Identification of Severe Weather Event with 3D meteorological radar and MLP
Tulipa Gabriela Guilhermina Juvenal da Silva1; Paulo Henrique Siqueira1; Cesar Beneti2; Maiko Buzzi3; Leonardo Calvetti4
1 Universidade Federal do Paraná; 2 Sistema Meteorológico do Paraná; 3 Universidade Tecnológica Federal do Paraná; 4 Universidade Federal de Pelotas
doi:10.20906/CPS/CILAMCE2017-0368
Resumo
Analyzes and meteorological studies make it possible to forecast weather and severe events such as heavy rain, electrical storms and tornadoes. These analyzes can be obtained by numerical models based on meteorological data. Meteorological radars have the characteristic of enabling the monitoring and prediction of severe weather events. This paper presents an application of a Multilayer Perceptron (MLP) machine learning technique for the prediction of Severe Weather Events (SWE) as detected by a meteorological radar. The data used as input of the model, characteristic vector, consists of data collected from a dual polarization radar in the south of Brazil, in Cascavel, Parana. After the training, a model was obtained that served as a support for the decision on SWE alerts in the state of Paraná. Results indicate 81,40% detection for the SWEs studied and an agreement of 68.58% for cases identified by a lightning detection network. This preliminary study showed that, for 75% of all cases, it is sufficient to evaluate threats up to 2.85 km of altitude. Therefore, it was decided to apply the MLP only to the data closest to 3 km altitude. This new MLP obtained results similar to the results including all altitudes, which implies that it is possible to study the ETSs in 2D, and not requiring the full volume of the radar to be analysed, facilitating its usage in an operational environment.
Palavras-chave: Polarimetric Radar; Nowcasting; Multilayer Perceptron; Machine Learning