The Application of Quickbird and Multi-Temporal Landsat TM Data for Coral Reef Habitat Mapping Case Study: Derawan Island, East Kalimantan, Indonesia
The Application of QuickBird and Multi-tem poral Landsat TM Data for Coral Reef Habitat Mapping Case Study: Derawan Island, East Kalim antan, Indonesia
Marlina Nurlidiasari March, 2004
The Application of QuickBird and Multi-temporal Landsat TM data for coral reef habitat mapping Case Study: Derawan Island, East Kalimantan, Indonesia
by
Marlina Nurlidiasari
Thesis submitted to the International Institute for Geo-information Science and Earth Observation in partial fulfilment of the requirements for the degree of Master of Science in Geo-Information Science and Earth Observation specialisation Coastal Zone Studies.
Degree Assessment Board
Prof. Dr. S. de Jong (External Examiner) Prof. Dr. F.D. van der Meer (Chairman) ESA Department, ITC Dr. T.W. Hobma (Supervisor) WRES Department, ITC Drs. M.C.J. Damen (Supervisor) ESA Department, ITC Dr. P.M. van Dijk (Member) ESA Department, ITC Drs. E. Westinga (Member), NRM Department, ITC
INTERNATIONAL INSTITUTE FOR GEO-INFORMATION SCIENCE AND EARTH OBSERVATION ENSCHEDE, THE NETHERLANDS Disclaimer
This document describes work undertaken as part of a programme of study at the International Institute for Geo-information Science and Earth Observation. All views and opinions expressed therein remain the sole responsibility of the author, and do not necessarily represent those of the institute. The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Abstract
Coral reefs are one of our earth‘s fundamental resources. Due to their rich habitat and large diversity of marine species and ecological complexity, they are frequently compared to tropical rainforest. They are distributed in the tropical and sub-tropical coastal waters, mostly in developing countries. However, the health of the world‘s coral reefs is in serious decline. Thus, information on the health of coral reefs status is crucial for their conservation and sustainable utilization.
Coral reefs in Derawan Islands are astonishingly rich in marine diversity. However, these reefs are threatened by humans. Destructive fishing methods, such as trawl, blasting and cyanide fishing practise, are found to be the main cause of this degradation. The coral reefs habitat reduction is also caused by tourism activities due to trampling over the reef and charging organic and anorganic waste.
Remote sensing offers an effective approach to compliment the limitation of field sampling, in particular the monitoring of the reefs in remote sites. Moreover, using achieved remotely sensed data; it is even possible to monitor the historic status of the coral reef environment.
The capabilities of satellite remote sensing techniques combined with field data collection have been assessed for generating coral reef habitat mapping and change detection of Derawan Island. Multi- temporal Landsat TM & ETM images (1991 & 2002) and a very high spatial resolution multi-spectral QuickBird image (October 2003) have been used.
The capability of QuickBird image to generate a coral reef habitat map with water column correction by applying the Lyzenga method, and also without water column correction by the applying maximum likelihood method, have been assessed. The classification accuracy of the coral reef habitat map increased after compensation of the water column effects. The classification of QuickBird image for coral reef habitat mapping increased 22% by applying a water column correction.
The classification of the coral reef habitat with multi-spectral QuickBird image data using in situ spectral reflectance has been tested. Wet bottom reflectance has been generated from the QuickBird image using the modified of Bierwirth method. Jupp method of depth or penetration zones has been applied to obtain the depth as the input parameter for the modified Bierwirth method. The result showed that the wet bottom reflectance values were over estimated; therefore it was not possible to link them to in situ spectral reflectance. It is concluded that the modified Bierwirth method is very sensitive to the depth input parameters and that therefore the crude bathymetric map derived from Jupp‘s depth of penetration zones method could not be used as input for the modified Bierwirth method. The classification using in situ spectral reflectance was tested to classify the full atmospheric correction of QuickBird image. In the classified image it was possible to differentiate sand, coral and algae. The produced habitat map has an accuracy of 84%.
Multi-temporal Landsat images of 1991 and 2002 have been used to assess the habitat shifting in Derawan Island. Comparison of the classified images of 1991 and 2002 shows spatial changes of the habitat. The changes were in accordance with the known changes in the reef conditions. The analysis shows a decrease of the coral reef and patchy seagrass percentage and on other hand the increasing of
i The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
algae composite and patchy reef percentage. The result of the study is a good example on how time series analysis of remotely sensed data can be a tool to monitor the historic status of the coral reef environment.
Key words: Coral reefs, classification, habitat changing, Derawan Island, QuickBird, Landsat.
ii The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Acknowledgm ents
I would like to thank the government of the Netherlands through the STUNED fellowship programme, Netherlands Education Centre, Jakarta, for the full financial support of my study in the Netherlands. Also thanks to Dr. Stefano Fazi, Dr. Jan Steffen and Prof. Dr. Stephen Hill who had given me recommendations for pursuing the scholarship.
I would like to express my gratitude to Drs. Michel Damen his guidance, reviews, effective research schedule and also his encouragement along the way. His critical comments had directed my research to be more focused.
I would like to express my great appreciation to Dr. Ir. Tjeerd Hobma for giving me the opportunities to pursue my research interest in his project, for his guidance during the fieldwork preparation, and also for the great efforts he made during the fieldwork.
My thanks go to Drs. Eduard Westinga from the Division of Natural Resources Management for his comments and suggestions regarding coral reefs issue.
My sincerely thanks for Drs. Boudewijn van Leeuwen from the ITC Remote Sensing Lab., for assisting me with the image acquisition, and the many questions I had with regard to atmospheric correction. Thanks to Drs. Harald van der Werff for his valuable comments regarding spectral reflectance.
My appreciation to Drs. Sam Purkis for his valuable inputs regarding the application of modified Bierwirth method.
Also thanks to all the staff of the ESA Department for their support and help. In particular Drs.Nanette Kingma, Drs. Robert.P.G.A. Voskuil, Dr. David. Rossiter and Dr. Paul van Dijk.
I also want to express my acknowledgements to the partner groups of the project, East Kalimantan Programme 2003, and the people I met during the fieldwork. Special thanks to Dr. Bert Hoeksema for his great coordination during the fieldwork and also his generosity to lend some research equipments. Thanks for Prof. Dr. Rolf P.M. Bak, Ibu Yosephine and Pak Suharsono and Dr. Emre Turak for interesting discussions in the field. Thanks to the dive guides of Derawan Dive Resort for their assistance during the data collection. Thanks to the people in The Nature Conservancy, Pak Budi, Irfan and Handoko for sharing their information.
Many thanks also to my friends and colleagues from the Indonesian Coral Reefs Foundation. For Mbak Via, Kiki, Cepo and Nugi who had been very helpful in the providing of research equipments. Also special thanks for Tony, for his incredible assistance during the field work preparation.
Thanks also for Drs. Sybrand van Beijma and Drs. Arjan Reesink for their great assistance in the field, particularly for spectral data collections.
iii The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
My special thanks to my fieldwork buddy, Ana Fonseca for her cooperative and very fruitful discussions during the fieldwork and data analyses.
My large gratitude also goes to friends in ITC, Maria, Brando, Nicky, Maria Fe, Umut and also my compatriot friends, ITC Indonesia (2001,2002 and 2003). In particular Dessi, Mas Syarif, Mbak Ida, Budi, Ismail, Mbak Retno,Mas Hartanto, Mas Bobby, Mbak Yanti, Anggoro, Nia and Indra. Thank you for the warm and friendly atmosphere during the entire study.
Last but not least, I would like to express my heart-felt thanks for cintaku Emmanuel Duguey and my beloved family. Thank you for all the encouragements, supports and that you prayed for me. You are important persons in my life.
iv The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Table of Contents ABSTRACT...... I ACKNOW LEDGMENTS...... III TABLE OF CONTENTS...... V LIST OF FIGURES...... VIII 1. INTRODUCTION...... 1
1.1. PROBLEM DEFINITION AND JUSTIFICATION ...... 2 1.2. RESEARCH OBJECTIVES...... 3 1.3. RESEARCH QUESTIONS AND HYPOTHESIS...... 4 1.3.1. Research Questions...... 4 1.3.2. Hypothesis...... 4 1.4. RESEARCH APPROACH...... 5 2. LITERATURE REVIEW ...... 6
2.1. REMOTE SENSING FOR SHALLOW WATER HABITAT ...... 6 2.2. REMOTE SENSING TECHNIQUES OF CORAL REEF HABITAT ...... 7 2.3. SPECTRAL DISCRIMINATION FOR CORAL REEF HABITAT ...... 9 2.4. REMOTE SENSORS FOR CORAL REEF HABITAT MAPPING ...... 10 3. THE STUDY AREA...... 12
3.1. INRODUCTION...... 12 3.2. HYDRO OCEANOGRAPHY AND CLIMATIC CONDITION ...... 12 3.3. SHALLOW MARINE HABITAT ...... 13 3.4. ENVIRONMENTAL THREATS IN DERAWAN ISLANDS ...... 14 4. AVAILABLE DATA AND FIELD W ORK ACTIVITIES...... 16
4.1. DESCRIPTION OF THE AVAILABLE DATA...... 16 4.1.1. Remote Sensing Data ...... 16 4.1.2. Aerial Visibility Data ...... 16 4.1.3. Bathymetric Data ...... 17 4.1.4. Tidal Data ...... 17 4.2. FIELDWORK ACTIVITIES...... 17 4.2.1. Ground Control Points Collection...... 18 4.2.2. Water Depth Measurements and Habitat Ground Truth Points Collection...... 18 4.2.3. Habitat Transect Data Collection...... 20 4.2.4. Spectral Measurements...... 20 5. RESEARCH M ETHODOLOGY...... 23
5.1. GEOMETRIC IMAGE CORRECTION ...... 23 5.2. ATMOSPHERIC CORRECTION ...... 24 5.2.1. ATCOR Approach for Landsat TM Images...... 24 5.2.2. Dark Reflectance Substraction Approach for QuickBird Image...... 28 5.3. LAND, CLOUD AND DEEP WATER MASKING ...... 32
v The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
5.4. WATER COLUMN CORRECTION ...... 33 5.4.1. Depth Invariant Index...... 33 5.5. MODIFIED OF BIERWIRTH METHOD ...... 34 5.6. ESTIMATION OF THE BATHYMETRY...... 35 5.7. CATEGORIZATION OF HABITAT CLASSES ...... 38 5.8. IMAGE CLASSIFICATION ...... 41 5.9. ACCURACY ASSESSMENT ...... 43 6. RESULTS AND DISCUSSION...... 44
6.1. GENERATING A CORAL REEF HABITAT MAP WITH AND WITHOUT WATER COLOUMN CORRECTION USING QUICKBIRD IMAGE OF 2003...... 44 6.1.1. Generating a coral reef habitat map without water column correction...... 45 6.1.2. Compensate for the water column effect on the image...... 45 6.1.3. Generating a coral reef habitat map with water column correction...... 47 6.1.4. Concluding Remarks...... 49 6.2. GENERATING A CORAL REEF HABITAT MAP USING IN SITU SPECTRAL DATA...... 50 6.2.1. Generating the bathymetric map...... 51 6.2.2. Deep water reflectance...... 56 6.2.3. Generating wet bottom reflectance from QuickBird Data ...... 56 6.2.4. In situ spectral reflectance data preparation...... 57 6.2.5. Generating a coral reef habitat map...... 58 6.2.6. Concluding Remarks...... 60 6.3. GENERATING A CORAL REEF HABITAT MAP FOR HABITAT CHANGING ASSESSMENT ...... 61 6.3.1. Concluding Remarks...... 64 7. CONCLUSIONS AND RECOM M ENDATIONS...... 65
7.1. ASSESSING THE CAPABILITY OF QUICKBIRD IMAGE TO GENERATE A CORAL REEF HABITAT MAP WITH WATER COLUMN CORRECTION ...... 65 7.2. INTEGRATING IN SITU SPECTRAL DATA WITH MULTI-SPECTRAL QUICKBIRD IMAGE DATA FOR CORAL REEF HABITAT CLASSIFICATION...... 66 7.3. ASESSING THE CHANGE OF THE CORAL REEF HABITAT BY COMPARING CLASSIFIED LANDSAT IMAGES OF 1991 AND 2002...... 66 7.4. RECOMMENDATIONS ...... 67 REFERENCES ...... 68 APPENDIX A - TEAM OF EAST KALIMANTAN PROGRAMME 2003...... 70 APPENDIX B œ VISIBILITY DATA...... 72 APPENDIX C œ TIDAL DATA OCTOBER 2003...... 73 APPENDIX D œ EXAMPLE OF CONVERTION MEASURED DEPTH TO THE DEPTH OF IMAGE ACQUISITION...... 74 APPENDIX E œ EXAMPLE OF SORTED THE DEPTH...... 75 APPENDIX F œ EXAMPLE TO DETERMINE DOP ZONE ...... 76 APPENDIX G œ EXAMPLE OF SAND SUBSTRATE PIXELS AT VARIOUS DEPTHS...... 78 APPENDIX H œ EXAMPLE TO DETEMINE LIMIN AND LIMAX OF EACH DOP ZONE (TABLE 6.7)...... 79
vi The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
List of Tables
3.1. Live coral cover percentage in transect locations… … … … … … … … … … … … … .. 14 4.1. Available Remote Sensing.Data… … … … … … … … … … … … … … … … … … 16 4.2. The field equipment and measured parameters… … … … … … … … … … … … … … . 18 5.1. RMS error of geometric corrected images… … … … … … … … … … … … … … … … . 23 5.2. Parameters of TM5 1991 and ETM7 2002 images… … … … … … … … … … … … … 25 5.3. Calibration parameter for Landsat TM5 and Landsat ETM7 images… … … … … … 26 5.4. Input parameters to calculate the satellite reflectance of QuickBird image, October 2003… … … … … … … … … … … … … … … … … … … … … … … … … … … . 30 5.5. Rg, rdw and x values of eachband… … … … … … … … … … … … … … … … … … … 31 5.6. The maximum DOP zi… … … … … … … … … … … … … … … … … … … … … … … ... 37 6.1. Parameter values to calculate ration attenuation coefficient between band 1 and band 2. 46 6.2. Error matrix of classified image without water column correction… … … … … … ... 49 6.3. Error matrix of classified image with water column correction… … … … … … … … 49 6.4. DN value for deep water of each band… … … … … … … … … … … … … … … … … .. 51 6.5. The maximum depth of penetration (zi) in metres… … … … … … … … … … … … . 52 6.6. A decision tree to assign pixels to depth zones… … … … … … … … … … … … … .. 52 6.7. Li min and Li max for each DOP zone i derived from QuickBird image… … … 53 6.8. The values of ki and Ai derived from QuickBird image… … … … … … … … … ... 54 6.9. Max, min and means plus-minus of standard deviation in band 1… … … … … … . 58 6.10. Error matrix of classified image derived from in situ spectral reflectance… … … 60
vii The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
List of Figures
1.1. Flow chart of research approach… … … … … … … … … … … … … … … … … … … 5
2.1. Factors influencing the amount of radiance reaching at sensor over a water mass… … … … … … … … … … … … … … … … … … … … … … … … … … … … … … 6 2.2. Scatter plot of sand values between two bands… … … … … … … … … … … … … .. 7 3.1. Map of Derawan Islands, Berau Region œ East Kalimantan… … … … … … … … . 15 3.2. Study Area… … … … … … … … … … … … … … … … … … … … … … … … … … … … . 15 4.1. Ground control points plot in QuickBird image panchromatic band… … … … … 19 4.2. Depth measurement sampling points and ground truth habitat points… … … … … .. 19 4.3. Transect habitat approach; sampling location sites and spectral reflectance measurements… … … … … … … … … … … … … … … … … … … … … … … … … … . 21
5.1. Spectral reflectance versus various visibilities and comparison with sea spectral 27 library… … … … … … … … … … … … … … … … … … … … … … … … … … … … … . 5.2. Comparison of histogram band 1 before and after applied atmospheric correction. 28 5.3. A Graphic of ground reflectance values of deep water (Rg)… … … … … … … … . 30 5.4. Comparison of the histogram of band 1 QuickBird before (a) and after (b) the application of atmospheric correction… … … … … … … … … … … … … … … … … 31 5.5. The masking land, deep water and cloud of QuickBird image for band 1… … ... 33 5.6. Depth of Penetration (DOP) zones… … … … … … … … … … … … … … … … … … 36 5.7. Four steps of derived bathymetric map… … … … … … … … … … … … … … … … … . 36 5.8. Description of habitat classes… … … … … … … … … … … … … … … … … … … … . 40 5.9. Four steps of supervised classification… … … … … … … … … … … … … … … … .. 42 6.1. Classification steps for generating habitat map without water column correction (a) and with water column correction (b)… … … … … … … … … … … … … … … .. 44 6.2. Scatter plot of sand substrate at various depth between band 1 and band 2 (a), and between band 2 and band 3 (b)… … … … … … … … … … … … … … … … … … … .. 45 6.3. Depth invariant index histogram and image of a pair band 1 and band 2… … … . 47 6.4. Coral reef habitat map of QuickBird image, without water column correction… .. 48 6.5. Coral reef habitat map of QuickBird image, with water column correction… .… 48 6.6. Some input parameters to apply modified Bierwirth algorithm… … … … … … … 50 6.7. (a) the masking map of DOP2, (b) the zone of DOP2… … … … … … … … … … … .. 53 6.8. DOP depth images of zone 1 (a), zone 2 (b), zone 3 (c) and zone 4 (d)… … … … 54 6.9. Crude bathymetric map of Derawan Island derived from QuickBird image 2003.. 55
viii The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
6.10. Scatter plot between predictive depth and measured depth (m)… … … … … … … 55 6.11. Curve of Rw measured in the field… … … … … … … … … … … … … … … … … … 56 6.12. Rb derived from QuickBird image of band 1 (a), band 2 (b) and band 3(c)… … .... 57 6.13. Means of measured spectral reflectance… … … … … … … … … … … … … … … … 57 6.14. Means of merged measured spectral reflectance… … … … … … … … … … … … ... 58 6.15. Coral reef habitat derived from wet bottom reflectance of band 1 (Rb1)… … … … . 58 6.16. Histogram of wet reflectance in band 1 (Rb1)… … … … … … … … … … … … … … .. 59
6.17. Classified habitat map derived from measured spectral reflectance and RA band 1 59 6.18. Coral reef habitat map of Derawan Island 1991 (1,2,3-a) and 2002 (1,2,3-b)… … . 62 6.19. Threats in Derawan Islands… … … … … … … … … … … … … … … … … … … … … .. 63 6.20. Coral reef habitat classes shifting between 1991 and 2002… … … … … … … … … .. 63
ix The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
1. Introduction
Coral reefs are one of our earth‘s fundamental resources. Due to their rich habitat and large diversity of marine species and ecological complexity, they are frequently compared to tropical rainforest. Coral reefs are an important asset for millions of people around the world, particularly for local communities who depend on these resources for their livelihood. The reef ecosystems provide pharmaceuticals, as a source of food, generating income from tourism, and as a buffer for coastal cities and settlements from storm damage.
