Comparative accuracy assessment of satellite (Sentinel-2, SMAP) and ground-based (TDR, PR2) sensors in soil moisture estimation

Introduction

Although soil moisture represents a small fraction of the world's available freshwater, it plays a crucial role as an effective water storage component within the hydrological cycle and is fundamentally important in hydrological, biological, and biogeochemical processes. The most critical aspect of soil moisture is the depth of water stored within the soil. Consequently, understanding the factors that influence soil moisture and its effects is essential for predicting future performance and ultimately enhancing agricultural production and food security. This understanding is also vital for optimizing irrigation water management and improving water use efficiency in agricultural fields. Therefore, various devices and equipment have been employed over the past few decades to estimate soil moisture. Their application requires consideration of several factors, including the need for calibration, accuracy of results, repeatability, spatial resolution, usability, and cost. Despite the significance of soil moisture in modeling hydrological, biogeochemical, and related dynamic processes, accurately measuring its temporal and spatial variations on a local or watershed scale presents challenges and high costs due to substantial fluctuations. Conversely, remote sensing data has provided opportunities for continuous and cost-effective monitoring of soil moisture estimates with appropriate temporal and spatial distribution, which must be controlled and calibrated with ground-based data. This study investigates and evaluates point-based sensors, including time-domain reflectometry (TDR) and the PR2 neutron probe, alongside sensors that offer suitable temporal and spatial distribution, such as Sentinel 2 and Soil Moisture Active Passive (SMAP), using weighted soil moisture measurement data.

Materials and Methods

In this study, soil moisture measurement as a direct method and TDR, PR2, Sentinel-2, and SMAP satellite images as indirect methods of soil moisture estimation were investigated. To assess and evaluate soil moisture sensors at the watershed scale, a part of the Telo region called Deh Sayid in Lavasanat, Tehran province, with an area of 328 hectares, was selected. Initially, due to the difficulty and high cost of measuring soil moisture by direct methods, data obtained from this method were considered the criterion for evaluating TDR and PR2 sensors. For this purpose, seven stations were established across this region in different land uses, where soil moisture monitoring was conducted through soil sampling and weight measurements in the SCWMRI laboratory, alongside simultaneous soil moisture measurements using TDR and PR2 sensors. Initially, the simultaneous data from these two sensors were evaluated against direct gravimetric observation values. Additionally, soil moisture was monitored by TDR and PR2 sensors during the passage of the Sentinel-2 and SMAP satellites from October 2020 to April 2022. During this period, TDR was used to evaluate PR2, Sentinel-2, and SMAP using statistical criteria, including correlation coefficient and percent deviation. Furthermore, considering the spatial scale of SMAP, the average of the simultaneous data from the seven stations was used for its evaluation.



Results and Discussion

The correlation between TDR and PR2 related to gravimetric method measurements showed that their coefficient of determination (R2) was 0.91 and 0.92, respectively, which showed a high correlation with absolute moisture values. Also, their percentage deviation was 6 and 23 percent, respectively. Therefore, TDR measured absolute soil moisture with much higher accuracy than PR2. The correlation coefficient of PR2 data with TDR varied between 0.826 and 0.933 at the monitoring stations, indicating a high correlation between them. To evaluate the moisture index of Sentinel 2 and SMAP images, simultaneous data from TDR were used due to the high accuracy of the TDR sensor. The results of the correlation analysis of the normalized differential water index (NDWI) obtained from Sentinel 2 images with TDR showed that the range of the coefficient of determination (R2) between the data of these two sensors was very high. The R² values vary from 0.003 at station P7 to 0.814 at station P2. The correlation coefficient at other stations is 0.094, 0.132, 0.587 and 0.723 at stations P3, P4, P6 and P5 respectively. However, at three stations P7, P5 and P6 the correlation coefficient is significant at the 5% level, but at three stations P2, P3 and P4 there is no significant correlation between them. The correlation coefficient of simultaneous SMAP data with the average TDR data at seven stations is 0.455 and the slope of their correlation line is 0.833. Therefore, SMAP shows more accuracy than Sentinel in estimating soil moisture

Conclusion

The result of this research indicated that the correlation between each of the two sensors, TDR and PR2, with gravimetric method is very high and close to each other with values of 0.956 and 0.962, respectively. Additionally, their absolute deviation is 6 and 23 percent, respectively. Consequently, TDR provides a more accurate estimate of the absolute value of soil moisture compared to PR2. Therefore, TDR is superior to PR2 as an appropriate estimator in the absence of the soil moisture data from gravimetric method in this region. Other findings from the study reveal that the indices obtained from Sentinel2 at various points yield different estimates of soil moisture; in pasture land with dense vegetation cover, the accuracy is significant, whereas in areas with tree cover, it lacks significant accuracy. Furthermore, the results demonstrated that the accuracy of Sentinel2 moisture indices is low, while SMAP offers higher accuracy for estimating moisture despite its lower spatial scale. Moreover, based on the findings of this study, the estimated absolute amount of soil moisture derived from remotely sensed data does not yet possess the necessary accuracy for practical applications, and further research is required to combine and test with other indices to enhance the accuracy of soil moisture estimates.