Project Topics
Get Complete Project Material File(s) Now! »
Coral reefs of Indonesia
Indonesia is the country with the highest presence of coral reefs worldwide, currently estimated at 50.200 km2 of reefal area although this number needs revision. Indonesia has numerous reef types, including barrier reefs, fringing reefs, and atolls for the main ones (Tomascik et al., 1997). Indonesia coral reefs are renowned for their biological diversity and ecological complexity, and are located at the epicentre of the global centre of tropical marine biodiversity, the so called Coral Triangle, which includes Indonesia, Malaysia, Philippines, Timor Leste, Papua New Guinea and Salomon Islands. Within this broad area, reef-building coral diversity exceeds 500 species which correspond to more than 70% of the total Indo-Pacific species (Veron,2000). This area also supports the highest diversity of reef-associated fishes (Allen and Steene, 1994), and is clearly of global significance as one of the main reservoir of tropical marine biodiversity (Turak and DeVantier, 2003).
Unfortunately, a comprehensive mapping of the geomorphology, or habitats of coral reefs in Indonesia remain unavailable. Much is needed in terms of understanding the ecology of Indonesian coral reefs, at all scales. As such, the INDESO project aims to contribute some of the gaps, in particular on several targeted National Parks that still lack remote sensing derived habitat maps and information on the dynamics and resilience of coral reefs.
Remote sensing of coral reef habitats and habitat mapping
Generally, a habitat can be defined as a spatial and functional entity characterized by various biological and abiotic parameters, at a specific spatial scale that will depend on the context (Galparsoro et al., 2012). For habitat mapping using remote sensing, habitats are described by four types of variables that refer respectively to geomorphology, architecture, benthic cover and taxonomy of the dominant structurally species for areas that may cover between several square meters to few thousands of square meters (Fig. 1) (Andréfouët, 2014, 2016). The inhabitants of interests are typically vertebrates and invertebrates fauna, and flora, because they are either ecological resources or functional entities that need to be mapped to study an important process, or architectural component of the habitat himself (coral, algae, seagrass…).
Figure 1: Definition of a habitat in a remote sensing context. A habitat is fully resolved when four variables are explicitly described: geomorphology, architecture, benthic cover and taxonomy (from Andréfouët 2014).
A reef habitat is thus explicitly considered here as a three dimensional benthic biophysical structure covering at least a few tens of square meters.
The geomorphology axis (Fig. 1) is the coarser component. It is a proxy of the depth, physical environment and formation of the location. Often, to map geomorphology, field work and ground-truthing are not necessary. Geomorphology can be mapped directly from the image without field data (Andréfouët et al., 2006). In contrast, the other component of the habitat (cover, architecture and taxonomy) need field investigation to be characterized qualitatively or quantitatively using appropriate field methods that will be presented in Chapter 2. Benthic cover describes how the area is covered by either biotic or abiotic entities. Architecture refers to different information that can be integrative (like the rugosity –or variation of height – of the habitat) or component specific (like the growth form of corals – tabular, massive, branching, etc., or the height of the canopy of an algal bed). Architecture can also refer, at another level, on the spatial topology of the components of the habitat, like patchiness or degree of fragmentation for instance. Examples of representative reef habitats are provided hereafter (Fig. 2) and details of their description are provided in Table 1.
The type of methods depends on the selected approach for image processing (classification, segmentation, physics-based, artificial intelligence, or a mix of these approaches), the type of sensor (spatial and spectral resolution), the feasibility of ground-truthing, and, most importantly the objectives of the mapping. Examples of these approaches applied to coral reefs can be found inPurkis (2005, physics-based), Andréfouët et al. (2003, classification-based) and Roelfsema et al. (2013, segmentation based), and Benfield et al. (2007, rule-based classification, related to artificial intelligence) for instance. Here, we can narrow the scope considering several constraints which are specific to the INDESO and Indonesia context briefly explained in the previous sections.
First, the priority of INDESO is to develop pilot projects using very high spatial resolution (2-4 meter), multispectral (4-8 bands), images. Second, Indonesian reefs being in the epicentre of the coral reef diversity, it is desirable to try to inventory and map the highest number of habitats, an exercise still not achieved in for this country. Third, it is expected that intensive ground-truthing is possible for the pilot INDESO sites. Fourth, INDESO is about capacity building so that as many reefs as possible can be mapped in the future. The ultimate priority is thus production of thematically relevant maps, not method development. Approaches that require limited technical skills, especially in radiative transfer physics, should be favoured, if they allow the production of thematically rich coral reef habitat maps.
