Predicting Cloud-to-Ground Lightning with Neural Networks
Abstract
Lightning is a hazard to ground operations, launch operations and recovery at Cape Canaveral. The Air Force is responsible for providing the forecasts of lightning for these operations. Neural networks (a sub-discipline of artificial intelligence which deals with the relationships between sets of data) are being applied in an effort to improve the forecasting of cloud to ground lightning. The excellent data sets from Cape Canaveral are used as inputs and lightning strike data are used as the output (predicted) data to train the networks. Wind data from 32 towers were used to predict lightning strikes in 16 blocks over Cape Canaveral for four time periods; 0-15 min., 30-60 min. and 1-2 hours. The network was trained by backpropagation using the data from one day and was verified on independent data. Comparisons made with the currently used convergence method gave similar results. The neural network results improved with larger training sets and with the addition of other readily available data. Five minute averaged field mill data did not improve the predictions, indicating that short term variations of the field mill data should be used.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 20, 1991
- Accession Number
- ADA239989
Entities
People
- Arnold A. Barnes Jr.
- Donald Frankel
- James S. Draper
Organizations
- Phillips Laboratory