ( 0.18) 0.28 ( 0.09) 0.46 ( 0.46) 1.20 ( 0.41) 0.16 ( 0.12) 0.15 ( 0.10) 0.41 ( 0.27) 0.52 ( 0.14) 0.15 ( 0.08) 0.24 ( 0.14) 0.28 ( 0.25) 0.71 ( 0.37) 1000hr Sound Fuel (kg/ha) 1.90 ( 1.75) 0.28 ( 0.35) 3.00 ( 7.16) 1.27 ( 0.37) 1.23 ( 1.02) 1.57 ( 0.89) 3.53 ( 2.87) 0.34 ( 0.37) 0.16 ( 0.24) 0.60 ( 0.38) 0.026 ( 0.04) 4.63 ( 9.42) 5.54 ( 11.37) 1.34 ( 1.63) 0.49 ( 0.62) 1000hr Rotten Fuel (kg/ha) 0.69 ( 1.13) 0.03 ( 0.07) 2.98 ( 3.96) 0.38 ( 0.58) 0.22 ( 0.29) 0.54 ( 0.45) 1.64 ( 1.97) 0.02 ( 0.04) 0.04 ( 0.09) 0.02 ( 0.04) 0.06 ( 0.09) 3.34 ( 4.00) 5.64 ( 6.56) 1.26 ( 0.71) 1.68 ( 0.13) 1000hr Total Fuel (kg/ha) 2.59 ( 2.81) 0.32 ( 0.37) 5.98 ( 10.16) 1.65 ( 0.75) 1.45 (.28) 2.11 ( 1.22) 5.17 ( 4.74) 0.36 ( 0.40) 0.20 ( 0.32) 0.62 ( 0.40) 0.09 ( 0.10) 7.97 ( 10.79) 11.18 ( 17.35) 2.6 ( 1.18) 2.18 ( 0.70) Notes: Recorded in the Juniper Prairi e Wilderness, Ocala National Fore st, Florida and listed by stand t ype and ocular severity class for each variable.

PAGE 84

84 Figure 2-1. Location of Pinus clausa stands and sample plots in the Juniper Prairie Wilderness of the Ocala National Fo rest, Florida, USA.

PAGE 85

85 Figure 2-2. Sample plots for recording burn seve rity, fuel loading, topographic and vegetative characteristics of sand pine scrub stands in the Juniper Prairie Wilderness of the Ocala National Forest, Florida. 5 m 4 m Vegetation Plot N 10 m 10 m Plot Center 7 m 3 m Fuel Loading Starting Points Fuel Loading Transect (25 m) Fuel Loading Transect(25 Fuel Loading Transect (25 m) 4 m Vegetation Plot

PAGE 86

86 Figure 2-3. Map of burn severity following a 2006 fire in the Juniper Pr airie Wilderness of the Ocala National Forest, Florida. Map derive d from a Landsat 5 TM scene (path/row: 16/40) captured November 19, 2006. Image classified using a minimum distance supervised technique.

PAGE 87

87 Figure 2-4. Percent of area burned for each severity class by Pinus clausa stand type within the Juniper Prairie Wilderness, Ocala Nationa l Forest, Florida. Severity classes indicated: 0 (unburned) 1 (low severity) 2 (mid severity) 3 (high severity). Stand type as indicated: DP (damaged pole) PP (pole) SS (seedling / sapling). Aral extent calculated from a Landsat supervised classification.

PAGE 88

88 Figure 2-5. Multiple regression model output us ing plot level data and CBI burn severity values derived from a supervised imag e classification. A) Pole Class B) Damaged Pole Class C) Seedling / Sapling Class 0 0.5 1 1.5 2 2.5 3 3.5 4 01020304050Burn Severity (CBI)Elevation (m)A Sapling Cover = 10% Sapling Cover = 20%R2= 0.77 RMSE = 0.52 0 0.5 1 1.5 2 2.5 3 3.5 4 05101520253035Burn Severity (CBI)ElevationB Sapling Cover = 10% Sapling Cover = 20%R2= 0.65 RMSE = 0.7 2 0 0.5 1 1.5 2 2.5 3 3.5 4 051015202530Burn Severity (CBI)Elevation (m)C Tree Cover = 20% Tree Cover = 40%R2= 0.73 RMSE = 0.60

PAGE 89

89 Figure 2-6. Plot of correlation be tween plot Percent Sapling Cove r and plot Total Percent Tree Cover recorded at the Juniper Prairie W ilderness, Ocala National Forest, Florida 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Total % Tree Cover R2: 0.89 RMSE: 0.095

