Abstract Convection‐permitting regional climate models (CP‐RCMs) often outperform coarser‐resolution models in simulating precipitation, but assessing their accuracy across resolutions and configurations remains challenging. This study evaluates the performance and sensitivity of different configurations of the sixth version of the Canadian Regional Climate Model in representing hourly precipitation over northeastern North America using a common 2‐year period (2016–2017). The configurations differ in horizontal grid spacing (2.5 or 12 km), condensation/microphysics schemes (Sundqvist or P3), land surface models (ISBA or CLASS), deep convection treatment (Kain‐Fritsch or explicit), driving strategy, and atmospheric model version (GEM5.0.2 or GEM5.1.1). The evaluation is based on radar‐ and satellite‐based precipitation data combined with the vertical velocity and the integrated water vapor from the ERA5 reanalysis. Precipitation errors are analyzed using the Environmentally Conditioned Intensity‐Frequency (ECIF) decomposition, which relates precipitation characteristics (frequency and intensity) to their environmental (dynamical and thermodynamical) conditions. Results indicate that all ECIF error components strongly interact, with notable compensations. Results indicate that grid spacing is the dominant driver of precipitation biases, with the 2.5‐km configuration showing the greatest improvements, particularly by reducing the overestimation of precipitation frequency seen in 12‐km simulations. Microphysics schemes exert the second‐largest influence. GEM version and driving strategies play an intermediate role, while land surface and deep convection schemes have minimal impact. Seasonal variations are substantial, with summer exhibiting the strongest sensitivity to configuration changes.
Lahaie et al. (Fri,) studied this question.