Datensatz

Zusammenfassung

A variety of surface water indices have been developed in the past to detect and monitor water resources on regional, national to continental scal. Less attention has been given to exploiting surface water resources using machine learning algorithms (MLAs). This work compares surface water detection by MLA with surface water indices. Using high-resolution sentinel-1/2 data in the Google Earh Engine cloud computing system we quantify seven widely established surface water indices. Alternatively, we test various MLAs as an integrated approach for enhanced surface water detection complemented by surrogate spatial information like hydro-meteorological and topographic data as additional predictor variables. We establish four MLAs, namely support vector machine (SVM), random forest (RF), gradient tree boost (GTB), and classification and regression tree (CART). All surface water indices and results of MLAs are evaluated for their capability in surface water detection by an assessment of qualitative and quantitative accuracy indicators, including the producer’s accuracy, user’s accuracy, overall accuracy, and kappa coefficients. Most surface water indices generally provide very good results with kappa coefficients of more than 0.90. MLA-based assessment of surface water detection is not superior in this regard, but provide further insight to dominating predictor variables, and therefore contribute to an improved process understanding of why and where surface water resources occur.