Research On Methods For Enhancements And Assessments Of Microwave Remotely Sensed Soil Moisture Products

As one of the core parameters in land-atmosphere interaction,soil moisture is an important component of the global water cycle,energy cycle and biochemical cycle of the earth.Microwave remote sensing has been regarded as the best technology for largescale soil moisture mapping,benefiting from its strong penetration,especially in densely vegetated areas.Based on these successfully launched microwave sensors and corresponding remote sensing inversion algorithms,a series of soil moisture products have been developed in recent years,including AMSR2 at C/X-band,SMOS-IC,SMOS-L3,SMAP-SCA-V,SMAP-DCA,SMAP-MTDCA and SMAP-IB at L-band,as well as the active-passive combined ESA CCI products,etc.These products have been derived from different satellite platforms and various inversion algorithms,and thus their applicability in the complex and variable surface environment is somewhat different.In this context,three questions were drawn in this paper:(1)How about the comprehensive performance of these remotely sensed soil moisture products?(2)Where are the uncertainties of the products and the error sources related to the retrieval algorithms?(3)How to improve existing soil moisture products?To answer the above three questions,this paper carried out the accuracy evaluation,error source analysis and improvement of the remotely sensed soil moisture products.Firstly,this paper proposed a framework of accuracy assessment on soil moisture products,in which their performance at global scale and various error sources involved in the inversion algorithms were fully considered.In contrast to previous studies,this paper focused more on the error source analysis on the inversion algorithms,i.e.,not only the performance of these products is evaluated,but also perturbation factors including surface roughness and vegetation optical depth(VOD)etc are considered.In addition,the effects of surface heterogeneity and climate type on the satellite-based soil moisture products were also comprehensively evaluated.The results indicate that SMAP outperforms other soil moisture products in capturing temporal trends in soil moisture,while ESA CCI can achieve higher absolute accuracy,and the complementary concerning error metrics of the two products provides an experimental basis for the further possible integration.Although the underestimation of the newly released SMOS-IC still exists,SMOS-IC achieves an effective improvement over the conventional SMOS-L3 soil moisture product at global scale.Considering several disturbances and climate types,although SMAP and SMOS-IC achieved satisfying correlations at moderate or intensive VOD,low surface roughness,and low heterogeneity,the enhancement of L-band soil moisture products remains challenging at low VOD,high roughness,and heterogeneity.Secondly,an assessment framework on L-band soil moisture products based on the integration of in situ and mathematical methods,was proposed.Then the uncertainty analysis of the five latest mainstream L-band products over tropics was conducted.From the results,all the L-band soil moisture products show acceptable accuracy except for the lowest and highest vegetation conditions,with high TCA/IVd-R2(Triple Collocation Analysis,TCA;double Instrumental Variable,IVd)values(over 0.6).This finding also experimentally demonstrates the promissing capability of current L-band satellites to estimate soil moisture in moderately vegetated areas.However,in most densely vegetated areas over tropics(e.g.,rainforests),all the L-band soil moisture products have some uncertainties.The climatology assessment of soil moisture suggests that L-band products,especially for SMOS-IC and SMAP-IB,are relatively reliable for climatological changes in tropical rainforests.As for specific products,the newly developed SMAP-IB achieves satisfactory performance with higher accuracy than other L-band products in tropical areas except for rainforests.SMAP-DCA shows comparable performance with SMAP-SCA-V.Based on the results of uncertainty analysis,this paper summarizes improvement strategies for current L-band soil moisture products from three perspectives:satellite observation,inversion strategy and forward model.Meanwhile,motivated by the needs of improvement remotely sensed soil moisture products,this paper proposed an evaluation framework for several model-/satellitebased soil temperature products.This paper evaluated five model-based soil temperature products(GEOS-5,MERRA-2,GLDAS Noah,ERA-Interim and ERA5)and one satellite-based soil temperature product(AMSR2 LPRM)using ground observations from about 800 stations worldwide.The results indicate that the GEOS-5 soil temperature product obtains the smallest ubRMSD of 1.84 K.All modeled soil temperature products generally exhibit lower temperature values than ground observations and show satisfying performance in capturing temporal trends of ground observations with an average correlation coefficient of more than 0.97.In addition,the impacts of land cover,climate type,altitude,soil temperature,and soil moisture conditions on these soil temperature products were systematically analyzed in this work.And suggestions for optimal temperature inputs are provided for the following work concerning improvement of soil moisture products.Finally,based on the above uncertainty evaluation and error analysis for soil moisture products,this paper developed an integration of active-passive for soil moisture enhancement method,taking the error sources into account.Specifcally,a soil moisture enhancement study by integration of the newest L-band SMAP passive observation and its only same scale operating in orbit ASCAT active obsevation was conducted.The consideration of various error sources is reflected in the specific implementation by inputting a variety of static,dynamic parameters as well as spatiotemporal information parameters.Specifically,this study firstly compares the capabilities of four mainstream machine learning approaches for soil moisture inversions through integration of active-passive satellite observations.The results show that the random forest(RF)obtains better results concerning all the error metrics and was selected as the optimal modeling approach for the soil moisture retrievals at global scale.Based on the validation results using global dense observation network,the proposed integration method achieves the expected results with an average ubRMSD of 0.042 m3/m3,the temporal correlation coefficient of 0.76,and an average deviation of-0.019 m3/m3.The developed integration method achieves a significant accuracy improvement relative to both the original active and passive products,and can obtain or even slightly outperform the accuracy of and ESA CCI products.Moreover,the temporal resolution of the joint soil moisture products in major regions of the world has been significantly improved.