CLIMATE VARIABILITY FORECASTING USING BAT ALGORITHM OPTIMISED ARTIFICIAL NEURAL NETWORK

Abstract

This paper presents a summary and results of a study that was conducted in an attempt to forecast climate variability in Zimbabwe using the BAT Algorithm optimised Artificial Neural Network (BAT-ANN) analysis technique. Forecasts of climate ahead of time can potentially allow governments, farmers and other players in private and/or public sectors to make decisions to reduce unwanted impacts or take advantage of expected favourable climate. However, potential benefits of climate forecasts vary considerably because of many physical, biological, economic, social, and political factors. In a developing country, like Zimbabwe where agriculture is the base of the national economy, climate conditions play leading role for progressive and sustainable development, therefore climate variability forecasts are very important. The BAT-ANN was adapted and tested using the Zimbabwean meteorological dataset and results confirm that our proposed model has the potential for reliable climate forecasting for a 25 year period. The mean percentage accuracy was used to evaluate the performance of the trained climate forecasting neural network and proved sufficient. Therefore, in this paper, we present a new technique to climate variability assessment namely; the BAT-ANN. In this study, the approach employed to achieve objectives was; collecting quantitative data, adapting a BAT-ANN for analysis, and developing a Java program that employs the BAT-ANN for forecasting. The objectives of the study were met.