Developing a Radial Basis Function Neural Networks to Predict the Working Days for Tillage Operation in Crop Production

Abstract

The aim of this study was to determine the probability of working days (PWD) for tillage operation using weather data with Multiple Linear Regression (MLR) and Radial Basis Function (RBF) artificial networks. In both models, seven variables were considered as input parameters, namely minimum, average and maximum temperature, relative humidity, rainfall, wind speed, and evaporation on a daily basis. The PWD was considered to be the output of the developed models. Performance criteria were RMSE, MAPE, and R². Results showed that the R²-valuewas 0.78 and 0.99 for MLR and RBF models, respectively. Both models had acceptable performance, but the RBF model was more accurate than the MLR model. The RMSE and MAPE values for the RBF model were lower than those for the MLR model. Thus, the RBF model was selected as the suitable model for predicting PWD. Moreover, the results of these models were compared to the prior soil moisture model. It was indicated that the results of the studied models had a good agreement with the results of the soil moisture model. However, the RBF model had the highest R² (99%). In conclusion, the developed RBF model could be used to predict the probability of working days in terms of agricultural management policies.