Learning ensembles of deep neural networks for extreme rainfall event detection

Abstract Accurate rainfall estimation is crucial to adequately assess the risk associated with extreme events capable of triggering floods and landslides. Data gathered from Rain Gauges (RGs), sensors devoted measuring intensity rain at individual points, are commonly used feed interpolation methods (e.g., Kriging geostatistical approach) estimate precipitation field over an area interest. However, information provided by RGs could be insufficient model complex phenomena, computationally expensive not in real-time environments. Integrating additional data sources radar geostationary satellites) effective solution for improving quality estimate, but it needs cope Big issues. To overcome all these issues, we propose a Rainfall Estimation Model (REM) based on Ensemble Deep Neural Networks (DeepEns-REM) that can automatically fuse heterogeneous sources. The usage Residual Blocks base models adoption Snapshot procedure build ensemble guarantees fast convergence scalability. Experimental results, conducted real dataset concerning southern region Italy, demonstrate proposal comparison technique other machine learning techniques, especially case exceptional events.