In a review paper, Andréfouët (2008) has made recommendations for exactly the context presented above: priority to production of thematically rich maps using very high spatial resolution images in a capacity building context. The methodology he recommends is a “user’ methodology, in the sense that the goal is to produce maps interesting for users and not focus on methodological aspects that can be of interest for map “producers”, but not users. Users can be managers, or scientists that need spatial information on habitats, but cannot produce it themselves.
A methodological flow chart was provided in Andréfouët (2008) (Fig. 3). The user and producer flow chart are compared. The user flow chart is simpler but still has several mandatory steps. The three main ones thematically speaking are the description of habitat typology (step 6), the photo-interpretation of the images (steps 7 – 8) and the accuracy assessment(step 10 (Andréfouët, 2008). Then, other steps are required and are related to image enhancement for photo-interpretation, sampling strategy for field work, and finally transfer into Geographical Information Systems (GIS) format.
Figure 3: Producers and users flow charts. Items in grey boxes show steps independent of the thematic scope. Arrows point to most frequent need of iterations to enhance accuracy and frequency of actions (1, 2, 3, 4).
Accuracy assessment is a necessary task in habitat mapping. The goal is to provide a quantification of the accuracy of the map, which is key information for all users. Many metrics exist (Foody, 2002) that require a set of observations independent from the set of observations used to take the map. (Congalton and Green, 1999) make recommendations in term of sampling effort, with 50 independent points per class for a robust assessment. However, in a coral reef context, this may not be easily feasible, and these guidelines need to be considered as guidelines only, not mandatory. Collecting 50 points per class may be impractical because some classes may be rare with limited coverage, or simply not enough time is allowed in the type of short expedition survey typically done for coral reefs. For a 50-classes map, collecting 500 points may be a costly dedicated full-week task in the field. More importantly it is also very easy to bias accuracy assessment, either positively or negatively depending on where the control points are selected (Andréfouët, 2008). Overall, it is acknowledged that a good accuracy is above 80% overall accuracy, meaning that 80% of all pixels are correctly classified. It also means that 20% of the pixels are mislabelled, which may still be inacceptable for some applications. Therefore, some applications may require values above 90%, if the map is used for precise management of resource stocks for instance. Most of the time, the level of accuracy is highly variable between classes, and overall accuracy is just an indicator that needs to be refined by other metrics provided per classes.
Generally, automatic methods (classification, segmentation) alone cannot reach this level of accuracy for more than 5 or 6 classes in a coral reef environment (Capolsini et al., 2003, Andréfouët et al., 2003). This is due to the inherent radiometric similarity between coral reef benthic classes, such as coral and algae (Hochberg et al., 2003). Higher complexity maps need to be manually edited, by contextual editing (Groom et al.,1996). This means that generic contextual rules (e.g., ‘seagrass are not on the forereef’) can be used a posteriori to detect misclassification and correct them. In practice, implementing automatically these rules is complex (with methods related to the field of artificial intelligence and automatic learning). Some software (e.g., Definiens) can be helpful (Benfield et al. 2007) or it is done manually by photo-interpretation (Andréfouët, 2008). Although this is site-complexity dependent, the trend is that aiming for a high accuracy for a high number of classes require lots of editing. As a result, direct photo-interpretation and manual digitization, which is in practicea priori contextual editing, is preferred as the main method to produce a map. Photo-interpretation has allowed good level of overall accuracy (>75%) for maps reaching more than 50 classes.
