عنوان مقاله [English]
Background and objectives: Soil temperature is an important and effective variable influencing different soil physical, chemical and biological processes. Soil temperature also has direct and indirect impacts on plant activities and has a remarkable effect on the mass and energy transfer rate between soil and atmosphere. With respect to having limited and non-complete soil temperature data, modeling the soil temperature is very important. In this context, SVAT models which are on the basis of the solving the coupled soil heat and water transfer equations have a high important and acceptable accuracy.
Materials and methods: To performing this research, a SVAT model was used to simulating hourly soil temperature in different depths. An old synoptic station (Saghez station) was selected to performing this research and on the basis of the measured data of the soil physical properties in different layers and several meteorological variables in a multi-year time period, the hourly soil temperature values was simulated in different soil depths. The Bayesian Calibration method was used to calibrating the model by considering 10 main model parameters and by performing 8000 model runs on the basis of the different combinations of the parameter values in their uncertainty range using the Monte-Carlo stochastic method in the 1994 to 2009 time period. Then, the calibrated model was validated in the 2010 to 2014 time period and to evaluating the model performance, three indices including mean absolute error (MAE), mean biased error (MBE) and coefficient of determination (R2) were applied.
Results: The bayesian calibration results showed a sensible change in the 10 selected parameter values compared to their default values with a good calibrating process. In the calibration period, the value of all of the performance indices in most cases was improved after calibration process in comparison to the default-value parameters case. The model validation results showed non-remarkable differences between the model performance indices for two calibration and validation cases and therefore, an acceptable performance of the calibrated model can be expected to forcasting the hourly soil temperature values in any desired period so that the coefficient of determination (R2) was obtained more than 0.9 in all of the studied times and soil depths. However, the model performance was not showed a similar behavior during different seasons so that based on the R2 criterion, the lowest model performance was occurred during the winter but on the basis of the MAE criterion, the best and lowest model performance was occurred in the winter and both spring and summer seasons, respectively. The results also showed that the overall tendency of the model simulations is to underestimate the spring and summer and overestimate the autumn and winter soil temperatures and in the annual scale, an overall model underestimation can be recognized. The daily mean soil temperature simulation results showed less error values in comparison to the hourly soil temperature simulations.
Conclusion: The overall results of this research showed that instead of getting appropriate results of the hourly and daily soil temperatures in different depths using the selected SVAT model, by collecting and involving more data such as soil moisture data in the calibration process, it is possible to getting more accurate results from the model. Also, with respect to existence of specific influential processes on the soil temperature during the summer and winter seasons, it is suggested to distinctly performing the model parameters calibration procedure for these two seasons.