Surface water detection in the Caucasus

Abstract

The Caucasus is an important global diversity hotspot and hosts a wide variety of surface water features, including major transboundary wetlands, in addition to large areas with irrigated agriculture and newly developed fishponds. In this study, we aim to establish the best performing methodology to produce surface water maps with a high degree of accuracy in the Caucasus. We evaluate optical data from Landsat 8 in both the dry and wet season for three study areas in the Caucasus. We test the performance of four different optical water indices derived from Landsat data, a method by Zou et al. (2017) also applied to Landsat data, and the European Commission Joint Research Centre (ECJRC) Global Surface Water dataset. We evaluate the performance of each water index using 5744 land cover validation/training points over all three study areas, which we manually classified by evaluating imagery from Google Earth. Using all validation points from all three study areas and both the wet and dry season, we find that the application of a logistic regression model using an optical surface water index (MNDWI) resulted in the most accurate open surface water maps. This approach achieved an overall accuracy of 93.0%, which is better than was found for freely available global surface water products.