Coral reefs are distributed in the tropical and sub-tropical coastal waters, mostly in developing countries. The greatest diversity of coral reefs is in South-east Asia, ranging from the Philippines to the Great Barrier Reef in Australia. Indonesia‘s coral reef coverage is estimated to be roughly 85.700 km2, which is about 14% of the world‘s total area (Tomascik et al. 1997). They have a high biological diversity in the marine environment and perform as an important basis for sustainable development.
The health of the world‘s coral reefs is in serious decline. Approximately 11 percent of coral reefs with a high level of marine diversity are under threat, including reefs in the Philippines, Indonesia, Tanzania, the Comoros, and the Lesser Antilles in the Caribbean (Bryant et al. 1998). Principally, human activities are the main cause of the coral reef degradation. These include destructive fishing methods, over fishing, inappropriate inland management; human and industrial waste disposal, coral mining and even careless diving activity. Global warming also may enforce reefs stress, such as increased flood and storm events and a rise of the seawater temperature.
Information on the health of coral reefs status is crucial for their conservation and sustainable utilization. Unfortunately, in most cases only a small amount of this information is available. The International Coral Reef Initiative (ICRI) Framework for Action emphasizes the needs of research and monitoring of the coral reefs status. Particularly the reefs in tropical ecosystem because reefs in this area are still not well understood compare to the temperate system (ICRI 1995).
A monitoring program to detect changes of the coral reef environment is essential to promote sustainable management. The monitoring has to be repeated over time at the same location. An appropriate time scale is needed to study statistically changes in the main variables but also to determine ecologically meaningful changes (Reese and Crosby 2000).
In the common approach for the monitoring of coral reef environments, field sampling, has its limitations. It requires large numbers of transects data to monitor the extent of the reef environment. It can be costly and labour intensive and can be even more complicated if the location is in a remote area. Remote sensing offers an effective approach to compliment the
1 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. limitation of field sampling, in particular the monitoring of the reefs in remote sites. Moreover, using achieved remotely sensed data; it is even possible to monitor the historic status of the coral reef environment.
1.1. Problem Definition and Justification Coral reefs in Derawan Islands are astonishingly rich in marine diversity. The high diversity of corals and fishes, the presence of turtles and Kakaban- salt lake makes Derawan Islands one of the most famous diving sites. One of the outcomes of the —World Heritage Marine Diversity“ Workshop, in Hanoi 2002, stated that the Derawan islands area has a high priority to be nominated as World Heritage Site. There are about 120 areas in the world identified as potential sites of outstanding universal values. Within those, 25 sites ranking in Southeast Asia, Derawan islands area among the top seven (UNESCO 2002).
However, these reefs are threatened by humans. WWF Indonesia (2002) reported there are 4 main threats in Derawan Islands: 1) Habitat reduction, 2) Fisheries activities, 3) Tourism and 4) Policy. East Kalimantan once had approximately 950,000 ha of mangrove-nipa forest formation and only 266,800 ha in 1980 (Mac Kinnon, et.al cited in Jompa and Pet-Soede 2002). These mangroves have been converted or degraded due to logging; the making of fishponds, human settlements and industrial facilities. As a result, the mangrove function as the barrier for erosion and as a filter of sediment also decreased. Destructive fishing methods, such as trawl, blasting and cyanide fishing practise, are found to be the main cause of this degradation. The coral reefs habitat reduction is also caused by tourism activities due to trampling over the reef and charging organic and anorganic waste. Moreover unwise practice development decisions also contribute to habitat degradation, such as stock collapse and habitat reduction due to intensification and industrial development (Ismuranty 2002; Jompa and Pet-Soede 2002).
To ensure the survival and well being of Derawan Islands‘s high biodiversity, it is essential to promote sustainable utilization. A monitoring program is necessary to assess the condition of the coral reefs. This information is very important for the identification of the management options of the Derawan Islands.
Until now there exist no monitoring program in this area, using geo-spatial information and remote sensing techniques. Some coral reefs studies have been conducted in Derawan Islands by different institutions. However, the assessments were conducted in different sites and only in one time period. Furthermore, there are only a few reefs studies on Derawan Islands that are using geo-spatial information.
The International Institute for Geo-Information Science and Earth Observation (ITC) is one of the counterparts of the East Kalimantan Programme for Coastal Zone Research Netherlands- Indonesia (EKP); a join research between the Netherlands and Indonesia institutions funded by The Netherlands Foundation for the Advancement of Tropical Research (WOTRO). East Kalimantan Program is accordance within a framework of international global change research program. The EKP research theme are; 1. ecosystem and biodiversity, 2. sedimentary systems and morphodynamic, and 3. Governance and management with special
2 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. focus on mangrove and fish-ponds. The institutions that involved in this project are listed in appendix A.
The ITC component more focuses on theme sedimentary and morphodynamic. This study aims at the development of a methodology, combining multi sensor satellite data and in-situ measurements, to detect trends in morphodynamics and explain processes of mixing and transport between river waters and ocean waters, along the EastKalimantan coast, which focus on the Mahakam and Berau delta. Although ITC component is under theme 2; sedimentary systems and morphodynamics, the remote sensing approach links with research theme 1. ecosystems and biodiversity and 3. governance and management.
This thesis is a contribution to the project which a combination under ITC objective component and ecology and biodiversity. It concentrates on remote sensing approach strongly links with ecosystem in the barrier reef habitat front of Berau delta.
1.2. Research Objectives The main objective of the study is to assess the capabilities of satellite remote sensing techniques combined with field data collection for coral reef habitat mapping and change detection of Derawan Island using multi-temporal Landsat TM & ETM images (1991 & 2002) and a very high spatial resolution of multi-spectral QuickBird image (October 2003).
In order to achieve the main objective, some specific objectives have been defined: ñ To assess the capability of QuickBird image to generate a coral reef habitat map with water column correction by applying the Lyzenga method; and without water column correction by applying maximum likelihood method. ñ To integrate in situ measured spectral data with multi-spectral the QuickBird image data for coral reef habitat classification. ñ To assess the coral reef habitat change by comparing classified images of Landsat TM of 1991 and 2002.
3 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
1.3. Research Questions and Hypothesis 1.3.1. Research Questions This research has the following specific objectives and research questions: Specific Objectives Research Questions
How accurate is the coral reef habitat map of the To assess the capability of QuickBird QuickBird image with and without water image to generate a coral reef habitat map column correction? with water column correction by applying the Lyzenga method; and without water How much will the accuracy of the coral reef column correction by applying maximum habitat map of the QuickBird image improve likelihood method after applying the water column correction technique?
Is it possible to integrate in situ reflectance of different coral reef habitats and to produce from To integrate in situ measured spectral data that a coral reef habitat map using QuickBird with multi-spectral the QuickBird image image data? data for coral reef habitat classification How much is the accuracy of predictive coral reef habitat map derived from the in situ spectral data classification?
Is there any coral reef habitat change in period To assess the coral reef habitat change by 1991 œ 2002, which can be detected by Landsat comparing classified images of Landsat remote sensing data? TM of 1991 and 2002. What are the trends of coral reef habitat changing in the period 1991 œ 2002?
1.3.2. Hypothesis. ñ Water column correction technique can generate a more accurate coral reef habitat map derived from QuickBird data. ñ It is possible to integrate in situ measured spectral data with QuickBird data for coral reef habitat classification. ñ There is a change in the coral reef habitat of Derawan Island between 1991 and 2002, which can be detected by multi-temporal Landsat images.
The applied methods to reach the objectives are described in chapter 5.
4 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
1.4. Research Approach
OFFICE DATA COL L ECTION
Satellite Literature Ancillary Images Review Data
IM AG E P R OCES S IN G FIEL D DATA
Geometric Ground Control Correction Points ATCOR for TM91 and ETM 02 Atmospheric Deep-water Correction reflectance Dark pixel Subtraction for QB
Masking
Water column correction
Bathymetric Map Using depth Water depth penetration zone Algorithm Calculation by Lyzenga Method For TM 91, ETM 02 Modification and QB Bierwirth Method For QB
Ground truth Habitat Classification points
2 Density slicing of image derived from Maximum Likelihood Density slicing of image Bottom feature Lyzenga method (Map QB1) derived from mod. reflectance (Map QB2, MapTM91 Bierwirh method & Map ETM02) (Map QB3)
DATA ASSESSMENT 1 Accuracy Assessments
Habitat Changing Assessment 3 (TM 91 & ETM 02)
EV AL U ATION AN D R EP OR T W R ITIN G Figure 1.1. Flow chart of research approach The number indicating research objective; 1. To assess the capability of QuickBird image to generate a coral reef habitat map with and without water column correction, 2. To integrate in situ measured spectral data with multi-spectral QuickBird image for coral reef habitat classification. 3. To assess habitat changing of the coral reef habitat by comparing classified images of Landsat 1991 and 2002.
5 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
2. Literature Review
2.1. Rem ote Sensing for shallow water habitat In remote sensing of shallow water habitat mapping one has to deal with the influence of the atmosphere and the water column. The radiation must pass through two media, the atmosphere and water and pass it again on its way back before being recorded at the sensor. Therefore to identify bottom reflectance, the image should be corrected for atmospheric and water column effects.
Figure 2.1. Factors influencing the amount of radiance reaching the sensor over a water mass (Edward, 1999)
6 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
2.2. Rem ote sensing techniques of coral reef habitat One of the most important requirements for the optimum growth of coral reef is clear water. This allows sunlight to the reach coral reef symbiotic algae living in coral tissues to photosynthesize. This condition allows optical remote sensing to detect the various coral reef habitats.
In general the reflectance of shallow water substrate to solar illumination is a function of; 1). Substrate reflectance, 2). Water depth and 3). Water optical properties (e.g. organic matter, suspended sediments, and dissolved substances). (Lyzenga 1981).
There exist various effects of scattering and absorption in the water column due to differences in the optical properties in water column, related to the water depth. Lyzenga (1981) proposed a method called depth-invariant index to enhance bottom type information. This concept is based on the fact that bottom reflected radiance is a linear function of the bottom reflectance and an exponential function of the water depth. To make a linear relationship between the radiance and the depth, the radiance values (which has been atmospheric corrected) are transformed using a natural algorithm (ln). Since the homogeneous substrate values are constant, when plotting two bands, the pixel values for each band will fall on a straight line. The variations of the values along the line occur due to the change of the depth (see figure 2.2). The slope of the line represents the ratio of the attenuation coefficient between the two bands. In this method the optical property assumed relatively similar in water due to horizontal mixture. (Lyzenga 1981).
Figure 2.2. Scatter plot of sand values between two bands (Lyzenga 1981)
The depth invariant index method has been tested by (Spitzer and Dirks 1987) for its algorithm sensitivity for turbidity. It was concluded that in the general the depth invariant index algorithm has a low sensitivity for the variation of suspended matter in the water column.
(Mumby et al. 1998a), combined the depth invariant index method and contextual editing for the mapping of Carabbian coral reefs using multi sensor satellite and airborne images. Contextual editing can improve the classification accuracy based on the knowledge that a
7 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. certain cover type is predicted to be present in a particular zonation. For example, in this study misclassified areas of seagrass that were present in fore reef slope were reclassified to the more appropriate reef category. These combinations of methods increased the classification accuracy of both satellite and airborne images by 17% and 22% respectively.
(Bierwirth et al. 1993) proposed a new method, in which one can simultaneously detect the water depth and the relative substrate bottom reflectance from remote sensing images. Water attenuation coefficients were derived from a regression analysis of bathymetric data against the radiances of remotely sensed data. The water optical property was assumed to be constant within the image.
The water attenuation coefficient (k), is an important factor for water column correction. It considers the effect that the absorption and scattering, within the water column, have on the radiance recorded at the sensor. (Bierwirth et al. 1993);(Maritorena 1996)
The attenuation coefficient can be estimated in two ways. The first estimation is from in situ irradiance measurements of down welling irradiances at different depths. A second estimation is derived from a depth-invariant index method. The first method can generate a better coefficient estimation; however the latter method is easier and more efficient to apply. (Maritorena 1996)
Another method is proposed by (Khan et al. 1992) for bottom substrate mapping based on axe rotation using principal component (PC) transformation. This PC transformation is based on the analysis of two water-penetrating bands, which have a high correlation; it rotates the axes to produces a set of images bands, which are uncorrelated. This method reduced the effect of water depth variation on bottom substrate, and is able to differentiate rock, coral rubble, seagrass and sand in Abu Ali bay, Arabian Gulf.
Another alternative for water column correction is the estimation of the depth from multi - spectral satellite imagery. (Mumby et al. 2000) assessed three methods to predict water depth; Benny and Dawson (1983) method, Jupp (1988) method and Lyzenga (1978) method. It is concluded that Jupp‘s method is the most accurate compared to the other tested methods.
Jupp‘s method assumes that; 1). Light attenuation is an exponential function of depth, 2). Water attenuation does not vary within the image, and 3). Albedo of the substrate is relatively constant. The depth is predicted by calculating depth of penetration (DOP) zones for each visible and near infra-red band of the multi-spectral image. DOP zone is determined from maximum deep-water radiance of each band. (Jupp, 1988 cited in Edward 1999).
(Purkis and Pasterkamp 2003) used a Landsat 5 TM image to investigate the advantage of five level image-processing methods in relation to the increase of accuracy of the generated habitat maps with its level processing. The five-level of image-processing consist of 1). atmospheric correction, 2). Water/air boundary refraction correction, 3). Water column correction œ influences of water optical properties 4). Water column correction œ influences of depth (depth values were derived from interpolation of depth data) and 5). Water column
8 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. correction œ influences of depth (in situ measurement of depth values). The habitat map accuracies of each processing level (1 to 5) are 19%, 19%, 40%, 47% and 71% respectively.
Although water column correction is an essential method for the detection of bottom substrate reflectances for coral reef habitat mapping, (Mumby et al. 1998a) noted that these pre- processing procedure is not commonly adopted. They found only four studies out of 45 relevant papers that have applied this method.
2.3. Spectral discrim ination for coral reef habitat
Another remote sensing method for coral reefs is spectral reflectance measurements, in situ or in the laboratory, using a spectrometer.
(Holden and LeDrew 1998), assessed the spectral discrimination between healthy and non- healthy coral. In situ spectral reflectances of 22 representative healthy and bleached corals, dead coral debris and dead coral with algae have been measured in a lagoon in Fiji. Principal components analysis and derivative spectroscopy were used to assess the spectral differences.
