Quantifying the change in soil moisture modeling uncertainty from remote sensing observations using Bayesian inference techniques


Corresponding author: K. Harrison, Hydrological Sciences Laboratory, Code 617, NASA Goddard Space Flight Center, 8800 Greenbelt Rd., Greenbelt, MD 20771, USA. (ken.harrison@nasa.gov)


[1] Operational land surface models (LSMs) compute hydrologic states such as soil moisture that are needed for a range of important applications (e.g., drought, flood, and weather prediction). The uncertainty in LSM parameters is sufficiently great that several researchers have proposed conducting parameter estimation using globally available remote sensing data to identify best fit local parameter sets. However, even with in situ data at fine modeling scales, there can be significant remaining uncertainty in LSM parameters and outputs. Here, using a new uncertainty estimation subsystem of the NASA Land Information System (LIS) (described herein), a Markov chain Monte Carlo (MCMC) technique is applied to conduct Bayesian analysis for the accounting of parameter uncertainties. The Differential Evolution Markov Chain (DE-MC) MCMC algorithm was applied, for which a new parallel implementation was developed. A case study is examined that builds on previous work in which the Noah LSM was calibrated to passive (L-band) microwave remote sensing estimates of soil moisture for the Walnut Gulch Experimental Watershed. In keeping with prior related studies, the parameters subjected to the analysis were restricted to the soil hydraulic properties (SHPs). The main goal is to estimate SHPs and soil moisture simulation uncertainty before and after consideration of the remote sensing data. The prior SHP uncertainty is based on the original source of the standard SHP lookup tables for the Noah LSM. Conclusions are drawn regarding the value and viability of Bayesian analysis over alternative approaches (e.g., parameter estimation, lookup tables) and further research needs are identified.