Quick Start Guide

Analysis for simulations produced with Model for Prediction Across Scales (MPAS) components and the Energy Exascale Earth System Model (E3SM), which used those components.

sea surface temperature

Installation

MPAS-Analysis is available as an anaconda package via the conda-forge channel:

conda config --add channels conda-forge
conda create -n mpas-analysis mpas-analysis
conda activate mpas-analysis

To use the latest version for developers, you will need to set up a conda environment with the following packages:

  • numpy

  • scipy

  • matplotlib >= 3.0.2

  • netCDF4

  • xarray >= 0.10.0

  • dask

  • bottleneck

  • basemap

  • lxml

  • nco >= 4.8.1

  • pyproj

  • pillow

  • cmocean

  • progressbar2

  • requests

  • setuptools

  • shapely

  • cartopy

  • geometric_features

These can be installed via the conda command:

conda config --add channels conda-forge
conda create -n mpas-analysis numpy scipy matplotlib netCDF4 \
    xarray dask bottleneck basemap lxml nco>=4.8.1 pyproj pillow \
    cmocean progressbar2 requests setuptools shapely cartopy \
    geometric_features
conda activate mpas-analysis
Then, get the code from:

https://github.com/MPAS-Dev/MPAS-Analysis

Download analysis input data

If you installed the mpas-analysis package, download the data that is necessary to MPAS-Analysis by running:

download_analysis_data -o /path/to/mpas_analysis/diagnostics

If you are using the git repository, run:

./download_analysis_data.py -o /path/to/mpas_analysis/diagnostics

where /path/to/mpas_analysis/diagnostics is the main folder that will contain two subdirectories:

  • mpas_analysis, which includes mapping and region mask files for standard resolution MPAS meshes

  • observations, which includes the pre-processed observations listed in the Observations table and used to evaluate the model results

Once you have downloaded the analysis data, you will point to its location (your equivalent of path/to/mpas_analysis/diagnostics above) in the config option baseDirectory in the [diagnostics] section.

Download Natural Earth data for cartopy

The cartopy package (used for creating inset maps) requires shapes of the land, ocean and coastline from Natural Earth. Typically, these data are downloaded automatically by cartopy. However, for systems with compute nodes that cannot reach the internet, you will need to download the data manually into your conda environment from a login node before launching any MPAS-Analysis jobs:

download_natural_earth_110m

(or if using the git repo: ./download_natural_earth_110m.py).

If the data have already been downloaded, you will see nothing. Otherwise, you should see a warning that the data are being downloaded.

Note: If you are having issues downloading the shape files (e.g., a time out error or forbidden error), follow these steps:

  1. Run the following in python on your local machine (i.e., one that has no trouble downloading these files): .. code-block:: python

    import cartopy.io.shapereader as shpreader for name in [‘ocean’, ‘coastline’, ‘land’]:

    shpfilename = shpreader.natural_earth(resolution=’110m’,

    category=’physical’, name=name)

    reader = shpreader.Reader(shpfilename)

  2. On your local machine, run python -c "import cartopy; print(cartopy.config['data_dir'])". This will print out the directory in which the natural earth shapefiles are being placed locally.

  3. Copy these files onto the remote machine you are working on. Include folders shapefiles/natural_earth/physical/* where * is the set of shapefiles that were downloaded.

  4. On your remote machine, run python -c "import cartopy; print(cartopy.config['data_dir'])". Copy the shapefiles folder and all contents over to this location.

  5. cartopy should now be able to find these files for MPAS-Analysis.`

List Analysis

If you installed the mpas-analysis package, list the available analysis tasks by running:

mpas_analysis --list

If using a git repository, run:

python -m mpas_analysis --list

This lists all tasks and their tags. These can be used in the generate command-line option or config option. See mpas_analysis/config.default for more details.

Running the analysis

  1. Create and empty config file (say config.myrun), copy config.example, or copy one of the example files in the configs directory (if using a git repo) or download one from the example configs directory.

  2. Either modify config options in your new file or copy and modify config options from mpas_analysis/config.default (in a git repo) or directly from GitHub: config.default.

  3. If you installed the mpas-analysis package, run: mpas_analysis config.myrun. If using a git checkout, run: python -m mpas_analysis config.myrun. This will read the configuraiton first from mpas_analysis/config.default and then replace that configuraiton with any changes from from config.myrun

  4. If you want to run a subset of the analysis, you can either set the generate option under [output] in your config file or use the --generate flag on the command line. See the comments in mpas_analysis/config.default for more details on this option.

    Requirements for custom config files:

  • At minimum you should set baseDirectory under [output] to the folder where output is stored. NOTE this value should be a unique directory for each run being analyzed. If multiple runs are analyzed in the same directory, cached results from a previous analysis will not be updated correctly.

  • Any options you copy into the config file must include the appropriate section header (e.g. ‘[run]’ or ‘[output]’)

  • You do not need to copy all options from mpas_analysis/config.default. This file will automatically be used for any options you do not include in your custom config file.

