Modeling of Soil Temperature by Generalized Regression Neural Network in Bathalagoda Area of Sri Lanka
Sri Lanka · 2013
Regression is the data analysis component of statistics. The application of neural networks for regression problems is a popular technique. Even though neural networks have complex structures, and various parameters, they perform well in regression problems. The generalized regression artificial neural network (GRNN) is a well-known and widely applied mathematical model for forecasting of nonlinear functions. This study presents the application of GRNN models to predict the weekly morning and evening soil temperatures at a depth of 5cm on the weekly agroclimatic data collected from Bathalagoda. This paper investigates the potential of using a GRNN for reference soil temperature estimation. Weekly data collected at Bathalagoda Rice Research and Development Institute wasanalyzed in this study. Serial cross correlation and autocorrelation analysis were used to select the most suitable input variables for the network models. Models with different combinations of input variables were tested and the best sets of input variables were selected based on the prediction accuracy. The Mean Square Error (MSE) and the coefficient of determination (R 2 ) of the model of predicted values on actual values were employed to compare the performances of different neural networks. Determination coefficients of GRNN models of morning and evening soil temperatures were found to be 0.71 and 0.68 respectively. The GRNN model yields the best result of minimum mean square errors values of 0.358 and 0.447 for the morning and evening soil temperatures respectively. The results suggest that using GRNN with minimum climatic variables for soil temperature forecasting is a reliable method.