WATER RESERVOIRS MONITORING THROUGH GOOGLE EARTH ENGINE: APPLICATION TO SENTINEL AND LANDSAT IMAGERY

Abstract

Water reservoirs are subjected to increasing hydrological stresses, therefore continuous and accurate monitoring of these resources is essential to ensure their sustainable management. This work proposes a methodology to remotely monitor the surface extent of water reservoirs through the analysis of satellite multispectral and Synthetic Aperture Radar (SAR) images. In particular, a segmentation strategy was implemented within Google Earth Engine (GEE) to distinguish water bodies from the surrounding land surface and measure their extension, by applying three different approaches to Sentinel-1, Sentinel-2, and Landsat-8 imagery. The first approach is based on the use of the Automatic Water Extraction Index (AWEI) and the self-adaptive Otsu’s thresholding method, the second approach is based on the image conversion from RGB (Red-Green-Blue) to HSV (Hue, Saturation, Value) and the use of a parametric threshold, the third approach is based on the use of SAR imagery and an empirically selected threshold. A “static” validation strategy was developed from scratch and standard segmentation metrics were computed to evaluate the accuracy of the three approaches. The average values of the F1 scores on the Sentinel imagery were equal to 0.95, 0.90, and 0.84 for the three approaches, respectively. The same metric on the Landsat imagery was 0.95 for the first approach and 0.93 for the second approach. The best approach, i.e. the AWEI-based method, was then applied to three water bodies in which the effects of the 2022 drought were particularly significant: Sawa lake (Iraq), Poyang lake (China), and Po river (Italy). The results visually highlighted the good performance of the approach in segmenting the water bodies from the surrounding areas.