Satellite-derived shallow water depths estimation using remote sensing and artificial intelligence models, a case study: Darbandikhan Lake Upper, Kurdistan Region, Iraq

Bathymetric mapping provides valuable information for the estimation of the depth and volume of enclosed inland water bodies that are useful in the planning and management of water resources. The use of conventional methods for the detection of shallow water depth, specifically in flooded areas, has been challenging. However, advances in remote sensing technology combined with artificial intelligence (AI) offer a reliable method. This study presents a reliable method to estimate water depth, using the Darbandikhan Lake Upper (DLU) as a test site. The novelty of this work lies in using a combination of Quantile Regression Forests (QRF), Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Networks (ANN) approaches together with the reflectance of Sentinel-2 and the ICESat-2 LiDAR data to estimate the depth of the water in the DLU during the 2019 spring flood. Our results gave the coefficient of determination (R2) and root mean square error (RMSE) between the actual depth obtained from the ICESat-2 and the estimated depth from the applied artificial intelligence models of 0.984, 0.983, 0.868, and 0.809; and 0.545, 0.569, 1.618, and 2.143 for the QRF, RF, SVM, and ANN models, respectively. This study, which applied the QRF model for the first time to determine the satellite-derived water depths, produced the most accurate result, with the maximum and mean estimated depth of DLU being 19.93 and 6.29 m, respectively. This study shows that the most sensitive bands to estimate the bathymetry are Band 9 (940 nm), Band 3 (560 nm), and Band 5 (705 nm) of the Sentinel-2, while the less sensitive bands are Band 2 (490 nm) and Band 11 (1610 nm). We argue that this technique can be applied to estimate the depth of shallow water bodies using passive satellite imageries in other regions of the world regardless of the full coverage availability of ICESat-2. •Quantile Regression Forests has the best performance in derived bathymetry.•The SVM has the best performance for when the number of training data is small.•The band 9, 3, and 5 of the Sentinel-2 are importance to derived bathymetry.