ارزیابی محصولات ماهواره‌ای رطوبت خاک مبنی بر مشاهدات میدانی در شبکه آبیاری و زهکشی درودزن

Abstract

Background and aim: Soil moisture is one of the main components of the terrestrial water and energy cycle and a key factor in the management of agricultural water resources, especially in semi-arid regions. Although satellite soil moisture products enable broad and continuous monitoring of this variable, their accuracy at local scales requires validation based on in-situ measurements. This study aims to select an appropriate dataset for irrigation monitoring in the Doroodzan irrigation network by evaluating the performance of four satellite soil moisture products—SMAP 9 km, SMAP 1 km, the ASCAT SWI index, and the downscaled SMAP1km RS product—in comparison with long-term measurements at the Zarghan station.

Materials and methods: For the evaluation, soil moisture data from the four satellite datasets were first extracted for the pixel corresponding to the location of the Zarghan station. This station, due to the absence of irrigation and its record of natural soil moisture variations, serves as a suitable reference for local validation. The data were analyzed at four temporal scales: daily, 10-day, 15-day, and 30-day intervals. Temporal aggregation was performed using non-overlapping averaging windows to reduce daily noise. The performance of the products was assessed using three statistical indices: correlation coefficient, root mean square error, and Nash–Sutcliffe efficiency. The SMAP1km RS product was generated through a downscaling procedure based on a random forest algorithm using remote sensing predictors including NDVI, land surface temperature, LAI, Sentinel-1 backscatter, and slope.

Results and discussion: The time-series results showed that at the daily scale, all four products captured the general wetting and drying trends of the soil; however, short-term discrepancies, especially during rainfall periods, were evident. The SMAP 1 km product exhibited the best agreement with ground measurements at this scale, while ASCAT showed the highest level of overestimation. With increasing temporal scale, the accuracy of all products improved significantly. At the 10- and 15-day scales, temporal averaging enhanced correlation and reduced error, with SMAP 1 km providing the highest accuracy. At the 30-day scale, the performance of the SMAP products reached their highest levels, and the 1-km version provided the most accurate representation of seasonal variability, whereas ASCAT continued to exhibit overestimation. The SMAP1km RS product also showed satisfactory performance but remained weaker than the official SMAP version. The findings indicated that the performance of satellite soil moisture products is strongly dependent on temporal scale, and temporal aggregation plays a crucial role in reducing noise and increasing the reliability of satellite data. The superior performance of the SMAP 1 km product stems from the efficiency of its downscaling method and the use of high-resolution auxiliary data. The acceptable performance of the SMAP1km RS product also demonstrates the potential of machine learning approaches for creating locally downscaled products, although limitations such as training data density and spatial heterogeneity may affect its accuracy. The persistent weakness of ASCAT across all scales reflects its sensitivity for local applications.

Conclusion: This study showed that the accuracy of satellite soil moisture products increases significantly with temporal scale, and the 15- and 30-day scales are the most suitable intervals for operational monitoring of soil moisture in the study area. It was also found that among the evaluated datasets, the SMAP 1 km product is the most reliable option for accurate soil moisture monitoring and for improving irrigation water-use estimation in the Doroodzan network. The SMAP1km RS product, despite its lower performance compared to the original product, confirmed the effectiveness and reliability of local downscaling frameworks and demonstrated that these methods can enhance data quality. The results of this study can be used to select appropriate datasets for optimized agricultural water-resource management, improve irrigation planning and scheduling, increase irrigation system efficiency, and enhance the accuracy of soil moisture monitoring in semi-arid regions and other areas with similar conditions.