Global Kilometer-Scale Simulations with ARP-GEM2: Effect of Parameterized Convection and Calibration

TL;DR

This work documents ARP-GEM2, a kilometer-scale global atmospheric framework with 72 vertical levels, calibrated for 2.6 km and 1.3 km resolutions and parallelized for high-performance computing. By comparing simulations with and without deep convection across resolutions, the study shows that parameterized convection remains influential at kilometer scales and that entrainment/detrainment tuning is crucial for balancing mean state accuracy and climate variability. The results demonstrate strong computational scalability, enabling multi-year to centennial-scale simulations at 2–3 km and highlighting that higher resolution can improve variability representation, though full removal of deep convection remains challenging. The findings provide practical guidance for high-resolution climate modeling, including calibration strategies, and support the viability of intercomparison efforts like DYAMOND at kilometer scales.

Abstract

The objective of this paper is twofold. First, it documents the second version of the global atmospheric model ARP-GEM and its calibration at kilometer-scale resolution. The model is currently able to run simulations at a resolution of up to 1.3 km. Second, this paper focus on multi-year global atmospheric simulations at a 2.6 km resolution with and without parameterized convection and associated calibration. Simulations without deep convection tend to be similar to those with infinite, or at least large, entrainment values. Consistently, entrainment and detrainment are used as primary drivers for the gradual reduction of convection as resolution increases. The results indicate that, with this hydrostatic model, parameterized convection still plays a significant role in the correct representation of the mean state at the kilometer scale. Additionally, they suggest some added value of high resolution in representing climate variability. However, a compromise between the adequate representation of the mean state and variability is necessary, as both are differently favored by the degree of parameterized convection. Finally, it is likely that even higher resolutions are necessary to achieve an unequivocal added value.

Figures (7)

• Figure 1: Simulated years per day (SYPD) normalized by the number of kilo-cores as a function of the cube of the inverse of the normalized grid spacing $(25/\Delta x)^3$ for ARP-GEM2 (72 vertical levels) configurations at 25 km (red), 12.6 km (orange), 6.3 km (green), 2.6 km (blue), and 1.3 km (violet) resolutions, and for ARP-GEM1 (50 vertical levels) configurations at resolutions from 25 km to 6.3 km (open circles). The line represents computational costs estimated by idealized scaling from the ARP-GEM2 25-km simulation.
• Figure 2: Annual normalized root-mean-square errors (RMSEs) in the climatology of precipitation (precip), top-of-atmosphere longwave (LW) and net shortwave (SW) radiation, total cloud cover (cloud), surface air temperature (temp), and 200-hPa zonal wind (u200). RMSE is normalized by the median value across 38 CMIP6 models (listed in Section 3 of the Supplementary Material of GS25). (a) RMSEs for ARP-GEM1 at 25-km (red circle) and 12-km (orange circle) resolutions, and ARP-GEM2 at 25-km (red dot), 12.6-km (orange dot), and 2.6-km (blue dot) resolutions, compared with the distribution of annual RMSEs for the 38 CMIP6 models during the period 2007–2009 (boxplot). (b) RMSEs for ARP-GEM2 at 2.6-km (blue dot) and 1.3-km (violet dot) resolutions for the year 2007, compared with the distribution of RMSEs for the 38 CMIP6 models (boxplot).
• Figure 3: Probability density functions (PDF) of daily mean precipitation (in mm/day) over the tropical domain ($20^{\circ}$S-$20^{\circ}$N) for IMERG and CMORPH datasets and for ARP-GEM2 simulations at 25 km (top), at 2.6 km (middle) and simulations with and without deep convection at 25, 12.6 and 2.6 km (bottom). Simulations are detailed in Table \ref{['tab:tuning']}. The period used is 2007-2009 for all datasets. Precipitation is conservatively interpolated to a to a 0.25$^{\circ}$$\times$ 0.25$^{\circ}$ grid. Left panels show low precipitation rates in the range [$10^{-1}$-$10^{2}$] mm day$^{-1}$, uniformly binned on a $\log_{10}$ scale (50 bins). Right panels show high precipitation rates, binned with a size of 5 mm day$^{-1}$.
• Figure 4: Annual mean of the OLR anomaly with respect to CERES observational dataset (period 2007-2009) for (a) ARP-GEM2-2.6km, for (b) ARP-GEM2-2.6km-ed$+$, (c) ARP-GEM2-2.6km-nodeep, and (d) ARP-GEM2-2.6km-nodeep-tun.
• Figure 5: Annual zonal mean of the specific humidity anomaly with respect to AIRS reanlysis dataset (period 2007-2009) for (a) ARP-GEM2-2.6km, for (b) ARP-GEM2-2.6km-ed$+$, (c) ARP-GEM2-2.6km-nodeep, and (d) ARP-GEM2-2.6km-nodeep-tun.