(Hochberg and Atkinson 2000) made a spectral discrimination of major types of reef benthic communities; coral, algae and sand. They measured 247 spectral reflectance in Kaneohe Bay, Oahu, Hawaii, representing three coral species, five algae species and three sand benthic communities. They applied a derivative analysis from Savitsky-Golay method and were able to detect the differences of reflectance between coral, algae and sand. The results of the spectral analysis were applied to narrow non-contiguous wavelength bands of the airborne hyperspectral image. To compare the result of habitat map from the image to the in situ habitat mapping, they subset particular patch reef area of the image. Ground truth measurements were conducted within the area of subset patch reef in the image. The ground truth consist of 144 underwater line transects with a total of 2090 data points. The comparison of the percentage cover of coral, algae and sand between ground-truth data analyses and airborne image are 85.4, 10.0 and 4.6 % respectively in ground-truth analysis and 92.4, 5.0 and 2.6% respectively for the airborne images.
A rigorous assessment of spectral reflectance for coral reef bottom-type was made by (Hochberg et al. 2003). They did 13,100 in situ measurements of the spectral reflectance in the Atlanctic, Pacific and Indian Oceans. The measured spectral reflectances were categorized in 12 basic reef-bottom types, such as fleshy (brown, green, red) algae; non-fleshy encrusting calcareous and turf algae; bleached, blue and brown hermatypic coral; soft/gorgonian coral; seagrass; terrigenous mud; and carbonate sand. They applied a derivative analysis from Savitsky-Golay method to examine the spectral pattern and partition method to separate the spectral characteristic of each bottom type. They found that the spectral reflectance of 12-reef bottom types could be separated and independent across bio- geographic regions.
(Purkis and Pasterkamp 2003) conducted a study to integrate in situ spectral reflectance measurement swith space-borne satellite imagery for habitat mapping. In situ spectral
9 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. reflectance of mixed features (reef-top reflectance) measurements have been collected in Marsa Shagra, Egyptian Red Sea. A subset of a Landsat TM image was corrected up to wet bottom reflectance values using a modification of Bierwirth method. Ground spectral measurements were used to classify the corrected Landsat image. Seven habitat classes were determined, consist of sand, aggregated rubble, two classes macro algae community, two classes of seagrass with different density and alga mats. The produced habitat map has an accuracy of 71% .
2.4. Rem ote sensors for coral reef habitat m apping
Besides correcting the image from atmospheric and water column effects, the accuracy of coral reef habitat mapping also depends on the spectral and spatial characteristics of the remote sensor itself.
(Maeder et al. 2002), assessed the capability of IKONOS satellite images (multi-spectral 4 by 4 m resolution) for coral reef mapping in Roatan island, Honduras. A Maximum likelihood classification was applied, by which it was possible to discriminate sand, coral reef of two different depths (< 5 m and more than 5 m) and seagrass. The overall accuracy of the habitat map was 89%.
(Mumby et al. 1998a), used various satellite images (MSS, Landsat TM, SPOT-XS and SPOT-Pan), airborne multi-spectral imagery (CASI) to assess the effect of water column correction and contextual editing for mapping coral reefs in the Turks and Caicos Islands, British West Indies. Extensive ground truth survey points (up to 600 points) were collected. They noted that the coarse spatial and spectral resolution, particularly MSS and SPOT pan, caused spectral mixture within the pixel, which made the classification difficult. The poor spatial and spectral resolution also had constrains if applied to water column correction due to its requirement of pairs of visible bands as input data.
(Dobson and Dustan 2000) examined shift detection for coral reefs in the Florida Marine Sanctuary Reefs. A time series of 20 Landsa-TM images were analysed using a temporal- texture deviation processing technique. This technique generates texture values in the spatial domain. A high texture values represent a change and low texture values represent stability. They concluded that with the Landsat sensor, it is possible to detect the changes in reef- community level and that can offer a diagnostic tool to monitor coral reef conditions. In term of temporal context, the changing only occurs in two directions: predecessor and successor, therefore the type of habitat change could not be identified.
(Hochberg and Atkinson 2003) investigated the capability of multi-spatial and spectral sensors to classify three basic classes of reef community (coral, algae and carbonate) based on in situ spectral reflectance measurements for assessing the global reef status. They used for the test, two airborne hyperspectral sensors (AAHIS and AVIRIS), three satellite broadband multi-spectral sensors (IKONOS, Landsat ETM and POT-HRV), and two hypothetical satellite narrow band multispectral sensors (Proto and CRESPO). The spatial resolutions of these sensors are 2 , 2, 20, 10, 4, 30, and 20 m for AAHIS, AVIRIS, Proto, CRESPO, Ikonos,
10 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Landsat ETM, and SPOT-HRV, respectively. The analysis showed uncertain coral class in large pixel images due to various levels of spectral mixing. However, small pixels (2 x 2 m) gave much less spectral mixing. In terms of spectral resolution, images from hyperspectral sensors can provide a very high spectral contrast between coral and algae, thus shows statistically a more accurate areal cover. Ikonos, Landsat ETM and SPOT-HRV gave errors in the estimation of the areas of coral cover due to misclassifications of pure or mixed pixels, between algae and coral.
11 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
3. The Study Area
3.1. Inroduction
The Derawan islands region is a sub-district (Kecamatan) of the Berau district of the East Kalimantan Province, Indonesia. Geographically, it is situated between118º 09‘ 53‘‘ - 118º46‘ 28‘‘ North and the 02º25‘45‘‘ - 02º03‘49“ East.
The islands are situated in two different systems: a shelf reefs system, which has developed as barrier reef, known as the Berau barrier reef, and tectonically uplifted atolls. The Berau barrier reef system is found in the Northern part of the Berau delta, part of Mangkalihat Peninsula; and it extends about 60 km to the South. It is located at the inner side of the 200 m iso-depth at the margin of the Sunda shelf (Tomascik et al. 1997).
The islands of Derawan, Rabu-rabu, Panjang, Samama and Sangalaki are part of Berau barrier reef. The two up lifted atolls are Maratua, with an open lagoon and Kakaban, with a closed lagoon. (See figure 3.1.)
Derawan island is a small island with a total area of 48,70 (ha) and a 2,7 km long shore-line (Based on the recent survey of The Nature Conservation, 2003); it has a flat topography. The sandy beach of the island has a width of 13,5 œ 20 m. The main substrate material of the island is coarse sand (45,7 %), composed of fragments of coral. (LIPI 1995)
This study focuses on the coral reefs of Derawan Island and Masimbung reef and Tubabinga reef in the southern part of Derawan Island. (See Figure 3.2).
3.2. Hydro Oceanography and Clim atic Condition
The climate and oceanography in the Derawan region is highly influenced by the climatic conditions of the Pacific Ocean. In general there are two kinds of climate: West season and East season, depending on the wind direction. The lowest wind speed in Derawan region are found in October and November and the maximum wind speed are in July and August.
Indonesian Through Flow (ITF) from the Pacific Ocean to the Indian Ocean, which passes trough Makassar Strait, influences the oceanographic condition of the Derawan region. ITF affects the dynamic process in Berau River and Makassar Strait such as the longs-shore current along the East Kalimantan coast toward the southward direction of Makassar Strait.
12 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Based on a survey by LIPI (1995), the hydro oceanographic aspects of Derawan Islands are divided in two areas: 1) The area influenced by the Berau river, and 2). The area influenced by the sea. a. Temperature The areas surrounding Derawan Island (Panjang and Derawan islands and reefs surrounding Derawan) have a large influence from the sea. The sea surface temperature ranges between 29,5èC œ 30èC and the sea bottom temperature is between 21èC œ 28èC. The area close to Berau river has a range of 29,5èC œ 30,5èC at the surface and 27,5èC œ 29èC at the bottom. There is no large different in surface temperature between the two areas. b. Salinity The surface salinity value in the area close to the mouth of the Berau river is 32,5 - 33è/èè and the average range in the offshore area is 33,5 è/èè. At the bottom part (100m depth), the area close to the river has a salinity value of 33,5 è/èè and 34-34,5 33,5è/èè for the offshore area. This difference of salinity between the two areas is caused by fact that the area close to Berau River has a low salinity value due to the mixture of fresh and sea water near the river out let. c. Oxygen There is no big different of dissolved oxygen value between the two areas either at the surface or at the bottom. The average value is 3,5 œ 4,5 ml/L. The homogenous dissolved oxygen values indicate that there is no influence from the river and the sea to the total amount of this element. d. Nitrate The average nitrate value (NO3) at the surface in the area close to the river and more offshore is similar 0,4 œ 1,8 mg/L. The value at the bottom (depth approximately 100 m) for close to the river area is 0 œ 1,2 mg/L and for more offshore area >than 1,2 mg/L. e. Phosphate The average phosphate concentration value at the surface is the same in the two areas, which is 0, 1,2 mg/L. The value at the bottom of the sea is higher in the offshore area (1,2 œ 2,4 mg/L) and <1,2 mg/L in the area close to the river mouth.
3.3. Shallow Marine Habitat a. Seagrass There are 8 species of seagrass found in the surrounding of Derawan Island. The dominant species is Thalasia hemprichii. In average, the seagrass coverage in Derawan (2001) is approximately 37%. This coverage is low compared to the observation of 1994, which was 56%. The substrate is a mixture of sand and rubble. Along the shoreline of Derawan island the seagrass is dominated by Halodule pinifolia. (LIPI 2001).
13 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. b. Coral Reefs The coral reef condition surrounding Derawan intermediate to good condition. Table 3.1, illustrates the percent coral cover in the Derawan region.
Table 3.1. Live coral cover percentage in transect locations Transect Location Live Coral Cover Percentage (% ) 1m 3m 10m 1. East part of Panjang Island 28 57 4 2. West part of Panjang Island 38 54 46 3. East part of Derawan Island 21 35 31 4. West part of Derawan Island 9 35 55 5. Samama Island 41 51 29 6. Sangalaki Island 72 27 36 (Source: LIPI-Ambon, 1995)
A LIPI survey (1995) found 50 kinds of coral species around Derawan Island. Coral reef growth was found up to a depth of 12 m. At a depth of 10 m, only patchy reef and at a depth of 20m, there was almost no coral reef encountered.
3.4. Environm ental threats in Derawan Islands The environmental threats for the coastal environment of East Kalimantan are in general grouped into several issues: land use issues, mining industrial and pollution issues, human settlements, and destructive fishing. Forest degradation due to logging or mangrove conversion (e.g to shrimp pond) caused increased the sediment discharge and decrease of dissolved organic. Some coastal areas of East Kalimantan, particularly in Balik papan have been decreased due to the rapid industrial and population growth. For example the oil and gas industry have two large disposal basins, one in Tarakan (North part of Derawan Islands) and one in the Mahakam Estuary (South part of Derawan Islands). The East Kalimantan urban centres are mainly located in or near the coast and near the rivers. It is still common practise to use the river and the sea as public waste disposal site. These domestic pollutions give additional problem for the ecosystem. For instance the disposal of plastic bags to the sea can cover coral and other filter feeder organism, thus caused their mortality. Blast fishing practises caused also rapid habitat degradation. Although illegal, this fishing method is still common in Indonesia including in East Kalimantan. (Jompa and Pet-Soede 2002)
A study about Derawan Islands, highlighteds that blast fishing is the major cause of coral reef degradation (Siahainenia 1999); Ismuranty 2002; (Jompa and Pet-Soede 2002);(Malik et al. 1999). This practise a high impact to coral reef degradation. A bottle of bomb, that explodes near or at coral reefs can destroys all corals within the radius of 1.2 m and can kill most marine organism within the radius of 77 m. This can be equal that the blast impact area is 1.9 ha. (Jompa and Pet-Soede 2002).
Siahainenia et.al (1999) conducted coral reef assessment in 1999, 14 bomb pits were found in the West part of Derawan Island during the survey. He also noted that coral reef degradation in Derawan Island is about 75% (comparing observations between 1994 and 1999). It takes decades or more for damaged coral reefs to recover.
14 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Panjang
Maratua Derawan
Kakaban Sam am a Sangalaki
Figure 3.1. M ap of Derawan Islands, Berau Region œ East Kalimantan Image ETM 2002 RGB 421
Derawan Island
1 km Masim bung Reefs
Tubabinga Reefs
Figure-3.2. Study Area. Derawan Island œ QuickBird 2003 RGB 321, Subset Derawan Islands region œ Landsat ETM 2002 RGB 421
15 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
4. Available Data and Field W ork Activities
4.1. Description of the Available Data 4.1.1. Rem ote Sensing Data Multi-temporal and multi-sensor remote sensing satellite data were used in the study. Table 4.1. describes the details of the available remote sensing data.
Table 4.1 Available Remote Sensing Data Acquisition Path/ Number of Spectral range* Sensor Resolution Row Time Bands (nm) Date (GMT) B1= 450 œ 520 16 June 09:43 30 m Landsat TM_5 116/58 (Band 1-7) B2= 520 œ 600 1991 a.m. 120 m (B6) B3= 630 œ 690 30 m B4= 760 œ 900 10:00 (Band 1-7) + Landsat ETM_7 116/58 8 July 2002 60 m (B6) B5= 1550 œ 1750 a.m. pan 15 m (pan) B6= 1040 œ 1250 User 2 October 10:23 (Band 1-4) 2.4 m B7= 2080 œ 2350 QuickBird defined 2003 a.m. + pan 0.6 m (pan) Pan= 450 œ 900 ñ Spectral ranges of Landsat TM 5, ETM 7 and QuickBird images are the same.
The landsat TM 5, of June 1991 has been downloaded from the Global land cover facilities achieved images data (http://glcfapp.umiacs.umd.edu). All 7 bands data were available, including the header files.
The Landsat ETM data, of July 2002 was obtained from the National Institute of Aeronatic and Space œ LAPAN, Indonesia. All 7 bands were available, including the panchromatic band and the header file.
The QuickBird image has been purchased by ITC for the East Kalimantan Programme (EKP) project. The image acquisition date (2 October 2003) corresponds approximately with the time of the field data collection. Two images of the Berau region were acquired: site 1 (consisting of Derawan Island, Masimbung Reefs and Tubabinga Reefs) and site 2 (consisting of Samama island, Sangalaki island, Beliulin Reefs and Pinaka Reefs). Due to the time constrain, this study only focuses on site 1.
4.1.2. Aerial Visibility Data Data on aerial visibility was acquired from the Meteorology and Geophysics Agency (BMG) in Balikpapan. The data were derived from flight information of Balikpapan airport. However it is only available for horizontal visibility (in km unit). The dates of the ordered
16 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. visibility data coincided with the image acquisition dates. The visibility data were used as an input parameter for the atmospheric correction using ATCOR software.
4.1.3. Bathym etric Data ñ Bathymetrical maps from the Hidro-Oceanography Service of Indonesian Navy (DISHIDROS) at various scales. ñ Historic bathymetrical maps from Expedition Snellius II (Naturalis and LIPI) at various scales.
4.1.4. Tidal Data ñ A time series of tidal data for the study of historic Landsat images (Landsat TM 5, 16 June 1991 and Landsat ETM 7, 8 July 2002) are obtained from the tidal database of ARGOSS. ARGOSS is a coastal mapping and marine information service for offshore and coastal engineering applications. It has a database with tidal data (tidal height, tidal current and current direction), which covers seas and oceans in any part in the world. The tidal-height data are derived from modeling based on the relationship between planetary orbit and moon position. This modeling data are calibrated to field measurements (buoyancies) and satellite imageries (www.argoss.nl). ñ Tidal data from the Hidro-Oceanography Service of Indonesian Navy (DISHIDROS) of October 2003, coinciding with the acquisition time of the QuickBird image.
Assessing the tidal data spread sheet from DISHIDROS in October 2003, the tide high range from 0,3 œ 2,4 m.
4.2. Fieldwork Activities The fieldwork activities were carried out in October 2003 for one month, including 1 week in Jakarta and Balikpapan (South-Kalimantan), for secondary data collection and research equipment preparation, and 3 weeks in Derawan Islands (4 œ 25 October 2003) for in situ data collection. The schedule of data collection in Derawan coincided with other research groups from The Netherlands and Indonesia (see section 1.1.).
The in situ data collections were conducted together with ITC student Ana Fonseca (Coastal Zone Studies) who studied the coral reef conditions of all the islands in the Berau region (see figure 3.1.), focusing on coral reef‘s health in relation to the influence of total sediment concentration and coral reef spectral discrimination. The in situ spectral data measurements were carried out together with temporary ITC staff from the Water Resources Department, Arjan Reesink and Sybrant van Beijma , who studied the total suspended concentration in the Berau area.
The field methods in Derawan Islands consisted of: 1. ground control points collection, 2. habitat ground truth points, 3. in situ spectral data measurements, 4. water depth measurements and 5. habitat data collection along transects.
The instruments that were used during the fieldwork and the measured parameters are listed in table 4.2.
17 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Table 4.2. The field equipment and measured parameters
No Equipment Name Parameters Coordinate of: 1. A GARMIN GPS 12 XL ñ ground control points; ñ ground truth points; ñ transect habitat sites.
2. ñ Scuba Diving equipment Transect habitat data ñ Transect quadrate (1 x 1m) ñ Line meter (50 m) ñ Digital underwater camera.
