Neural Network for Lightning Forecasting in Southeastern Brazil


  • Gisele dos Santos Zepka Instituto Nacional de Pesquisas Espaciais
  • Sebastião Cícero Pinheiro Gomes Fundação Universidade Federal do Rio Grande



Lightning, forecasting, MM5 model, neural network.


This paper introduces a lightning forecasting system using Neural Networks (NN) based on correlations betweencloud-to-ground (CG) lightning flash data and meteorological variables obtained from MM5 (Fifth-GenerationPenn State/NCAR Mesoscale Model) simulations in southeastern Brazil. Besides the area of nineteen high densitylightning subregions selected from non-severe thunderstorms cases, the following model parameters were also usedas inputs for the NN: divergence at 850 hPa and 200 hPa, temperature advection at 500 hPa, vertical velocity at500 hPa, and water vapor-mixing ratio at the surface and 850 hPa. The target output of the neural system was thenumber of CG lightning flashes that would occur one hour in advance. Four cases, not previously trained, of highdensity lightning subregions were used to test the system efficiency. Although the testing cases were sufficientlydifferent, a case with no occurrence of lightning has also been included in order to not introduce any bias on theanalysis. The results of the lightning forecasting were promising, indicating that the NN technique associated withmodel variables from numerical simulations can be an efficient tool for predicting flash occurrence appropriately.



Download data is not yet available.



How to Cite

Zepka, G. dos S., & Gomes, S. C. P. (2014). Neural Network for Lightning Forecasting in Southeastern Brazil. Ciência E Natura, 36(3), 538–547.




Most read articles by the same author(s)