Título: Lightning Forecast Using Data Mining Techniques On Hourly Evolution Of The Convective Available Potential Energy
Autores: Sá, J. A. S.; Almeida, A. C.; Rocha, B. R. P.; Mota, M. A. S.; Souza, J. R. S.; Dentel, L. M.
Resumo: This study presents a method developed for lightning forecasting in eastern Amazonia, based on the estimates of the hourly evolution of the convective available potential energy (CAPE). The CAPE is a computed index of the air stability situation over a given area of the Earth. This parameter is determined from vertical profiles of temperature and humidity of the atmosphere, obtained through radiosondes. The CAPE values may also be estimated during the period between soundings, by using the meteorological variables observed continuously at surface weather stations. Two data mining techniques were used for the forecasts: k-Nearest Neighbor and Decision Tree. For the calculation of the CAPE and its estimated hourly evolution, we used radio soundings data made available by a site of the University of Wyoming, in addition to surface temperature data provided by the METAR code, both collected at the Belém- Brazil airport, during 2009. The CAPE index levels, indicative of strong convection in the area were compared to data of actual lightning activity, provided by the STARNET detection system, in a circular area of 100 km radius, centered at that airport. The angular coefficient of the adjusted line equation to the hourly evolution values of the CAPE and the average value of the CAPE were the predicting attributes, while the number of lightning flashes detected by the STARNET was the classification attribute. The results indicated that it is possible to predict the lightning class of occurrences with an accuracy of the 70%, in this research area.
Palavras-chave: Lightning Forecast; K-Nearest Neighbor; Decision Tree; Convective Available Potential Energy (CAPE)
Código DOI: 10.21528/CBIC2011-27.1
Artigo em pdf: st_27.1.pdf
Arquivo BibTex: st_27.1.bib