Estimation and evaluation of high spatial resolution surface soil moisture using multi-sensor multi-resolution approach

Abstract

Various soil moisture products (AMSR-E at 25 km, SMAP at 9 km, CCI at 25 km) available from different space agencies, derived using the active microwave remote sensing, may not be suitable for studies at a local level due to their coarse spatial resolution. Hence, an attempt has been made in this study to estimate the surface soil moisture at high spatial resolution using different high-resolution multispectral imageries from Landsat 8 OLI (Optical Land Imager) and Sentinel 2A MSI (Multi Spectral Instrument), of similar date. The thermal (10.9 µm) and short wave infra-red (SWIR) (2.2 µm) bands of the Landsat image and the SWIR band (2.19 µm) of Sentinel image were used to estimate the soil moisture at a spatial resolution of 30 m and 20 m respectively, whereas the red and near infrared (NIR) bands of both the images were used to estimate the dryness. As the thermal band at high spatial resolution (at 20 m or below) is not available, this study attempted to bridge the gap and integrate multi-sensor and multi-resolution feature space utilizing various scalable procedures so as to estimate soil moisture and dryness. Limitation due to unavailability of the thermal bands in Sentinel multispectral image was overcome through a bridging procedure using Landsat thermal bands derived LST and Sentinel derived NDVI to estimate Sentinel LST at 20 m. The distribution pattern of high-resolution estimated soil moisture in each coarse spatial resolution grid (1 km) was taken into consideration to downscale the available CCI and SMAP soil moisture data. The estimated soil moisture and dryness were upscaled to 1 km and then validated using the downscaled soil moisture products (SMAP and CCI), and a significant correlation having R2 greater than 0.8, 0.5 and 0.4 was observed for the thermal band derived soil moisture, short wave band derived soil moisture, and the red and NIR band derived dryness, respectively. To understand the temporal variability of the soil moisture, data from three different seasons representing extreme soil moisture conditions (winter, summer and monsoon) were analyzed and found that the independently estimated soil moisture had errors in temporal continuity and spatial spread. Hence, the estimated soil moisture was standardized using annual information which resulted in more accurate products, and the SWIR Transformed Reflectance (STR) derived soil moisture was more accurate than LST derived product.