Large Eddy Simulations Refine Cold Pool Parameterizations in Global Climate Modeling

Atmospheric cold pools, formed beneath convective clouds when falling precipitation evaporates into unsaturated air, play a central role in organizing and sustaining convective systems. As evaporating raindrops extract latent heat, the cooled air sinks rapidly toward the surface in unsaturated downdrafts and spreads horizontally as density currents bounded by turbulent gust fronts. Despite their fundamental importance in driving the diurnal cycle of rainfall and triggering new convective updrafts, global climate models have historically omitted explicit cold pool dynamics or evaluated them solely through indirect impacts on large-scale precipitation. In a landmark study published in Atmospheric Chemistry and Physics, researchers led by Mamadou Lamine Thiam conducted the first direct evaluation of the internal physical formulations of the cold pool parameterization in the LMDZ climate model by comparing simulated internal variables against high-resolution Large Eddy Simulations.

Atmospheric General Circulation Models, which operate on coarse horizontal grid spacings between thirty and three hundred kilometers, cannot resolve individual convective clouds or their associated boundary layer currents directly. Consequently, these models must parameterize subgrid convective transport, a necessity that frequently leads to severe structural biases such as underestimating extreme rainfall rates and predicting premature midday precipitation peaks that closely track peak solar insolation. In nature, convective precipitation over land often peaks in late afternoon or at night, a delayed timing governed largely by cold pools maintaining convection past sunset. To address these deficiencies, developers incorporated an explicit cold pool parameterization into the LMDZ climate model, coupling it with the Emanuel deep convection scheme and the thermal plume model. Although this coupling shifted simulated tropical precipitation peaks toward mid-afternoon, the internal variables governing the simulated cold pools remained unvalidated against resolved physical dynamics until now.

The physical framework formulated by Grandpeix and Lafore conceptualizes a subgrid population of circular cold pools expanding across a horizontally homogeneous surface. The entire population within a model column is approximated by an effective mean wake characterized by a uniform radius, top height, and vertical anomaly profile of potential temperature and specific humidity. The primary driver of cold pool expansion is the Wake Available Potential Energy, defined as the vertical integral of the virtual potential temperature anomaly from the surface up to the cold pool top, where thermal differences between the wake interior and environment vanish. The mean radial spreading rate of the wake boundary is proportional to the square root of two times this available potential energy multiplied by a dimensionless tuning coefficient. Furthermore, the parameterization assigns the evaporative cooling from unsaturated convective downdrafts to the wake interior while restricting environmental heating from saturated drafts and boundary layer thermal plumes to the exterior.

To establish interactive feedback with the atmospheric column, the parameterization couples spreading cold pools directly to convective triggering and closure formulations. Convective clouds trigger when the Available Lifting Energy exceeds ambient convective inhibition, where the lifting energy provided by cold pools equals the kinetic energy associated with the maximum spreading speed along the wake contour. The parameterization sets this maximum speed proportional to the square root of twice the Wake Available Potential Energy with a coefficient of unity, postulating that cold pool lifting energy matches its total collapse energy. Once triggered, the convective mass flux at cloud base depends on the combined Available Lifting Power supplied by thermal plumes and expanding cold pools. This lifting power is computed by integrating horizontal kinetic energy flux over the wake perimeter and depth, adjusted by an empirical lifting efficiency factor nominally fixed at twenty-five percent to account for dissipation.

To rigorously evaluate these conceptual equations, the researchers employed high-resolution, three-dimensional Large Eddy Simulations across two contrasting meteorological environments. The first test environment represented an oceanic Radiative-Convective Equilibrium, a controlled and quasi-stationary numerical setup without large-scale dynamic forcing, simulated over a domain of two hundred kilometers by two hundred kilometers at a fine horizontal resolution of two hundred and fifty meters using both the SAM and Meso-NH models. The second benchmark was a realistic continental diurnal case derived from observational field campaign data collected during the African Monsoon Multidisciplinary Analysis over Niamey, Niger, on July 10, 2006. This continental case simulated a short-lived Sahelian convective storm over an identical domain at two hundred meters resolution using Meso-NH, capturing the complete diurnal evolution of deep convection triggered by intense daytime surface heating.

Evaluating internal parameterization variables against resolved simulation fields required an objective sampling methodology to separate cold pool interiors from the undisturbed ambient environment across the domain. Because the simulations employed idealized, horizontally uniform surface boundary conditions, the investigators identified cold pool boundaries by applying specific thresholds to the horizontal temperature anomaly at ten meters altitude. Comprehensive sensitivity testing determined that temperature anomaly thresholds of minus zero point two Kelvin for the oceanic equilibrium and minus one point zero Kelvin for the continental case reliably demarcated cold pools from surrounding air masses. To quantify lifting dynamics, the authors constructed a gust front mask by horizontally smoothing vertical velocity at cloud base, applying thresholds of zero point six meters per second for the marine regime and two point zero meters per second for the continental system. This mask confirmed that powerful updrafts coincided with spreading thermal boundaries.

