Forecasting precipitation from multi-platform remote sensing systems using wavelet-based neural network models

This paper explores spectral decomposition of environmental data for use in ad hoc artificial neural networks for predicting precipitation patterns by exploiting the nonlinear dynamic signals of oceanic teleconnection patterns found in the Northern Atlantic and Pacific. Using sophisticated ground and satellite remote sensing, including the Advanced Very High Resolution Radiometer (AVHRR) instrument onboard the NOAA satellites for sea surface temperature detection and the GOES geostationary satellite for precipitation correction of in-situ data, high predictive skill is demonstrated during the winter months within the Adirondack state Park in upstate New York, USA. Results show winter months with up to 67% of the land area accurately forecasting precipitation trends with a lead time of 3 months.