Rainfall Forecasting: An Artificial Neural Network Approach

Abstract

An innovative technique is utilized for rainfall forecasting using Artificial Neural Networks based on feed-forward back-propagation architecture. Focus is set upon making successful predictions from the available data, not on incorporating the physical aspects of the atmosphere or the actual process of rainfall occurrence. Both short term and long term forecasting was attempted for ground level data collected by the meteorological station in Colombo, Sri Lanka (Lat: 79.87 E, Long: 6.90 N, Altitude: 7.3 m). Three Neural Network models were developed; a one-day-ahead model for predicting the rainfall occurrence of the next day, which was able to make predictions with a 74.25% accuracy, and two long term forecasting models for monthly and yearly rainfall depth predictions with 58.33% and 76.67% accuracies within a 5% uncertainty level. Each of these models was extended to make predictions several time steps into the future, where accuracies were found to be decreasing with the number of time steps. The success rates and rainfall trends within the monsoon seasons were also studied and presented.