In this study, spatially and temporally distributed soil moisture were estimated using high resolution multiple sensors. The MODIS(MODerate Resolution Imaging Spectroradiometer) and Sentinel-1A/B based soil moisture values in South Korea were estimated based on the regression model. Then, we estimated the long-term(2001 to 2018) daily soil moisture values at the spatial and temporal scales using Sentinel-1A/B Synthetic Aperture Radar(SAR) sensor images and soil moisture data assimilation technique. Also, we tested a deep learning approach in deriving Sentinel-1A/B based soil moisture estimates. Additionally, SMDI(Soil Moisture Deficit Index) and FFDI(Forest Fire Danger Index) were estimated to evaluate drought and wildfire vulnerability using the daily estimated soil moisture values.

1. We estimated the spatially-distributed soil moisture using MODIS and Sentinel-1A/B satellite images. The monthly MODIS soil moisture values were highly influenced by precipitation amounts. Also, the results showed that the soil moisture has the strong correlation with DEM(Digital Elevation model) while the temperature was inversely correlated with the soil moisture. Additionally, the Sentinel-1A/B based soil moisture(10m×10m) were estimated based on the linear relationship between the TDR(Time Domain Reflectrometry) measurements and backscatter coefficients. The regression equations were derived based on the relationships between the TDR measurements and backscatter coefficients. The TDR measurements at the 51 monitoring sites from RDA(Rural Development Administration) were used to derive the regression equations. Then, the Sentinel-1A/B based soil moisture values were estimated using the derived regression equations. The Sentinel-1A/B based soil moisture matched the TDR measurements with the high correlations(more than 0.7), although uncertainties exist, especially for forest and urban areas.

2. We estimated the spatially/temporally distributed soil moisture using Sentinel-1A/B SAR(Synthetic Aperture Radar) sensor images and soil moisture data assimilation technique in South Korea. The soil moisture data assimilation technique adapting GA(Genetic Algorithm) can extract the soil hydraulic parameters from soil moisture measurements. The SWAP(Soil Water Atmosphere Plant) model integrated with the soil moisture assimilation technique simulates the soil moisture using the estimated soil hydraulic parameters and meteorological data. The soil moisture data assimilation technique derived the soil hydraulic parameters from the individual Sentinel-1A/B based soil moisture images. Then, the derived soil hydrological parameters as the input data to SWAP were used to simulate the daily soil moisture values from 2001 to 2018 with the TRMM/GPM satellite rainfall data. Overall, the simulated soil moisture estimates matched well with the Sentinel-1A/B based soil moisture under various land surface conditions(bare soil, crop, forest, and urban).

3. A deep learning(Deep Neural Network-DNN) model was used to estimate soil moisture using Sentinel-1A/B based backscatter coefficients and environmental input data(DEM, landuse, soil types, temperature, and rainfall). The number of hidden layers/nodes and the radio of drop-out were determined based on the least error empirically through the given generation(epoch). Also, we tested the DNN performance in estimating soil moisture based on various antecedent rainfall conditions from the present(n) to past 5 days(n-5). The correlation(R: 0.974) of estimated soil moisture with the antecedent rainfall(n-1) was higher than others. Although this study was conducted under the limited conditions, the performance of DNN could be improved with more input data available.

4. We estimated SMDI and FFDI to assess drought and wildfire vulnerability in South Korea. SMDI was estimated using the daily soil moisture estimates based on the Sentinel-1A/B SAR images and soil moisture data assimilation technique while FFDI was estimated using KBDI and the meteorological data. Overall, the spatial distributions of SMDI showed the similar trends at the spatial domain with GPM, but FFDI showed differences compared to SMDI and GPM. Although this study was validated under the limited conditions, our findings showed the potential that the soil moisture values can be used to evaluate drought and wildfire vulnerability.