Direct analysis of the resolved simulation data provided striking verification of several foundational relationships while revealing critical parameter misalignments. The sampled vertical velocities at cloud base confirmed the theoretical hypothesis that Available Lifting Energy equals the Wake Available Potential Energy, with parameterized calculations matching simulation estimates within thirty percent across all regimes. However, when evaluating the mean horizontal spreading velocity from sampled surface wind divergence, the parameterization using its standard velocity coefficient of zero point three three systematically underestimated cold pool propagation speeds. Increasing this coefficient to zero point five six brought parameterized spreading rates into excellent quantitative alignment with resolved wind fields in both marine and continental settings. Similarly, calculated Available Lifting Power was initially underestimated by more than a factor of three, but adopting the adjusted velocity coefficient of zero point five six while retaining the nominal twenty-five percent lifting efficiency successfully reconciled the formulation with resolved gust front power.

When the full cold pool scheme was subsequently tested in the Single-Column Model version of LMDZ against the simulations, the comparison highlighted prominent systematic biases in the baseline control configuration. Although the model accurately reproduced the relative depth of cold pools—reaching approximately six hundred meters over the ocean and roughly two thousand meters over the Sahelian continent—the simulated thermodynamic anomalies exhibited marked errors. In the oceanic equilibrium, simulated cold pools were noticeably warmer than those in the simulation, generating less than half the expected collapse energy and underestimating Available Lifting Power by a factor of twenty. Furthermore, while the model captured near-surface moistening and upper-level drying, it produced an excessively severe dry anomaly near the cold pool top. This localized desiccation was directly traced to a rigid assumption in the original scheme, which fixed the upper boundary where dry air subsides into the wake at six hundred hectopascals regardless of cold pool depth.

To resolve these structural deficiencies, the research team developed a succession of refined model configurations. Version one updated the spreading velocity coefficient to zero point five six, which substantially improved the vertical profiles of velocity difference below the cold pool top and increased lifting power by more than a factor of five, although a warm bias remained. Version two introduced a dynamically calculated upper boundary pressure that scaled directly with the diagnosed height of the cold pool top, replacing the arbitrary six-hundred-hectopascal threshold. This dynamic formulation eliminated the severe dry bias near the cold pool summit by preventing unphysical entrainment of dry air from upper levels into shallow marine wakes. A third experimental version tested the activation of boundary layer thermal plumes across the entire grid cell, which helped reduce excessive near-surface humidity through vertical mixing but simultaneously exacerbated the surface cold bias in the continental regime.

Recognizing that parameterizations interact nonlinearly within a full climate model, the investigators implemented a comprehensive multi-parameter calibration using the automated High-Tune Explorer software tool. Applying history matching algorithms across thirteen iterative tuning waves, the tool searched multi-dimensional parameter spaces to find parameter combinations that simultaneously matched thirteen target metrics covering vertical thermodynamic profiles and energetic variables. The automated calibration identified optimal parameter bounds that closely mirrored the values derived from direct simulation sampling, selecting a velocity coefficient between zero point five six and zero point five seven and an effective lifting efficiency between twenty-six and forty-six percent. Crucially, the calibration revealed that expanding the fractional area allocated to unsaturated convective downdrafts from zero point zero zero three to between zero point zero four two and zero point zero four eight intensified evaporative cooling, which resolved the cold pool temperature deficit in the oceanic equilibrium while correcting severe specific humidity biases.

The resulting calibrated configuration achieved remarkable quantitative agreement with Large Eddy Simulations, successfully reproducing the collapse energy, spreading rates, and lifting power of continental cold pools at their peak development. Nonetheless, important scientific challenges remain before full atmospheric implementation is complete. The single-column model initiated deep convection approximately three hours earlier than the Large Eddy Simulation in the continental case, a timing discrepancy reflecting unresolved physics in the transition from shallow to deep convective clouds that operates independently of cold pool dynamics. Furthermore, simulated cold pools still exhibit excessive surface moisture unless a dedicated parameterization of shallow convective mixing within wake interiors is introduced. Future developmental efforts within the LMDZ modeling group are actively incorporating prognostic formulations for cold pool number density, introducing interactions with ambient vertical wind shear, and enabling lateral cold pool propagation across adjacent model grid cells to support the IPSL-CM7 coupled climate model for upcoming CMIP7 evaluations.

Subject of Research: Evaluation and improvement of convective cold pool parameterization in the LMDZ climate model

Article Title: Evaluation and improvement of the cold pool parameterization in the LMDZ climate model against Large Eddy Simulations

Article References: Thiam, M. L., Hourdin, F., Grandpeix, J.-Y., Rio, C., & Coulon-Decorzens, M. (2026). Evaluation and improvement of the cold pool parameterization in the LMDZ climate model against Large Eddy Simulations. Atmospheric Chemistry and Physics, 26(21), 15975-16002. https://doi.org/10.5194/acp-26-15975-2026

Image Credits: AI Generated

DOI: 10.5194/acp-26-15975-2026

Keywords: cold pool parameterization, LMDZ climate model, Large Eddy Simulation, convective downdrafts, atmospheric convection, wake available potential energy, available lifting power, available lifting energy, single-column model, radiative-convective equilibrium, AMMA campaign, htexplo, Meso-NH, SAM model, diurnal cycle of precipitation, thermal plume model, gust fronts