xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001915112
007 cr mnu|||uuuuu
008 071101s2006 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001778
Palandro, David A.
Coral reef habitat change and water clarity assessment (1984-2002) for the Florida Keys national marine sanctuary using landsat satellite data
h [electronic resource] /
by David A. Palandro.
[Tampa, Fla] :
b University of South Florida,
ABSTRACT: The decline of coral reef habitats has been witnessed on a global scale, with some of the most dramatic decline occurring in the florida keys. as remote sensing can provide a synoptic view of coral reef ecosystems, 28 landsat images (1984-2002) were utilized to study water clarity and habitat change. first, the data were used to derive the diffuse attenuation coefficient (kd, m-1), a measure of water clarity, for 29 sites throughout the florida keys national marine sanctuary (fknms). landsat-derived kd values from bands 1 (blue) and 2 (green) provided useful information for 26 of 29 sites, whereas band 3 (red) provided no consistent data due to the high absorption of red light by water. it was not possible to assess long-term temporal trends as data were acquired, at most, twice a year. spatial variability was high between sites and between regions (upper, middle and lower keys) for bands 1 (0.019 m-1 0.060 m-1) and 2 (0.036 m-1 0.076 m-1). the highest kd values were f ound in the upper keys, followed by the middle and lower keys, respectively. this trend was corroborated by in situ monitoring of kd(par). second, the data were used to assess benthic habitat changes in eight coral reef sites located in the fknms. a mahalanobis distance classification was trained for four classes using in situ ground-truth data. overall coral habitat decline was 61% (3.4%/y), from 19% (1984) to 7.7% (2002). in situ monitoring data acquired by the coral reef evaluation and monitoring project (cremp) for the eight reef sites (1996-2002) showed a loss in coral cover of 52%, whereas the landsat-derived coral-habitat cover declined 37% for the same time period. a trend comparison between the full cremp percent coral cover data (1996-2004) and the full landsat-derived coral habitat class (1984-2002) showed no significant difference between the rates of change (ancova f-test, p = 0.303). The derivation of Kd and benthic habitat maps produced from Landsat data could provide c oastal marine managers another tool to help in the decision-making process.
Dissertation (Ph.D.)--University of South Florida, 2006.
Includes bibliographical references.
Text (Electronic dissertation) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 116 pages.
Adviser: Frank E. Muller-Karger, Ph.D.
x Marine Science
t USF Electronic Theses and Dissertations.
bbtnfrrbrbfrbbrtrb f !b"rf#fb$%r&'f(r$rbfrb)r )*f+, fbrrf&frrbf-rft&tfbr trbb.&fbbrtrbbbbt )%rt,fbbt#fb$%fb%b 'f*bfrt$&r f #/,tb0 2+#&b!b3,+)+ $bb4t&5r3,+)+ 6+7%13,+)+ ,ff-)&r,+)+ ,bn%1#&b3,+)+ )rbt--*0 "*bb83 !b0brbbf%bbrb%rf%%*b39rb.&frrrb&rf %bttf%fbr :-fr3)*f+,
Dedication This is dedicated to my wife Marlena and kids, Conner and Marina. As well as to my parents and family, without whose help th is work would not have been possible.
Acknowledgements I thank my major advisor, Dr. Frank Mu ller-Karger and my committee members; Dr. Serge Andrfout, Dr. John C. Brock, Dr. Ph illip Dustan, and Dr. Pamela Hallock for their guidance, support and pa tience throughout this study. This work was accomplished with the assistance from others at the College of Marine Science, especially Dr. Edward VanVleet, as well as everyone at the Institute for Marine Remote Sensing, with special thanks to Dr. Chuanmin Hu, Damaris To rres-Pulliza, Christine Kranenburg, Laura Lorenzoni and Dr. Christopher Moses. I w ould also like to th ank the Coral Reef Evaluation and Monitoring Program (CREMP) for their fi eld support and expertise, specifically Dr. Carl Beaver and Michael Ca llahan. This work was primarily supported by NASAs Earth System Science Fellows hip Grant NGT5-30414. Funding was also provided by a NOAA Hurricane A lliance grant (PI-Dr. Pamela Hallock), NASA Grant NNG04GG04G (PI-Dr. Frank Muller-Karge r) and NASA Grant NNG04GO90G (PI-Dr. Serge Andrfout). Landsat and IKONOS Imagery were provided by the NASA Scientific Data Purchase Program (contact : Martha Maiden) and NOAA Special Projects Group (contacts: Dr. Steven Rohmann and Aurelie Shapiro) Research activities performed in the Florida Keys National Ma rine Sanctuary were under permits FKNMS2002-067, FKNMS 2003-015 and FKNMS 2004-007 (latter two permits to CREMP).
i Table of Contents List of Tables iii List of Figures iv Abstract vi Chapter One Introduction Why Coral Reefs? 1 Coral Reef Communities of the Florida Keys 2 Remote Sensing Technology and Coral Reef Ecosystems 6 Landsat 6 Focus of Dissertation 8 References 10 Chapter Two Water clarity in the Flor ida Keys, USA, as observed from space (1984-2002) Abstract 14 Introduction 15 Data and Methods 16 Derivation of Bathymetry 17 Atmospheric Correction of Landsat Imagery 18 Derivation of K d From R rs 22 Results 26 Spatial Variability in Water Clarity 28 Discussion 29 Water Clarity by Region 33 Conclusion 35 References 36 Chapter Three Quantification of coral reef habitat decline in the Florida Keys National Marine Sanctuary determined from satellite data (1984-2002) Abstract 41 Introduction 42 Methods 44 Study Sites 44 CREMP 47 In Situ Ground-truth Data 48
ii Image Processing 49 Image Classification 50 Results 53 Comparison to In Situ Spectral Data 53 Classification Analyses 55 Seasonal Variability 56 Change Detection 56 Change Detection Image Progression 75 Carysfort Reef 75 Grecian Rocks 75 Molasses Reef 76 Conch Reef 76 Sombrero Reef 77 Looe Key Reef 77 Western Sambo 78 Sand Key Reef 78 Comparison to CREMP 79 Discussion 83 Seasonal Variability 83 Coral Habitat Pixels in the Backreef 83 Progression of Coral Habitat D ecline and Class Separability 84 Change in Coral Habitat Cover 85 Comparison to CREMP 87 Conclusions 89 References 90 Chapter Four Conclusions Synopsis 95 Summary of Conclusions 96 Remote Sensing and Corals 97 K d Study Portability 99 Coral Habitat Change Study Portability 100 Benefits to Other Research Topics 101 Water Clarity for Shallow-water Ecosystems 102 Marine Park Designation 102 Fisheries 103 Carbonate Production 104 References 104 Appendix 107 About the Author End Page
iii List of Tables Table 1.1. Summary of specially designated areas encompassed by the Florida Keys National Marine Sanctuary. 4 Table 1.2. Spectral and spatial resolutions of the Landsat 7 Enhanced Thematic Mapper Plus. 7 Table 1.3. List of Landsat imag es used for dissertation. 9 Table 2.1. Bathymetric datasets and associated attributes. 17 Table 2.2. Mean Landsat-derived K d values (m -1 ) per region and year. 27 Table 2.3. Mean, minimum and maximum of Landsat-derived K d values (m -1 ) 30 per band and site. Table 3.1. Reef site ge neral locations and asso ciated regions. 44 Table 3.2. In situ ground-truth benthi c cover major classes and ancillary information. 48 Table 3.3. Mean ground-truth data for the visually es timated percentage of benthic constituents found in each set of derived training pixels per class (n = 192). 52 Table 3.4. Percent of Landsat-derived co ral habitat cover by location and region per year (fall and spring values have been averaged where possible), as well as percent change from 1984-2002). 65 Table 3.5. Relative ranking of reef site s between percent CREMP coral cover and Landsat coral habitat change (1996-2002). 81 Table 4.1. List of ten most cited coral reef remote sens ing references since 1995. 98
iv List of Figures Fig. 1.1. Location map and extent of the Fl orida Keys and Florida Keys National Marine Sanctuary including specially designated areas. 3 Fig. 1.2. Underwater photographs of Carysf ort Reef taken from the same location and vantage point from 1975, 1985, 1995 and 2004, respectively (Dustan 2003). 5 Fig. 2.1. Derived bathymetry map showing geographic extent and zoomed area. 19 Fig. 2.2. Color-stretched RGB composite of a Landsat image from path/row 15/43, pre-atmospheric correction (lef t) and post-atmospheric correction (right). 21 Fig. 2.3. Landsat image of Florida Keys with locations of the 29 K d -derived sites, numbered from northeast to southwest (locations listed in Table 2.3). 24 Fig. 2.4. Example of typical fitting curve used in derivation of K d for Sombrero Reef (spring, 2002). 25 Fig. 2.5. Temporal and seasonal variability of K d values for Landsat band 1 and band 2 for sites averaged by region; Upper, Middle and Lower Keys. 28 Fig. 2.6. Spatial variability of K d values for Landsat band 1 and band 2 for all 29 reef sites studied, with a ssociated RME error bars. 29 Fig. 2.7. Example of a high turbidity pl ume (bottom) passing through channels located around Long Key (Middle Keys) from 1996. 31 Fig. 2.8. Comparison of trends between WQMP K d ( PAR ) and Landsat-derived K d data (band 1 and 2 averaged). 33 Fig. 3.1. Location map of the eight r eef sites used in this study. 45 Fig. 3.2. RGB images of each reef site with Sanctuary Preservation Area (SPA) extent overlaid in white. 46
v Fig. 3.3. Remote sensing reflectance ( Rrs sr -1 ), for each Landsat band wavelength mean, for all four classes based on tr aining pixels utilized in spectral library. 53 Fig. 3.4. Comparison of Landsat-derived reflectance ( R %) from the spectral library (black solid) to in situ R values (grey solid) based on values published by Hochberg et al. (2003). 54 Fig. 3.5. Spring 2002 Landsat images (left) with appropriate 2006 IKONOS images (right) for the eight reef sites, namely: Carysfort Reef (a), Grecian Rocks (b), Molasses Reef (c), Conch Reef (d), Sombrero Reef (e), Looe Key Reef (f), Western Sambo (g) and Sand Key Reef (h). 57 Fig. 3.6. Complete classified dataset for spring images for Carysfort Reef (a), Grecian Rocks (b), Molasses Reef (c), Conch Reef (d), Sombrero Reef (e), Looe Key Reef (f), Western Sambo (g) and Sand Key Reef (h). 66 Fig 3.7. CREMP percent cover data for live coral, zoanthid ( Palythoa spp.) and total (coral + zoanthid). 79 Fig. 3.8. Correlation of Landsat percent coral habitat versus CREMP percent coral cover for all concurrent da ta points (1996-2002) in the Upper (!), Middle (0 ) and Lower () Keys, R 2 = 0.704 (n = 32). 81 Fig. 3.9. Percent coral habitat for Landsat (!) and CREMP (/) full-timeline data (y-axis) per year (x-a xis) with linear regression line added for each dataset, for selected Florida Keys reef sites. 82
vi Coral Reef Habitat Change and Water Clar ity Assessment (1984-2002) for the Florida Keys National Marine Sanctuary Using Landsat Satellite Data David A. Palandro ABSTRACT The decline of coral reef habitats has been witnessed on a global scal e, with some of the most dramatic decline occurring in the Flor ida Keys. As remote sensing can provide a synoptic view of coral reef ecosystems, 28 Landsat images (1984-2002) were utilized to study water clarity and habitat change. First, the data were used to derive the diffuse attenuation coefficient ( K d m -1 ), a measure of water clarit y, for 29 sites throughout the Florida Keys National Marine Sa nctuary (FKNMS). Landsat-derived K d values from bands 1 (blue) and 2 (green) provided useful information for 26 of 29 sites, whereas band 3 (red) provided no consistent da ta due to the high absorption of red light by water. It was not possible to assess long-term temporal trends as data we re acquired, at most, twice a year. Spatial variability was high between s ites and between regions (Upper, Middle and Lower Keys) for bands 1 (0.019 m -1 0.060 m -1 ) and 2 (0.036 m -1 0.076 m -1 ). The highest K d values were found in the Upper Keys, followed by the Middle and Lower Keys, respectively. This trend was corroborated by in situ monitoring of K d ( PAR ). Second, the data were used to assess benthic habitat change s in eight coral reef sites located in the FKNMS. A Mahalanobis distance classification was trained for four classes using in situ ground-truth data. Overall coral hab itat decline was 61% (3.4%/y), from 19% (1984) to 7.7% (2002). In situ monitoring data acquired by the Coral Reef Evaluation and Monitoring Project (CREMP) for the eight r eef sites (1996-2002) showed a loss in coral cover of 52%, whereas the Landsat-derived coral-habitat cover declined 37% for the same time period. A trend comparison between the full CREMP percent
vii coral cover data (1996-2004) a nd the full Landsat-derived co ral habitat cl ass (1984-2002) showed no significant difference between th e rates of change (ANCOVA F-test, p = 0.303). The derivation of K d and benthic habitat maps produced from Landsat data could provide coastal marine managers another tool to help in the decision-making process.
1 CHAPTER ONE Introduction Coral reef ecosystems have been studied usi ng satellite remote se nsing for about three decades. Over that time, the availabili ty of new technology and new methods has allowed widespread mapping and monitoring efforts to take place. With coral reefs in apparent decline worldwide, the use of sa tellite remote sensin g may provide the only means to rapidly map the shallow-water extent of coral reef ecos ystems worldwide. Remote sensing is also the only technology capable of providing historic data for areas that were previously unmapped, and provide a baseline against which to measure change. Why Coral Reefs? Coral reefs are one of the most remarkable eco systems on the planet. They are one of the most productive and diverse ecosystems in the sea (Birkeland 1997). Although fragile, coral reefs, and associated lagoons, are globally distribute d, and are estimated to cover 600,000 km 2 (Birkeland 1997). Their di stribution is controlled by geological, biological, physical and chemical factors (Hallock 1997). Coral reefs provide many benefits to huma ns, including physical shoreline protection from storm surge and high-energy wave action that would increase rates of erosion. They provide habitat for fish, mollusks and crusta ceans, and serve as nurseries for several species of pelagic fish. Coral reefs constitu te approximately 0.2% of the global marine ecosystem (surface area), but account for 1.8% ($375 billion) of its annual value (Costanza et al. 1997). Entir e economies are dependent on corals reefs for commercial fisheries, tourism (e.g., scuba and recreationa l fishing) and aquaculture (e.g., exotics and
2 aquarium trade) (Birkeland 1997). In the Unite d States, coral reef-r elated tourism is a significant economic engine estimated to gene rate $1.6 billion a year (Birkeland 1997). Coral Reef Communities of the Florida Keys, USA Discovered by Ponce De Leon in 1513, the Flor ida Keys are located at the southeastern tip of the Florida peninsula and extend over 355 km in a s outhwestern direction (Fig. 1.1). They are comprised of 822 low-lying isla nds, from Key Largo to the Dry Tortugas (Dustan 2000). The Florida Keys lie within Monroe County (Key Largo to Key West, 203 km) with a resident human population of 80,000 and nearly 4 million visitors a year (Johns et al. 2001). Between June 2000 a nd May 2001, the economic contribution of reef-related expenditures (s norkeling, scuba, boating and fishing) was nearly $490 million, supplying 9,848 jobs (Johns et al. 2001). The Florida Keys Reef Tract is comprised of both patch and bank reefs and falls under the jurisdiction of the Na tional Oceanic and Atmospheric Administrations (NOAA) Florida Keys National Marine Sanctuary (F KNMS). Created in 1990 by act of the US Congress, the FKNMS encompasses 9500 km 2 from Biscayne National Park to Dry Tortugas National Park, encircling but not incl uding the latter (Fig. 1.1). Within the confines of the FKNMS, there are eighteen Sanctuary Preservation Areas (SPA), 27 Wildlife Management Areas (WMA), six Existing Management Areas (EMA), four Special Use Areas (SUA) and two Ecological Re serves (ER) (Fig. 1.1), each with varying levels of restrictions, jurisdiction and enforcement beyond that of the Sanctuary (Table 1.1). The Environmental Protection Agency (EPA) and the State of Florida began the Water Quality Protection Plan to monitor wate r quality parameters in 1995 as the Water Quality Monitoring Project (WQMP) and coral habitats in 1996 with the Coral Reef Monitoring Project (CRMP), currently the Co ral Reef Evaluation a nd Monitoring Project (CREMP). Shinn et al. (1989) divided the Florida Keys Reef Tract into four distinct reef regions: Upper Keys (Fowey Rocks to Molasses Reef ), Middle Keys (Molasses Key to Looe
Fig. 1.1. Location map and extent of the Fl orida Keys and Florida Keys National Marine Sanctuary including specially designate d areas. Map created by Kevin Kirsch. 3
Table 1.1. Summary of specially designated areas encompassed by the Florida Keys National Marine Sanctuary. This is accompanied by a list of the major regulations beyond those of the normal Sanctuary regulations. Designation Number Major Regulations Sanctuary Preservation Area (SPA) 18 No contact or take zone (e.g., fishing) Discharging any matter Anchoring (mooring balls available) Wildlife Management Area (WMA) 27 Regulations compliment those already in place by U.S. Fish and Wildlife Service Existing Management Area (EMA) 6 Regulations compliment those already in place by local entity Special Use Area (SUA) 4 No entrance into area without permit Ecological Reserve (ER) 2 No contact or take zone Discharging any matter Anchoring Key), Lower Keys (Looe Key Reef to Cosgrove Shoal) and the Dry Tortugas. Historically, the coral reefs of the Upper (from Carysfort Reef south) and Lower Keys (between Looe Key Reef and Western Sambo Reef) have been the most robust (Shinn et al. 1989). The Middle Keys reefs tend to be poorly developed, lacking Acropora palmata (Elkhorn coral), which once were the major reef-building coral of the Keys. The Florida Bay Hypothesis, first set forth by Ginsburg and Shinn (1964), explained this difference in reef structure based on the physical influence of cooler water from Florida Bay bathing the corals in the Middle Keys region. This hypothesis was later expanded to include the chemical influences (e.g., increased nutrients) of Florida Bay (Dustan 1977, LaPointe 1999, Porter et al. 1999). Although coral decline has been noted in other locations of the world (Hughes et al., 2003), the degradation of the coral reefs of the Florida Keys has been staggering over the past three decades (Fig. 1.2) (Causey et al. 2000, Porter et al. 2002, Jaap et al. 2003). Decline of the coral reef ecosystems of the Florida Keys has been attributed to many factors, from local to global. They include: disease proliferation (Porter and Tougas 2001), nutrification (LaPointe and Clark 1992), tropical storms (Kleypas et al. 2001), 4
Fig. 1.2. Underwater photographs of Carysfort Reef taken from the same location and vantage point from 1975, 1985, 1995 and 2004, respectively (Dustan 2003). 5
6 mass mortality of the Diadema antillarum (Lessios 1988), and the increase in atmospheric CO 2 and subsequent global warming that has increased sea-surface temperature (bleaching, Hoegh-Guldberg 1999) and acidification (Kleypas et al. 1999). Remote Sensing Technology and the Coral Reef Ecosystem With coral reefs threatened worldwide, remote sensing is the only t ool that can provide current and historic synoptic views of reefs around the world. Several medium to high resolution satellite and aircra ft sensors exist that can pr ovide benthic habitat cover mapping and monitoring, water co lumn and depth derivation st udies for reef ecosystems. Coral reef benthic habitat mapping has been well documented using sa tellite and airborne data from Landsat (e.g., Purkis and Pa sterkamp 2004), Systeme Probatoire ( or Satellite Pour) lObservation de la Terre (SPOT) (e.g., Andrfout et al. 2001), IKONOS (e.g., Riegl and Purkis 2005), aerial photography (e .g., Lyzenga 1978), Airborne Imaging Spectroradiometer (AISA) (e.g., Torres-Pulliza 2004) and the Compact Airborne Spectrographic Imager (CASI) (e.g., Mumby et al. 2004). The mapping performed is primarily based on two methods, visual interpre tation of the imagery or identification of different spectral signatures of various benthic classes. Mapping studies have further led to the utilization of historic remote sensing data for coral reef ch ange detection studies (Dustan et al. 2001, Palandro et al. 2003a, Palandro et al. 2003b). Beyond mapping and monitoring there is also great interest in the fact that estimates of the diffuse attenuation coefficient ( K d m -1 ) can be derived by remotely sensed methods. K d is a measure of water clarity and, as such, can be a proxy for water quality in coral reef environments (Maritorena 1996, Palandro et al. 2004). Landsat Designed to study the terrestrial environmen t, the Landsat missions are a series of satellites with medium spatia l resolution sensors designed to image land surface areas. The series began with Landsat 1 in 1972 a nd there have been six follow-on missions (Landsat 2-7) with progressively improved sensors; Landsat 6 fa iled to reach orbit. Currently, NASA maintains operational status for Landsats 5 and 7, both of which are
7 still transmitting data archived by the United States Geological Survey (USGS) at the Earth Resources Observation and Science (E ROS) Data Center (EDC). The Landsat series has provided 34 years of uni nterrupted environmental data. The Landsat 7 Enhanced Thematic Mapper Plus (ETM+) is a whiskbroom sensor operated at an altitude of 705 km with a swath width of 185 km, capable of acquiring approximately 250 scenes per day. The ETM+ has a spatial resoluti on of 30 m in the visible (VIS) range and a temporal resolution, or revisit time, of si xteen days. Table 1.2 lists the spectral characteristics of the eight ETM+ bands. Landsat data have allowed researchers to map global coral reef ecosystems and have become the backbone for the Millennium Global Coral Reef Mapping Project, which intends to characterize, map and estimate th e extent of shallow coral reef ecosystems worldwide (Andrfout et al. 2006). Limitations to using Landsat for shallow-wate r environments have been found to be its spatial (Mumby and Edwards 2002) and spectra l resolutions (Hochberg and Atkinson 2000). However, no sensor designed specifically to observe coral r eefs exists. Some of the other mediumto high resolu tion satellite sensors that have been used include the Table 1.2. Spectral and Spatial resolutions of the Landsat 7 Enhanced Thematic Mapper Plus. Band Spectral range (nm) Pixel size (m) Visible 1 450 520 30 2 520 600 30 3 630 690 30 Infrared 4 760 900 30 5 1550 1750 30 7 10.4 12.5 m 60 Thermal 6 2080 2350 30 Panchromatic 8 500 900 15
8 Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) (Capolsini et al. 2003), Hyperion (Arnol d Dekker, personal communication), IKONOS (Palandro et al. 2003a), QuickBird (Mishr a et al. 2005) and SPOT 5 XS multispectral scanner (Chauvaud et al. 2001). Of these, only Hyperion, IKONOS and QuickBird have spectral coverage in the blue range, critical for the capability to examine shallow-water submerged lands in clear water conditions. Hyperion and QuickBird have smaller area coverage and require several images to pr ovide the same geographic coverage as one Landsat image. This is also true for air borne sensors, e.g., AISA, making them costly alternatives. Focus of the Dissertation Two Landsat images (path/row: 15/43 and 16/43) are required to cover the geographic extent of the Florida Keys Reef Tract. Th is dissertation uses fourteen Landsat image pairs (Table 1.3) to cover the Florida Keys from 1984 to 2002. These images provide data for the spring season (March-May) every two years and the fall season every six years (September-November). Each image underwent a series of calibration and corrections to best represent true in situ measurements of remote sensing reflectance. A detailed 30 m spatial resolution and 1 m ve rtical resolution bathymetric dataset was produced for the Florida Keys (0-20 m dept h) to provide a depth layer to the study. In Chapter Two, these images are used to estimate the diffuse attenuation coefficient ( K d ), water clarity, for 29 s ites. Valid estimates of K d were derived for Landsat band 1 (blue) and band 2 (green), but not for band 3 (red). Seasonal variability of the satellitederived K d for each site was analyzed. A spatial comparison was made between reef sites and among the three distinct geological regions of the Florida Keys (Upper, Middle and Lower). An emphasis on the Landsat-derived K d between regions was undertaken to help determine the dominant water column clar ity influences in those regions. K d variability was also analyzed spatially. Rigorous temporal trend analysis was not possible due to the disparity of data points; however a simple trend comp arison was made to an ongoing in situ monitoring dataset (WQMP data).
