SMAP passive microwave soil moisture spatial downscaling based on optical remote sensing data: A case study in Shandian river basin

Soil Moisture (SM) is not only an important variable in land surface processes, but also a key parameter in global water cycle. In this paper, the objectives are: (1) downscaling SMAP (Soil Moisture Active Passive) SM (SMAP SM) from spatial resolution of 9 km to 1 km, with the using of the auxiliary data from MODIS (Moderate-Resolution Imaging Spectroradiometer) products (land surface temperature and normalized difference vegetation index) by a downscaling method based on self-adaptive window in Shandian river basin; (2) validating the downscaled SM with the in-situ SM and the airborne passive microwave SM (airborne SM); and (3) analyzing the uncertainty caused by auxiliary data and SM estimated model in the downscaling process. The downscaling method used in this paper involves two steps. The SM model was established by using geographically weighted regression model between SMAP SM and the auxiliary data to calculate the 1-km estimated model SM (SMR). Then the 9-km residual (RC) generated by the SM estimated model is downscaled to 1-km spatial resolution (RF) by area-to-point kriging. Finally, the downscaled SM (SMF) is the sum of SMR and RF. It's worth noting that to derive the robust downscaled SM, self-adaptive windows are adopted in these two steps. Visual assessment shows that the downscaling method can not only improve the spatial resolution of SMAP SM, but also retain the consistency between the spatial distributions of the downscaled SM and of the original SMAP SM. The validation results of the airborne SM, the SMAP SM and the downscaled SM against the in-situ SM are not satisfactory. On Sep 24, the correlation coefficient (R) between the three SM data and the in-situ SM are less than 0.5, and on Sep 26, the root mean squared errors (RMSE) are greater than 0.08 m3/m3. By analyzing these data, we found that the limited amount of valid data used in validation was one of the reasons for the poor validation. In addition, the different spatial representativeness and the inconsistent spatial matching of point-scale data and pixel-scale data are also the factors caused the uncertainty in the validation results. Compared with the in-situ SM, the SMAP SM and the downscaled SM have better correlations with the airborne SM. The RMSEs between the downscaled SM and the airborne SM are about 0.04 m3/m3, while the RMSEs between the SMAP SM and the airborne SM are less than 0.04 m3/m3. The correlation between the SMAP SM and auxiliary data (the absolute values of Rs are greater than 0.6) is higher than that between the airborne SM and the auxiliary data (the absolute values of Rs are less than 0.53). It can be seen that there are some differences between the SMAP SM and the airborne SM, which is mainly affected by different spatial scales, observation configurations, SM derived algorithms and auxiliary data using in algorithms of these two SM data. However, more studies are needed on the mechanism of the relationship between auxiliary data and SM in the downscaling process. By adding auxiliary data (land surface albedo) or changing the SM estimation model, the validated results of the downscaled SM against the airborne SM did not improve obviously. This is mainly because more auxiliary data and higher polynomials caused overfitting in the downscaling process, which will be still the focus of future research. © 2021, Science Press. All right reserved.