|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 A SPATIAL TEMPORAL ANALYSIS OF VEGETATION CHANGE, LAND COVER CHANGE, AND HEALTH IMPACTS IN FLORIDA By HUIPING TSAI A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2012
2 2012 Huiping Tsai
3 To my family and friends, who, always support me
4 ACKNOWLEDGMENTS I would like to express my deepest appreciation to Professor Jane South worth, who served as my research advisor over the past four years With her broad knowledge, diligent working attitude, and most importantly her optimistic and kind sprit in daily life, Professor Jane Southworth to me is not only my research mentor; she is my role model both in academic and daily life. My special thanks go to Profess or Pete R. Waylen, who served as my Co chair and helped me a lot on developing valuable ideas to approach and resolve the challenging research questions. His personality and res earch spirit really encourage me and make m e always feel passionate on my work Many appreciations go to Professor Youliang Qiu, who served as my committee member and always helped me a lot on developing and solving technique issues of my research. His kin dness and open mind always amaze me and ease my anxiety. More thanks to Professor Xiaohui Xu, who served as my external committee member and educated me a lot on air pollution issues. His patience and kind ness encourage me keep on tract and pursue my acad emic goal. Finally and most importantly, I would like to thank for my supported parents, my wonderful husband and my cute son. With their endless support on this journey, I feel energetic every day and become a more confident research er with positive atti tude both in life and work I am always going to be grateful to have them by my side.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 12 2 SPATIAL PERSISTENCE AND TEMPORAL TRENDS IN VEGETATION COVER ACROSS FLORIDA, 1982 2006 ................................ ............................... 18 2.1 Background ................................ ................................ ................................ ....... 18 2.2 Methods ................................ ................................ ................................ ............ 20 2.2.1 AVHRR Vegetation Indices ................................ ................................ ..... 20 2.2.2 NOAA CCAP Land Cover Classification Data ................................ ......... 22 2.2.3 Mean Variance Analysis ................................ ................................ .......... 23 2.2.4 Spatial Persistence ................................ ................................ .................. 25 2.2.5 Statistical Test for Persistence Analysis ................................ .................. 26 220.127.116.11 Persistence analysis of direction of change ................................ ... 27 18.104.22.168 Persistence analysi s of absolute amount of change ...................... 28 2.3 Results ................................ ................................ ................................ .............. 30 2.3.1 Mean Variance Analysis ................................ ................................ .......... 30 2.3.2 Spatial Persistence with Statistical Test ................................ .................. 31 22.214.171.124 Persistence analysis of direction of change ................................ ... 31 2.3.2 .2 Persistence analysis of absolute amount of change ...................... 32 2.4 Discussion and Summary ................................ ................................ ................. 34 3 NDVI, CLIMATE VARIABILITY AND LAND COVER IN FLORIDA ......................... 50 3.1 Background ................................ ................................ ................................ ....... 50 3.2 Study Area ................................ ................................ ................................ ........ 52 3.3 Methods ................................ ................................ ................................ ............ 55 3.3.1 Remote Sensing Data ................................ ................................ ............. 55 3.3.2 Precipitation Data ................................ ................................ .................... 56 3.3.3 Climatic Divisions ................................ ................................ .................... 56 3.3.4 Land Cover Data ................................ ................................ ..................... 57 3.3.5 Time Series Approach and Wavelets ................................ ...................... 58 3.4 Results ................................ ................................ ................................ .............. 60 3.4.1 Climate Divisions ................................ ................................ ..................... 61 3.4.2 CCAP Land Cover Types ................................ ................................ ........ 62
6 3.5 Discussion ................................ ................................ ................................ ........ 62 3.6 Summary and Finding ................................ ................................ ....................... 65 4 IMPACTS OF LAND COVER AND LAND USE ON AIR QUA LITY IN FLORIDA .... 88 4.1 Background ................................ ................................ ................................ ....... 88 4.2 Data and Methods ................................ ................................ ............................. 90 4.2 .1 Particulate Matter (PM 2.5 ) ................................ ................................ ........ 90 4.2.2 Ozone (O 3 ) ................................ ................................ .............................. 91 4.2.3 Normalized Difference Vegetation Index (NDVI) ................................ ..... 92 4.2.4 Precipitation and Maximum Temperature Data ................................ ....... 93 4.2.5 NOAA CCAP Land Cover Classification Data ................................ ......... 94 ................................ ........................... 95 4.3 Results and Discussion ................................ ................................ ..................... 96 4.4 Summary and Finding ................................ ................................ ....................... 99 5 CONCLUSION ................................ ................................ ................................ ...... 109 LIST OF REFERENCES ................................ ................................ ............................. 111 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 119
7 LIST OF TABLES Table page 2 1 CCAP Land Cover Classification ................................ ................................ ........ 37 3 1 El Nio/La Nia Impacts ac ross the Southeast U.S. ................................ ........... 67 3 2 Selected C CAP Land Cover Classification Scheme ................................ .......... 68 3 3 Significant period CCAP land cover class ................................ ...................... 71 3 4 Significant period Climate Division ................................ ................................ .. 72 4 1 National ambie nt air quality standards ................................ .............................. 100
8 LIST OF FIGURES Figure page 2 1 Hypothetical relationship between mean variance and vegetation status, ........ 38 2 2 Mean Variance plot for annual NDVI ................................ ................................ .. 38 2 3 Mean Variance plot for winter NDVI ................................ ................................ ... 39 2 4 Mean Variance plot for summer NDVI ................................ ................................ 39 2 5 Mean Variance plot for monthly NDVI ................................ ................................ 40 2 6 Example maps for persistence analysis ................................ ............................. 42 2 7 Comparison of observed value from persistence analysis direction of change with simulated normal distribution result ................................ ............................. 43 2 8 Cumulative frequency plots for comparison of obser ved value from persistence analysis direction of change with simulated normal distribution ...... 44 2 9 Example of areas that identified by above or below critic classes ...................... 45 2 10 Example for developed land confidence bounds maps in January. .................... 46 2 11 Developed land confidence category percentage maps ................................ ..... 47 2 12 Agricultural land confidence category percentage maps ................................ .... 48 2 13 Palustrine wetlands confidence category percentage maps ............................... 49 3 1 Study area. ................................ ................................ ................................ ......... 73 3 2 Florida statewide average monthly precipitation ................................ ................. 74 3 3 Florida stat ewide average monthly precipitation anddistribution ........................ 74 3 4 Climate divisions in Florida ................................ ................................ ................. 75 3 5 Land cover classification data fr om C CAP ................................ ........................ 76 3 6 Dominant land cover types at different percentages ................................ ........... 77 3 7 Precipitation and NDVI time series present in monthly order July 1981 to December 2006. ................................ ................................ ................................ 78 3 8 ................................ ................................ ...... 79
9 3 9 ................................ ................................ .... 80 3 10 Wavelet analysis of NDVI for Florida climate divisions. ................................ ...... 81 3 11 Wavelet analysis of NDVI for different land cover types ................................ ..... 82 3 12 Wavelet analysis of precipitation of Florida. ................................ ....................... 84 3 13 Wavelet analysis of precipitation for Florida climate divisions ............................ 85 3 14 Wavelet analysis of precipitation fordifferent land cover types ........................... 86 4 1 Florida air monitor site location for particulate matter (PM 2.5 ) and ozone (O 3 ) with climate division boundary embedded ................................ ........................ 101 4 2 Florida air monitor site location for particulate matter (PM 2.5 ) and ozone (O 3 ) that locate in developed land use category ................................ ...................... 102 4 3 The correlation coefficients between PM 2.5 and O 3 with rain, maximum temperature and NDVI for all monitor sites ................................ ....................... 103 4 4 The c orrelation coefficients between PM 2.5 O 3 mean and O 3 maximum with rain, maximum temperature and NDVI for all monitor sites by land cover types ................................ ................................ ................................ ................. 104 4 5 The correlation coefficients between PM 2.5 and O 3 with rain, maximum temperature and NDVI for all monitor sites ................................ ....................... 105 4 6 The correlation coefficients between PM 2.5 and O 3 with rain, maximum temperature and NDVI for all monitor sites by climate divisions ....................... 106 4 7 Seasonal variation of the correlation coefficient by climate divisions. ............... 107 4 8 Correlation coeffi cients of pollutants with Buffer ................................ ............... 108
10 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy A SPATIAL TEMPORAL ANALYSIS OF VEGETATION CHANGE LAND COVER CHANGE AND HEALTH IMPACTS IN FLORIDA By Huiping Tsai December 2012 Chair: Jane Southworth Cochair: Peter R. Waylen Major: Geography The present dissertation contributes to the overall better understanding of vegetation change; land cover and land use change and health impacts in Florida, but would also have implications and significance beyond just Florida. By using the Normalized Difference Vegetation Index (NDVI) as vegetation representatio n, a time series approach is applied in order to assess vegetation dynamics across the state from 1982 2006. In addition, the response of vegetation to climate variability and land cover is investigated with remote sensing based land cover classification d ata. At last, the impacts of land cover and land use change on air quality is examined. Overall, three conclusions can be drawn from this dissertation. First, there is an increasing NDVI value during winter months from 1995 onward and this phenomenon cou ld be explained by the Atlantic Multidecadal Oscillation (AMO) switched into its warm phase around 1995. The result also corresponding with an increased winter rainfall proportion has been found from a previous study. Second, precipitation and land cover types both influence NDVI behavior. The NDVI responds to precipitation is found to be stronger in natural land cover like estuarine wetlands than in human manipulated land cover such as
11 developed land. Third, air pollution is highly correlated to weather conditions especially precipitation. Future research opportunities are plenty based on the findings from this dissertation.
12 CHAPTER 1 INTRODUCTION oints out that the global environment was changed dramatically and human beings need to face the challenges resulting from these changes. The global environment is changing due to human and natural causes (GLP 2005) and human causes have been recognized as a major contributor to global environment changes (Vitousek et al. 1997, Dale 1997, Foley et al. 2005, GLP 2005). A conceptual model of how human alterations to the Earth system operate through interacting processes is developed by Vitousek and colleagu es in 1997. They argue that the growth of human population and growth in the resource base used by humanity is maintained by a suite of human enterprises such as agriculture, industry, fishing, and international commerce (Meyer and Turner 1992). These ente rprises transform the land surface (through cropping, forestry, and urbanization), alter the major biogeochemical cycles (Kalnay and Cai 2003, Foley et al. 2005, Bala et al. 2007, Bonan 2008, Grimm et al. 2008, McCarthy et al. 2010, Lambin et al. 2003, Dav idson and Janssens 2006, Pongratz et al. 2009), and alters species and genetically distinct populations in most substantial, are well quantified, and all are ongoing. These well do cumented changes cause further alteration to the functioning of the Earth system, most notably by driving global climatic change (Dale 1997, Bounoua et al. 2002, Feddema et al. 2005, Ebi and McGregor 2008, Jacob and Winner 2009) and causing irreversible lo sses of biological diversity (Vitousek et al. 1997, Lambin et al. 2001, Foley et al. 2005, Moss et al. 2010).
13 Realizing human activities are responsible for the majority of changes on Earth across a variety of spatial and temporal scales, there is a need t o understand the interaction between humans and the environment and the way these have affected, and may yet affect, the sustainability of the Earth System (GLP 2005). Much of the international global change research is facilitated by four programmes, inc luding the International Geosphere Biosphere Programme (IGBP), the International Human Dimensions Programme on Global Environmental Change (IHDP), DIVERSITAS (an international programme of biodiversity science) and the World Climate Research Programme (WCR P). In 2005, the Global Land Project (GLP) Science Plan and Implementation Strategy was built upon the extensive heritage of the joint IGBP IHDP project on land use and land cover change (LUCC) and worked on improving the understanding of land system dynam ics in the context of Earth System functioning (GLP 2005). Numerous researchers collaborated and contributed observations, study results, experiments and modeling techniques to improve the ability to explain and predict global environmental changes (Turn er et al. 1990, Lambin et al. 2001, Tilman et al. foundational element of global environment chan ge and sustainability science. The focus of LCS requires the integration of social, natural, and geographical information sciences (Rindfuss et al. 2004). Researchers of LCS utilize environmental, human, and remote sensing/geographical information system (GIS) science to solve various questions about land use and land cover changes. They also study the impacts of these
14 changes on humankind and the environment as an integrated science (DeFries et al. 1999, Lambin et al. 2001, Bounoua et al. 2002, Gutman et al. 2004, Rindfuss et al. 2004, Southworth et al. 2006, Turner et al. 2007). Satellite based observations of the Earth have provided a spatially and temporally consistent picture of the state of global land cover and it has been used in LCS as a valuable r esource (Townshend et al. 1991, Bartholome and Belward 2005, Lambin and Geist 2006, Herold et al. 2008, Friedl et al. 2010). Integration of Geographical Information Systems (GIS) and Global Positioning Systems (GPS) with advanced remote sensing techniques improves the ability and efficiency in assessing land cover change than by remote sensing data only (Xiao et al. 2006, Shalaby and Tateishi 2007, Friedl et al. 2010). However, the challenge of thoroughly understanding the complexity of global environment changes remains (Lambin and Geist 2006). As the world population approaches 10 billion in the next ninety years (UN 2010), the intensity and extent of human alterations to the Earth system as well as their interacting processes are only expected to increas e Additionally, the Earth system responds to changes at different speeds and extents and that varies greatly through time and from place to place (Turner et al. 1990, Vitousek et al. 1997, Steffan 2004, Rockstorm et al. 2009). Thus, a more detailed inve stigation is needed across different time (short term and long term) and spatial scales (global, regional, community) to provide explicit monitoring, analyzing, and problem solving strategies, and to link changes across systems, such as changes in climate to resultant vegetation, or changes in climate on human health. As a regional scaled investigation, the southeast US has drawn more attention from researchers across diverse disciplines. Sohl and Sayler (2008) pointed out the
15 southeast US has experienced massive land use change since European settlement and continues to experience extremely high rates of forest cutting, significant urban development, and changes in agricultural land use. Based on a report from the U.S. Global Change Research Program (USGCR P 2009), the southeast US has also experienced an increase in annual average temperature by 2F since 1970. There has been a 30 percent increase in fall precipitation over most of the region since 1901 but a decrease in fall precipitation in South Florida. The decline in fall precipitation in South Florida contrasts strongly with the regional average. However, there has been an increase in heavy downpours in many parts of the region (Keim 1997, Karl and Knight 1998), while the percentage of the region exper iencing moderate to severe drought increased since the mid 1970s. Additionally, sea level rise and the likely increase in hurricane intensity and major hurricane frequency with associated storm surge pose a severe risk to the southeast US (Mulholland et al 1997, Esterling et al. 2000, Pielke et al. 2005). Furthermore, as the population of Florida more than doubled during the past three decades and it receiving over 80 million visitors every year, the overall ecosystem is under an elevated pressure caused b y its dwellers. From 2000 to 2010, there was a population coupled with an increase in societa l demand is a big contributor to agricultural development, land cover change/urbanization and air pollution in Florida (Samet et al. 2000, Solecki and Walker 2001, USGS 2004, Hu et al. 2008, Zanobetti and Schwartz 2009). However, anthropogenic activities have not only transformed the landscape of the Florida peninsula, but also altered the regional climate (Marshall et al.
