Monthly rainfall prediction using artificial neural network: A case study of Kano, Nigeria

Abstract

Rainfall continues to be the major source of moisture for agricultural activities over Nigeria, therefore accurate and timely rainfall prediction is essential for food availability and improved water resources management over this region. In this study, Artificial Neural Network (ANN) was applied to predict monthly rainfall over Kano, Nigeria. Three months lagged climate indices for monitoring El Niño–Southern Oscillation (ENSO) namely; Southern Oscillation Index (SOI), Niño1+2, Niño3, Niño3.4 and Niño4 monthly values for 37 years were used as predictors. A Linear Model (LM) was first developed to serve as a yardstick. The ANN was trained using neuralnet package in R statistical software, 25 years data (1981-2005) was used for model training while the remaining 12 years data (2006- 2017) was used for model evaluation. Results indicated that both ANN and LM replicated the actual pattern of monthly rainfall, although with some disparities. ANN has a correlation coefficient value of 0.73 which is higher than 0.70 recorded by LM, a lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were also observed for ANN as compared with LM. Therefore indicating ANN is more preferable and could confidently be used with ENSO indices for subsequent monthly rainfall prediction over Kano, Nigeria. 1.