Predictive uncertainties in climate models remain high, and confidence in the magnitudes of these uncertainties are low for many quantities of interest, especially for regional precipitation and temperature. In the Accelerated Climate Model for Energy (ACME) project, a key research focus is to improve model representations of hydrology and biogeochemical cycling to reduce uncertainties. Uncertainty quantification (UQ) techniques are being applied to the land and atmosphere component models, but new methods are needed to address the computationally expensive coupled model. A large number of land and atmosphere observations also exist but have not been incorporated in ACME or other models for the purpose of estimating predictive uncertainties. Here we propose to address two key science questions related to the application of new UQ techniques:
What is the optimal design of a measurement network to minimize climate model prediction uncertainties?
How much do these uncertainties in land and atmosphere physics contribute to predictive uncertainty and which processes are most responsible?
While the methods below will allow for analysis of any quantity of interest, we will focus on uncertainties in regional-scale near-surface air temperature and precipitation in the midwestern United States, the Amazon region of South America, and northwestern North America. To answer these science questions, we propose the following steps that will comprise a new UQ framework for ACME:
Calibrate a network of coupled land-atmosphere single-column simulations using observations from intensively studied sites
Apply statistical methods to upscale these network simulations to regional scale
Optimize the placement of new simulation points for calibration and upscaling
Propagate uncertainties from the calibrated model forward using a multiple-resolution ensemble of the fully coupled ACME model
The resulting advance from these methods in understanding about the sensitivity of model outputs to specific processes, and how model uncertainties contribute to prediction uncertainty, will be critically important for future model development and for providing actionable information from models to policymakers.