PROMETHEUS: A Process-Oriented Framework to Accelerate UFS Model Development Cycles
Augustin Vintzileos, Benjamin Cash, and Jim Kinter
ESSIC-University of Maryland, and COLA-George Mason University
20 July, 2026, 12-1pm
Abstract:
The calibration of physical parameterizations in Earth system models remains a
challenging part of the model development workflow, as improvements in one performance metric (e.g., mean temperature over the CONUS or 500-hPa height forecast skill) often come at the expense of others. This arises because a single set of parameter values is expected to represent multiple process regimes governed by different physical processes. One example is air–sea turbulent exchange, which is governed primarily by interfacial turbulence under moderate winds but becomes increasingly influenced by wave breaking, foam, and sea spray in hurricane conditions. Similar transitions occur in convection, cloud microphysics, and boundary-layer turbulence. These transitions motivate a process-oriented model development strategy in which distinct process regimes are identified and optimized separately.
Here, we propose replacing traditional global parameter tuning with a process-oriented model development framework, PROMETHEUS (PROcess-oriented Model dEvelopment Through dEcision tRee HeUristic System). PROMETHEUS uses interpretable decision trees to partition atmospheric columns into objectively defined Column States. Column States are groups of atmospheric columns exhibiting statistically distinct behavior in a target variable or model error, providing a physically interpretable representation of process regimes (e.g., shallow versus deep convection). This enables targeted diagnosis, investigation, and optimization of physical parameterizations within individual Column States, for example using ensembles of Single Column Models. The framework transforms parameterization development from a single global optimization problem into a set of process-specific optimization problems.
We illustrate the approach using tropical outgoing longwave radiation tendency as a proxy for convective development. The resulting tree identifies coherent combinations of moisture profiles, vertical motion, thermodynamic structure, and surface conditions associated with different rates of convective evolution. In particular, a familiar oceanic skin-temperature threshold near 28°C emerges only after the atmosphere has been preconditioned by large-scale moisture divergence and upper-level ascent, demonstrating how the framework objectively identifies conditional process dependencies. Finally, we discuss how PROMETHEUS can be incorporated into the UFS development workflow to provide objective process-level diagnostics, guide targeted parameterization development, and accelerate physics development within the UFS.