search

GCEL / ILAMB

(This is the GCEL development fork of ILAMB - it is not intended to replace the ornl bitbucket site which should be referred to first.)
2 stars 1 forks source link

Uncertainty stippling plots #22

Open dvalters opened 5 years ago

dvalters commented 5 years ago

Based on discussion at meeting, priority is to be able to produce uncertainty plots for ILAMB based on a stippling overlay over spatial outputs. Stippling uncertainty ranges would come from the CARDAMOM dataset.

We would need to make sure this is implemented in a general form in ILAMB, so that other datasets could be used that also had a source of uncertainty attached to them.

I.e. in the config file, you could specify an uncertainty variable(s) in the input netCDF file:

[CARDAMOM]
source = "blah/foo/caradamom.nc"
uncertainty = gpp_uncertainty_75pc

For example in the Exbrayat et al (2018) paper:

https://www.earth-syst-dynam.net/9/153/2018/esd-9-153-2018.pdf

In this case _"Stippling indicates regions where there is an agreement on the sign of DeltaNPP through the uncertainty." (Fig. 3, Exbrayat 2018), but we could adapt the definition of what the stippling means.

ratio-of-the-uncertainty-from-each-rea-to-the-uncertainty-in-the-isi-mip-ensemble-for_w640

dvalters commented 5 years ago

It is worth having a read of the Wilks (2017) paper (about statistical stippling!) before implementing this: https://journals.ametsoc.org/doi/10.1175/BAMS-D-15-00267.1

dvalters commented 5 years ago

Stippling proof of concept in matplotlib:

poc_stipple

dvalters commented 5 years ago

To clarify: the stippling would show disagreement/agreement (agreement more intuitive?) where the variable was outside the XXth percentile confidence interval, for example. I.e. if say the RT modelled value was within the 95th percentile of the CARDAMOM output, stipple that pixel, etc. User could choose confidence interval if multiple are provided in the dataset. (Currently only Residence Time exists in the sample dataset.)