A successful GCHP run produces three categories of output files: diagnostics, restarts (also called checkpoints), and logs. Diagnostic and restart files are always in netCDF4 format, and logs are always ascii viewable with any text editor. Diagnostic files are output to the
OutputDir directory in the run directory. The end-of-run restart file is output to the
Restarts directory. All other output files, including periodic checkpoints if enabled, are saved to the main level of the run directory.
It is important to be aware that GCHP 3D data files in this version of GCHP have two different vertical dimension conventions. Restart files and Emissions diagnostic files are defined with top-of-atmospheric level equal to 1. All other data files, meaning all diagnostic files that are not Emissions collections, are defined with surface level equal to 1. This means files may be vertically flipped relative to each other. This should be taken into account when doing data visualization and analysis using these files.
Below is a summary of all GCHP output files that you may encounter depending on your run directory configuration.
Standard output log file of GCHP, including both GEOS-Chem and HEMCO. The date in the filename is the start date of the simulation. Using this file is technically optional since it appears only in the run script. However, the advantage of sending GCHP standard output to this file is that the logs of consecutive runs will not be over-written due to the date in the filename. Note that the file contains HEMCO log information as well as GEOS-Chem. Unlike in GEOS-Chem Classic there is no
If you use a job scheduler to submit GCHP as a batch job then you will have a job log file. This file will contain output from your job script unless sent to a different file. If your run crashes then the MPI error messages and error traceback will also appear in this file.
GCHP logging output based on configuration in
logging.yml. Treat this file as a debugging tool to help diagnose problems in MAPL, particularly the ExtData component of the model which handles input reading and regridding.
Log file for advection. It includes information such as the domain stack size, stretched grid factors, and FV3 parameters used in the run.
This file is both input and output. As an input file it contains the simulation start date. After a successful run the content of the file is updated to the simulation end date. As an output file it is therefore the input file for the next run if running GCHP simulations consecutively in time.
GCHP restart file output at the end of the run. This file is actually the GCHP end-of-run checkpoint file that is moved and renamed as part of the run script. Unless including the code to do that in your run script you will instead get
gcchem_internal_checkpointin the main run directory. Moving and renaming is a better option because (1) it includes the datetime to prevent overwriting upon consecutive runs, (2) it enables using the
gchp_restart.nc4symbolic link in the main run directory to automatically point to the correct restart file based on start date and grid resolution, and (3) it minimizes clutter in the run directory. Please note that the vertical level dimension in all GCHP restart files is positive down, meaning level 1 is top-of-atmosphere.
Optional restart files output mid-run. In order to generate these you must configure the run directory to output with a specific frequency that is less than the duration of your run. Note that unlike the end-of-run restart file, these files are not copied to
Restartsin your run script and are not renamed.
GCHP diagnostic data files. Each file contains the collection name configured in
HISTORY.rcand the datetime of the first data in the file. For time-averaged data files the datetime is the start of the averaging period. Please note that the vertical level dimension in GCHP diagnostics files is collection-dependent. Data are positive down, meaning level 1 is top-of-atmosphere, for the Emissions collection. All other collections are positive up, meaning level 1 is surface.
Summary of settings in
This file is empty and can be ignored. It is an artifact of the MAPL software used in GCHP.
This file is empty and can be ignored. It is an artifact of configuration in
Memory statistics are printed to the GCHP log each model timestep. As discussed in the run directory configuration section of this user guide, this includes percentage of memory committed, percentage of memory used, total used memory (MB), and total swap memory (MB) by default.
To inspect the memory usage of GCHP you can grep the output log file for string
Mem/Swap. For example,
grep "Date:|Mem/Swap" gchp.log. The end of the line containing date and time shows memory committed and used. For example,
42.8% : 40.4% Mem Comm:Used indicates 42.8% of memory available is committed and 40.4% of memory is actually used. The total memory used is in the next line, for example
Mem/Swap Used (MB) at MAPL_Cap:TimeLoop= 1.104E+05 0.000E+00. The first value is the total memory used in MB, and the second line is swap (virtual) memory used. In this example GCHP is using around 110 gigabytes of memory with zero swap.
These memory statistics are useful for assessing how much memory GCHP is using and whether the memory usage grows over time. If the memory usage goes up throughout a run then it is an indication of a memory leak in the model. The memory debugging option is useful for isolating the memory leak by determining if there if it is in GEOS-Chem or advection.
Timing of GCHP components is done using MAPL timers. A summary of all timing is printed to the GCHP log at the end of a run. Configuring timers from the run directory is not currently possible but will be an option in a future version. Until then a complete summary of timing will always be printed to the end of the log for a successful GCHP run. You can use this information to help diagnose timing issues in the model, such as extra slow file read due to system problems.
The timing output written by MAPL is somewhat cryptic but you can use this guide to decipher it. Timing is broken in up into several sections.
GCHPctmEnv, the environment component that facilitates exchange between GEOS-Chem and FV3 advection
GCHPchem, the GEOS-Chem component containing chemistry, mixing, convection, emissions and deposition
DYNAMICS, the FV3 advection component
GCHP, the parent component of GCHPctmEnv, GCHPchem, and DYNAMICS, and sibling component to HIST and EXTDATA
HIST, the MAPL History component for writing diagnostics
EXTDATA, the MAPL ExtData component for processing inputs, including reading and regridding
Total model and MPI communicator run times broken into user, system, and total times
Full summary of all major model components, including core routines SetService, Initialize, Run, and Finalize
Model throughput in units of days per day
Each of the six gridded component sections contains two sub-sections. The first subsection shows timing statistics for core gridded component processes and their child functions. These statistics include number of execution cycles as well as inclusive and exclusive total time and percent time.
Inclusive refers to the time spent in that function including called child functions.
Exclusive refers to the time spent in that function excluding called child functions.
The second subsection shows from left to right minimum, mean, and maximum processor times for the gridded component and its MAPL timers. If you are interested in timing for a specific part of GEOS-Chem then use the timers in this section for
GCHPchem, specifically the ones that start with prefix
GC_. For chemistry you should look at timer
GC_CHEM which includes the calls to compute overhead ozone, set H2O, and calling the chemistry driver routine.
Beware that the timers can be difficult to interpret because the component times do not always add up to the total run time. This is likely due to load imbalance where processors wait (timed in MAPL) while other processors complete (timed in other processes). You can get a sense of how large the wait time is by comparing the
Exclusive time to the
Inclusive time. If the former is smaller than the latter then the bulk of time is spent in a sub-process and the
Exclusive time may be at least partially due to wait time.
If you are interested in changing the definitions of GCHP timers, or adding a new one, you will need to edit the source code. Toggling
GC_ timers on and off are mostly in file
geos-chem/Interfaces/GCHP/gchp_chunk_mod.F90, but also in
geos-chem/Interfaces/GCHP/Chem_GridCompMod.F90, using MAPL subroutines
MAPL_TimerOff. When in doubt about what a timer is measuring it is best to check the source code to see what calls it is wrapping.