16 2004, Pielke and Niyogi 2010) and potentially the sustainability of Florida ecosystems (Harwell et al. 1996, Solecki 2001). The chara cteristics of the southeast US make Florida a desirable study region. A long term spatial and temporal variability analysis is needed to provide insights into the dynamic ecological and societal system (Magnuson et al. 1991). Climatologically, Florida is experiencing changes in weather patterns and precipitation tends to be a big factor influencing the natural landscape/vegetation. Ecologically, Florida is confronting a massive land cover change with aggressive anthropogenic activities coupled with adver se environmental and health impacts like air pollution. The present dissertation research utilizes advanced remote sensing and geographical information system (GIS) techniques to analyze the changes in Florida in terms of its vegetation cover, climate vari ability, land cover and land use with associated health impacts especially on air pollution as well as the associated linkages/exchanges among these different variables. The overarching object of this present dissertation is to contribute the overall bet ter understanding of vegetation change; land cover and land use change and health impacts in Florida, but would also have implications and significance beyond just Florida. The order of main chapters is organized by three research papers. The first paper a pply the normalized difference vegetation index (NDVI) to investigate long term vegetation trend and the spatial and temporal variations of vegetation cover (as represented by NDVI) within the landscape which are influenced by different climate variables, in particular precipitation and large scale circulation patterns such as the Atlantic Multidecadal Oscillation (AMO). The second paper analyzes the responds of NDVI to climate variability and land cover and how the responds varied spatially by
17 using the wa velet analysis. The third paper combined the results from both previous papers with an emphasis on the associated air pollution impacts for developed areas in Florida.
18 CHAPTER 2 SPATIAL PERSISTENCE AND TEMPORAL TRENDS IN VEGETATION COVER ACROSS FLORIDA, 1982 2006 2.1 Background As modern advanced remote sensing techniques have been proven to provide an efficient tool to provide a spatially and temporally consistent picture of land surface conditions (Curran 1989, Gould 2000, Stow et al. 2004, Townshend e t al. 1991, Bartholome and Belward 2005, Lambin and Geist 2006, Herold et al. 2008, Friedl et al. 2010), they have been increasingly used to monitor and detect patterns and variations in vegetation worldwide(Gao 1996, Gould 2000, Wu et al. 2009, Tang et al 2010) In order to enhance our understanding of intra and inter annual variations in vegetation, time series approach of analyzing continuous Earth Observation (EO) based estimates of vegetation are being adapted by many scholars (Myneni et al. 1998; Ekl undh & Olsson 2003; Olsson et al. 2005; McCloy et al. 2005; Anyamba & Tucker 2005, Jeyaseelan et al. 2007, Helldn & Tottrup 2008, Fensholt et al. 2009, Neeti et al. 2011). The Advanced Very High Resolution Radiometer (AVHRR) on board the National Oceani c and Atmospheric Administration (NOAA) satellite series provides an excellent opportunity to employ time series approaches to vegetation research because of its consistent fine spatial and temporal coverage. Florida is well known of its diverse ecosyste ms and natural resources (Brockway et al. 1998, Gentile et al. 2001, Menges et al. 2009). Its vegetation cover is influenced by numerous factors not the least of which is the 600 percent increase in population since 1940. Land cover and land use change re sulting from global climate change/variability are big concerns in terms of its indirect and direct impacts on vegetation cover. Vegetation structure and pattern in Florida has been studied
19 intensively in several regions such as Everglades National Park (H uang et al. 2008, Todd et al. 2010, Gaiser et al. 2011), South Florida (LaRoche and Ferriter 1992, Menges et al. 2001, Ross et al. 2009); or in specific ecosystems (Sitch et al. 2003, Smith et al. 2009, Haile et al. 2010). Less recognized, however, is a de termination of the overall vegetation pattern and trend across the state. During the twentieth century, the natural landscape of the Florida peninsula was transformed extensively by agriculture, urbanization, and the diversion of surface water features (Ma rshall et al. 2004). Therefore, it is necessary to evaluate the statewide spatial and temporal trends of vegetation cover in order to enhance our understanding of a long term vegetation changes that may be as background information for future research. T his study applies advanced remote sensing techniques especially the use of Normalized Difference Vegetation Index (NDVI) derived from AVHRR in order to assess a long term time series study on vegetation cover in Florida from 1982 2006. NDVI is a sufficien t and the most commonly used vegetation index to monitor the healthiness of vegetation (Anyamba and Eastman 1996, Carlson and Ripley 1997, Li and Kafatos 2000, Gurgel and Ferreira 2003, Jarlan et al. 2005, Pettorelli et al. 2005, Barbosa et al. 2006, Phili ppon et al. 2007, Meng et al. 2011, Begue et al. 2011). Therefore, NDVI is reasonable and applicable means of representing vegetation in this study. Pickup and Foran (1987) developed a mean variance analysis by utilizing a graphical analysis of a dynamica l system to describe the motion or trajectory of states through time. Later on Zimmermann et al. (2007) and Washington Allen et al. (2003 & 2008) applied the mean variance analysis on their research. From their results, they concluded that mean variance a nalysis is an effective method to describe the seasonal and interannual
20 response of vegetation to climate and disturbance. Therefore, the mean Variance analysis is employed to describe the trajectory of the state of vegetation across the 25 year time peri od in terms of the mean NDVI, characterizing the overall amount of vegetation, and its simultaneous spatial variance, describing the spatial heterogeneity, in a two dimensional plane. The persistence analysis is applied to assess changes in NDVI from 1982 2006 in a spatially explicit manner. Based on the value of every pixel, two spatial persistence analyses are performed. One of them is based on a logic of NDVI is increase/decrease to the next time step and another one is based on calculating the absolute amount of NDVI value changes through time. Statistical tests are performed in order to characterize vegetation patterns and trends more objectively. Overall, this study proposes a unique method by which to explicitly detect vegetation cover patterns and trends both spatially and temporally. The results provide valuable information and enhance the understanding of statewide temporal changes in vegetation cover change. Additionally, the results from this study will also fed into further investigations of la nd cover change combining climate variability of Florida. 2.2 M ethods 2.2.1 AVHRR Vegetation Indices Satellite derived vegetation indices have proven to be an efficient way to characterize explicitly surface vegetation conditions (Gutman 1991 & 1999, Huete et al. 2002, Anyamba and Tucker 2005, Fensholt et al. 2009, Meng et al. 2011, Begue et al. 2011). Analysis of the seasonal and interannual vegetation dynamics and trends in Florida is based on the normalized difference vegetation index (NDVI) data derived from measurements made by the Advanced Very High Resolution Radiometer (AVHRR) sensor aboard the National Oceanic and Atmospheric Administration (NOAA) polar
21 orbiting satellite series (NOAA 7, 9, 11, 14, 16 and 17). The AVHRR sensor was originally designe d as a weather satellite. However from the early 1980s, AVHRR data have found increasing use to monitor the type and condition of land vegetation. Therefore there is a large archive of AVHRR data, providing a very rich source of information for multi tempo ral studies (Warner and Campagna 2009). NDVI is a vegetation index commonly used in remote sensing research whose signal response is a function of absorbed radiation by chlorophyll in the red band and scattering by cellulose in the near infrared (NIR) and it has been widely accepted as an indicator for providing vegetation properties and assessing ecological response to environmental changes for large scale geographic regions (Anyamba and Eastman 1996, Carlson and Ripley 1997, Li and Kafatos 2000, Gurgel an d Ferreira 2003, Jarlan et al. 2005, Pettorelli et al. 2005, Barbosa et al. 2006, Philippon et al. 2007, Meng et al. 2011, Begue et al. 2011). NDVI is a variant of simple ratio of near infrared and red band, which is defined: NDVI = (Near Infrared Red)/(Ne ar Infrared+Red). However, by constructing the ratio as the difference of the two wavelengths over the sum of the two wavelengths, NDVI is normalized, so that the range falls between 1 and +1. Furthermore, NDVI is designed such that high values of the ra tio (closer to +1) indicate abundant green vegetation, and values near zero or less indicate an absence of vegetation (Warner and Campagna 2009). The AVHRR NDVI record is calculated based on its red band (band 1 covered the 0.58 to 0.68 m radiation) and i nfrared band (band 2 covered the 0.72 to 1.10 m radiation).
22 AVHRR GIMMS (Global Inventory Modelling and Mapping Studies) NDVI dataset is available from July 1981 to December 2006 and has been used for numerous regional to global scale vegetation studies ( Los et al. 1994, Hogda et al. 2001, Tucker et al. 2005, Anyamba et al. 2005, White and Nemani 2006, Fensholt et al. 2009, Milesi et al. 2010, Raynold et al. 2012). AVHRR GIMMS NDVI dataset has been corrected for residual sensor degradation and sensor inter calibration differences; distortions caused by persistent cloud cover globally; solar zenith angle and viewing angle effects due to satellite drift; volcanic aerosols; missing data in the Northern Hemisphere during winter using interpolation due to high so lar zenith angles; and low signal to noise ratios due to sub pixel cloud contamination and water vapor. The spatial resolution is 8 km and the data record is based on 15 day composites in order to construct cloud free views of the Earth with the maximum ND VI during regularly spaced intervals. According to its intensive 25 year monthly coverage, AVHRR GIMMS NDVI dataset has been chosen in this study. The Florida state boundary covering the domain 25~30N, 79~87W was subset from the dataset for the period J anuary 1982 December 2006. Since the AVHHR GIMMS NDVI dataset is constructed based on 15 dat maximum composites, so for every month, there are two 15 day composites. For this study, the statewide higher NDVI value composite is chosen from those two 15 day composites at a monthly basis to represent the NDVI value for the month. 2.2.2 NOAA CCAP L and C over C lassification D ata Florida land cover classification data are available from the Coastal Change Analysis Program (C CAP) developed by the National Oceanic and Atmospheric Administration (NOAA) with collaborations of the Department of Commerce (DOC), National Ocean Service (NOS) and NOAA Coastal Services Center (CSC). Current
23 production of the Coastal Change Analysis Program (C CAP) land cover datasets is acc omplished through closely coordinated efforts with the U.S. Geological Survey (USGS) as it produces the National Land Cover Dataset (NLCD). C CAP data are developed, primarily, from Landsat Thematic Mapper (TM) satellite imagery. The smallest feature si ze (spatial resolution) that can be mapped is 30 meter pixels (1/4 acres) on the ground. Current C CAP datasets are available for Florida in the years 1996, 2001, and 2006 by 23 classes. In order to simplify for the following analysis, 23 classes were regr ouped into 10 classes based on their classification scheme (Table 2 1). 2.2.3 Mean Variance A nalysis In order to characterize the spatio temporal behavior of NDVI, the NDVI values derived from AVHRR GIMMS NDVI dataset are analyzed using a mean variance app roach. A mean variance analysis is developed in late 1980s by Pickup and Foran (1987) to characterize the spatiotemporal behavior of a remotely sensed vegetation index (VI) and researchers utilize it to describe the seasonal and inter annual response of ve getation to climate and disturbance at several regions across the globe (Wachington Allen et al. 2003 and 2008, Zimmermann et al. 2007). The mean can be interpreted as vegetation presence or the overall amount of vegetation within the landscape and the va riance is representative of the degree of landscape heterogeneity. Figure 2 1 shows the hypothetical relationship between mean variance and vegetation status (Washington Allen et al. 2008). Each quadrant demonstrates a relative measurement of heterogeneity (variance) and vegetation status (mean). Quadrant 1 (low mean and low variance) can be considered as the most degraded landscape because the amount of vegetation is relatively low and
24 homogeneous. Quadrant 2 (low mean and high variance) indicates that a greater proportion of the landscape tend to be bare ground and thus high susceptibility to disturbance. Quadrant 3 (high mean and low variance) shows a relatively higher vegetation cover with lower vegetation variability. Quadrant 4 (high mean and high va riance) indicates the landscape has higher vegetation cover with a higher variability of vegetation cover. The approach is employed to describe the trajectory of vegetation state across the 25 year time period in terms of the mean NDVI characterizing the o verall amount of vegetation and the simultaneous variance of NDVI describing the heterogeneity by a two dimensional plane. Three temporal scales are examined in order to depict trend and pattern if any. First, the annual total NDVI value is cumulated by summing up all NDVI value from January to December for a specific year. Second, the seasonal NDVI value is represented by two non overlapped seasons, winter (October, November, December, January, February, March) and summer (April, May, June, July, August September). Winter NDVI value is calculated by summing up all NDVI value from the defined winter months and summer NDVI value is calculated by summing up all NDVI value from the defined summer months. A special note here is winter NDVI value is named bas ed on the year of January March, for instance, winter 1983 is calculated by summing up the NDVI value from October 1982 through March 1983. Third, a monthly NDVI value is derived from the monthly composite that chosen from AVHRR GIMMS NDVI dataset for the period January 1982 to December 2006 to represent the maximum NDVI for a particular month.