PAGE 90

90 CHAPTER 3 HIGH SEVERITY PRESCRIBED FIRE IN FLOR IDA: SAND PINE SC RUB MANAGEMENT IN THE OCALA NATIONAL FOREST Introduction Prescribed fire has becom e one of the prim ary land management practices used in many ecosystems throughout the southern United States. Pr escribed fire is commonly used to maintain ecosystems in specific seral stages, to aid in site preparation and regeneration, to reduce hazardous fuel loads associated with wildfire ris k, and to perpetuate fire dependent ecosystems. In the southern US, prescribed fire is most commonly associated w ith the longleaf pine ( Pinus palustris P. Mill.) savannahs and the slash pine ( Pinus elliottii Engelm.) flatwoods that dominate much of the upland landscape. These relatively open forest types, under low to moderate fuel loads, are conducive to the applica tion of prescribed fire. Such fires are typically low intensity, windand fuel-driven surface fires that can be relatively safely and easily controlled by competent and skilled crews. The USDA Forest Service Ocala National Forest (ONF) in north central Florida is home to the largest remaining tract of sand pine ( Pinus clausa var. clausa D.B. Ward) scrub in the world. The sand pine scrub of the ONF contrasts with the fuels and fire regi mes found in most other ecosystems in the southeastern US. Typically burning at much longer fire return intervals than the proximate slash and longleaf pi ne forests (from 1-6 year fre quencies), sand pine scrub often contains higher fuel loads at the time of bur ning. These high fuel loads (woody fuel load: 10.76 Mg/ha (sand pine scrub) v. 1.57 Mg/ha (longleaf pine sand hill), along with the high density (trees > 10.2 cm / dbh: 536 (sand pine scrub) v. 161 (longleaf pine sand hi ll) contribute to the high-intensity, stand-replacing fire regime (Ottmar et al. 2008). Sand pine scrub fires are often described as catastrophic. Fires such as this are generally feared, rather than prescribed, by managers and burn bosses. However when sa nd pine scrub falls within the management

PAGE 91

91 boundaries of agencies charged with mainta ining the natural, historical components and processes specific to the ecosystem, managers mu st face the challenge of somehow incorporating fire into management. In the case of the 1,572 km2 ONF, managers are tasked with maintaining ecosystem processes while accommodating multiple use objectives, including the 56 km2 Juniper Prairie Wilderness area (JPW) dominated by dense stands of sand pine scrub. The JPW is the largest and most protected tract of sand pine scrub in the world (Gre enburg, 1996). Federally designated as a wilderness area in 1984, the JPW falls under the resource management protection of the United States 1964 Wilderness Act (Wilder ness Act, 1964). In an effort to further preserve the character of the ar ea, the Act dictates that there shall be no temporary road, no use of motor vehicles, motorized equipment or motorboats, no landing of aircraft, no other form of mechanical transport, and no structure or instal lation within the desi gnated Wilderness Area (Wilderness Act of 1964). These acc ess restrictions apply both to users of the area and to the land management authority, in this case, the ONF. Therefore, many of the tools managers commonly use to control and manipulate prescribed fire in the southern US are not permissible within the JPW. This significantly increases th e complexity of firing operations and control, along with the risk to neighbor ing land owners and property. Understanding the challenges, as well as potential solutions, to using prescribed fire in the ONF sand pine scrub is critical, as the survival of the unique ecotype is cl osely linked to its fire regime. This paper thus establishes the platfo rm for fire use in the sand pine scrub through a description of its fire ecology. Ne xt, the challenges associated with high intensity prescribed fire both within and outside of wilder ness areas are identified, and fi nally, the future of sand pine scrub management at the ONF is investigated.

PAGE 92

92 Sand Pine Scrub A regionally endem ic natural community, sand pi ne scrub is found only in Florida and in a small coastal region in southern Alabama (Myers, 1985). Occupying a geographic region comprised of Miocene-Pleistocene marine and ae olian depositions, sand pine scrub exists on dry, well-drained sandy soils with low nutrient levels (Ca rrington, 1999; Myers, 1985). Predominantly, this region is an ancient coas tal dune running along the central Brooksville Ridge of peninsular Florida stretching from St. Johns County to Dade County. Within the JPW, sand pine scrub is the dominant natu ral community type. Sand pine scrub is typically dominated by an even-aged monoculture overstory of sand pine. Sand pine scr ub regions of inland peninsular Florida tend to be comprised of stands of fi re-dependent Ocala sand pine, while regions of coastal sand pine scrub tend to be comprised of stands of fire independent Choctahatchee sand pine ( P. clausa var. immuginata D.B. Ward). The understory and midstory are generally dominated by scrubby sclerophyllic oaks, ( Quercus chapmanii Sarg. Q. myrtifolia Willd Q. geminate Small), saw palmetto ( Serenoa repens (Bart.) Small), scrub palmetto ( Sabal etonia Swingle ex Nash), tree lyonia ( Lyonia ferruginea (Walt.) Nutt.), red bay ( Persea humilis Nash) and rosemary ( Ceratiola ericoides Michx.) (Outcalt and Gree nberg, 1998; Greenberg, 1996; Menges et al. 1993; Myers, 1985; Veno, 1976). Re lying on late-spring thunderstorm lightning ignitions, sand pine scrub typically burns every 30-60 years, much less frequently than the neighboring slash pine flatwoods or longleaf pine sandhills (Fonda, 2001; Menges et al. 1998; Richardson et al. 1988; Myers, 1985;). The fire -resisting nature and the extreme differences between scrub and its neighboring sandhills have b een noted by naturalists and researchers since the late 1800s (Myers, 1985; Nash, 1895). The ab ility to stop fires coming from adjacent vegetative communities is in part due to th e patchy ground cover and compact, shallow litter layer (Outcalt, 2003). This discon tinuous understory lacks the fine fuels needed to carry surface