Remote sensing of Indonesian coral reefs
Indonesia has vast areas of coral reefs, and published remote sensing studies describe imageprocessing in several localities. Covering the entire country, national initiative included a COREMAP product derived for Landsat 7 images (2004). The Millennium Mapping Project has released some detailed geomorphological products (Andréfouët et al., 2006), also derived from Landsat 7, after the 2004 Tsunami. Comparison with COREMAP products showed many missing reefs and errors in the COREMAP product (Brian Long and Serge Andréfouët, unpublished data). The fully validated Millennium product is not yet distributed and a simplified, unvalidated version of the Indonesian products have been used by UNEP-WCMC to release a global “coral reef’ product in one single layer without thematical detail. This product also suffers from many errors (Cros et al., 2014). The Table 2 provides recent high spatial resolution studies on Indonesian coral reefs published in the peer-reviewed literature. Other studies can be found in conference proceedings and student thesis. These studies represent a variety of coral reef theme, including change detection (Table 2). These studies looked at SPOT and Landsat imagery, thus analysed changes using 20-30 meter resolution. We did not find example of change detection study using very high spatial resolution images (2-4 m).
Resilience of coral reefs and mapping resilience
Resilience is the ability of a system to absorb or recover from disturbance and change, while maintaining its functions and services (Carpenter et al., 2001): for example a coral reef ’s ability to recover from a hurricane(Grimsditch and Salm, 2006). It is often opposed to resistance, which is the ability of an ecosystem to withstand disturbance without undergoing a phase shift or losing neither structure nor function (Odum, 1989): for example a coral reef ’s ability to withstand bleaching and mortality (Grimsditch and Salm, 2006).
Resilience and the resilience concept have been a significant focus in the past 20 years, triggered by the on-going, obvious, degradation of corals reefs that seemed to be unable to bounce back to their initial state, or even shift to a seemingly different system, for instance dominated by algae. Understanding resilience and managing the resilience capacities of a reef have appeared as new priorities for science and management. The difficulty is that managing resilience implies a holistic, ecosystem, view of how a reef is functioning and the consequences of all interactions between all its diversity of components.
Resilience of coral reefs has been studied as a theoretical concept (Nyström and Folke, 2001), empirically in the field (e.g., Wakeford et al., 2008), and with models (Mumby et al., 2006). Resilience has three critical components (1) biodiversity, (2) connectivity and (3) spatial heterogeneity (Nyström et al., 2008). Biodiversity allows redundancy of important ecosystem functions. Connectivity between reefs allows population flux and population renewal after disturbances. Spatial heterogeneity implies that the resilience factors are variable in space across a reef or series of reef. It is recognized that habitat diversity, connectivity and spatial heterogeneity are important resilience modulators, yet these variables remain poorly quantified for most reefs worlwide.
While there is a general consensus on all the factors, from local to global, that can affect coral reef resilience (McClanahan et al., 2012), which factors are the most important for any given reef remains poorly understood (Obura and Grimsdith, 2009). Some modelling studies suggest universal recipes to manage resilience (e.g., the management of herbivores to limit algal overgrowth), but empirical evidences suggest that these recipes cannot cover the range of situation (e.g., Carassou et al., 2013). Models remain invalidated, non-spatial, with arguable parameterization, and the related sensitivity studies can only show the importance of the pre-selected parameters, not those who have been dismissed or neglected. In practice, little is known on what factors contribute to the resilience of coral reef communities and habitats for a particular reef, before it can be studied intensively.
Considering the most likely factors of resilience, remote sensing has been used to map variables that can affect resilience (Table 3). Both stressors and factors of recovery can be mapped and combined in these approaches. A combined index is then used to define areas prone to resilience or not. This is a fairly pragmatic and common-sense approach that could serve well management due to its spatially-explicit approach, yet it remains also very difficult to validate.
Another use of remote sensing to characterize resilience is based on revisiting the history of a reef using time-series of images. This is somewhat in opposition to the modelling-forecasting approach but it is even more interesting, because it can actually show if a reef has been resilient or not after some disturbances, and at which time-scale resilience can be observed. Revisiting the history of reefs can be useful to understand which factors have been at played. However, this approach ideally requires historical field data that are often missing. It is also practically limited to shallow areas. This aspect is the focus of the next section.
Change detection of coral reefs using remote sensing
Habitat mapping is a one-time mapping exercise, but scientists and managers may also want to know how a reef has changed, and if the habitats are stable, degrading, or enhancing in quality. Change detection of habitats using remote sensing has been the subject of several papers, but far less than habitat mapping. Table 4reviews the characteristics of representative studies, published in peer-review journals, and focussing on coral habitats (not on seagrass habitats).
