Enable SIPNET to write PEcAn standard netCDF files

[dlebauer]

• Write model outputs directly to PEcAn standard
• Keep legacy SIPNET output files optional
• Write tests
• Update documentation

Before closing:

• Update this issue with results of performance before and after changes; and with and without new feature turned on.

[avinxshKD]

can i work on this

[dlebauer]

@avinxshKD Thank you for your interest. But this isn't a well defined task - more of a reminder - also because it is both a large change and a lower priority, it doesnt make a very good first issue.

That said, if you are still interested, we would welcome help pushing this forward. Two first steps before getting started:

1. Check how easy it is to find documemtation of the PEcAn standard for model output. Suggest improvements (if any are needed) to the documentation.
2. Scope out this issue- suggest the approach you would use and request feedback.

Then it will be easier for all involved to reassess the feasibility and effort involved before proceeding.

Thanks!

[dlebauer]

@avinxshKD Just to follow up - I don't mean to discourage you from taking on the work, just wanted to point out that it was under specified and would require some initial work to define requirements.

[ashiklom]

My two cents: After thinking about this some more, I might advise against doing this, at least as a default. This is mostly because reliably writing to NetCDF requires the NetCDF C library, which is a moderately heavy dependency that would make SIPNET significantly harder to install and test.

Also, conceptually, having SIPNET output dependent on the PEcAn standard means that if we ever update the PEcAn standard, we have to update code in at least 1 more repository. More generally, I think it's better to stick firmly to the current abstraction of "Models do whatever they want, and PEcAn post processes their output to its own standard".

If we do want to have this capability, I would keep it optional --- it should be possible to compile and test SIPNET without NetCDF.

[dlebauer]

@ashiklom your point about keeping netcdf optional and writing to text as default is a good one.

I take your point about allowing the model to do what it wants. I can see the argument for keeping the variable names and units consistent across text and netcdf, and changing variable names and units is probably a relatively small computational cost compared to the status quo of txt-->netcdf. But I think what SIPNET really wants is the ability to optimize for PEcAn by writing PEcAn standard netcdf.

[infotroph]

what SIPNET really wants is the ability to optimize for PEcAn

To me one of the key lessons of PecanProject/pecan#3925 is that when we want to optimize we need to be really sure we're attacking the real bottleneck. It's possible that writing directly to Sipnet to single-site netCDF and then combining those into a multi-site netCDF will be faster than combining many single-site plaintext outputs into one NetCDF in one single-shot conversion, but I think that's far from proven.