Spatial And Temporal Variability Analysis For Microwave Remote Sensing Of Soil Moisture And Hydrological Data Assimilation

Soil moisture is a key parameter to simulate hydrological processes such as evapotranspiration,runoff and groundwater recharge,and is also one of the key variables to understand the interactions between land and atmosphere.It is an important research object in many disciplines such as weather forecasting,drought and flood monitoring and crop yield estimation.With the continuous development and popularization of remote sensing satellites,remotely sensed soil moisture has gradually become an important technic for obtaining soil moisture information.Soil moisture can be applied in many disciplines(such as meteorology and agroforestry)and various remote sensing applications(such as data assimilation and data fusion studies).However,remotely sensed soil moisture observations inevitably contain errors,which including instrument observation error,inversion model error,and representative error caused by subsequent data processing.Understanding the temporal and spatial variability of remotely sensed soil moisture errors is crucial to correctly interpret the observation information in satellite-based soil moisture and successfully assimilate them into hydrological models.The main contents of this thesis are as follows:(1)Consistency analysis of the L-band brightness temperature(Tb)observations for SMAP and SMOS satellites.Pearson’s correlation coefficient,bias and Hovm?ller diagram were used to compare and analyze the L-band SMAP and SMOS Tbs in different land cover types and different climate zones.Results indicate that SMAP and SMOS Tbs are highly consistent over the global landmass,because the correlation coefficient between SMAP and SMOS Tbs is generally greater than 0.8 and their bias values varied from-5 K to 5 K.The Hovm?ller diagram shows that the seasonal dynamic changes of SMAP and SMOS Tbs are highly consistent.In addition,the correlation coefficient between SMAP and SMOS Tbs is smaller than 0.8 in deciduous coniferous forest,mixed forest,bare/sparse vegetated area and several continental climate zones.SMOS Tbs are generally warmer than SMAP Tbs,with a bias value of 0.7 K at the horizontal polarization and 0.16 K at the vertical polarization.It should be noted that the RFI(Radio Frequency Interference)contamination may explain the significant difference between SMAP and SMOS Tbs in Eurasia and eastern North Africa.(2)Analysis for temporal variability in remotely sensed soil moisture errors.At present,literature about the temporal variability of satellite-based soil moisture errors is very limited.To address this problem,we applied the window-based TCA with ASCAT and SMAP soil moisture products to obtain their monthly climatological errors in the four latitudinal climate zones.Results indicate that soil moisture errors have a large temporal variability.No matter for ASCAT or SMAP soil moisture,the relative difference between static and monthly climatological errors is generally greater than 10% and is about 40% on global average.In general,we found that monthly climatological errors were smaller than static errors.The static error provides an upper boundary rather than an average reference for the monthly climatological errors.In the tropics,the static error is largely deviated from the monthly climatological error in the dry seasons whereas is small deviated from the monthly climatological error in the wet seasons.On the other hand,we also investigated correlations between SMAP daily errors with LAI(Leaf Area Index)and rainfall in the four climate zones and seven land cover types.Results reveal that rainfall is a better indicator than LAI to predict temporal variability in remotely sensed soil moisture products.Compared with LAI,rainfall exhibits higher correlation with SMAP daily errors in 68% of total land pixels(N=57926)when Pearson’s correlation is used.Lag correlation analysis showed that soil moisture error peak typically occurs before LAI and rainfall peak.Soil moisture error peak generally agrees well with rainfall peak.However,savannas and woody savannas is a special case,in these two land cover types,soil moisture error peak comes before LAI and rainfall peaks,then follows rainfall peak,and LAI peak appears at last.These results can help community better understand time-variant error characteristics in remotely sensed soil moisture products and provide instructions for predicting time-variant soil moisture errors.(3)Analysis for spatial variability in satellite-derived soil moisture errors.Firstly,the classical Triple Collocation Analysis(TCA)was used to estimate the SMAP and SMOS soil moisture error variances over the global landmass.Then,multiple linear regression method was used to analyze the impacts of six factors(including vegetation optical depth,water body fraction,urban fraction,RFI contamination,clay and sand fractions)on SMAP and SMOS retrieval error variances.Results show that vegetation has the largest influence on microwave remote sensing soil moisture error variance as compared with other alternative factors.The RFI contamination has a significant influence on the SMOS soil moisture error variance in Eurasia and Africa.Secondly,we proposed a new methodology(i.e.the TC＿Cov method)to estimate spatially correlated error covariance for remotely sensed soil moisture products.The effectiveness of the TC＿Cov method was verified by a practical experiment with the Tobler’s First Law of Geography.The spatial variability analysis of satellite-derived soil moisture errors provides a new study direction for data assimilation and data fusion research.(4)Hydrological data assimilation.Firstly,we implemented three point-scaled data assimilation experiments in the USA continent to illustrate the basic procedures in a remote sensing data assimilation and to explain the validity of the data assimilation method.In the three point-scaled data assimilation experiments,soil moisture errors were defined by the prior knowledge.CCI soil moisture data were assimilated into the CLM 4.0 model simulations via the En KF(Ensemble Kalman Filter)method.The assimilation results were validated with in situ observations.Secondly,we estimated observation error covariance(off-diagonal matrix)for AMSR-E soil moisture in the Murrumbidgee basin using the TC＿Cov method proposed in the third chapter.The estimated off-diagonal observation error covariance matrix and the En KF method were used to assimilate AMSR-E soil moisture into the VIC model simulations.A control experiment was also implemented.The two assimilation experiments were exactly identical except for the observation error covariance matrix.By comparing diagonal and off-diagonal assimilation results,we found that off-diagonal assimilation results have higher data quality than diagonal assimilation results.The off-diagonal assimilation results are more accurate and more precise than diagonal assimilation results.Our results provide a framework and reference for considering spatially correlated observation error covariance in hydrological data assimilation systems.