A reliable linear stochastic daily soil temperature forecast model

Forecasting soil temperature profile is recognized as vital information for irrigation demand forecast in a modem/efficient agricultural water management framework in arid regions. A new linear stochastic model is proposed to more accurately forecast daily soil temperature (DST) profile at depths of 5, 10 and 20 cm below ground surface. The data used to test the proposed new method is collected from two stations in Tabriz and Jolfa, located in the East Azerbaijan Province of Iran. The proposed new method uses four preprocessing techniques, including spectral analysis, standardization, trend removing and differencing. A total of 1680 different modelling scenarios were performed in this study. The results show the superior ability of the proposed methodology in DST estimation, compared to existing nonlinear methods such as the multilayer perceptron neural network (MLPNN), with excellent performance indicators such as the coefficient of determination, mean relative error and the Nash-Sutcliffe index. Moreover, the Akaike Information Criterion (AICc) index is employed to compare the performance of the proposed method with MLPNN in terms of both accuracy and easy-of-use. The AICc of the proposed method at Jolfa at a depth of 5, 10 and 20 cm were 176, -2 and -184, respectively, in comparison with 1991, 30 and -57 for MLPNN. Similarly, the AICc index for Tabriz at 5, 10 and 20 cm are 200, 17 and -152, respectively, for the proposed method and 202, 33 and -62 for MLPNN. Consequently, the proposed new linear method is recommended for forecasting daily soil temperature profiles.