Machine Learning Approaches to Extreme Weather Events Forecast in Urban Areas: Challenges and Initial Results
Fábio Porto, Mariza Ferro, Eduardo Ogasawara, Thiago Moeda, Claudio Daniel Tenorio de Barros, Anderson Chaves da Silva, Rocío Zorrilla, Rafael Silva Pereira, Rafaela Castro, João Victor Silva, Rebecca Salles, Augusto Fonseca, Juliana Hermsdorff, Marcelo Magalhães, Vítor J. Sá, Antônio Adolfo Simões, Carlos Cardoso, Eduardo Bezerra
Abstract
Weather forecast services in urban areas face an increasingly hard task of alerting the population on extreme weather events. The hardness of the problem is due to the dynamics of the phenomenon, which challenges numerical weather prediction models and opens an opportunity for Machine Learning (ML) based models that may learn complex mappings between input-output from data. In this paper, we present an ongoing research project which aims at building ML predictive models for extreme precipitation forecast in urban areas, in particular in the Rio de Janeiro City. We present the techniques that we have been developing to improve rainfall prediction and extreme rainfall forecast, along with some initial experimental results. Finally, we discuss some challenges that remain to be tackled in this project.