Applying machine learning in the investigation of the link between the high-velocity streams of charged solar particles and precipitation-induced floods.

Abstract

This study explores a possible link between solar activity and floods caused by precipitation. For this purpose, discrete blocks of data for 89 separate flood events in Europe in the period 2009-2018 were used. Solar activity parameters with a time lag of 0-11days were used as input data of the model, while precipitation data in the 12days preceding the flood were used as output data. The level of randomness of the input and output time series was determined by correlation analysis, while the potential causal relationship was established by applying machine learning classification predictive modeling. A total of 25 distinct machine-learning algorithms and four model ensembles were applied. It was shown that in 81% of cases, the designed model could explain the occurrence or absence of precipitation-induced floods 9days in advance. Differential proton flux in the 0.068-0.115MeV and integral proton flux > 2.5MeV were found to be the most important factors for forecasting precipitation-induced floods. The study confirmed that machine learning is a valuable technique for establishing nonlinear relationships between solar activity parameters and the onset of floods induced by precipitation.