  • You should not modify mpas_analysis/config.default directly.

List of MPAS output files that are needed by MPAS-Analysis:

  • mpas-o files:

    • mpaso.hist.am.timeSeriesStatsMonthly.*.nc (Note: since OHC anomalies are computed wrt the first year of the simulation, if OHC diagnostics is activated, the analysis will need the first full year of mpaso.hist.am.timeSeriesStatsMonthly.*.nc files, no matter what [timeSeries]/startYear and [timeSeries]/endYear are. This is especially important to know if short term archiving is used in the run to analyze: in that case, set [input]/runSubdirectory, [input]/oceanHistorySubdirectory and [input]/seaIceHistorySubdirectory to the appropriate run and archive directories and choose [timeSeries]/startYear and [timeSeries]/endYear to include only data that have been short-term archived).

    • mpaso.hist.am.meridionalHeatTransport.0001-03-01.nc (or any hist.am.meridionalHeatTransport file)

    • mpaso.rst.0002-01-01_00000.nc (or any other mpas-o restart file)

    • streams.ocean

    • mpaso_in

  • mpas-seaice files:

    • mpasseaice.hist.am.timeSeriesStatsMonthly.*.nc

    • mpasseaice.rst.0002-01-01_00000.nc (or any other mpas-seaice restart file)

    • streams.seaice

    • mpassi_in

Note: for older runs, mpas-seaice files will be named:

  • mpascice.hist.am.timeSeriesStatsMonthly.*.nc

  • mpascice.rst.0002-01-01_00000.nc

  • streams.cice

  • mpas-cice_in Also, for older runs mpaso-in will be named:

  • mpas-o_in

Purge Old Analysis

To purge old analysis (delete the whole output directory) before running run the analysis, add the --purge flag. If you installed mpas-analysis as a package, run:

mpas_analysis --purge <config.file>

If you are running in the repo, use:

python -m mpas_analysis --purge <config.file>

All of the subdirectories listed in output will be deleted along with the climatology subdirectories in oceanObservations and seaIceObservations.

It is a good policy to use the purge flag for most changes to the config file, for example, updating the start and/or end years of climatologies (and sometimes time series), changing the resolution of a comparison grid, renaming the run, changing the seasons over which climatologies are computed for a given task, updating the code to the latest version.

Cases where it is reasonable not to purge would be, for example, changing options that only affect plotting (color map, ticks, ranges, font sizes, etc.), rerunning with a different set of tasks specified by the generate option (though this will often cause climatologies to be re-computed with new variables and may not save time compared with purging), generating only the final website with --html_only, and re-running after the simulation has progressed to extend time series (however, not recommended for changing the bounds on climatologies, see above).

Running in parallel via a queueing system

If you are running from a git repo:

  1. If you are running from a git repo, copy the appropriate job script file from configs/<machine_name> to the root directory (or another directory if preferred). The default cript, configs/job_script.default.bash, is appropriate for a laptop or desktop computer with multiple cores.

  2. If using the mpas-analysis conda package, download the job script and/or sample config file from the example configs directory.

  3. Modify the number of parallel tasks, the run name, the output directory and the path to the config file for the run.

  4. Note: the number of parallel tasks can be anything between 1 and the number of analysis tasks to be performed. If there are more tasks than parallel tasks, later tasks will simply wait until earlier tasks have finished.

  5. Submit the job using the modified job script

If a job script for your machine is not available, try modifying the default job script in configs/job_script.default.bash or one of the job scripts for another machine to fit your needs.

Instructions for creating a new analysis task

Analysis tasks can be found in a directory corresponding to each component, e.g., mpas_analysis/ocean for MPAS-Ocean. Shared functionality is contained within the mpas_analysis/shared directory.

  1. create a new task by copying mpas_analysis/analysis_task_template.py to the appropriate folder (ocean, sea_ice, etc.) and modifying it as described in the template. Take a look at mpas_analysis/shared/analysis_task.py for additional guidance.

  2. note, no changes need to be made to mpas_analysis/shared/analysis_task.py

  3. modify mpas_analysis/config.default (and possibly any machine-specific config files in configs/<machine>)

  4. import new analysis task in mpas_analysis/<component>/__init__.py

  5. add new analysis task to mpas_analysis/__main__.py under build_analysis_list, see below.

A new analysis task can be added with:

analyses.append(<component>.MyTask(config, myArg='argValue'))

This will add a new object of the MyTask class to a list of analysis tasks created in build_analysis_list. Later on in run_analysis, it will first go through the list to make sure each task needs to be generated (by calling check_generate, which is defined in AnalysisTask), then, will call setup_and_check on each task (to make sure the appropriate AM is on and files are present), and will finally call run on each task that is to be generated and is set up properly.

Generating Documentation

To generate the sphinx documentation, run:

conda env create -f docs/environment.yml
conda activate mpas-analysis-docs
python setup.py install
cd docs
make html