We found methodological papers that look at correction of images, quantify noise, and test various methods of analysis and correction, sometimes using only one pair of images. There are also a number of thematic papers that have tried to understand and explain the causes of the changes that have occurred on a reef, sometimes across several decades and using up to a dozen of images. Various types of images have been used, including aerial photographs (color and black and white) which allow very long time-series, and sometimes with the analysis of multi-sensor series of images. All study sites were shallow, in less than 10 meter deep at the maximum.
The characteristics of change detection analysis, especially for long periods spanning several decades is that the accuracy of the treatment is often difficult to quantify (Scopélitis et al., 2009). Often, no historical data exists to be able to quantify accuracy with a confidence similar to a present-time habitat mapping exercise. Many areas may remain undocumented, hence the level of analysis may be limited to some variables of the habitat that can be photo-interpreted (geomorphology) or related to known unambiguous spectral signatures (cover) while the other variables remain unavailable without historical surveys (architecture, taxonomy).
High thematic richness could be achieved by Scopélitis et al. (2009) even with limited ground-truthing, at least to the point that they could demonstrate, using photo-interpretation techniques that an assemblage of coral communities at Saint-Leu fringing reef in La Reunion has recovered after a hurricane and moderate bleaching event across a 35-year period. In contrast, also using photo-interpretation techniques, Andréfouët et al. (2013) showed for the barrier reef of Toliara in Madagascar that coral communities have steadily decreased due to destructive artisanal fishing, without any signs of recovery. These two Indian Ocean stories highlight two different conclusions in term of resilience: based in the trajectory of their coral habitats, Saint-Leu has been a resilient reef in the face of acute short term disturbance while Toliora appears to be a non-resilient reef in the face of chronic disturbance. Thus, time-series of remote sensing images have the potential to inform on the capacity of a reef to be resilient depending on the type of disturbances the reef had to face during the study period.
General research objective& research questions
The objective of the study is to study for the first time the resilience of an Indonesian coral reef and its habitats, using a multi-sensor time-series of very high spatial resolution (VHR) multispectral satellite images. Bunaken Island, in the Bunaken National park in North Sulawesi, is the study area.
The focus is on thematic interpretation, not image processing method development because the goal is also to provide practical recommendations for Park managers in term of using remote sensing to monitor reefs more effectively, especially the shallow reef flats which have been neglected by monitoring programs.
The INDESO project provides the imagery by purchasing all cloud free images available from the IKONOS, Quickbird, Geoeye and Worldview multispectral archive of images between 2001 and 2015.
The thesis can be divided in 3 series of important questions and steps, inspired by the information presented throughout the previous sections:
1. What are the present day coral reef habitats found in Bunaken National Park and Bunaken Island? How are they distributed? What is the habitat diversity?
2. Can we detect changes in these habitats using a multi-sensor time-series VHR images? Can we reconstruct the recent history of changes around Bunaken Island? Are the habitats resilient?
3. If there are changes on reef habitats after answering Question 2, what are the causes of these changes? If there are no changes after answering Question 2, what resilience factors could be at play?
After answering these questions specific to Bunaken Island, the potential for generalization to other reefs and practical recommendations for mapping and monitoring Indonesian reefs will be discussed.
Thesis structure
This thesis has been divided into six chapters. Three of them are presented in the form of submitted papers to peer-reviewed journal. We refer to a multi-source approach in the title considering first the use of images acquired by different satellite vectors, but also the use of in situ data, and also the use of altimetry data to explain some of the observed changes.
The current Chapter 1 has presented here the research background and key information used to define the subject, with brief presentation on the INDESO project, Indonesia coral reefs, remote sensing of coral reefs and habitat mapping, remote sensing of Indonesia coral reefs, resilience of coral reefs and its mapping, and change detection of coral reefs. Then the general research objectives and the main research questions are given.
Chapter 2 presents in more detail the field and image processing methods used in the following chapters, and justify the choice of these methods. Bunaken National Park and island are also presented.
Chapter 3 presents the results of the field survey, the creation of a detailed habitat map and the analysis of the map in term of habitat richness and distribution. The chapter is a paper entitled Revisiting Bunaken Island (Indonesia): a habitat stand point using very high spatial resolution remotely sensed, which is submitted to the journal Marine Pollution Bulletin, for a special issue on the project INDESO.
