Severe Precipitation in Brazil: Data Mining Approach

Abstract

Data mining approach is applied to evaluate extreme rainfall events in the Brazil. Statistical analysis is combined with an artificial intelligence technique to identify the most relevant meteorological variables for a local severe precipitation in the Rio de Janeiro state (Brazil): Rio de Janeiro and Nova Friburgo cities. The p-value statistical technique is employed to select a much smaller subset of climatic variables, preserving the information associated with extreme meteorological events. A decision tree algorithm is used as a model to identify the precipitation severity. The method is tested with the events at Apr/2010 (Rio de Janeiro city) and at Jan/2011 (Nova Friburgo city). In both cases, our results show a good local analysis for extreme precipitation episodes.