Physics-Informed Machine Learning for Uncertainty Quantification in Land Models

• Activity Number: 309 - Interface Between Machine Learning and Uncertainty Quantification
• Type: Topic Contributed
• Date/Time: Wednesday, August 5, 2020 : 10:00 AM to 11:50 AM
• Sponsor: Uncertainty Quantification in Complex Systems Interest Group
• Author(s): Khachik Sargsyan* and Cosmin Safta and Vishagan Ratnaswamy
• Companies: Sandia National Laboratories and Sandia National Laboratories and Sandia National Laboratories
• Keywords: Machine learning; Uncertainty quantification; Surrogate models; Bayesian inference; Variance decomposition; Climate models

Abstract

Complex physical models are computationally expensive, challenging ensemble-intensive studies such as parameter estimation, uncertainty quantification, and optimal experimental design. In order to make such studies tractable, we build efficient and accurate surrogate approximations to maps from input parameters to output quantities of interest (QoIs).

This study focuses on building neural network surrogates for climate land models. Due to the temporal nature of the model, we construct the surrogate with a recurrent neural network (RNN) architecture with long-short term memory (LSTM) units. We then augment the architecture with a hierarchical structure of the relationships between the various inputs, state variables, and QoIs. The resulting hierarchical LSTM RNN mimics the physical constraints and relationships between the underlying processes and provides a natural structure for temporal evolution. Using the hierarchical LSTM surrogate, we then perform global sensitivity analysis to identify the most influential input parameters for dimensionality reduction and Bayesian model calibration when observational data is provided.