25 2.2.4 Spatial P ersistence The concept of persistence analysis is to detect changes based on the value of every pixel. Total two spatial persistence analyses are perf ormed; one of them is based on a logic of NDVI is increase/decrease to the next time step and another one is based on calculating the absolute amount of NDVI value changes through time. The persistence analysis is applied to assess NDVI change from 1982 2 006 in a spatially explicit manner on a monthly basis. The spatial persistence layers characterize the direction of change and the absolute amount of change of NDVI value during successive years on a pixel basis. The direction of change of NDVI value is d etermined using the following nomenclature: Where the NDVI, t, in year I,s t_i and t_(i+1) is the value in the following year, e.g. t_i=October NDVI in 1982, then t_(i+1)=October NDVI in 1983. A value +1 is assigned to pixels which have had an increase in successive values of NDVI; value 1 is assigned to pixels which have had a decrease in NDVI, and zero indicating no change. However, theoretically, there is an infinitesimally very small chance that two successive NDVI values are truly id entical as NDVI is a continuous variable. Limiting by the AVHRR GIMMS NDVI dataset structure, NDVI value are stored and scaled from 1000 to +1000, which means there are only three decimal places if it is scaled back to the original theoretical NDVI range 1 to +1. In order to handle the occasional issue a
26 zero change, a neighborhood evaluation of GIS technique is applied. A 5*5 window is placed on pixels that returning a value of zero value from the above nomenclature, a new value either 1 or +1 is assi gned based on the majority value from the neighborhood. For example, if the majority of the neighboring pixels return values of +1, this pixel itself is assigned a new value +1 instead of the original value zero. This adjustment had to be made in order t o overcome the issue of data truncation in the AVHRR GIMMS NDVI dataset. The absolute change of NDVI is calculating by subtracting successive values. Thus if a pixel, returns an NDVI value of 0.647 in January 1982 and 0.691in January 1983 ; the persiste nce analysis for direction of change will assign this pixel a value of +1 and the analysis for absolute amount of change will yield a value of +0.044;. Retain the convention that positive values of both variables imply an increase in NDVI. There are 24 t ime steps maps for the persistence analysis procedure for each month, for a 25 year record. By summing up each of the individual maps, one persistence layer representing the cumulative direction of change and one representing the absolute amount of cha nge in NDVI are produced for each month. 2.2.5 Statistical Test f or Persistence Analysis No appropriate statistical test of significance exists for these two variables. However, if NDVI values in the absence of any trends or jumps induced by climate chang e/variability or land use change, are assumed to be normally distributed and serially independent (Independent and identically distributed, aid, random variable), compa red, and from which critical values of the persistence variables may be derived via simulation. On a purely theoretical basis, the assumption of normality runs into
27 difficulties as NDVI is a bounded variable ( 1 to +1), and because it is computed as a rat io of two negatively correlated variables, it will become increasingly negatively (positively) skewed, as the mean values of NDVI approach the upper (lower) end of the scale. However, empirical evidence drawn from pixels whose land us e categories remained constant, suggests that, the assumption of normality, for LULC types found in Florida is not totally unreasonable, and may at least provide a reasonable set of null conditions. Similarly, in the absence of any climate change/variability and LULC change the phenomena that we are trying to identify, it is also reasonable to assume serial independence from one month to the same month in the next year. 126.96.36.199 Persistence analysis of direction of change The persistence analysis of direction of changes in N DVI from 1982 to 2006 (25 years) yields a possible range of 24 to +24. Under the null hypothesis, if one specific term mean, the probability of a positive change in direction in 1990, is +1 If the Ja nuary 1990 value is exactly one standard deviation greater than the mean, then the probability of a positive step to January 1991 has now dropped to 0.16. The persistence measures used here therefore are not analogous to a standard random walk process (se e for example, Wilson and Kirkby, 1980) as the probabilities of successes/failures (positive steps/negative steps) is not fixed with every iteration or transition but is controlled by the magnitude of the preceding observation. Therefore the likelihoods o f smaller (larger) directional sums positive or negative increases (decreases) in comparison to the random walk. For the persistence analysis of direction of change in NDVI, 10,000 sequences of normally distributed data, each of 25 observations are sim ulated and analyzed using the same methodology to yield a relative frequency count of the sums of directional change
28 under conditions of the null hypothesis. A comparison of this simulated distribution with the cumulative persistence layer result is condu cted to examine which pixel yields unusually high (positive or negative) observed frequencies. As the directional variable is by definition a discrete variable, and because of the number of transitions involved, only possible to take on even numbers or ze ro, traditional significance values (e.g. 0.05) cannot be used. However the simulated distributions do allow approximations to these values. In this case, values of +/ 6, represent 4.3% of the area under the distribution in either tail, and will be used as critical values of the variable throughout. 188.8.131.52 Persistence analysis of absolute amount of change As the absolute change variable requires a magnitude term in addition to the sign of the change, the establishment of the null conditions becomes more difficult. However regardless of the observed values of mean and variance of NDVI (for instance, Developed Land, compared to Forest) each distribution could itself be reduced to a standard normal distribution. Simulations and computations can then be carr ied out and appropriate critical values determined based upon the mean and variance of the original data set. It is well known that the sum (Z) of variables drawn from two Normal distributions (X and Y) is itself normally distributed with mean (Z) = mean (X) + mean (Y), and variance (Z) = variance (X) + variance (Y). In this study, there are 25 independent observations from 1982 to 2006, the sum, Z, can be extending to 1+Xn, for n=25 with mean (n. mean(X)) and variance (n. variance(X)). Th en confidence bounds could be set up based on the standard normal distribution at any interested level such as 10% to identify extreme values in the distribution. However, as in the case of the directional variable, the definition of this variable does not match the established theory. It is the differences in NDVI from the
29 previous value that a re being summed not the values e. g NDVI themselves. Extension of the previous 10,000 simulated series allows the computation of the distribution of values of this variable under the null conditions. It was found that the sums themselves approximated to a normal distribution, with a mean of zero and a standard deviation of 1.41 (based on the standard normal distribution employed in the simulation). These properties can them be applied to the observed sum of magnitudes derived from a set of data with a specific mean and variance by multiplying the observed variance by 1.41. Under null conditions the most likely sum of changes was zero This approach presents the prob lem of identifying the expected variance, which was not important in computations for the directional change variable. This could only be done by observing typical values of variance in pixels that appeared to meet the conditions of the hull hypothesis. Guided initially by the results from the analysis of directional changes, pixels were also selected that met the following criteria: It should stay the same land cover type (1982 2006); and The land cover type should dominant at least 50% within this NDV patial resolution 8 km square. Examples, if they existed, were also to be drawn separately from each of the 6 NOAA climate divisions of the state in order to avoid establishing unrealistic null conditions resulting from spatial variations in ann ual and monthly rainfall totals across the region. Accompany with NOAA CCAP land cover classification data at three dates 1996, 2001, and 2006 (spatial resolution 30 meter square), selecting criteria for a particular pixel including: It should stay the sa me land cover type; and It can be found across the state from climate division 1 to climate division 6; and
30 resolution 8 km square. The series of NDVI values for these lan d cover/ month/climate division type are derived from the AVHRR GIMMS NDVI dataset from 1982 to 2006 and utilized for comparative purpose and to provide the basic variance measures to be used in establishing significance levels. Examples from three land co ver types; developed land, agricultural land, and palustrine wetlands were found in each of six climate divisions. 2.3 R esults 2.3.1 Mean Variance A nalysis For the annual temporal scale that have been examined (Figure 2 2), the NDVI values from January to December are summed to produce only one single mean and variance of NDVI for a specific year. A grand mean and a grand variance are marked in the plot as a vertical straight line and a horizontal straight line. For a convenient visualization purpose, 198 0s (1982 1989) are marked as purple colored circles, 1990s (1990 1999) are marked as green colored squares, and 2000s (2000 2006) are marked as yellow colored triangle. As a result, an increasing trend in mean NDVI (represent vegetation presence) tends to been seen after 1995 in quadrant 3 (higher mean, lower variance) and quadrant 4 (higher mean, higher variance) except 1999. For the seasonal temporal scale that have been examined (Figure 2 3 and Figure 2 4), the NDVI values from October to March are sum med to produce a winter total NDVI and the NDVI values from April to September are summed to produce a summer total NDVI. The same plotting scheme is applied to these seasonal temporal results. As a result, winter NDVI are showing a very clear clustered p attern after 1995 in quadrant 4
31 (higher mean, higher variance) except 1999. However, summer NDVI do not have a clear pattern and the values are more spread out in the plot. For the monthly temporal scale that have been examined (Figure 2 5), NDVI values are plotted following the same plotting scheme. The months from October to March are showing an increasing trend in mean NDVI after 1995. In general, the grand mean NDVI values (vertical straight line) are higher during the summer months (April to Septembe r) and lower during the winter months (October to March). 2.3.2 Spatial P ersistence with Statistical T est 184.108.40.206 Persistence analysis of direction of change Total twelve persistence layers are produced to represent the summation of direction of change fro m January to December 1982 2006. Only January and August are shown in Figure 2 6 (a) as examples. Green gradient colors represent positive summation values (+1 to +24) and red gradient colors indicate negative summation values ( 1 to 24). Gray colors pres ent zero summation values. Numbers in each possible classes ( 24 to +24) are extracted for each month and served as observed value and then compare to the simulated normally distributed data mentioned before (Figure 2 7). The comparison results are plotte d in F igure 2 8. For each plot the black dashed line represents the cumulated frequencies for the simulated normally distributed data (null hypothesis), and the black dots with connecting red line represents the cumulated frequencies for the observed dat a from the direction of change persistence results for each month. If the null hypothesis (black dashed line) describes the actual process then the black dots with connecting red line should fall right along the clack dashed line. As
32 a result, the null h ypothesis could not be rejected in March and May and it looks dubious in November. In addition, the plots are also representing a degree of upward/downward trend form the data. If the red line plots up above the black dashed line (like April), there are probably more negative classes than one would expect (downward trend in NDVI). If the red line plots below the black dashed line (like November) then there are more positive classes than expected (upward trend in NDVI). This upward trend in NDVI can also been seen in the months September, October, November, January and February. Figure 2 9 are examples of areas that are identified above or below the cumulative direction of critical change classes +6 and 6 in January and August respectively. The areas th at have been marked should be interpreted as areas that experience a significant increasing (green gradient) or a decreasing trend (red gradient) in NDVI from 1982 2006. A further investigation of why these areas experience either increased or decreased tr ends of NDVI is needed. In general, these may be areas in which climate classification trajectory may have been altered dramatically by human activity. 2.3.2. 2 Persistence an alysis of absolute amount of change Total twelve persistence layers are produced to represent the summation of absolute amount of change from January to December 1982 2006. Only January and August are shown in Figure 2 6 ( b ) as examples. Green gradient col ors represent positive summation values (above zero) and red gradient colors indicate negative summation values (below zero). A confidence bound has been set up at 10% (5% at either tail) by climate division and by month in order to identify areas that may experience an increased or decreased
33 trend in terms of the absolute amount of change in NDVI value for three land cover type (developed land, agricultural land, and palustrine wetlands). Figure 2 10 represents the confidence bounds for developed land. Places that below 5% is represented in red color while upper 95% is represented in green color. For each month, three dates NOAA CCAP land cover classification data are examined by these three land cover type. Additionally, a series maps are produced in or der to show the distribution of percentage of either tails (lower 5 and upper 95) by climate divisions (Figure 2 11 Figure 2 12 and Figure 2 13) for three land cover classification dates. The benchmarks which is the null hypothesis of totally random (stat ionary) changes is true, then 5% of the pixels in each division are expected to fall into these two categories (lower 5 and upper 95) purely by chance. On the maps, the colored circles represent the actual percentage of pixels in that division/month that f ell into each category. Green color is assigned to the lower 5 category while red color indicates the upper 95 category. Each vertical arrangement of circles represents the time of land cover classification data are utilized (1996, 2001 and 2006). The pr oportional circles have a radius equal to the square root of the observed percentage and powered down proportionate to the radius used to represent 5%. In this way, the size of the circles is actually displaying the percentage. Since developed land does n ot have difference between three land cover classification dates in terms of the percentages in either tails, only one date data are present here (Figure 2 11). Agricultural land and palustrine wetlands are both presented for all three dates (Figure 2 12 and Figure 2 13).
34 Generally speaking, despite of different land cover types, all the winter months (October to March) especially October and November are showing larger green circles than in the summer months (April to September). This result imply a green er trend occurs during winter months and it matches with the mean variance analysis results which also reveal that the mean NDVI is increasing during winter months after 1995. 2.4 D iscussion and Summary The results from mean variance analysis and persisten ce analysis (both direction of change and absolute amount of change) have shown a consistent pattern from 1982 2006 in vegetation trend of Florida. The consistent pattern of vegetation trend is illustrated by overall higher NDVI values are observed durin g the winter months, October to March. The results are coincident with Waylen and Qiu (2008) unpublished work. Waylen and Qiu conduct their work on winter rains to annual precipitation totals in Florida based on selected stream flow data from 1979 2000. The conclusion from their work indicates that winter rainfall proportion to annual precipitation totals in Florida is increasing during the studied period. A possible physical explanation of this found pattern can be drawn based on the Atlantic multidecada l oscillation (AMO). The AMO is an ongoing series of long duration changes in the sea surface temperature of the North Atlantic Ocean, with cool and warm phases that may last for 20 40 years at a time and a difference of about 1F between extremes. These c hanges are natural and have been occurring for at least the last 1,000 years (NOAA, 2011). According to Enfield et al. (2001), During AMO droughts in the 1930s and 1950s. Between AMO warm and cool phases, the inflow t o Lake Okeechobe varied by 40% They concluded that the geographical pattern of
35 variability is influenced mainly by changes in summer rainfall. The winter patterns of interannual rainfall variability associate d with ENSO are also significantly cha nged between AMO phases According to the monitoring data from National Oceanic and Atmospheric Administration (NOAA), there are many impacts of the AMO to air temperatures and rainfall in Florida. First of all, rain fall in central and south Florida becomes more plentiful when the Atlantic is in its warm phase and droughts and wildfires are more frequent in the cool phase. As a result of these variations, the inflow to Lake Okeechobee changes by 40% between AMO extrem es In northern Florida the relationship begins to reverse less rainfall when the Atlantic is warm. Secondly, during warm phases of the AMO, the numbers of tropical storms that mature into severe hurricanes is much greater than during cool phases, at le ast twice as many. Since the AMO switched to its warm phase around 1995, severe hurricanes have become much more frequent and this has led to a crisis in the insurance industry. Compare this to the results from mean variance and persistence analysis of ND VI, it is reasonable to assume that the increasing vegetation trend during winter months is driven by the increased rainfall while AMO is switched to warm phase after 1995. However, rainfall variability may not be the only driver of vegetation trend. Flor ida is experiencing a massive land cover change and anthropogenic influences because its population is increasing by 600 percent since 1940 (Schmidt et al. 2001). With the growth in population and increased societal demands during the twentieth century, th e natural landscape of the Florida peninsula was transformed extensively by agriculture, urbanization, and the diversion of surface water features (Marshall et al. 2004).
36 Therefore, human induced land cover change is another important factor while examinin g vegetation trend over time and space. dynamic is more complex and rich for scientific research. A further investigation of climate variability with land cover change anal ysis is needed in order to gain more deep understanding of the drivers of vegetation dynamics in Florida.