3. Speedtech Depthmate Portable The water depth Sounder Range: 0.6 œ 79 m
4. Analytical Spectral Devices (ADS) Spectral reflectance of: FieldSpec® Pro spectroradiometer ñ Bottom features Range: 350 œ 2500 nm ñ Deep water Optical resolution: 3 nm
4.2.1. Ground Control Points Collection Ground control points were measured with a garmin 12 XL receiver for the geometric correction of the images. Examples are the corners of jetties surrounding the islands and the corner of a helicopter base. WGS 84 and UMT 50 North were set up as the datum and projection. GARTRIP software was used to transfer the data to the computer. The locations of the ground control points are shown in figure 4.1.
4.2.2. W ater Depth Measurem ents and Habitat Ground Truth Points Collection A Portable Speedtech Depthmate was used to measure the water depth, simply by putting the head of the sounder into the water (see figure 4.2.) The measurements were taken from the boat at every 30 to 200 m distance; depending on the depth profile. About 900 depth points were collected and each coordinate point was recorded with the GPS, including the time of acquisition. From about 100 depth points a habitat description has been made to be used as habitat ground truth points. This was done when the water was not too deep and clear enough to describe the habitat from the boat.
18 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Figure 4.1. Ground control points plot in QuickBird image panchromatic band
Legend: Depth point
Ground truth point
P h
o t o g
r a p h
b y
T .
H o
b m a
Figure 4.2. Depth measurement sampling points and ground truth habitat points. Insert: Picture of measuring the depth
19 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
4.2.3. Habitat Transect Data Collection Habitat transect data were collected to obtain more details of the coral reef condition, such as live coral percentage, the dominant species etc. A quadrate transect technique was used to collect the data. A 20 m line was placed parallel to the coastal line. The data was collected at three different depths; 10, 6 and 2 m, starting at the deeper part (10 m) until the shallow part (2m). The transect quadrate (1 x 1 m) was placed along this line, with 2 m intercept; so there were 10 transect quadrates along each line (see Figure 4.3-a). A digital underwater camera was used to record the data in each transect quadrate (Figure 4.3-b).
4.2.4. Spectral Measurem ents The reflectance was measured with an Analytical Spectral Devices (ADS) FieldSpec® Pro spectroradiometer (range 350 œ 2500 nm, 1,4 nm sample interval and 3 nm optical resolution) which was linked to a laptop computer to monitor and save the data. The ADS is provided with various optic (25°, 8°, 10° and 1° field of view), including a cosine receptor for irradiance measurements, an extension with a 10 m long fiber cable and a white reference panel, which has approximately 100 % reflectance across the entire spectrum.
The protocols for spectral measurement are: 1. irradiance measurement using the cosine receptor, 2. white reference panel measurement and 3. reflectance measurement. In each spectral measurement of either specific or mixed specimens, a photograph was taken. This picture has been used to confirm the description of the measured specimen.
Two types reef substrate reflectance were measured: 1. Specific specimen measurement: The measurements of each specific specimen, classified as dominant coral species, dead coral at different stage (bleached dead coral, dead coral cover with turf algae and dead coral cover with coralline algae), sponge, soft coral, macro-algae and sand. This measurement used a 10è fields-of view; at a distance of 10 cm (used 10 cm stick as reference) and covers ~4.2 cm2 of the target specimen surface. Of each site approximately 25 œ 30 specimens reflectance were measured and each measurement was stored in one file. Each category; life coral, sand, dead coral, soft coral and algae, has 5 times of measurements.
2. Mixed specimens measurement: This approach was used to measure the reflectance of mixed specimens. This measurement used a 25è field-of view; at a distance of 2 m (used a 2 m stick as reference) and covers ~60 cm2 of the mixed specimens within a 1 x 1 m transect quadrate. Of each site about 10 mixed spectral data were measured.
At least three people were needed for the measurements: One person to do the underwater measurements, another person to operate the computer in the boat and one person to look at a sign from the diver after each measurement to give then to the computer operator.
20 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
P h
o t o g r 1 m a p h
1 m b
20 m y
M .
N u r l i d i
a s a r i
a. Position transect quadrate
b. Photo of transect quadrate data
M a p
p r e p a r
e d
b y
A .
F
o n s e c
a
c. Sampling location sites of transect habitat data and spectral measurement
P h P o h t o o g t o r a g p r a h p
b h y
b
M y
. M
N .
u
N r l u i d r l i i a d
s i a a r s i a
r i
d. Specific spectral reflectance e. Mixed spectral reflectance measurement measurement
Figure 4.3. Transect habitat approach; sampling location sites and spectral reflectance measurements.
21 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
The collections of the habitat transect data and the spectral measurements were carried out in a continuous way. Due to logistic constrains, it was not possible to measure both specific specimen and mixed specimen at each site. So the protocol was, in every site, first collect the transect habitat data, after this do either specific specimen or mixed specimen measurements.
Deep-water reflectance measurements were collected at a depth of 75 m; both above and beneath the water surface. This reflectance was used as a reference for the atmospheric correction and as an input parameter in the modified Bierwith method. (see section 4.4.4.2). ViewSpec± Pro. Software is used to store the reflectance data.
The processing and the analyses of the spectral reflectance and habitat transect data are describe in the MSc thesis of Ana Fonseca, —Spectral discrimination and mapping of coral reef environments Berau reef system, East Kalimantan, Indonesia“.
22 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
5. Research Methodology
5.1. Geom etric Im age Correction A ground control point is a position in the field that can be recognised in an image for geometric correction. Infrastructure (e.g. intersection of the road, flight based line etc.) or the boundary or shape of different land covers can be used as ground control point. The type of ground control point also depends on the spatial resolution and the size of study area. The aim of ground control points is to give coordinates to the image (Landsat TM, ETM and QuickBird). After this one will be able to plot ground truth habitat points on the images and compare images of different years.
As a topographic map of the study area was not available, for the geometric correction, ground control points were used as tie points. For resampling, the affine transformation was used, because it is a simple one and commonly used for satellite images.
It was difficult to find suitable ground control points. Derawan Island is only a very small island, approximately 1 km long and 0,5 km wide. The high spatial resolution of the panchromatic QuickBird band (resolution 0,6 m), gave the possibility to use infrastructural elements as ground control points. In the field, the corner of some jetties surrounding the island, corners of a helicopter base and sand pit in the east part of the island were collected as ground control points.
However, those points could not be used for the geometric correction of the Landsat image, due to its relative low spatial resolution. Thus, for the geometric correction of Landsat TM (1991) and Landsat ETM (2002) the geometric corrected image of QuickBird was used as master image. The two Landsat images were subset first to the same area as the QuickBird image.
Table 5.1. gives information about the Root Means Square Error (RMSE) of the geometric image correction.
Table 5.1. RM S error of geometric corrected images Images Number of points Resolution RM SE QuickBird panchromatic 5 0.6 m 4.667 QuickBird MS 10 2.4 m 0.468 Landsat ETM 12 30 m 0.557 Landsat TM5 7 30 m 0.544
23 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
5.2. Atm ospheric Correction
Satellite sensors record the intensity of electromagnetic radiation (EMR) as digital number (DN) values. The DN value of each image is specific to the type of sensor and the atmospheric condition during the image acquisition. To make different images comparable, the DN values should be converted to a physical unit, using the calibration data given in the header file of the image (Richter, R. 2002).
Since one of the objectives of this study is to compare images in time; atmospheric correction is important for the analysis: ñ To remove or reduce the influence of the atmospheric and the solar illumination ñ To be able to compare multi-temporal satellite images with different times of acquisition. After atmospheric correction, changes observed are due to different features on the earth‘s surface rather than differences of the atmospheric condition; ñ To improve the results of change detection and classification algorithms; ñ To be able to compare multi-sensor images with similar spectral bands; ñ To be able to compare ground reflectance data retrieved from satellite imagery to the ground reflectance from in situ measurements in the field. (Richter, R. 2002)
Two approaches were used for atmospheric correction: ñ Atmospheric and Topographic Correction (ATCOR), an extension software of ERDAS 8.5, for Landsat TM and ETM images ñ In situ dark reflectance subtraction for the QuickBird image.
5.2.1. ATCOR Approach for Landsat TM Im ages The ATCOR approach refers to ATCOR user manual (Richter 2002). The following steps were performed for atmospheric correction: 1. Specify input parameters; 2. Estimate the visibility and the atmosphere model with Spectra module; 3. Apply atmospheric correction with constant atmospheric conditions using ATCON module.
ATCOR2 for ERDAS IMAGINE (v. 8.5) software was used for the radiometric and atmospheric correction. There are several options under ATCOR; Calculate Sun Position, ATCOR 2 workstation, ATCOR3 Derive Terrain File and ATCOR 3 workstation. For our analyses ATCOR2 was used because it is a model for the atmospheric correction of flat terrain.
24 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
1. Specify input parameter Most of the parameters are provided by the image header file, which is attached to the image by the provider. The parameters of the acquired TM images are displayed in Table 5.2:
Table 5.2. Parameters of TM 5 1991 and ETM 7 2002 images Sensor TM-5 ETM-7 Date of Acquisition 16 June 1991 8 July 2002 Bands 1-7 exclude b6 1-7 exclude b6 and pan Pixel size 30 m 30 m Factor scale 4 (default) 4 (default) Solar zenith angle 40.32 35,6 Calibration file Tm51991.cal Tm7-2002.cal Model for solar Others. Type: tropical ocean Others. Type: tropical ocean region Visibility 30 km 30 km Ground elevation 0.1 km (default) 0.1 km (default)
Some parameters should be prepared before conducting atmospheric correction such as: a. Image format To apply t ATCOR2 application, the images to be corrected should be saved with an —img“ file extension. For example if the images are in ILWIS format, one can create a map list then export with LAN extension. The benefit of using LAN extension is that when it is imported to —img“ format, all the bands are contained in one file. So the atmospheric correction can be done at once. b. Solar zenith angle In the header file, the information regarding sun illumination is sun elevation. However, the parameter that ATCOR2 requires is solar zenith angle. Sun/solar elevation angle and solar zenith angle are complement to each other. So to obtain the solar zenith angle, the sun elevation angle has to be subtracted from 90è. c. Calibration file The calibration file contains the calibration coefficients c0 (bias) and c1 (gain) for each image band. These coefficients are used to convert the DN values to radiance at sensor. The revised Landsat TM5 calibration parameters (USGS, 2003) were used to create the calibration file for Landsat TM5, 16 June 1991, image. The calibration values for Landsat ETM, 8 July 2002, were extracted from the header file. The bias and gain values of Landsat TM5 and ETM7 that are used in this study are listed in table 5.3.
25 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Table 5.3. Calibration parameter for Landsat TM 5 and Landsat ETM 7 images Landsat TM5 Landsat ETM7 8 July 2002 Band [mW/cm2 /sr/micron] [mW/cm2 /sr/ micron] C0 C1 C1 C0 1 -0.15200 0.06024314 -0.6200 0.0775686 2 -0.28399 0.11750981 -0.6400 0.0795686 3 -0.11700 0.08057647 -0.5000 0.0619216 4 -0.15100 0.08145490 -0.5100 0.0965490 5 -0.03700 0.01080784 -0.1000 0.0125725 6 0.12378 0.12377996 0.0000 0.0066824 7 -0.01500 0.00569804 -0.0350 0.0043725
d. Model for solar region Standard atmospheres are used to model the atmosphere as it was during the image acquisition. ATCOR provides several options of aerosol type and atmospheric type. For this study —tropical ocean“ and —humid“ options are selected for aerosol and atmospheric types respectively. e. Visibility The visibility is used to determine the appropriate atmospheric condition due to aerosol and humidity during the image acquisition. The spectra module chart (figure 5.1.) was used to determine the appropriate visibility. Ancillary visibility data was obtained from the Meteorology and Geophysics Agency in Balikpapan. Based on this data, the visibility during the image acquisition of Landsat TM5 and ETM7 was estimated at 15 km (see appendix B). However based on the spectral chart (see figure 5.1.) the visibility is 30 km for Landsat TM5 and ETM7, showing relatively less negative reflectance values and the closest values to the sea spectra reference. Beside, the visibility data from meteorology represented the atmospheric condition in Balikpapan, which is located at distance approximately 400 km away from Berau. Therefore the visibility estimation model with spectral were used. d. Ground Elevation (km) Derawan island has a flat ground, so for the ground elevation parameter, the value of 0,1 km was used.
2. Estimate The Visibility And The Atmosphere M odel W ith Spectra The purpose of this module is to determine the appropriate atmospheric condition (aerosol and humidity) and visibility. Moreover this spectra module can also be applied to confirm the influence of the calibration file.
26 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Comparison of several spectra, using different visibility, with the sea_spectral library for TM 5
Comparison of several spectra, using different visibility, with the sea_spectral library for ETM 7 Figure 5.1. Spectral reflectance versus various visibilities and comparison with sea spectral library
Figure 5.1. shows spectral signatures of deep sea at various visibilities. These spectral signatures are compared to deep sea from the spectral library as a reference (the red line). The graph shows that the visibility of 30 km has a relatively similar pattern compare to the spectral library signature.
3. ATmospheric correction with CONstant atmospheric condition œ ATCON
There are three options available for ATCON: ñ No Haze Removal ñ Haze Removal, Omitting Visibility Index File ñ Have Removal, Including Visibility Index File.
The option —No Haze Removal“ was chosen because applying —haze removal“ option caused misidentified reef area as haze.
27 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
TM5 band 1 before atmospheric correction (DN) TM5 band 1 after atmospheric correction (X)
ETM7 band 1 before atmospheric correction (DN) ETM7 band 1 after atmospheric correction (X) Figure 5.2. Comparison of histogram band 1 before and after applied atmospheric correction
Figure 5.2. illustrate that in the histogram before the application atmospheric correction the mean DN values are 34,88 and 80,86 and after atmospheric correction the means reflectance values (X) are 4,31 and 9,35. These lower mean values indicate that the effects of the atmosphere have been reduced.
5.2.2. Dark Reflectance Substraction Approach for QuickBird Im age For the atmospheric correction of the QuickBird image the ATCOR program was not used because the area of the QuickBird image has a polygon shape, as ATCOR calculates area in square shape.
There are three steps in the atmospheric correction procedure (Edward, 1999): 1. Conversion of DN to satellite radiance. 2. Conversion of satellite radiance to satellite reflectance 3. Conversion of satellite reflectance to ground reflectance
1. Conversion of DN to satellite radiance. The DN conversion to satellite radiance is based on a technical note from Digital Globe (Krause 2003). The purchased QuickBird image has already radiometrically corrected image pixels (qpixel, Band). This corrected only count specific to the QuickBird instrument, thus for
28 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. e.g. comparison to in situ spectral reflectance, further atmospheric correction have to be applied.
For images generated after June 6, 2003 at 0:00 GMT, the equation is:
absCalFactorBand*q Pixel,Band L= E.q. (1)
Z[Band Where: L= satellite radiance (W m-2 ster-1 mm-1) q Pixel,Band = digital number values (counts) absCalFactorBand= absolute radiometric calibration factor (W m-2 sr-1 count-1), which is provided in header file. Z[Band= effective bandwidth of each band (\m), which is refers to Digital Globe (Krause, 2003).
2. Conversion of satellite radiance to satellite reflectance The conversion of satellite radiance to satellite reflectance refers to BILKO module seven (Edwards 1999) Reflectance is the ratio of the radiance to the irradiance, which depends on the wavelength. The following equation is used to calculate the apparent reflectance (r).
p.L.d 2 r = E.q. (2) ESUN.cos(SZ) r = satellite reflectance (range values of 0-1.) p = 3.14152 L = satellite radiance (mW cm-2 ster-1 mm-1) d2 = the square of the Earth-Sun distance in astronomical units ESUN = Mean solar irradiance in mW cm-2 mm-1. SZ = sun zenith angle in radians.
The square of the Earth-Sun distance in astronomical units (d2) is calculated using the following equation: d2 = (1 - 0.01674 cos(0.9856 (JD-4)))2 E.q. (3)
Where JD is the Julian Day (day number of the year) of the image acquisition. [Note: the units of the cosine function of 0.9856 x (JD-4) will be in degrees. However, it is expecting the unit in radians. To convert the unit from degrees to radians is simply multiply by p/180].
The input parameters for the conversion of satellite radiance to satellite reflectance of the QuickBird image of 2 October 2003 are listed in table 5.4.
29 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Table 5.4. Input parameters to calculate the satellite reflectance of QuickBird image, October 2003
Parameter Value Remark The unit should be converted The output values from equation from (W m-2 ster-1 mm-1) to L (Ll Pixel,Band ) 1 (mW cm-2 ster-1 mm-1).
