Coral reefs are a globally important resource with well recognised beauty, biological diversity and economic significance. Also well recognised are the host of natural and anthropogenic impacts that threaten future of reef ecosystems. To understand the effects of impacts on the health and status of reefs globally, methods to rapidly and effectively map and monitor changes are required. The cover of live coral and its change over time has been identified as an ecologically important indicator for coral reefs, though its effective mapping with remote sensing remains a challenge.
One of the challenges in mapping live coral cover is due to the complex and heterogeneous nature of a coral reef environment. Even with the increasing spatial resolution of contemporary imaging devices, it is common for the features of interest (eg. live coral) to be smaller than the spatial dimensions of the imaging sensor. This means that more than one substrate type will contribute to the signal recorded in individual digital image pixels. As such, assigning a substrate class to a pixel that is dominated by live coral cover (for example) reveals little information about the true amount of coral cover within that pixel. Recognising that spectral indices have been used in various terrestrial and oceanic environments to address similar challenges, this thesis aimed to develop, test and assess the effectiveness of a spectral index for mapping live coral cover.
Initial determination of the wavelengths and band combinations that were sensitive to variations in live coral cover was based on extensive modelling using in situ field spectrometer data. Spectral transformations in the form of simple band ratios and first and second order derivatives were also analysed. Six different scenarios were then developed with Hydrolight 4.1 radiative transfer model for a range of water depths (1, 5, and 10m) and water quality parameters (chlorophyll = 0.33 and 0.91 mg/m3), representative of GBR waters. Similar modelling was then undertaken using a hyperspectral (19 bands) high resolution (1m) CASI-2 image mosaic of Heron Reef, southern Great Barrier Reef (GBR), Australia.
Following the spectral modelling, the second derivative around 564nm was selected for its sensitivity to live coral cover variations within a window relatively insensitive to water column and atmospheric attenuation. After comprehensive pre-processing (radiometric, atmospheric, air-sea interface and geometric corrections), the transform was then applied to the CASI-2 image mosaic. Extensive field survey of substrate cover across the reef was used to select calibration and validation sites, and to calculate a regression equation to transform the index from relative to absolute live coral cover (r2 = 0.63).
The final stage of the project was to evaluate the results of the live coral cover spectral index map with other commonly used image mapping techniques. While the results from the spectral angle mapper supervised classification demonstrated the highest correlation with field survey data, the continuous mapping methods mixture-tuned matched filter and the live coral cover index produced more realistic maps that corresponded to extensive though qualitative field experience on Heron Reef. The coral cover index was the fastest mapping technique to perform, requiring significantly less computer hard drive space or processing power than the other methods tested.
Based on the results obtained here using both simulated spectral mixtures and image data, the coral cover index is presented as an effective method for producing a map of live coral cover using high spatial and spectral resolution data. With the addition of field survey data, the index output can be converted from a relative measure to an even more useful absolute estimate of live coral cover. While the index was predominantly developed and tested using data acquired from Heron Reef, previous studies have demonstrated that reef features have similar optical properties regardless of their location. Therefore, the method presented here should be applicable in other reef systems, limited only by water column properties and sensor dimensions.
The work presented in this thesis represents a significant contribution to the field of coral reef remote sensing. This is demonstrated not only through the results achieved, but through new and innovative approaches in the techniques used for achieving the aims and objectives. In addition, this study highlights several critical areas and key research problems to be addressed in future studies.