37 Table 2 1. CCAP Land Cover Classification Original Regroup Value Class Value Class 0 Background 1 Unclassified 2 Developed, High Intensity 1 Developed Land 3 Developed, Medium Intensity 4 Developed, Low Intensity 5 Developed, Open Space 6 Cultivated Crops 2 Agricultural Land 7 Pasture/Hay 8 Grassland/Herbaceous 3 Grassland 9 Deciduous Forest 4 Forest Land 10 Ev ergreen Forest 11 Mixed Forest 12 Scrub/Shrub 5 Scrub Land 13 Palustrine Forested Wetland 7 Palustrine Wetlands 14 Palustrine Scrub/Shrub Wetland 15 Palustrine Emergent Wetland 16 Estuarine Forested Wetland 8 Estuarine Wetlands 17 Estaurine Scrub/Shrub Wetland 18 Estaurine Emergent Wetland 19 Unconsolidated Shore 9 Barren Land 20 Bare Land 6 Barren Land 21 Open Water 10 Water and Submerged Lands 22 Palustrine Aquatic Bed 23 Estuarine Aquatic Bed
38 Figure 2 1. Hypothetical relationship between mean variance and vegetation status, [Adapted from Washington Allen, R. A. Ramsey, R., West, N. E. and B. E. Norton, B.E. (2008) Quantification of the ecological resilience of drylands using digital remote sensing. Ecology and Society 13, pp. 33. ] Figure 2 2 Mean Variance plot for annual NDVI
39 Figure 2 3 Mean Variance plot for winter NDVI Figure 2 4 Mean Variance plot for summer NDVI
40 a) b) c) d) e) f) Figure 2 5 Mea n Variance plot for monthly NDVI a) to l ) represents January to December
41 g) h) i) j) k) l) Figure 2 5 Continued
42 a) b) Figure 2 6 Example maps for persistence analysis of a) direction of change of NDVI.in January from 1982 to 2006 and b ) absolute amount of change of NDVI in January from 1982 to 2006
43 a) b) c) Figure 2 7. Comparison of observed value from persiste nce analysis direction of change with simulated normal distribution result a) January to April b) May to August and c) September to December
44 a) b) c) Figure 2 8. Cumulative frequency plots for comparison of observed value from persistence analysis direction of change with simulated normal distribution a) January to April b) May to August and c) September to December
45 Figure 2 9 Example of a reas that identified by above or below critic classes
46 a) b) c) Figure 2 10. Example for developed land confidence bounds maps in January. a) 1996 land cover classification b) 2001 land cover classification and c) 2006 land cover classificatio n
47 a) b) c) d) e) f) g) h) i) j) k) l) Figure 2 11. Developed land confidence category percentage maps a) to l) represents January to December
48 a) b) c) d) e) f) g) h) i) j) k) l) Figure 2 12. Agricultural land con fidence category percentage maps a) to l) represents January to December
49 a) b) c) d) e) f) g) h) i) j) k) l) Figure 2 13. Palustrine wetlands confidence category percentage maps a) to l) represents January to December
50 CHAPTER 3 NDVI, CLIMATE VARIAB ILITY AND LAND COVER IN FLORIDA 3.1 Background As a regional scaled investigation, the southeast US has drawn more attention from researchers across diverse disciplines. Sohl and Sayler (2008) pointed out the southeast US has experi enced massive land use change since European settlement and continues to experience extremely high rates of forest cutting, significant urban development, and changes in agricultural land use. The subtropical and tropical climate makes Florida an ideal pla ce for agriculture, tourism, water recreation, industry, etc. On the other hand, Florida appears to be a particular vulnerable environment subject to major changes due to climate variability and anthropogenic influences. Much evidence has been presented regarding climate variability and anthropogenic influences Schmidt et al. 2001, Enfield et al. 2001, Cronin 2002), however, a broad picture of the ation coupled with climate variability still remains unclear. The potential impacts of climate variability such as the El Nino Southern Oscillation (ENSO) phenomena in Florida have been evaluated by many researchers. From an agricultural aspect, Hansen et al. (1998) found that during the winter season (months) in Florida, quarterly yields, prices, production, and value for crops such as tomato, bell pepper, sweet corn, and snap bean are related to ENSO phase and its relationship to rainfall, temperature, a nd solar radiation. From a more hydrological rainfall and river discharge from 1950 98 and report that Florida does not respond as a
51 uniform region to ENSO, particularly wi th respect to precipitation in the Panhandle and the southernmost areas of Florida. Additionally, anthropogenic influences in Florida have been amplified since 1940 due to a 600 percent increase in population (Schmidt et al. 2001). From 2000 to 2010, thes growth in population and increased societal demands during the twentieth century, the natural landscape of the Florida peninsula was transformed extensively by agriculture, urbanization, and the diversion of surface water features (Marshall et al. 2004). According to the National Resources Inventory (NRI) from the USDA Natural Resources Con servation Service (2001), the largest increases in US developed areas between 1982 and 1997 were in the south and Florida was one of the top three states with the largest average annual additions of developed area (Alig et al. 2004). Furthermore, anthrop ogenic activities have not only transformed the landscape of the Florida peninsula, but also altered the regional climate (Marshall et al. 2004, Pielke al. 1996, Soleck i 2001). They concluded that there was a 9% decrease in rainfall averaged over south Florida with the 1973 landscape and an 11% decrease with the 1993 landscape, as compared with the model results when the 1900 landscape is used. Marshall et al. (2004) pre sent a numerical model to study the possible impacts of land cover change on the warm season climate. Their results are in reasonable agreement with an analysis of observational data that indicates decreasing regional precipitation and increasing daytime m aximum temperatures during the twentieth century.
52 Although the relationship between NDVI variability and climate variability has been examined at varied spatial and temporal scales globally (Nicholson et al. 1990, Farrar et al. 1994, Richard and Poccard 19 98, Lotsch et al. 2003, Fensholt et al. 2004) and the relationship between precipitation and NDVI has been well established. For Florida, there still missing a statewide analyze of NDVI with climate variability. Thus, coupling the climate variability an d anthropogenic influences, we ask, what are the changes in variability? And what are the dominant human induced land cover conversions Impacting the state? In order to get more insight of the vegetation response to climate variability and land cover and land use, this study utilizes NDVI derived from Global Inventory Modeling and Mapping Studies (GIMMS) group as vegetation presence indicator and applies a time series approach by performing the wavelet analysis in order to study the how NDVI respond to precipitation across different locations and different land cover type. 3.2 Study A rea The state of Florida (25~30N, 79~87W) is located in the southeastern USA (Figure 3 1) with a geographical area of approximately 1,398,600 square kilometers and population densities of around 135 persons per square kilometer in 2010 (U.S. Census Bureau 2011). Climate to Floridians is undisputed as being very important from its well known officia distribution, prevailing winds, storms, pressure systems and ocean currents. Most of the
53 S subtropical climate zone, noted for its long hot and humid summers and mild and wet winters. The southernmost portion of the State is generally designed as belonging to the tr opical savanna region, which is sometimes called the wet and dry tropics (NCDC 2011). The annual average precipitation is approximately 1371.6 mm with high peak during the summer time (Figure 3 2, NCDC 2011). The panhandle and southeastern Florida are the wettest parts of the state (Figure 3 3, Fernald and Patton 1984). The driest portions are the Florida Keys and the offshore bar of Cape Canaveral. The principle precipitation generating mechanisms operating in Florida are varied across space and time. The panhandle receives rainfall during winter when the fronts pass through and during summer when convective rain falls. Frontal influence is reducing southward for the state (Jordan 1984, Henry 1994). Additionally, the position and intensity of Azores Bermud a High Pressure system also exerts a powerful influence on normally begins in the southeastern part in late April and the moves northward. The fall dry season begins in North Florida in September, and spreads southward, arriving in extreme South Florida in mid November. Moreover, tropical storms play an important role in the summer rain fall and it can also postpone the arrival of the dry season. On average, the hurricane season reaches its peak in September, and the length of coastline of Florida makes it more prone to hurricane impacts and landfalls than other states.
54 According to the N ational Climate Data Center, Florida is divided into seven climate divisions: Northwest, North, North Central, South Central, Everglades and Southwest Coast, Lover East Coast, and Keys (Figure 3 4). The present study will exclude the Keys because its relat ively small land mass. Spatial variations are expected to be seen in each climate division. The influence of frontal is stronger up north and weaker when it moves southward. As a result, climate division 1 and 2 are influence by frontal activities the mos t and division 5 and 6 may not get a pronounced effect from frontal activities. Depending on the location, the influences of the low frequency climate phenomena, such as the ENSO and the Atlantic Multidecadal Oscillation (AMO), Quasi Biennial Oscillation (QBO), North Atlantic Oscillation (NAO), Pacific North America pattern (PNA), Pacific Decadal Oscillation (PDO), Madden Julian Oscillation (MJO) have been identified with aggregate annual or seasonal rainfall variations in Florida (Enfield et al. 2001, Kwo n et al. 2009). Among those large scale atmospheric extensively (Hansen et al. 1998, Schmidt et al. 2002, Gubler et al. 2001, Cronin et al. 2002). ENSO is a physical pheno menon that occurs in the equatorial Pacific Ocean where the water temperature oscillates between being unusually warm (El Nino) and unusually cold (La Nina). These two oceanic events shift the position of the jet streams across the North America continent, which act to steer the fronts and weather systems. During El Nino, it typically brings 30 to 40 percent more rainfall and cooler temperatures to Florida in the winter, while La Nina brings a warmer and much drier than normal winter and spring. According t o that, La Nina is frequently a trigger to periodic drought in
55 Florida (NCDC 2011). Table 3 1 is the summary of El Nino and La Nina impacts for the southeast part of the US. 3. 3 Methods 3.3.1 Remote Sensing Data NDVI is calculated from the visible red wave band (RED) and near infrared (NIR) waveband reflected by vegetation as equation below (Eidenshink 1992). NDVI = (NIR RED)/(NIR+RED) This relatively simple algorithm produces output values in the range of 1.0 to 1.0. Increasing positive NDVI values indicat e increasing amounts of healthy green vegetation. NDVI values near zero and decreasing negative values indicate non vegetated features such as barren surfaces (rock and soil) and water, snow, ice, and clouds (USGS 2010). The Global Inventory Monitoring and Modeling System (GIMMS) group Normalized Difference Vegetation Index (NDVI) dataset was used in this study. The GIMMS dataset is a NDVI product available for a 25 year period spanning from 1981 to 2006 with its spatial resolution 8 km. The dataset is deri ved from imagery obtained from the Advanced Very High Resolution Radiometer (AVHRR) instrument onboard the NOAA satellite series 7, 9, 11, 14, 16 and 17. The GIMMS data set has been corrected for calibration, view geometry, volcanic aerosols, and other eff ects not related to vegetation change. The GIMMS data set is composited at a 15 day time step. For each month, the first composite is the maximum value composite from the first 15 days of the month and the second is from days 16 through the end of the mont h. For each month, the highest NDVI value composite is chosen as the NDVI for the month because it represents the maximum NDVI value for the month.
56 The time series of NDVI value are extracted for Florida from July 198 1 to December 2006 and processed in ERD AS 2011 and ESRI ArcGIS 10 software. 3.3.2 Precipitation Data Precipitation data were gathered from the PRISM Climate Group, Oregon State University. PRISM stands for parameter elevation regression on independent slopes model, which is a climate mapping sy stem developed by Dr. Christopher Daly, PRISM climate group director. The PRISM model allows for the incorporation of expert knowledge about the climate and can be particularly useful when data points are sparse. With this method, one can explicitly accoun t for the effects of coastal influences, terrain barriers, temperature inversions, and other factors on spatial climatic patterns (Daly et al. 2002). PRISM data sets are recognized world wide as some of the highest quality spatial data sets currently avail able (Hijmans et al. 2005, Hamann and Wang climatological data. All monthly precipitation data were downloaded from the PRISM website for the study area from 1981 to 2006 for the state of Florida (Figure 3 7 ). 3.3.3 Climatic D ivisions The climate divisions represent regions within states that are considered to be climatically homogeneous (Karl and Riebsame 1984). Although extreme climate variations can occur in areas of comple x terrain, such as mountainous areas (Karl and According to the National Climate Data Center, Florida is divided into seven climate divisions: Northwest, North, North Central, South Central, Everglades and Southwest Coast, Lower East Coast, and Keys (Fig ure 3 convenience, Northwest division will be denoted as division 1; North division will be
57 denoted as division 2; North Central will be denoted as d ivision 3; South Central will be denoted as division 4; Everglades and Southwest Coast will be denoted as division 5; Lower East Coast will be denoted as division 6 and Keys will be excluded because it has relatively small land mass and vegetated area (Fig ure 3 4). Time series of NDVI value and precipitation data by each of the climate divisions are created for further analysis (Figure 3 7). 3.3.4 Land Cover Data Florida land cover classification data are available from the Coastal Change Analysis Program ( C CAP) developed by the National Oceanic and Atmospheric Administration (NOAA) with collaborations of the Department of Commerce (DOC), National Ocean Service (NOS) and NOAA Coastal Services Center (CSC). Current production of the Coastal Change Analysis P rogram (C CAP) land cover datasets is accomplished through closely coordinated efforts with the U.S. Geological Survey (USGS) as it produces the National Land Cover Dataset (NLCD). CCAP land cover classification data are developed, primarily, from Lands at Thematic Mapper (TM) satellite imagery. The smallest feature size (spatial resolution) that can be mapped is 30 meter pixels (1/4 acres) on the ground. Current C CAP datasets are available for Florida in the years 1996, 2001, and 2006 by 23 classes (Fig ure 3 5) In order to simplify for the following analysis, 23 classes were regrouped into 10 classes based on their classification scheme (Table 2 1 ). CCAP land cover classification data for the years 1996, 2001, and 2006 were analyzed under a superimpo sed GIMMS NDVI pixel grid, which is an 8km squared grid. Then for each grid, the proportions of each land cover type (regrouped 10 classes) are recorded. Practically, there is no single one grid that only has solo land cover type.
58 According to that and in order to find a proper NDVI time series to represent a typical land cover type, a determine process was proposed as follows. First of all, the land cover type need to be consistent from 1996 through 2006 (land cover classification stay the same). Second, we determined the land cover type occupied more than 70 percent of a grid can represent a typical behavior of NDVI for this specific land cover type. The determine process just as the climate zones are thought to exhibit patterns representative of the cli mate region it is found in. As a result, seven land cover types were selected as they met the criteria (Figure 3 6). These seven land cover types are developed land, agricultural land, forest land, scrub land, palustrine wetlands, estuarine wetlands and w ater and submerged lands. The detailed classification scheme can be found in Table 3 2 Time series of NDVI value and precipitation data by eac h of the seven land cover types are created for further analysis (Figure 3 7). 3.3.5 Time Series Approach a nd Wa velets The decomposition of time series into time frequency space permits not only the identification of the dominant modes of variability, but also the determination of how these modes vary in time. This can be done by using either windowed Fourier transf orm or wavelet transform (Coulibaly 2006). Fourier transform has traditionally been used to analyze relationship between oscillating time series and decomposes time series into their different periodic components (Klvana et al. 2004). This method was init ially developed for the analysis of physical phenomena but is not always appropriate when dealing with complex biological and climatic time series (Chatfield 1989). A major shortcoming of standard Fourier transform is that it does not provide an accurate time frequency localization of dynamical processes (Coulibaly 2006).
59 Figure 3 8 illustrates the main difference of time frequency windows usage between Fourier transform (FT), windowed Fourier transform (WFT) and wavelet transform (WT). In some way, WT is a generalized form of FT and WFT (Gabor 1946). The Fourier transform uses sine and cosine base functions that have infinite span and are globally uniform in time. For a stationary time series with a pure sine wave signal, its FT is a line spectrum (Figure 3 8, left panel). FT does not contain any time dependence of the signal and therefore cannot provide any local information regarding the time evolution of its spectral characteristics. In a WFT, a time series is examined under fixed time frequency window with constant intervals in the time and frequency domains. The middle panel in Figure 3 8 shows when a wide range of frequencies is involved, the fixed time window of the WFT tends to contain a large number of high frequency cycles and a few low frequency cycles or parts of cycles. One big disadvantage of WFT is it often results in an overrepresentation of high frequency components and underrepresentation of the low frequency components. A WT uses generalized local base functions (wavelets) that can stretch ed and translate with a flexible resolution in both frequency and time (Figure 3 8, right panel). As a result, high precision in time localization in the high frequency band can be achieved at the expense of reduced frequency resolution (Lau and Weng 1995) Wavelet analysis is notably free from the assumption of stationary and offers several advantages (Daubechies 1990, Lau and Weng 1995, Torrence and Campo 1998, Cazelles et al. 2008, Martinez and Gilbert 2009). It overcomes the problems of non stationar y in time series by performing a local time scale decomposition of the signal, i.e., the estimation of its spectral characteristics as a function of time (Lau et al.
