Seagrass Habitat Suitability Models using Multibeam Echosounder Data and Multiple Machine Learning Techniques

Abstract Seagrass beds are important habitats in the marine environment by providing food and shelter to dugongs sea turtles. Protection conservation plans require detail spatial distribution of these such as habitat suitability maps. In this study, machine learning techniques were tested using Multibeam Echo Sounder System (MBES) ground truth datasets produce seagrass models at Redang Marine Park. Five bathymetric predictors seven backscatter from MBES data used representing topography features sediment types study area. Three algorithms; Maximum Entropy (MaxEnt), Random Forests (RF), Support Vector Machine (SVM) tested. The results revealed that MaxEnt RF achieved highest accuracy (93% 91%, respectively) with SVM produced lowest (67%). Depth was identified most significant predictor for all three models. contributions more central model. High showed suitable is distributed around shallow water areas (<20 m) between fringing reef habitats. findings highlight acoustic capable predict how spatially which provide information managing resources.