Estimation of evapotranspiration rate in the Sahelian region of Nigeria using Generalized Regression Neural Network and Feed Forward Neural Network

Artificial Neural Network (ANN) was employed by researchers in obtaining accurate estimates of evapotranspiration rate. Generalized Regression Neural Network (GRNN) and Feed Forward, Back Propagation Neural Network (FFBP NN) were used to estimate evapotranspiration rate in Kano State, Northern Nigeria to ascertain its modelling accuracy under less input parameters. A 25-year monthly-time step of climatological data was collected from International Institute of Tropical Agriculture (IITA) station. The data were grouped into 12 different input combination with training and validation sets. Based on performance ranking of input combinations used in different neural networks, the solar radiation (GRNNSr) with a Root Mean Square Error (RMSE) of 1.982 ranked lowest while the temperature and wind speed combination input (GRNNTW) ranked highest with a Root Mean Square Error (RMSE) of 0.7777. Observations indicated similar input combinations ranking with the two–layered Feed Forward Neural Network (FFNN) (with 10 hidden neurons). The input combination of temperature, wind speed and solar radiation had the best performance under the Feed Forward Back Propagation Neural Network (FFBP NN) with RMSE as low as 0.6333. This is contrast to the input combination of solar radiation and humidity, which had the lowest performance under the FFBP NN with RMSE of 1.3512. The input combination of temperature and wind speed is the most preferred input combination, having less data input and higher performance. Observations indicated that wind speed provides the best estimate of evapotranspiration in the region than all other lone inputs. Overall, FFBN showed the highest potential in estimating evapotranspiration in the Sahelian region of Nigeria under limited climatological input parameters.