Speed up a slow simulation

GEOS-Chem Classic performance is continuously monitored by the GEOS-Chem Support Team by means of benchmark simulations and ad-hoc timing tests. In this chapter, we provide some practical tips that you can use to speed up your simulations.

Use a coarser chemistry timestep

The table below contains our recommended GEOS-Chem Classic timestep settings.

GEOS-Chem Classic Resolution	Transport	Chemistry
\(4^{\circ}{\times}5^{\circ}\)	600s (10m)	1200s (20m)
\(2^{\circ}{\times}2.5^{\circ}\)	600s (10m)	1200s (20m)
\(0.5^{\circ}{\times}0.625^{\circ}\)	300s (5m)	600s (10m)
\(0.25^{\circ}{\times}0.3125^{\circ}\)	300s (5m)	600s (10m)
\(0.125^{\circ}{\times}0.15625^{\circ}\)	150s (2.5m)	300s (5m)

The Courant limit on the latitude-longitude grid constrains the choice of transport timestep for a given horizontal resolution. We choose a chemistry timestep that is double the transport timestep (i.e. Strang operator splitting).

If you wish to speed up your simulation, try increasing the chemistry timestep. Chemistry is the GEOS-Chem operation that takes the longest to execute, so increasing the interval between calls to the chemistry solver will reduce the run time accordingly. But you should also verify that the increased chemistry timestep allows your simulation to faithfully capture diurnal variations, etc.

See Philip et al. [2016] for a comprehensive study on GEOS-Chem timesteps.

Use the Rosenbrock solver with auto-reduction

The GEOS-Chem full-chemistry mechanism uses the KPP Rosenbrock solver, which has an automatic mechanism reduction option as described in Lin et al. [2023]. This automatic mechanism reduction feature separates species into “fast” and “slow” categories based on their chemical production or loss rates. “Fast” species are integrated with the full Rosenbrock algorithm, while “slow” species will have a simple 1st-order loss applied to them. This approach has been shown to reduce the time spent in chemistry by 20 to 30 percent.

The automatic mechanism reduction option is disabled by default, but can be enabled by toggiing this switch in geoschem_config.yml:

autoreduce_solver:
  activate: false  # <=== set to true to activate auto-reduction

Turn off unwanted diagnostics

Several diagnostics are turned on by default in the HISTORY.rc configuration file. The more diagnostics that are turned on, the more I/O operations need to be done, resulting in longer simulation execution times. Disabling diagnostics that you do not wish to archive can result in a faster simulation.

Disable debugging options

If you previously configured GEOS-Chem with the : CMAKE_BUILD_TYPE option set to Debug, then several run-time debugging checks will be activated. These include:

Checking for array-out-of-bounds errors
Checking for floating-point math exceptions (e.g. div-by-zero)
Disabling compiler optimizations

These options can be useful in detecting errors in your GEOS-Chem Classic simulation, but result in a much slower simulation. If you plan on running a long Classic simulation, make sure that you configure and build GEOS-Chem Classic so that CMAKE_BUILD_TYPE is set to Release.

Reduce the amount of files that need to be read

If you are developing a new data set (such as an emissions inventory) for GEOS-Chem, we recommend that you prepare data files with multiple timestamps rather than one file per timestamp. For example, if your data set has hourly time resolution, consider creating one file for each day, with each file containing 24 hours of data, etc.

The greatest amount of overhead in I/O occurs when new data files (in netCDF format) are opened. This also usually involves decompression of the file contents, which is computationally intensive. Reducing the number of times that GEOS-Chem has to open and close netCDF files can substantially improve performance.

Speeding up GEOS-Chem Classic nested-grid simulations

Use these tips to speed up your GEOS-Chem nested-grid simulations:

Crop nested-grid meteorology inputs

Your simulation should not read global high-resolution (\(0.5^{\circ}{\times}0.625^{\circ}\) or finer) meteorology fields. The overhead in reading and regridding these global fields can significantly impact your simulation. Instead, consider cropping high-resolution meteorology fields to the extent of your nested domain. This can easily be done with the netCDF operators or the Climate Data Operators; see our Work with netCDF files supplemental guide for more information.

Increase the size of the transport buffer zone

By default, nested buffer_zone_NSEW option is set to 3 boxes in each cardinal direction. Increasing this number will reduce the amount of grid boxes in which transport will be performed, which should also reduce the overall run time.