PAGE 93

93 fires into the scrub. Additionally, the flammab ility characteristics of sand pines are different from other Florida pine species. Sand pine has been classified as a fir e evader and existing in a scrub community classified as fire-resili ent (Fonda, 2001; Rowe, 1984). Fire evaders are species that do not have the th ick protective bark, self-prunin g, or other resistance-conferring physical characteristics to shield them from cambial or crown damage caused by fire (Rowe, 1983). Fire-resilient communities are generally comprised of fi re evader species that are typically killed by infrequent in tense fires. The heat from su ch fires opens the serotinous (literally, Late-opening) cones of the fire evad ers to release seeds, which were stored in the canopy prior to the passing of fire (Fonda, 2001). As a fire evader, sand pine initial needle combustion is very slow, requiring more heat for flaming combustion than needles of other southern pines. In one study, sand pine needle f lame time (the duration of the visible flaming period of needle combustion when exposed to an ignition source in a fi re chamber; Fonda, 2001), was second to last of the species tested, followed only by jack pine (Pinus banksiana Lamb.), another crown-fire regime species (Fonda 2001) Eventually however, sand pine scrub does burn, following long fire-free periods that result in extremely hea vy fuel loads, which coincide with extended droughty conditions. Th is typically occurs in the la te spring months from March to May, before peninsular Florida summer weather patterns bring consistent rain. When fires do occur in sand pine scrub, they are often high severity stand-replacement burns that dramatically alter the structure of the scr ub until stands recover via natura l regeneration (Abrahamson and Abrahamson, 1996; Myers, 1986). Understanding and maintaining na tural fire regimes in sand pi ne scrub is important for species richness, wildlife habi tat, and the perpetuation of this endemic ecotype. Following disturbance, the diversity and ri chness of sand pine scrub herbaceous species have been found to

PAGE 94

94 be highest (Greenberg et al. 1995). Prolonged fire suppression or high fire frequency in sand pine scrub has been shown to lead to tran sitions to different community composition and structure, possibly resulting in the loss of endemic or rare sp ecies (Menges et al. 1998, Myers, 1985; Veno, 1976; Laessle, 1968). Veno (1976) found that following 40 years of fire suppression, over a study period of 20 years, scrub basal area increased 683%, indicating extreme community structure chan ges (Veno, 1976). In long term studies of fire suppressed sandhill and sand pine scrub, succes sional shifts to a xeric oak ha mmock have been observed as shade-tolerant oaks invade previously pine-dominated sites (Abrahamson and Abrahamson, 1996; Myers, 1985; Veno, 1976; Webber, 1933). Ocala sand pine ( P. clausa var. clausa) is an obligate seeder that releases seeds from serotinous cones when sufficiently heated by fire similar to jack pine, lodgepole pine ( P. contorta Dougl. ex Loud.) and giant sequoia ( Sequoiadendron giganteum (Lindl.) Buchh.) (Custer and Thorsen, 1996; Myers, 1986). Without fire, Ocala sand pine releases only limited seed and has very limited regene ration. Under natural, fire-free stand conditions, sand pi ne mortality increases dramatica lly after 40 years (Ross, 1970). A study of old growth sand pine stands found that the average age of old growth trees was only 55 years likely due to root rot fungi, a common cause of sand pine mortality (Outcalt 1997). Periodic strong winds from hurricanes also contri bute to sand pine mortal ity, both in stands of Choctahatchee and Ocala sand pine (Drewa et al 2007). The effects of hurricanes in Ocala sand pine however are not as significant in shaping st and age and structure as periodic high intensity fire (Drewa et al. 2007). Short fi re return intervals in the range of 5-15 year s are likely to result in a natural community shift from sand pine sc rub to longleaf pine or slash pine dominated sandhill or flatwoods (depending on nearby seed sources and topographic location) (Myers, 1985). This species shift due to fr equent fires is the result of sa nd pine seedling mortality (which

