Flood Prediction and Susceptibility Mapping Using Deep Learning and Geosystem Approach

The article isdevoted to solving the problem of flood prediction and susceptibilitymapping. A test site on the territory of Russia, including the floodplain of the Pechora River, waschosen as the study area. The relevance of the approachpresented in the article lies in the application of deep machine learning technologies and the geosystemapproach to the analysis of geospatial data. The task of collectinghistorical data necessary to predict the susceptibility of the territory to floodingduring the spring flood wassolved, whichmakesit possible to predictflooding of territories in order to makedecisionsnecessary to ensure the safety of life and health of thepopulation, as well as territoriesfromflooding. The presentedFloodNET model takes as input data aboutthe territoryduring the low-water period andhistoricalmeteorological observationsandpredicts the susceptibility of the territory to flooding. The model achieves an accuracy of 92%, itsincreaseisinfluenced by the informativeness of the analyzedgeoinformational model and the quality of the model fine-tuning.