0.2.0 release: PARIS updates
Past due by 4 months
100% complete
This release will contain the following changes based on work on the PARIS project:
- "Minimum model error" is implemented as a true minimum, rather than a constant addition to the likelihood's uncertainty
- The model error from scaled pollution events can use pollution event size calculated as (obs Y) - (baseline H_bc * xbc), or as (modelled pollution event…
This release will contain the following changes based on work on the PARIS project:
- "Minimum model error" is implemented as a true minimum, rather than a constant addition to the likelihood's uncertainty
- The model error from scaled pollution events can use pollution event size calculated as (obs Y) - (baseline H_bc * xbc), or as (modelled pollution event H * x)
- Filtering updates:
- PBLH filter now filters obs whenever the PBL is below (inlet height) + 50m.
- There is a new filter for times when the PBL is below a fixed threshold (200m?)
- Filters can be specified by site using a dictionary in the ini file
- Log normal priors can specified via
mean
(default = 1) andstdev
, rather than themu
andsigma
parameters from PyMC - Log normal priors can be "reparameterised" as
np.exp(mu + sigma * Z)
whereZ
is a standard normal variable. This can help with divergences. - numpyro can be used for sampling
- the model can be run without model error (i.e. only obs error), for comparison with InTEM and Elris
There are no open issues in this milestone.
Add issues to milestones to help organize your work for a particular release or project.
Create new issueOr find and add issues with no milestone in this repo.