Comparative Analysis of Machine Learning Algorithms in Automatic Identification and Extraction of Water Boundaries

Monitoring open water bodies accurately is important for assessing the role of ecosystem services in context human survival and climate change. There are many methods available body extraction based on remote sensing images, such as normalized difference index (NDWI), modified NDWI (MNDWI), machine learning algorithms. Based Landsat-8 this study focuses effects six algorithms three threshold used to extract bodies, evaluates transfer performance models applied images different periods, compares differences among these models. The results follows. (1) Various require numbers samples reach their optimal consequence. logistic regression algorithm requires a minimum 110 samples. As number increases, order model support vector machine, neural network, random forest, decision tree, XGBoost. (2) accuracy evaluation each test set cannot represent local area performance. (3) When directly AUC indicators regions all show significant decline, with decrease range 0.33–66.52%, performances areas obvious. Generally, tree has good under curve (AUC) indexes 0.790, 0.518, 0.697 areas, respectively, average value 0.668. Otsu methods, 0.970, 0.617, 0.908 respectively an 0.832.