Spatial downscaling of gridded soil moisture products using optical and thermal satellite data: the effects of using different vegetation indices

Abstract: Soil moisture (SM) plays an important role in the exchange of heat and water between the surface and atmosphere, impacting water and energy cycles and the climate. Satellite remote sensing offers a global-scale estimation of SM; however, the coarse resolutions of satellite SM products, typically ranging from 25-50 km, are unsuitable for regional analysis. To overcome this limitation, various spatial downscaling methods have been developed to disaggregate SM products at coarse resolution to estimates at higher resolution. One commonly used approach is the optical and thermal-based method, which utilizes higher resolution ancillary data, such as land surface temperatures (LST) and vegetation indices (VI), within a triangular feature space. Previous studies have primarily relied on the use of NDVI (Normalised difference Vegetation Index) or EVI (Enhanced Vegetation Index) as VIs, neglecting the potential benefits of newly proposed VIs for spatial downscaling. Consequently, few studies have investigated the influence of different VIs on the downscaling of gridded soil moisture. This study aims to investigate the influences of using different VIs on spatial downscaling of the coarse resolution SM product. Two study areas are focused on in this study (1) an area around the SMOSMANIA network in southern France with 18 SM measurement stations and (2) an area surrounding the REMEDHUS network in northern Spain with 17 measurement stations. The daily ESA CCI SM product at 0.25° resolution was spatially downscaled using four different VIs including the NDVI, EVI, the kernel NDVI (kNDVI) and Plant Phenology Index (PPI) to produce a higher resolution SM product at 1 km and 16-day resolutions. All four VIs and the LST were obtained from MODIS products. The Vegetation Temperature Condition Index (VTCI) based downscaling approach was used for this study, in which wet and dry edges of the triangular feature space were determined by fitting a linear line to the maximum and minimum temperatures, respectively, for each VI interval. Evaluation showed that using PPI showed better consistency between two study areas, having the good correlation and ubRMSD against the in-situ measurements, whilst the performance of using other VIs particularly EVI and kNDVI varied in the study area. Using NDVI generally yielded the poorest overall performance in terms of ubRMSD and correlation, but it outperformed kNDVI in areas with generally sparser vegetation within the SMOSMANIA network. Comparison of SM product at the original coarse resolution and spatially downscaled SM, the ESA CCI SM product generally outperformed the downscaled SM products, with only 12 out of 35 stations showing superior performance for the downscaled products in terms of correlation and 10 out of 35 stations in terms of ubRMSD.