9 Table 1.3. List of Landsat images used for dissertation. Entity ID Landsat / Sensor Format Acquisition Date Path/Row 15/43 LT5015043008411210 5 /TM NLAPS 21-Apr-1984 LT5015043008430410 5 /TM NDF 30-Oct-1984 LT5015043008610110 5 /TM NLAPS 11-Apr-1986 LT5015043008807510 5 /TM NLAPS 15-Mar-1988 LT5015043009009610 5 /TM NLAPS 6-Apr-1990 LT5015043009028810 5 /TM NLAPS 15-Oct-1990 LT5015043009213410 5 /TM NLAPS 13-May-1992 LT5015043009409110 5 /TM NLAPS 1-Apr-1994 LT5015043009608110 5 /TM NLAPS 21-Mar-1996 LT5015043009625710 5 /TM NLAPS 13-Sep-1996 LT5015043009807010 5 /TM NLAPS 11-Mar-1998 L71015043_04320000527 7 / ETM+ HDF 27-May-2000 L71015043_04320020314 7 / ETM+ GeoTIFF 14-Mar-2002 L7101504304320021109 7 / ETM+ FAST 9-Nov-2002 Path/Row -16/43 LT5016043008413510 5 /TM GeoTIFF 14-May-1984 LT5016043008431110 5 /TM GeoTIFF 6-Nov-1984 LT5016043008612410 5 /TM GeoTIFF 4-May-1986 LT5016043008808210 5 /TM GeoTIFF 22-Mar-1988 LT5016043009011910 5 /TM GeoTIFF 29-Apr-1990 LT5016043009026310 5 /TM GeoTIFF 20-Sep-1990 LT4016043009214910 5 /TM GeoTIFF 28-May-1992 LT5016043009408210 5 /TM GeoTIFF 23-Mar-1994 LT5016043009607210 5 /TM GeoTIFF 12-Mar-1996 LT5016043009632810 5 /TM NDF 23-Nov-1996 LT5016043009812510 5 /TM GeoTIFF 5-May-1998 L71016043_04320000502 7 / ETM+ GeoTIFF 2-May-2000 L71016043_04320020321 7 / ETM+ GeoTIFF 21-Mar-2002 L71016043_04320021015 7 / ETM+ FAST 15-Oct-2002 Chapter Three examines changes in benthic habitat change detected over tim e using the series of Landsat images for eight reef sites located in the FKNMS. A watercolumn/depth correction was completed utilizing the Landsat-derived Kd data from the previous chapter. A spectral library was produced based training pixels developed from in situ ground-truthing data. Each Landsat image per reef site was classified using four distinct classes to produce a series of classification maps to determine change in coral
10 habitat. Analyses were made on seasonal cove rage variability and class separability, and then a comparison to in situ reflectance data was performed. The Landsat-derived coral habitat data were compared to in situ monitoring efforts, spec ifically those of CREMP (1996-2004), to help assess the accuracy of the remote sensing derived trends in benthic change. Chapter Four serves as a synthesis chapter to conclude the study. The chapter includes a summary of conclusions and places into context, in terms of publications, the results of this study. There is also a disc ussion of the portability of the methods utilized here and the benefits of these methods to other research topics. References Andrfout S, Claereboudt M, Matsakis P, Pages J, Dufour P (2001) Typology of atoll rims in Tuamotu Archipelago (French Polynesia) at landscape scale using SPOT HRV images. International Journal of Remote Sensing 22:987-1004 Andrfout S, Muller-Karger F, Robinson J, Kranenburg C, Torres-Pulliza D, Spraggins S, Murch B (2006) Global assessment of mode rn coral reef extent and diversity for regional science and management applicati ons: a view from space. 10th International Coral Reef Symposium:28 June-23 July 2004. Birkeland C (1997) Introduction. In: Birkeland C (ed) Life and death of coral reefs. Chapman and Hall, New York, pp 1-12 Capolsini P, Andrefouet S, Rion C, Payri C (2003) A comparison of Landsat ETM+, SPOT HRV, Ikonos, ASTER, and airborne MASTER data for coral reef habitat mapping in south pacific islands. Canadian Journal of Remote Sensing 29:187-200 Causey B, Delaney J, Diaz E, Dodge D, Garcia JR, Higgins J, Jaap W, Matos CA, Schmal GP, Rogers C, Miller MW, Turgeon DD (2000) Status of coral reef in the US Caribbean and Gulf of Mexico : Florida, Texas, Puerto Ri co, U.S. Virgin Islands and Navassa. In: Wilkinson C (ed) Status of Co ral Reefs of the World: 2000. AIMS, Cape Ferguson, Australia, pp 239-259 Chauvaud S, Bouchon C, Maniere R (2001) Thematic mapping of tropical marine communities (coral reefs, seagrass beds and mangroves) using SPOT data in Guadeloupe Island. Oceanologica Acta 24:S3-S16
11 Costanza R, d'Arge R, de Groot R, Farber S, Grasso M, Hannon B, Limburg K, Naeem S, O'Neill RV, Paruelo J, Raskin RG, Sutton P, van den Belt M (1997) The value of the world's ecosystem services and na tural capital. Nature 387:253-260 Dustan P (1977) Vitality of reef populations off Key Largo, Florida: recruitment and mortality. Environmental Geology 2:51-58 Dustan P (2000) Florida Keys. In: Sheppard C (ed) Seas at the Millennium: An Environmental Evaluation. Elsevier, London, UK, pp 405-414 Dustan P (2003) Ecological perspective: the decline of Carysfort Reef, Key Largo, Florida 1975-2000. In: Valette-Si lver NJ, Scavia D (eds) Ec ological forecasting: new tools for coastal and ecosystem management. NOAA NOS NCCOS 116 Dustan P, Dobson E, Nelson G (2001) Landsat Thematic Mapper: detection of shifts in community composition of coral reefs. Conservation Biology 15:892-902 Ginsburg RN, Shinn EA (1964) Distribution of the reef-building community in Florida and the Bahamas ( abstract ). American Association of Petroleum Geologists Bulletin 48:527 Hallock P (1997) Reefs and reef limestones in Earth history. In: Birkel and C (ed) life and death of coral reefs. Chapman and Hall, New York, pp 13-42 Hochberg EJ, Atkinson MJ (2000) Spectral discrimination of coral reef benthic communities. Coral Reefs 19:164-171 Hoegh-Guldberg O (1999) Climate change, co ral bleaching and the future of the world's coral reefs. Marine Fres hwater Resources 50:839-866 Hughes TP, Baird AH, Bellwood DR, Card M, Connolly SR, Folke C, Grosberg R, Hoegh-Guldberg O, Jackson JB, Kleypas J, Lough JM, Marshall P, Nystrom M, Palumbi SR, Pandolfi JM, Rosen B, Roughgarden J (2003) Climate change, human impacts, and the resilience of coral reefs. Science 301:929-933 Jaap WC, Porter JW, Wheaton J, Beaver CR, Callahan MK, Kidney J, Hackett K, Lybolt M, Kupfner S, Torres C, Sutherland K ( 2003) EPA/NOAA coral reef evaluation and monitoring project, 2002. Florida Marine Re search Institute, St Petersburg, FL 28 Johns GM, Leeworthy VR, Bell FW, Bonn MA (2001) Socioeconomic study of reefs in Southeast Florida. Hazen and Sawyer, Hollywood, FL 348 Kleypas JA, Buddemeier RW, Archer D, Gattuso J, Langdon C, Opdyke BN (1999) Geochemical consequences of increased carbon dioxide on coral reefs. Science 284:118120
12 Kleypas JA, Buddemeier RW, Gattuso JP (2001) The future of coral reefs in an age of global change. International Jour nal of Earth Sciences 90:426-437 Lapointe BE (1999) Simultaneous top-down a nd bottom-up forces control macroalgal blooms on coral reefs. Limnology and Oceanography 44:1586-1592 LaPointe BE, Clark MW (1992) Nutrient inputs from the watershed and coastal eutrophication in the Florid a Keys. Estuaries 15:465-476 Lessios HA (1988) Ma ss mortality of Diadema antellarum in the Caribbean: what have we learned? Annual Review of Ecology and Systematics 19:371-393 Lyzenga D (1978) Passive remote sensi ng techniques for mapping water depth and bottom features. Applied Optics 17:379-383 Maritorena S (1996) Remote sensing of the wa ter attenuation in cora l reefs: a case study in French Polynesia. International Journal of Remote Sensing 17:155-166 Mishra D, Narumalani S, Rundquist D, Lawson M (2005) Benthic Habitat maping in tropical marine environments using quic kbird multispectral data. Photogrammetric Engineering & Remote Sensing 72:1037-1048 Mumby PJ, Edwards AJ (2002) Mapping mari ne environments with IKONOS imagery: enhanced spatial resolution can deliver grea ter thematic accuracy. Remote Sensing of Environment 82:248-257 Mumby P, Hedley J, Chisholm J, Clark C, Ripley H, Jaubert J (2004) The cover of living and dead corals from airborne remote sensing. Coral Reefs 23:171-183 Palandro D, Andrfout S, Dustan P, Mulle r-Karger FE (2003a) Ch ange detection in coral reef communities using Ikonos satellite sensor imagery and historic aerial photographs. International Journa l of Remote Sensing 24:873-878 Palandro D, Andrfout S, Muller-Karger FE, Dustan P, Hu C, Hallock P (2003b) Detection of changes in coral reef co mmunities using Landsat 5/TM and Landsat 7/ETM+ data. Canadian Journal of Remote Sensing 29:201-209 Palandro D, Hu C, Andrfout S, Muller-Karger FE (2004) Synoptic water clarity assessment in the Florida Keys using diffuse attenuation coefficient estimated from Landsat imagery. Hydrobiologia 530-531:489-493 Porter JW, Kosmynin V, Patterson KL, Port er KG, Jaap WC, Wheaton JL, Hackett K, Lybolt M, Tsokos CP, Yanev G, Marcinek DM, Dotten J, Eaken D, Patterson M, Meier OW, Brill M, Dustan P (2002) Detection of co ral reef change by the Florida Keys Coral Reef Monitoring Project. In: Porter JWPaKG (ed) The Everglades, Florida Bay, and
13 Coral Reefs of the Florida Keys An Ecosystem Source Book. CRC Press, Boca Raton, pp 749-769 Porter JW, Lewis SK, Porter KG (1999) The effect of multiple stressors on the Florida Keys coral reef ecosystem: A landscape hypot hesis and a physiological test. Limnology and Oceanography 44:941-949 Porter JW, Tougas JI (2001) Reef ecosystems: threats to their biodiversity. In: Levin S (ed) Encyclopedia of Biodivers ity. Academic Press, pp 73-95 Purkis SJ, Pasterkamp R (2004) Integrating in situ reef-top refl ectance spectra with Landsat TM imagery to aid shallow-tropical benthic habitat mapping. Coral Reefs 23:520 Riegl BM, Purkis SJ (2005) Detection of shallow subtidal corals from IKONOS satellite and QTC View (50, 200 kHz) single-beam sonar data (Arabian Gulf; Dubai, UAE). Remote Sensing of Environment 95:96-114 Shinn EA, Lidz BH, Halley RB, Hudson JH, Kindinger JL (1989) Reefs of Florida and the Dry Tortugas Field Trip Guidebook T176. AGU, Washington D.C. 57 Torres-Pulliza D (2004) A multi-sensor comp arison for coral reef habitat mapping: A case study using a tropical patch reef environm ent in Biscayne National Park, Florida. Master's thesis, University of Puerto Rico, p 62
14 CHAPTER TWO Water clarity in the Florida Keys, USA, as observed from space (1984-2002) Abstract Landsat TM and ETM+ data were used to de rive the diffuse atte nuation coefficient ( K d m -1 ), a measure of water clarity, for 29 sites throughout the Florida Keys Reef Tract. A total of 28 individual Landsat images between 1984 and 2002 were used, with imagery gathered every two years for spring seasons a nd every six years for fall seasons. Useful information was obtained by Landsat bands 1 (blue) and 2 (green), except when sites were covered by clouds or showed turbid water. Landsat band 3 (red) provided no consistent data due to the high absorption of red light by water. Because image sampling represented only one or two samples per year on specific days, and because water turbidity may change over short time scales, it was not possible to assess temporal trends at the sites with the Landsat data. K d values in band 1 were higher in the spring (mean spring = 0.034 m -1 mean fall = 0.031 m -1 ) and band 2 were higher in the fall (mean spring = 0.056 m -1 mean fall = 0.058 m -1 ), but the differences were not statistically significant. Spatial variability was high between sites and between regions (Upper, Middle and Lower Keys), w ith band 1 ranges of 0.019 m -1 0.060 m -1 and band 2 ranges of 0.036 m -1 0.076 m -1 The highest K d values were found in the Upper Keys, followed by the Middle Keys and Lower Keys, respectively. This trend was corroborated by in situ monitoring of K d ( PAR ). The Landsat-derived K d values, and inherent variability, may be influenced by the dominant water mass associated with each Florida Keys region, as well as localized oceanic variables. The methodology used here may be applied to other reef areas and used with satellites th at offer higher temporal resolution to assess temporal change and variability.
15 Introduction Shallow tropical coral reef ecosystems of the world are experiencing a number of concurrent stresses that threaten their health (Klein and Orlando 1994, Boyer and Jones 2002, Aronson et al. 2003, Pandolfi et al. 2003, LaPointe et al. 2004). The Florida Keys Reef Tract, as well as the r eefs of much of the Caribbea n, has undergone a significant decline in live coral cover over the past three decades (Cause y et al. 2000, Porter et al. 2002, Gardner et al. 2003, Buddemeier and Wa re 2003, Grigg et al. 2005, Pandolfi et al. 2005). Indeed, there is growing evidence that anthropogenic impacts may increasingly be affecting water quality in th e Florida Keys (Hallock 2001, Koop et al. 2001, Lipp et al. 2002). Water quality can fluctuate depending upon the presence and concentration of nutrients (which affect cons tituents such as phytoplankton), colored dissolved organic material (CDOM) and suspended sediment. The Florida Keys Reef Tract, excluding the Dr y Tortugas, has historically been separated into three major geologic regions, namely th e Upper Keys (north Key Largo to Conch Reef), Middle Keys (Alligator Reef to Mo lasses Keys) and Lower Keys (Looe Key to Smith Shoal) (Porter et al. 2002; see also LaPointe and Clark 1992, Klein and Orlando 1994, Smith 1998, Causey et al. 2000, Boyer and Jones 2002 for similar geographical classification schemes). The Florida Keys National Marine Sanctuary Protection Act designated the Florida Keys National Marine Sanctuary (FKNMS) in 1990. The Water Quality Protection Plan (WQPP) was implemented by the Environmental Protection Agency (EPA), the National Oceanic and Atmospheric Administration (NOAA) and the state of Florida to monitor wate r quality and benthic habitats (i.e., corals and seagrass). Coral cover has been m onitored since 1996 at 40 sites in the FKNMS under the Coral Reef Evaluation and Monitori ng Project (CREMP) (Porter et al. 2002). The Water Quality Monitoring Project (WQM P) started in 1995 and monitors water quality parameters for 154 sites up to four times a year. Water quality is assessed through measurements of nutrient concentra tions, turbidity, temperature, salinity, and photosynthetically active radiation ( PAR Em -2 sec -1 ). The downwelling irradiance
16 measurements are used to compute a diffuse attenuation coefficient or K d (PAR) (m -1 ) (Boyer and Jones 2002). Remote sensing from satellites is an eff ective means to study and monitor the marine environment because it provides the capability for frequent synoptic observations (Smith and Baker 1978, Durnad et al. 2000, Morel a nd Maritorena 2001, Palandro et al. 2004). The Thematic Mapper (TM) sensors flown on the series of U.S. Landsat satellites have been used to map shallow-water benthic habitats (Smith et al. 1975, Jupp et al. 1985, Ahmad and Neil 1994, Palandro et al. 2003) based on visible spectral reflectance measurements at a spatial reso lution of 30 m per pixel. The spectral signatures of various benthic communities may serve as input to perform supervised or unsupervised classifications and to generate thematic classi fication maps. Landsat data have also been used to estimate water clarity (Lyzenga 1981, Palandro et al. 2004, Phinn et al. 2005) as a means to assess water quality over synoptic scales. As a proof of concept, Palandro et al. 2004 utilized three Landsat images to determin e seasonal water clarity variability in the Florida Keys. Here we derive the diffuse attenuation coefficient ( K d m -1 ), a measure of water clarity, from a series of 28 Landsat images for 29 sites throughout the Florida Keys Reef Tract from 1984 through 2002. The goal of the study is to determine whether there are persistent differences in water clarity among the three major regions of the Florida Keys (Upper, Middle and Lower Keys) as well as determine the variability per individual reef site over time. Data and Methods Two Landsat sensors were used for this study, namely the Thematic Mapper (TM) onboard the Landsat 5 satel lite and the Enhanced Them atic Mapper Plus (ETM+) onboard Landsat 7. Both sensors provide data in three discrete vi sible spectral bands, specifically band 1 (450 nm 520 nm), band 2 (520 nm 600 nm) and band 3 (630 nm 690 nm). The images have a nominal spatial resolution of 30 m. A total of 28 Landsat
17 images were obtained for this study spanning the period 1984 2002, with individual images every two years for the spring season and every six years for the fall season. Two scenes are required to cover the Florida Ke ys, path/rows 15/43 and 16/43. To derive K d from Landsat data, it is also necessary to ha ve a detailed and accurate bathymetric map, and to correct the Landsat data for atmospheri c effects. We describe how we addressed these requirements below. Derivation of Bathymetry Even though the Florida Keys are one the world's most studi ed coral reef ecosystems, no high-resolution (30 m horizontal) bathymet ric map available covered the region of interest. To establish a regional bathymetr y, three different bathymetric datasets were combined, each with specific spatial resolution, depth resolu tion and geographic coverage (Table 2.1). The first was extracted from the joint NOAA and Florida Fish and Wildlife Conservation Commission (FWC) "Benthic Habitats of the Florida Keys (BHFK)" (Department of Commerce 1998). The second was extracted from NOAAs "Geophysical Data System (GEODAS)". The third was a subset from a separate FWC bathymetric dataset (contact: Chris Anderson, FW C). The metadata for all these datasets suggests that they were orig inally derived from the same hydrographic soundings held by the NOAA National Ocean Service (NOS) and preceding agencies, although they were processed and interpolated in different manners. Table 2.1. Bathymetric datasets and associated attributes. A 20 m lower depth limit was set by this study. Dataset Spatial Resolution Depth Interval Original Source File Format BHFK 30 m 1 m (< 10 m) 5 m (10 m 20 m) hydrographic soundings shapefile FWC 30 m 3 ft, 6 ft, 12 ft, 18 ft, 30 ft, 60 ft nautical charts shapefile GEODAS 90 m 1 m (< 20 m) hydrographic soundings .a98 raster
18 The various bathymetric datasets were first geo-referenced to the BHFK, and then fused through interpolation using Arc-GIS (Envir onmental Systems Res earch Institute/ESRI) Arc Geographic Information System), and ENVI /IDL (Research Systems, Inc.). For the FKNMS, Landsat imagery could not provi de quantitative benthic reflectance observations for depths greater than 20 m. Depths shallower than 2 m were difficult to define in the imagery due to changes in shor e topography. Therefore, depths near shore shallower than 2 m and great er than 20 m were excluded from the analysis. Depth intervals of 5 m were refined to 1 m resolution using an inverse distance weighted interpolation. The BHFK dataset lacked information in and around the Marquesas Islands, hence the other datasets were used to fill the gap. The result was a bathymetric map offering complete coverage of the Florid a Keys Reef Tract, with vertical resolution of 1 m between 2 and 20 m, and a horizontal resolution of 30 m (Fig. 2.1). It should be noted that the tidal range for this area ranges from 15 cm to 30 cm, which is within the 1 m vertical resolution of this bathymetry data set. This new bathymetric dataset provided sufficient accuracy for the current study. In no way does this imply that an accurate, spatially and vertically high resolution map of the Florid a Keys is no longer needed. Atmospheric Correction of Landsat Imagery The total visible radiance signal reaching a satellite sensor is comprised of photons reflected from the atmosphere, the water su rface and water column, and from the benthos in shallow water. The atmosphere can contri bute as much as 80-90% to the total signal over deep waters of the open ocean (Gordon 1997). In shallow water environments, the atmosphere's contribution may be smaller relative to that derived from the water and benthos, but still may dominate the total signa l. In either case, the color of the atmosphere must be removed to derive an accu rate estimate of the sea spectral reflectance (ocean color) (Gordon 1997). The simplest approach to remove atmospheric effects over shallow waters is to determine the signal over adjacent deep waters and then subtract the signal from the shallow-water pixels (Chavez 1988). This assumes that th e atmosphere is homogenous over the study
Fig. 2.1. Derived bathymetry map showing geographic extent and zoomed area. Areas shown in black are land or > 20 m in depth. 19
20 area and that negligible red light is eman ating from the surface of the deep ocean. Moreover, this method may lead to an overc orrection because it does not account for the water-leaving radiance derived from the deep water column, even at red wavelengths where the ocean is assumed to be black or "dark". A more sophisticated approach is to transfer to the Landsat observations the atmospheric reflectance properties estimated from concurrent observations from satellite sensors such as the Sea-viewing Wide Field-of-View Se nsor (SeaWiFS). The SeaWiFS atmospheric correction parameters estimated over the deep ocean are applied to adjacent shallow water pixels (Hu et al. 2001). To accomplish this requires first a vicarious crossreference of the calibration between the Landsat and the other sensor (Hu et al. 2001). The vicarious cross-reference calibration was not performed here so that each of the Landsat images could be atmospherically co rrected via the same method; SeaWiFS was not launched until 1997 and therefore could not be utilized on the Landsat images acquired before that time. For the particular application described in this paper, we relaxed this rigorous approach and derived the atmospheric properties over deep water with Landsat data only, and then propagated th ese estimates to shallow water, as follows. First, Landsat images were georectif ied and calibrated to radiance (mW cm -2 m -1 sr -1 ) using factors provided with each file. I then corrected the radiance for ozone absorption effects using either the near-concurrent satell ite-derived ozone path radiance estimates or climatological data from the National Ae ronautics and Space Administration (NASA). Next, two static masks were defined. The firs t one defined the area of interest, i.e., the shallow-water pixels covering the reef tract. The second mask defined the adjacent deep, clear-water pixels from which the atmosphe ric properties were derived. The shortest distance between the shallow and clear-water pi xels was specified to be > 20 pixels to minimize contamination of the deep water si gnals from light scattered from shallow water areas. The atmospheric reflectan ce over deep, clear water was derived by estimating a Rayleigh component using si ngle-scattering approximations, estimating
