Multisite calibration of a multifractal downscaling model for satellite-based soil moisture estimates

The current resolution of soil moisture (s) estimates provided by satellite-borne passive microwave sensors is still too low (25-50 km) to allow their direct use in land surface models simulating water and energy fluxes. Downscaling (or disaggregation) strategies are often required to characterize the sub-grid s variability. In a recent study, we successfully calibrated a downscaling model based on the multifractal theory using 800 m aircraft-based s estimates collected in the southern Great Plains experiment (SGP97) in 1997. The model is based on a log-Poisson generator depending on two parameters and simulates the small-scale (800 m) s probability distribution, starting from the mean soil moisture value in coarse domains of 25.6 km (approximately a satellite footprint), and ancillary predictors accounting for soil texture, land cover and topography. In this work, we present the applicability of this disaggregation scheme to other s datasets collected in different environmental settings and climatic regions. These include: (i) SGP99 data collected in the same area of SGP97 during drier conditions, (ii) Soil Moisture Experiment in 2002 (SMEX02) collected in an area with moderate to heavy water content in Iowa, and (iii) SMEX04, conducted in arid regions in Arizona and Sonora (Mexico). We first calibrate the model at each site as a function of local ancillary factors, and then we explore the possibility to estimate a global calibration relation able to characterize, with minimal computational demand, the sub-grid s variability within satellite estimates in different regions, for use in land-surface models and data assimilation systems.