MODELLING THE IMPACTS OF CLIMATIC VARIABILITY ON UPLAND RICE YIELD USING ARTIFICIAL NEURAL NETWORKS (ANN) IN SOUTH WEST, NIGERIA

Abstract

Declining rice productivity in recent times have been responsible for rising food insecurity and most extreme forms of famine. In the light of this, some climatic variables were modelled as it affects rice yield in Ibadan, Nigeria for the purpose of increasing rice yield production. The variables rainfall, temperature, relative humidity, solar radiation, wind speed were obtained from the International Institute for Tropical Agriculture (IITA) Ibadan, Nigeria; rice yield data for a 30 year period (1980-2010) while prediction was carried out using Multiple Linear Regression (MLR) and Artificial Neural Network (ANN). Trends and forecast analyses were the tools used. From the study, minimum and maximum rainfall were 794.2 mm and 1926.3 mm recorded in the years 1998 and 2010 respectively while 8% fluctuation was observed for the forecast analysis for the next 20 years (2010-2030). Minimum and maximum temperature 26.22 and 27.70 were recorded in the years 1986 and 1998 respectively and the trend was insignificant (P < 0.05) while forecast prediction for the next 20 years (2010 – 2030) shows root mean square error (RSME) of 0.19. Minimum and maximum relative humidity were 64.29% and 77.83% recorded in the year 2009 and 1999 respectively while the forecast prediction for the next 20 years (2010 - 2030) with root mean square error (RMSE) of 4.06 was obtained. Minimum and maximum solar radiation were 12.07 and 24.68 recorded in the years 1980 and 1984 respectively while the forecast prediction for the next 20 years (2010 -2030) shows the root mean square error (RMSE) as 1.01. Minimum and maximum rice yield were 1.40 t/ha and 2.40 t/ha recorded in the years 1994 and 1987 respectively. The data was subjected to multiple linear regression (MLR) analysis and artificial neural network (ANN) and MLR further confirmed that the four weather parameters have insignificant effect on rice yield. The model obtained gives a low coefficient of determination (R 2 ) of 0.206. The rice yield data was also subjected to ANN model using different types of approach architecture of both LOGSIG and TANSIG function by adjusting the parameters such as numbers of neurons and number of hidden nodes to enhance the accuracy of rice yield predictions. Comparing the predicted rice yield with both regression and ANN models, it was observed that ANN models consistently produced more accurate yield predictions than regression models based on coefficient of determination (R 2 ), root mean square error (RMSE), Standard deviation (SD) and Acute error (AE). ANN rice yield models resulted in R 2 (0.8215), RMSE (0.1122), SD (0.2552) and AE (0.3782) against Multiple linear regression (MLR) R 2 (0.0062), RMSE (0.2772), SD (0.0888) and AE (2.3052) respectively. The results obtained from the modelling will assist farmers to plan rice growth and development to achieve optimum yield and ensure food security especially within the study area.