Choosing the Right Storm Formula: RegCM5 Test Reveals What Makes or Breaks Rainfall Forecasts in West Africa

In the heart of Côte d’Ivoire, the Bandama River winds its way from the northern savannas to the Gulf of Guinea, draining a watershed that sustains agriculture, hydropower, and millions of people. How much rain falls over this basin, and when, is a matter of livelihood. Yet simulating that rainfall with a computer model turns out to hinge on a set of choices most people never hear about: which mathematical recipes, called parameterization schemes, the model uses to represent processes too small to resolve directly. A new study published in Theoretical and Applied Climatology has now systematically tested those choices in the fifth-generation Regional Climate Model, RegCM5, and the results show just how dramatically the wrong recipe can distort the picture of West African rainfall.

The research team, led by Mamadou Doumbia of Université Félix Houphouët-Boigny in Abidjan, together with colleagues from Ivorian and French institutions, focused on a deceptively simple question: if you keep everything else in the model the same, how much does the rainfall simulation change when you swap out the schemes that represent convection, cloud microphysics, and the planetary boundary layer? The answer, quantified across thirteen successful simulations, is that the convection scheme alone can swing annual rainfall totals by several hundred percent. In a region where the seasonal monsoon determines planting calendars and reservoir levels, that is not a technical footnote but a headline finding.

To understand why convection matters so much, it helps to know what the scheme actually does. Thunderstorms over West Africa are typically far smaller than the grid cells of a regional climate model, which in this study resolve the watershed at a scale where individual storm systems cannot be explicitly simulated. Instead, the model relies on a parameterization: a set of equations that estimates, on average, how much moisture rises, condenses, and rains out from each grid cell. Four such schemes were tested here: the Emanuel scheme, which bases its calculations on quasi-equilibrium thermodynamics; the Kain–Fritsch scheme, which tracks entrainment and detrainment in rising plumes; the Grell scheme, which treats convection as a stabilizing process between cloud and environment; and the Tiedtke scheme, a mass-flux approach that accounts for both deep and shallow convection.

The team built sixteen possible model configurations by combining these four convection schemes with two cloud-microphysics options, the Subgrid Explicit moisture scheme known as SUBEX and the Nogherotto–Tompkins scheme known as NoTo, and two planetary boundary layer schemes, Holtslag and the University of Washington scheme, or UW. All simulations were driven by the ERA5 reanalysis, a state-of-the-art global dataset that constrains the large-scale atmospheric circulation, and covered the period 2001 to 2009. Thirteen of the sixteen configurations completed successfully and were analyzed, giving the researchers a robust matrix of comparisons over the Bandama watershed.

Evaluating the simulations required trustworthy observations, and in a data-sparse region that is itself a challenge. The researchers turned to two independent satellite-based rainfall products: CHIRPS, the Climate Hazards Group InfraRed Precipitation with Station data, which blends satellite estimates with ground observations, and IMERG, the Integrated Multi-satellitE Retrievals for the Global Precipitation Measurement mission. Against these references, the team applied a battery of statistical measures, including bias, temporal correlation, and distribution- and event-based skill scores. This dual-reference approach matters because satellite rainfall products each carry their own uncertainties; agreement between CHIRPS and IMERG lends confidence that the model rankings reflect genuine differences in physical realism rather than quirks of a single dataset.

The verdict on the convection schemes was unambiguous. The Tiedtke and Emanuel schemes, particularly when paired with the UW boundary layer scheme, performed best. These configurations reproduced the main phases of the seasonal rainfall cycle, from the onset of the monsoon rains in the south to the single-peaked regime of the northern savanna, with moderate annual biases generally below thirty percent. For a watershed-scale climate simulation, that level of agreement is respectable and suggests these configurations could serve as a sound foundation for future climate projections and hydrological studies in the region.

