Large-scale surface water change observed by Sentinel-2 during the 2018 drought in Germany
Marc Wieland, Sandro Martinis
Abstract
Monitoring and understanding the spatio-temporal dynamics of hydrological droughts with seamless geographical coverage over large areas is essential for an assessment of impacts on water resources, industry, transport, and human health. This became particularly relevant during the heat and drought of 2018 in Germany, which affected the country across various sectors and caused significant interruptions to ship traffic on rivers and lakes with negative impacts on tourism, transportation, and supply chains. In this study, we provide a spatially and temporally consistent view on the 2018 hydrological drought in Germany as seen from Sentinel-2 satellite images. We extract waterbodies with national coverage at different timestamps using an automated processing chain, which is based on a convolutional neural network and has originally been developed for near-real time flood monitoring. The method produces water segmentations with consistently high Overall Accuracy (≥0.95) and Kappa Coefficient (≥0.89), despite varying topography, land-use/land-cover and atmospheric conditions. Furthermore, we identify hotspots of change in water extent at a national scale by comparing monthly water maps for 2018 with the respective maps of the previous year 2017. For the change hotspots, we map change gradients and produce water extent time-series with mapping frequencies <5 days along a timeline of 12 months.