a model comparison between predicted soil temperatures using anfis model and regression methods in three different climates

abstract soil temperature is one of the key parameters affecting most hydrologic and agricultural processes. therefore, its measurement and prediction is very crucial. so far, the statistical regression methods have been used for estimation of soil temperature for specific location encountering with lack or shortage of data. in this work, soil temperature data are estimated at six different depths for three typical climates (zahedan, tehran, ramsar) by a new approach namely adaptive neuro-fuzzy inference system (anfis), and the results are compared with those of estimated by regression methods. in addition, the most important meteorological parameters (maximum temperature, minimum temperature, mean daily temperature, relative humidity, sunshine hour, and wind speed) which influence soil temperature at the study sites are used during the 15-years period (1992-2006) of study. the comparison of soil temperature data predicted by anfis and regression methods indicated that the performance of anfis model is 4% more accurate than regression methods. it was found that the accuracy of prediction using anfis model for arid climates of zahedan and tehran was 12% and 4.5% better than ramsar (humid), respectively. the statistical comparison of the estimations derived by anfis model and the observed soil temperature data of drier climates showed that the coefficients of correlation (r) are reduced (up to 10%) for deeper layers. in contrast, for the humid climate of ramsar, the model accuracy for near surface layers (5 and 10 cm) was up to 18% less than deeper layers (100 cm).