Seasonal Rainfall Prediction in Lagos, Nigeria Using Artificial Neural Network

Abstract

Deliberating the importance of rainfall in determining process such as agriculture, flood and water management, these study aim at evaluation of non-linear techniques on seasonal rainfall prediction (SRP). One of the non-linear method widely used is the Artificial Neural Networks (ANN) approach which has the ability of mapping between input and output patterns. The complexity of the atmospheric processes that generate rainfall makes quantitative forecasting of rainfall an extremely, difficult task. The research goal is to train/develop Artificial Neural Network model using backward propagation algorithm to predict seasonal Rainfall. Using some meteorological variables like, sea surface temperature (SST), U-wind at (surface, 700, 850 and 1000), air temperature, specific humidity, ITD and relative humidity. The study adopt  monthly June-October (JJASO) rainfall data and January-May (JFMAM) monthly data of SST, U-wind at (surface, 700, 850 and 1000), air temperature, specific humidity and relative humidity for a period of 31 years (1986-2017) over Ikeja. The proposed ANN model architecture (9-4-1) in training the network using back-propagation algorithm indicated that the statistical performance of the model for predicting 2013 to 2017 (JJASO) rainfall amount indicated as follows; MSE, RMSE, and MAE were 7174, 84.7 and 18.6 respectively with a high statistical coefficient of variation of 94% when the ANN model prediction is validated with the observed rainfall. The result indicated that the propose ANN built network is reliable in prediction of seasonal rainfall amount in Ikeja with a minimal error.