Using Agricultural Production to Evaluate the Novel Method of Reducing the Uncertainty in GNSS-R Soil Moisture Retrieval

Soil moisture (SM) is one of the important driving factors for promoting photosynthesis of vegetation to produce organic carbon. Global satellite navigation system reflectometry (GNSS-R) has considerable potential in monitoring regional SM due to its penetrability and high temporal resolution. However, the existing researches have two problems: 1) the GNSS-R SM retrieval methods lack the uncertainty quantification for results; 2) the application potential of the retrieved GNSS-R SM has not been evaluated from the perspective of agricultural production. The core scientific basis for this evaluation is the tight biogeochemical coupling mechanism existing among SM, vegetation (crops grown in plain) physiology and atmospheric exchange (i.e., soil-plant-atmosphere continuum), which forms a quantitatively describable response relationship. This paper first proposes a Natural Gradient Boosting (NGBoost)-based framework to quantify the uncertainty and feed it back to improv the accuracy of retrieved results. Secondly, gross primary productivity (GPP) and evapotranspiration (ET) are used to conduct the potential evaluation of the retrieved GNSS-R SM. Compared with the frameworks based on the existing advanced uncertainty quantification techniques, such as Monte Carlo (MC) dropout-based and Deep Ensembles-based frameworks, the correlation (R) is improved by 4% 5%, the root mean square error (RMSE) is reduce by 17% 25%, and the uncertainty is decreased by 21% 25%. Compared with the official SM released by Cyclone GNSS, the R is increased by 0.2 and the RMSE is decreased by 0.01 cm3/cm3. From the responses of SM to GPP and ET, the retrieved SM has better response (9% 17% enhancement) to vegetation productivity compared with that of other frameworks. Meanwhile, compared with the responses of other remote sensing SM data to vegetation productivity, such as Soil Moisture Active and Passive, Climate Change Initiative, and the official SM released by Cyclone GNSS, R has increased by 20% to 60%. The results shows that the proposed framework can improve the accuracy of GNSS-R SM retrieval by quantifying uncertainty and the retrieved GNSS-R SM can better reflect the agricultural production. In addition, by analyzing the importance of physical parameters in the GNSS-R SM retrieval, we conclude that the signal reflectivity under different roughness conditions should be considered at the regional scale.