Estimating By Thresholds Optimization And Modeling Soil Moisture Based On ATI And Modified TVDI For The Chinese Loess Plateau

Soil moisture(SM)is widely recognized as a key land surface parameter that associates with the hydrological cycle,and accurately and timely SM estimation at a continually spatiotemporal scale can better understand the water-energy balance and the land-atmosphere interaction.SM estimation with the high spatiotemporal resolution is a sound strategy to forecast flooding/droughts,climate prediction,and schedule irrigation for the sustainability and productivity of agriculture,particularly in arid and semi-arid regions like the Chinese Loess Plateau(CLP).At present,various methods of SM estimation at smaller spatial scales have been proposed using visible/thermal-infrared remote sensing data.However,it is inappropriate to employ a single model to estimate SM throughout a larger study area because the single method generally neglects the applicability resulting in lower accuracy.This study estimated SM information at a high spatiotemporal scale using the Moderate Resolution Imaging Spectroradiometer(MODIS)data and SM observations at the stations.Firstly,based on the modified temperature vegetation dryness index(TVDI)-based model,the method we have proposed allows us to identify optimal normalized difference vegetation index(NDVI)thresholds by performing 10 rounds 10-fold cross-calibration and thus to estimate SM using the apparent thermal inertia(ATI)-based and TVDI-based models.Then the proposed method was used to estimate SM for each 8-day imagery at 500m spatial resolution over the CLP in 2017 and monthly,seasonal,and yearly SM maps were produced via the 8-day SM maps.In order to improve the spatial coverage of the estimated SM and explore the affecting factors on the changes of SM,based on reference SM(estimated using proposed method),34 candidate variables from multisource data were then processed and modeled for SM via stepwise multilinear regression(SMLR).After accuracy assessment,monthly,seasonal,and annual spatiotemporal patterns of the modeled SM were mapped and analyzed.The estimated SM using proposed method and the modeled SM via SMLR were compared and evaluated lastly.The key findings and main conclusions are summarized in the following:(1)The modified TVDI improved the accuracy of the SM estimation.The new TVDI value was calculated based on a given NDVI₀(the low limit of NDVI to derive dry/wet edges)in iteration and the optimal NDVI₀ was identified with the best validation results between the estimated and observed SM.Results of estimated SM in 2017 over the CLP indicate that the optimal NDVI₀ for 8-day periods were not fixed and fluctuated irregularly throughout the year.There were few periods(3/43periods)of the optimal NDVI₀ equaling to zero(consistent with the original TVDI).In addition,the NDVI₀ corresponding to the highest fitting coefficient of the dry/wet edges for each period is greater than zero as well.Thus,the modified TVDI presents a better fitting relationship between dry and wet edges,thereby truly reflecting the SM variation mechanism in the study area and improving the accuracy of SM estimation.(2)SM was estimated using the ATI-based and modified TVDI-based models by identification of thresholds optimization over the CLP in 2017.The results indicate that there was a better correction between estimated and observed SM and the accuracy of SM estimation based on the ATI-based and TVDI-based models was generally higher than that of SM estimation solely using the single model.In terms of the subregional SM estimation,the ATI/TVDI joint model had higher applicability(accounting for 40/45 periods)and accuracy(maximum (?)of 0.82±0.007)than the ATI-based and TVDI-based models.In addition,the 10 times 10-fold cross-calibration method in the iteration procedure improves overall reliability and effectiveness in terms of identifying optimal NDVI thresholds.(3)Modeled SM using multivariable(34 candidate variables)extracted from multisource data via SMLR based on reference SM.Results indicate that there was a high potential of SMLR to model SM with the desired accuracy(best fit of the model with Pearson's r=0.969 and root mean square error=0.761%in December)over the CLP.The SMLR-modeled SM for almost all periods except for January and February achieved better spatial coverage than that of reference SM.The variables of elevation(0-500 m and 2000-2500 m),precipitation,soil texture of loam,and nighttime land surface temperature can continuously be used in the regression models for all seasons.(4)The spatiotemporal variation characteristics of SM in 2017 over the CLP were analyzed based on the multivariable modeled SM.From the perspective of SM temporal variation,the seasonal characteristics of SM were obvious and there were two peaks in April(SM:13.982%)and August(SM:18.610%)throughout the year.The mean modeled SM was highest(?13.810%)in autumn and summer.Spatially,SM decreased gradually from southeast to northwest of the CLP,which was consistent with the distribution of total precipitation,average annual air humidity,average annual evapotranspiration,the diurnal difference in land surface temperature,and soil texture in 2017.In general,the western,southern,and southeastern regions of the CLP are wetter from April to August(spring and autumn)than other regions,while the southern region and northwest region are wetter and dryer,respectively,throughout the year.There are 52 figures,28 tables,and 229 references in this dissertation.