aerosol properties using the red band 3 ("dark pixel"), and then extrapolated to other bands assuming white-aerosols (Hu et al. 2001). For each pixel in the area of interest, a nearest-neighbor pixel from the clear-water mask was found (Hu et al. 2000), and all clear water pixels within < 15 pixels distance of this pixel were pooled to obtain the median values of the atmospheric properties. These median values were used to remove the atmospheric effects from the total radiance at the shallow water pixel by subtracting from the total radiance. The estimates of water-leaving radiance (L w ) were divided by an estimate of the downwelling irradiance (E d ) at the ocean's surface in that spectral band, to obtain the remote sensing reflectance (R rs =L w /E d sr -1 ) at each Landsat band (1, 2 and 3, for each band; Fig. 2.2) (Kirk 2003). Fig. 2.2. Color-stretched RGB composite of a Landsat image from path/row 15/43, pre-atmospheric correction (left) and post-atmospheric correction (right). Image inset displays zoom location. 21
Derivation of K d From R rs The R rs contains the signal derived from both the water column and the benthos. For benthic habitat mapping, it is often desirable to remove the contribution of the water column, as done with the signal from the atmosphere, so that the data include contributions only from the benthos. Here I am interested by the radiance associated with the water column to provide insight into the oceanographic and water quality conditions at the time the image was acquired. The downwelling diffuse attenuation coefficient is an apparent optical property (Kirk 1994) that is used to characterize how light propagates through the water column, typically from the surface toward the bottom. K d is also frequently used as a proxy for water quality (Smith and Baker 1978, Palandro et al. 2004). In principle, K d is proportional to the sum of total absorption (a, m -1 ) and the backscattering coefficients (b b m -1 ) of the water column, with a modulation factor that depends slightly on the solar and viewing geometry. Therefore, K d is often regarded as a quasi-inherent optical property (Kirk 1994, Yamano and Tamura 2000). K d can be estimated with a mathematical inversion using the radiance measured by a satellite sensor in one or several channels or bands (Lee et al. 1999, Palandro et al. 2004). This is done by applying Beers Law (Kirk 1994, Maritorena 1996) as follows. Beers Law states that: (1) z-KdddEzEe*)0()( where E d (z) and E d (0) are the downward irradiance at depth z and 0 (just below the surface of the water). K d is the average attenuation coefficient between 0 and z. Thus, if E d is measured at two different depths, K d can be derived. The radiance detected by a remote sensor over water after the removal of the atmospheric effects is L w Assuming the water is optically shallow (i.e., the bottom affects the reflectance observed from above the water's surface) and the diffuse attenuation in the 22
upward direction can be approximated by that for the downward direction (i.e., K u K d ), then L w as a function of bottom depth can be approximated as: (2) zKwdeC (z)L2* Where C is a constant if the bottom type does not change. Here the factor is to account for the attenuation in both the downward and upward directions (Maritorena 1996). C depends only on the surface irradiance (E d (0)) and bottom reflectance, which for each Landsat image do not vary from pixel to pixel over the same bottom type. Note that z is the depth of the bottom and L w is measured at the surface only. Hence, from L w measured at two locations (pixels) with the same bottom type but different bottom depths, z1 and z2, K d can be derived as: (3) )]12(2/[)]2(/)1(ln[zzzLzLKwwd Because E d is assumed not to vary from pixel to pixel, this is equivalent to: (4) )]12(2/[)]2(/)1(ln[zzzRzRKrsrsd We examined each target in the Florida Keys (Fig. 2.3) to locate where sand areas (bright pixels) were found. Then, for each depth as defined by the bathymetric map derived earlier, a histogram of R rs from all pixels was computed. Those pixels where R rs was within the top 5% were binned to derive a histogram (cloud pixels had been discarded using a predefined threshold of R rs 5% sr -1 ). Because several depths were used, an exponential fitting was used to determine K d and a relative error term was generated to describe the fitting quality. This method was applied to the series of Landsat images (bands 1-3) to derive K d at reef sites throughout the Florida Keys Reef Tract between the depths of 2 m and 20 m. 23
Fig. 2.3. Landsat image of Florida Keys with locations of the 29 K d -derived sites, numbered from northeast to southwest (locations listed in Table 3). The color of the sites denote geologic region; white Upper (1-10), green Middle (11-18) and yellow Lower (19-29). 24
A typical case of how K d was derived for one site is shown in Fig. 2.4. The majority of the observations followed the exponential shape expected for R rs (z). For a particular site, from all valid data (i.e., non-cloudy, sandy bottom, known depths), the following exponential equation was used to fit the data to determine K d : (5) zKrsdeCzR2*)( To assess errors, a mean relative error (MRE) for the regression was estimated as: (6) N/))ln(/))ln()(ln((MREfitrsrsfitrsRRR Where N is the total number of valid data points. For all 29 sites, the average MRE was 0.016 (band 1) and 0.015 (band 2). The average MRE decreased to 0.013 for both bands when clearly erroneous K d values were removed. Fig. 2.4. Example of typical fitting curve used in derivation of K d for Sombrero Reef (spring, 2002). Black line displays the linear fit between the ln(R rs (z)) (x axis) and depth (y axis). 25
26 Results In total, up to 14 observations were obtained at each site over 18 years, with four fall images and ten spring images per path/row. As expected, Landsat band 3 (red) did not provide reliable K d data. Red wavelengths are rapidly attenuated in wate r, and since the upper water column signal dominates R rs (Palandro et al. 2004), Eq. 2 is no longer valid. As a result, estimated K d values were often negative. Landsat bands 1 (blue) and 2 (green) provided useful K d data for 26 of the 29 sites studied. In some instances, some pixels yielded negative or low K d These artifacts were at tributed to possible high turbidity events when the bottom was not vi sible, causing failure of the algorithm. The K d values derived here for the Florida Keys correspond to comparatively clear waters, with low concentrations of both C DOM and chlorophyll (Smith and Baker 1978). Band 1 K d ranged from 0.019 m -1 (Lower Keys, 1984 spring) to 0.060 m -1 (Upper Keys, 1998), and showed an overall mean K d of 0.033 m -1 with a standard deviation of 0.010. Band 2 K d ranged from 0.036 m -1 (Lower Keys, 1984 spring) to 0.076 m -1 (Upper Keys, 1988) with an overall mean of 0.056 m -1 and standard deviation of 0.011. Regional mean K d values for each year are shown in Table 2.2. There is high variability in K d over the period of the study (1984-2002) (Fig. 2.5). Because for each site I obtained between zero to, at most, two instantaneous observations per year, the study of inter-annual variation in water clarity was not possible, i.e., no clear trend over time could be gleaned from th is dataset. Fig. 2.5 shows that band 1 K d values (mean of 0.034 m -1 in spring, 0.031 m -1 in fall) were always lower than band 2 values (0.056 m -1 in spring, 0.058 m -1 in fall). Further, the ratio of band 2 to band 1 was relatively constant over most of the reef sites, except in the lowest of the Lower Keys, where the band 2 to band 1 ratio nearly doubl ed. Variability between Landsat images from the same year but different se ason (1984, 1990, 1996 and 2002) did not yield a significant difference (paired ttest, p = 0.596). The Upper Ke ys exhibited the greatest difference between seasons (0.006 m -1 band 1) but was still not significant.
Table 2.2. Mean Landsat-derived K d values (m -1 ) per region and year. Year Region 1984s 1984f 1986 1988 1990s 1990f 1992 1994 1996s 1996f 1998 2000 2002s 2002f Upper Band 1 0.039 0.030 0.042 0.057 0.035 0.042 0.028 0.032 0.049 0.031 0.060 0.029 0.042 0.038 Band 2 0.054 0.053 0.068 0.076 0.061 0.073 0.043 0.048 0.071 0.063 0.073 0.046 0.071 0.062 Middle Band 1 0.033 0.023 0.039 0.041 0.030 0.037 0.026 0.027 0.030 0.040 0.024 0.041 0.039 0.022 Band 2 0.046 0.053 0.061 0.062 0.058 0.059 0.041 0.056 0.048 0.063 0.037 0.062 0.061 0.053 Lower Band 1 0.019 0.020 0.021 0.022 0.028 0.035 0.033 0.041 0.017 0.030 0.043 0.029 0.032 0.020 Band 2 0.036 0.043 0.040 0.048 0.053 0.063 0.059 0.067 0.041 0.065 0.067 0.051 0.062 0.042 27
0.000.010.020.030.040.050.060.070.081984198619881990199219941996199820002002YearKd (m1 ) Upper band 1 Upper band 2 Middle band 1 Middle band 2 Lower band 1 Lower band 2FFF F Fig. 2.5. Temporal and seasonal variability of K d values for Landsat band 1 and band 2 for sites averaged by region; Upper, Middle and Lower Keys. The letter F denotes fall, for those years comprised of two sets of images. Spatial Variability in Water Clarity Fig. 2.6 shows the mean Landsat-derived K d value dataset for all 29 reef sites. The Upper Keys (stations 1-10) showed the highest average K d in bands 1 and 2 (0.039 m -1 and 0.062 m -1 respectively; Table 2.3). The Middle Keys (stations 11-18) showed lower K d (averages of 0.032 m -1 for band 1 and 0.054 m -1 for band 2). Finally, the Lower Keys (stations 19-29) showed the clearest water (average K d of 0.027 m -1 for band 1 and 0.053 m -1 for band 2). Molasses Reef (Upper Keys) had the highest average band 1 K d (0.059 m -1 ) and Alligator Reef (Middle Keys) had the highest average band 2 K d value (0.076 m -1 ). The lowest average K d values for band 1 were detected at Eastern Dry Rocks (0.023 m -1 Lower Keys) and for band 2 west of Western Dry Rocks (0.042 m -1 Lower Keys). 28
0.000.010.020.030.040.050.060.070.080.090.10Upper 12345678910Middle 1112131415161718Lower 1920212223242526272829LocationKd (m-1) band 1 band 2 Fig. 2.6. Spatial variability of K d values for Landsat band 1 and band 2 for all 29 reef sites studied, with associated RME error bars. Upper, Middle and Lower Keys regions are noted. The Middle Keys displayed the highest variability in band 1 K d with a relative standard deviation of 33.4%, compared to 24.6% in the Upper Keys and 17.1% in the Lower Keys. Band 2 K d relative standard deviations were more similar for all three regions (i.e., 17.3% for Upper, 24.5% for Middle, and 21.5% for the Lower Keys). Carysfort Reef (Upper Keys) showed the largest range of K d for both bands. The ratio of band 2 to band 1 was nearly twice as high for the lower portion of the Lower Keys, i.e. Eastern Sambo (station 24) through Western Dry Rocks (station 28), as in the rest of the keys. Discussion Three sites presented the most difficulty with deriving K d values, namely south of Tennessee Reef (site 14), south of Sombrero Reef (site 18) and north of Looe Key Reef (site 19). These sites are hard bottom (CREMP) and have limited live coral coverage 29
30 Table 2.3. Mean, minimum and maximum of Landsat-derived K d values (m -1 ) per band and site. Note that lines delineate Upper, Middle and Lower Keys, respectively. Denotes reef sites where errone ous data have been removed. Band 1 Band 2 Site min max mean min max mean 1 Turtle 0.011 0.062 0.027 0.026 0.081 0.050 2 Carysfort / South Carysfort 0.013 0.110 0.041 0.028 0.126 0.063 3 South of Carysfort 0.026 0.087 0.046 0.051 0.103 0.075 4 Elbow / Dry Rocks 0.013 0.075 0.048 0.037 0.097 0.070 5 Grecian Rocks 0.017 0.067 0.041 0.043 0.079 0.065 6 Molasses 0.024 0.089 0.059 0.033 0.109 0.076 7 South of Molasses 0.018 0.039 0.029 0.029 0.058 0.045 8 Conch 0.018 0.045 0.032 0.043 0.074 0.059 9 Davis 0.019 0.042 0.035 0.037 0.073 0.050 10 Hens and Chickens 0.014 0.052 0.038 0.048 0.082 0.064 11 Alligator 0.024 0.065 0.046 0.037 0.093 0.078 12 Long Key 0.015 0.048 0.034 0.033 0.100 0.054 13 Tennessee 0.012 0.048 0.026 0.036 0.060 0.049 14 South of Tennessee* 0.012 0.018 0.012 0.026 0.042 0.033 15 Coffins Patch 0.013 0.050 0.027 0.032 0.068 0.047 16 Dustan Rocks 0.019 0.046 0.028 0.030 0.061 0.046 17 Sombrero 0.013 0.055 0.032 0.031 0.071 0.052 18 South of Sombrero* 0.015 0.067 0.026 0.036 0.099 0.050 19 North of Looe Key* 0.014 0.057 0.029 0.032 0.078 0.044 20 Looe Key 0.012 0.053 0.027 0.041 0.080 0.050 21 South of Looe Key 0.016 0.051 0.030 0.032 0.088 0.047 22 Western Washer Woman 0.013 0.044 0.024 0.030 0.063 0.044 23 Pelican Shoal 0.011 0.043 0.028 0.027 0.058 0.048 24 Eastern Sambo 0.010 0.053 0.026 0.027 0.105 0.067 25 Western Sambo 0.011 0.065 0.024 0.036 0.121 0.064 26 Eastern Dry Rocks 0.011 0.047 0.023 0.024 0.077 0.052 27 Rock Key / Sand Key 0.015 0.054 0.039 0.027 0.098 0.072 28 Western Dry Rocks 0.015 0.058 0.035 0.043 0.094 0.073 29 West of Western Dry Rocks 0.018 0.050 0.032 0.024 0.066 0.042 (i.e., < 2% live coral cover). They are also adjacent to large gaps in the Florida Keys, therefore are regularly influenced by turbid water intrusions from Florida Bay (Smith 1994, Porter et al. 1999, Lee and Smith 2002). Several images (7) illustrate highturbidity plumes emanating from the large channels in the Middle Keys toward the Atlantic Ocean, obscuring the benthos even in 3m of water (Fig. 2.7).
Fig. 2.7. Example of a high turbidity plume (bottom) passing through channels located around Long Key (Middle Keys) from 1996. Also shown is a time with no plume present in 2002 (top). 31
32 The Florida Water Quality Monitoring Pr oject (WQMP) has estimated the diffuse attenuation coefficient of photosynt hetically available radiation ( K d ( PAR ), 400-700 nm) for 154 sites four times per year throughout the FKNMS, starting in 1995 (Boyer 2004). These data are not directly comparable with the Landsat-derived K d values because they are not measured concurrently. Also, the WQMP K d values span a broader set of wavelengths than Landsat bands 1 or 2 ( PAR is 400-700 nm, compared to band 1: 450520 nm and band 2: 520-600 nm). Absolute K d values will obviously be higher for the WQMP as the Landsat data do not include 600700 nm (a range strongly influenced by the attenuation of pure water) (P ope and Fry 1997). Therefore, K d ( PAR ) will be several times larger than K d estimates for Landsat bands 1 and 2. In an attempt to link both datasets, Landsat K d data from bands 1 and 2 were averaged. The result was compared to the spring WQMP values for sites that ove rlap both datasets. However, without information about the water column, specifica lly in the red wavele ngth range, it is not possible to accurately estimate K d (PAR) from the Landsat information. There was no long-term trend apparent in K d of either the Landsat or the WQMP data (Fig. 2.8). Both the WQMP and the La ndsat data show the lowest overall K d values for Lower Keys, followed by the Middle and Uppe r Keys, respectively. Unfortunately, there is no instance when both datasets provided data concurrently for a direct comparison. However, both datasets exhibit a high variab ility, not only between regions, but between reef sites located in the same region. Some of my K d estimates were lower than those of pure water absorption (22.8%) (Pope and Fry 1997). The mean values below those of pure water absorption were 0.0016 m -1 and 0.0124 m -1 for bands 1 and 2 respectively. This can likely be attributed to the ETM+ sensor having a target radiometric accur acy of 5% (Chander and Markham 2003). A vicarious calibration of Landsat ETM+ data over clear water scenes using nearly concurrent SeaWiFS data and a radiative tran sfer model showed that bands 1 and 2 atsensor signals may be a few percent hi gher than those estimated from SeaWiFS measurements (Hu et al. 2001). Assuming SeaW iFS is well calibrated, the excessive few
0.000.050.100.126.96.36.19919841986198819901992199419961998200020022004YearKd (m1 ) Landsat SERC slope = 0.0003 slope = 0.0021 Fig. 8. Comparison of trends between WQMP K d (PAR) and Landsat-derived K d data (band 1 and 2 averaged). Error bars are one standard deviation. percent signal in Landsat ETM+ bands 1 and 2 will be added to the water-leaving radiances and result in a positive offset after atmospheric correction. The effect is that R rs (z1) and R rs (z2) in Eq. (4) are both overestimated by approximately the same offset, resulting in an underestimate in K d Nevertheless, the objective of this study is to investigate the relative changes in Landsat-derived K d across the region and time per reef site, so these values were utilized. Water Clarity by Region The FKNMS has been geographically delineated based on the needs of different water quality studies, specifically those utilized for water quality management (Klein and Orlando 1994, Boyer and Jones 2002). This spatial framework is based on the relative contribution of adjacent water masses (e.g., Florida Bay, oceanic water), residence time and circulation patterns. This framework follows the same zonation for the Upper and 33
34 Middle Keys. The delineation for Lower Keys va ries slightly. Sites west of Key West are placed in the Marquesas region. The Lower Keys were expected to have the lowest K d values (Klein and Orlando 1994, Szmant and Forrester 1996, Boyer and Jones 2002), and did. The relatively constant K d band 2 to band 1 (green:blue) ratio observ ed throughout the Florida Keys (Fig. 2.6) changed to some extent in the lower portion of the Lower Keys. The mean ratio changed from 1.63 (sites 1-23, 29) to 2.25 for Eastern Sa mbo (24) to Western Dry Rocks (28), an increase of 63% in the green to blue ratio. Baker and Smith (1982) found that K d variations in the blue (e.g., Landsat band 1) can be mainly attributed to chlorophyll and CDOM, whereas K d variations in the green (e.g., La ndsat band 2) can be mainly attributed to chlorophyll. The change in ratio may indicate an increase in the chlorophyll concentration for this region. However, because the increase in K d band 2 is not accompanied by a corresponding increase in K d band 1, and because of the lack of field measurements, this assertion cannot be conf irmed. Still, Boyer and Jones (2000) found the Marquesas region to have the highest chlo rophyll concentration in the Florida Keys. These results agree with the sp atial framework utilized for water quality for the Florida Keys mentioned above (Boyer and Jones 2002). The Marquesas region is a transitional area that differs from the adjacent Lower Keys region in that it has a shorter residence time and is less influenced by adjacent areas (Klein and Orlando 1994). The change in ratio may indicate the relative importance of the different water masses (i.e., input of different water column constituents) to these reef sites. It was unexpected to find that the Upper Keys had the highest K d values (least clear water). This is, however, in agreement with the WQMP data, when all sites are included (Boyer and Jones 2002). This may be in cont rast with Szmant and Forrester (1996) who found the highest nutrient concentr ations in the Middle Keys. K d values derived from Landsat bands 1 and 2 are controlled by chlor ophyll and CDOM and therefore related to nutrient concentrations (Baker and Smith 1982). However there was a greater gradient of nutrient concentrations, at least in the inshore to offs hore direction, for the Upper Keys
35 (Szmant and Forrester 1996). The results from this study cannot show the gradient as the study areas were taken as a si ngle unit, i.e., regardless of cross-shelf distance. A significant onshore to offshore gradient in nutrient concentration, which would be averaged within a single reef site in this study, would show a high level of variability in water clarity for the region if east-west tran sport occurred. Transport in the east-west direction is prevalent for part of the year (Smith 1998). As the Middle Keys historica lly (Shinn et al. 1989) and curr ently (Porter et al. 2002) have the lowest percent coral cover, it was assume d that this region would have the highest K d values. The three sites that provided the most difficulty in deri ving consistent valid K d estimates due to high turbidity are located in the Middle Keys. As the unreliable data from these three sites were excluded from the overall study due to low K d values. The low K d values were a function of the bent hos being obscured by the water column, therefore Eq. 4 was no longer valid because it could not utilize accurate depth data ( z ). The turbid water column had a high Rrs independent of depth, therefore over deep water the high Rrs value would cause a lower slope of the line used to define K d from the Landsat image data (Fig. 2.4). All three sites are adjacent to channels to Florida Bay, and as such are susceptible to th e highest turbidity (Boyer and Jo nes 2002). If consistent data from these sites were available, both the mean K d values, a well as their variability would be expected to have been higher for the region as a whole and the relative K d rank of Upper Keys versus Middle Keys would be reversed (i.e., Middle Keys with a higher mean Landsat-derived K d ). Conclusions Landsat image data from 1984-2002 were anal yzed to determine the validity of a Landsat-derived K d for 29 discrete sites in the Florida Keys Reef Tract. These results were then compared to ongoing in situ monitoring efforts. Landsat-derived K d provided consistent data for 26 of 29 sites for bands 1 (blue) and 2 (green). Erroneous and often negative values were acquired for band 3 (red); this is attributed to the rapid attenuation of red light in water. The th ree sites that provided unreliable data were all located in the
