Testing Kilometer-Scale Model Against Observations: Microphysical Insights from the ORCESTRA Campaign

The emergence of kilometer-scale atmospheric models represents more than just increased resolution: it offers three key advantages. First, by explicitly resolving deep convection, these models eliminate the convective parameterisation that previously masked deficiencies in other processes. This reduces the problem to a finite set of well-defined processes (microphysics, turbulence, radiation, surface interactions, and forcing data), each testable and improvable independently. Second, they enable direct comparison between models and observations by matching spatio-temporal sampling rates, avoiding artifacts from upscaling and time-averaging. This synergy allows direct assessment of the model's underlying physics against real-world observations. Finally, since regional and global models now share the same core physics, any insight gained from regional simulations directly translate to better global climate projections.

The ORCESTRA campaign over the tropical Atlantic (August-September 2024) provided such an opportunity for model development. We ran ICON in its Sapphire-configuration at 1.25 km resolution in parallel with field operations. Overlapping 48-hour simulations forced by IFS analyses generated high-frequency output matching observational sampling from space (EarthCARE satellite), the air (HALO plane and dropsonde) and the surface (METEOR ship and radiosondes). ICON captures large-scale circulation well, but with some important caveats: persistent atmospheric drying, insufficient upper-tropospheric ice clouds with weak humidity contrasts, undersized systems failing to organise into mesoscale clusters, and reduced surface wind variability. Critically, cloud radar measurements reveal an obvious microphysical flaw: rain falls twice as fast as observed. This excessive fall speed plausibly connects all biases through premature moisture depletion. Rapidly falling drops reach the surface before evaporating (weakening cold pools), before detraining moisture upward (reducing ice clouds), and before enabling mesoscale organisation.

Guided by these observations, we revised ICON's rain microphysics by 1) incorporating lognormal particle size distributions (Feingold & Levin, 1986), 2) evaluating fall velocities based on Van Boxel (1998) and, 3) consistently adjusting evaporation and accretion rates. Targeted reruns along EarthCARE overpasses show that this revision successfully re-aligns the Doppler velocities with observations and appreciably affect the model's representation of convective organisation. Overall, this work illustrates the synergy between kilometer-scale models and field measurements: by operating at observational scales and explicitly resolving convection, models can be physically interpreted and improved, with these refinements directly serving global climate projections.