USING ARTIFICIAL NEURAL NETWORKS TO FORECAST RAINFALL OVER GUINEA ECOLOGICAL ZONE, NIGERIA

Abstract

Rainfall forecast is very crucial in view of the fact that Nigeria is not left out in the global rainfall variation and climate change occurrence. This research used the artificial neural networks to forecast rainfall for 2019 over the Guinea Ecological Zone, Nigeria (GEZN) and comparing its results with the Nigerian Meteorological Agency (NiMet) Seasonal rainfall Prediction (SRP) for the same region in the corresponding year. Daily rainfall data which spanned through 1981 to 2015 and obtained from NiMet, Oshodi, Lagos; were used. The NiMet rainfall forecast with its margin of error were obtained from 2019 SRP by NiMet. The data were trained using artificial neural networks (ANNs). The test dataset was used to evaluate the performance of the networks by computing the Root Mean Square Errors (RMSEs). Fuzzy logics were also used for the neural networks training. The neural networks outputs were probability of density of rainfall, regression and margin of error. Margin of errors were also calculated for the data collection points. Results were shown in figures. According to the results from ANNs were able to forecast rainfall over the study area. Onset and cessation of rains are April and October, while August has the highest mean rainfall. Both NiMet and ANN forecasts showed near accurate annual rainfall over Abuja, Ibi, Kaduna, Lokoja and Makurdi while discrepancies were observed over Ilorin, Minna, Lafia and Jos. It was then recommended that rainfall review should be carried out at the end of 2019 to evaluate the performance of the ANNs so as to ascertain its suitability for future use.