36 Middle Keys and adjacent to passes between the Florida Keys, where the input of high turbidity water is prevalent. The Lower Keys we re found to have the lowest K d values (clearest water), followed by the Middle and U pper Keys, respectively. This pattern was similar to K d (PAR) derived from ongoing in situ monitoring efforts. Comparisons between seasons showed small differences in mean Landsat-derived K d that were not statistically significant, but data may have been biased due to only four fall images. Water clarity and water quality are highly va riable between and within regions of the Florida Keys. Inter-region variability has been linked to the influence of the dominant water mass feature, e.g., residence time a nd nutrient concentra tion. Intra-region variability must be due to sm aller scale processes acting at the site level. Satellite monitoring every two years prov ide snapshots into the relati ve spatial trends in water clarity in the Florida Keys, but cannot pr ovide data to infer temporal change. In situ monitoring up to four times a year can pr ovide inter-site co mparison and seasonal variation but lacks the repeat c overage to detect event-driven changes in water clarity and may be biasing data for certain time periods. The results found here are consistent with those of the in situ (WQMP) monitoring effort. Daily satellite coverage, like that provided by the Moderate Imaging Spect roradiometer (MODIS) launched in December1999, can provide a tool at a medium spatial resolution to fill gaps in the in situ monitoring datasets. An estimate of water clarity, i.e., K d (PAR) or K d 490 from the satellite would provide a robust data set that is comparable to the in situ data and provide coastal zone managers another tool to detect changes in water qual ity of their protected ecosystem. References Ahmad W, Neil DT (1994) An Ev aluation of Landsat Thematic Mapper (TM) digital data for discriminating coral reef zonation: He ron Reef (GBR). Intern ational Journal of Remote Sensing 15:2583-2597 Aronson RB, Bruno JF, Precht WF, Glynn PW, Harvell CD, Kaufman L, Rogers CS, Shinn EA, Valentine, F. J, Pandolfi JM, Br adbury RH, Sala E, Hughes TP, Bjorndal KA,
37 Cooke RG, McArdle D, McClenachan L, Newman MJH, Paredes G, Warner RR, Jackson, C. JB, Hughes TP, Baird AH, Bellwood DR, Connolly SR, Folke C, Grosberg R, Hoegh-Guldberg O, Jackson JBC, Kle ypas J, Lough JM, Marshall P, Nystrom M, Palumbi SR, Pandolfi JM, Rosen B, Roughga rden J (2003) Causes of Coral Reef Degradation. Science 302:1502b-1504 Baker KS, Smith RC (1982) Bio-optical classification and model of natural waters. 2. Limnology and Oceanography 27:500-509 Boyer JN (2004) FY2004 Annual report of the water quality monitoring project. Florida International University, Miami, FL 84 Boyer JN, Jones RD (2002) A view from the bri dge: external and internal forces affecting the ambient water quality of the Florida Keys National Marine Sanctuary (FKNMS). In: Porter JWP and KG (ed) The Everglades, Fl orida Bay, and Coral R eefs of the Florida Keys An Ecosystem Source Book. CRC Press, Boca Raton, pp 609-628 Buddemeier RW, Ware JR (2003) Coral reef decline in the Cari bbean. Science 302:391392 Causey B, Delaney J, Diaz E, Dodge D, Garcia JR, Higgins J, Jaap W, Matos CA, Schmal GP, Rogers C, Miller MW, Turgeon DD (2000) Status of coral reef in the US Caribbean and Gulf of Mexico : Florida, Texas, Puerto Ri co, U.S. Virgin Islands and Navassa. In: Wilkinson C (ed) Status of Co ral Reefs of the World: 2000. AIMS, Cape Ferguson, Australia, pp 239-259 Chander G, Markham B (2003) Revised Landsat-5 TM radiometric calibration procedures and postcalibration dynamic range s. IEEE Transactions on Geoscience and Remote Sensing 41:2674-2677 Chavez PS (1988) An Improved Dark-Object Subtraction Technique for Atmospheric Scattering Correction of Multispectral Data Remote Sensing of Environment 24:459-479 Department of Commerce (DOC) NOAA (1998) Benthic Habitats of the Florida Keys NOAA's Ocean Service, National Centers fo r Coastal Ocean Scie nce (NCCOS), Silver Spring, MD Durand D, Bijaoui J, Cauneau F (2000) Op tical remote sensing of shallow-water environmental parameters: a feasibility study. Remote Sensing of Environment 73:152161 Gardner TA, Cote IM, Gill JA, Grant A, Wa tkinson AR (2003) Long-term region-wide declines in Caribbean corals. Science 301:958-960 Gordon HR (1997) Atmospheric correction of ocean color imagery in the Earth Observing System era. Journal of Geophysical Research 102:17081-17106
38 Grigg RW, Dollar, J. S, Huppert, Amit, Causey BD, Andrews, Katherine, Kruczynski, L. W, Precht WF, Miller SL, Aronson RB, Bruno JF, Kaufman, Les, Jackson JBC, Ogden JC, Pandolfi JM, Baron N, Bradbury RH, Gu zman HM, Hughes TP, Kappel CV, Micheli F, Possingham HP, Sala E (2005) Reassessing U.S. Coral Reefs. Science 308:1740c-1742 Hallock P (2001) Coral reefs, carbonate sedi ments, nutrients, and global change. In: Stanley GD (ed) The history and sediment ology of ancient reef systems. Kluwer Academic/Plenum Publishers, New York, pp 387-427 Hu C, Carder KL, Muller-Karger FE ( 2000) Atmospheric correction of SeaWiFS imagery: assessment of the use of alte rnative bands. Applied Optics 39:3573-3581 Hu C, Muller-Karger FE, Andrfout S, Card er KL (2001) Atmospheric correction and cross-calibration of Landsat-7/ETM+ im agery over aquatic environments: a multiplatform approach using SeaWiFS/MODIS. Remote Sensing of Environment 78:99107 Jupp DL, Mayo KK, Kuchler DA, Claasen DV, Kenchington RA, Guerin PR (1985) Remote sensing for planning and managing the Great Barrier Reef of Australia. Photogrammetria 40:21-42 Kirk JT (1994) Light and photosynthesis in aquatic ecosystems. Cambridge University Press, Cambridge Kirk JT (2003) The vertical attenuation coefficient of irradi ance a function of the optical properties of the water. Limnology and Oceanography 48:9-17 Klein CJI, Orlando SPJ (1994) A spatial framework for water-quality management in the Florida Keys National Marine Sanctuary. Bulletin of Marine Science 54:1035-1044 Koop K, Booth D, Broadbent A, Brodie J, Bucher D, Capone D, Coll J, Dennison W, Erdmann M, Harrison P, Hoegh-Guldberg O, Hutchings P, Jones GB, Larkum WD, O'Neil J, Steven A, Tentori E, Ward S, Williamson J, Yellowless D (2001) ENCORE: The effect of nutrient enrichment on coral reef. Synthesis of results and conclusions. Marine Pollution Bulletin 42:91-120 Lapointe BE, Barile PJ, Matzie WR (2004) Anthropogenic nutrient enrichment of seagrass and coral reef communities in the Lower Florida Keys: discrimination of local versus regional nitrogen sources. Journal of Experimental Marine Biology and Ecology 308:23-58 LaPointe BE, Clark MW (1992) Nutrient inputs from the watershed and coastal eutrophication in the Florid a Keys. Estuaries 15:465-476
39 Lee TN, Smith N (2002) Volume transport va riability through the Florida Keys tidal channels. Continental Shelf Research 22:1361-1377 Lee Z, Carder KL, Mobley CD, Steward RG Patch JS (1999) Hyperspectral remote sensing for shallow waters: 2. deriving bottom depths and water properties by optimization. Applied Optics 38:3831-3843 Lipp EK, Jarrell JL, Griffin DW, Lukasik J, Ja cukiewicz J, Rose JB (2002) Preliminary evidence for human fecal contamination in corals of the Florida Keys, USA. Marine Pollution Bulletin 44:666-670 Lyzenga DR (1981) Remote sensing of bottom reflectance and water attenuation parameters in shallow water using aircraft and Landsat data. International Journal of Remote Sensing 1:71-82 Maritorena S (1996) Remote sensing of the water attenuation in co ral reefs: a case study in French Polynesia. International Journal of Remote Sensing 17:155-166 Morel A, Maritorena S (2001) Bio-optical properties of oceanic wa ters: a reapprisal. Journal of Geophysical Research 106:7163-7180 Palandro D, Andrfout S, Muller-Karger FE, Dustan P, Hu C, Hallock P (2003) Detection of changes in coral reef co mmunities using Landsat 5/TM and Landsat 7/ETM+ data. Canadian Journal of Remote Sensing 29:201-209 Palandro D, Hu C, Andrfout S, Muller-Karger FE (2004) Synoptic water clarity assessment in the Florida Keys using diffuse attenuation coefficient estimated from Landsat imagery. Hydrobiologia 530-531:489-493 Pandolfi JM, Bradbury RH, Sala E, Hughes TP, Bjorndal KA, Cooke RG, McArdle D, McClenachan L, Newman MJ, Paredes G, Warner RR, Jackson JB (2003) Global trajectories of the long-term decline of coral reef ecosystems. Science 301:955-958 Pandolfi JM, Jackson JBC, Baron N, Bradbury RH, Guzman HM, Hughes TP, Kappel CV, Mitchell F, Ogfen JC, Possingham HP, Sa la E (2005) Are U.S. coral reefs on the slippery slope to slime? Science 307:1725-1726 Phinn SR, Dekker AG, Brando VE, Roelfsema CM (2005) Mapping water quality and substrate cover in optically complex coastal and reef waters: an integrated approach. Marine Pollution Bulletin 51:459-469 Pope RM, Fry ES (1997) Absorption spectrum (380-700nm) of pure water. II. Integrating cavity measurements. Applied Optics 36:8710-8723 Porter JW, Kosmynin V, Patterson KL, Port er KG, Jaap WC, Wheaton JL, Hackett K, Lybolt M, Tsokos CP, Yanev G, Marcinek DM, Dotten J, Eaken D, Patterson M, Meier
40 OW, Brill M, Dustan P (2002) Detection of co ral reef change by the Florida Keys Coral Reef Monitoring Project. In: Porter JWPaKG (ed) The Everglades, Florida Bay, and Coral Reefs of the Florida Keys An Ecosystem Source Book. CRC Press, Boca Raton, pp 749-769 Porter JW, Lewis SK, Porter KG (1999) The effect of multiple stressors on the Florida Keys coral reef ecosystem: A landscape hypot hesis and a physiological test. Limnology and Oceanography 44:941-949 Shinn EA, Lidz BH, Halley RB, Hudson JH, Kindinger JL (1989) R eefs of Florida and the Dry Tortugas Field Trip Guidebook T176. AGU, Washington D.C. 57 Smith NP (1994) Long-term Gulf-to-Atlantic transport through tidal channels in the Florida Keys. Bulletin of Marine Science 54:602-609 Smith NP (1998) Tidal and long-term exch anges through channels in the middle and upper Florida Keys. Bulletin of Marine Science 62:199-211 Smith RC, Baker KS (1978) The bio-optical st ate of ocean waters and remote sensing. Limnology and Oceanography 23:247-259 Smith VE, Rogers RH, Reed LH (1975) Them atic mapping of coral reefs using Landsat data. 10th International Symposium on Remo te Sensing of the Environment 1:585-594. Szmant AM, Forrester A (1996) Water colu mn and sediment nitrogen and phosphorus distribution patterns in the Florid a Keys, USA. Coral Reefs 15:21-41 Yamano H, Tamura M (2000) Can satellite se nsors detect coral r eef bleaching? A feasibility study using radiative transfer models in air and wa ter. 9th International Coral Reef Conference V:2199-2102.
41 CHAPTER THREE Quantification of coral reef habitat declin e in the Florida Keys National Marine Sanctuary determined from satellite data (1984-2002) Abstract The decline of coral reef habitats has been w itnessed on a near global scale, with some of the most dramatic decline taking place in the Florida Keys and the Caribbean. Remote sensing can provide a synoptic view of these habitats, a difficult task for in situ monitoring efforts attempting to detect ch anges over large areas. Previous remote sensing studies have focused on single reef sites or a small time series of images to detect change in the benthic community. Here, we utilize an 18-year time series of Landsat 5/TM and 7/ETM+ images as the basis to asse s changes in eight coral reef sites in the Florida Keys National Marine Sanctuary, namely Carysfort Reef, Grecian Rocks, Molasses Reef, Conch Reef, Sombrero Reef Looe Key Reef, Western Sambo and Sand Key Reef. Twenty-eight Landsat images ( 1984-2002) were used, with imagery gathered every two years during spring, and every six years during fall. The image dataset was georectified, calibrated to re mote sensing reflectance and corrected for atmospheric and water-column effects. A Mahalanobis distance classification was trained for four classes using in situ ground-truth data from 2003-2004 and us ing the spectral statistics from a 2002 image. The red band was considered useful for benthic habitats in depths less than 6m only. The classes are coral habitat sand, bare hardbottom and covered hardbottom. Results showed that there was no significant differenc e in coral habitat cover between paired spring-fall images (p aired t-test, p = 0.535). Overall mean coral habitat decline for all sites was 61% ( 3.4%/y), from 19% (19 84) to 7.7% (2002). In situ monitoring data acquired by the Coral Reef Evaluation and Monitoring Project (CREMP) for the eight reef sites be tween 1996 and 2002 showed a loss in coral cover of 52%
42 (8.7%/y), whereas the Landsat-derived coral -habitat cover declin ed 37% (6.2%/y) for the same time period. A direct trend comp arison between the full CREMP percent coral cover data (1996-2004) and the full Landsat-d erived coral habitat class showed no significant difference between the two time series (ANCOVA; F-test, p = 0.303, n = 32). Beyond these comparable trend results, I also more precisely investigated the classification results for the eight different si tes. A detailed pixel by pixel examination of the spatial patterns across time suggests that the classification results range from reliable and ecologically plausible, to spatially inc onsistent and ecologically improbable. Coral habitat pixels were found in the backreef r ubble zone, an area traditionally low in coral cover. This may be due to the zoanthid Palythoa spp., which thrives in shallow-water environments and appears optically identical to scleractinian corals in this habitat. Introduction There is a consensus among cora l reef scientists that co ral reefs worldwide are under multiple stresses (Pandolfi et al. 2003). In the Caribbean Sea (Buddemeier and Ware 2003) and Florida Keys (Dustan 2003), live cora l cover has declined markedly over the past 30 years. Repeated independent surveys for the shallow zone of Carysfort Reef show a decline in live coral cover of 90% between 1974 and 2000 (Dustan 2003). There are multiple stressors associated with co ral reef decline (Hughes and Connell 1999), including poor water quality (Boyer and Jones 2002), overfishing and changes in water temperature (above and below coral local th reshold) (Dustan 1999). These stresses may cause an increased frequency in coral dise ases (Patterson et al. 2002), bleaching (HoughGuldberg 1999) and algal overgrowth (Koop et al. 2001). Over time, there have been many calls for systematic monitoring programs to asses these issues over the large spatial extent of the Florida Keys National Ma rine Sanctuary (FKNMS) (Ogden et al. 1994, Dustan 1999, Murdoch and Aronson 1999). The FKNMS was established in 1990. Subs equently, the Environmental Protection Agency (EPA), the National Oceanic and Atmospheric Administration (NOAA) and the state of Florida established the Water Quality Protection Plan (WQPP) in 1995 to monitor water quality and benthic habitats (i.e., corals and seagrass) in the FKNMS. As part of
43 the WQPP, coral reef health has been monitored at 40 sites in the FKNMS under the Coral Reef Evaluation and Monitoring Proj ect (CREMP), formerly the Coral Reef Monitoring Project (CRMP) (P orter et al. 2002, Beaver et al. 2006). This dataset is unique and provides the basis fo r a comprehensive study of change in coral cover over time. CREMP uses permanently placed stakes at 40 different reef sites in the FKNMS to construct transects, which are revisited y early. The annual surv eys provide live coral percent cover by species, as well as the per cent cover of broader benthic categories (e.g., substrate, sponges, macroalgae). Beyond the CREMP data, a few change detection studies have been performed in the FKNM S (Dustan and Halas 1987, Porter and Meier 1992, Cockey et al. 1996, Dustan et al. 2001, Miller et al. 2002, Porter et al. 2002, Hallock et al. 2003, Palandro et al. 2003a, Pala ndro et al. 2003b). Of these, only two studies have looked at more than one site (Porter and Meier 1992, Porter et al. 2002). While the CREMP effort is significant and collects precise information, monitoring the entire FKNMS is simply not possible due to the sheer size of the sanctuary. Remote sensing technology has been used to map shallow-water ecosystems worldwide (Ahmad and Neil 1994, Mumby et al. 1997, Andrf out et al. 2006). In particular, the Landsat series of satellites carrying the Them atic Mapper (TM, Landsats 4 and 5) and the Enhanced Thematic Mapper Plus (ETM+, La ndsat 7) sensors are the longest running, continuous series of satellites that is useful for coral reef benthic cover studies. Landsat provides 16-day repetitive coverage for site s at a 30 m spatial re solution. Although the resolution is too coarse to identify coral sp ecies, TM and ETM+ data allow the study of overall benthic cover at the habitat level (Andrfout et al. 2 003). Landsat data have also been used to detect changes to coral habita t cover (Andrfout et al 2001, Dustan et al. 2001, Palandro et al. 2003b). The synoptic view historic dataset and repetitive coverage provided by satellites make remote sensing a useful tool to provide ongoing synoptic monitoring data on co ral habitat cover. Here an assessment of coral habitat change using an 18-year (1984-2002) time series of Landsat TM and ETM+ images for eight sites in the FKNMS is made. The results are compared with the percent coral cover es timates of CREMP (1996-2002). The primary
44 goal was to evaluate the utility of Landsat for such studies as a first step to more comprehensive regional assessments of cora l reef habitat cover using remote sensing tools. Methods Study Sites There are three major regions associated with the Florida Keys Reef Tract (Porter et al. 2002). Four sites were chosen for this st udy (Table 3.1, Fig. 3.1, Fig. 3.2) in the Upper Keys (Carysfort Reef, Grecian Rocks, Molass es Reef and Conch Reef), one site in the Middle Keys (Sombrero Reef) and three si tes in the Lower Keys (Looe Key Reef, Western Sambo and Sand Key Reef). The sites were chosen because they all fit a desired set of criteria, they are: 1) monitored by CREMP, 2) considered Sanctuary Preservation Areas (SPAs) and 3) classified by CREMP as Offshore Shallow, with reef crest depths less than 6m. The 6m threshold was selected to accommodate the use of all three Landsat bands. Only 10% of light the red band (630 nm 690 nm) can reach 5.6 m in depth, even in pure water (Kirk 1994, Pope and Fry 1997). The selected locations are representative of other FKNMS reefs and are representative of the reef habitat zonation found throughout the Caribbean Sea (Jaap and Hallock 1990). Although Acropora palmata (Elkhorn Coral) was once the major reef-building coral of Table 3.1. Reef site general lo cations and associated region. Reef Site Location Region Carysfort Reef 25.20 -80.25 Upper Grecian Rocks 25.10 -80.30 Molasses Reef 25.00 -80.42 Conch Reef 24.94 -80.49 Sombrero Reef 24.61 -81.09 Middle Looe Key Reef 24.55 -81.40 Lower Western Sambo 24.47 -81.75 Sand Key Reef 24.43 -81.92
Fig. 3.1. Location map of the eight reef sites used in this study. They are, from north to south; Upper Keys (white) Carysfort Reef, Grecian Rocks, Molasses Reef, Conch Reef: Middle Keys (green) Sombrero Reef: Lower Keys (yellow) Looe Key Reef, Western Sambo, Sand Key Reef. 45
Sand Key Reef We stern S am bo Looe Key Reef Sombrero Reef Molasses Reef Conch Reef Grecian Rocks Carysfort Reef 1 km E 46 Fig. 3.2. RGB images of each reef site with Sanctuary Preservation Area (SPA) extent overlaid in white. The area of interest used for the quantitative analyses of this study is outlined in black. Note that the full geographic extent of the SPAs for Carysfort Reef and Western Sambo have been reduced in size for display purposes.
47 the Florida Keys (Porter et al. 2002), that is no longer true. The remains of the A. palmata are largely low-relief rubble covered with turf algae ( Ceramium spp.). The current dominant live coral is Montastraea cavernosa (Great Star Coral) and Montastraea annularis species complex (Boulder St ar Coral). Other common hardbottom constituents include Millepora spp. (Fire Coral), gorgonians ( Gorgonia ventalina Sea Fan) and zoanathids ( Palythoa caribaeorum White Encrusting Zoanthid) (Jaap and Hallock 1990). There are two very similar Palythoa species found in the Florida Keys, P. caribaeorum and P. mammilosa. There are unresolved taxonomic i ssues related to these species, therefore they will be discussed jointly as Palythoa spp. (Haywick and Mueller 1997). Palythoa spp. are colonial cnid arians that harbor z ooxanthellae of the genus Symbiodinium within their cells, providing for the animals yellow-brown color (Haywick and Mueller 1997, Acosta 2001). Spectrally, Palythoa spp. are nearly identical to Acropora spp. and Millepora spp (Eric Hochberg, personal communication). Palythoa spp. thrive in shallow-water coral reef envir onments, and can dominate the reef crest as mats (Haywick and Mueller 1997). Their abundance in shallow-water hardbottom zones and similar spectral qualities to scleractinian coral make Palythoa spp. unique and an important aspect to this study that wi ll be discussed at length below. CREMP The process by which CREMP data are gathered a nd analyzed is described in Porter et al. (2002) and is only briefly outlined here. E ach CREMP site characterizes a single reef zone and is comprised of one to four stations located on the shallow (~6 m) or deep forereef (~15 m). Stations, demarcated by a pair of permanently placed stakes, consist of three parallel video transects spaced 0.6 m apart. Each tr ansect is approximately 0.6 m wide and approximately 22 m long. Video transects are analyzed with PointCount software to estimate benthic cover. The st ation-level CREMP per cent coral cover data were pooled to provide a site-l evel percent coral c over mean for all eight reef sites. These data were then compared to the Landsat-derived data.
