Machine-learning regression for coral reef percentage cover mapping

Abstract

Coral reef live percent cover (LPC) mapping has always been a challenging application for remote-sensing. The adoption of machine-learning algorithm in remote-sensing has opened-up the possibility of mapping coral reef at higher accuracy. This paper presents the application of machine-learning regression in the empirical modeling of coral reef LPC mapping. Stepwise regression, Support Vector Machine (SVM) regression, and Random Forest (RF) regression were used model the percentage of live coral cover in optically shallow water of Parang Island, Central Java, Indonesia using field photo-transect data to train the PlanetScope image. PlanetScope multispectral bands were transformed into water column corrected bands, Principle Component bands, and Cooccurrence texture analysis bands to be used as predictors in the regression process. The results indicate that the accuracy of machine learning algorithm to map coral reef LPC is relatively low due to the radiometric quality issue in the PlanetScope image (RMSE = 15.43%). We could not yet fairly justify the performance of machine learning algorithm until we applied the algorithms in other images.