How Do Tropical Cyclones Directly Simulated in High-Resolution Climate Models Differ from Statistically Dynamically Generated Storms?

We compare the simulation of landfalling tropical cyclones (TCs) using two popular tools for studying TC climatology: high-resolution climate models (which directly simulate TCs that can be tracked in model output) and statistical–dynamical downscaling (SDD) models (which generate synthetic storms based on a model’s large-scale climatology). Objectively tracked TCs in data from the High-Resolution Model Intercomparison Project (HighResMIP) are compared with observed landfalls in the North Atlantic Ocean using the International Best Track Archive for Climate Stewardship and reanalysis storm tracks. Using an SDD TC model, we generate a parallel set of tracks with HighResMIP subdaily kinematic and monthly thermodynamic fields as forcings from the same climate simulations. We find that while SDD is advantageous in its low computational cost and improves upon several metrics relative to using global climate models (GCMs), SDD also introduces unphysical behaviors in TCs that are not observed in GCMs. Downscaling results in more uniform behavior across models but also overrepresents the proportion of landfalling storms, even when correcting for genesis location. This has important implications for use in risk assessment. SDD TC climatologies are more sensitive to the choice of model forcing than to the grid spacing of the model forcing. While each technique has distinct advantages and disadvantages, comparing them provides insights into the biases inherent in HighResMIP TC climatology. A simple metric for TC seed survival reveals that the mechanisms contributing to biases in TC climatology vary among HighResMIP models. An increased understanding of the strengths of these techniques is crucial for enhancing confidence in results of future studies.