Impacts of stochastic parameterizations on forecast error growth and predictability in convection‐permitting simulations over the Maritime Continent

This is the authors' abstract. We don't add key points for this paper.

This study investigates how different stochastic parameterizations influence forecast error growth in convection‐permitting simulations over the tropical Maritime Continent. First, we introduce perturbations to boundary layer, microphysics, and convective schemes to identify the dominant sources of uncertainty in high‐resolution tropical forecasts. A clear hierarchy in error growth is found: boundary‐layer perturbations produce the strongest and fastest growth, microphysical perturbations yield intermediate growth, and convective perturbations the weakest. Independent of perturbation source, the regions of largest forecast uncertainty align with areas of active convection, highlighting the strong coupling between diabatic processes and error amplification. Second, we examine sensitivity to perturbation amplitude, temporal decorrelation time‐scale, spatial correlation length scale, and tendency versus parameter‐based formulations. Results show that perturbation amplitude is the dominant control on forecast error growth for both formulations, while temporal decorrelation time‐scale plays a secondary role. Notably, this sensitivity is particularly pronounced for tendency perturbations, which directly project onto dynamically growing modes, and weaker for parameter perturbations, whose impacts are mediated indirectly through the physical processes. Overall, this study provides guidance for the design and tuning of stochastic parameterizations in tropical ensemble forecasting systems, with boundary‐layer processes identified as the dominant source of subgrid‐scale model uncertainty. While physical consistency considerations remain important, this study indicates that this uncertainty is most effectively represented by applying perturbations to tendencies rather than model parameters, except at early lead times when weak turbulent activity limits subgrid tendencies.