PAGE 95

95 are vulnerable to fire, and do not produce viable seed until approximately 5 years of age) (Fonda, 2001; Veno, 1976). Hence, maintaining the appropriate fire frequency, as well as fire type, is essential for the perpetuation of the sand pine scrub ecotype. Sand Pine Scrub Management at the ONF The fire regim e and management of sand pine scrub is similar to the jack pine ( Pinus banksiana ) stands of the upper Lower Peninsula Mich igan. Jack pine in habits well drained glacial deposits in dense even-a ged stands that are periodically burned in high-severity stand replacement burns (Houseman and Anderson, 2002). For the benefit of the Kirtlands warbler ( Dendroica kirtlandii S. F. Baird) an endangered species with habita t requirements similar to the Florida Scrub Jay (Aphelocoma coerulescens Bosc)(FSJ), some jack pi ne stands are maintained in a landscape mosaic of stands aged 50 years or younger. The sand pine scrub of the ONF has been under federal management since 1908, when it was ceded to the government by an act of Presid ent Theodore Roosevelt (S ekerak and Hinchee, 2001). The ONF sand pine scrub is managed un der an extensive silviculture program for multiple use objectives including sand pine timber harvests, Florida FSJ habitat management and public recreation (USDA Forest Service, 2006 Monitoring and Evaluation Report). The FSJ was listed as a federally threatened species in 1987 with its population decline primarily attributed to habitat loss. The FSJ has been chosen as an indicator species of healthy sand pine scrub by the ONF (USDA Forest Serv ice, 2006 Monitoring and Evaluation Report). Since 1987, the ONF has managed sand pine scrub to provide suitable habitat for the FSJ while continuing to manage for clear-cut timber harves ts of sand pine. The FSJ has very specific habitat requirements, primarily defined by the hei ght of the scrub oak understory and mid-story. Optimal FSJ habitat is defined as oak scr ub heights of 1.2 1.7 m, interspersed among a landscape mosaic of varying vegetation heights (the result of multiple scrub seral stages)

PAGE 96

96 (Breininger, 1992). The ONF has determined that clear-cutting mature sand pine results in scrub regeneration that provides suitable nesting hab itat for the FSJ for 10-12 years following the cut (USDA Forest Service, 2006 Monitoring and Evalua tion Report). FSJ populations are monitored and recorded in surveys covering approximately 25% of the suitable habitat within the ONF, per year (USDA Forest Service, 2006 Monitoring and Evaluation Report). Current management of sand pine scrub stands in the ONF calls for retention of 5% of stands in older age classes (Sek erak and Hinchee, 2001). This ag e class retention plan is an effort to maintain sustainable early seral stage sand pine scrub for the benefit of the FSJ. Mature or old growth Ocala sand pine has been categorized as tree s in 60-70 years old range with maximum diameter at breast height (DBH) of 43-44 cm; fairly young and diminutive as compared to longer lived longleaf an d slash pine (Parker et al. 2000). Through decades of fire suppression and lands cape fragmentation, current scrub conditions likely do not reflect historical composition and st ructure (Parker et al. 2000). Fire suppression has likely led to the development of older, mo re mature stands of sand pine scrub wherein historical stands were likely comprised of younger, earlier seral stage scrub (Parker et al. 2000). These stands also have a well-developed mids tory comprised of oaks, suggesting eventual succession to xeric oak hammock if not burned. Pres ervation of the current old-growth stands as a percent of the overall vegetation type may not be mimicking historical conditions due to the varied nature of a landscape shaped high-seve rity stand-replacement fire (Parker et al. 2000). As a forest type that is associated with stand-replacing disturbanc es, 24 ha clear-cuts are justified by the Forest Service as mimicking natural disturbances (s tand-replacement fire) (R. Shelfer, USFS, pers. comm., 2008). The size of these clear cu ts however may not actually be mimicking the size of historic st and replacement burns. Followi ng the clear cuts, which ideally

PAGE 97

97 take place in late summer, surveys for adequate natural regeneration are conducted. In some instances, on-site sand pine serotinous cones contai ned within in the logging slash, open due to the summer heat, and reseed the site. If natu ral regeneration is deem ed successful, no other management steps are taken. However, if regene ration is unsuccessful, or insufficient to meet desired stocking levels, site pr eparation and direct seeding of sand pine seedlings (from commercial nurseries) is completed in the follo wing fall or winter season (R. Shelfer, USFS, pers. comm. 2008). In some cases, prescribed fire has been used as a site preparation tool, but this is not a common practice in the ONF. Outside of expe riments and demonstrations, prescribed fire is not used in the mana gement of sand pine scrub at the ONF. Management within the sand pine scrub stands of the JPW has been very different from the rest of the ONF. Due to the protections offere d by the wilderness designa tion, these stands have not been managed using mechanical silvicul ture practices. A 2006 wildfire burned 44 km2 of the JPW, originating as an escaped prescribed pr airie fire. A multi-disciplinary study of the 2006 JPW fire is currently underway through the collaborative efforts of the University of Florida and the Ocala National Forest. Challenges to Prescribed Fire in Sand Pine Scrub There are many challenges to c onducting prescribed fires in sa nd pine scrub that continue to lim it and complicate the use of fire for ma nagement. First and foremost, under certain conditions, particularly following se vere drought, sand pine scrub can burn with very active fire behavior and great intensity. In fact, the fastest spreading fire in U.S. history occurred in the ONF in 1935. Starting from a burning stump in th e SW corner of the ONF, the Big Scrub Fire burned 14,000 ha across 58 km in 4 hours (Seker ak and Hinchee, 2001). The 2006 JPW fire began in July as a prescribed fire operation in the open, grassy prairies within the wilderness area. During burning operations, the fire escaped the prairie and moved in to the then drought-

