Sentinel-1 és Sentinel-2 felvételek belvízveszélyeztetettségi idősoros elemzése konvolúciós neurális hálózatokkal

Abstract

Presently, climate change is expected to cause extreme weather more often, which may result in increasingly frequent and widespread inundations. Shallow floods, also defined as inland excess water, are complex phenomena, which makes it very difficult to initiate preventive action and to identify areas at risk. It is therefore important to develop a "near-continuous" monitoring system based on images from the Sentinel constellation of active and passive satellites, available from the European Space Agency (ESA), covering large areas and with sufficient spatial resolution. We used convolutional neural networks (CNNs -Convolutional Neural Networks) to delineate water surfaces, which were temporally interpolated to create a frequency map over the whole year by filling temporal and spatial gaps in the time series. In the 1600 km 2 study area, we delineated 17,80 km 2 of permanent water surface, and 5,64 km 2 of highly, 3,70 km 2 of moderately and 7,79 km 2 of land slightly vulnerable to inland excess water.