امکان‌سنجی برآورد دمای اعماق مختلف خاک و تبخیر از تشت با به‌کارگیری تصاویر ماهواره‌ای

Background and Objective: Soil surface temperature is one of the key variables in the surface energy balance, which is influenced by net radiation, sensible heat flux, and latent heat flux, and in turn, it affects the flux of heat entering the soil. Land surface temperature (LST) is one of the most important products that can be extracted by sensors operating within the thermal infrared spectral range. Among these sensors, MODIS is one of the most significant, installed on both the Aqua and Terra satellites, capable of providing land surface temperature measurements at four different times throughout the day and night. Given the close relationship between land surface temperature and hydroclimatic variables, in this study, LST data derived from MODIS sensors were utilized to estimate soil temperatures at various depths and evaporation from pans.
Materials and Methods: For conducting this study, six synoptic meteorological stations located in Kurdistan Province, Iran, were initially selected. For these six stations, in addition to soil temperature data measured at different depths and pan evaporation observations, four land surface temperature variables derived from MODIS sensors were extracted. These included LSTTerra-Night, LSTTerra-Day, LSTAqua-Night, and LSTAqua-Day. By averaging the daytime and nighttime LST values from the Aqua and Terra satellites, two additional mean daily LST variables were calculated, namely LSTTerra-Mean-Night&Day and LSTAqua-Mean-Night&Day. Initially, daily time series of all the above variables were compiled for the statistical period from 2002 to 2021. Then, by employing a stepwise multiple linear regression model, the six MODIS-derived LST variables were used as predictor variables to estimate soil temperatures at various depths and evaporation from pans. This overall process was carried out at two scales: the station scale, where each of the six stations was modeled individually, and the regional scale, where all six stations were considered collectively to develop integrated models. In order to improve the accuracy of the models at deeper soil layers, daily lagged soil temperatures at depths of 50 and 100 centimeters were incorporated as additional independent variables. This approach accounted for the delayed response of deeper soil layers to surface thermal fluctuations. Model validation at the station scale was performed using 75 percent of the total data from each station (for the statistical period 2002–2016) for calibration, and the remaining 25 percent of the data (for the statistical period 2017–2021) for independent validation. At the regional scale, leave-one-out cross-validation (LOOCV) was applied in six separate iterations, with one station excluded in each iteration, to assess the robustness of the regression models under spatial variability. Model performance was evaluated using the coefficient of determination (R²) and mean absolute error (MAE). Additionally, the mean bias error (MBE) was used to assess systematic overestimation or underestimation by the regression models.
Results: The results obtained from the validation of the regression models at both the station and regional scales indicated that among the six MODIS-derived land surface temperature variables, the multiple linear regression models developed in most cases used four surface temperature variables to simulate soil temperatures at various depths, and three surface temperature variables were sufficient to predict pan evaporation. The two mean daily LST values derived from the Aqua and Terra satellites played a prominent and significant role in all resulting models, both for soil temperatures at different depths and for pan evaporation. Validation results for regression models simulating soil temperatures at various depths showed that, at both station and regional scales, based on both the coefficient of determination and the mean absolute error, the regression models performed very well in simulating soil temperatures in shallower layers (5 to 30 centimeters from the soil surface), with R² values close to 0.95. At a depth of 50 centimeters, a slight decrease in model performance was observed, with R² values close to 0.90. At a depth of 100 centimeters from the soil surface, the reduction in model performance was more pronounced, with R² values close to 0.75. This decline in model performance was attributed to the fact that soil temperatures at deeper layers are strongly dependent on the amount of net energy received at the soil surface. As heat penetrates from the soil surface to deeper layers, the energy and heat wave gradually dissipate, resulting in a weaker dependence of deep soil temperatures on surface temperature. The results also demonstrated that incorporating lagged daily surface temperatures into the regression models reduced the MAE and improved model performance at deeper soil layers. Regarding pan evaporation, the validation results indicated that the regression models performed less accurately in simulating pan evaporation compared to soil temperature predictions. Furthermore, results showed that regression models, for both soil temperature simulation and pan evaporation simulation, generally performed slightly better at the station scale than at the regional scale, highlighting the importance of localized calibration for improved model accuracy.
Conclusions: Overall, the results of this study showed that the six MODIS-derived land surface temperature variables obtained from the Aqua and Terra satellites possess very high potential for simulating soil temperatures at multiple depths, particularly in the shallower layers. These surface temperatures can be directly used as primary predictor variables to simulate soil temperatures at both the station and regional scales. Despite the satisfactory accuracy achieved by the regression models in estimating soil temperatures at various depths, it is recommended that, in order to enhance model accuracy in future studies, other satellite-derived products that influence soil temperature, particularly surface soil moisture, should also be used in combination with land surface temperatures for a more comprehensive simulation of soil temperature at different depths. Regarding pan evaporation simulation, the results of the regression models indicated that direct high-accuracy estimation of pan evaporation using surface temperature alone is not feasible. However, considering the significant role and degree of influence that land surface temperatures exhibited on pan evaporation, there is potential to use these surface temperatures as useful auxiliary variables in combination with other factors that influence pan evaporation. In summary, this study demonstrated the strong capability of MODIS-derived LST products for soil temperature modeling at multiple depths and highlighted their potential as reliable predictive variables for both station- and regional-scale environmental modeling applications.