APPLICATION OF ARTIFICIAL NEURAL NETWORKS (ANNs) FOR FORECASTING RAINFALL IN ILORIN, KWARA STATE, NIGERIA

Abstract

The study is aimed at applying Artificial Neural Networks for forecasting rainfall in Ilorin, Kwara state with a view that artificial neural networks (ANNs) is an emerging computationally powerful technique with very high degree accuracy and widely used as forecasting models in many areas such as engineering, social, finance, economic, stock and foreign exchange problems. The objectives of this study are to examine the rainfall trend and distribution in the study area (1999-2018), to develop a reliable rainfall forecast for the period under study using ANNs and to examine the reliability of the developed rainfall forecast over the study area. In this research, we attempt to study the application of artificial neural networks for forecasting rainfall using some dependent weather variables such as temperature, rainfall, wind speed and sunshine hour in Ilorin metropolis Kwara state from June to October, all for the period of 19 years (1999-2018). The first part of the methodology to carry out this research was the collection of rainfall data (from the Nigerian Meteorological Agency) which serves as the fundamental input for statistical computations. The second aspect was the data processing then followed by the presentation of relevant outputs. From the monthly rainfall data, computation of mean rainfall and percentage mean rainfall for the period under study was carried out. Decadal charts were plotted to ascertain the maximum mean rainfall for each decade and the degree of variation in the amount of fluctuation in rainfall recorded over the period. Developing an artificial neural network (ANNs) as a reliable rainfall forecast essentially involve a nonlinear modeling approach that provides a fairly accurate universal approximation to any function. This is done both visually (using plotted graphs) and statistical measurements such as root-mean-square error (RMSE), mean square error (MSE), Mean Absolute Percent Error (MAPE) and the coefficient of correlation (CORR) to test the degree of error and examine the model performance. The results indicate that the trend and pattern of rainfall movement with respect to its amount and time is such that the rainfall amount either ascends gradually or fluctuates. It was discovered that much of the amount of rainfall in all the years under study is received in the month of June, July, August and September which are largely variant and characterized with fluctuations. Generally, a decreasing trend in rainfall is observed within the first three years of the first decade, with the highest amount of rainfall experienced in the first year of the decade totaled up to 1539.3mm. Also a significant increase in the rainfall amount for the last year of the first decade was observed resulting in an upward trend with values close to what was experienced at the beginning of the decade. The trend in rainfall for the second decade is a little similar to the previous decade with respect to the first, second, eight, ninth and tenth year. However, the eighth year of this decade is most significant as it recorded the highest value of 2552.6mm compared to the previous years in the decade. The highest mean annual rainfall experienced in the first decade was 128.3mm and 212.7mm in the second decade which also correspond to the highest value within the period under study. It was recommended that meteorological stations should be established to cushion the effect and challenge of sparse meteorological data and further reduce the representativeness of a system which can also have significant effect on the results of subsequent analysis. Government should support and encourage private organizations to key into establishment of more automatic weather stations.