60 1995, Torrence and Campo 1998). Through this approach one can track how the different sca les related to the periodic components of the signal change over time and represent time series into a finer scale time domain without a window with arbitrary limited length (Coulibaly 2006). Detailed reviews of wavelet analysis can be found from Daubechie s 1990, Farge (1992), Meyers et al. (1993), Weng and Lau 1994, Lau and Weng (1995), an Torrence and Compo (1998). Many softwares such as Interactive Wavelets, MATLAB TimeStat, and R all have wavelets related packages. This present study adapted the MAT LAB codes developed by Drs. Christopher Torrence and Gilbert P. Compo (Torrence and Compo 1998) to perform wavelet analysis in M ATLAB environment. 3. 4 Results A time series of NDVI and precipitation data is analyzed for the whole state of Florida (Figure 3 9 and Figure 3 12). This is done to provide information as a controlled baseline. Upon this controlled baseline, it would be easier to detect the changes and fluctuations at different scales climate divisions and land cover types. Furthermore, a more informative analysis from comparisons among different spatial scales provides insight of ecosystems functions. Figure 3 9 represent the wavelet analysis of NDVI for the whole state of Florida. The upper panel is the time series data. The middle left panel present the wavelet power spectrum in a contour map. Areas below the curved black line indicate the region of cone of influence (COI), where zero padding has reduced the variance. Which means the areas below the curved black line should not be considered into analysis. The bold black contour line is the 10% significance level using a red noise (autoregressive lag1) background spectrum. In the global wavelet spectrum (lower
61 panel on the right), blue line represent the overall wavelet power at each period and the dashed line is the significance for the global wavelet spectrum assuming the same significance level and background spectrum. Any blue line come cross the dashed line indicate the power at the specific period is significant. Generally speaking, r eading the results from the wavelet analysis need to pay more attention to two parts. First, is the significant period (above the dashed line in global wavelet spectrum); second, is the time of this significant period in the power spectrum (bold black cont our line in wavelet power spectrum). For all the analysis, the significant period are extracted and present in a tabulation form with value 1 represents significant. The timing of significant are also present in a tabulation form with value 0 represents n ot significant and value 1 represent significant. Form the wavelet analysis, this study propose a hypothesis that if two time series data have identical significant period, it is reasonable to assume that these two time series data have similar periodic pa ttern/cycle and can be a driven relationship between these two. For example, if the wavelet analysis of NDVI in climate division 3 is having identical significant period pattern/cycle with the rainfall in climate division 3, it is demonstrating that the r ainfall in climate division 3 is an important driver for the NDVI. However, the timing of significant period/cycle needs to be examined between NDVI and rainfall time series as well in order to identify a more clear relationship in time. 3.4.1 Climate Div isions The results from wavelet analysis for climate divisions are shown in Figure 3 10 and Figure 3 13. The significant period and the timing of significant period in the power spectrum are shown in Table 3 3 and Table 3 4 From Table 3 4 climate divisi on three and four are showing identical significant periods of NDVI and rainfall (2.46 month, 5.84
62 month, and 11.69 month). For a convenience expression purpose, 2.46 month can be considered as 3 month; 5.84 month can be considered as 6 month and 11.69 mon th can be considered as 12 month. Generally speaking, a 2.5 3 month (2.46 2.92 month) period pattern/cycle is seen in rainfall across all climate divisions. A 6 month (5.84 month) period pattern/cycle appear in rainfall from division 1 to 5 but not seen in division 6. An annual 12 month (11.69 month) period pattern/cycle appears in divisions 2 to 6. There are spatial variation exist in NDVI across climate divisions. The NDVI in climate division 1 and 2 are identical. Additionally, the NDVI in climate divis ion 3 and 4 are also identical. However, NDVI in division 5 and 6 are not similar. 3.4.2 CCAP Land Cover Types The results from wavelet analysis for land cover types are shown in Figure 3 11 and Figure 3 14. The significant period and the timing of signifi cant period in the power spectrum are shown in Table 3 3 From Table 3 3 only estuarine wetlands are showing identical significant periods of NDVI and rainfall (2.46 month, 4.13 month, and 11.69 month). This can be intepret that the NDVI in estuarine wet lands are mainly driven by its rainfall. Agricultural land, forest land, scrub land and palustrine wetlands are showing significant periodic pattern/cycle at about 6 month (5.84 month) for both NDVI and rainfall. Almost all land cover types shown an annua l 12 month (11.69 month) period pattern/cycle. A 2.5 3 month (2.46 2.92 month) cycle in rainfall can be seen at all land cover types. 3. 5 Discussion Many researchers have pointed out precipitation is a primary control on vegetation dynamics in many tropica l and subtropical biomes (Nicholson et al. 1990, Li and Kafatos
63 2000, Wang et al. 2001, Ichii et al 2002, Gurgel and Ferreira 2003, Lotsch et al. 2003, Jarlan 2005, Phillippon et al. 2007, Neeti et al. 2012). Climate induced disturbances in both the freque ncy and timing of precipitation result in observable ecosystem responses (Knapp and Smith 2001). However, the variability in precipitation regimes at seasonal and longer time scales strongly influences ecosystem dynamics and is varied by location. Additio nally, many researchers found out other than precipitation, land cover type is also responsible for vegetation dynamics. For example, Yang et al. (1997) use 1 km multitemporal AVHRR derived NDVI data to examine the eco climatological relations in Nebraska U.S.A. from 1990 1991.They conclude that NDVI precipitation and NDVI potential evapotranspiration relations exhibited time lags, although the length of lag varied with land cover type, precipitation, and soil hydrologic properties. Moreover, they find th at NDVI response to precipitation was stronger in natural grasslands and grassland/wet meadows than in areas of irrigated cropland and mixed crop/ grass. Florida is experiencing a massive land cover alteration due to its increasing population and climate variability has been a big concern raised by global climate change. Therefore, the main goal of this study is to investigate the vegetation dynamics with precipitation and land cover variations in Florida by performing the wavelet analysis. From Figure 3 9 and Figure 3 and precipitation (controlled baseline), it is clearly to see a strong activity present at 6 month (5.84 month) and 12 month (11.69 month) cycle for both. However, a 2.5month (2.46 month) c ycle can only be found in rainfall. However, the pattern and timing of
64 black contour line (10% significance level) does not match with each other. This can be explained by two reasons. First, since Florida has a big north south variation of precipitation pattern, mixing precipitation information may result in an averaged presentation in the power spectrum. Secondly, the NDVI precipitation relationship is not a perfect linear trend, more information such as soil moisture, land cover types, temperature need to be considered. For climate divisions, the analysis is done based on averaging the NDVI and rainfall within each division. As a result, spatial variations can be seen from its significant periodic pattern/cycle (Table 3 4 ) and especially pronounced on r ainfall pattern. The rainfall pattern in climate division 1 is explainable by its expected influences by frontal activities during winter time and convective rainfall during summer time. However, the NDVI in climate division 1 are not having exactly ident ical significant period with rainfall. Non identical situation also happen in climate division 5 and 6 which suggest that the rainfall may not be the only driver that influence the performance of NDVI. According to this climate division subgroup are mixi ng all the land cover types within climate divisions, it is reasonable to discover a mixing signal of NDVI from a diverse land cover types. Thus, the response from NDVI to precipitation is more complicated. For different land cover types, only estuarine wetlands have identical significant periodic/cycle between NDVI and rain. Theoretically, it is reasonable to assume that if land cover type is not manipulated by human and maintain its more natural state, the relationship between NDVI and rain should be di rectly related and showing identical periodic pattern/cycles from wavelet analysis results. For instance, in estuarine
65 wetlands (Figure 3 11 (e) ), the NDVI pattern and rainfall pattern on its global wavelet spectrum are very similar in terms of the shapes Additionally, their significant period results are also showing identical cycles as well (Table 3 3 ). If land cover type is manipulated by human, the relationship between NDVI and rain may not be found directly related, as a result, the periodic patter ns/cycles are expected to be found different between NDVI and rain. For example, in agricultural land (Figure 3 11 (b) ), a very strong 12 month cycle (11.69 month) is a strong evidence of human manipulation. In developed land, the NDVI patterns are very scattered and it suggest that even within one single land cover type, spatial variation also plays a role. This part of analysis is based on the same land cover group, however, the same land cover can be found from different location. Thus, the spatial va riation is not been reflected. 3. 6 Summary and Finding The NDVI precipitation relationship has not previously been investigated explicitly in Florida. The present study applies wavelet analysis to monthly NDVI and precipitation data from July 1981 to Decem ber 2006 and links the inter annual and intra annual variability of precipitation to the NDVI responses accordingly. Additionally, considering anthropogenic influence is a big factor contributing to ecosystem dynamics, different land cover types are analy zed to examine the response in terms of NDVI and precipitation variability. The wavelet analysis appears to be an efficient technique to depict periodic patterns/cycles from a time series data. The results from this study provide strong evidences that NDV I and precipitation are related in terms of its period patterns. However, spatial variations are present from the results. Additionally, land cove types
66 play an important role of NDVI pattern. We found out that NDVI response to precipitation was stronger in estuarine wetlands than in areas of agricultural land. The present study present a novel time series approach to investigate the relationships between precipitation and NDVI variability with explicit consideration of land cover types. We conclude that spatial variations exist on NDVI and precipitation in Florida and they both have strong influences on the behavior of NDVI
67 Table 3 1. El Nio/La Nia Impacts across the Southeast U.S. (Source: Florida Climate Center 2011) Phase Region Oct Dec Jan Mar Ap r Jun Jul Sep El Nino Peninsular Florida Wet & cool Very Wet & cool Slightly dry Slightly dry to no impact Tri State Region Wet Wet Slightly wet No impact Western Panhandle No impact Wet Slightly Dry No impact Central and North Ala. & Ga. No impact No impact No impact Slightly Dry La Nia Peninsular Florida Dry & slightly warm Very dry & warm Slightly wet Slightly cool Tri State Region Slightly dry Dry Dry No impact Western Panhandle Slightly dry Dry Dry No impact Central and North Ala. & Ga. Dry Dry in the south, wet in NW Ala. No impact Wet in NW Ala. Neutral All Regions No impact No impact No impact No impact
68 Table 3 2 S elected C CAP Land Cover Classification Scheme (Source: NOAA 2011) Class name Description Developed Land Developed, High Intensity contains significant land area is covered by concrete, asphalt, and other constructed materials. Vegetation, if present, occupies < 20 percent of the landscape. Constructed materials account for 80 to 100 percent of the total cover. This cl ass includes heavily built up urban centers and large constructed surfaces in suburban and rural areas with a variety of land uses. Developed, Medium Intensity contains areas with a mixture of constructed materials and vegetation or other cover. Cons tructed materials account for 50 to 79 percent of total area. This class commonly includes multi and single family housing areas, especially in suburban neighborhoods, but may include all types of land use. Developed, Low Intensity contains areas wi th a mixture of constructed materials and substantial amounts of vegetation or other cover. Constructed materials account for 21 to 49 percent of total area. This subclass commonly includes single family housing areas, especially in rural neighborhoods, bu t may include all types of land use. Developed, Open Space contains areas with a mixture of some constructed materials, but mostly managed grasses or low lying vegetation planted in developed areas for recreation, erosion control, or aesthetic purpos es. These areas are maintained by human activity such as fertilization and irrigation, are distinguished by enhanced biomass productivity, and can be recognized through vegetative indices based on spectral characteristics. Constructed surfaces account for less than 20 percent of total land cover. Agricultural Land Cultivated Crops contains areas intensely managed for the production of annual crops. Crop vegetation accounts for greater than 20 percent of total vegetation. This class also includes all la nd being actively tilled. Pasture/Hay contains areas of grasses, legumes, or grass legume mixtures planted for livestock grazing or the production of seed or hay crops, typically on a perennial cycle and not tilled. Pasture/hay vegetation accounts for gr eater than 20 percent of total vegetation.
69 Table 3 2 Continued. Class name Description Forest Land Deciduous Forest contains areas dominated by trees generally greater than 5 meters tall and greater than 20 percent of total vegetation cover. More tha n 75 percent of the tree species shed foliage simultaneously in response to seasonal change. Evergreen Forest contains areas dominated by trees generally greater than 5 meters tall and greater than 20 percent of total vegetation cover. More than 75 p ercent of the tree species maintain their leaves all year. Canopy is never without green foliage. Mixed Forest contains areas dominated by trees generally greater than 5 meters tall, and greater than 20 percent of total vegetation cover. Neither deci duous nor evergreen species are greater than 75 percent of total tree cover. Both coniferous and broad leaved evergreens are included in this category. Scrub Land Scrub/Shrub contains areas dominated by shrubs less than 5 meters tall with shrub canopy typically greater than 20 percent of total vegetation. This class includes tree shrubs, young trees in an early successional stage, or trees stunted from environmental conditions. Palustrine Wetlands Palustrine Forested Wetland includes tidal and non tidal wetlands dominated by woody vegetation greater than or equal to 5 meters in height, and all such wetlands that occur in tidal areas in which salinity due to ocean derived salts is below 0.5 percent. Total vegetation coverage is greater than 20 percen t. Palustrine Scrub/Shrub Wetland includes tidal and non tidal wetlands dominated by woody vegetation less than 5 meters in height, and all such wetlands that occur in tidal areas in which salinity due to ocean derived salts is below 0.5 percent. Tot al vegetation coverage is greater than 20 percent. Species present could be true shrubs, young trees and shrubs, or trees that are small or stunted due to environmental conditions. Palustrine Emergent Wetland (Persistent) includes tidal and nontidal wetlands dominated by persistent emergent vascular plants, emergent mosses or lichens, and all such wetlands that occur in tidal areas in which salinity due to ocean derived salts is below 0.5 percent. Total vegetation cover is greater than 80 percent. Pla nts generally remain standing until the next growing season.
70 Table 3 2 Continued. Class name Description Estuarine Wetlands Estuarine Forested Wetland includes tidal wetlands dominated by woody vegetation greater than or equal to 5 meters in height, and all such wetlands that occur in tidal areas in which salinity due to ocean derived salts is equal to or greater than 0.5 percent. Total vegetation coverage is greater than 20 percent. Estuarine Scrub / Shrub Wetland includes tidal wetlands domina ted by woody vegetation less than 5 meters in height, and all such wetlands that occur in tidal areas in which salinity due to ocean derived salts is equal to or greater than 0.5 percent. Total vegetation coverage is greater than 20 percent. Estuarin e Emergent Wetland Includes all tidal wetlands dominated by erect, rooted, herbaceous hydrophytes (excluding mosses and lichens). Wetlands that occur in tidal areas in which salinity due to ocean derived salts is equal to or greater than 0.5 percent and th at are present for most of the growing season in most years. Total vegetation cover is greater than 80 percent. Perennial plants usually dominate these wetlands. Water and Submerged Lands Open Water include areas of open water, generally with less tha n 25 percent cover of vegetation or soil. Palustrine Aquatic Bed includes tidal and nontidal wetlands and deepwater habitats in which salinity due to ocean derived salts is below 0.5 percent and which are dominated by plants that grow and form a cont inuous cover principally on or at the surface of the water. These include algal mats, detached floating mats, and rooted vascular plant assemblages. Total vegetation cover is greater than 80 percent. Estuarine Aquatic Bed includes tidal wetlands and deepwater habitats in which salinity due to ocean derived salts is equal to or greater than 0.5 percent and which are dominated by plants that grow and form a continuous cover principally on or at the surface of the water. These include algal mats, kelp be ds, and rooted vascular plant assemblages. Total vegetation cover is greater than 80 percent.