The square of the Earth-Sun Calculated by eq.4, with image distance in astronomical units 1.001696 atronomical unit acquisition is 2 October 2003, (d2) thus JD is 275 ESUN was obtained using the B1=198 ESUN QuickBird parameter that B2=183 ESUN provided in the ERDAS- B3=156 ATCOR program. B4= 104 The unit is in mW cm-2 mm-1 SZ obtained by subtracted 90è- solar elevation (listed in header Solar Zenith (SZ) 24è ö 0.41867 radians file, which is 66è). To convert to radiance is multiplied SZ by p/180
3. Conversion of satellite reflectance to ground reflectance In situ spectral reflectance data of deep water above the surface was used as a reference to correct the image for atmospheric effects.
B1=0.07
B2=0.05 B3=0.04
B4=0.04
Figure 5.3. A Graphic of ground reflectance values of deep water (Rg)
Figure 5.3. A graphic of ground reflectance mean values of deep water (Rg)
Figure 5.3. shows the mean reflectance values of the deep-water reflectance (depth= ~75 m) above the surface of the water, which was measured during the fieldwork. These values were used as a reference of in situ ground reflectance of dark pixels (deep water pixel) to convert satellite reflectance to ground reflectance retrieved from the image (Rg ≈ R, where Rg is in situ ground reflectance and R is ground reflectance derived from image).
The mean of in situ ground reflectance (Rg) and deep water value of satellite reflectance (rdw), were used to calculate the value of x. The x value is a certain values representing the influence of the atmosphere to reflectance recorded at the satellite.
30 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
The following formula was used to calculate the x value:
X= rdw - Rg E.q. (4)
The information of mean ground reflectance value (Rg), deep water value of satellite reflectance (rdw ) and the value representing the effects of the atmosphere (x) are listed in table 5.5.
Table 5.5. Rg, rdw and x values of each band Band Deep water value of M ean ground The value represents reflectance at satellite reflectance (Rg) of effects of atmosphere
(rdw ) deep water (x) B1 0.23 0.07 0.16 B2 0.09 0.05 0.04 B3 0.07 0.04 0.03 B4 0.05 0.04 0.01
The formula to obtain ground reflectance retrieved from the image is:
R = r œ x E.q. (5)
Where; R ground reflectance values retrieved from the image in band, r at satellite reflectance, and x is the additional values due to the influence of the atmosphere, which was recorded at the satellite.
Figure 5.4. shows that the mean value of band 1 of QuickBird before (a) and after (b) the application of atmospheric correction has decreased from 43,07 to 6,39.
(a) before atmospheric correction (DN) (b) after atmospheric correction (X)
Figure 5.4. Comparison of the histogram of band 1 QuickBird before (a) and after (b) the application of atmospheric correction
31 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
4. Accuracy assessment of ground reflectance values derived from the image
The mean of in situ ground spectral measurement of bright object (eg. sand beach) was compared to the mean of ground spectral of sand beach on the image. The ratio between the mean of in situ ground reflectance to ground reflectance on the image is shown in the following formula:
Rg 0.59 = = 0.95 R 0.62
Where: Rg = mean in situ measurement o ground spectral reflectance of sand beach. R = mean ground spectral reflectance of sand beach on the image.
The ratio means spectral reflectance of bright object between the field data and from the image shows correlation of 0.95.
5.3. Land, Cloud and Deep W ater Masking The aim of masking the image for land, clouds and deep water is to consider only the area of interest, which is shallow water. The masked area (land, cloud and deep water) is set as —undef“ and the shallow water pixels remain having their own values. The processes of masking followed the steps: 1. Creation of a masked image in raster format. a). Creating a segment map by digitising the land, cloud and deep water part using a false colour composite of band combination RGB 321; b). Creating a label point map; c).vectorization of the segment map; and e). rasterization of the vector map. 2. To use the masked image to mask the atmospheric corrected image by applying map calculation.
For example, the algorithm to mask band 1 of the QuickBird image is: Maskb1=iff((mask=“deep“)or(mask=“cloud“)or(mask=“land“),?,atcorb1)
Where; Maskb1 = the masked image in raster format Atcorb1=the atmospheric corrected image
Figure 5.5. shows the example of masked land, deep water and cloud of QuickBird image for band 1.
32 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
land
deep water
cloud
Figure 5.5. The masking land, deep water and cloud of QuickBird image for band 1.
5.4. W ater Colum n Correction 5.4.1. Depth Invariant Index The concept is, that bottom reflected radiance is a linear function of the bottom reflectance and an exponential function of the water depth. The intensity of light penetration decreases exponentially with increasing water depth; this process is known as attenuation. The approach proposes a method to compensate light attenuation due to the influence of depth by using the information ratio attenuation of two bands (Lyzenga 1981).
The processing steps of depth invariant index are (Lyzenga 1981; Edward, 1999): 1. Selection of homogenous substrate at various depth; 2. Calculate the ratio attenuation coefficient of a pair of bands; 3. Linearise the relationship between depth and radiance; 4. Generate a depth invariant index of bottom type within a pair of bands.
1. Selection of hom ogenous substrate at various depth Homogenous substrate at various depths is selected from the image following the steps: Firstly, display a false colour composite of RGB 321 of the atmospheric corrected image, secondly assess the homogenous substrate values (in this case sand) in the three bands and list them in a EXCELL spreadsheet. These values will be the input for the calculations in step 2.
33 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
2. Calculation of the Ratio Attenuation Coefficient of a Pair of Bands
Ratio attenuation is obtained using the following equations:
Ki = a + (a2 +1) Kj E.q. (6) where:
sii-sjj a = E.q. (7) 2sij
and:
`ij = X X œ X X i j i j E.q. (8)
`ii is the variance of Xi measurements, `jj is the variance of Xj measurements and `ij is the covariance of Xi and Xj. The variance and covariance values are calculated using the values from step 2.
3. Linearise the Relationship Between Depth And Radiance Water attenuation has an exponential function to the increase of the depth. By applying a natural algorithm (ln) to the atmospheric corrected image (in radiance units), this relationship will be linear. This step is written as:
E.q. (9) Xi=ln(Li)
Where; Xi = normalized image in bi
Li = atmospheric corrected image of bi (radiance unit)
4. Generate A Depth Invariant Index To generate a depth invariant index of bottom types within a pair of bands, the following formula is used:
»≈ ki ’ ÿ E.q. (10) depth - invariant index ij = x i - …∆ ÷ * (x j )Ÿ « kj ◊ ⁄
5.5. Modified of Bierwirth Method The output of the Lyzenga method is an index value of dept-invariant, which enhances the different bottom types. However this value is derived from a pair of image bands and the output is not in reflectance. In order to link in situ spectral reflectance measurements to the image, the image should be in reflectance values. Therefore another approach to compensate for the water column effects was carried out by applying a modification of the Bierwirth method (Purkis and Pasterkamp 2003) The advantage of this method that the output value is
34 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. wet bottom reflectance of each band thus it is assumed to be possible to link the image to field spectral reflectance.
Bierwirth (1993) proposed a method to estimate bottom substrate reflectance of shallow water derived from the image. Estimation of bottom substrate reflectance uses the following equation: E.q. (11) Rb = RA exp(2kz)
(Purkis and Pasterkamp 2003) modified the Bierwirth method by applying a compensation for the refraction of the water/air boundary and water column reflectance of deep water. The modified algorithm is:
œ2kz 1/0.545*RA œ (1- e )*Rw Rb = œ2kz e E.q. (12)
Where; Rb = (wet) substrate reflectance 0.545 = ”Water-Air-Transmission‘ factor RA = ground reflectance value from atmospheric corrected image k = effective attenuation of the water body z = water depth (in metres) Rw = water column reflectance of optically deep water
Ground surface reflectance (RA) is the atmospheric corrected image values in reflectance units; water column reflectance of optically deep water (Rw) is the reflectance of deep water just beneath the water surface, which is measured during the fieldwork. The Rw value used the means of the deep-water reflectance measurements. Depth values cannot be derived from an existing bathymetric map; the information is not detailed enough in the areas with shallow water, such as the coral reefs. Thus water depth points are used. For that reason, the estimation of the bathymetry was done using the —Jupp“ method. Besides the depth estimation, the output of this method is coefficient attenuation of water body (k), which is also used as input parameter in equation 12.
5.6. Estim ation of the bathym etry The concept of bathymetric mapping using optical remote sensing is, that more light is reflected from a shallow sea floor because less has been absorbed in the water column. Contrary, a deeper sea floor reflects less light due to a larger absorption by the water column. So, in the image, the shallow areas appear bright and the deeper areas will appear darker.
The estimation of the bathymetry was done by applying the Jupp method (Jupp, 1988 cited in Edward, 1999). This method has three assumptions, which are: 1. Light attenuation is an exponential function of depth; 2. Water attenuation does not vary within the image; 3. The albedo of the substrate is fairly constant.
The third assumption in fact is not really suitable for coral reef environments. However (Purkis and Kenter 2001) applied this method in a coral reef environment in the Egyptian Red Sea and had a correlation of 0.96 between predictive and measured depth. Moreover, (Mumby et al. 2000) assessed three methods to predict the water depth; the Benny and Dawson (1983) method, the Jupp (1988) method and the Lyzenga (1978) method. It is
35 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. concluded that Jupp‘s method is the most accurate compared to the other tested methods. Therefore, Jupp method is applied for this study.
The attenuation value (k) varies between the different wavelengths. For example, the blue band (450-520 nm) can penetrate further up to ~30 m, but red band (630-690 nm) attenuates rapidly in water and only can penetrate up to ~5 m. So the concept of Jupp method is to determined the depth penetration zone of each band. Figure 5.6. shows the rationale behind the concept of Depth Of Penetration (DOP) zones.
TM band 1 TM band 2 TM band 3 TM band 4
L1 deep max+1 L2 deep max+1 L3 deep max+1 L4 deep max+1 surface 4
e n o z
P Z 3
O 4
D e n o z
P O D Z3 2
e n o z
P O D
Z2 1
e n o z
P O D
Z 1 Figure 5.6. Depth of Penetration (DOP) zones
The processing steps of depth of penetration (DOP) method are shown in figure 5.7. (Jupp, 1988 in Edward, 1999 and Mumby, 2000).
Convert in situ depth points to the depth at the time of image acquisition
Create depth of penetration (DOP) zones zones
Interpolation of DOP zones
Accuracy assessment of the derived bathymetric map
Figure 5.7. Four steps of derived bathymetric map (Jupp method)
36 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
1. Convert in situ depth points to the depth at the time of image acquisition. The first step is to convert in situ bathymetric points to a standard datum. In order to get standard datum depth, the depth values measured from the field were subtracted to the high tide values at the time of measurements. The high tide values can be retrieved from ancillary tidal data. The second step is to add the high tide value at the time of image acquisition, to standard datum depth values. Thus in situ depth points have been converted to the depth at the time of image acquisition.
2. Creating depth of penetration (DOP) zones, which consist of: ñ Select an area of deep water, which represents the optical properties of the study area. Then calculate the maximum, minimum and mean pixel values over deep water of this area. ñ Estimate the maximum depth of penetration of each band, using the in situ depth points which have been converted to the depth at the time of image acquisition (output of step 1). The first step is to use UMT coordinate of in situ depth points and plot it on the image to retrieve the values of each band (band 1-4) at measured points locations. Then sorted the depth starting from the deepest to progressively shallower depths. After that, one can start to estimate DOP zone.
ñ The estimation of maximum depth of penetration is also taking into account of the reference depth of penetration (listed in table 5.6.). For example to estimate DOP 1, only consider DN values in Band 1 which have the depth approximately between 25m œ 15m. Then sorted the DN in band 1 together with its depth value starting from the higher DN value. After, the mean value of depth from the DN value more than
maximum deepwater DN value was calculated (>V deep maxb1). Then, the mean value of depth from DN values which is same as deepwater DN value was calculated too
(=V deep maxb1). The average between the depth value (>V deep maxb1) and the depth value
(=Vdeep maxb1) is DOP zone 1.
Table 5.6. The maximum DOP zi (Jupp, 1988). band i Depth of penetration zi 1 25 m 2 15 m 3 5 m 4 1 m
ñ Make a —DOP zone mask“ for each band. This is done by coding all pixels within the DOP of zone of band 1 to a value of 1 and other pixels to a value of 0. This procedure was also applied to band 2-4.
3. Interpolating DOP zones, which consist of: ñ After generating a mask of each band, the next step is to multiply the DOP zone mask with the atmospheric corrected image. So the pixels within DOP zone will have its original values and those outside of DOP zone will have 0 value.
37 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
ñ Use the histogram of each DOP zone map to obtain the Limax and Limin . Then Ximax
and Ximin can be calculated since Lideep mean is known. The calculation of Ximax and
Ximin is done with the following equations:
Xi max = ln(Li max œ Ldeep mean) E.q. (13) and:
Xi min = ln(Li min œ Ldeep mean) E.q. (14)
ñ The coefficient attenuation (ki) and Ai are calculated using the following equations:
(Ximax - Xi min ) ki = E.q.(15) 2(zi - zj)
Where ki is coefficient attenuation of bandi (eg. Band 2), zi is DOP zone in bi (eg.DOP zone
b2), zj is DOP zone in bj (eg. DOP zone b3).
Ai = Xi min + 2kizi E.q.(16)
Once Ai and ki are known, the depth of the water (z) for any pixel with value xi=ln(Li-Ldeepmean) can be calculated using this following equation:
(A - X ) zi = i i E.q. (17) 2ki ñ Applying equation 17, the depth will be assigned to each pixel in each DOP band and produce four separate interpolated DOP depth images. These DOP depth images are added together to produce a final depth image.
4. Accuracy assessment of the crude bathymetric map derived from the DOP zones method. Using the depth values of in situ measurements, which have been converted to the depth at the time of image acquisition, one can compare the depths predicted by the image at each UTM coordinate with those measured at the site. In EXCELL spread sheet, the values of measured depths can be plotted against predicted depth to obtain the correlation between them as the parameter of accuracy.
5.7. Categorization of Habitat Classes There are at least five ways to classify coral reef habitat with remote sensing mapping (Mumby, 1998 & 2000): 1. Ad hoc definition of habitats. 2. Application-specific studies. 3. Geomorphological classifications. 4. Ecological classification of habitats. 5. Combined hierarchical geomorphological and ecological classifications.
38 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
1. Ad hoc definition of habitats. Ad hoc habitat classification can be used if the analyst is familiar with the area or if a comparable habitat classification scheme is available. Usually this approach does not include the collection new field data. Thus the advantage of this approach the relative low cost. However, several disadvantages might occur: The fundamental definition of classes may be incorrect and not applicable to the area concern. Although the scheme of classes is appropriate, the habitats may be misidentified, for example due to inaccuracy of the image interpretation, and it is not possible to assess the accuracy without independent field data.
2. Application-specific studies. This classification approach may be applied if the study is focused on specific surface features and therefore not concerned to map all habitats in the area. For example particular area on the image is dominated with one species, and the classification is only focussed at that particular area.
3. Geomorphological classifications The geomorphological classification approach is commonly used in remote sensing studies for coral reef mapping. This classification is relatively straightforward because there exist standard classification scheme. Examples of geomorphological classes are: Backreef, Reef crest, Spur and Groove, Fore reef, Escarpment, Patch reef, Lagoon floor. For a detail description of geomorphological classes see (Mumby et al. 1998b).
4. Ecological classification of habitats Ecological classification is not as straightforward as the geomorphological approach. The ecology of a habitat may be limited to the assemblage of plant and animal species and the substrate. Examples of ecological classification classes are: coral, algal dominated, bare substratum dominated and seagrass dominated (Mumby, 1998).
5. Combined hierarchical geomorphological and ecological classification This is a merged habitat classification scheme of a geomorphological and ecological habitat classification. An example of a class in this classification approach is shallow lagoon floor with seagrass (the ecological class can be defined in more detail including the species and its density).
The categorization of habitat classes in this study is more based on the ecological classification approach. Figure 5.8. shows descriptions of the habitat classes for this study which were represents the habitats in the study area. For seagrass habitat, only patchy seagrass is determined since high density seagrass was not found.
39 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Alg= algae composite of turf algae, coraline algae and algae assemblage
S=sand, bare substratum which composite of sand beach and mixed sand with fragment of carbonate
Ps=patchy seagrass Sparse seagrass with density < ~35 %
Cr=coral reefs, live coral dominated >30% with reefs surrounding
Pr=patchy reef Aggregation of soft corals and/or hard coral located sparsely on bare substratum
Dw=Deep water The area >~ 10 m which is dominated
by bare substratum.
Figure 5.8. Description of habitat classes
40 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
5.8. Im age Classification There are two types of image classification, unsupervised and supervised classification. Unsupervised classification is applied when the knowledge of the area is not available or not sufficient enough or the classes are not yet defined. This approach requires minimal input from the user, who only has to decide the number of classes to be produced. Contrary, supervised classification requires a user to have knowledge of the area. The user controls the process of classification by selecting training samples.
The general steps of supervised classification are (Mumby et al. 2000): 1. Define habitat classes; 2. Select training samples; 3. Evaluate the signatures; 4. Decide the decision rules.
