Abstract:

The focus of the study was to investigate the capability of linear and non-linear regression techniques for long-term rainfall prediction. Of the linear regression techniques, multiple linear regression method was employed. One of the non-linear regression techniques being widely used in time series prediction is Artificial Neural Networks (ANN) approach which has the ability of mapping between input and output patterns. ANNs are among the numerous empirical models available and have proven to be especially good in modelling time-dependent systems. The study area was restricted to Benin City. Monthly rainfall data, wind speed, evaporation, temperature and relative humidity for the period of 1981 to 2015 spanning to about thirty-four (34) years was collected processed and used for the analysis. Data analysis tools, namely; EViews, SPSS, and MATLAB were employed to conduct the analysis. Results of the descriptive statistics show a marked variation in the mean and standard deviation of the data used. Rainfall, for example, had a mean value of 459.643 and standard deviation of 1.0655E2. The bell-shaped configuration observed in the histogram plot of the variables revealed that the climatic variables used in the study are statistically normally distributed. On the performance of multiple linear regression (MLR) and artificial neural network (ANN), it was observed that artificial neural network performed better than multiple linear regressions. This conclusion was based on the calculated value of the coefficient of determination (R2) for which ANN was 0.9999 and MLR was 0.1755. The performance of ANN compared to MLR was based on the non-linear dependence of rainfall on other climatic variables such as temperature, wind speed, relative humidity, and vapour pressure.

Keywords: