Optimization of Sensor Networks for Improving Climate Model Predictions
Funded jointly by the Office of Science Biological and Environmental Research (BER) and Applied Scientic Computing Research (ASCR) programs, OSCM, in the context of DOE's flagship climate model E3SM, addresses two major science questions: a) what is the optimal design of a measurement network to minimize climate model prediction uncertainties, and b) how much do the uncertainties in land and atmosphere physics contribute to predictive uncertainty and which processes are most responsible?
The Energy Exascale Earth System Model (E3SM) is central to many of the Climate and Environmental Sciences Division activities, as it is developing a computationally advanced coupled climate-energy model to investigate the challenges posed by the interactions of weather-climate scale variability with energy and related sectors. The OSCM project develops and applies state-of-the-art UQ methods to the coupled land-atmosphere system with the goal of quantifying predictive uncertainty in regional to global scale climate variables. Two of the major UQ tasks, listed below, in this project will be carried out by FASTMath members, with a target to produce a self-contained E3SM UQ product.
- Apply novel UQ methods to calibrate a network of coupled land-atmosphere single-column simulations using land-surface flux and atmospheric observations. These network simulations are estimated to be 2-4 orders of magnitude faster than global coupled simulations. Polynomial chaos expansion (PCE)-based surrogate modeling will be applied to perform global sensitivity analysis (GSA), dimensionality reduction and model calibration that incorporates parametric, structural and observational uncertainties in a Bayesian framework. This task is carried out by Drs. Khachik Sargsyan and Cosmin Safta (SNL).
- Develop a framework to optimize the placement of new simulation points for use in model calibration and reconstruction of global outputs, using existing or synthetic observation data for calibration to minimize the total QoI uncertainty. This task is led by Prof. Youssef Marzouk (MIT).
The project will advance the current understanding about the sensitivity of model outputs to specific processes, and how model uncertainties contribute to prediction uncertainty, providing critically important and actionable information from models to policymakers.