1. Define habitat classes: The way to define habitat classes for a coral reef habitat map has been describe in section 5.7.
2. Select training samples: The group of pixels that represent a known feature or habitats are known as a training sample. It is important that the training samples are representative of the habitat class; otherwise the user will introduce misclassification errors. The training samples may be defined from field sampling points, user knowledge of the area and/or some other ancillary data (e.g. aerial photographs and maps).
3. Evaluate the signatures: When the training samples have been selected, the signature can be evaluated before deciding the decision rules of classification. This step is important because, there may be some errors during the selection of the training samples. Some examples or of errors are: signature overlap between classes, position error of the GPS which caused a mismatch between the position recorded in the field and on the image, misinterpretation of field data or aerial photographs/maps. Once the evaluation of the signatures is acceptable, the next step in decision-making stage can be performed.
4. Deciding the decision rules: The final step of classification uses the evaluated signatures to assign every pixel within the image to a particular class. Decision rules are types of statistical rules to assign pixels into a particular class, which are also known as classification algorithms. There are two types of decision rules: 1. Non-parametric rules and 2. Parametric decision rules.
The common example of non parametric rules is —Parallelpiped decision rule“. Parallelpiped decision rule: the data values in each band of training samples pixels are compared to upper and lower limits as the threshold. The limits can be maximum and minimum values, means plus and minus of standard deviation values, or other limits set by the user based on the knowledge of data acquired during signature evaluation. The advantage of this approach is
41 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. fast and simple. However if the spectral pixels are quite distance from the mean of a class within the specified boundary, may lead to a less a accurate accuracy.
The common example of parametric rules is —Maximum likelihood decision rule“. The maximum likelihood classifier assumes that the statistic of the class (cluster) has a —normal“ (bell shape œ Gaussian) distribution. Beside the shape, this approach also considers size and orientation of the cluster. This is achieved by calculating a statistic distance based on the mean values and covariance matrix of the clusters. Then the pixel is assigned to the class, which has the highest probability.
Figure 5.9. show a flow chart of general supervised classification approach.
Corrected Image Field data
Supervised Training
Selection of training samples
Evaluate the signature: Are they No acceptable?
Yes
Decision rules
Classified Image
Figure 5.9. Four steps of supervised classification.
42 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
5.9. Accuracy assessm ent The most common accuracy assessment of classified remotely sensed data is error matrix, sometimes known as confusion matrix. This is done by compare the classified image, as the result map, to the ”true world class‘. The true world class are preferable derived from field observations (e.g. ground truth sampling points). But sometimes sources which assumed has higher accuracy, such as aerial photograph, is used as a reference.
Other measures derived from the error matrix are ”error of omission‘ and ”error of commission‘. Error of omission represents an error from including a pixel to a particular class, which is actually not part of that class. Commision error represents that a pixel which should be part of a particular class but is not included.
There are three types of accuracy can be generated from error matrix: overall accuracy, producer accuracy and user accuracy. Overall accuracy represents the number of correctly classified pixels. The producer accuracy indicates the probability that a sampled point on the map is that particular class. The user accuracy indicates the probability that a certain reference class has also been labelled that class indicates (Janssen and Huurneman 2001).
43 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
6. Results and Discussion
6.1. Generating a coral reef habitat m ap with and without water coloum n correction using QuickBird im age of 2003 One of the hypothesis of the study is, that the water column correction technique can generate a more accurate coral reef habitat map from QuickBird data of October 2003. In order to assess effect of the water column correction for coral reef habitat mapping, a coral reef habitat classification without water column correction have to be made as a comparison. The classification steps used to generate the two types of classified images are shown in figure 5.9. However, there are differences in input corrected images and decision rules. Figure 6.1. shows the specific classification steps for generating the habitat maps with and without water column correction.
Atmospheric W ater column corrected image corrected image
Supervised Training Field data Supervised Training Field data
Selection of training samples Selection of training samples
Decision rules: M ax.likelihood Decision rules: Parallelpiped
Classified Image Classified Image
Evaluate the map: Evaluate the map: No is it acceptable? No is it acceptable?
Yes Yes Final Final Classified Image Classified Image
(a) (b)
Figure 6.1. Classification steps for generating habitat maps without water column correction (a) and withwater column correction (b).
44 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
6.1.1. Generating a coral reef habitat m ap without water colum n correction. The classification steps are shown in figure 6.1. (a). Habitat classes have been defined, which include: algae composite, coral reef, patchy reef, patchy seagrass, sand and deep water. The input for this classification is a maplist of the atmospheric corrected QuickBird image of 2003, which only covers the shallow water part of the study area. The maplist consist of bands 1,2 and 3 of the QuickBird image. The training samples have been selected using half of the total number of habitat ground truth points. The selected training samples were evaluated using the feature space option.
After the evaluation of the training set, the maximum likelihood algorithm was used to generate the classified image. During the classification processes, the first result did not meet the expectation, and the classification was repeated by re-selection of the training samples.
The results of the coral reef habitat map without water column correction is presented in figure 6.4.
Once the classification result was acceptable, the next step was to assess its accuracy. This process of accuracy assessment is described in section 5.9. For the accuracy assessment, half of the total numbers of habitat ground truth points were compared with the classified image. The result of the accuracy assessment is presented in table 6..2.
6.1.2. Com pensate for the water colum n effect on the im age For generating a habitat map with water column correction, a water column corrected image is required. To compensate for the water column effect, —depth invariant index“ method (also known as Lyzenga method) is used. The description of this method is given in section 5.4.1.
1. Selection of hom ogenous substrate at various depth First, a false colour composite of a RGB 321 QuickBird atmospheric corrected image of October 2003 has been displayed. The next step was to assess the homogenous substrate values (in this case sand) in the three bands image using pixel information and to list them in an EXCELL spread sheet (see appendix G).
Scatter plot of sand substrate between band 1 and band 2 Scatter plot of sand substrate between band 2 and band 3
50 100
40 80
30 60 3 2 B B 20 40 10 20 0 0 0 10 20 30 40 50 60 70 0 10 20 30 40 50 B1 B2
(a) (b) Figure 6.2. Scatter plot of sand substrate at various depth between band 1 and band 2 (a), and between band 2 and band 3 (b).
45 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
On the scatter plot, most of the pixels values between band 1 and band 2 were falling on a strait line (Figure 6.2.a), however this is less the case between band 2 and band 3 (Figure 6.2.b). This maybe because of the larger light attenuation of band 3 in the water column. Based on this result, for the next step, only a pair of band 1 and band 2 were used.
2. Calculation of the Ratio Attenuation Coefficient (ki/kj) of a Pair of Bands The ratio attenuation coefficients are calculated using the formulae 7, 8 and 9 (see section 5.4.1.). The results are listed in table 6.1.
Table 6.1. Parameter values to calculate the ration attenuation coefficient between band 1 and band 2. Parameter Values Variance b1 156,51 Variance b2 84,02 Covariance b1,2 113,36 a b1,b2 0,32 k1/k2 0,73
3. Linearise Relationship Between Depth And Radiance Band 1,2 and 3 of the atmospheric corrected and masked images of QuickBird are linearised using equation 6 (see section 5.4.1.).
For example, the algorithm to linearise band 1 of QuickBird image is: lnb1=ln(Maskb1)
Where: Lnb1= linearised of band1; Maskb1= masked band1 of the atmospheric corrected image. This step is also applied to band 2 and band 3 of the QuickBird image.
4. To generate a Depth Invariant Index
After the ratio attenuation coefficient (ki/kj) is obtained and the input bands have been linearised, a depth invariance index map was generated by applying equation 10 (see section 5.4.1). In this way, an image has been obtained in which the water column effect has been compensated for. The result of the depth invariant index map is shown in figure 6.3.
46 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
(a) (b)
Figure 6.3. Depth invariant index histogram and image of QuickBird band 1 and band 2.
6.1.3. Generating a coral reef habitat m ap with water colum n correction. The next in the classification processes, are shown in figure 6.1.b. The input classification map is a depth invariant index of a pair of band 1 and 2 (DI_12). The training samples are selected, using half the number of the ground truth habitats points (the same points that were used in section 6.1.1.). After having selected the training samples, a paralellpiped classification decision rule is used to classify the image.
Using ILWIS v.3.1. software, the parallelpiped classification approach was carried out by using the density slicing option. The values of the training sample pixels were used as the thresholds. The limits were maximum and minimum values, means plus and minus of standard deviation values. After the threshold boundary of each class was determined, the image was classified. The output of the classified image was evaluated; if this was not acceptable the classification processes was repeated by reselecting training samples or/and threshold boundaries.
The output of the classified image, which is a coral reef habitat map with water column correction, is presented in figure 6.5.
For accuracy assessment, the same approach is used as mentioned in section 6.1.1. The result of the accuracy assessment is presented in table 6.3.
47 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Figure 6.4. Coral reef habitat map of QuickBird image of October 2003, without water column correction
Figure 6.5. Coral reef habitat map of QuickBird image of October 2003, with water column correction
48 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Table 6.2. Error matrix of the classified image without water column correction (cr) (dw) (ps) (pr) (s) (alg) User Accuracy (% ) Coral reef (cr) 5 2 0 0 0 0 71 Deep water (dw) 0 2 0 0 0 0 100 Patchy seagrass (ps) 0 0 3 2 0 0 60 Patchy reef (pr) 2 0 0 12 0 0 80 Sand (s) 0 0 2 0 7 0 78 Algae (alg) 2 0 0 8 0 9 47 Producer 56 50 60 55 100 90 Accuracy (% )
Total accuracy: 67 %
Table 6.3. Error matrix of the classified image with water column correction (cr) (dw) (ps) (pr) (s) (alg) User Accuracy (% ) Coral reef (cr) 5 2 0 0 0 0 71 Deep water (dw) 0 2 0 0 0 0 100 Patchy seagrass 0 0 4 1 0 0 80 (ps) Patchy reef (pr) 0 0 1 14 0 0 93 Sand (s) 0 0 0 0 9 0 100 Algae (alg) 1 0 0 1 0 17 89 Producer 83 100 80 88 100 100 Accuracy (% )
Total accuracy: 89 %
6.1.4. Concluding Rem arks In this section, a remote sensing technique to compensate for the water column effect by using the depth invariant method has been tested to the QuickBird image. It is concluded that the depth invariant index can be applied to enhance bottom type information and compensate the water column effect.
The results demonstrate that water column correction can increase the quality of the coral reef habitat map. The accuracy of patchy seagrass, patchy reef, sand and algae have increased after water column correction. This was respectively 60%, 80%, 78% and 47% before water column correction and 80%,93%,100% and 89% after the correction. The increase of the accuracy of patchy seagrass, patchy reef, sand and algae shows that by applying depth invariant index method, the bottom type information has been enhanced.
However, the accuracy of coral reef and deep water are similar (71% and 100% respectively), before and after water column correction. This similarity may be because these bottom types are quite distinct compared to other habitats so there was no improvement in term of accuracy. Another reason may be due to insufficient validation points (total 7 and 2 points for
49 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. coral reef and deep water respectively). Total accuracy of coral reef habitat map has increased before and after water column correction from 67 % to 89% respectively.
6.2. Generating a coral reef habitat m ap using in situ spectral data The output of the Lyzenga method is an index value of dept-invariant, which enhances the different bottom types. However these values are derived from a pair of bands. In this section is explained how the modified Bierwirth method is applied. This method compensates for the water column effects and the output result of wet bottom reflectance, thus it is possible to link it to field spectral reflectance data.
In order to compensate for the water column effect, the modified Bierwirth method requires several input parameters, which are shown in figure 6.6.
In situ deep-water reflectance measurement below the water
œ2kz surface 1/0.545*RA œ (1- e )*Rw
R= œ2kz e
Water depth and attenuation coefficient are derived from Jupp method
Rb = (wet) substrate reflectance 0.545 = ”Water-Air-Transmission‘ factor
RA = apparent surface reflectance k = effective attenuation coefficient of the water body z = water depth (in metres) Rw = water column reflectance of optically deep water
Figure 6.6. Input parameters of the modified Bierwirth algorithm (Purkis and Pasterkamp 2003).
In the modified Bierwirth method are the water depth and the attenuation coefficient derived from the Jupp method (also known as DOP zone method).
The Jupp method was a method to retrieve a bathymetrical map from optical remote sensing data. Beside the water depth values of each pixel, the output of this method is also the water column attenuation coefficient. The description of the Jupp method is given in section 5.6. and the application of this method is presented in the following section (section 6.2.1.).
50 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
6.2.1. Generating the bathym etric m ap There are four processing steps in the Depth Of Penetration (DOP) method: 1. Convert in situ depth points to the depth at the time of image acquisition. 2. Creating depth of penetration (DOP) zones. 3. Interpolation of the DOP zones. 4. Accuracy assessment of the crude bathymetric map derived from the DOP zones method.
1. Convert in situ depth points to the depth at the time of image acquisition. Depth measurements were carried out during the fieldwork. The locations of the depth measurements are presented in figure 4.2. The depth values measured in the field were subtracted from the high tide values at the time of measurement, to obtain standard datum values. The tidal data are retrieved from the Hidro-Oceanography Service of the Indonesian Navy (DISHIDROS) of October 2003 (see appendix C).
Once the standard datum values were obtained, the next step was to add the tidal height value at the time of image acquisition, to standard datum depth values. The QuickBird multi-spectral images were acquired on 2 October 2003, at 10:23 (GMT). The tidal height data retrieved from DISHIDROS at the time of image acquisition is 1,4 m. This value was added to the standard datum values. The results of the conversion are listed in appendix D.
2. Creating depth of penetration (DOP) zones: a. Select an area of deep water, representing the optical properties of the study area. To select the deep-water area, the QuickBird image before the masking was used. Then calculate the maximum, minimum and mean pixel for each image band by displaying the histogram of the deep-water subset. The results of the deep water DNs of each band are presented in table 6.4.
Table 6.4. DN value for deep water of each band
QuickBird band 1 2 3 4
Maximum deepwater DN value (Ldeep max) 11 8 8 6
Minimum deepwater DN value (Ldeep min) 7 5 4 4
Mean deepwater DN value (Ldeep mean) 9 6.5 6 5
b. Estimate the maximum depth of penetration of each band, using the in situ depth points which have been converted to the water depth at the time of image acquisition (output of step 1). The first step is to plot the points on the image to retrieve the values of bands1-4 at measured points locations based on UTM coordinates. Then sort the depth starting from the deepest to progressively shallower depths. After that, one can start to estimate DOP zone. (see Appendix E).
The estimation of the maximum depth of penetration took also into account the reference depth of penetration (listed in table 5.6.). For example, to estimate DOP 1, one has to consider only DN values in Band 1 which have a depth of approximately 25m œ 15m.
51 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Then sort the DN in band 1 together with its depth value, which has DN value higher than
the maximum deep water DN value (>V deep maxb1). For example, the maximum deep- water value for band 1 is 11, so the depth with a DN value of 12 was sorted. After this, the mean value of the depth from the DN value more than the maximum deepwater DN
value was calculated (>V deep maxb1). Then, the mean value of the depth from the DN
values which are the same as deepwater DN value was calculated too (=V deep maxb1). The
average between the depth value (>V deep maxb1) and the depth value (=Vdeep maxb1) is DOP zone 1. The process to estimate the DOP zones is presented in appendix F, and the results of the DOP zones are listed in table 6.5.
Table 6.5. The maximum depth of penetration (zi) in metres. Depths (m) QuickBird band 1 2 3 4
Depth of deepest pixel with DN> L deepmax 23.4 16.3 4.4 2.1
Depth of shallowest pixel with DN ≤ L deepmax 17.1 7.1 3.9 1.4
Average depth of boundary pixels with DN values> L deepmax 19.9 10.8 4.5 1.6
Average depth of boundary pixels with DN values= L deepmax 19.4 10.4 3.9 1.6
Estimated maximum depth of penetration (DOPzi) 19.6 10.6 4.2 1.6
The maximum deepwater DN values (Table 6.4.) and maximum Depth Of Penetration
(DOPzi) in table 6.5. can be combined together to construct a decision tree for assigning pixels within the DOP zones. The decision tree values are listed in table 6.6.
Table 6.6. A decision tree to assign pixels to depth zones QuickBird band 1 2 3 4 DOP zones Deepwater maximum 11 8 8 6 If DN value of pixel Ç11 Ç8 Ç8 Ç6 then depth > 19.6 m If DN value of pixel >11 Ç8 Ç8 Ç6 then depth = 10.6-19.6 m (zone 1) If DN value of pixel >11 >8 Ç8 Ç6 then depth =4.2-10.6 m (zone 2) If DN value of pixel >11 >8 >8 Ç6 then depth = 1.6-4.2 m (zone 3) If DN value of pixel >11 >8 >8 >6 then depth = 0-1.6 m (zone 4)
After the DOP zones are determined, the next step is the masking of the image based on the range of DOP. The pixels, which are within the value of DOP, are assigned value 1, otherwise 0.
