ARTIFICIAL NEURAL NETWORKS APPLIED TO RAINFALL PRECIPITATION SERIES

Abstract

The key feature of artificial neural networks (RNs) is their ability to learn from experience. The main neural learning processes are: supervised learning and unsupervised learning. In supervised learning the output behavior is known a priori and the network is driven to reproduce the desired outputs. In unsupervised learning, the network alone must extract regularities in the input stimuli. Principles of unsupervised neural learning tend to be more biologically plausible than the supervised techniques normally used in control. The RNs stand out as general non-linear models capable of learning complex patterns present in data sets of certain phenomena (for example, time series) and, for this reason, make a great contribution to the study of time series prediction. Multiple-layer RNs (with supervised learning) and radial bases will be applied as adjustment and prediction models of rainfall as a methodology of time series for daily data, averages of five, ten and thirty days of time lags. The multiple layer RNs will also be applied as regression models considering as inputs in the network the variables: atmospheric pressure; maximum, minimum and average temperatures; maximum, minimum and average relative humidity; number of days per month with water deficiency, average wind speed, wind gust and global radiation, and as output from the network the precipitation. They will also be applied as Principal Nonlinear Component Analysis. Precipitation anomalies, droughts or excesses, affect the cultivation and yield of the annual crops, interfering throughout the agricultural calendar and consequently in the agricultural zoning. Brazil is a huge country with four climatic seasons, a vast water network and an enviable agricultural potential. Hence, the importance of understanding precipitation. In this project the methodology of artificial neural networks is proposed as a complement to other already existing and established techniques that assist rural managers in the decision making processes. (AU)