PAGE 98

98 stricken sand pine scrub. The scrub fire was reclassified as a wildfi re and required extensive manpower and equipment (including aircraft) to maintain containment within the wilderness area boundary. The high intensity nature of the fire resulted in multiple spot-overs and hazardous road conditions (flames and smoke) which required the closing of two state highways (J. Nobles, ONF staff, pers. comm., 2007). Such active fire behavior de monstrates the in creased perceived and real risks associated with burning in sand pine scrub. Prescribed burning in high-intensity, high-severity ecotypes often require more personnel and more equipment to manage than is required to burn in low-intensity, low-severity fire prone ecotypes. Such demands in equipment and personnel mean that the management costs of prescribed fire activity are significantly higher in sand pine sc rub than in low-severity, lowintensity fire type ecosystems wherein small crews can reasonably burn tens of hectares in a day. Prescribed fire operations are conducted under weather and fuels conditions conducive both to meeting management objectives and to maintaining containment of the fire. The temporal window within which prescribed fire ca n be applied and managed in sand pine scrub is very limited. Weather and fuel conditions th at are conducive to burning sand pine scrub are infrequent, as suggested by the 3060 year fire return interval Within the burnable window are even fewer days in which prescribed fire operations could be conducted. Forecasting and applying fire within such limited burning windows is a challenge unlikely to be mitigated. Landscape fragmentation and the encroachme nt of wildland urban interface (WUI) are expected to pose additional challe nges to the use of prescribed fire at the ONF well into the future (Menges, 2007; Radeloff et al. 2005). Florida has nearly 28,000 km2 classified as WUI, a region that is showing decadal increases (Radelo ff et al. 2005). Fueling those WUI increases is a state population increasing at ne arly twice the national averag e (US Census, Bureau, 2005).

PAGE 99

99 Prescribed fire operation complexities increase when the adjacent lands are heavily populated and when the intensity of the fire is high. An accident partially attributed to smoke from a nearby prescr ibed fire on Interstate 4 in Central Florida, in January of 2008, resulted in 38 injuries and 4 deaths. Incidents such as this demonstrate some of the real dangers associat ed with the juxtaposit ion of populations and managed wildlands. Florida has a significant population over the age of 65 (17.6% of total population) (US Census Bureau, 2005). Many elderly Floridians have legitimate health concerns associated with smoke created by prescribe fi re. Expected regional air quality regulation changes may also impact the ability of managers to apply fire to the landscape, especially in areas proximate to non-attainment zones. Smoke based concerns (health, traffic and air quality) are issues prescribed fires may have on surrounding populations that managers must contend with and mitigate, even beyond the fe nces of their management areas. In some cases, management directives designe d to protect sand pine scrub may have the opposite effect. The JPW has been a federally de signated wilderness area since 1984. In an effort to preserve the character of wilderness areas, strict regu lations are imposed preventing the use of wheeled or tracked machinery and gasoline powered engines. In the southeastern US, prescribed fire operations and wildfire suppres sion activities go handin-hand with bulldozers and plows. The terrain, vegetation, fuels and cult ure simply dont support the use of hand-crews. Limiting the use of these key land management tools has led to wilderness areas remaining relatively un-managed, as managers are unable to use their traditional tools (Murray, 1996). In the case of the JPW and other wilderness areas, th e actions taken to preserve the wilderness areas may be in-fact hurting the lands they we re designed to protect (Murray, 1996).

PAGE 100

100 The repercussions of the legacy of fire suppression in sand pine scrub may also prove to be a challenge (Menges, 2007). As has been demo nstrated in the western US and elsewhere, effective fire suppression programs often result in unnaturally high fuel loads accumulating over time. Sand pine scrub that has had fires supp ressed for extended periods is unlikely to be considered by land managers for prescribed fire due to the perceived risks associated with burning the high fuel loads. One of the most critical challenges to prescribed burning in sand pine scrub is an unintentional result of management policy. Sand pine scrub in the ONF that has been prescribed burned (experimentally), was burned in relatively small spatial blocks. These small burn units contrast with the vast landscape burns conducted annually in ecosystems with less severe fire regimes elsewhere in Florida. The prevailing agen cy metric of prescribed fire success in Florida and elsewhere is based on the areal extent of burned lands. Land management agencies, including the ONF, annually track their total hectar es burned, which are used as an indicators of land management success. For example, USDA Forest Service administrators set prescribed burn acreage goals for each FS Region, and each mana gement unit within that region must take responsibility for its own portion. The more acres burned from prescribed fire, the better. Harkening to the early 20th century forest management directives to get the cut out agency management doctrine now seems push to get the burn out This system favors the burning of low-intensity, moderate fire behavior ecoty pes which can be burned in larger blocks: contributing more towards annual hectares burned. The result of this spatially based success metric is that certain ecotypes, such as sand pi ne scrub, are continually pushed to the bottom of the priority list for burning. The multitudinous ecos ystem effects and services of prescribed fire are marginalized by the current spatial success metric.