71 Table 3 3 Significant period CCAP land cover class The significant period are represented in star mark Land Cover / period 1_Developed Land 2_Agricultural La nd 4_Forest Land 5_Scrub Land 7_Palustrine Wetlands 8_Estuarine Wetlands 10_Water and Submerged Lands Florida NDVI Rain NDVI Rain NDVI Rain NDVI Rain NDVI Rain NDVI Rain NDVI Rain NDVI Rain 2.07 2.46 * * 2.92 3.47 4.13 4.91 5.84 * * * 6.95 8.26 9.83 11.69 * * *
72 Table 3 4 Significant period Climate Division The significant peri od are represented in star mark Climate Division / period 1 2 3 4 5 6 Florida NDVI Rain NDVI Rain NDVI Rain NDVI Rain NDVI Rain NDVI Rain NDVI Rain 2.07 2.46 * 2.92 3.47 4.13 4.91 5.84 * * * 6.95 8.26 9.83 11.69 * * *
73 Figure 3 1 Study area: Florida State, U.S.A.
74 Figure 3 2 Florida statewide average monthly precipitation (Source: NCDC 2011) Figure 3 3 Florida statewide aver age monthly precipitation and distribution [A dapted from Fernald, E. A. and Patton, D. J. 1984 Water resources atlas of Florida (Page 121, Figure 3 1). Florida State University Florida ]
75 Figure 3 4 Climate divisions in Florida (Source: NCDC 2011)
76 a) b) c) F igure 3 5 Land cover classification data from C CAP for the year a) 1996, b) 2001, and c) 2006 (Source: NOAA 2011)
77 Figure 3 6. Dominant land cover types at different percentages (superimposed black line: climate divisions)
78 (a) (b) (c) (d) F igure 3 7 Precipitation and NDVI time series present in monthly order July 1981 to December 2006. (a) Precipitation for climate divisions and Florida; (b) NDVI for climate divisions and Florida (c) Precipitation for land cove r types and Florida; (d) NDVI for land cover types and Florida.
79 Figure 3 8 windowed Fourier transform (WFT), and (c) a wavelet transform (WT), and their time series represented in time space and frequency space. (Source: Lau and Weng 1995)
80 Figure 3 9 Wa hatched region is the cone of influence where zero padding has reduced the variance. Black contour is the 10% significance level, using a red noise (autoregressive la g1) background spectrum. c) The global wavelet power spectrum (black line). The dashed line is the significance for the global wavelet spectrum, assuming the same significance level and background spectrum as in b).
81 a) b) c) d) e) f) Figure 3 10 Wavelet analysis of NDVI for Florida climate divisions a) climate division 1 to f) climate division 6
82 a) b) c) d) e) f) Figure 3 11 Wavelet analysis of NDVI for different land cover types a) developed land b) agricultural land c) forest la nd d) scrub land e) palustrine wetlands f) estaurine wetlands
83 Figure 3 11 Continue d. Wavelet analysis of NDVI for water and submerged land
84 Figure 3 12 Wavelet analysis of precipitation of Florida: a) Precipitation data. b) The wavelet power spectru m. The contour levels are chosen so that 75%, 50%, 25%, and 5% of the wavelet power is above each level, respectively. The cross hatched region is the cone of influence, where zero padding has reduced the variance. Black contour is the 10% significance lev el, using a red noise (autoregressive lag1) background spectrum. c) The global wavelet power spectrum (black line). The dashed line is the significance for the global wavelet spectrum, assuming the same significance level and background spectrum as in (b).
85 a) b) c) d) e) f) Figure 3 13 Wavelet analysis of precipitation for Florida climate divisions a) climate division 1 to f) climate division 6
86 a) b) c) d) Figure 3 14 Wavelet analysis of precipitation for different land cover type s a) developed land b) agricultural land c) forest land d) scrub land
87 e) f) g) Figure 3 14 Continued. Wavelet analysis of precipitation for different land cover types. e) palustrine wetlands f) estaurine wetlands and g) water and submerged la nd
88 CHAPTER 4 IMPACTS OF LAND COVE R AND LAND USE ON AI R QUALITY IN FLORIDA 4.1 Background The impacts of land cover and land use are studied intensively from many aspects (Lambin et al. 2001, Patz et al. 2005, Felet et al. 2005, Lambin et al. 2006). Utiliz ing modern techniques with concentration on assessing public health issues become a hot topic in academic (Gimms et al. 2008, Zhou et al. 2011). Air quality concerns have been raised according to the degree of land cover change coupled with consequences fr om anthropogenic influences (Kinney 2008, Yuan 2008, Jacob et al. 2009). Florida is experiencing a massive land cover and land use change during the twentieth century (Marshall et al. 2004). From 2000 to 2010, there was a 17.6 percent increase of Florid an increase in societal demand is a big contributor to agricultural development, land cover change/ur banization and air pollution in Florida (Samet et al. 2000, Solecki and Walker 2001, USGS 2004, Hu et al. 2008, Zanobetti and Schwartz 2009). In general, air pollution contributing factors include emissions from vehicles, industrial facilities, electric u tilities like power plants and other combustion sources. All those factors are exaggerated by increased population and land cover change. Particulate matter (PM) and ground level ozone (O 3 ) are two of the six common air pollutants that have adverse effect s on human and environmental health. Of the six pollutants, particle pollution and ground level ozone are the most widespread health threats (US EPA 2012). Fine particles, or so called PM 2.5 are defined as particle diameter less than 2.5 micrometers. P M 2.5 can be directly emitted from sources such as
89 forest fires, or they can form when gases emitted from power plants, industries and automobiles react in the air (US EPA 2012). Ground level ozone sometimes referred as nto the air, but is created by chemical reactions between oxides of nitrogen (NOx) and volatile organic compounds (VOC) in the presence of sunlight. Common emission sources of NOx and VOC are from industrial facilities and electric utilities, motor vehicle exhaust, gasoline vapors, and chemical solvents. O 3 pollution is becoming a concern during summer months especially in many urban and suburban areas throughout the United States (US EPA 2012). PM 2.5 and O 3 have been continued monitored by EPA since 1999 a nd 1980s, respectively. In addition, since Florida is characterized by humid tropical climate for a majority state with the southernmost part by tropical savanna climate (NCDC 2011), climate variability also plays an important role in terms of its effect s on air quality. According to the previous studies about vegetation and precipitation variability, evidences are found across the state indicating that Florida is a place that having very active interactions between environmental factors and human induce d factors. Therefore, understanding the impacts of land cover change and the relationship between air pollutants becoming an important issue when a lot of complicate factors all contributing to the amount of air pollutants. Therefore, it is important to u nderstand the relationship between pollutants and their interaction with environmental factors (such as rainfall and temperature) and human induced factors (such as land cover and land use change) in Florida. In this s applied to investigate the relationship between land cover and land use with air pollutants, PM 2.5 and ground level O 3 A more detailed
90 correlation analysis on pollutants measurements from different land cover types is also investigated in this study. 4 .2 Data and M ethods 4.2.1 Particulate Matter (PM 2.5 ) Particulate matter, also known as particle pollution or PM, is a complex mixture of extremely small particles and liquid droplets. Particles that are 10 micrometers in diameter or smaller can pass throu affect human health by influencing heart and lungs functions with associated serious health problems (US EPA 2011). People with heart or lungs diseases, children and older adults are the most likely to be affected by particle pollution exposure (US EPA 2011, Brook et al., 2010, Ebi and McGregor 2008, Harrison and Yin, 2000). As well as PM affecting human health, the environmental effects of PM are also enormous such as reducing visibility, making lakes and streams acidic, changing the nutrient balance in coastal waters, damaging sensitive forests and farm crops, and affecting the diversity of ecosystems (US EPA 2011). Furthermore, air pollution concentrations are the result of interactions among local weath er patterns, atmospheric circulation features, wind, topography, human activities (i.e., transport and coal fired electricity generation), human responses to weather changes (i.e., the onset of cold or warm spells may increase heating and cooling needs and therefore energy needs), and other factors (Ebi and McGregor, 2008). In order to reduce the impacts of air pollution, the Clean Air Act Amendments, the layer was ena cted by Congress in 1990. Under the Clean Air Act, EPA sets and reviews national air quality standards (NAAQS) for wide spread pollutants like PM (Table 4 1).
91 Using a nationwide network of monitoring sites, EPA has developed ambient air quality trends for particle pollution. Although average PM concentrations have decreased over the years nationally (US EPA 2011), geographic variations play an important role on public health (Harrison and Yin, 2000). Daily PM 2.5 data are derived from US EPA Air Quality Sys tem (AQS) and subset to the geographical area of Florida for the year 2001 and 2006. Monitor sites are selected based on its data quantity and quality. Total twenty four monitor sites are selected for PM 2.5 data source and PM 2.5 values are aggregated at mo nthly basis in order to match with other materials (Figure 4 1). 4.2.2 Ozone (O 3 ) Ozone is one of the air quality concerns that have been set as one of six common air pollutants by the United States Environmental Protection Agency (EPA). Ozone can be found one in the upper regions of the atmosphere called stratosphere and the other one at ground level. Ozone in both layers is containing the same chemical composition (O 3 ). However, while the upper atmospheric ozone the Earth from harmful rays from the sun (biologically damaging ultraviolet sunlight), the e the harmful effect to human health. According to the negative health effect, the ground level ozone standards has been set up by EPA guided by the Clean Air Act Amendments since 1997 in order to reduce ozone air pollution and protect human/environment he alth. The ground level ozone is created by chemical reactions between oxides of nitrogen (NOx) and volatile organic compounds (VOC). Thus, ozone concentration
92 elevated on hot sunny days in urban environments and could be transported long distances by wind which means during a hot sunny day ground level ozone are more likely to reach unhealthy levels. Therefore, the ground level ozone is focused in this study and states as ozone (O 3 ) only. Very similar to PM, O 3 also affects both sensitive human population and environment. Vegetation and ecosystems including forests, parks, wildlife refuges and wilderness areas experience higher exposure to ozone can have adverse impacts including loss of species diversity and changes to habitat quality and water and nutrien t cycles. Daily O 3 data are derived from US EPA Air Quality System (AQS) and subset to the geographical area of Florida for the year 1996, 2001 and 2006. Monitor sites are selected based on its data quantity and quality. Total twenty one monitor sites are selected for O 3 data source and ozone values are aggregated at monthly basis in order to match with other materials. The arithmetic mean value and the maximum value of O 3 are extracted for analysis (Figure 4 1). 4.2.3 Normalized Difference Vegetation Index (NDVI) NDVI is calculated from the visible red waveband (RED) and near infrared (NIR) waveband reflected by vegetation as equation below (Eidenshink 1992). NDVI = (NIR RED)/(NIR+RED) This relatively simple algorithm produces output values in the range of 1.0 to 1.0. Increasing positive NDVI values indicate increasing amounts of healthy green vegetation. NDVI values near zero and decreasing negative values indicate non vegetated features such as barren surfaces (rock and soil) and water, snow, ice, and cl ouds (USGS 2010).
