Extended Abstract
Introduction: Although numerous studies have investigated the relationships between soil moisture and various environmental and climatic variables, the spatial relationship between soil moisture and these variables has not yet been fully identified. This is primarily because traditional statistical methods present parameter values as averages across the study area, thereby ignoring spatial variations in the relationships between soil moisture and independent variables. To overcome this limitation, it is necessary to use an appropriate spatial analysis approach. In this context, spatial statistical methods such as Ordinary Least Squares (OLS) and Geographically Weighted Regression (GWR) can be employed to model spatial relationships between different variables. The Halilrud Basin is a key agricultural region in Kerman Province, Iran, with the local economy heavily dependent on crop production. Soil moisture is a critical factor affecting agricultural drought. Therefore, this study aimed to estimate soil moisture in the Halilrud watershed using field observations and laboratory analyses, and to evaluate its spatial relationship with remotely sensed indices—specifically the Normalized Difference Vegetation Index (NDVI) and Land Surface Temperature (LST)—using OLS and GWR models.
Material and Methods: Soil moisture measurements were taken in the plains of the Halilrud watershed using a TDR device at a depth of 30 cm across 72 sampling points in May 2019. To prepare NDVI and LST maps, Landsat 8 (OLI) and MODIS (MOD11A1) satellite images from May 2019 were acquired and preprocessed. The NDVI and LST indices were then extracted. To assess the spatial relationship between soil moisture and each independent variable (NDVI, LST) as well as their combination, both GWR and OLS regression models were applied.
Results and Discussion: The results showed that the GWR model outperformed the OLS model based on evaluation criteria. The GWR model yielded R² values of 0.28 for LST, 0.44 for NDVI, and 0.58 when both variables were combined, indicating improved model performance. Additionally, the GWR model demonstrated higher efficiency across all scenarios due to lower AICc values and higher local and adjusted R² values. While the OLS model indicated a general correlation between soil moisture and the independent variables, the GWR model revealed that this relationship varies spatially. In particular, the northern regions of the watershed exhibited a stronger correlation between soil moisture and the independent variables. This spatial variability illustrates the advantage of the GWR model, which accounts for local variations in the relationships, unlike the OLS model that assumes a uniform relationship across the study area.
Conclusion: The maps generated in this study can be used to identify areas with significant increases or decreases in soil moisture, which is valuable for decision-making, watershed management, and irrigation planning in the agricultural sector. The methodology and objectives applied here can be extended to other watersheds, offering practical and research value. For future studies, it is recommended to include additional independent variables—such as topographic features and other satellite-derived indices—to identify the most influential factors affecting soil moisture.