Spatial Downscaling Of SMAP Passive Soil Moisture Product Based On Value Consistency

Soil moisture(SM)is one of the key parameters in practical applications such as climate changing,sustainable development of agroforestry,water resource management and natural disaster monitoring,so it is of great significance to obtain spatial-temporal continuous SM product at the regional or global scale.Traditional site-based SM monitoring method failed of measuring the spatial distribution of SM at the large scale.With the development of remote sensing technology,microwave remote sensing has become the primary way to monitor the SM of large scale due to its observations’ strong physical connection with SM,ability of penetrating soil and all-weather and all-day operation.At present,most remote sensed SM products are mainly derived from passive microwave remote sensing observations.However,the spatial resolution of these passive microwave SM products is very low(ranging from several kilometers to dozens of kilometers),which greatly limits the application of remote sensed SM products.SM spatial downscaling is one of the effective means to improve the spatial resolution of passive microwave SM products.However,most SM relationship model in current downscaling methods are without strong applicability.At the same time,less attention was paid to the residuals in the SM relationship model,which makes the SM value range of pixel in the downscaled results and the corresponding original coarse-resolution SM value data are not consistent.Moreover,the influence of the cloud in the auxiliary data makes the incompletely spatial continuity in the downscaled results.On this basis,with the aid of the 1-km Moderate Resolution Imaging Spectroradiometer(MODIS)products(land surface temperature and normalized differential vegetation index),passive microwave SM downscaling model with the characteristic of value consistency was constructed in Iberian Peninsula,which downscaled SMAP(Soil Moisture Active Passive)Passive microwave SM(36 km)into the spatial resolution of 1 km.Then,the downscaled results were evaluated by the in-situ SM data and airborne passive microwave SM data.Finally,based on the reconstructed daily LST(land surface temperature)data,a spatial-temporal continuous passive microwave SM downscaling model was constructed and a 1-km spatial-temporal continuous downscaled SM dataset of study area was produced.The main conclusions of this study are as follows:(1)Two land surface parameters,LST and NDVI(normalized differential vegetation index),had high correlation with the SMAP SM,and can construct high-accurate regression fitting with SMAP SM.In the local linear fitting,the annual average and annual standard deviation of the fitting correlation coefficient of LST and NDVI against SMAP SM are 0.88 and 0.02,respectively.Therefore,MODIS(Moderate-Resolution Imaging Spectroradiometer)LST(8 days,1 km)and MODIS NDVI(16 days,1 km)were selected as the auxiliary data in the process of downscaling 36-km SMAP SM.(2)The passive microwave SM downscaling model constructed in this study can produce the downscaled SM of subpixels consistent with the SM value of corresponding SMAP cell.In the network-scale validation with the in-situ SM,the R and RMSE of the downscaled SM are 0.87 and 0.045 m3/m3.In station-scale validation,the mean values of R,RMSE and ub RMSE of all stations are 0.79,0.085 m3/m3 and 0.043 m3/m3.These validation results indicated that the accuracy of the downscaled SM is acceptable.(3)There was deviation between airborne passive SM and SMAP SM,which was mainly affected by different spatial scale,observation configuration,parameter setting of algorithms and selected auxiliary data.Using the airborne SM,we also implemented the uncertainty analysis of the downscaling model and found that the SM relationship model could express the relationship between SMAP SM and the auxiliary data well already.By increasing auxiliary data or changing the mathematical formula in our SM relationship model,the accuracy of the downscaled SM could not increase effectively.(4)We used ATC(annual temperature cycle)model to reconstruct the cloud covered pixels in MODIS LST data.Then,we established a spatial-temporal continuous passive SM downscaling model and produced the daily and 1 km downscaled SM for three years(2016-2018).The ub RMSE of the downscaled SM against the in-situ SM was 0.039 m3/m3,which reached the target accuracy of the SMAP SM product(ub RMSE = 0.04 m3/m3).Validation results also indicated that the downscaled SM has a good accuracy level and a great application potential.After CDF(cumulative distribution function)matching,the time variation of the downscaled SM was closer to that of the in-situ SM,which indicating that the distribution of the downscaled SM on time scale was with reliable accuracy.