At the other end of the spectrum, the results were sobering. The Grell scheme, in every pairing tested, produced massive overestimates of rainfall, with annual biases exceeding plus 440 percent. The Kain–Fritsch scheme fared poorly as well whenever it was not combined with the UW boundary layer. In practical terms, a modeler who selected Grell without prior calibration would simulate a Bandama watershed drenched by more than five times its actual rainfall, rendering any downstream analysis of floods, soil moisture, or water resources meaningless. The authors’ implicit warning is clear: these schemes should not be used over this region without careful validation, a caution that applies broadly to climate modeling in West Africa, where the interaction between the monsoon circulation, the African easterly jet, and mesoscale convection stresses parameterizations designed elsewhere.

The boundary layer scheme emerged as the second most influential factor. The UW scheme consistently outperformed Holtslag, indicating that how the model represents the turbulent mixing between the surface and the free atmosphere feeds directly into the triggering of convection and hence into rainfall. This finding aligns with earlier work by some of the same authors, who assessed boundary layer parameterization sensitivity over the wider West African domain using RegCM5 and found the choice to be consequential for the mean climate. By contrast, the cloud-microphysics schemes, SUBEX and NoTo, had only a secondary and non-systematic effect on precipitation in this configuration, suggesting that at parameterized-convection resolution the fate of moisture inside resolved clouds matters less than how convection itself is triggered and modulated.

One limitation united all thirteen configurations: a systematic excess of light rainfall, or drizzle. Every simulation produced too many days with small precipitation amounts compared with the satellite references. This drizzle bias is a well-known generic weakness of parameterized convection at these resolutions, arising because the schemes tend to moisten and rain out grid cells too readily, producing frequent weak events instead of the intense, organized squall lines that deliver much of West Africa’s rainfall. The study’s authors note this as a shared limitation rather than a scheme-specific failure, and it points toward the broader trajectory of the field: convection-permitting modeling, in which storms are simulated explicitly at kilometer-scale grid spacing, is increasingly seen as the path forward, and RegCM5 itself has already demonstrated such capability in European-wide experiments.

For Côte d’Ivoire, the practical stakes are considerable. The Bandama watershed feeds agriculture and water infrastructure across the country, and climate change is expected to intensify both drought stress and extreme rainfall in West Africa. Reliable regional projections require a model configuration that can be trusted, and this study provides exactly that kind of evidence-based guidance: favor Tiedtke or Emanuel convection with the UW boundary layer, treat Grell and unpaired Kain–Fritsch with skepticism, and account for the drizzle bias when interpreting simulated rainfall statistics. The work also carries a message for the global modeling community, demonstrating that sensitivity testing at the watershed scale, grounded in independent satellite observations and run on Ivorian supercomputing infrastructure at the Centre National de Calcul de Côte d’Ivoire, can deliver actionable results from the Global South that speak directly to international efforts to improve climate models for the regions that need them most.

Subject of Research: Sensitivity of the RegCM5 regional climate model's physical parameterization schemes for precipitation simulation over the Bandama watershed in Côte d'Ivoire

Article Title: Assessment of the sensitivity of the RegCM5 model to physical parameterization schemes in precipitation simulation in the Bandama watershed (Côte d’Ivoire)

Article References: Doumbia, M., Yoroba, F., Koné, B., Tiémoko, T. D., Kouassi, B. K., Kouadio, K., Diédhiou, A., Salifou, I. T., & Diawara, A. (2026). Assessment of the sensitivity of the RegCM5 model to physical parameterization schemes in precipitation simulation in the Bandama watershed (Côte d’Ivoire). Theoretical and Applied Climatology, 157(10), Article 653. https://doi.org/10.1007/s00704-026-06591-3

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06591-3

Keywords: RegCM5, regional climate modeling, precipitation simulation, convection schemes, Bandama watershed, Côte d'Ivoire, West African monsoon, parameterization, CHIRPS, IMERG, planetary boundary layer, climate model evaluation