#L06 Forward and Inverse Uncertainty Quantification for ALM Single Point Model

Abstract

The poster will highlight recent advances and results relevant to uncertainty quantification (UQ) method development and application to ACME Land Model (ALM). We are developing and applying two broad categories of UQ algorithms:

• Forward UQ encompasses input parameter uncertainty representation, propagation, as well as global sensitivity analysis (GSA), otherwise known as variance-based decomposition. We will highlight Polynomial Chaos (PC) expansions and Bayesian compressive sensing (BCS) for high-dimensional model surrogate (proxy, metamodel, emulator) construction as the main ingredient of any forward UQ study.
• Inverse UQ is essentially parameter estimation (calibration, tuning) given observational data. We will highlight Bayesian machinery, and a recently developed model error embedding approach for simultaneous estimation of physical parameters and structural errors.

The poster will demonstrate surrogate model construction and GSA (forward UQ) based on ensemble of simulations of ALM point model for a select set of FLUXNET sites, as well as initial exploratory results of calibration (inverse UQ) with structural errors given monthly latent heat flux data at the Missouri Ozark flux tower.