48 In Situ Ground-truth Data Ground-truth data for this study were collect ed during two summer field seasons (20032004). Much of the data were collected by joining the CREMP team during their surveys. Using SCUBA, data were acqui red by extending a 90 m underwater transect line on a given bearing. At each 10 m in crement, a visual assessment provided qualitative information on the percent bent hic cover of four major classes (sand, hardbottom, seagrass and coral) (Table 3.2) Observations included benthic cover for 10m on either side of the measuring tape (i.e., 10 m along-tape x 20 m across-tape measure). Three 10 m increments were averaged to produce a single point of groundtruthed data. All transect line starting poi nts were geo-located using a hand-held GPS Table 3.2. In situ ground-truth benthic cover major classes and ancillary information. Major Class Sand Hardbottom Size Cover <30cm Gorgonian 30cm 1m Rhodophyta >1m Phaephyta pavement Chlorophyta Porifera Zoanthid Other Seagrass Density Type <30% Thalassia testudinium 30-70% Syringodium filiforme >70% Halophila baillonis Coral Height Type <30 cm Acropora spp. 30cm 1m Agaricia spp. >1m Diploria spp. Millepora spp. Montastrea spp. Porites spp. Siderastrea spp. Other
49 (Garmin GPS 12XL or Garmin GPS 76, dependi ng on field season) an d ship-borne GPS (Leica MX412). A total of 192 points were acquired in a haphazard sampling method in areas on or near specific CREMP sampling sites, covering 115,200 m 2 Image Processing Twenty-two (22) Landsat 5 TM and six La ndsat 7 ETM+ images were used for this study. Both sensors provide data in three disc rete visible spectral bands, specifically the blue (band 1, 450 nm 520 nm), green (ba nd 2, 520 nm 600 nm) and red (band 3, 630 nm 690 nm). The images have a nominal spa tial resolution, or pixe l size, of 30 m. The study spanned the period from 1984 to 2002, with individual images every two years for the local spring season (March-May) and every six years for the local fall season (September-November). Two scenes were requ ired to cover the Florida Keys Reef Tract (path/row 15/43 and 16/43). The images were georectified and calib rated to at-sensor radiance (L mW cm -2 m -1 sr -1 ) using calibration paramete r factors provided with each file (Chander and Markham 2003). An atmospheric correction was performed (see Ch apter 2, this dissertation). Briefly, the atmospheric reflectance over deep (z > 20m), clear water was derived by estimating a Rayleigh component using single-scattering ap proximations, and aerosol properties were estimated using information from the red band ("dark pixel") and th en extrapolated to other bands assuming white-aerosols (Hu et al 2001). The estimates of water-leaving radiance (L w ) were divided by an estimate of the downwelling irradiance ( E d ) at the ocean's surface in that spectral band to provide the remote sensing reflectance ( R rs = L w /E d sr -1 ) at each Landsat band (Kirk 1994). A water column correction was performed by deriving a diffuse attenuation coefficient ( K d m -1 ) from each of the Landsat images for bands one and two (Chapter 2). K d is an apparent optical property (Kir k 1994) that is used to char acterize how light propagates through the water column and is therefore a measure of wate r clarity (Smith and Baker 1978, Palandro et al. 2004). K d was derived for sand pixels w ithin each reef site using each image, in an effort to compensate for the spatial variance in water clarity observed
in the Florida Keys, specifically between regions (Klein and Orlando 1994). K d data was incorporated by using the equation: 50zKrsrsdeRzR2*)0()( (1) Where z is a bathymetry derived from combining a variety of digital bathymetric datasets (Chapter 2) and R rs (0) is the remote sensing reflectance just above the oceans surface and the factor is used to account for the attenuation in both the downward and upward directions (Maritorena at al. 1994). Due to the rapid attenuation of red light in water, no K d value for the Landsat band 3 could be derived (Palandro et al. 2004). Therefore the absorption coefficient (a, m -1 ) for pure water was used, 0.41 m -1 (Pope and Fry 1997). Finally, an empirical line calibration (ELC; Jensen 2004) was performed to account for interand intrasensor variability (Telliet et al. 2001). Normally used for an atmospheric correction when in situ spectroradiometer data are available, the ELC uses spectral data from optically bright (sand) and optically dark (seagrass) targets to calibrate an image. Here the ELC was utilized to further reference all historical images radiometrically to a more recent base image. This was possible as both sand and seagrass areas were found to be stable throughout the time period of Landsat imagery. The spring 2002 images were used as the base images. Image Classification Four classes were used to separate benthic habitat cover for the areas located in 6m water depth within and adjacent to each reef sites SPA demarcation. The training pixels per each class were derived from the in situ ground-truth data. The classes are coral habitat, sand, covered hardbottom and bare hardbottom. Each Landsat pixel covers a mix of different benthic constituents. Therefore the spectral signal received by the satellite sensor is a mixture of the signals from those different constituents. It is for this reason that I have selected the term coral habitat to best describe the class that includes the spectral signal inclusive of coral. A 30% threshold of actual live coral cover was used in
defining the lower threshold of the coral habitat class. In other words the ground-truth transect sections with 30% actual live coral cover were used to derive training pixels for the coral habitat class. Thirty percent (30%) was used due to the low level of live coral cover observed during field activities. The delineation between bare and covered hardbottom was set at 20% benthic cover. A transformed divergence class separability algorithm was performed to determine the level of overlap between the classes. All pixels used to train the classification scheme were binned into each of the four classes and then located and selected on the spring 2002 images. Using ENVI image processing software, an endmember spectral library was built. The spectral library contained minimum, maximum, mean and covariance data from Landsat bands 1-3 for each of the four classes. These values were then used to perform a classification on all of the images using the Mahalanobis Distance classification algorithm. The change detection was done by comparing the change in percent of coral habitat-classed pixels over time. These Landsat-derived coral habitat pixels were compared to percent coral cover data from CREMP, and other in situ studies. The Landsat-derived percent coral habitat cover is an estimate of coral habitat pixels, which can be a mix of different benthic constituents, over a large spatial extent (30 m). For CREMP, as well as other in situ studies, coral cover is an estimate of the presence of actual live coral cover over a smaller spatial extent. Rrs data from the spectral library data produced from the training pixels were averaged for each class and band. These provided three mean spectral points for the blue (485 nm), green (560 nm) and red (660 nm), for each class. Above-water Rrs values were converted to below-water reflectance (R, %) (Gordon et al. 1988) by: (2) 51100**54.0/rsRR and compared to species-level in situ R measurements derived by Hochberg and Atkinson (2000) for different coral reef components. Field data from this study were used to determine the mix (linear) of benthic constituents within each of the ground-truthed
52 training pixels, e.g., what per centage of covered hardbottom is found in a pixel used to define a sand training pixel? The benthi c field data were averaged for benthic constituents per training pi xel class (Table 3.3). The in situ R values were provided by E. Hochberg (UH, personal communication) and are the basis for the results published in Hochberg et al. 2003. These values were used to more accurately describe the spectral signal for the training pixel. This was done to establish the link between the optical characteristics of the classified benthic constituents and the Landsat-derived classification of those constituents. High spatial resolution IKONOS satellite im agery acquired in the spring of 2006 with 4 m pixel size and visibl e band wavelengths equivalent to Landsat were made available by NOAA after the classification and change dete ction analyses were completed for this study. IKONOS imagery for the eight reef site s were georectified to the Landsat data by using common recognizable points between the images and a GIS layer of navigational aids (e.g., lighthouse tower), which were de tectable in the IKONOS imagery. The IKONOS imagery provided a means to better se parate coral reef zones and was used to carry out a first order error analysis to determin e if pixels classified as coral habitat were present in ecologically pl ausible reef locations. Table 3.3. Mean ground-truth da ta for the visually estimated percentage of benthic constituents found in each set of derived training pixels per class (n = 192). Percent of benthic constituent present Training Pixel Class Sand Bare HB Covered HB Coral Habitat Sand 75.00 12.19 8.75 0.31 Bare Hardbottom 7.50 62.25 23.00 5.75 Covered Hardbottom 6.70 15.09 72.39 5.83 Coral Habitat 7.74 12.50 44.52 32.74
Results Comparison to In Situ Spectral Data Fig. 3.3 shows the mean Rrs values for the four Landsat-derived classes from the three Landsat bands. Sand has the highest value in the blue band (centered at 485 nm), green band (centered at 560 nm) and the red band (centered at 660 nm). Bare hardbottom has the second highest Rrs value for all three bands, followed by coral habitat and covered hardbottom, respectively. Landsat-derived R values were compared to the in situ R values (Hochberg et al. 2003) calibrated from the values derived in Table 3.3 for all four benthic classes (Fig. 3.4). There was a strong agreement between the two datasets for each class. Covered hardbottom, bare hardbottom and coral habitat showed the most similar data as certain 00.010.020.030.040.050.06485560660Wavelength (nm)Rrs (sr-1) coral habitat sand bare hardbottom covered hardbottom Fig. 3.3. Remote sensing reflectance (Rrs, sr -1 ), for each Landsat band wavelength mean, for all four classes based on training pixels utilized in spectral library. Error bars are 95% confidence intervals. 53
0510152025303540400425450475500525550575600625650675700Wavelength (nm)R (%) 0102030405060400425450475500525550575600625650675700Wavelength (nm)R (%) 051015202530400425450475500525550575600625650675700Wavelength (nm)R (%) 051015202530400425450475500525550575600625650675700Wavelength (nm)R (%) Fig. 3.4. Comparison of Landsat-derived reflectance (R, %) from the spectral library (black solid) to in situ R values (grey solid) based on values published by Hochberg et al. (2003). Actual spectral data were provided by E. Hochberg, (UH, personal communication) and integrated over Landsat spectral wavelengths for Sand, Bare Hardbottom, Covered Hardbottom and Coral Habitat, with plus/minus one standard deviation (dashed lines). 54
55 points intersected. Sand show ed the least similar data; where the Landsat-derived R values are all lower than that of the in situ R values. However, only the data at 560 nm (green) fell outside one standard deviation for both datasets. When the in situ R data were averaged around 485 nm (450-520 nm), 560 nm (520-600 nm) and 660 nm (630690 nm) for each of the four classes a nd compared to the Landsat-derived R at these wavelengths R 2 = 0.891, showing a significant correlation between the R from each dataset. Classification Analyses An analysis was performed on the 30% fieldacquired coral habitat threshold. The field data from this study were compared to CREMP in situ data for each reef site. The mean values from both datasets were plotted agains t each other and the derived slope and offset were applied to this study s field-acquired 30% threshold value of coral habitat cover (i.e., normalization). The normalized field-acq uired coral habitat threshold was reduced to 22% (R 2 = 0.727), showing that there was an ove restimate of live coral cover by the in situ visual-estimate method. The transformed divergence (TD) class sepa rability data provided a mean value of 1.91 for all sites, years and classes. TD is an empirical measure of real values between 0 and 2, where 0 indicates complete class overlap and 2 indicates complete class separation; therefore a TD value of 1.91 denot es a near complete separati on of classes. The larger the class separability, the more accurate the classification results (Richards 1986). Certain reef sites performed better than othe rs, with a range of values from 1.84 (Sand Key Reef) to 1.99 (Sombrero Reef ). Between classes, the highest mean TD value was between sand and coral habitat (1.99) (i.e., the two classes were the most separate). The remaining mean TD values, from high to low, were sand and covered hardbottom (1.98), sand and bare hardbottom (1.96), covered ha rdbottom and coral habitat (1.93), bare hardbottom and coral ha bitat (1.90) and finally bare ha rdbottom and covered hardbottom (1.70).
56 The 2002 Landsat-derived coral habitat pixel locations were overlaid on top of the 2006 IKONOS image data (Fig. 3.5). From this co mparison certain reef sites displayed coral habitat pixels that were clearly misclassified. Specifically, Conch Reef had two pixels located in an area of sand and seagrass to the south. Sombrero Reef showed three individual coral habitat pixe ls some distance behind the r eef crest along the sand, bare hardbottom and covered seagrass channel that runs from the southeast to the northwest. The IKONOS imagery, due to its higher spatial resolution, provided a means to more accurately delineate the backreef rubble zone from the reef crest and forereef zones. There are Landsat-derived coral habitat pixels located in the backreef rubble zone of the eight reef sites for the full time series of Landsat classified image data. The backreef zone has historically low concentrations of live stony coral (Wheaton and Jaap 1988); therefore it is unlikely that there is a significant amount of live coral present in these images. However, Palythoa spp. can exist in abundance in this zone and may be the cause for the coral habitat pixel classification. Seasonal Variability Image pairs from the spring and fall for 1984, 1990, 1996 and 2002 were analyzed for seasonal variability in percent coral habitat. Where paired reef site values were present (n = 20), there was no significant difference in Landsat-derived per cent coral habitat due to season (paired t-test, p = 0.535). This being the case, th e spring and fall percent coral habitat values have been averaged to derive a single value (Table 3.4). Change Detection A total of 89 reef site imag es over time provided usable im age data, out of a possible 112 opportunities (8 reef sites x 14 images per re ef site). The 23 missed image opportunities were due to the presence of cloud cover over pa rticular sites. The sequence of classified images for each reef site, which was used in the change detection analysis, is shown in Fig. 3.6.
Fig. 3.5(a). Spring 2002 Landsat images (left) with appropriate 2006 IKONOS images (right) for the eight reef sites, namely: Carysfort Reef (a), Grecian Rocks (b), Molasses Reef (c), Conch Reef (d), Sombrero Reef (e), Looe Key Reef (f), Western Sambo (g) and Sand Key Reef (h). The white line delineates the area of interest from this study and the black line displays coral habitat extent. 57
Fig. 3.5b Grecian Rocks 58
Fig. 3.5c. Molasses Reef 59
Fig. 3.5d. Conch Reef 60
Fig. 3.5e. Sombrero Reef 61
Fig. 3.5f. Looe Key Reef 62
Fig. 3.5g. Western Sambo Reef 63
Fig. 3.5h. Sand Key Reef 64
Table 3.4. Percent of Landsat-derived coral habitat cover by location and region per year (fall and spring values have been averaged where possible), as well as percent change from 1984-2002. (*Note that Western Sambo percent change is from 1988-2002) Year Reef Site 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 % Change Carysfort Reef 20.0 19.1 17.7 15.6 13.9 13.0 10.0 8.3 6.5 5.7 -71.5 Grecian Rocks 32.9 28.0 29.4 26.6 23.8 23.1 21.4 17.5 14.7 13.9 -57.8 Molasses Reef 19.1 17.8 16.5 15.4 13.7 12.7 11.9 9.2 7.8 -59.4 Conch Reef 10.1 9.5 9.1 7.6 8.2 6.9 5.7 4.3 3.4 -66.8 Upper Keys 20.5 18.6 18.2 16.3 14.9 18.1 12.7 10.9 8.7 7.7 -62.6 Sombrero Reef / Middle Keys 7.8 7.2 7.0 7.1 6.5 6.1 4.8 4.5 -42.9 Looe Key Reef 20.3 20.4 19.0 16.2 14.7 14.7 14.1 12.5 -38.4 Western Sambo 21.5 20.3 18.0 15.9 14.6 8.6 6.1 -71.9* Sand Key Reef 19.5 13.4 14.2 15.0 11.8 11.0 9.8 7.7 7.7 -60.5 Lower Keys 19.9 20.4 17.5 17.8 16.4 14.1 12.9 12.8 8.2 8.8 -59.0 65
E 100m 1984 1986 1988 1990 1992 Fig. 3.6(a). Complete classified dataset for spring images for Carysfort Reef (a), Grecian Rocks (b), Molasses Reef (c), Conch Reef (d), Sombrero Reef (e), Looe Key Reef (f), Western Sambo (g) and Sand Key Reef (h). Classification color codes are: Red = coral habitat, Brown = covered hardbottom, Yellow = bare hardbottom and Green = sand. 1994 1996 2002 1998 2000 66
E 1984 1986 1988 Fig. 3.6b. Grecian Rocks 1990 1992 1994 1996 100m 1998 2000 2002 67
E 1984 Fig. 3.6c. Molasses Reef 68 1986 1988 1990 1992 1996 1998 200m 2000 2002
E 200 1984 1986 1988 1990 1994 1996 1998 2000 2002 69 Fig. 3.6d. Conch Reef
E 1984 19861990199419922000 2002 200m 70 Fig. 3.6e. Sombrero Reef
1984 19861990199619982002 19921994 200m E Fig. 3.6f. Looe Key Reef 71
Fig. 3.6g. Western Sambo 1988199219961990199820002002 200m E 72
1984 1988 199019961998 20021992 1994 2000 100m E 73 ig. 3.6h. Sand Key Reef F
74The total mean Landsat-derived percent coral habitat data for all reef sites was 7.7% in 2002 down from 19% fol ha r 1984 (adjusted). These values are a mean of the Landsat-derived percent corabitat data from each and 1984, respectively. The adjusted mean value was due to no image data being available for Western Sambo for 1984. An adjustment was deemed necessary as this reef has the second highest percent coral habitat in 1988 (the first year that data were available). The adjustment was made by assuming a linear trend could be used to extrapolate coverage back to 1984. The 18-year decline for all reef sites was 61%, with an average of 3.4% loss per year. Every time period (2y), except 1986-1988, showed a decline in percent Landsat-derived coral habitat. The percent decline per time period ranged from 2.2% (1994-1996) to 32% (1998-2000). The Lower Keys had the highest percent coral habitat throughout the study, followed by the Upper and Middle Keys, respectively. Grecian Rocks in the Upper Keys started with the highest percent coral habitat in 1984 with 33% and also had the highest percent coral habitat cover with in 2002 with 14%. The lowest percent coral habitat in 1984 was found at Sombrero Reef (Middle Keys) with 7.7%, but was only the second lowest in 2002 (4.5%). The lowest percent coral habitat in 2002 was Conch Reef (Upper Keys) with 3.4%. It is interesting to note that the sites with the highest and lowest percent coral habitat are both located in the Upper Keys. The Upper Keys reef sites yielded the greatest decline with 63% (3.5%/y), followed by the Lower Keys with 59% (3.3%/y) and finally, the single Middle Keys reef site with 43% (2.4%/y). The decline in the Upper Keys was stable among the sites, with only 12% change variability. Western Sambo showed the greatest loss in coral habitat at 72% (1988-2002) and Looe Key Reef exhibited the least change at 38% (1984-2002). Again, it is interesting to note that the reef sites with the highest and lowest percent coral habitat decline are both located in the same region, in this case the Lower Keys. reef site for 2002
75Change Detection Im As a means to asses the validity of the change detection study, an analysis was made based solely on the visual interpretation of the Landsat-derived classified image data (Fig. 3.6). The interpretation was based upon the likelihood of the logical ecological progression and spatial consistency and coherence for each of the four benthic classes. In other words, the goal was to determine whether the changes detected were ecologically feasible or due to a misclassification. This was a qualitative assessment meant to determine if certain images within a time series need to be reviewed with caution. Carysfort Reef The time series of Carysfort Reef (Fig. 3.6a) showed some areas of little to no change in most of the images. Thee areas included a constant sand patch in the northwest, the coral habitat clustered in the southern center, bare hardbottom area south of the sand patch, a second bare hardbottom area west of the coral habitat cluster and covered hardbottom along the eastern edge of the reef site. Certain images lack all or part of these common threads and tended to have a more heterogenic appearance. Specifically, the 1986, 1990 and 1998 images had a lack of continuity of coral habitat class. The class was more disperse for 1986 and 1990, extended too far south in 1990 and the central area had widened by 1998. The bare hardbottom area south of the northwest sand patch was greatly altered in the 1988, 1990 and 1998 images, where sparse coral habitat or sand is present. The bare hardbottom area to the west of the central coral habitat tends to be more of a mixed area (sand and covered hardbottom) in the 1990, 1992 and 1998 images. The eastern edge of covered hardbottom was altered in the 1986, 1990 and 1998 images, displaying a mix of bare hardbottom and sand. Overall, the 1986, 1990 and 1998 images lacked the consistency displayed by the other images within the time series; therefore the classification for these images must be used with caution. Grecian Rocks The time series of Grecian Rocks (Fig. 3.6b) also had a central coral habitat area, with the surrounding areas classified as a mix of bare and covered hardbottom. The coral habitat progressed from nearly the full width of the central area to just the central western age Progression s
76 wo here are several areas located on Molasse s Reef (Fig. 3.6c) that are consistent covered hardbottom, except in the 1992 and 998 images, which had a mix of bare hardbottom and sand in this area. There was a tant sand patch the northeast and southwest, as well as bare hardbottom around each of the two coral abitat areas. Covered hardbot tom surrounded much of the reef site, specifically in the orth, east into the center. The coral habita t areas remained tightly clustered with the portion of the reef site. The coral habitat became more disparate in the 1988 image with areas of bare hardbottom mixed into the central region and the 1998 image showed t distinct coral habitat areas. The northern section was a mix of sand (western half) and bare and covered hardbottom in all of the images, making it difficult to assess its consistency. However, the 1994 image was the only image that displayed covered hardbottom at the northwestern most tip. The southern section was dominated by covered hardbottom in most of the imag es, except in the 1996 image, which was dominated by bare hardbottom. There was a progression through the time series that showed a lack of consistency mainly in the 1988, 1996 and 1998 images. Molasses Reef T throughout the time series. There is a central southwest to northeas t coral habitat area, covered hardbottom areas in the northwest, s outh and southeast, and a sand area in the west. The coral habitat ar ea showed a wide heterogeneity for the 1984, 1992, 1998 and 2002 images. The western sand area was inconsistent for the 1986, 1990 and 1996 images, and was actually classified as bare or covered hardbottom. The northwest covered hardbottom area is inc onsistent only in the 1984 image, whereas the south area was uniform throughout the time series. The southeast area, which ran parallel to the coral habitat area, was solidly dominated by 1 progression for Molasses Reef with inconsis tencies from the 1992 and 1998 images that may make these data suspect. Conch Reef The time series on Conch Reef (Fig. 3.6d) was the most stable among the reef sites and showed a logical progression of the loss of coral habitat, wh ich occurred in two areas; one centrally in the north and one centrally in the south. There was a cons in h n
77 s, where th e latter displayed the southern coral rea time d n that eterogeneous pattern throughout the tim e series, specifically in the 1992, 1994, 1996 s. There were two uniform cove red hardbottom areas located in the north uth exception of the 1984 and 1998 image habitat area as two separate areas. There was a lack of a coherent northeastern sand a in the 1994 image, whereas the southeaste rn sand area was missing in the 1986 image (both were classified as ba re hardbottom). The 1994 imag e lacked the eastern covered hardbottom, which was replaced by a mix of sand and bare hardbottom. The 1984, 1986 and 1990 images had a mix of sand and covere d hardbottom for the same area. Because of the inconsistencies observe d, the 1994 image should be used with caution in this series. Sombrero Reef There were only seven time series images for Sombrero Reef (Fig. 3.6e). The coral habitat was clustered in the s outhern area with a mix of vary ing levels of sand, bare an covered hardbottom surrounding it. There was a dense covered hardbottom in the northeast, with an area also in the northwest. Bare hardbottom dominated the regio runs from the coral habitat area northwest to the end of the reef site. The 1984 and 1990 images displayed a number of coral habitat pi xels outside of the southern cluster and were also the most heterogeneous. The 1986 image showed a sand-dominated central region that was dominated by ba re hardbottom in the other images, as well as sand along the northeastern edge of the reef site. The suspect images in this time series were the 1984, 1986 and 1990 images. Looe Key Reef The Looe Key Reef (Fig. 3.6f) time series has a distinct sand area in th e west of the reef site. That area was interspersed with c overed hardbottom in the 1998 and 2002 images. The coral habitat extended along the southern edge, with small clus ters that ran along a central southeast to northwest area. Co ral habitat pixels al so possessed a highly h and 1998 image and west for the time series. However, the 1996 image lacked the northern covered hardbottom area, whereas the 2000 image lacked the western area. The central southeast to northwest area was dominated by bare ha rdbottom and extended farther to the so
78 g. 3.6g) habitat was a ix of covered hardbottom and sand, which showed a progression from sand dominated bottom dominated. As both ar eas of coral habitat declined through time, habitat at ries. e eef (Fig. 3.6h) included a single clus ter of coral habitat cated in the southern centra l area, a covered hardbottom dominated area to the east and re hardbottom area to the north. The west was either sand dominated (1984, upon the decline of coral hab itat. This was not the case for the 1996 and 1998 image, which displayed a mix of covered hardbottom and coral habitat instead. The 1996 and 1998 images showed the least consistency for this time series and should be used with caution. Western Sambo There were two dense areas of coral habitat detected from the Western Sambo (Fi time series. The larger coral habitat area wa s located in the southwest and the smaller coral habitat was located in the southeast. The large area behind th e coral m to covered hard the appearance of bare hardbottom increased in those areas. In the western coral area, the bare hardbottom increased in size in the center, thereby spli tting the coral habit area in two, northern and southern. The sma ller eastern coral hab itat area decreased as bare hardbottom increased to the south. The 1990 and 2000 images had coral habitat pixels in the large area to the north, inconsistent with the other images in the time se Also, the 2002 image showed sand along the sout hernmost edge of the reef site. Th 1998 image had bare hardbottom in the northea st, at the same location where the other images showed a mix of sand and covere d hardbottom. The 1990 and 2000 were the least consistent images within this time series. Sand Key Reef The time series of Sand Key R lo a sand and ba 1988, 1994, 1996 and 2002) or covered har dbottom dominated (1990, 1992, 1998 and 2000). There was also an area of bare hardbotto m in the north central of the reef site. However, this area was not observed or greatly reduced in si ze in the 1992 and 1998 images. The 1984 image was inconsistent in the northern area, displaying coral habitat pixels as well as having the most disparate central coral habitat area, making this image suspect.
