Depth-less netCDF fields about parcels HOT 6 CLOSED

Good point, @Jacketless

Similarly, at some point we will probably also encounter time-less netCDF files (think spatial maps of diffusivity).

This needs some thinking about how general we want our netcdf files to be.

from parcels.

Ok, I think ultimately we want a hierarchy of specialised Field sub-classes to encapsulate our internal data layout, similar to the set of grid classes discussed in issue #39. I would think something along the lines of Field2D, Field2DVertical, Field2DTime and Field2DVerticalTime with a constructor hierarchy, appropriate from_netcdf() initialisers and cached interpolator routines should be sufficient. Once we have that we can even write smart wrapper functions that detect data layout from file, but let's first get the data structures right.

from parcels.

Yes, I totally agree. For completeness, don't forget the Field0DVertical class of column models.

Also, we would ideally also accomodate time-evolving grids. The adaptive horizontal meshes are one type, but there are also quite few ocean models that have temporally-varying vertical grids (i.e. horizontal surfaces go up and down over time). Something to remember as we're defining these classes?

from parcels.

Ok, that sounds like a fairly intuitive way to build in this functionality.

To get things working in the short window I have in France, I've hacked my
netCDFs using NCO to have the appropriate form, but I'll try and write some
of these example subclasses as I go and you adjust.

@erikvansebille now you describe it in that way, I realise that I am
already trying to use time-less fields with tuna biomass densities that I'm
using as 'initial conditions' for the distribution of particles. In this
case, there is an arbitrary time dimension of length one.

On 3 March 2016 at 19:28, Erik van Sebille [email protected] wrote:

Yes, I totally agree. For completeness, don't forget the Field0DVertical
class of column models.

Also, we would ideally also accomodate time-evolving grids. The adaptive
horizontal meshes are one type, but there are also quite few ocean models
that have temporally-varying vertical grids (i.e. horizontal surfaces go up
and down over time). Something to remember as we're defining these classes?

—
Reply to this email directly or view it on GitHub
#41 (comment)
.

from parcels.

I have created a new script in the landmass_kernel branch, related to this topic. I was hoping to create a kernel capable of determining a particle's distance from the coast, for which I would require a grid of values that are 'land'. Such a grid doesn't have a time dimension, whereas field.py dictates that this must be that case.

from parcels.

Completed implementation in #184

from parcels.