search

eco4cast / unconf-2023

Brainstorming repo to propose and discuss unconference project ideas!
12 stars 0 forks source link

Terrestrial forecast challenge office hours / code sprint #16

Open mdietze opened 1 year ago

mdietze commented 1 year ago

In the interest of getting more models into the Terrestrial forecast challenge, I think it would be fun to have some dedicated time to helping folks get models running for this challenge, similar to the success the Aquatics challenge has had at recent conferences (e.g. GLEON).

I can share the simple model that I use at the Flux Course, which runs in Github Actions, that anyone could build on and modify, even if you don't come with your own land model or want to eschew land models in favor of statistical or machine learning approaches.

I can also share the more complex workflow we use in PEcAn. PEcAn already supports ~20 models but only 1 of these has been submitting to the challenge.

Could see subsuming #15 (CLM forecasts) into this effort as one example model that we want to get forecasting.

mdietze commented 1 year ago

Wanted to follow up to give some additional detail about what this project would consist of. Here's the Flux Course model that I envisioning using at a jumping off point https://github.com/mdietze/FluxCourseForecast

Some options for things this group might do:

  1. Get additional models submitting to the Terrestrial NEON forecast challenge. I think this is a particularly good option if any participants have a model they already work with
  2. Continue to refine/improve the tutorial model and its documentation to make it easier for independent teams to use as a template (similar to how https://github.com/eco4cast/neon4cast-example has facilitated getting many more aquatic models in). For example, the existing code only runs one site.
  3. Create forks that improve the underlying simple model to produce better forecasts
  4. Create additional guidelines to help terrestrial teams engage in the challenge (e.g. how to get process-based models Dockerized and running in Github Actions)
  5. Extend Flux Course model beyond NEON sites (e.g. TERN, Ameriflux)
  6. Add more data constraints to the existing example (currently just uses tower fluxes, could add a wide range of remote sensing [e.g. MODIS LAI] or NEON field-based constraints)