Mapping Seagrass Percent Cover And Biomass In Nusa Lembongan, Bali, Indonesia

Abstract

Seagrass meadow is one of the blue-carbon ecosystems capable of absorbing and storing carbon more effectively in the bodies and sediments than terrestrial ecosystems. However, nationwide data on its carbon stock remains elusive due to limitations and challenges in data collection and mapping. Seagrass percent cover and biomass, which were closely related with above-ground carbon stock, can be effectively mapped and monitored using remote sensing techniques. Therefore, this study aimed to compare the accuracy of 4 scenarios as well as assess the performance of random forest and stepwise regression methods, for mapping seagrass percent cover and biomass in Nusa Lembongan, Bali, Indonesia. The scenarios were experimented using only atmospherically corrected images, sunglint, water, as well as sunglint and water column corrected images. Furthermore, WorldView-3 images and in-situ seagrass data were used, with the image corrected by applying the scenarios. Random forest and stepwise regression methods were adopted for mapping and modelling. The optimum mapping scenario and method were chosen based on R2, RMSE, and seagrass spatial distribution. The results show that the atmospherically corrected image produced the best seagrass percent cover and biomass map. Range of R2 using random forest and stepwise regression model was 0.49–0.64 and 0.50–0.58, with RMSE ranging from 18.50% to 21.41% and 19.36% to 20.72%, respectively. Based on R2, RMSE, and seagrass spatial distribution, it was concluded that the random forest model produced better mapping results, specifically for areas with high seagrass percent cover.