Spatial Downscaling of Remotely Sensed Soil Moisture Using Support Vector Machine in Northeast Asia
Thursday, 18 December 2014
Recent advances in remote sensing of soil moisture have broadened the understanding of spatiotemporal behavior of soil moisture and contributed to major improvements in the associated research fields. However, large spatial coverage and short timescale notwithstanding, low spatial resolution of passive microwave soil moisture data has been frequently treated as major research problem in many studies, which suggested statistical or deterministic downscaling method as a solution to obtain targeted spatial resolutions. This study suggests a methodology to downscale 10 km and 25 km daily L3 volumetric soil moisture datasets from Advanced Microwave Scanning Radiometer 2 (AMSR2) in 2013 in Northeast Asia using Support Vector Machine (SVM). In the presented methodology, hydrometeorological variables observed from satellite remote sensing which have physically significant relationship with soil moisture are chosen as predictor variables to estimate soil moisture in finer resolution. Separate downscaling algorithms optimized for seasonal conditions are applied to achieve more accurate results of downscaled soil moisture. A comparative analysis between in-situ and downscaled soil moisture is also conducted for quantitatively assessing its accuracy. Further application can be carried out in hydrological modeling or prediction of extreme weather phenomena in fine spatial resolution based on the results of this study.