Chapter 4 addresses the mortality of corals related to the 2015-2016 El-Niño that we could witness during the study period. This was an opportunistic event that brought new insights on the processes that control the resilience of Bunaken Island coral reef flats. This chapter is a paper entitled Coral mortality induced by the 2015-2016 El Niño in Indonesia: the effect of rapid sea level fall, which is in press with the journal Biogeosciences and also available as a discussion paper open to comments (DOI:10.5194/bg-2016-375).
Chapter 5 presents the change detection analysis of coral reef habitats in Bunaken National Park using an original scenario-based approach. This chapter is a paper entitled Assessment of the resilience of Bunaken Island coral reefs using 15 years of very high spatial resolution satellite images: a kaleidoscope of habitat trajectories, which is under review with the journal Marine Pollution Bulletin, for a special issue on the project INDESO.
Chapter 6 reviews the results, highlight the main findings and put them in the broader context of understanding the resilience of coral reefs, especially in Indonesia, and make suggestions for the future monitoring of these reefs using a combined remote sensing and monitoring approach.
Table of contents :
CHAPTER 1. INTRODUCTION
1.1. The INDESO project
1.2. Coral reefs of Indonesia
1.3. Remote sensing of coral reef habitats and habitat mapping
1.4. Remote sensing of Indonesian coral reefs
1.5. Resilience of coral reefs and mapping resilience
1.6. Change detection of coral reefs using remote sensing
1.7. General research objective& research questions
1.8. Thesis structure
CHAPTER 2. SETTINGS, MATERIAL AND METHODS
2.1. Introduction
2.2. Study site: Bunaken National Park and Bunaken Island
2.3. Field work method for habitat typology
2.4. Training and accuracy assessment points
2.5. Photo-interpretation, digitization, thematic simplification, and final typology of mapped habitats
2.6. Accuracy assessment for habitat mapping
2.7. Multi-sensor image data sets for change detection analysis
CHAPTER 3. REVISITING BUNAKEN NATIONAL PARK (INDONESIA): A HABITAT STAND POINT USING VERY HIGH SPATIAL RESOLUTION REMOTELY SENSED
3.1. Introduction
3.2. Material and Methods
3.2.1. VHR Image
3.2.2. Habitat typology
3.2.3. Habitat mapping for a very high thematic resolution of habitats
3.2.4. Accuracy assessment for a very high thematic resolution of habitats
3.4. Results
3.4.1. Habitat typology
3.4.2. Habitat mapping
3.4.3. Accuracy assessment
3.5. Discussion and conclusion
CHAPTER 4. CORAL MORTALITY INDUCED BY THE 2015-2016 El-Niño IN INDONESIA: THE EFFECT OF RAPID SEA LEVEL FALL
4.1. Introduction
4.2. Material and Methods
4.3. Results
4.3.1. Mortality rates per dominant coral genus
4.3.2. Map of occurrences of mortality
4.3.3. Absolute Dynamic Topography time series
4.3.4. Sea Level Anomaly trends
4.4. Discussion
4.5. Conclusion
CHAPTER 5. ASSESSMENT OF THE RESILIENCE OF BUNAKEN ISLAND CORAL REEFS USING 15 YEARS OF VERY HIGH SPATIAL RESOLUTION SATELLITE IMAGES: A KALEIDOSCOPE OF HABITAT TRAJECTORIES
5.1. Introduction
5.2. Material and methods
5.2.1. Study site
5.2.2. Remote sensing data set
5.2.3. Mining georeferenced historical in situ data
5.2.4. Image and habitat-scenario analyses
5.2.5. Interpretation of changes
5.3. Results
5.3.1. Image time-series quality and correction
5.3.2. Historical data search
5.3.3. Changes for selected polygons
5.4. Discussion
5.4.1. Towards an ecological view of changes using VHR images
5.4.2. A kaleidoscope of habitat trajectories
5.4.3. A kaleidoscope of processes difficult to reconcile
5.4.4. The influence of management and recommendations
5.4.5. Methods for change detection beyond a scenario-based approach
5.4.6. Bunaken, a resilient reef?
5.5. Conclusion
CHAPTER 6. DISCUSSION, PERSPECTIVES AND RECOMMENDATION CONCLUSION (in French)
REFERENCES
