13:45 〜 15:15
[ACG41-P02] Estimation of benthic fractional cover using nonlinear unmixing of multispectral data for shallow coral reef analysis in Ishigaki Island, Okinawa, Japan
キーワード：Spectral unmixing, Shallow coral reefs, Ishigaki Island, Multispectral images
Spectral unmixing is a technique used on remote sensing images to analyze the composition of ground cover and can be applied in shallow coral reef environments, such as Ishigaki Island in Okinawa, Japan. However, obtaining the reflection of radiation from shallow benthic environments through remote sensing instruments presents challenges (Paringit & Nadaoka, 2011). The linear mixture model for spectral unmixing has been widely used for characterizing mixed pixels in remote sensing images due to its simplicity, but its limitations have led to the need for nonlinear techniques in more complex scenarios (Chen et al., 2013).
This study aims to determine the optimal ratios of benthic cover that explain satellite reflectance in shallow coral reef areas and to create a more effective satellite dataset for monitoring coral reefs. To achieve this, we developed a model for nonlinear unmixing of multispectral data to estimate high accuracy of the benthic coverage in shallow coral reef. The accuracy of the model was validated with in situ observations. Rather than using the typical three or four endmembers, we used a set of six endmembers, including bathymetry, in the mixing process across the entire image.
Moreover, we stress the importance of considering the quality of the dataset or remote sensing images when developing models. While the model development is crucial, the radiometric accuracy of the images is equally important. Therefore, we chose Sentinel-2 satellite imagery with five bands centered at 490, 560, 665, 705, and 865 nm and high spatial resolution (10m) as our data source, as it is deemed reliable enough for accurate mapping of benthic habitats in shallow coastal reef areas (Wicaksono et al., 2021).
Although our preliminary results still require improvement, we anticipate that our final outcome will demonstrate high accuracy in estimating benthic cover in shallow coral reef areas, thus providing a valuable satellite dataset for ongoing benthic health monitoring.
This study aims to determine the optimal ratios of benthic cover that explain satellite reflectance in shallow coral reef areas and to create a more effective satellite dataset for monitoring coral reefs. To achieve this, we developed a model for nonlinear unmixing of multispectral data to estimate high accuracy of the benthic coverage in shallow coral reef. The accuracy of the model was validated with in situ observations. Rather than using the typical three or four endmembers, we used a set of six endmembers, including bathymetry, in the mixing process across the entire image.
Moreover, we stress the importance of considering the quality of the dataset or remote sensing images when developing models. While the model development is crucial, the radiometric accuracy of the images is equally important. Therefore, we chose Sentinel-2 satellite imagery with five bands centered at 490, 560, 665, 705, and 865 nm and high spatial resolution (10m) as our data source, as it is deemed reliable enough for accurate mapping of benthic habitats in shallow coastal reef areas (Wicaksono et al., 2021).
Although our preliminary results still require improvement, we anticipate that our final outcome will demonstrate high accuracy in estimating benthic cover in shallow coral reef areas, thus providing a valuable satellite dataset for ongoing benthic health monitoring.