93 The Global Inventory Monitoring and Modeling System (GIMMS) group Normalized Difference Vegetation Index (NDVI) dataset was used in this study. The GIMMS dataset is a NDVI product available for a 25 year period spanning from 1981 to 2006 with its spatial resolution 8 km. The dataset is derived from imagery obtained from the Advanced Very High Resolution Radiometer (AVHRR) instrument onboard the NOAA satellite series 7, 9, 11, 14, 16 and 17. The GIMMS data set has been corrected for calibra tion, view geometry, volcanic aerosols, and other effects not related to vegetation change. The GIMMS data set is composited at a 15 day time step. For each month, the first composite is the maximum value composite from the first 15 days of the month and t he second is from days 16 through the end of the month. For each month, the highest NDVI value composite is chosen as the NDVI for the month because it represents the maximum NDVI value for the month. The monthly NDVI value are extracted for Florida for th e year 1996, 2001 and 2006 based on the locations of PM 2.5 and ozone monitor sites. 4.2.4 Precipitation and Maximum Temperature Data Precipitation and Maximum Temperature data were gathered from the PRISM Climate Group, Oregon State University. PRISM stand s for parameter elevation regression on independent slopes model, which is a climate mapping system developed by Dr. Christopher Daly, PRISM climate group director. The PRISM model allows for the incorporation of expert knowledge about the climate and can be particularly useful when data points are sparse. With this method, one can explicitly account for the effects of coastal influences, terrain barriers, temperature inversions, and other factors on spatial climatic patterns (Daly et al. 2002). PRISM data sets are recognized world wide as some of the highest quality spatial data sets currently available (Hijmans et al. 2005,
94 Hamann and Wang 2005, Wang et al. 2006, Daly 2006, Loarie et al. 2009) and is the tation and maximum temperature data were downloaded from the PRISM website and extracted for each monitor sites for the year 1996, 2001, and 2006. 4.2.5 NOAA CCAP Land Cover Classification Data Florida land cover classification data are available from the Coastal Change Analysis Program (C CAP) developed by the National Oceanic and Atmospheric Administration (NOAA) with collaborations of the Department of Commerce (DOC), National Ocean Service (NOS) and NOAA Coastal Services Center (CSC). Current production of the Coastal Change Analysis Program (C CAP) land cover datasets is accomplished through closely coordinated efforts with the U.S. Geological Survey (USGS) as it produces the National Land Cover Dataset (NLCD). C CAP data are developed, primarily, fr om Landsat Thematic Mapper (TM) satellite imagery. The smallest feature size (spatial resolution) that can be mapped is 30 meter pixels (1/4 acres) on the ground. Current C CAP datasets are available for Florida in the years 1996, 2001, and 2006 by 23 clas ses. For this study, the original 23 classes are kept in order to analysis the relationship between pollutants with more detailed land cover and land use. Additionally, 1 km and 5 km buffers are created for each PM 2.5 and ozone monitor sites. The percen tages of developed land area (four levels of developed intensity, Table 3 2 and Figure 4 2 ) within each buffer are recorded in order to analysis the relationship According to Xian and Crane (2005), a strong relationship has been found between impervious surface area (ISA) and urban land use. ISA is usually defined as
95 roofs, roads, parking lots, driveways, and sidewalks (Xian 2007), thus this study proposed that the developed land classification of NOA A CCAP land cover classification dataset can be considered as a substitute variable to represent ISA. More specifically, based on the classification scheme, the developed high intensity class has constructed materials account for 80 to 100 percent of the t otal cover, so the level of ISA is the highest among other developed land cover classes. The medium intensity class has constructed materials account for 50 to 79 percent of the total cover, so it can be considered as the second highest ISA level. The low intensity class has constructed materials account for 21 to 49 percent of the total cover, so it can be considered as the third highest ISA level. The open space has constructed materials account for less than 20 percent of the total cover, so it is the l owest ISA among others. 4.2.6 Correlation Analysis cover and land use on air quality. First of all, the relationship between air pollutants (PM 2.5 and O 3 ) with NDVI, rain, and maximum temperature are examined. Then PM 2.5 and O 3 monitor sites are categorized by their land cover type in order to evaluate land cover and land use influences on air pollutant variation. In this step, the percentages of developed land area within 1 km and 5 km buffers are included. Theoretically, the pollutants sources of PM 2.5 and O 3 include emissions from vehicles, industrial facilities, electric utilities like power plants and other combustion sources. This study propose a hypothesi s that surrounding areas that contain more constructed areas (higher percentage) may have a higher chance to contribute air pollutants. Additionally, the timing of the year may have a function in terms of pollutants concentration and the
96 interaction betwee n pollutants with climate variables. Therefore, all the variables are organized again in order to investigate the relationship by month. 4.3 Results and Discussion Figure 4 3 2.5 O 3 (arith metic mean and maximum) with NDVI, rain, and maximum temperature. Significant statistical negative correlations are found between PM 2.5 with rain and O 3 with rain. From many literatures (Querol et al. 2001, Pillai et al. 2002, Hien et al. 2002, Latha and Badarinath 2005), it is reasonable to expect a removal process like washout and rainout. The relationship between NDVI with PM 2.5 and O 3 arithmetic mean are not significant related. However, O 3 maximum is found significantly positively related to NDVI a nd maximum temperature. Maximum temperature is found significantly positive related to O 3 max which can be explained by O 3 concentration are more likely to increase during a hot sunny day due to its active photochemical reaction. Figure 4 4 shows catego rized monitor site by land cover type and their relationship with NDVI, rain and maximum temperature. Rainfall again is found significantly negative related to almost all PM 2.5 and O 3 between NDVI and pollut ants are more fluctuated by the land cover classes of the monitor sites. For monitor sites located in developed high intensity location (Figure 4 5) NDVIs are found to be negative related to pollutants which can be interpret that the higher the NDVI the lower the pollutant concentration (PM 2.5 and O 3 arithemetic mean). For monitor sites located in developed medium intensity location, NDVIs are found to be positive and significant related to pollutants concentration (PM 2.5 and O 3 arithmetic mean). This co uld be understandable if in this particular land cover, there are a lot of
97 contributions sources like electric facilities or vehicles emissions level are elevated through commute. For monitor sites located in developed low intensity location, NDVIs are fou nd to be negative related to pollutants but not significant. This could be reasonable to assume that in this low intensity area; it is commonly includes single family housing areas especially in rural neighborhoods, there are sparse sources that would con tribute to air pollution (Figure 4 6 and Figure 4 7) According to almost monitor site measure PM 2.5 and O 3 are located in developed land classes (and in different intensity level), the further analysis is focus on the relationship between pollutants wi th developed land. Additionally, O 3 maximum concentration has shown more clearly relationship with pollutants, so from now on the analysis is concentrate on O 3 maximum instead of O 3 arithmetic mean. First of all, the relationship of pollutants with NDVI, rain and maximum temperature are examined by simple charts visually In general, rainfall in all monitor sites located in developed land classes is reaching its maximum during summer time (Jul to Sep) and its minimum during late winter (Mar to Apr). Maxi mum temperature in all monitor sites reaches its highest during summer time and minimum during winter time. However, NDVI value stay pretty consistent does not matter what time of the year. It can be explained by human maintaining system are mainly control ling the landscape of developed land. For PM 2.5 the concentration reaches its highest point in May with minimum during late fall (Oct Dec). For O 3 max, the concentration reaches its highest point in May and minimum in Nov to Feb. A seasonal variation can be expected since the weather obviously influences the pollutants. Variation can also been seen between observations
98 years for pollutants, it can be assume that regional and local weather patterns, as well as anthropogenic sources, play a significant role Seasonal variation can be found from the monthly correlation analysis ( Figure 4 7 ). For the overall PM 2.5 (including all land cover classes), rainfall has been shown a constant negative correlation with pollutants with a stronger correlation in May and Sep. Maximum temperature appears to be negative correlated with pollutants from Aug to Feb and positive from May to July. It could be considered that during May to July, the temperature is generally higher and the usage of air conditions is expected to be higher. the higher chance for power plant to generate more contribution to PM 2.5 For O 3 the relationship with meteorological par ameters, rain and maximum temperature are more complex to explain. However, there are more significant positive correlation be found In O 3 max with maximum temperature which can be understandable from its photochemical process relationship. Figure 4 8 repr area. As a result, rainfall has shown a significant negative relationship with pollutants. The high intensity land cover has negative relationship with pollutants at 5 km buffer but not 1 km buffer (weak positive correlation). The medium intensity land cover has presented a constant negative relationship with pollutants in both 1 km and 5km buffer. The low intensity land cover has shown a positive correlation with PM 2.5 and O 3 arithmetic b ut negative correlation with O 3 max. The open space land cover holds consistent positive correlation with pollutants. Figure 4 8 represents the correlation coefficients of pollutants with the percentage developed lands cover (different
99 intensities) within 1 km and 5 km buffers. The single star indicates correlation is significant at the 0.05 level (2 tailed) and double stars indicate correlation is significant at the 0.01 level (2 tailed). 4.4 Summary and Finding The analyses of patterns and results reveal that air pollutants levels in Florida exhibit strong seasonal variations. However, NDVI with pollutants cannot be found a strong correlation in this study. Couple possible reasons including the limitation of in urban area is responsible for the failure of establish relationship. Additionally, the data quality is another issue with air pollutants monitor system. From the process of data collection, a significant amount of missing data situation is been found c onstraining the further investigation. Furthermore, the spatial resolution of GIMMS AVHRR NDVI and CCAP land cover classification data are much coarser than the point observation of air pollutants monitor sites. It could be also an issue of lacking more de tailed local information due to its pixel size. In this study, an inverse relationship between rainfall and pollutants PM 2.5 and O 3 are found. It provides strong evidences that weather conditions play an important role in ambient air quality regardless la nd cover types. However, future study and further investigation are needed in order to understand the pollutants spatial pattern and what are the spatial variations of the relationship between pollutants with environment conditions.
100 Table 4 1. National ambient air quality standards (NAAQS) (s ource: EPA. (2011). National ambient air quality standard. from http://www.epa.gov/air/criteria.html ) Primary Standards Secondary Standards Pollutant Level Avera ging Time Level Averaging Time Particulate Matter (PM 2.5 ) 15.0 g/m3 Annual (Arithmetic Average) Same as Primary 35 g/m3 24 hour Same as Primary Ozone (O 3 ) 0.075 ppm 8 hour Same as Primary
101 F igure 4 1 Florida air monitor site location for p articulate matter (PM 2.5 )and ozone (O 3 ) with climate division boundary embedded
102 Figure 4 2. Florida air monitor site location for particulate matter (PM 2.5 )and ozone (O 3 ) that locate in developed land use category
103 Figure 4 3 The correlation coef ficients between PM 2.5 and O 3 with rain, maximum temperature and NDVI for all monitor sites. Bar with solid black boundary indicates the correlation is significant at the 0.01 level (2 tailed)
104 Figure 4 4 The correlation coefficients between a) PM 2.5 b) O 3 mean and c) O 3 maximum with rain, maximum temperature and NDVI for all monitor sites by land cover types. Bar with solid black boundary indicates the correlation is significant at the 0.01 level (2 tailed). Bar with dashed black boundary indica tes the correlation is significant at the 0.05 level (2 tailed)
105 Figure 4 5 The correlation coefficients between PM 2.5 and O 3 with rain, maximum temperature and NDVI for all monitor sites. Bar with solid black boundary indicates the correlation is sign ificant at the 0.01 level (2 tailed)
106 Figure 4 6 The correlation coefficients between PM 2.5 and O 3 with rain, maximum temperature and NDVI for all monitor sites by climate divisions. Bar with solid black boundary indicates the correlation is signif icant at the 0.01 level (2 tailed). Bar with dashed black boundary indicates the correlation is significant at the 0.05 level (2 taile )
107 a) b) c) d) e) f) g) h) i) j) k) l) Figure 4 7 Seasonal variation of the correlation coefficient by cli mate divisions. a) to l) represents January to December. The size of black hollow circle represents the correlation coefficient equals 0.3. The circle with upward diagonal indicate the correlation is significant at the 0.05 level (2 tailed) and the solid b lack circle indicate the correlation is significant at the 0.01 level (2 tailed). Green color circle represents positive correlation and red color circle represents negative correlation
108 a) b) c) Figure 4 8 Correlation coefficients of pollutants wi th Buffer a) PM 2.5 b) O 3 mean and c) O 3 maximum
109 CHAPTER 5 CONCLUSION The overarching object of this present dissertation is to evaluate the overall vegetation trend/change; land cover and land use change and health impacts especially on air pollution in F lorida from 1982 to 2006. Numerous evidences have been pointed out that Florida is experiencing big challenges such as the effects of global climate change and anthropogenic activities (Mulholland et al. 1997, Esterling et al. 2000, Pielke et al. 2005, USG CRP 2009). According to that, an overall evaluation of the environmental changes including vegetation trend, land cover and land use, and associated impacts on air pollution is needed for Florida. First of all, by using the Normalized Difference Vegetatio n Index (NDVI) as vegetation representation, a time series approach is applied in order to assess vegetation dynamics across the state from 1982 2006. We argue that there is an increasing NDVI value during winter months from 1995 onward and this phenomenon could be explained by the Atlantic Multidecadal Oscillation (AMO) switched into its warm phase around 1995. This result also corresponding with an increased winter rainfall proportion has been found from a previous study. In addition, the response of vege tation to climate variability and land cover is investigated with remote sensing based land cover classification data. From the wavelet analysis result, it provide evidence that the NDVI responds to precipitation is found to be stronger in natural land cov er like estuarine wetlands than in human manipulated land cover such as developed land. Moreover, the impact of land cover and land use on air pollution is evaluated. The results point out there are seasonal variations exist in air pollution. However, air
110 pollution is found highly correlated to weather conditions especially an inverse relationship with precipitation. Future research opportunities are plenty based on the findings from this dissertation. This dissertation recommends that since modern remot e sensing satellites like MODIS (start at 2000) provides a better spatial resolution than AVHRR; it could be beneficial to use a finer spatial scaled data when the records are sufficient to support a long term investigation. Areas that have been identified having extreme increasing/decreasing NDVI can be another interesting research topic if considering its associated land cover change over time. Additionally, air pollution assessment can be improved if the monitored data can be gathered at low cost in the place of interest or the analysis could apply the other satellite based air pollutants information.
111 LIST OF REFERENCES Anyamba, A., & Tucker, C. J. (2005). Analysis of Sahelian vegetation dynamics using NOAA AVHRR NDVI data from 1981 2003. Journal of Ari d Environments, 63 (3), 596 614. Bala, G., Caldeira, K., Wickett, M., Phillips, T. J., Lobell, D. B., Delire, C., & Mirin, A. (2007). Combined climate and carbon cycle effects of large scale deforestation. Proceedings of the National Academy of Sciences, 1 04 (16), 6550 6555. Barbosa, H A, Huete, A. R., & Baethgen, W. E. (2006). A 20 year study of NDVI variability over the Northeast region of Brazil. J. Arid Environ., 67 (2), 288. Bartholom, E., & Belward, A. S. (2005). GLC2000: A new approach to global lan d cover mapping from Earth observation data. International Journal of Remote Sensing, 26 (9), 1959 1977. Bonan, Gordon B. (2008). Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science, 320 (5882), 1444 1449. Bounoua, L., DeFries, R., Collatz, G. J., Sellers, P., & Khan, H. (2002). Effects of land cover conversion on surface climate. Climatic Change, 52 (1), 29 64. Brook, Robert D., Rajagopalan, Sanjay, Pope, C. Arden, Brook, Jeffrey R., Bhatnagar, Aruni, Diez Roux, An a V., Kaufman, Joel D. (2010). Particulate matter air pollution and cardiovascular disease. Circulation, 121 (21), 2331 2378. Dale, Virginia H. (1997). The relationship between land use change and climate change. Ecological Applications, 7 (3), 753 76 9. Daly, Christopher. (2006). Guidelines for assessing the suitability of spatial climate data sets. International Journal of Climatology, 26 (6), 707 721. Davenport, M. L., & Nicholson, S. E. (1993). On the relation between rainfall and the normalized di fference vegetation index for diverse vegetation types in East Africa. International Journal of Remote Sensing, 14 2369 2389. Davidson, Eric A., & Janssens, Ivan A. (2006). Temperature sensitivity of soil carbon decomposition and feedbacks to climate cha nge. Nature, 440 (7081), 165 173. DeFries, R. S., Field, C. B., Fung, I., Collatz, G. J., & Bounoua, L. (1999). Combining satellite data and biogeochemical models to estimate global effects of human induced land cover change on carbon emissions and primary productivity. Global Biogeochem. Cycles, 13 (3), 803 815. Defries, R. S., Hansen, M. C., Townshend, J. R. G., Janetos, A. C., & Loveland, T. R. (2000). A new global 1 km dataset of percentage tree cover derived from remote sensing. Global Change Biology, 6 (2), 247 254.
112 Defries, R. S., & Townshend, J. R. G. (1994). NDVI derived land cover classifications at a global scale. International Journal of Remote Sensing, 15 (17), 3567 3586. Easterling, David R., Meehl, Gerald A., Parmesan, Camille, Changnon, Stanl ey A., Karl, Thomas R., & Mearns, Linda O. (2000). Climate extremes: Observations, modeling, and impacts. Science, 289 (5487), 2068 2074. Ebi, Kristie L., & McGregor, Glenn. (2008). Climate change, tropospheric ozone and particulate matter, and health impa cts. Environ Health Perspect, 116 (11), 1449 1455. Eidenshink, J. C. (1992). The 1990 conterminous US AVHRR dataset. Photogrammetric Engineering and Remote Sensing, 58 809 813. EPA, US. (2011). Particulate matter. Retrieved September 7, 2011, from http://www.epa.gov/pm/ Fabricante, I., Oesterheld, M., & Paruelo, J. M. (2009). Annual and seasonal variation of NDVI explained by current and previous precipitation across Northern Patagonia. Journal of Arid Environments, 73 (8), 745 753. Foley, Jonathan A., DeFries, Ruth, Asner, Gregory P., Barford, Carol, Bonan, Gordon, Carpenter, Stephen R., Snyder, Peter K. (2005). Global consequences of land use. Science, 309 (5734), 570 574. Friedl, Mark A., Sulla Menashe, Damie n, Tan, Bin, Schneider, Annemarie, Ramankutty, Navin, Sibley, Adam, & Huang, Xiaoman. (2010). MODIS collection 5 global land cover: Algorithm refinements and characterization of new datasets. Remote Sensing of Environment, 114 (1), 168 182. GLP. (2005). Gl obal land project: Science plan and implementation strategy. In B. Young (Ed.), (pp. 64). Gong, Dao Yi, & Shi, Pei Jun. (2003). Northern hemispheric NDVI variations associated with large scale climate indices in spring. International Journal of Remote Sens ing, 24 (12), 2559 2566. Grimm, Nancy B., Faeth, Stanley H., Golubiewski, Nancy E., Redman, Charles L., Wu, Jianguo, Bai, Xuemei, & Briggs, John M. (2008). Global change and the ecology of cities. Science, 319 (5864), 756 760. Group, PRISM Climate. (2011). US precipitation. from Oregon State University http://prism.oregonstate.edu Guo, W. Q., Yang, T. B., Dai, J. G., Shi, L., & Lu, Z. Y. (2008). Vegetation cover changes and their relationship to climate variation in the source region of the Yellow River, China, 1990 2000. International Journal of Remote Sensing, 29 (7), 2085 2103.