PAGE 101

101 Experiments in Sand Pine Fire Management Multip le experiments within recent years have attempted to apply prescribed fires to sand pine scrub. These efforts are novel, as the reig ning dogma in most sand pine scrub management programs is to suppress fires at all cost. An experiment to prescribe burn sand pine scrub in the winter months in 1986 raised questions as to the effectiveness at achievi ng targeted sand pine regeneration and desired structural outcomes when attempting to burn out side of the typical la te spring-early summer burning period (Abrahamson and Abrahamson, 1996). Burning sand pine scrub in the winter months allowed managers to conduct fire operati ons under less extreme fuel moisture conditions resulting in lower intensity fires that are easier to cont rol (Abrahamson and Abrahamson, 1996). Such fires, however, may not have the resulting burn severity needed to achieve objectives, such as stand replacement through extensive overstory mortality and natural regeneration via the opening of serotinous cones (Abrahamson and Abrahamson, 1996). An experiment conducted on a 12 ha site in May of 1993 by the ONF attempted to prescribe a stand-replacement burn on sand pine scrub (Custer and Thorsen, 1996; Outcalt and Greenberg, 1996). Using the National Forest Fire Laboratory (NFFL) fuel model 4, in conjunction with the BEHAVE Fire Behavior Predictions System the ONF staff were able to determine the ideal fuel and meteorological condition parameters needed to achieve a controllable stand-replacement bur n with minimal spot fires (Custer and Thorsen, 1996; Outcalt and Greenberg, 1996; Andrews, 1986). The resulti ng fire consumed 98% of the understory and killed many of the sand pines (Custer and T horsen, 1996; Outcalt and Greenberg, 1996). The researchers documented after 30 days a significant number of sand pine seedlings emerging from the burned areas. This experiment in stand-replacement prescribed fire was deemed a success, due to the safe execution and containment of th e burn; but further stand replacement prescribed

PAGE 102

102 fires have not been conducted in the ONF (C uster and Thorsen, 1996; Outcalt and Greenberg, 1996). Fire and the Future Management of Sand Pine Scrub Maintaining the processes that shape sand pine scrub is m ore important for preserving endemic species and the overall ecotype than preserving existing stands and old growth sand pine (Parker et al. 2000). This notion is reflec ted in the USFS plans for the future management of sand pine scrub at the ONF. Future manageme nt plans call for a stand level mosaic of cutblocks designed to mimic wildfire disturbances and create a patch-work of various scrub seral stages. These cut-blocks will be strategically oriented to provide natural fuels reductions around stands targeted to be burned instead of cut (including wilderness area s) (R. Shelfer, USFS pers. comm., 2008). Cut-block sizes during 2000-2006 av eraged 23.0 ha with the 2006 mean cut block size at 29 ha. The desired larger futu re cut-block sizes will range from 65 ha (USDA Forest Service, 2006 Monito ring and Evaluation Report). Management of sand pine scrub, especially in wilderness areas where mechanical equipment is restricted, will require some sort of fire. Prescribed fires or wildland fires for resource benefit (WFR) have been suggested as management tools in wilderness areas (Parker et al. 2000; Outcalt, 1997). However, in the ONF a nd elsewhere, wilderness areas often exist as islands within landscapes of other manageme nt regimes. Due to landscape and fuels fragmentation, and wildfire suppr ession policies, WFR fire return intervals are unlikely to be sufficient to meet management and ecosystem n eeds. The current ONF plans will be to use prescribed fire to manage wilderness areas only after surrounding stands have been cut to provide fuel breaks and reduce risk (R. Shelfer, USFS, pers. comm. 2008).

PAGE 103

103 Research Needs Research co ntinues to be needed to guide th e future of sand pine scrub management, both within and outside of wilderness areas. Current and planned management practices need to be studied in terms of impacts on threatened species populations and overall biodiversity over both the short and long term. Future research will be needed to study the efficacy of the USFS shifting sand pine scrub management programs as well as current management in sand pine scrub wilderness areas. It is likely that prescribed fi res will be needed to maintain the sand pine scrub ecotype within wilderness areas where mechanical stand manipulations are not possible. Given the limitations and challenges of high-intensity prescribed fire, research is needed to provide managers with the knowledge of how to apply the right types of fires, to achieve the specific effects that will sustain ecosystem processes and preserve the original (JPW) wilderness character. To this effect, investigations of burn se verity and ecosystem response will be key. Further understanding of burn severity mapping me thods and burn severity drivers will allow managers to better predict the re sults of sand pine scrub prescrib ed fire as well as quantify and assess the results of both prescribed fires and w ildfires. Further understa nding of the drivers of burn severity and fire intensity in sand pine scru b may allow for safer application of fire while still attaining management objectives.

