Application of artificial intelligence methaods to estimate soil daily temperature in arid and semi-arid climates

Estimation of the soil temperature in arid regions is one of the most important issues in planning the projects of desertification, water resource management, and establishment of vegetation.The present study aimed to compare the accuracy of artificial intelligence methods in order to estimate soil daily temperatures using meteorological data (daily minimum and maximum of temperatures, sunshine, and evaporation), as well as, identifying the most important factors on soil temperature in Zabol and Shiraz synoptic stations. For this purpose, soil daily temperature was estimated at 5, 10, 20, 30, 50 and 100 cm depths by using the three-year period data (2011-2014) and artificial neural network, neuro-fuzzy adaptive genetic programming, and combined neural network-genetic algorithm approaches. Thae results were evaluated using the root mean square error (RMSE) and mean absolute error (MAE) and determination coefficient. Based on the results, there was more dependence between the air temperature and soil temperature at the topsoil so that, the highest and the lowest correlation between actual and simulated data were observed at 5 cm (mean R2=0.92) and 100 cm depths (mean R2=0.56), respectively. The accuracy of the methods used was different from each other in estimating the soil daily temperature. Based on  results, in Zabol and Shiraz stations, combined neural networks - genetic algorithm approach and artificial neural network methods provided the most accurate estimation of soil daily temperature, respectively (the mean RMSE=3.69, 2.86 and mean MAE=3.23, 2.57 respectively). According to the results of the present study, it is suggested that in order to choose appropriate time and depth of seeding in the vegetation reclamation projects in arid regions, through considering of climatic conditions of each region, precise artificial intelligence techniques could be used to estimate the soil daily temperature.