Hybrid satellite monitoring and AI platform for rapid identification of flood conditions

A methodology for constructing a hybrid operational satellite monitoring platform for the early detection of flood zones is presented. The approach combines optical (Sentinel-2), radar (Sentinel-1), and geostationary (SEVIRI) observations, precipitation estimation products (MPE), and ground-based verification. Unified data preprocessing is performed, including georeferencing, reduction to a single spatial resolution, cloud filtering, and windowing of large scenes. A set of index indicators (NDWI, MNDWI, AWEI, SWI, etc.), texture characteristics, and radar features are calculated. The results are integrated by a neural network module, which generates probabilistic flood maps and confidence indicators for each selected polygon. The implementation is performed in the Python environment using standard libraries for working with raster data and training neural networks. At the same time, modularity is emphasized: individual stages (downloading and saving scenes, preprocessing, generating a training set, model training, and postprocessing) can be performed independently and reused for other events and regions. Testing on flood data in Crimea (June 2021) demonstrated improved detection accuracy and robustness compared to standard index-based methods and confirms the practical applicability of the proposed solution for operational monitoring and alerting, provided the labeled set is further expanded and local landscape features are taken into account.