Title: Uncertainty Analysis in the Decadal Survey Era: A Hydrologic Application using the Land Information System (LIS)
Author: Ken Harrison
Organization: UMD ESSIC/NASA Goddard Space Flight Center
Co-Authors: Sujay Kumar, Christa Peters-Lidard, Joseph Santanello

Abstract:
Computing and algorithmic advancements are making possible a more complete accounting of errors and uncertainties in earth science modeling. Knowledge of uncertainty can be critical in many application areas and can help to guide scientific research efforts. Here, we describe a plan and progress to date for a fuller accounting of hydrologic modeling uncertainties that addresses the challenges posed by decadal survey missions. These challenges include the need to account for a wide range of error sources (e.g., model error, stochastically varying inputs, observational error, downscaling) and uncertainties (model parameters, error parameters, model selection). In addition, there is a need to incorporate into an assessment all available data, which for decadal survey missions includes the wealth of data from ground, air and satellite observing systems. Our core tool is NASAís Land Information System (LIS), a high-resolution, high-performance, land surface modeling and data assimilation system that supports a wide range of land surface research and applications. Support for parameter and uncertainty estimation was recently incorporated into the software architecture, and to date three optimization algorithms (Levenberg-Marquardt, Genetic Algorithm, and SCE-UA) and two Markov chain Monte Carlo algorithms for Bayesian analysis (random walk, Differential Evolution-Monte Carlo) have been added. Results and discussion center on a case study that was the focus of Santanello et al. (2007) who demonstrated the use of remotely sensed soil moisture for hydrologic parameter estimation in the Walnut Gulch Experimental Watershed. We contrast results from uncertainty estimation to those from parameter estimation alone. We demonstrate considerable but not complete uncertainty reduction. From this analysis, we identify remaining challenges to a more complete accounting of uncertainties.