79e p. where Palythoa spp. usually thrives. MP stony coral cover data for the selected reef sites showed a loss of a Comparison to CREMP CREMP data include monitoring data for zoanthid cover (e.g., Palythoa spp.). These data were analyzed but not used in combination with the CREMP coral cover data in thanalyses below for two reasons. First, CREMP data show no net change in Palythoa sppercent cover between 1996 and 2004 (Fig. 3.7). Average values range from 4.1% (1996) to 4.2% (2004), with a range of 4.1% (2001) to 5.1% (2002). Second, CREMP stations are not located in and do not include any information on the shallowest zone of the reef The pooled CRE 52% (8.7%/y) between 1996 and 2002, whereas the Landsat-derived coral habitat datshowed a decline of 37% (6.2%/y) for the same time period. CREMP data had the 051996199719981999200020012002200320Year 1004cen 1520t Cover Per total coral zoanthid Fig 3.7. CREMP percent cover data for live coral, zoanthid (Palythoa spp.) and total (coral + zoanthid), for the eight selected reef sites.
80 ys wed the greatest decline per time period (2y) between 1998 and 2000 (CREMP 37%, Landsat = 32%). The least decline was 2000-2002 for both data sets (CREMP = 3.7). Percent co ral cover data and Landsat-der ived percent coral habitat nd st percent changed), two sites differed by ne rank and two other sites differed by two ra nks (Table 3.5). The two datasets were or the four reef sites that exhibited the greatest and the four reef sites that as eef sites where the slopes (i.e., rate s of change) were significantly different (Fgreatest decline in the Middle Ke ys, and systematically lower declines in the Upper Ke and Lower Keys, respectively. This differed fr om the Landsat derived coral habitat data in which the single Middle Keys reef site examined had the least decline. Both time series sho = 2.4%, Landsat = data matched between the two datasets (2002) : the highest percentage was found in the Lower (CREMP = 9.5%, Landsat = 8.8%), Up per (CREMP = 5.3%, Landsat = 7.7%) a Middle Keys (CREMP = 3.2%, Landsat = 4.5%), respectively. For all concurrent data points (n = 32), ther e was no significant diffe rence (paired t-test, p = 0.468) between the CREMP percent coral cove r and the Landsat-derived percent coral habitat. A correlation analysis between the two datasets provided the same result, R 2 = 0.704 (Fig. 3.8). It was difficu lt to rigorously compare CREMP and Landsat data for each reef site as there were only four concurrent points (1996, 1998, 2000 and 2002). Instead the focus was placed on the percent coral cover (CREMP) versus percent coral habitat (Landsat) trend over time for each reef site. Relative percent change rank between data sets, i.e., how each reef site co mpared to the others in terms of percent change over time, was examined. This comp arison showed that four sites agreed in relative ranking (including the highest and lowe o also consistent f exhibited the least change. A trend comparison was also made for the full timelines of each dataset; that is, Landsat coral habitat 1984-2002 and CREMP coral cove r 1996-2004 (Fig. 3.9). An analysis of covariance (ANCOVA) was used to compar e the regression lines for homogeneity of slope (Sokal and Rohlf 1981). For data from a ll eight reef sites taken together, there w no significant difference betw een Landsat-derived change in coral habitat and CREMP change in coral cover (F-test, p = 0.303). Sombrero Reef and Sand Key Reef were the only two r
81 y = 0.6109x + 4.1094 R2 = 0.704, n = 32 10 15 20 25oral Habitat (Landsat) 0 5 0 5 10 15 20 25 % Coral Cover (CREMP)% C Fig. 3.8. Correlation of Landsat percent coral habitat versus CREMP percent coral cover for all concurrent data poin ts (1996-2002) in the Upper (!), Middle ( 0 ) and Lower () Keys, R2 = 0.704 (n = 32). Table 3.5. Relative ranking of reef site s between percent CREMP coral cover and Landsat coral habitat change (1996-2002). Ranking is highes t percent change (1) to lowest (8). CREMPReef Site Landsat 1 Western Sambo 1 2 Molasses Reef 4 3 Carysfort Reef 3 4 Conch Reef 2 5 Sand Key Reef 5 6 Sombrero Reef 7 7 Grecian Rocks 6 8 Looe Key Reef 8
82 Fig. 3. 9. Percent coral habitat for Landsat (!) and CREMP (/) full-timeline data (yaxis) per year (x-axis) with linear regressi on line added for each da taset, for selected Florida Keys reef sites. Carysfort Reef0 5 10 15 20 25 19841986198819901992199419961998200020022004 Grecian Rocks0 5 10 15 20 25 30 35 40 1984198619881990199219941996199820002002200 4 Molasses Reef20 25 0 5 10 19841986198819901992199419961998200020022004 15 Conch Reef8 2004 0 2 4 6 1984198619881990199219941996199820002002 10 12 Sombrero Reef0 1 2 3 4 5 6 7 8 9 19841986198819901992199419961998200020022004 Western Sambo0 0 5 10 15 19841986198819901992199419961998200020022004 2 25 Sand Key Reef 0 5 10 15 2 0 2 5 19841986198819901992199419961998200020022004 Looe Key Ree0 5 10 15 20 25 1984198619881990199219941996199820002002 f2004
83test, p = 0.001 and p = 0.034, respectively). Grecian Rocks was found to have the most similar slopes (F-test, p = 0.967), whereas Western Sambo had the least similar slopes compared to CREMP (F-test, p = 0.059). Discussion Seasonal Variability No significant variability in percent coral habitat between paired fall and spring images from the same year was detected. A significant negative change on the scale detectable by Landsat would require a natural (e.g., hurricane) or man-made (e.g., ship grounding) catastrophic event (Porter and Meier 1992). No significant weather events or ship groundings occurred during the five to six month intervals between any of the paired fall-spring images studied here (1984, 1990, 1996 and 2002). Although coral spawning does take place during the interval between images (August), the growth rates would not ccount for changes at the habitat scale in such a short period of time. NOS imagery from NOAA provided for a more accurate eans to separate the backreef, reef crest and shallow forereef (Fig. 3.5). Pixels ified as coral habitat from the Landsat data exist in the backreef, an area that heaton and Jaap (1988) found whole and fragmented colonies of Acropora palmata and A. cerv the backre Key Reef, which they hypothesized were likely transported inshore from ef. If conditions are favorable, transported pieces of coral can thrive hin veneer of Porites astreoides was also found in the bareef of Looeaton and p 1988). A survey performed at Carysfort ef in 1981colonies f four different species of stony corals on the shallow (90 m) reef flat (Dustan and Halas 1987). A. palmata reefts were deset al. (1989) for Grecian Rocks. owever, most of the A. palmata has undergone decline in recent years (Eugene Shinn, ersonal communication). a Coral Habitat Pixels in the Backreef The late availability of IKO m class traditionally has very low live coral cover. Unfortunately, very few studies extend transects or perform monitoring in the backreef rubble zones of the Florida Keys. However, W icornes in ef of Looe the shallow forere in this zone. A t ck e h Key Reef (W Jaa Re 1982 found coral rom (~1 m) extende d fla cribed by Shinn H p
84Palythoa spp. are common in, and can dominate, shallow-water hardbottom environments (e.g., reef crest and reef flat) in the Florida Keys and throughout the Caribbean (Haywick and Mueller 1997, Acosta 2001). Dustan and Halas (1987) documented Palythoa spp. in ten of eleven transects in waters from 0-4 m at Carysfort Reef. Although CREMP data showed almost no change in zoanthid cover for Carysfort Reef between 1996 (2.7%) and 2004 (4.1%) for Carysfort Reef, Dustan and Halas (1987) found an overall increase of 85% in coverage between 1975 and 1982. Surveys performed at Looe Key Reef in 1983 found an abundance of Palythoa spp. in waters ranging from 0-7 m in depth (Wheaton and Jaap 1988). The abundance of Palythoa spp., accompanied by Millepora complanata, in this zone led to the designation Millepora/Palythoa zone. In fact, Palythoa spp. accounted for 11% of all cnidarians sampled in depths ranging from 0-11 m by Wheaton and Jaap (1988). As Palythoa spp. have essentially the same spectral signal due to its incorporation of zooxanthellae and the fact that it thrives in the shallowest zonation of coral reef environments, it is possible that at least part of the coral habitat classification witnessed in the backreef may be due to Palythoa spp. Acquisition of quantitative field data for the backreef is required to test this hypothesis. It is also possible that some misclassification may have occurred in the backreef. The class separability measure of TD between bare hardbottom, which is the class traditionally associated with the backreef rubble zone, and coral habitat was 1.9. This TD value denotes that the classes are statistically separable; however 1.9 is on the lower end of this scale (Jensen 2004). Progression of Coral Habitat Decline and Class Separability By performing a visual qualitative assessment on the progression of the classified change detection images, it was possible to infer the likeliness of misclassification and relevance of specific images within the dataset. Understanding the separability between the classes be attributed to the similarity between the two classes; bare hardbottom has much of the provided insight as to why misclassifications may occur. The highest occurrence of likely misclassification was between covered hardbottom and bare hardbottom. This can
85 f misclassification occurred between sand and are hardbottom. This appears logical as these two classes posses the highest albedo lso, among the three classes not sand, bare hardbottom had the lowest TD nd. On the opposite end, there were no observed occurrences of e reef Landsat path 16 ages had no single year that produced que stionable results between both reef sites bitat line loss of t ent e, ral habitat at reef sites, but none were significant. same benthic constituents, just less spatial c overage of them. Also, the TD between the two classes was the lowest, which may result in misclassification between the two classes. The second highest likely event o b (Fig. 3.4). A value in relation to sa coral habitat and sand misclassifications. The assessment also showed that certain image years provided consistently questionable results. For Landsat path 15, the 1998 image resu lts were suspect for f our of the fiv sites covered by that image (no image data we re available for Conc h Reef due to cloud cover). The 1998 image data was not unique in the context of the time series, but this image did have the highest K d variance among reef sites a nd did require the greatest water column correction for the Upper and Lowe r Keys reef sites. The im covered. Change in Coral Habitat Cover The dramatic decline in the number of pixels classified as Landsat-derived coral ha described in this study is not novel. Florida and Caribbean reefs have exhibited dec for many years (Pandolfi et al. 2003) What is novel is the quan titative measure of coral habitat gathered over a regular time pe riod (2y) and over an 18-year timeline. Only one time period showed an increase in coral habitat when the data from the eight reef sites were combined, 1986-1988. This result is artificial and can be attributed to the introduction of image data for Western Sambo in 1988, when no image data were available for 1984 or 1986. Western Sambo was classified as havi ng the second highes percent coral habitat (22%) in 1988, which skewed the da ta to an overall higher perc coral habitat for that year, in comparison to 1986. If the linear regr ession derived earlier for Western Sambo to hindcast coral habita t cover for 1984 and 1986 was utilized abov the change between 1986 and 1988 would be nega tive. There were also six individual instances of increases in co
86 he Upper toric tributed h ernmost eys s with the of on ted for 47.6% of the to tal variance (Murdoch and Aronson 1999). 2) e Key eaton and Jaap 1988). Porter and Meier (1992) showed verage loss on Looe Key Reef to be 29% for all coral species between 1984 and 1991, ecline in coral cover of 26 MP. The Lower Keys sites showed the highest percent coral habitat, followed by t and Middle Keys sites, respectively. This obs ervation is in agreement with both his (Shinn et al. 1989) and current in situ studies (Beaver et al. 2006) and has been at to the relative influence of Florida Bay on the three regions. The outfl ow of nutrient-ric and turbid waters from Florida Bay has cont ributed to lower coral cover (Shinn et al. 1989, Porter et al. 1999) at Sombrero Reef (Middle Keys) and Conch Reef (Upper Keys) with 3.4% and 4.5% coral habitat, respectively. Conch Reef is this studys south Upper Keys reef site and may be more affect ed by Florida Bay than the other Upper K reef sites. Grecian Rocks (Upper Keys) a nd Looe Key Reef (Lower Keys) display the highest coral habitat, 14% and 13%, respectively. The fact th at the reef site highest and lowest percent coral habitat reside in the same region (Upper Keys) demonstrated considerable variability and pr ovided no clear rule about coral reef health among the three major geologic regions of th e Florida Keys. Further, Murdoch and Aronson (1999) found that studying coral cover at different spatial sc ales led to varying estimates of coral cover. They found, base d on the Klein and Orla ndo (1994) division regions, that among-reef site variation accounted for 34.8% of the total and among-regi variation accoun Looe Key Reef had the lowest percent decline in coral habitat (38%). Miller et al. (200 analyzed loss in acroporid coral species ( A. palmata and A. cervicornis ) on Loo Reef from 1983-2000 and found a decline of 93% and 98%, respectively. However this is solely for the A. palmata and A. cervicornis which were already scarce in 1983 (6.8% and 2.7%, respectively) (Wh a with an average loss per year of 4.1%. CREMP data show a d % (3.3%/y) for Looe Key Reef between 1996 and 2004. Both of these coral cover decline rates are comparable to our results for Landsat-derived coral habitat of 2.1%/y. Carysfort Reef is the other reef site with cha nge detection data other than that of CRE In situ studies found a decline at Carysfort Reef averaging from 20% (2.9%/y) for 19841991 (Porter and Meier 1992) to 72% (9.0%/y) for 1996-2004 (CREMP) to 90%
87 l % ed alibrated from 30%). Although each of these studies employed different means to er les the reef sites but smaller than the regions (Murdoch and Aronson 1999). For xample, an environmental factor (e.g., se dimentation) may affect a small geographic sites locat ed within a geologic region that includes ntative rwent the (3.5%/y) for 1974-2000 (Dustan 2003). Remote sensing studies on Carysfort Reef estimate coral habitat decline at 89% fo r 1981-2000 (4.7%/y) from Ikonos and aeria photography (Palandro et al. 2003a) and 88% (5.5%/y) for 1984-2000 from Landsat (Palandro et al. 2003b). This st udy estimated coral habitat loss at Carysfort Reef at 72 (4.0%/y). The different results obtained fr om the two Landsat studies are due to the different thresholds used to define the coral habitat class. Palandro et al. (2003b) utiliz a coral-dominated pixel (i.e., > 50% coral) whereas this study utilized the 22% (c derive an estimate of coral cover over different spatial and temporal scales, the results are comparable. The high variability in coral habitat cover betw een reefs sites within one region is furth demonstrated by the change in percent coral habitat for the Lower Keys. While Looe Key Reef had the least change, Western Sa mbo had the highest pe rcent decline (72%) from 1988-2002. The difference may be due to environmental gradie nts acting on sca larger than e area that encompasses four reef twelve total reef sites. Water clarity data at the spatial scale of th e reef site also show wide variability in the diffuse attenuation coe fficient, a measure of water clarity (Chapter 2). Based on the assumption that poor wate r quality input from Florida Bay leads to greater decline in coral cover, Looe Key Reef due to its location, would be expected to be the most affected reef site. However, th e reef site with the s econd lowest decline in coral habitat was the single Middle Keys r eef site, Sombrero Reef (42.9%), which is directly influenced by Florida Bay. However, th is one reef site may not be represe of the entire Middle Keys. Comparison to CREMP From 1996-2002, Landsat-derived coral habitat data from this study revealed that the greatest percent loss occurred at the Upper Keys sites, followed by Lower and Middle Keys sites, respectively. CREMP data showed that the Middle Keys site unde
88 (37%) od he concurrent coral cover data betw een CREMP and this study yielded an R2 of 0.704, ve e at the changes taking place at the fine sp atial resolution of CE MP were also taking greatest change, followed by the Upper and Lo wer Keys sites, respectively. The data used for the Middle Keys in this study come s from only one reef s ite (Sombrero Reef), and, as mentioned above, may not be representa tive of all Middle Keys sites. However, when the entire CREMP dataset was analy zed, the same percent loss pattern as the Landsat data emerged, i.e., the greatest lo ss was observed in the Upper, Lower and Middle Keys, respectively. These results are intuitive, as the Middle Keys historically have the least coral cover to lose (Shinn et al. 1989). The single greatest decline per time period fo r both datasets occurred between 1998 and 2000. This can be attributed to extensive bleaching, and subsequent mortality, which occurred during the summer of 1998. The ra tes of decline derived from CREMP are comparable to that of Landsat (32%), showing that both techniques were able to detect this significant event. Both datase ts also identified 20002002 as the time peri of least change (Landsat, 4% and CREMP, 3%). Both monitoring methods also provided a means to detect relatively small changes to the coral reef habitat at different spatial scales. T showing a good agreement between the datase ts. This indicates that, although the information was acquired at different spatia l scales, they were comparable across the scales of measurement, which suggests that th e different datasets are representative of the reef site as a whole. It is possible to normalize the Lands at-derived data to the CREMP data using the 2002 data values. If this were performed, R 2 would equal 0.905 for all eight reef sites. In fact, if each reef site is taken separately, R 2 would average 0.894 (range = 0.733 0.997). Of course, the caveat he re is that there we re only four data points per reef site. The sim ilarity in datasets was furthe r corroborated by the relati rankings of percent change over ti me per reef site (Table 3.5). From this we can surmis th place at the habitat resolution of Landsat.
89 ral ecline. cian only two sites had slopes signifi cantly different, Sombrero Reef and Sand ey Reef. A cause for this may be that the image dataset for Sombrero Reef was the nd ed ral ation likelihoods. Although most of the image data provided logical ecological progression of classified habitats, some we re spatially incoherent and me cora l habitat pixels were classifi ed in the backreef zone. The comparison of full time series data for both satellite-derived and in situ datasets provides a strong argument for continuing the us e of remote sensing for monitoring co reef habitats. Both datasets showed a si gnificant decline in coral cover (or Landsatderived coral habitat). Also, the rates of ch ange (i.e., slopes) in Landsat-derived percent coral habitat cover and CREMP coral cover we re not significantly different. In other words, both datasets, which utilize different monitoring techniques and different spatial resolutions, and were gathered over different time periods, had similar rates of d In all cases the rates of decline were greater for CREMP data, with the slopes for Gre Rocks being very close (CREMP, m = -1.043 and Landsat, m = -1.035). Taken individually, K smallest with only seven images. Also, Sombrero Reef was unique among the other seven reef sites in that CREMP data show that Palythoa spp. percent cover was significantly greater than that of live coral cover (16%-zoanthid and 3% coral for 2004) Sand Key Reef had the lowest TD value ( 1.84), which may have led to some pixel misclassifications. Conclusions Twenty-eight Landsat images from 1984-2002 were atmospherically, radiometrically a bathymetrically corrected to perform a change detection analysis for coral habitat cover for eight reef sites in the FKNMS. A spect ral library was produced from ground-truth training pixels for four distinct spectral cl asses: coral habitat, sand, bare hardbottom and covered hardbottom. Images acquired every two years for the spring were compared to images acquired every six years for the fall. There was no significant difference in co habitat between the two seasons for each site. The progression of classified images revealed a range of classific a ecologically unlikely. So This may be due to some live coral cover but is more likely due to the existence of Palythoa spp. located in this zone, as well as some misclassification due to the spectral class separability of coral ha bitat and bare hardbottom.
