[MITgcm-support] Off line particle tracking with MITgcm

[Andrea Cimatoribus]

A final message on this for future reference.

The simple particle tracking code I developed, partly deriving it form 
tracmass, is freely available at:

  https://c4science.ch/diffusion/3351/

The code does 3D particle tracking (off line) based on MITgcm results in 
cartesian and curvilinear grids. It runs in parallel (multithreaded), 
and stores the results in netcdf4.

For those interested, the code is written in python, with numerical 
parts in C via cython. I am testing it and can reproduce tracmass 
results (apart from some bugs I spotted in the 3D tracmass 
computations). Bugs may still be present in particular in untested parts 
of the code (e.g. cartesian grids), but I would be surprised if there 
were any major one. I/O could probably be improved in terms of 
efficiency, and is not fully tested, but it seems to deal fine with 
O(10^5) particle at least.

If you ever use the code, please let me know. Any other feedback is of 
course welcome too.

Best,


Andrea Cimatoribus
postdoctoral researcher
EPFL ENAC IIE ECOL
https://people.epfl.ch/andrea.cimatoribus

On 21/04/17 09:19, Andrea Cimatoribus wrote:
> Dear all,
> as a follow up to my previous messages, and given the relatively strong
> interest in the topic I see lately, I thought I should somehow
> "advertise" the work I am doing on the tracmass code
> (http://tracmass.org/) by Döös et al.
>
> I am working on a fork of the original code, here:
>
> https://github.com/sambarluc/tracmass
>
> (sooner or later it will have to be moved to a new repo on the
> institutional git server, but for now it's there)
>
> I maintained the same numerical core, and focused on:
> - enabling a more straightforward use of the MITgcm cartesian and
> curvilinear grids (done and relatively well tested on my MITgmc results)
> - simplifying the treatment of the time axis (I basically removed all
> calendar functionality, since I found it overly complicated for my needs
> and error prone, I am planning to deal with that via python, but haven't
> written anything yet)
> - providing netcdf output (works nicely, with minimal overhead)
> - adding a python interface to run the code and load the results (works,
> but not yet well tested and far from complete in terms of running
> tracmass from python)
>
> I want to stress that the code is still mostly the original by Döös et
> al. Still, I think it can provide some advantages to MITgcm and python
> users now. I should also stress that only parts of it have been tested.
> For instance, I never used any of the turbulence parameterizations - and
> doubt I will ever -, the "analytical time stepping" works ok only for 2D
> advection, for 3D I think numerical errors are an issue but did not
> investigate any further. In any case, since I am approaching the stage
> at which "it works fine for what I need now," I thought it was good to
> share with the world. Maybe others are interested, are willing to test,
> find bugs, contribute, etc.
>
> Best, Andrea
>