PAGE 104

104 LIST OF REFERENCES Abraham son WG (1984) Species responses to fire on the Florida Lake Wales Ridge American Journal of Botany 71, 35-43. Abrahamson WG, Abrahamson JR (1996) Effects of a low-inte nsity winter fire on longunburned Florida sand pine scrub Natural Areas Journal 16, 171-183. Andrews PL (1986) BEHAVE: fire behavior prediction and fuel modeling system-BURN Subsystem, part 1 U.S. Department of Agriculture, Fo rest Service, Intermountain Research Station, General Technical Report INT-194. (Ogden, Utah) Breininger DR, Schmalzer PA (199 0) Effects of fire and disturbance on plants and birds in a Florida oak / palmetto scrub community. American Midland Naturalist 123 64-74. Breininger DR, Smith RB (1992) Relationships betw een fire and bird density in coastal scrub and slash pine flatwoods in Florida. American Midland Naturalist 127, 233-240. Brown JK (1974) Handbook for inventorying downed woody material. U. S. Department of Agriculture, Forest Service Intermountain Forest and Range Experiment Station, General Technical Report INT-16. (Ogden, Utah) Carrington ME (1999) Post-fir e seedling establishment in Florida sand pine scrub. Journal of Vegetation Science 10, 403-412. Carrington ME, Keeley JE (1999) Comparison of post-fire seedling establishment between scrub communities in Mediterranean and non-Mediterranean climate ecosystems Journal of Ecology 87, 1025-1036. Clark J, Bobbe T (2007) Using remote sensing to map and monitor fire damage in forest ecosystems. In Understanding Forest Disturba nce and Spatial Pattern Remote Sensing and GIS Approaches. (Eds MA Wulder SE Franklin) (CRC Press: Florida) Cocke AE, Fule PZ, Crouse JE (2005) Comp arison of burn severity assessments using Differenced Normalized Burn Ratio and ground data International Journal of Wildland Fire 14, 189-198. Congalton RG, Green K (1999) Assessing the Accuracy of Remotely Sensed Data: Principles and Practices. (CRC Press: Florida) Custer G, Thorsen J (1996) Stand-replacement burn in the Ocala National Forest a success. Fire Management Notes 56, 7-12. Drewa PB, Platt WJ, Kwit C (2007) Stand structure and dynamics of sand pine differ between the Florida panhandle and peninsula Plant Ecology 196, 15-25.

PAGE 105

105 Epting J, Verbyla D (2005) Landscape-level inter actions of prefire vegetation, burn severity, and postfire vegetation over a 16-year period in interior Alaska Canadian Journal of Forest Research 35 ,1367-1377. Fonda RW (2001) Burning characteristics of needles from eight pine species Forest Science 47, 390-396. Greenberg CH, Neary DG, Harris LD, Linda SP (1995) Vegetation recovery following highintensity wildfire and silvicultu ral treatments in sand pine scrub American Midland Naturalist 133 149-163. Hawkes CV, Menges ES (1996) The relationship between open space and fire for species in a xeric Florida shrubland Bulletin of the Torrey Botanical Club 123, 81-92. Henry MC (2007) Comparison of singleand multidate Landsat data for mapping wildfire scars in Ocala National Forest, Florida Photogrammetic Engineering & Remote Sensing 74 881-893. Houseman GR, Anderson RC (2002) Effects of j ack pine plantation management on barrens flora and potential Kirtland s warbler nest habitat. Restoration Ecology 10, 27-36. Jensen JR (2005) Introductory digital image pr ocessing: a remote sensing perspective. In Geographic Information Science. (Ed KC Carke) (Prentice Hall: New Jersey) Key CH, Benson NC (2006) Landscape assessment: ground measure of severity, the Composite Burn Index. In FIREMON: Fire Effects Monitoring and Inventory System. (Eds DC Lutes, RE Keane, JF Caratti, CH Key, NC Benson, S Sutherland, LJ Gangi) USDA Forest Service, Rocky Mountain Research Stati on, General Technical Report RMRS-GTR-164CD: LA1-51. (Ogden, UT) Laessle AM (1968) The origin and successional relationship of sandhill vegetation and sand pine scrub Ecological Monographs 28, 361-387. Lentile LB, Smith FW, Shepperd WD (2006) In fluence of topography an d forest structure on patterns of mixed fire severity in ponderosa pine forests of the South Dakota Black Hills, USA. International Journal of Wildland Fire 15, 557-566. Liu C, Frazier P, Kumar L (2007) Comparativ e assessment of the measures of thematic classification accuracy Remote Sensing of Environment 107, 606-616. Menges ES, Abrahamson WG, Givens KT, Gallo NP, Layne JN (1993) Twenty years of vegetation change in five long-unburned Florida plant communities Journal of Vegetation Science 4, 375-386. Menges ES, Hawkes CV (1998) Inte ractive Effects of fire and m icrohabitat on plants of Florida scrub Ecological Applications 8, 935-946.