90 s per, during that time frame, which rther stressed the coral communities. The similarity in results provided by these two e J, Kranenburg C, Torres-Pulliza D, Spraggins Murch B (2006) Global assessment of mode rn coral reef extent and diversity for Results also showed a dramatic decline in La ndsat-derived coral hab itat cover, which wa strongly correlated to the in situ CREMP live stony coral cover data for the eight reef sites. Overall analysis of both datasets found that the greatest decline was in the Up Lower and Middle Keys, respectively. Sim ilarly, the time period with the most significant decline (1998 -2000) was found to be he same for both datasets and has been attributed to the severe bleaching event that occurred fu different datasets show that the changes witnessed at the smaller in situ monitoring level are mirrored at the larger reef habitat scale. The study perf ormed here shows the utility of remote sensing technology for shallow-wate r coastal environments (e.g., coral reefs). Furthermore, the technology and methodology used here may augment a number of research initiatives by providing baseline benthi c data to coastal marine managers. Thes data could be used for fisheries (e.g., essent ial fish habitats), marine park designation (e.g., location) and metabolism (scaling-up) studies. References Acosta A (2001) Disease in Zoanthids: dynamics in space and time. Hydrobiologia 460:113-130 Ahmad W, Neil DT (1994) An Ev aluation of Landsat Thematic Mapper (TM) digital data for discriminating coral reef zonation: Her on Reef (GBR). International Journal of Remote Sensing 15:2583-2597 Andrfout S, Kramer P, Torres-Pulliza D, Joyce KE, Hochberg EJ, Garza-Perez R, Mumby PJ, Riegl B, Yamano H, White WH (2 003) Multi-site ev aluation of IKONOS data for classification of tropical coral reef environm ents. Remote Sensing of Environment 88:128-143 Andrfout S, Muller-Karger FE, Hochberg EJ, Hu C, Carder KL (2001) Change detection in shallow coral reef environm ents using Landsat 7/ETM+ data. Remote Sensing of Environment 78:150-162 Andrfout S, Muller-Karger F, Robinson S
91 S ing s Lidz BH (1996) Decadal-scale changes in benthic foraminiferal Largo, Fl orida. Coral Reefs 15:237-248 9) Coral reefs under stress: s ources of mortality in the Florida Keys. rce Forum 23:147-155 ) regional science and management applicati ons: a view from space. 10th International Coral Reef Symposium:28 June-23 July 2004. Beaver C, Callahan MK, Johnson D, Kidney J, Kupfner S, Porter JW, Torres C, Wade (2006) 2005 CREMP Executive Summary. FWRI, St. Petersburg, FL. 16 pp Boyer JN, Jones RD (2002) A view from the bri dge: external and internal forces affect the ambient water quality of the Florida Ke ys National Marine Sanctuary (FKNMS). In: Porter JW and KG (ed) The Everglades, Florid a Bay, and Coral Reefs of the Florida Key An Ecosystem Source Book. CRC Press, Boca Raton, pp 609-628 Buddemeier RW, Ware JR (2003) Coral reef decline in the Cari bbean. Science 302:391392 Causey BD, Dodge RE, Jaap WC, Banks K, Delaney J, Keller BD, Spieler R (2002) Florida, The state of coral reef ecosystem s of the United States and Pacific Freely Associated States: 2002. NOAA, Silver Spring, MD, pp 101-118 Chander G, Markham B (2003) Revised Landsat-5 TM radiometric calibration procedures and postcalibration dynamic range s. IEEE Transactions on Geoscience and Remote Sensing 41:2674-2677 Cockey E, Hallock P assemblages off Key Dustan P (199 Natural Resou Dustan P (2000) Florida Keys. In: Sheppard C (ed) Seas at the Millenium: An Environmental Evaluation. Elsevier, London, UK, pp 405-414 Dustan P (2003) Ecological perspective: the decline of Carysfort Reef, Key Largo, Florida 1975-2000. In: Valette-Si lver NJ, Scavia D (eds) Ec ological forecasting: new tools for coastal and ecosystem management. NOAA NOS NCCOS 116 Dustan P, Halas JC (1987) Changes in the reef-coral community of Carysfort Reef, Key Largo, Florida: 1972-1982. Coral Reefs 6:91-106 Dustan P, Dobson E, Nelson G (2001) Landsat Thematic Mapper: detection of shifts in community composition of coral reefs. Conservation Biology 15:892-902 Gordon HR, Brown OB, Evans RH, Brown JW Smith RC, Baker KS, Clark DK (1988 A Semianalytic Radiance Model of Ocean Color. Journal of Geophysical ResearchAtmospheres 93:10909-10924
92 allock P, Lidz BH, Cockey-Burkhard EM, Donnelly KB (2003) Foraminifera as nd 99ughes TP, Connell JH (1999) Multiple stressors on coral reefs: A long-term perspective. imnology and Oceanography 44:932-940 llock P (1990) Coral Reefs. In : Myers RL, Ewel JJ (eds) Ecosystems of 04) Introductory Digital Image Processing. Prentice Hall, Upper Saddle ity he ent A, Brodie J, Bucher D, Capone D, Coll J, Dennison W, E: 6891703 H bioindicators in coral reef assessmen t and monitoring: The FORAM Index. Environmental Monitoring and Assessment 81:221-238 Haywick DW, Mueller EM (1997) Se diment retention in encrusting Palythoa spp.--a biological twist to a geological process. Coral Reefs 16:39-46 Hochberg EJ, Atkinson MJ (2000) Spectral discrimination of coral reef benthic communities. Coral Reefs 19:164-171 Hochberg EJ, Atkinson MJ, Andrefouet S ( 2003) Spectral reflectan ce of coral reef bottom-types worldwide and implications for coral reef remote sensing. Remote Sensing of Environment 85:159-173 Hoegh-Guldberg O (1999) Climate change, co ral bleaching and the future of the world's coral reefs. Marine Fres hwater Resources 50:839-866 Hu C, Muller-Karger FE, Andrefouet S, Card er KL (2001) Atmospheric correction a cross-calibration of Landsat-7/ETM+ im agery over aquatic environments: a multiplatform approach using SeaWiFS/MODIS. Remote Sensing of Environment 78: 107 H L Jaap WC, Ha Florida. UCF Press, Orlando, FL, pp 574-616 ensen JR (20 J River, NJ Kirk JT (1994) Light and photosynthesis in aquatic ecosystems. Cambridge Univers ress, Cambridge P Klein CJI, Orlando SPJ (1994) A spatial framework for water-quality management in t Florida Keys National Marine Sanctuary. Bulletin of Marine Science 54:1035-1044 oop K, Booth D, Broadb K Erdmann M, Harrison P, Hoegh-Guldberg O, Hutchings P, Jones GB, Larkum WD O'Neil J, Steven A, Tentori E, Ward S, Williamson J, Yellowless D (2001) ENCOR The effect of nutrient enrichment on coral reef. Synthesis of results and conclusions arine Pollution Bulletin 42:91-120 M Maritorena S, Morel A, Gentili B (1994) Diffuse reflectance of oceanic shallow waters; influence of the water depth and botto m albedo. Limnology and Oceanography 39:1
93 ls eefs 21:179-182 (1994) A long-term terdisciplinary study of the Florida Keys Seascape. Bulletin of Marine Science alandro D, Andrfout S, Dustan P, Mulle r-Karger FE (2003a) Ch ange detection in alandro D, Andrfout S, Muller-Karger FE, Dustan P, Hu C, Hallock P (2003b) t alandro D, Hu C, Andrfout S, Muller-Karger FE (2004) Synoptic water clarity D, s G, Warner RR, Jackson JB (2003) Global ajectories of the long-term decline of coral reef ecosystems. Science 301:955-958 eters EC, Santavy DL, mith GW (2002) The etiology of white pox, a lethal disease of the Caribbean elkhorn nited ope RM, Fry ES (1997) Absorption spectrum (380-700nm) of pure water. II. Integrating orter JW, Lewis SK, Porter KG (1999) The effect of multiple stressors on the Florida orter JW, Kosmynin V, Patterson KL, Port er KG, Jaap WC, Wheaton JL, Hackett K, Lybolt M, Tsokos CP, Yanev G, Marcinek DM, Dotten J, Eaken D, Patterson M, Meier Miller MW, Borque AS, Bohnsack JA (2002) An analysis of the loss of acroporid cora at Looe Key, Florida, USA:1983-2000. Coral R Mumby PJ, Green EP, Clark CD, Edwards AJ (1997) Coral reef habitat mapping: how much detail can remote sensing provide? Marine Biology 130:193-202 Murdoch TJ, Aronson RB (1999) Scale-dependent spatial vari ability of coral assemblages along the Florida Reef Tract. Coral Reefs 18:341-351 Ogden JC, Porter JW, Smith BP, Szmant AM Jaap WC, Forucci D in 54:1059-1071 P coral reef communities using Ikonos satellite sensor imagery and historic aerial photographs. International Journa l of Remote Sensing 24:873-878 P Detection of changes in coral reef co mmunities using Landsat 5/TM and Landsa 7/ETM+ data. Canadian Journal of Remote Sensing 29:201-209 P assessment in the Florida Keys using diffuse attenuation coefficient estimated from Landsat imagery. Hydrobiologia 530-531:489-493 Pandolfi JM, Bradbury RH, Sala E, Hughes TP, Bjorndal KA, Cooke RG, McArdle McClenachan L, Newman MJ, Parede tr Patterson KL, Porter JW, Ritchie KB, Polson SW, Mueller E, P S coral, Acropora palmata Proceedings of the National Academy of Sciences of the U States of America 99:8725-8730 P cavity measurements. Applied Optics 36:8710-8723 P Keys coral reef ecosystem: A landscape hypot hesis and a physiological test. Limnology and Oceanography 44:941-949 Porter JW, Meier OW (1992) Quantification of loss and change in Floridian Reef coral populations. American Zoologist 32:625-640 P
94 oral Bay, and ook. CRC Press, Boca Raton, pp 49-769 introduction. Spingererlag, Berlin n JH, Kindinger JL (1989) Reefs of Florida and e Dry Tortugas Field Trip Guidebook T176. AGU, Washington D.C. 57 y 23:247-259 eillet PM, Barker JL, Markham BL, Irish RR, Fedosejevs G, Storey JC (2001) aton JL, Jaap WC (1988) Corals and other prominent benthic Cnidaria of Looe Key ational Marine Sanctuary. Florida Marine Research Institute, St. Petersburg, FL 31 OW, Brill M, Dustan P (2002) Detection of co ral reef change by the Florida Keys C Reef Monitoring Project. In: Porter JW and KG (ed) The Everglades, Florida Coral Reefs of the Florida Keys An Ecosystem Source B 7 Richards JA (1986) Remote sensing digital image analys is: an V Shinn EA, Lidz BH, Halley RB, Hudso th Smith RC, Baker KS (1978) The bio-optical st ate of ocean waters and remote sensing. Limnology and Oceanograph Sokal RR, Rohlf FJ (1981) Biom etry. W.H. Freeman, New York T Radiometric cross-calibration of the Landsat -7 ETM+ and Landsat-5 TM sensors based on tandem data sets. Remote Sensing of Environment 78:39-54 Whe N
95 CHAPTER FOUR Conclusions Synopsis This study utilized Landsat sa tellite data to provide quant itative data about the water clarity and benthic habitat for the coral r eef ecosystems of the Florida Keys National Marine Sanctuary (FKNMS). Specifically it sought to: Derive the diffuse attenuation coefficient ( K d ) for 29 reef sites from Landsat images between 1984 and 2002 to determin e: 1) whether there was a seasonal variability, 2) the temporal variability of each reef site and 3) the spatial variability between individual r eef sites and geologic regions. Map benthic habitats for ei ght reef sites located in the FKNMS using Landsat images from 1984-2002 to derive: 1) a singl e spectral library for four benthic classes, 2) a time series of change for those benthic classes, specifically coral habitat, 3) a rate of change for the individual reef sites and geologic regions and 4) a comparison to an in situ database for the same reef sites. The Landsat image data comprised a total of 28 images (1984-2002), fourteen images from path 15, row 43 and fourteen images from path 16, row 43. Twenty-two of the images were acquired by Landsat 5 Thema tic Mapper (TM) and six were acquired by Landsat 7 Enhanced Thematic Plus (ETM+). Images were processed for the local spring season (March May) every two years a nd the fall season (September November) every six years.
96 Summary of Conclusions 1. Landsat image data provided valid K d data for clear shallow-water ecosystems in the blue (450 520 nm) and green (520 600 nm) bands, but not in the red (630 690 nm). 2. Seasonality between spring and fall imag e data in the region of study showed no significant difference (paired t-test, p = 0.596). 3. Localized plume events that obscured th e benthos could be detected from the image data but confounded the algorithm, therefore no K d could be derived. 4. There was high variability in K d values for the individua l reef sites and within the three geologic regions in whic h those reef sites were located. 5. Both in situ K d (PAR) and the Landsat-derived K d found that the lowest average K d values were found in the Lower Keys (of 0.027 m -1 and 0.053 m -1 ), followed by the Middle (0.032 m -1 and 0.054 m -1 ) and Upper Keys (0.039 m -1 and 0.062 m -1 ), for bands 1 and band 2 respectively. 6. Landsat image data can provide a change detection time seri es for coral reef environments. 7. Reflectance acquired in situ at the benthos can be li nked to that of Landsatderived spectral data for specific benthi c habitats when then the imagery has been corrected for atmospheric and water column influences. 8. There was no seasonal variability det ected for coral habitat cover in the Florida Keys (paired t-test, p = 0.535). 9. In situ live coral cover data at a fine spat ial resolution (Coral Reef Evaluation and Monitoring Program, CREMP) correlate d to Landsat-deriv ed coral habitat data at a medium spatial resolution for the eight reef sites analyzed (R 2 = 0.704). a. CREMP coral cover and Landsat-deriv ed coral habitat (2002) both found the highest coral cover in the Lower (CREMP = 9.5%, Landsat = 8.8%), Upper (CREMP = 5.3%, Landsat = 7. 7%) and Middle Keys (CREMP = 3.2%, Landsat = 4.5%), respectively. b. Rates of change were similar over CREMP and Landsat concurrent data (1996-2002), CREMP = -52% (-8.7%/y) and Landsat = -37% (-6.2%/y).
97 c. The level and rates of change were not statistically different for the full timelines of each dataset (F-test, p = 0.303). d. Both datasets had the greatest rate of change associated with the 19971998 ENSO event (CREMP = -37%, Landsat = -32%). e. Relative rankings of change between th e eight reef sites were similar for both datasets. 10. The presence of benthic constituents (i.e., Palythoa spp.) with zooxanthellae shared with Scleractinian coral species ma y lead to the classification of pixels in the backreef area as cora l habitat. However, field data is required to test this hypothesis. Remote Sensing and Coral Reefs The number of coral reef remote sensi ng peer-reviewed published manuscripts has significantly increased annually over the past decade, from three manuscripts in 1995 to eighteen in 2005, reaching a cumula tive total of 114. This can be attributed to advances in technology, methodology and th e understanding of the benefits that remote sensing can provide to coral reef research. The ten most cited references are listed in Table 4.1 and show a range of coral reef topics from sp ectral characterization to mapping as well as different remote sensing technologies that range from 1 km sea-surface temperature (SST) data to 1 m airborne data. Other of ten cited manuscripts provide information on bleaching, remote sensing for management, water attenuation over coral reefs and modeling or scaling-up of results. The derivation of K d over coral reefs performed in th is study (see Chapter 2) provides information similar in scope but at a larger spatial extent when compared to manuscripts covering the same topic (total = 28). From the literature, there are four independent reasons to derive remotely sensed K d for coral reef habitats: determination of water column constituents (Roelfsema et al. 2006), as a proxy for water qual ity (Palandro et al. 2004), to derive bathymetry (Isoun et al. 2003) and to perform water column correction to compute bottom reflectance (Maritorena 1996) This study combines two of these reasons, incorporating K d as a proxy for water quality as well as its use for a water
98 Table 4.1. List of ten most cited coral reef remote sensing references since 1995. Topic and Location Citation # Cited Review of remote sensing for managing tropical ecosystems, Global Green et al. 1996 Coastal Management 68 Determination of SST and its correlation to delta 18 O for coral, Ecuador Wellington et al. 1996 Paleoceanography 61 Spectra of coral reef constituents and remote sensing, Hawaii Hochberg et al. 2000 Coral Reefs 48 What information can remote sensing provide for coral reefs, Caribbean Mumby et al. 1997 Marine Biology 47 Spectral discrimination of healthy versus nonhealthy corals, Fiji Holden & LeDrew 1998 RSE 43 Comparison of remote sensing technologies for coral reefs, Caribbean Mumby et al. 1998 Coral Reefs 41 Development of a single classification scheme for coral reefs, Caribbean Mumby & Harborne 1999 Biological Conservation 38 Fluorescence of corals, Caribbean Mazel 1995 Marine Ecology Prog S 36 Mapping marine environments with high resolution data, Caribbean Mumby & Edwards 2002 RSE 34 Spectral characterization of coral reef benthic constituents, Caribbean Myers et al. 1999 Coral Reefs 33 column correction (completed in Chapter 3). Although the results derived from this study are only snapshots in time for, at most, twic e a year, this study represents the largest spatial and temporal study of its kind for coral reef environments. Coral reef habitat mapping by remote sensing has been the leading topic in coral reef manuscripts (total = 84). However, only ten manuscripts have been published on coral reef remote sensing change detection or m onitoring, five of which used Landsat (Zainal et al. 1993, Andrfout et al. 2001, Dustan et al., 2001, Palandro et al. 2003b, Shapiro et al. 2005). The results from this coral habita t change study (see Chapter 3) encompass the largest temporal dataset and the second largest spatial dataset. Neither of the studies in this project nor any of the studies cited a bove has been used to further coral reef remote sensing into an operational phase setting. Only the acquisition
99 of SST data from the Advanced Very High Resolution Radiometer (AVHRR) sensor to derive a hot spot map is utilized operati onally by the National Oceanic and Atmospheric Administration (NOAA) for a coral reef bleach ing index (Strong et al. 2002). However, Landsat image data are currently being us ed to derive a global shallow-water geomorphological dataset for coral r eefs (Andrfout et al. 2006). K d Study Portability Portability, as used here, is defined as th e likelihood that the same methods can be utilized elsewhere in the world with simila r results. The methods utilized in the K d study are highly transportable to ot her coral reef ecosystems worldwide, with some caveats. For a single image, a consistent benthos (e.g., sand) must be discernable within the image, so that a relative comparison can be made for that benthos over several different depths. Any difference between the variably-d epthed benthic signals is assumed to be due solely to the water column influence and change in depth, not fr om different benthic spectral signatures. Secondly, bathymetry at or near the spatial resolution of the image must be available to facilita te the steps mentioned above. The greater the accurate depth interval resolution, the greater likelihood of a more consistently accurate K d If a time series is desired (e.g., operati onal status), a rigorous atmosphe ric correction is required to remove influence of Rayleigh and aeroso l effects on the satellite signal. A major concern of the portability of this study is the fact that Lands at image data have a temporal resolution of 16 days and still pr ovide only a snapshot of water clarity, which can vary significantly on shorter time scales. This can be especially true in geographic areas where rainfall can cause increased runo ff of terrestrial sediment or in locations where wind can lead to increased water column mixing. Both of th ese variable physical parameters can be present at any given locati on and change over time periods of to hours or days. Therefore, the use of a satellite with greater temporal resolution would be preferred, e.g., Moderate Imaging Spectro radiometer (MODIS), flown onboard two satellites (Terra and Aqua), can provide wate r clarity information up to four times a day at a spatial resolution of 250 m.
100 Coral Habitat Change Study Portability The coral habitat change study ma y be less portable than the K d study. As a first step, the data provided by the K d study need to be made available. Specifically, there needs to be an accurate atmospheric and water column correction performed. For a single image, these steps need to be completed so that be nthic spectral data from one site within the image can be made comparable to other sites, irregardless of depth or distance from the original site. For multiple images the need is greater as image data includes variability within an image and variab ility between images. The portability of the classes must also be considered. The goal of this study was to determine the level of change specifically to one benthic class (i .e., coral habitat). However, actual live coral cover was not the dominant benthic class present in any 30 m pixel. The results of this st udy would differ if the four classe s were analyzed equally. If a pixel was dominated by a benthic class (> 50%), that pixel would be classified accordingly and none of the Landsat-derived coral habitat pixels would have been classified. Therefore, without field data to specifically trai n the coral habitat class this study would not have been possible. The addition of other cl asses would be necessary for other geomorphologic zones. For example, a detailed analysis of the backreef would require the inclusion of a seagrass class. The spectral library derived for this study woul d likely only be useful for the coral reef ecosystems within the Florida Keys, with the caveat that any new Landsat images would need to be calibrated to the base image used in this study. All of the images underwent an Empirical Line Calibration (ELC) to th e 2002 image. This was done to compensate for variable noise, sensor cal ibration, changes in the relative spectral response and intersensor calibration (see Chapter 3). The benthi c constituents found in the Florida Keys are also found in the coral reef environments of th e Caribbean Sea, therefor e it is feasible that if an ELC were performed on Caribbean Lands at images and the same benthic classes were desired, the spectral library derived for this study could be used.
101 Even though the classes and sp ectral library may not be porta ble to other shallow-water coral reef environments of the world, the me thods can be. This statement assumes that bathymetry data are available at a spatial a nd depth interval resolution to be useful, an accurate atmospheric correction can be applie d and the number of benthic classes desired are spectrally separable at Landsat spatial resolution. Knowledge of coral reef zonation, morphology and optics is also suggested. In May 2003, there was a scan line correct or malfunction on the Landsat 7 ETM+ sensor, causing data gaps to the outer two-thirds of an image; ap proximately 22% of data is lost in each image. Landsat 5 TM has been ope rational since 1984 and has suffered two near fatal systems crashes in the past twelve months. Therefore, long-term continuity of both operational Landsat satellites is uncertain. Two high spatia l resolution satellites are currently in operation that po ssess visible bands centered at the same wavelengths as Landsat TM and ETM+. IKONOS and QuickBir d are privately owned and have visible spatial resolutions of 4 m and 2.4 m, resp ectively, and are logical choices to expand shallow-water coral reef habitat mapping. A lthough no data exist for these sensors before 1999 (IKONOS launch), it is possible to use hist oric aerial photography to build a time series (Palandro et al. 2003a). However, the image data are expensive; NOAA recently purchased IKONOS data for the entire Florida Ke ys Reef Tract at a cost of approximately $400,000 or $50 km -2 (Aurelie Shapiro-NOAA, personal communication). As a means of comparison, Landsat image data for the sa me geographic area costs $1200 or $0.02 km -2 (two individual images). Also, there is a lack of published information on the sensor radiometric responses and possi ble degradation over time. Benefits to Other Research Topics The utility of the coral habitat change time series to other res earch topics can be significant. The study provides two components to other studies; a map that can serve as a base layer with other data la yers overlaid and a series of temporal change base layers that may provide a means to reconstruct other data layers. Below is a summary of research areas that could bene fit from this studys data.
