PyGChem

A tentative brigde between GEOS-Chem and the Python Scientific (SciPy) Stack.

Benoît Bovy -- GIRPAS (AGO) Université de Liège

bbovy@ulg.ac.be | https://github.com/benbovy

SEAS - Harvard (September 2014)

About myself...

• PhD in geosciences (geomorphology)
  • Numerical modelling of erosion processes (C/Python dev)
• Since 2012 at the InfraRed Group of Atmospheric and Solar Physics (GIRPAS)
  • At the same time, the group joined the GC community
• Part of my job:
  • Set up a computing environment for the modelling tasks
  • Facilitating the integration and the use of various softwares (including GC)

**DISCLAIMER**:

• I don't have much experience in running GC simulations (so far)
• I'm not a specialist in the field of atmospheric chemistry (so far)
• I use Python (and its scientific libraries) a lot in my research

Interacting with GEOS-Chem

GC user-interface: for what purposes?

• Pre/post processing, format conversion, output analysis

• Visualization

• Run / manage GC simulations

• Model development: prototyping, testing

• Interfacing with other codes (models, libraries, applications...)

• ...

GC user-interface: what matters (ideally)?

• Interactivity interactive

• Richness of the computing environment features

• Ease-of-use, less error-prone user-friendly

• Public availability availability

• Hardware / Platform availability availability

• Interaction with the GC community collaboration

• Reproducible experiments reproducibility

• Reliability maturity

• Dealing with big amounts of data parallel

GC user-interface: Existing tools

Referenced on the GEOS-Chem wiki:

• IDL: GAMAP maturity features user-friendly availability
• NCL: NC4GC
• R: a few scripts for dataset analysis
• Python: a few packages (mainly for reading bpch files)

And also:

• IDL, R, Matlab, Python, etc...: GC community researchers's own recipes reproducibility collaboration

The Python Scientific (SciPy) Stack

The Python language (and the CPython implementation)

• Open-source, cross-platform availability

• Language inherent characteristics user-friendly interactive
  • Readability, introspection, multi-paradigm...

• Integrates well with other languages (C/C++, Fortran...)

• Includes natively a huge number of libraries features
  • OS, network, web, testing, debugging, profiling, file formats...

• Very active, growing scientific community features collaboration

The SciPy Stack

The SciPy Stack

• Dependencies user-friendly ?? reproducibility ??
  • but recent solutions...

• A lot of new packages in development maturity ??

Packages specific to the atmospheric sciences community

• UV-CDAT (http://uvcdat.llnl.gov/)
  • Software bundle (visualization and data analysis)

• Iris (http://scitools.org.uk/iris/)
  • Data abstraction layer based on the CF-Conventions

• cf-python (http://cfpython.bitbucket.org/)
  • Similar to Iris

• ESMPy (https://www.earthsystemcog.org/projects/esmpy/)
  • Python interface to ESMF (in development)

• OpenClimateGis (https://www.earthsystemcog.org/projects/openclimategis/)
  • Geospatial manipulation of climate datasets

PyGChem

• Avoid reinventing the wheel: "just" connect GEOS-Chem to the SciPy Stack (in a "pythonic fashion")

• Provide a consistent, full data model (and its implementation) for GEOS-Chem outputs and inputs
  • not only for datasets (HEMCO settings, globchem, input file...)

• Provide access to GEOS-Chem components from within Python ?
  • prototyping, tests...

Demo

Let's start

On the roadmap

• Needs more testing (unit tests)
• Compatibility with Python 3.x
• Datasets
  • CF-Conventions vs. COARDS Conventions ?
  • Handling datasets with CDAT (cdms), cf-python, ESMPy ?
• Chemistry (KPP format)
• Input parameters, managing GC runs

• Suggestions, pull requests and/or issue reports are welcome!!

https://github.com/benbovy/PyGChem