PAGE 106

106 Menges ES, Kohfeldt N (1995) Life History Strate gies of Florida Scrub Plants in Relation to Fire Bulletin of the Torrey Botanical Club 122, 282-297. Murray MP (1996) Natural processes: Wilderness management unrealized. Natural Areas Journal 16, 55-61. Myers RL (1995) Fire and the dynamic relationshi p between Florida sandhill and sand pine scrub vegetation Bulletin of the Torrey Botanical Club 112, 241-252. Myers RL, Ewel JJ (1990) Ecosystems of Florida. (University of Central Florida Press: Florida) Nash, GV (1895) Notes on Some Florida Plants Bulletin of the Torrey Botanical Club 22, 141161. Ottmar RD,Vihnanek RE, Mathey JW (2003) Ster eo photo series for quantifying natural fuels. Volume VIa: sand hill, sand pine scrub, and hardwoods with white pine types in the Southeast United States with supplemental sites for volume VI National Wildfire Coordinating Group, National Interagency Fire Center, PMS 838. (Boise, Idaho) Outcalt KW (1997) An old-growth definition for sand pine forests. U.S. Department of Agriculture Forest Service, Southern Res earch Station, General Technical Report SRS-12. (Asheville, North Carolina) Outcalt KW (2000) Decay of fire-caused snags in Ocala sand pine. In Fire Conference 2000: The First National Congress on Fire Ecology, Prevention, and Management. (Eds KEM Galley, RC Klinger, NG Sugihara) (Tall Ti mbers Research Station: Florida) Outcalt KW, Greenberg CH (1998) A stand-replacement prescribed burn in sand pine scrub. In Proceedings of the 20th Tall Timbers Fire Ecology Conference, Fire in Ecosystem Management: Shifting the Paradigm from S uppression to Prescrip tion. (Eds TL Pruden, LA Brennan) pg. 141-145 (Tall Timbers Research: Florida) Parker AJ, Parker KC, McCay DH (2001) Disturbance-mediated variation in stand structure between varieties of Pinus clausa (Sand Pine). Annals of the Association of American Geographers 91, 28-47. Radeloff VC, Hammer RB, Stewart SI, Fried JS, Holcomb SS, McKeefry JF (2005) The wildland-urban interface in the United States Ecological Applications 15 799-805. Richardson DR, Williamson GB (1988) Allelopathic effects of shrubs of the sand pine scrub on pines and grasses of the sandhills Forest Science 34, 592-605. Romme WH (1982) Fi re and landscape diversity in subalp ine forests of Yellowstone National Park. Ecological Monographs 52, 199-221.

PAGE 107

107 Rowe JS (1983) Concepts of fire effects on plant i ndividuals and species, In The Role of Fire in Northern Circumpolar Ecosystems. (Eds RW Wein, DA MacLean) p. 134-154 (Wiley: New York) Sekerak CM, Hinchee JK (2001) The evolution of Ocala National Forest's current sand pine scrub management program. In Florida Scrub Symposium 2001 (Ed D Zattau) p. 21-25 (U.S. Fish and Wildlife Service: Florida) Turner MG, Romme WH (1994) Landscap e dynamics in crown fire ecosystems Landscape Ecology 9, 59-77. USDA Forest Service (2006) Monitoring and Evaluation Report: National Forests in Florida USDA Forest Service, Region 8, Annual Report (Talla hassee, Florida) Veno PA (1976) Successional relationships of five Florida plant communities Ecology 3, 498508. Webber HJ (1933) The Florida scr ub, a fire-fighting association American Journal of Botany 22, 344-361. White JD, Ryan KC, Running SW (1996) Remote sens ing of forest fire se verity and vegetation recovery International Journal of Wildland Fire 6, 125-136. Wulder MA, Franklin SE (2007) Understanding Fo rest Disturbance and Spatial Pattern: Remote Sensing and GIS Approaches (CRC Press: Florida)

PAGE 108

108 BIOGRAPHICAL SKETCH David Robert Godwin was born in Jacksonvill e, Florida, in 1981, to Robert and Sarah Godwin. He grew up in Tallahasse e, Florida, after his fam ily moved from Jacksonville when he was an infant. The older of two, he has one brot her, Daniel, of Gainesville, Florida, also a graduate of the University of Florida. When David was growing up, his family always had a focus and commitment to continuing education and to understanding the natural world. These aspects profoundly influenced David as a yout h and have shaped his perspectives and aspirations. David graduated from the Florida State Un iversity School (Florida High) in 2000 and Florida State University in 2003 with a Bachelor of Science in geography. Before entering graduate school, he worked as a recreation trails specialist for the Florida Fish and Wildlife Conservation Commission, on state wildlife management areas where he developed an interest in land management and fire ecology. Upon completion of his Master of Science degr ee, David will begin a program leading to a Doctor of Philosophy at the School of Forest Res ources and Conservation at the University of Florida. On January 12, 2008, David married Mary Kath erine Kight of Orange Park and St. Augustine, Florida. They currently reside in St. Augustine, Florida.