For example, the algorithm applied to mask DOP zone2 using band 2 and band 3: MaskDOP2= iff ((atcorb2>8) and (atcorb3<=8),1,0)
Where: MaskDOP2 is the image to mask-out the area outside the DOP of zone 2 Atcorb2 and atcorb3 are the atmospheric corrected images.
52 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
After generating a mask of each DOP zone, the next step is to multiply the DOP zone masks with the atmospheric corrected images. So the pixels within the DOP zone will have its original values and those outside the DOP zone will have 0 value.
For example, the mask of the DOP2 image is multiplied by the atmospheric corrected image of band2, so the pixels outside DOP zone 2 will be assigned zero and the pixels within DOP zone 2 will have the values of the atmospheric corrected image of band2. This step was conducted by applying the following algorithm:
DOPZ2=MaskDOP2*atcorb2
Figure 6.7 shows the output map of MaskDOP2 and DOPZ2.
(a) (b) Figure 6.7. (a) the masking map of DOP2, (b) the zone of DOPZ2
The mask DOP2 (a) has DN value 0 and 1 and the DOPZ2 (b) has DN values of atmospheric corrected image of band 2.
3. Interpolation of the DOP zones. By calculating DOP zones the pixels have not yet assigned a depth value. Interpolation depths within each DOP zone assigned the depth value of each pixel. By using the table histogram of DOPzi, (eg. DOPZ2), the maximum and minimum DN values (Li min and Li max)
within the DOP can be determined. The process to determine Li min and Li max is described in appendix-H .
The DN (Li min and Li max) values within each zone are presented in table 6.7.
Table 6.7. Li min and Li max for each DOP zone i derived from QuickBird image. Band 1 Band 2 Band 3 Band 4 DOP 1 11 - 17 DOP 2 8 - 15 DOP 3 8 - 12 DOP 4 4 - 16
53 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
After the DN (Li min and Li max) values are determined, the attenuation coefficient (ki) and Ai were calculated using equations 15 and 16 (section 5.6.). The values of ki and Ai are listed in table 6.8.
Table 6.8. The values of ki and Ai derived from QuickBird image Band 1 Band 2 Band 3 Band 4
ki 0.04507 0.15650 0.14714 1.11405
Ai 3.781609 5.16939 4.14338 3.28129
Once Ai and ki are known, the depth of the water (z) can be estimated for any pixel in the image by using equation 17 (section 5.6). By applying equation 17, the depth is assigned to each pixel in each DOP band to produce four separate interpolated DOP depth images (Z1,Z2,Z3 andZ4). These DOP depth images are added together to produce a final depth image from the study area. The four separate interpolated DOP depth images and the merged depth image are presented in figure 6.8.
(b) (a)
(e)
(d) (c)
Figure 6.8. DOP depth images Z1 (a), Z2 (b), Z 3 (c), Z4 (d) and Ztotal (e)
54 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
The depth image within DOP zone 1 (Z1) has a depth range of 10,6-19,6 m, the image Z2 has a depth range of 4,2-10,6 m, the image Z3 has a depth range of 1,6-4,2 m and the image Z4 has a depth range less than 1,6 m. So the merged of these depth images (Z1-Z4) is Ztotal, which has a depth range <1,6m-19,6 m.
The final step was to mask the Ztotal image, which only considers the shallow water part.. For this the masked image from section 5.3. was used to obtain only the shallow water part.
The final bathymetric map is presented in figure 6.9.
Figure 6.9. Crude bathymetric map of Derawan Island derived from QuickBird image of 2003
4. Accuracy assessment of the crude bathymetric map derived from DOP zones method. An accuracy assessment of the crude bathymetric map was carried out by comparing the predicted depths of the DOP depth image at each UTM coordinate with those measured in the field (after having converted them to the depth at the time of image acquisition).
Figure 6.10. Scatter plot between predictive depth and measured depth (m)
55 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
The crude bathymetric map gave a correlation of only 0,6 between the predicted and measured depth. The different substrates seem to influence the values of the predicted depth. For example, the coral reef in shallow water will be estimated as a rather deep area. The concept of bathymetric mapping from optical remote sensing is, that more light is reflected from the shallow sea floor because relative less has been absorbed in the water column and on the contrary, less light from the deeper sea floor because more absorption. Coral reef in shallow water absorbs more light compared to sand at the same depth. As a consequence, it is predicted by the DOP method to have a deeper depth than it actual depth has. This also represent that the assumption of this method that the albedo of the substrate is fairly constant (see section 5.6) is correct.
Although the quality of the predictive bathymetric map from the Jupp method was not so good, it was still used to test the effect of even this low accuracy map on the modified Bierwirth method. The parameters that were used were the water depth (z) and the attenuation coefficient (k).
6.2.2. Deep water reflectance Another input parameter as input to the modified Bierwirth method is the deep-water reflectance at location where the water had a depth of ~75m.,Rw, (see figure 6.6.). This parameter was measured in the field just below the surface. Figure 6.11. show the curve of measured deep-water reflectance and Rw mean value of each band.
B1=0,0324 B2=0,0122 B2=0,0016
Figure 6.11. Curve of Rw measured in the field.
6.2.3. Generating wet bottom reflectance from QuickBird Data Once the parameters of Rw and the depth values of each pixel (z) were obtained, the wet substrate (bottom) reflectance (Rb) can be calculated, using equation 12 (section 5.5.). The output maps of Rb for each band are presented in figure 6.12.
56 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
(a) (b) (c)
Figure 6.12. Rb derived from QuickBird image of band 1 (a), band 2 (b) and band 3(c).
The output of Rb in band 1 represents the whole area, however the output Rb in band 2 and band 3 produced a lot of undef pixels. (Figure 6.12). Thus for the next step, only Rb from band1 was considered.
6.2.4. In situ spectral reflectance data preparation Before to link in situ spectral reflectance date to the image, they were grouped into the same classes as used in this study. Figure 6.13. Represents the mean spectral signatures of more detailed classes, prepared by Ana Fonseca.
Note: DCBL=Dead Coral Bleaching; DCCA= Dead Coral Coraline Algae; DCTA=Dead Coral Turf Algae;ACROPO= Coral genus Acropora sp.; BRANPOR= Coral genus Porites sp (branching form); CF= Coral in Folious form;MASSPOR=Coral genus Porites sp. (Massive form); OTHERC= Other coral species; CA=Coraline Algae;CYAN=Cyanobacteria; GALAX= Coral genus Galaxea sp.; HALIMEDA=Macro Algae genus Halimeda.; PADINA=Macro Algae genus Padina; HELIOPORA: Blue Coral; SPONGE; XENIA=Soft Coral genus Xenia;SINULARIA=Soft coral genus Sinularia; OTHER SC=Other Soft Coral; MUD;SAND.
Figure 6.13. M eans of measured spectral reflectance
The mean spectral reflectances presented in figure 6.13. were merged into broader classes such as sand, algae composite and coral reef. These categories also refer to the habitat classes, which have been determined (see section 5.7). However the habitat classes seagrass and patchy reef were not used due to a lack of measured spectral reflectance data. Not all classes that were shown in figure 6.13. were used. The merged means spectral reflectances of algae composite consist of the mean spectral reflectance of DCBL, DCCA and DCTA and the means of spectral reflectance of coral consist of the mean reflectance of Acropora sp. and Massive Porites sp. The mean spectral reflectance of sand is the same as shown in figure 6.13. .The merged mean spectral reflectances into broad classes are presented in Figure 6.14.
57 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Figure 6.14. M eans of merged measured spectral reflectance.
The maximum and minimum values, means plus and minus of standard deviation values spectral of means spectral reflectance of algae composite (alg), sand (s) and coral reef (cr), were used as the threshold for the classification step (Table 6.9.).
Table 6.9. max, min and means plus-minus of standard deviation in band 1 cr alg s B1 means (+) 0.097 0.148 0.262 B1 means 0.087 0.134 0.241 B1 means (-) 0.077 0.119 0.220 B1 max 0.106 0.167 0.278 B1 min 0.073 0.112 0.205
6.2.5. Generating a coral reef habitat m ap The general steps of the classification processes are presented in figure 6.1.b, without training samples selection. The input classification map is the wet bottom reflectance of band 1 (Rb1). A Parallelpiped classification decision rule is used to classify the image.
The parallelpiped classification approach was carried out using the density-slicing option. The data values of the measured spectral reflectances are used as the thresholds (Table 6.9). The limits are the maximum and minimum values, means plus and minus of standard deviation values. The output of the classified image is presented in figure 6.15.
Figure 6.15. Coral reef habitat derived from wet bottom reflectance of band 1 (Rb1).
58 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
The result of the habitat map (figure 6.15) shows that all the habitats are classified as sand. Assessing the histogram of Rb1 (figure 6.16) shows that the minimum pixel value is 0,25. This value is in the range of sand (table 6.9). Therefore all the pixels of Rb1 are classified as sand.
Figure 6.16. Histogram of wet reflectance in band 1 (Rb1)
It is assumed that the low accuracy of the estimated depth and water coefficient attenuations are not good enough to produce wet bottom reflectance with the modified Bierwirth method. To support this assumption the spectral reflectance (table 6.9) were used to classify the atmospheric corrected image in band 1 (indicated as RA in figure 6.6.).
Table 6.9. only indicate the threshold values of habitat class coral reef, algae and sand. The pixel which has the value less than 0,073 (which is min value of coral reef) was assigned to the —deep water“ habitat class. To determine the pixel into class —other“ was used the gap between the upper boundary of algae and lower boundary of sand.
The result of the classified image derived from RA band 1 is presented in figure 6.17.
Figure 6.17. Classified habitat map derived from measured spectral reflectance and RAband 1
59 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
The measured spectral reflectance can be used to classify the atmospheric corrected image of band 1. (Figure 6.17.) The next step is to assess the accuracy of this classified habitat map.
For accuracy assessment, the same approach has been used as described in section 6.1.1. The coral reef habitat map is independent from ground truth habitat points. The result of the accuracy assessment is presented in table 6.10.
Table 6.10. Error matrix of the classified image using in situ spectral reflectance (cr) (dw) (ot) (s) (alg) User Accuracy (% ) Coral reef (cr) 3 0 0 0 4 43 Deep water (dw) 0 2 0 0 0 100 Other (ot) 0 0 16 0 4 80 Sand (s) 0 0 1 8 0 89 Algae (alg) 0 0 0 0 19 100 Producer Accuracy (% ) 100 100 94 100 70
Total accuracy 84%
6.2.6. Concluding Rem arks In this section, optical remote sensing data of QuickBird image has been used to produce a Bathymetric map of Derawan Island by applying the Deph of Penetration Zone method (also known as Jupp method). The produced bathymetric map has only a correlation 0.6 when plotted against field depth measurements.
The different bottom substrates in the study area influence the value of the crude bathymetric map. However, one of the assumptions of this method is that the albedo of the substrate should remain constant. In this case coral reef in shallow water absorbs more light compared to sand at the same depth. As a consequence, it is predicted as a deeper depth than it actual is. So it is assumed that the crude bathymetric map derived from the Jupp method will be more accurate if the bottom substrates are relatively homogenous.
Although the predictive bathymetric map from the Jupp method is not so good, it was still used as input parameter for the modified Bierwirth method. This was done to test how much the effect of low accuracy of bathymetric map was on the result. The test result demonstrates that the low accuracy of the water depth and water attenuation coefficient caused an error in the output of wet bottom reflectance. Furthermore, it was not possible to link the image to measured spectral reflectance in order to generate a reliable coral reef habitat map. It is concluded that the crude bathymetric map and water coefficient attenuation derived from Jupp‘s method are not accurate enough as input parameters for the modified Bierwirth method.
The atmospheric corrected image of band 1 was tested and classified using in situ spectral reflectance. The classified image was able to determine coral reef, sand and algae with an accuracy of 43%, 89% and 100% respectively. The results indicate that there was an over estimate of algae habitat class and an under estimate of the coral habitat class. The total
60 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia. accuracy of the coral reef habitat map generated from measured spectral reflectance to full atmospheric corrected image is 84%.
6.3. Generating a coral reef habitat m ap for habitat changing assessm ent In order to assess the capability of remote sensing data for change detection of coral reef habitats, Landsat TM5 of 1991 and Landsat ETM 7 of July 2002 were used.
The classification steps for both Landsat images are presented in figure 6.1.b. The input classification map is a depth invariant index of band 1 and band 2 (DI_12). Ground truth habitat points were not used during the selection of training samples. When the ground truth habitat points were used as training samples, their feature did not match with its label class. This missmatch may be due to a GPS position error when plotted on the Landsat image. Therefore training sample points were selected based on user knowledge of the study area, the output habitat map of QuickBird 2003, and the evaluation of feature space. After the selection of the training samples, a paralellpiped classification decision rule is used to classify the image. The coral reef habitat maps derived from Landsat 1991 and Landsat 2002 are presented in figure 6.18.
61 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
1-a 1-b
2-a 2-b
3-a 3-b
Figure 6.18. Coral reef habitat map of Derawan Island 1991 (1,2,3-a) and 2002 (1,2,3-b)
62 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
The comparison of the coral reef habitat maps of 1991 and 2002 shows a decrease of the coral reef habitat. This is clearly shown in the Western part of Derawan Island (figure 6.18-1a,b). Based on the reports and interviews with local people, there was no significant natural disturbance occurred in Derawan Islands.
Figure 6.19. Threats in Derawan Islands (Source: The Nature Conservancy in progress)
According to the thematic map of threats in Derawan Islands, produced by The Nature Conservancy, this particular area is affected by fish bombing. (Figure 6.19). Siahainenia et.al (1999) conducted a coral reef assessment in 1999; 14 bomb pits were found in the Western part of Derawan Island during their survey. He also noted that coral reef degradation in Derawan Island is about 75% (comparing observations between 1994 and 1999). It takes decades or more for damaged coral reefs to recover.
Beside the bombing, other human activity such as boat transportation may also cause the degradation of the coral reef. For example, the dive resort in the Southern part of the island (indicated with a circle in figure 6.18-3a,b) started its operation in 1992. Based on the habitat map of 1991, there was more coral reefs in this area compared to the map of 2002, when the area was dominated by algae composite and patchy reef. This habitat degradation might be due to the fact that this zone is used as a dive boat transportation track. The overall habitat change between 1991 and 2002 is presented in figure 6.20.
Figure 6.20. Coral reef habitat classes shifting between 1991 and 2002.
63 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Based on figure 6.20. the percentage of coral reef and patchy seagrass have decreased between 1991 and 2002, whereas the percentage of algae composite and patchy reef has increased.
6.3.1. Concluding Rem arks It has been demonstrated in this section how remotely sensed data can be used to detect the changes of coral reef habitat. Using data of Landsat TM5 and Landsat ETM7 of 1991 and 2002, it is possible to monitor the historic status of the coral reef environment in Derawan Island.
There was a coral reef habitat change between 1991 and 2002, which represents a decrease of the percentage of coral reef and patchy seagrass and on the other hand an increase of the algae composite and patchy reef percentage. The threats of coral reef in Derawan Islands were mainly from human activities in particular blast fishing. In addition the accumulation of oil spills from the boats and domestic pollution were also contributing to the degradation of the coral reef in this area.
64 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
7. Conclusions and Recom m endations
The thesis describes the results of the research on assessing the capabilities of satellite remote sensing techniques combined with field data collection for coral reef habitat mapping and change detection. The research has been tested in Derawan Island, Berau, East Kalimantan, Indonesia. Several input data have been collected and determined, such as Landsat TM-5 of 1991, Landsat ETM-7 2002 and QuickBird of 2003. Secondly, information on the physical coastal environment area retrieved from the reports. Thirdly, fieldwork data, which was conducted in Derawan Island in October 2003.
The conclusions of this thesis are structured based on the research objectives in section 1.3.2.
7.1. Assessing the capability of QuickBird im age to generate a coral reef habitat m ap with water colum n correction ñ The total accuracy of coral reef habitat map has increased before and after water column correction from 67 % to 89% respectively. ñ The mapping accuracy of patchy seagrass, patchy reef, sand and algae have increased before and after water column correction, which is 60%, 80%, 78% and 47% respectively before correction and 80%,93%,100% and 89% after water column correction. ñ The increasing of the accuracy of patchy seagrass, patchy reef, sand and algae shows that by applying depth invariant index method, the bottom type information has been enhanced. ñ However, the accuracy of coral reef and deep water is similar (71% and 100% respectively), before and after water column correction. This similarity may be due to the fact that bottom types are quite distinct compared to other habitats Another reason may be due to insufficient number of validation points (total 7 and 2 points for coral reef and deep water respectively). ñ The classification of coral reef habitat map of QuickBird image increased 22% by applying a water column correction using the depth invariant index method. ñ Depth invariant index can be applied to enhance bottom type information by compensate for the water column effect; this approach is relatively easy to apply.