113 Gutman, Garik et al. (2004). Land change science: observing, monitoring and understanding trajectories of change on the Earth's surface (Vol. 6). Dordrecht ; Boston ; London : Kluwer Academic Publishers, c2004. Gmez Mendoza, L., Galicia, L., Cuevas Fernndez, M., Magaa, V., Gmez, G., & Palacio Prieto, J. (2008). Assessing onset and length of greening period in six vegetation types in O axaca, Mexico, using NDVI precipitation relationships. International Journal of Biometeorology, 52 (6), 511 520. Hamann, A., & Wang, T. L. (2005). Models of climatic normals for genecology and climate change studies in British Columbia. Agricultural and Fo rest Meteorology, 128 (3 4), 211 221. Harrison, Roy M., & Yin, Jianxin. (2000). Particulate matter in the atmosphere: Which particle properties are important for its effects on health? The Science of The Total Environment, 249 (1 3), 85 101. Harveson, Patr icia Moody. (2006). The impacts of urbanization on endangered Florida key deer. (Doctoral dissertation), Texas A&M University. Harwell, Mark A., Long, John F., Bartuska, Ann M., Gentile, John H., Harwell, Christine C., Myers, Victoria, & Ogden, John C. (1 996). Ecosystem management to achieve ecological sustainability: The case of South Florida. Environmental Management, 20 (4), 497 521. Herold, M., Mayaux, P., Woodcock, C. E., Baccini, A., & Schmullius, C. (2008). Some challenges in global land cover mappi ng: An assessment of agreement and accuracy in existing 1 km datasets. Remote Sensing of Environment, 112 (5), 2538 2556. Herrmann, Stefanie M., Anyamba, Assaf, & Tucker, Compton J. (2005). Recent trends in vegetation dynamics in the African Sahel and thei r relationship to climate. Global Environmental Change, 15 (4), 394 404. Hijmans, Robert J., Cameron, Susan E., Parra, Juan L., Jones, Peter G., & Jarvis, Andy. (2005). Very high resolution interpolated climate surfaces for global land areas. International Journal of Climatology, 25 (15), 1965 1978. Hu, Zhiyong, Liebens, Johan, & Rao, K. Ranga. (2008). Linking stroke mortality with air pollution, income, and greenness in northwest Florida: an ecological geographical study. International Journal of Health Ge ographics, 7 (1), 20. Jacob, Daniel J., & Winner, Darrell A. (2009). Effect of climate change on air quality. Atmospheric Environment, 43 (1), 51 63. Janssen, Marco A., Schoon, Michael L., Ke, Weimao, & Brner, Katy. (2006). Scholarly networks on resilienc e, vulnerability and adaptation within the human dimensions of global environmental change. Global Environmental Change, 16 (3), 240 252.
114 Jordan, Charles L. (1984). Water resources: Atlas of Florida : Florida State University. Justice, CO, Townshend, JRG, H olben, BN, & Tucker, CJ. (1985). Analysis of the phenology of global vegetation using meteorological satellite data. International Journal of Remote Sensing, 6 (8), 1271 1318. Kalnay, E., & Cai, M. (2003). Corrigendum: Impact of urbanization and land use c hange on climate. Nature, 425 (6953), 102 102. Karl, T. R., & Knight, R. W. (1998). Secular trends of precipitation amount, frequency, and intensity in the United States (Vol. 79). Boston, MA, ETATS UNIS: American Meteorological Society. Kawabata, A., Ichi i, K., & Yamaguchi, Y. (2001). Global monitoring of interannual changes in vegetation activities using NDVI and its relationships to temperature and precipitation. International Journal of Remote Sensing, 22 (7), 1377 1382. Keim, Barry D. (1997). Prelimina ry analysis of the temporal patterns of heavy rainfall across the southeastern United States. The Professional Geographer, 49 (1), 94 104. Lambin, E. F., & Geist, H. J. (2006). Land use and land cover change: Local processes and global impacts Berlin: Spr inger. Li, Ainong, Wang, Angsheng, Liang, Shunlin, & Zhou, Wancun. (2006). Eco environmental vulnerability evaluation in mountainous region using remote sensing and GIS -A case study in the upper reaches of Minjiang River, China. Ecological Modelling, 192 ( 1 2), 175 187. Loarie, Scott R., Duffy, Philip B., Hamilton, Healy, Asner, Gregory P., Field, Christopher B., & Ackerly, David D. (2009). The velocity of climate change. Nature, 462 (7276), 1052 1055. Magnuson, JJ, Kratz, TK, Frost, TM, Bowser, CJ, Benson BJ, & Nero, R. (1991). Expanding the temporal and spatial scales of ecological research and comparison of divergent ecosystems: roles for LTER in the United States. In P. Risser (Ed.), Long term Ecological Research (pp. 45 70). Sussex, England: John Wile y and Sons. Mao, Liang, Qiu, Youliang, Kusano, Claudia, & Xu, Xiaohui. (2012). Predicting regional space time variation of PM2.5 with land use regression model and MODIS data. Environmental Science and Pollution Research, 19 (1), 128 138. McCarthy, Mark P. Best, Martin J., & Betts, Richard A. (2010). Climate change in cities due to global warming and urban effects. Geophys. Res. Lett., 37 (9), L09705. Meyer, William B., & Ii, B. L. Turner. (1992). Human population growth and global land use/cover change. A nnual Review of Ecology and Systematics, 23 39 61.
115 Milesi, Cristina, Elvidge, Christopher D., Nemani, Ramakrishna R., & Running, Steven W. (2003). Assessing the impact of urban land development on net primary productivity in the southeastern United State s. Remote Sensing of Environment, 86 (3), 401 410. Moss, Richard H., Edmonds, Jae A., Hibbard, Kathy A., Manning, Martin R., Rose, Steven K., van Vuuren, Detlef P., Wilbanks, Thomas J. (2010). The next generation of scenarios for climate change resea rch and assessment. Nature, 463 (7282), 747 756. Mulholland, Patrick J., Best, G. Ronnie, Coutant, Charles C., Hornberger, George M., Meyer, Judy L., Robinson, Peter J., Wetzel, Robert G. (1997). Effects of climate change on freshwater ecosystems of the south eastern United States and the Gulf Coast of Mexico. Hydrological Processes, 11 (8), 949 970. NASA. (2009). NDVI: Satellites could help keep hungry populations fed as climate changes. 2011, from http://www.nasa.gov/topics/earth/features/obscure_da ta.html Nicholson, S. E., & Farrar, T. J. (1994). The influence of soil type on the relationships between NDVI, rainfall, and soil moisture in semiarid Botswana. I. NDVI response to rainfall. Remote Sensing of Environment, 50 (2), 107 120. Nicholson, Sharo n E., Davenport, Michael L., & Malo, Ada R. (1990). A comparison of the vegetation response to rainfall in the Sahel and East Africa, using normalized difference vegetation index from NOAA AVHRR. Climatic Change, 17 (2), 209 241. NOAA. (2011). Coastal chan ge analysis program regional land cover. O'Reilly, Catherine M., Alin, Simone R., Plisnier, Pierre Denis, Cohen, Andrew S., & McKee, Brent A. (2003). Climate change decreases aquatic ecosystem productivity of Lake Tanganyika, Africa. Nature, 424 (6950), 76 6 768. Oerlemans, J. (2005). Extracting a climate signal from 169 glacier records. Science, 308 (5722), 675 677. Pielke, R. A., Walko, R. L., Steyaert, L. T., Vidale, P. L., Liston, G. E., Lyons, W. A., & Chase, T. N. (1999). The influence of anthropogeni c landscape changes on weather in south Florida. Monthly Weather Review, 127 (7), 1663 1673. Pielke, Roger, & Niyogi, Dev. (2010). The role of landscape processes within the climate system landform structure, evolution, process control. In J. C. Otto & R Dikau (Eds.), (Vol. 115, pp. 67 85): Springer Berlin / Heidelberg. Poccard, Y. R. I. (1998). A statistical study of NDVI sensitivity to seasonal and interannual rainfall variations in Southern Africa. International Journal of Remote Sensing, 19 (15), 2907 2920.
116 Pongratz, J., Reick, C. H., Raddatz, T., & Claussen, M. (2009). Effects of anthropogenic land cover change on the carbon cycle of the last millennium. Global Biogeochemical Cycles, 23 Propastin, P. A., & Kappas, Martin. (2008). Reducing uncertain ty in modeling the NDVI precipitation relationship: A comparative study using global and local regression techniques. GIScience and remote sensing, 45 (1), 47 67. Propastin, P. A., & Muratova, N. R. (2006). Spatial non stationarity and time dependency in t he relationship between NOAA AVHRR NDVI and climatic determinants: Sensitivity to landuse/landcover change. Paper presented at the Geoscience and Remote Sensing Symposium, 2006. IGARSS 2006. IEEE International Conference. Rockstrom, Johan, Steffen, Will, N oone, Kevin, Persson, Asa, Chapin, F. Stuart, Lambin, Eric F., Foley, Jonathan A. (2009). A safe operating space for humanity. Nature, 461 (7263), 472 475. Romero, H., Ihl, M., Rivera, A., Zalazar, P., & Azocar, P. (1999). Rapid urban growth, land us e changes and air pollution in Santiago, Chile. Atmospheric Environment, 33 (24 25), 4039 4047. Rosenzweig, Cynthia, Karoly, David, Vicarelli, Marta, Neofotis, Peter, Wu, Qigang, Casassa, Gino, Imeson, Anton. (2008). Attributing physical and biologic al impacts to anthropogenic climate change. Nature, 453 (7193), 353 357. Samet, Jonathan M., Dominici, Francesca, Curriero, Frank C., Coursac, Ivan, & Zeger, Scott L. (2000). Fine particulate air pollution and mortality in 20 U.S. cities, 1987 1994. New En gland Journal of Medicine, 343 (24), 1742 1749. Steffen, W. L. (2004). Global change and the earth system : a planet under pressure / W. Steffen ... [et al.] Berlin ; New York: Springer. Shalaby, Adel, & Tateishi, Ryutaro. (2007). Remote sensing and GIS f or mapping and monitoring land cover and land use changes in the Northwestern coastal zone of Egypt. Applied Geography, 27 (1), 28 41. Sohl, Terry, & Sayler, Kristi. (2008). Using the FORE SCE model to project land cover change in the southeastern United S tates. Ecological Modelling, 219 (1 2), 49 65. Solecki, W.D., & Walker, R.T. (2001). Transformation of the South Florida landscape Growing populations, changing landscapes Studies from India, China, and the United States (pp. 237 273). Washington, D.C.: Na tional Academy Press. Tilman, David, Fargione, Joseph, Wolff, Brian, D'Antonio, Carla, Dobson, Andrew, Howarth, Robert, Swackhamer, Deborah. (2001). Forecasting agriculturally driven global environmental change. Science, 292 (5515), 281 284.
117 Townshen d, John, Justice, Christopher, Li, Wei, Gurney, Charlotte, & McManus, Jim. (1991). Global land cover classification by remote sensing: present capabilities and future possibilities. Remote Sensing of Environment, 35 (2 3), 243 255. Turner Ii, B. L., Kasper son, Roger E., Meyer, William B., Dow, Kirstin M., Golding, Dominic, Kasperson, Jeanne X., Ratick, Samuel J. (1990). Two types of global environmental change: Definitional and spatial scale issues in their human dimensions. Global Environmental Chang e, 1 (1), 14 22. United Nations. (2010). United nations press release USGCRP. (2009). Global climate change impacts in the United States USGS. (2004). Pollution. Environmental Quality. 2011, from http://sofia.usgs.gov/publications/reports/rali/eqpollution.html USGS. (2010). What is NDVI. 2011, from http://ivm.cr.usgs.gov/whatndvi.php Vitousek, Peter M., Mooney, Harold A., Lub chenco, Jane, & Melillo, Jerry M. (1997). Human domination of Earth's ecosystems. Science, 277 (5325), 494 499. Wang, J., Rich, P. M., & Price, K. P. (2003). Temporal responses of NDVI to precipitation and temperature in the central Great Plains, USA. Inte rnational Journal of Remote Sensing, 24 (11), 2345 2364. Wang, Jue, Price, K. P., & Rich, P. M. (2001). Spatial patterns of NDVI in response to precipitation and temperature in the central Great Plains. International Journal of Remote Sensing, 22 (18), 3827 3844. Wang, T., Hamann, A., Spittlehouse, D. L., & Aitken, S. N. (2006). Development of scale free climate data for Western Canada for use in resource management. International Journal of Climatology, 26 (3), 383 397. Xian, George, & Crane, Mike. (2005). Evaluation of urbanization influences on urban climate with remote sensing and climate observations. Paper presented at the 5th International Symposium. Xiao, Jieying, Shen, Yanjun, Ge, Jingfeng, Tateishi, Ryutaro, Tang, Changyuan, Liang, Yanqing, & Huang Zhiying. (2006). Evaluating urban expansion and land use change in Shijiazhuang, China, by using GIS and remote sensing. Landscape and Urban Planning, 75 (1 2), 69 80. Yang, Limin, Wylie, Bruce K., Tieszen, Larry L., & Reed, Bradley C. (1998). An analysi s of relationships among climate forcing and time integrated NDVI of grasslands over the U.S. Northern and Central Great Plains. Remote Sensing of Environment, 65 (1), 25 37.
118 Zanobetti, Antonella, & Schwartz, Joel. (2009). The effect of fine and coarse particulate air pollution on mortality: A national analysis. Environ Health Perspect, 117 (6), 898 903. Zeng, Ning, Neelin, J. David, Lau, K. M., & Tucker, Compton J. (1999). Enh ancement of interdecadal climate variability in the Sahel by vegetation interaction. Science, 286 (5444), 1537 1540.
119 BIOGRAPHICAL SKETCH Huiping Tsai received her bachelor degree in l andscape a rchitecture from Fu J en Cath o lic University and a master de gree in b ioenvironmental s ystems e ngineering from National Taiwan University in 2004. She began her graduate studies at the University of Florida in August 2008 and join ed group to pursue a doctorate degree. She will gr aduate in the fall of 2012 after spending four years being educated in Geography Department.