Uncertainty Quantification for E3SM Land Component using Low-Rank Surrogate Models
Abstract
Uncertainty quantification studies for earth system models are challenged by the large number of parameters that control these models and by the prohibitive computational cost required for each model evaluation. The non-linear input-output dependencies compound these challenges by limiting the number of reduced-order techniques that could be leveraged in these studies. In this presentation we propose an approach that relies on sparse surrogate models via low-rank functional tensor train approximations. The construction of these surrogates is designed to exploit sparse connectivities between model components and produce a model parsimony commensurate with the flow of information between processes. We propose a set of functional representations that includes both the E3SM land model stochastic parameters and the physical space components by considering approximations spanning several adjacent land cells. We compare the efficiency of this approach for global sensitivity analysis with other sparse approximations in a Bayesian setting.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.H43J2168S
- Keywords:
-
- 0430 Computational methods and data processing;
- BIOGEOSCIENCES;
- 1805 Computational hydrology;
- HYDROLOGY;
- 1846 Model calibration;
- HYDROLOGY;
- 1873 Uncertainty assessment;
- HYDROLOGY