65 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
7.2. Integrating in situ spectral data with m ulti-spectral QuickBird im age data for coral reef habitat classification. ñ The crude bathymetric map has a correlation of only 0.6 when plotted against field depth measurements. ñ The different bottom substrates in the study area influenced the outcome values of crude bathymetric mapping. Coral reef in shallow water absorbs more light compared to sand at the same depth. As a consequence, it is predicted as deeper than it actual is. ñ The crude bathymetric map derived from the Jupp method will be more accurate in case of relative homogenous bottom substrates are. ñ The crude bathymetric map and water coefficient attenuation derived from Jupp‘s method are not accurate enough as input parameters for the modified Bierwirth method in order to generate wet bottom reflectance. ñ The low accuracy of the water depth and water attenuation coefficient as input parameter for the Bierwirth method caused an error in the output of the wet bottom reflectance. Therefore it was not possible to link the image to to measured spectral reflectance in order to generate a coral reef habitat map. ñ In Derawan Island, in a situation with only shallow and clear water condition, it is possible to link the measured spectral reflectances to the full atmospheric corrected image (without water column correction). ñ The total accuracy of the coral reef habitat map generated from measured spectral reflectance to full atmospheric corrected image is 84%. And it was able to determine coral reef, sand and algae with the accuracy of 43%, 89% and 100% respectively.
7.3. Asessing the change of the coral reef habitat by com paring classified Landsat im ages of 1991 and 2002. ñ There is a coral reef habitat change between the periods 1991 to 2002, which can be detected by satellite remote sensing data. ñ The habitat change consist of a decrease of the percentage of coral reef and patchy seagrass and at the other hand an increase of the algae composite and patchy reef percentage. ñ The threats of coral reef in Derawan Islands were mainly from human activities in particular blast fishing.
66 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
7.4. Recom m endations ñ The collection of ground truth habitat points should consider the spatial resolution of the remotely sensed data be used and also the possibility of errors of the GPS measurements. ñ A good geometric and atmospheric image correction are very important in order to make a coral reef habitat map using ground truth habitat points and in situ spectral measurement. ñ In order to improve the accuracy of the modified Bierwirth method, more accurate bathymetric map should be available. Using an acoustic instrument can be an alternative to obtain bathymetric data. With a more advanced acoustic instrument with the capability to define substrate types, a substrate habitat map can be generated automatically. ñ To make a detailed coral reef habitat map using in situ spectral measurements, high spatial and spectral image data are needed.
67 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
References
Bierwirth PN, Lee TJ, Burne RV (1993) Shallow Sea-Floor Reflectance and Water Depth Derived by Unmixin Multispectral Imagery. Photogrammetric Engineering & Remote Sensing 59: 331-338 Bryant D, Burke L, McManus JW, Spalding MD (1998) Reefs at Risk - A map based indicator of threats to the world's coral reefs. World Resources Institute, Washington, D.C. USA Dobson EL, Dustan P (2000) The Use of Satellite Imagery for Detection of Shifts in Coral reef Communities American Society for Photogrammetry and Remote Sensing, Washington D.C. Edwards AJ (1999) Applications of Satellite and Airborne Image Data to Coastal Management. UNESCO, Paris Hochberg EJ, Atkinson MJ (2000) Spectra discrimination of coral reef benthic communities. Coral reefs 19: 164-171 Hochberg EJ, Atkinson MJ (2003) Capabilities of remote sensors to classify coral, algae, and sand as pure and mixed spectra. Remote Sensing of Environment 85: 174-189 Hochberg EJ, Atkinson MJ, Andrefouet S (2003) Spectral reflectance of coral reef bottom- types worldwide and implications for coral reef remote sensing. Remote Sensing of Environment 85: 159-173 Holden H, LeDrew E (1998) Spectral Discrimination of Healthy and Non-Healthy Corals Based on Cluster Analysis, Principal Components Analysis, and Derivative Spectroscopy. Remote Sensing of Environment 65: 217-224 ICRI (1995) International Coral Reef Initiative Framework For Action. International Coral Reef Initiative (ICRI), Dumaguete, Philippines Ismuranty C (2002) Build the co-management for the conservation and suistainable use of the Derawan Islands, East Kalimantan, Indonesia. KEHATI, Jakarta Janssen LLF, Huurneman GC (2001) Principles of Remote Sensing. ITC, Enschede Jompa H, Pet-Soede L (2002) The coastal fishery in East Kalimantan - A rapid assessment of fishing patterns, status of reef habitat and reef stocks and socio-economic characteristics. WWF Indonesia, Indonesia Khan MA, Fadlallah YH, Al-Hinai KG (1992) Thematic mapping of subtidal coastal habitats in the western Arabian Gulf using Landsat TM data-Abu Ali Bay, Saudi Arabia. International Journal Remote Sensing 13: 605-614 Krause K (2003) Radiance Conversion of QuickBird data. Digital Globe, Colorado, USA LIPI (1995) Marine survey to support Marine Resources Evaluation and Planning (MREP) in Derawan Islands region. Indonesian Scientific Institute - LIPI, Ambon (Indonesian Language). LIPI (2001) Inventory and Evaluation for the Potency of New Marine Conservation Area of Derawan, Kakaban and Maratua Island, Sub-district of Derawan Islands, Berau Region, East Kalimantan Province. Indonesian Scientific Institute - LIPI, Jakarta. (Indonesian Language). Lyzenga DR (1981) Remote sensing of bottom reflectance and water attenuation parameters in shallow water using aircraft and Landsat data. International Journal Remote Sensing 2: 71-82
68 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Maeder J, Narumalani S, Rundquist DC, Perk RL, Schalles J, Hutchins K, Keck J (2002) Classifying and Mapping General Coral-Reef Structure Using Ikonos Data. Photogrammetric Engineering & Remote Sensing 68: 1297-1305 Malik R, Kasmawaty, Mursidi, Muhammad A, Setiabudi A (1999) Field Survey Report, (Derawan, Samama and Sangalaki Islands). Coastal Resources Management Project- CRMP, Jakarta. (Indonesian Language). Maritorena S (1996) Remote sensing of the water attenuation in coral reefs: a case study in French Polynesia. International Journal Remote Sensing 17: 155-166 Mumby PJ, Clark CD, Green EP, Edwards AJ (1998a) Benefits or water column correction and contextual editing for mapping coral reefs. International Journal Remote Sensing 19: 203-210 Mumby PJ, Green EP, Clark CD, Edwards AJ (2000) Remote Sensing handbook for Tropical Coastal Management. In: 3 (ed) Coastal Management Sourcebooks. UNESCO, Paris, pp 316 Mumby PJ, Harborne AR, Raines PS (1998b) Classification Scheme for Marine Habitats of Belize. Coral Cay Conservation Purkis S, Kenter JAM (2001) REEF RECAP - Remote monitoring of shallow, tropical carbonate platform habitats as an indicator of ecosystem change Purkis S, Pasterkamp R (2003) Integrating in situ reef-top reflectance spectra with Lndsat TM imagery to aid shallow-tropical benthic habitat mapping. Vrije University Amsterdam, Amsterdam, The Netherlands Reese ES, Crosby MP (2000) Assessment and monitoring of coral reefs: asking the right question. University of Hawaii Richter R (2002) ATCOR for ERDAS IMAGINE - Atmospheric and Topographic Correction ATCOR2 and ATCOR3 (Ver.2.0) User Manual. Geosystems, Germering Siahainenia AJ (1999) Study of Spatial Planning of Derawan Islands. Coastal Resources Management Project - CRMP, Balikpapan. (Indonesian Language). Spitzer D, Dirks RWJ (1987) Bottom influence on the reflectance of the sea. International Journal Remote Sensing 8: 279-290 Tomascik T, Mah AM, Nontji A, Moosa MK (1997) The ecology of the Indonesian Seas - Part two. Periplus Editions, Singapore UNESCO (2002) The Hanoi Statement. UNESCO, Paris
69 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
APPENDIX A - TEAM OF EAST KALIMANTAN PROGRAMME 2003
Contact Person Institutions Biodiversity Dr. BERT WILLEM HOEKSEMA Naturalis Darwinweg, Leiden P.O. Box 9517 NL-2300 RA Leiden The Netherlands tel. +31 71 5687600 fax +31 71 5687666 Internet: www.naturalis.nl Dr. SUHARSONO LIPI- Indonesian Institute of Science Jl. Pasir Putih 9 Ancol œ Jakarta Indonesia Internet:www.lipi.go.id
Reef Ecology Prof. Dr. ROLF PIETER MARTIN BAK NIOZ - The Royal Netherlands Institute For Sea Research P.O. Box. 59 1790 AB Den Burg Texel The Netherlands Internet: www.nioz.nl
Seagrass Dr. TJEERD JORIS BOUMA NIOO-KNAW CEM E P.O. Box. 40 4400 AC Yerseke The Netherlands Tel. +31 133 57 73 Fax: +31 133 57 36 16 Internet: www.nioo.knaw.nl Drs. WAWAN KISWARA LIPI - Indonesian Institute of Science Jl. Pasir Putih 9 Ancol œ Jakarta Indonesia Internet:www.lipi.go.id
Remote Sensing
70 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
Dr. TJEERD HOBMA ITC œ International Institute for Geo- Information Science and Earth Observation Hengelosestraat 99 PO Box. 6 7500 AA Enschede The Netherlands Tel. +31 53 487 44 44 Fax: +31 53 487 44 00 Internet: www.itc.nl
71 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
APPENDIX B œ VISIBILITY DATA Source: Meteorology Balikpapan - INDONESIA
72 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
APPENDIX C œ TIDAL DATA OCTOBER 2003 Source: DISHIDROS -INDONESIA
73 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
APPENDIX D œ EXAMPLE OF CONVERTION MEASURED DEPTH TO THE DEPTH OF IMAGE ACQUISITION.
UTM UTM Measured Measured Tidal Datum Im age (QB) Date Easting N o rthing Tim e Depth(m ) Height (m ) Depth (m ) Depth (m ) 14-OCT-03 641783 252163 11:45 14.2 0,5 13,7 15,1 641862 252025 10.8 0,5 10,3 11,7 641862 251957 9.2 0,5 8,7 10,1 641893 251863 10.1 0,5 9,6 11 641896 251795 14 0,5 13,5 14,9 641880 251747 12:00 9.9 0,5 9,4 10,8 641873 251693 12 0,5 11,5 12,9 641847 251710 4.9 0,5 4,4 5,8 641839 251507 9.8 0,5 9,3 10,7 641850 251431 13.2 0,5 12,7 14,1 641851 251364 10.3 0,4 9,9 11,3 641858 251282 4.8 0,4 4,4 5,8 641850 251252 4.6 0,4 4,2 5,6 641849 251172 4.1 0,4 3,7 5,1 641863 251119 2.9 0,4 2,5 3,9 641932 250989 12:28 4.4 0,4 4 5,4 641888 251052 14:50 2.3 0,6 1,7 3,1 641959 250959 4.6 0,6 4 5,4 642003 250935 15 0,6 14,4 15,8 642002 250849 4.4 0,6 3,8 5,2 642008 250812 1.8 0,6 1,2 2,6 642026 250750 1.7 0,6 1,1 2,5 642056 250725 15:00 2.1 0,7 1,4 2,8 642116 250679 2.1 0,7 1,4 2,8 642165 250690 6.6 0,7 5,9 7,3 642193 250655 7.7 0,7 7 8,4 642228 250641 15:11 19.7 0,7 19 20,4 642248 250612 15:15 7.7 0,8 6,9 8,3 642245 250580 3.4 0,8 2,6 4 642235 250550 4.4 0,8 3,6 5 642218 250506 1.2 0,8 0,4 1,8 642276 250448 15:24 2,9 0,9 2 3,4 642294 250424 4.4 0,9 3,5 4,9 642321 250394 16.6 0,9 15,7 17,1 642309 250371 15:40 8.4 1 7,4 8,8 642304 250343 4 1 3 4,4 642270 250293 1.2 1 0,2 1,6
74 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
APPENDIX E œ EXAMPLE OF SORTED THE DEPTH
Easting Northing QB depth (m) QB-b1 QB-b2 QB-b3 QB-b4 641035 244812 27,5 12 7 5 5 638640 248057 27,4 13 7 6 5 638306 251647 27,3 13 9 7 6 640966 245128 27,3 12 9 5 5 638576 248060 27,2 12 7 4 4 638299 248005 27,1 12 8 5 4 641091 244793 27,1 14 8 6 5 638805 248142 27,1 12 7 6 5 638740 248066 26,9 12 7 6 5 639748 248002 26,7 14 8 5 4 638518 248054 26,7 13 7 5 4 638757 248069 26,7 13 7 5 5 638717 248064 26,7 13 7 5 4 638670 248058 26,4 13 7 5 4 638550 248058 26,3 13 7 5 5 638695 248064 26,3 11 7 5 4 641122 244780 26,2 14 7 5 4 638852 248110 25,7 14 7 6 4 638787 248084 25,4 13 7 5 4 639527 251740 25,3 13 7 5 5 639806 247973 25,3 12 9 6 4 644014 247125 24,1 13 7 5 4 641013 245090 24,9 13 8 5 4 644109 246940 24,9 12 7 5 5 639953 247929 23,9 12 8 5 5 638885 249526 23,8 12 9 6 6 638918 249540 23,4 14 10 7 6 639558 249735 23,4 12 7 5 4 643014 249123 23,4 12 8 7 6 638280 247990 23,1 14 7 6 5 638743 252202 22,9 13 7 5 4 638215 247966 22,7 13 7 6 5 639566 251717 22 13 7 5 4 644098 246920 22 13 7 5 4 639026 249592 21,7 13 7 5 4 644184 246501 21,4 11 7 5 4 643967 247143 21 14 7 5 4 639612 251650 21 14 8 5 4
75 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
APPENDIX F œ EXAMPLE TO DETERMINE DOP ZONE
depth over pass (m ) QB1 (L)
25,3 12 depth over 24,1 13 pass (m ) QB1 (L) 24,9 13 25,3 12 24,9 12 24,9 12 23,9 12 23,9 12 23,8 12 23,8 12 23,8 14 23,4 12 23,4 12 23,4 12 23,4 12 19,2 12 23,1 14 19,1 12 22,9 13 19 12 22,7 13 19 12 22 13 18,9 12 22 13 18,2 12 21,7 13 18,1 12 21,4 11 17,7 12 21 14 17,7 12 21 14 17,6 12 20,6 13 17,4 12 20,4 11 17,3 12 Average DOP 20,3 11 17,2 12 depths Z1 19,9 12 >V deep m ax 17,1 12 19,91 19,63 19,8 12 <=V deep m ax 21,4 11 19,36 19,6 13 20,4 11 19,4 11 20,3 11 19,2 12 19,4 11 19,1 12 18,6 11 19 12 18,3 11 19 12 17,1 11 19 14 18,9 12 18,9 14 NOTE: 18,7 13 V deep max B1= 11 18,6 11 (see table 6.4) 18,4 13 18,3 11 18,3 14 18,2 12 18,1 13 18,1 12 18 13 17,8 13 17,7 12 17,7 12
76 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
17,6 13 17,6 12 17,4 13 17,4 12 17,3 12 17,2 13 17,2 12 17,1 13 17,1 13 17,1 11 17,1 12 16,9 13 15,8 13 16,3 15 16,4 14 16,7 13 15,4 14 15,1 13 15 13
77 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
APPENDIX G œ EXAMPLE OF SAND SUBSTRATE PIXELS AT VARIOUS DEPTHS
QB-b1 QB-b2 QB-b3 19 14 12 14 12 11 13 11 11 15 13 12 15 13 11 15 12 11 14 11 10 14 12 11 15 13 11 15 13 11 17 14 12 16 13 11 16 14 11 16 13 11 16 13 11 17 14 11 17 14 11 17 14 11 20 15 13 18 14 11 17 14 11 18 14 12 18 14 12 16 13 11 18 14 12 16 12 11 14 11 11 15 12 11 15 11 11 16 13 11 17 14 12 17 14 12 15 13 11 13 10 11 16 13 12 18 15 14 17 14 13 16 14 12 18 15 14 16 13 12 17 13 13
78 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
APPENDIX H œ EXAMPLE TO DETEMINE Lim in AND Lim ax OF EACH DOP ZONE (Table 6.7).
Table histogram of DOPZ1
The zero values was not considered Because it represents the area Outside DOPZ1.
The total number of pixel within The DOP zone is: Total sum œ npix (value 0) =20641536-20497868 =143668 number of pixel
To determine Li min and Li max Only consider approximately 0,1% Of the total number of pixels within The DOP zone. So 0,1% from 143668 is ~143.
According to column —npix“ the value for Lmin is 11 and the value for Lmax is 17. The value of 17 was considered to be selected instead of the value of 16 because the number of pixel in value 17 is relatively close to 143.
79 The application of QuickBird and multi-temporal Landsat TM data for coral reef habitat mapping. Case study: Derawan Island, East Kalimantan, Indonesia.
1