In June 2023, a manuscript documenting the development and evaluation of a doubly-periodic (DP) configuration for the convection-permitting E3SM (SCREAM) was accepted for publication in the Journal of Modeling Earth Systems (doi: 10.1029/2022MS003466). This manuscript serves as the main reference for the Simple Convection-Permitting E3SM Atmosphere Model (SCREAM) when run in DP mode (henceforth, DP-SCREAM), and is also the first work to examine the horizontal resolution sensitivity for SCREAM.
- DP-SCREAM makes running the SCREAM code base at convection permitting resolutions accessible to all users and serves as an efficient framework to replicate and understand biases in the global model.
- Changing the horizontal resolution in DP-SCREAM is trivial, making it the ideal framework to study SCREAM’s resolution sensitivity.
- The representation of stratiform clouds in SCREAM is remarkably robust to changes in the horizontal resolution, however there is relatively large sensitivity for shallow convective clouds.
- SCREAM exhibits the behaviors of a scale aware model, therefore the resolution of SCREAM can be increased (either globally or RRM mode) without the need of modifying the parameterizations used.
The development of the high resolution (~3 km) E3SM model (Caldwell et al. 2021), known as SCREAM or next generation E3SM Atmosphere Model (EAMxx), is a monumental advance that avoids the uncertainties and biases often associated with lower resolution models. However, global convection permitting models are expensive to run and require computational resources that the common user simply cannot afford, even for relatively short simulations. Historically, general circulation models have supported “efficient configurations” known as single column models (SCMs; Bogenschutz et al. 2020). However, a SCM does not make sense for a global convection permitting model as clouds and convection are expected to be resolved over several grid columns. Therefore, a doubly-periodic configuration has been developed which allows users to run cases of the cloud/turbulence regime of their choice and makes running the SCREAM code base accessible to all.
The Doubly-Periodic SCREAM configuration represents a “single point” three-dimensional cloud resolving model configured on a planar grid (Fig 1). To run DP-SCREAM a user selects one of 30 cases from the E3SM Intensive Observational Period (IOP) library, which represents a diverse library ranging from deep convection, warm boundary layer cloud cases, arctic mixed-phase clouds, cold air outbreaks, etc. The majority of these cases are contributed from ARM field campaigns while the remainder are based on GCSS intercomparison studies and idealized cases. Each of these cases comes with a supported script that will run out-of-the box on any machine that E3SM can run on. The model is initialized in a similar manner to a SCM, where a single point is extracted from the global input files to determine the location; however in DP-SCREAM a domain of columns is generated. After initialization, these columns are subjected to random temperature perturbations to encourage heterogeneity and spin-up turbulence and convection.
All DP-SCREAM cases are configured to run with SCREAM’s default resolution of 3.25 km, with a domain size that is deemed appropriate for that case. However, a major advantage of DP-SCREAM is that it is trivial to change the domain size AND resolution; both of which are controlled by simple namelist settings. This is counter to changing the resolution in global E3SM/SCREAM runs, as adding a new resolution often requires time-consuming generation and testing of several necessary input files. This makes DP-SCREAM the ideal vehicle to examine resolution sensitivity.
Running DP-SCREAM requires negligible computational resources, when running at the default 3.25 km resolution. For example, running a boundary layer cloud case will take about 4 minutes of wall clock time on one computational node for one simulated day (timing is given on the now-retired Cori-knl), whereas a deep convection case (which requires a larger domain to capture organization) will take about 4 minutes wall clock on 6 nodes. These represent fairly meager expenses, especially when compared to the 5 hours on 1536 nodes it took global SCREAMv0 to simulate one day.
Horizontal Resolution Sensitivity
DP-SCREAM was run for five well established and diverse cases to understand how SCREAM responds to changing the horizontal resolution. These cases were run with horizontal grid sizes as fine as 100 m and as coarse as 5 km. We find that for deep convection SCREAM is remarkably scale insensitive for upper-level clouds (Fig. 2). This is significant because it means that as SCREAM’s resolution is changed globally, there will likely not be a need for time-consuming retuning that is often associated with these clouds. However, SCREAM does have a rather large resolution sensitivity associated with lower-level shallow convective clouds (Fig. 2 and bottom row of Fig. 3). It is found that SCREAM’s comparison with large eddy simulation (LES) and observations improves with increasing resolution for these clouds, thus suggesting that sub-grid scale (SGS) parameterization improvements are required to reduce this sensitivity. In general, however, SCREAM produces very similar results for resolutions ranging from 1 km to 5 km for convective clouds.
The insensitivity to horizontal resolution for stratiform clouds is a significant achievement for SCREAM, as demonstrated by the simulation of the DYCOMS-RF01 case (Stevens et al. 2005; top row Fig. 3). This case is often viewed as the gold standard case to test model performance for sub-tropical marine stratocumulus as most models have a tendency to produce too little cover. However, SCREAM has very good performance for DYCOMS-RF01 when compared to LES benchmarks and observations, not only for the default 3.25 km resolution but for all resolutions across the “gray zone” of turbulence.
Finally, it is found that SCREAM exhibits the behaviors of a scale aware model. This means that as SCREAM’s resolution increases towards 100 m, the SGS physics gradually shuts off and the resolved dynamics takes over. While preliminary results suggest that SCREAM may shut off its SGS physics a bit too abruptly when compared to LES filtering, this result is significant as it indicates that as SCREAM’s resolution increases (to at least 100 m) it will not be necessary to modify the model’s equation set or parameterization suite. Therefore, SCREAM appears to be well equipped to tackle even more ambitious resolutions either globally or with a regionally refined mesh (RRM).
When running DP-SCREAM users are encouraged to run with the most up-to-date SCREAM master branch. The E3SM IOP wiki provides support, scripts, and instruction to users wishing to run DP-SCREAM. The exact model code used to produce results in this paper is archived at Software for Doubly Periodic SCREAM Horizontal Resolution Sensitivity,, while the output data can be found at Data for Doubly Periodic SCREAM Horizontal Resolution Sensitivity.
- Peter A Bogenschutz, C Eldred, P M Caldwell. Horizontal Resolution Sensitivity of the Simple Convection-Permitting E3SM Atmosphere Model in a Doubly-Periodic Configuration. Journal of Advances In Modeling Earth System, 15, doi: 10.1029/2022MS003466, 2023.
- SCREAM featured on the cover of JAMES, Feb 2022
- Introduction to EAMxx and its Superior GPU Performance, Aug 2022
- Exascale Performance of the Simple Cloud Resolving E3SM Atmosphere Model, Feb 2023
- Peter Bogenschutz, Lawrence Livermore National Laboratory