102 Water Clarity for Shallow-water Ecosystems Data from both MODIS and the Sea Wide Fi eld-of-view Sensor (SeaWiFS) are used to derive K d (490) data on a daily basis. However the satellite-derived K d (490) over shallowwater ecosystems (e.g., coral reefs) are directly influenced by benthic communities. The benthos is visible to the sensor due to its sh allow depth and relatively clear water column. However, the influence of the benthos could be assumed to be uniform if consistent benthic coverage (e.g., sand) ove r variable depths are known. This is essentially what was performed in this study for Landsat data. Another approach would be to acquire in situ data on the reflectance at th at wavelength of a number of different dominant benthic habitats. Again by knowing depth, the data could be used to empirically account for the benthos so that that it can be removed from the total signal received by the sensor. This could be accomplished for each wavelength that the sensors acquire data for to better characterize the water column for its diffe rent constituents (e.g., sediment versus chlorophyll). Water clarity data would be useful to coas tal zone managers as it would provide a proxy for water quality on a daily basis. More rigorous research could be performed utilizing these data, as there would be several more data points available to produce statistical analyses. Operational satellit e-derived water clarity data w ould be analogous to the SST hot spot data currently being output by NOAA. The data could be used conjunction with the SST data to enhance the early warning of possible bleaching events. For example, long periods of high water transparency may augment the bleaching response in corals by allowing greater levels of ultraviolet ra diation to penetrate the water column. Marine Park Designation Designation of marine parks, reserves and sanc tuaries is a management tool utilized to reduce anthropogenic effects on marine res ources. Such designations are most commonly used to either protec t benthic habitats (e.g., coral r eefs) or fish populations. In order to effectively delineate a marine par k, the location and extent of the habitat of interest to protect needs to be known. R ecent work shows that global marine park designations may not be located in the optimal locations to best protect coral reef
103 ecosystems (Mora et al. 2006). The image data from this study can provide the regions where coral habitat colonization is possible, but more importantly, where coral habitat currently exists and occurred historically. The data can also provide a mechanism by which marine parks designated to promote coral reef ecosystem protection can be validated. This could lead to an interesting research study. Two coral reef s ites could be selected that have similar current and historic coral habitat cover. The experiment would designate one site as a marine park and retain the other site as a control. Over time the negative impacts caused by shipping, fishing and scuba diving (i.e ., anthropogenic) could be gleaned. If similar rates of change in coral habitat persist for both sites, it may be assumed that sources acting on a larger scale than those occurring lo cally, directly on the area, are the cause for decline. By utilizing the methods from the K d study here, water clarity for both sites can be derived and a determ ination can be made as to its effect in the change. Fisheries Many fish species, important to fisheries (i.e., the grouper-snapper complex) rely on shallow-water coral reef habitats during the juvenile phase of th eir life cycle, i.e., essential fish habitats (EFH) (Ault et al. 2005). Still other fish speci es spend their entire life on one reef site and others show site fi delity for foraging (Humston et al. 2005). If known EFHs for a target fish species are asso ciated with life cycle phases, foraging or spawning aggregations then information on changes to those hab itats could provide useful data to fisheries scientis ts studying that species. This is especially true if there is a well documented relationship between a distinct feature that the habitat may provide. For example, coral reef environments in the Dry Tortugas that have a higher rugosity are associated with the highest reef fish populations (Ault et al 2006). Coral habitat can be used as a proxy for reef rugosity. The coral ha bitat data could also be applied to other reef inhabitants besides fish that are economically important (e.g., Caribbean Spiny Lobster, Panulirus argus ).
104 Carbonate Production The role of coral reefs in the global carbon cycle is about 1% of global anthropogenic input, but is critical to understand for accura te modeling of global climate change (Ware et al. 1991) and ocean acidification (Kleypas et al. 2006). Previous research has scaledup carbonate metabolism rates from the combination of in situ metabolism rate data and remote sensing-derived maps of different benthic habitats (Andr fout and Payri 2000, Brock et al. 2006). This study can provide similar spatially e xplicit results as the scalingup studies as well as data on the temporal va riability of carbonate metabolic rates for a specific reef zone or site, based on the changes in benthic communities over time. The topics summarized above are some research areas wher e the applicab ility of the remote sensing tools and products discussed in this dissertation can be applied. The methods and results provided herein may provide the greatest usage in developing countries that may lack the f unds to sustain long-term ongoing in situ monitoring projects. The remote sensing data acquired for coral reef environments could present a means by which to monitor, what is in some nations, a major economic entity. References Ault JS, Bohnsack JA, Smith SG, Luo JG (2005) Towards sustainable multispecies fisheries in the Florida, USA, coral reef ecosystem. Bulletin of Marine Science 76:595622 Ault JS, Smith SG, Bohnsack JA, Luo JG, Harper DE, McClellan DB (2006) Building sustainable fisheries in Florida's coral reef ecosystem: Positive signs in the Dry Tortugas. Bulletin of Marine Science 78:633-654 Andrfout S, Muller-Karger FE, Hochberg EJ, Hu C, Carder KL (2001) Change detection in shallow coral reef environm ents using Landsat 7/ETM+ data. Remote Sensing of Environment 78:150-162 Andrfout S, Muller-Karger F, Robinson J, Kranenburg C, Torres-Pulliza D, Spraggins S, Murch B (2006) Global assessment of mode rn coral reef extent and diversity for regional science and management applicati ons: a view from space. 10th International Coral Reef Symposium: 28 June-23 July 2004
105 Andrfout S, Payri C (2000) Scaling-up car bon and carbonate metabo lism of coral reefs using in-situ data and remote sensing. Coral Reefs 19:259-269 Brock JC, Yates KK, Halley RB, Kuffner IB, Wright CW, Hatcher BG (2006) Northern Florida reef tract benthic metabolism scaled by remote sensing. Marine Ecology-Progress Series 312:123-139 Dustan P, Dobson E, Nelson G (2001) Landsat Thematic Mapper: detection of shifts in community composition of coral reefs. Conservation Biology 15:892-902 Humston R, Ault JS, Larkin MF, Luo JG ( 2005) Movements and site fidelity of the bonefish Albula vulpes in the northern Florida Keys determined by acoustic telemetry. Marine Ecology-Progress Series 291:237-248 Isoun E, Fletcher C, Frazer N, Gradie J (2003) Multi-spectral mapping of reef bathymetry and coral cover; Kailua Bay, Hawaii. Coral Reefs 22:68-82 Kleypas J, Feely RA, Fabry VJ, Langdon C, Sabine C, Robbins LL (2006) Impacts of ocean acidification on coral reefs and other marine calcifiers: A guide for future research. NSF, NOAA, and the USGS, St. Petersburg, FL 88 Maritorena S (1996) Remote sensing of the water attenuation in co ral reefs: a case study in French Polynesia. International Journal of Remote Sensing 17:155-166 Mora C, Andrefouet S, Costello MJ, Krane nburg C, Rollo A, Veron J, Gaston KJ, Myers RA (2006) ECOLOGY: Enhanced: Coral Reef s and the Global Network of Marine Protected Areas. Science 312:1750-1751 Palandro D, Andrfout S, Dustan P, Mulle r-Karger FE (2003a) Ch ange detection in coral reef communities using Ikonos satellite sensor imagery and historic aerial photographs. International Journa l of Remote Sensing 24:873-878 Palandro D, Andrfout S, Muller-Karger FE, Dustan P, Hu C, Hallock P (2003b) Detection of changes in coral reef co mmunities using Landsat 5/TM and Landsat 7/ETM+ data. Canadian Journal of Remote Sensing 29:201-209 Palandro D, Hu C, Andrfout S, Muller-Karger FE (2004) Synoptic water clarity assessment in the Florida Keys using diffuse attenuation coefficient estimated from Landsat imagery. Hydrobiologia 530-531:489-493 Roelfsema CM, Phinn SR, Dennison WC, Dekker AG, Brando VE (2006) Monitoring toxic cyanobacteria Lyngbya majuscula (Gomont) in Moreton Bay, Australia by integrating satellite image data a nd field mapping. Harmful Algae 5:45-56
106 Shapiro AC, Rohmann SO (2005) Summit-to-s ea mapping and change detection using satellite imagery: tools for conservation and management of coral reefs. Revista De Biologia Tropical 53:185-193 Strong A, Liu G, Meyer J, Hendee J, Sas ko D (2004) Coral Reef Watch 2002. Bulletin of Marine Science 75:259-268 Ware JR, Smith SV, Reaka-Kudla ML (1991) Coral reefs: Sources or sinks of atmospheric CO2. Coral Reefs 11:127-130 Zainal A, Dalby DH, Robinson IS (1993) Monitoring marine ecologi cal changes in the east coast of Bahrein with Landsat TM. Photogrammetric Engineering and Remote Sensing 59:415-421
Appendix A. Location and date of visually estimated percent benthic cover ground-truth data. Reef Site Z Sand Bare Hardbottom Covered Hardbottom Coral Habitat Seagrass Latitude Longitude Date Molasses Reef 5 5 15 70 10 25.0113 -80.3749 08/18/03 60 10 25 5 25.0112 -80.3749 08/18/03 20 50 20 10 25.0109 -80.3749 08/18/03 50 30 20 25.0104 -80.3749 08/18/03 50 30 20 25.0104 -80.3746 08/18/03 50 30 20 25.0104 -80.3743 08/18/03 Molasses Reef 4 5 90 5 25.0139 -80.3730 08/18/03 5 5 75 15 25.0141 -80.3728 08/18/03 5 10 75 10 25.0141 -80.3725 08/18/03 80 5 15 25.0147 -80.3731 08/18/03 60 30 10 25.0150 -80.3731 08/18/03 70 20 10 25.0152 -80.3731 08/18/03 Molasses Reef 6 15 70 15 25.0117 -80.3743 08/18/03 10 65 25 25.0117 -80.3740 08/18/03 10 80 10 25.0117 -80.3737 08/18/03 60 25 15 25.0096 -80.3788 08/18/03 30 60 10 25.0096 -80.3785 08/18/03 60 35 5 25.0096 -80.3782 08/18/03 30 50 20 25.0096 -80.3779 08/18/03 25 70 5 25.0096 -80.3776 08/18/03 50 50 25.0096 -80.3773 08/18/03 Grecian Rocks 2 10 80 10 25.1092 -80.3048 08/19/03 108
Appendix A (Continued) Reef Site Z Sand Bare Hardbottom Covered Hardbottom Coral Habitat Seagrass Latitude Longitude Date Grecian Rocks 2 30 70 -80.3048 -80.3051 08/19/03 10 90 25.1092 -80.3054 08/19/03 20 80 -80.3054 -80.3057 08/19/03 25 70 5 25.1092 -80.3057 08/19/03 5 10 80 5 25.1092 -80.3060 08/19/03 10 90 25.1092 -80.3063 08/19/03 10 15 75 25.1092 -80.3066 08/19/03 5 10 70 15 25.1092 -80.3069 08/19/03 Grecian Rocks 3 30 20 50 25.1127 -80.3039 08/19/03 50 50 25.1130 -80.3039 08/19/03 20 80 25.1133 -80.3039 08/19/03 10 90 25.1135 -80.3039 08/19/03 40 60 25.1138 -80.3039 08/19/03 100 25.1141 -80.3039 08/19/03 Grecian Rocks 3 10 60 20 10 25.1116 -80.3036 08/19/03 75 20 5 25.1116 -80.3036 08/19/03 20 70 10 25.1114 -80.3036 08/19/03 10 60 30 25.1114 -80.3039 08/19/03 90 10 25.1111 -80.3039 08/19/03 70 30 25.1108 -80.3039 08/19/03 30 60 10 25.1106 -80.3039 08/19/03 60 30 10 25.1103 -80.3039 08/19/03 109
Appendix A (Continued) Reef Site Z Sand Bare Hardbottom Covered Hardbottom Coral Habitat Seagrass Latitude Longitude Date Grecian Rocks 3 30 20 50 25.1100 -80.3039 08/19/03 CarysfortReef 4 5 95 25.2245 -80.2089 08/20/03 100 25.2244 -80.2092 08/20/03 100 -80.2092 -80.2092 08/20/03 5 5 90 -80.2092 -80.2092 08/20/03 30 10 60 25.2236 -80.2092 08/20/03 10 5 85 25.2233 -80.2092 08/20/03 Carysfort Reef 6 30 10 60 25.2195 -80.2125 08/20/03 30 5 65 25.2195 -80.2122 08/20/03 5 95 25.2195 -80.2119 08/20/03 10 10 80 25.2195 -80.2119 08/20/03 15 15 70 25.2195 -80.2116 08/20/03 5 95 25.2195 -80.2113 08/20/03 Molasses Reef 5 15 70 15 25.0109 -80.3743 06/11/04 60 15 25 25.0112 -80.3743 06/11/04 20 60 20 25.0114 -80.3743 06/11/04 50 30 20 25.0114 -80.3738 06/11/04 45 25 30 25.0112 -80.3740 06/11/04 50 30 20 25.0109 -80.3740 06/11/04 10 90 25.0106 -80.3740 06/11/04 5 60 30 25.0104 -80.3740 06/11/04 25 50 25 25.0101 -80.3741 06/11/04 110
Appendix A (Continued) Reef Site Z Sand Bare Hardbottom Covered Hardbottom Coral Habitat Seagrass Latitude Longitude Date Molasses Reef 5 5 80 15 25.0098 -80.3741 06/11/04 MolassesReef 3 60 30 10 25.0152 -80.3752 06/11/04 70 20 10 25.0150 -80.3752 06/11/04 15 40 25 15 5 25.0145 -80.3749 06/11/04 10 80 10 25.0139 -80.3731 06/11/04 20 60 20 25.0139 -80.3731 06/11/04 20 55 25 25.0141 -80.3731 06/11/04 5 25 55 15 25.0141 -80.3734 06/11/04 30 15 55 25.0144 -80.3734 06/11/04 15 25 60 25.0147 -80.3734 06/11/04 Conch Reef 5 10 90 24.9562 -80.4599 06/14/04 20 80 24.9560 -80.4599 06/14/04 20 80 24.9557 -80.4599 06/14/04 20 80 24.9554 -80.4599 06/14/04 20 80 24.9552 -80.4599 06/14/04 30 70 24.9549 -80.4599 06/14/04 25 75 24.9546 -80.4599 06/14/04 30 70 24.9543 -80.4599 06/14/04 30 70 24.9541 -80.4599 06/14/04 40 60 24.9541 -80.4599 06/14/04 90 5 5 24.9538 -80.4599 06/14/04 40 10 50 24.9538 -80.4599 06/14/04 111
Appendix A (Continued) Reef Site Z Sand Bare Hardbottom Covered Hardbottom Coral Habitat Seagrass Latitude Longitude Date Conch Reef 3 100 24.9511 -80.4635 06/14/04 100 24.9514 -80.4635 06/14/04 20 80 24.9516 -80.4635 06/14/04 90 10 24.9519 -80.4635 06/14/04 90 10 24.9519 -80.4635 06/14/04 100 24.9522 -80.4635 06/14/04 100 24.9525 -80.4632 06/14/04 15 5 80 24.9525 -80.4629 06/14/04 40 60 24.9525 -80.4626 06/14/04 80 20 24.9524 -80.4623 06/14/04 85 15 24.9524 -80.4620 06/14/04 90 10 24.9524 -80.4617 06/14/04 Grecian Rocks 2 10 50 40 25.1122 -80.3039 06/15/04 15 45 40 25.1119 -80.3039 06/15/04 20 40 40 25.1116 -80.3039 06/15/04 Grecian Rocks 3 70 30 25.1081 -80.3072 06/15/04 10 50 30 10 25.1081 -80.3069 06/15/04 5 40 25 30 25.1081 -80.3066 06/15/04 20 40 40 25.1081 -80.3063 06/15/04 30 25 35 10 25.1081 -80.3060 06/15/04 20 25 25 15 15 25.1081 -80.3057 06/15/04 Western Sambo 6 50 30 20 24.4800 -81.7181 07/16/03 112
Appendix A (Continued) Reef Site Z Sand Bare Hardbottom Covered Hardbottom Coral Habitat Seagrass Latitude Longitude Date Western Sambo 6 60 25 15 24.4800 -81.7184 07/16/03 90 70 15 15 24.4800 -81.7187 07/16/03 80 20 24.4800 -81.7190 07/16/03 90 10 24.4800 -81.7193 07/16/03 90 10 24.4800 -81.7196 07/16/03 Western Sambo 5 10 40 20 30 24.4803 -81.7172 07/16/03 40 20 10 30 24.4803 -81.7170 07/16/03 50 15 10 25 24.4803 -81.7167 07/16/03 50 15 10 25 24.4803 -81.7164 07/16/03 80 5 5 10 24.4803 -81.7161 07/16/03 25 20 55 24.4803 -81.7158 07/16/03 30 40 30 24.4806 -81.7152 07/16/03 30 40 30 24.4806 -81.7149 07/16/03 40 50 10 24.4806 -81.7146 07/16/03 Western Sambo 6 20 10 55 15 24.4807 -81.7140 07/16/03 10 70 20 24.4808 -81.7137 07/16/03 20 65 15 24.4808 -81.7134 07/16/03 30 50 20 24.4808 -81.7131 07/16/03 15 15 65 5 24.4808 -81.7128 07/16/03 20 15 60 5 24.4808 -81.7125 07/16/03 Looe Key Reef 5 10 30 40 20 24.5462 -81.4087 06/03/04 20 55 25 24.5462 -81.4084 06/03/04 113
Appendix A (Continued) Reef Site Z Sand Bare Hardbottom Covered Hardbottom Coral Habitat Seagrass Latitude Longitude Date Long Key Reef 5 50 35 15 24.5462 -81.4081 06/03/04 20 75 5 24.5462 -81.4078 06/03/04 20 70 10 24.5462 -81.4075 06/03/04 20 70 10 24.5462 -81.4072 06/03/04 75 25 24.5467 -81.4066 06/03/04 15 50 35 24.5465 -81.4066 06/03/04 5 55 40 24.5462 -81.4066 06/03/04 45 45 10 24.5459 -81.4090 06/03/04 45 55 24.5459 -81.4087 06/03/04 5 25 55 15 24.5459 -81.4084 06/03/04 Looe Key Reef 4 20 25 55 24.5476 -81.4031 06/03/04 10 35 45 10 24.5473 -81.4031 06/03/04 35 30 35 24.5470 -81.4031 06/03/04 5 30 35 30 24.5467 -81.4042 06/03/04 35 45 20 24.5470 -81.4042 06/03/04 35 50 15 24.5473 -81.4042 06/03/04 Looe Key Reef 6 15 30 50 5 24.5470 -81.4033 06/03/04 10 55 35 24.5470 -81.4036 06/03/04 10 50 40 24.5470 -81.4039 06/03/04 10 55 35 24.5462 -81.4060 06/03/04 50 25 25 24.5465 -81.4060 06/03/04 5 25 55 15 24.5467 -81.4060 06/03/04 114
Appendix A (Continued) Reef Site Z Sand Bare Hardbottom Covered Hardbottom Coral Habitat Seagrass Latitude Longitude Date Looe Key Reef 6 20 65 25 24.5481 -81.4031 06/03/04 20 70 10 24.5478 -81.4031 06/03/04 10 60 30 24.5476 -81.4031 06/03/04 Western Sambo 5 10 5 50 30 24.4814 24.4814 07/02/04 70 30 24.4811 -81.7119 07/02/04 5 65 30 24.4808 -81.7119 07/02/04 30 50 20 24.4806 -81.7119 07/02/04 15 15 65 5 24.4803 -81.7119 07/02/04 5 60 30 24.4800 -81.7119 07/02/04 5 5 60 30 24.4803 -81.7175 07/02/04 10 75 15 24.4800 -81.7175 07/02/04 10 60 30 24.4797 -81.7175 07/02/04 5 70 25 24.4814 -81.7110 07/02/04 15 50 30 24.4814 -81.7107 07/02/04 20 60 20 24.4814 -81.7104 07/02/04 Sand Key Reef 6 80 20 24.4526 -81.8772 07/22/04 35 45 20 24.4526 -81.8769 07/22/04 45 50 5 24.4526 -81.8769 07/22/04 40 60 24.4529 -81.8769 07/22/04 20 55 25 24.4529 -81.8766 07/22/04 65 30 24.4529 -81.8763 07/22/04 25 65 10 24.4532 -81.8749 07/22/04 115
Appendix A (Continued) Reef Site Z Sand Bare Hardbottom Covered Hardbottom Coral Habitat Seagrass Latitude Longitude Date Sand Key Reef 6 15 70 15 24.4532 -81.8746 07/22/04 15 80 5 24.4532 -81.8743 07/22/04 30 70 24.4532 -81.8740 07/22/04 15 65 20 24.4532 -81.8737 07/22/04 5 90 5 24.4532 -81.8734 07/22/04 Sand Key Reef 4 15 85 24.4526 -81.8787 07/22/04 20 60 20 24.4526 -81.8784 07/22/04 15 70 15 24.4526 -81.8781 07/22/04 70 30 24.4526 -81.8778 07/22/04 35 65 24.4526 -81.8775 07/22/04 20 65 15 24.4526 -81.8772 07/22/04 15 75 10 24.4529 -81.8781 07/22/04 15 85 24.4526 -81.8781 07/22/04 25 70 5 24.4523 -81.8781 07/22/04 116
About the Author David Anthony Palandro received his Bachelors degree in Biology an d History from the State University of New York, College of Brockport in 1993. While in Brockport, he completed a field course on coral reef ecosy stems on San Salvador Island, the Bahamas. Upon completion of his B.S. he moved to Florid a and worked as a contract consultant at the US Geological Survey Center for Coasta l Geology in St. Peters burg. David began pursuing a graduate degree in 1997, formally entering the M.S. program in 1998 at the University of South Florida in St. Petersburg, Fl orida. He received his Masters degree in Marine Science in 2000 from USF. David bega n his Ph.D. in 2001 at USF. While in St. Petersburg, he has participated in research activities in Florida, Hawaii, Australia and the Bahamas. David has published in peer-revie wed journals, presented at international conferences and participated extensiv ely in education outreach programs.