Improved Hydrologic Parametrization and Spatial Uncertainty Characterization of the Community Land Model over the Conterminous United States
Abstract
Hydrologic parameter uncertainties in large-scale land models within Earth System Models (ESMs) remain a major challenge for improving predictions of water storage in the land surface and water fluxes, which subsequently affects ESMs ability to predict water, energy, carbon, and nutrient cycles. While deterministic calibration based on global optimization methods is commonly employed in hydrology, the costly nature of these approaches makes it rarely feasible for large-scale land surface models with hundreds of parameters such as the Community Land Model (CLM) and are limited in their treatment of parametric uncertainties. To address the challenges, we contribute a diagnostic model evaluation and a Monte Carlo regionalization framework for parameterizing CLM. The framework initially exploits global sensitivity analysis to identify the dominant parameters for different hydrologic processes (e.g., surface water, snow, evapotranspiration, groundwater) over different climate regimes of the Conterminous United States (CONUS). We use machine learning to divide 464 CAMELS (Catchment Attributes and Meteorology for Large-Sample Studies) basins over the CONUS into 7 clusters based on their physiographic and climate features and then select 7 representative basins in each cluster for parametric screening. We then employ a diagnostic evaluation of several different hydrologic processes and signatures to identify the dominant parameters in each cluster and then use a Monte Carlo pre-calibration approach to select behavioral parameters. We regionalize CLM predictions in ungauged basins over the CONUS for the 1/8-degree grid cells within the classified 7 zonal clusters. We build and evaluate regression relationships between hydrologic signatures and physiographic/climate features using the CAMELS reference basins within each cluster. Finally, for each grid cell over the CONUS, the regression relationships are used to constrain behavioral parameters and consequently characterize the spatial parametric uncertainty. The new CLM parameterization we develop here improves CLM predictive skill and allows for the first time the CONUS-scale characterization of parametric uncertainty in key hydrologic processes and signatures.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H52C..05Y