mpvginde/pysteps

Python framework for short-term ensemble prediction systems.

pysteps's Issues

Correct noise cascade initialisation

Task related to this pySTEPS issue: 332

Now, the all noise cascade levels are initialized to zero.
It is probably better to initialize the noise with the appropriate noise field as provided by the EPS term.

The noise cascade has holds 2 time levels of noise. Both time levels should be initialized with the same noise field. If not the overall power of the noise cascade level will not converge to the correct power.

Experiment with localized AR processes

The currently implemented estimate_ar_params_yw_localized function is rather slow.
Let's look into ANVIL as suggested by S. Pulkkinen (slack).

Find and implement fix for the interpolation of smallest level cascades

Check impact of number of cascade levels /do we need 8 cascade levels for Belgian domain?

Prepare flooding cases suggested by Ricardo

The 10 flooding events that occurred in Antwerp as proposed by @ReinosoRadar:

• 10/08/2014 ⬅ No appropriate NWP data
• 13/08/2015 ⬅ No appropriate NWP data
• 30/05/2016
• 05/09/2018 ⬅ No appropriate NWP data
• 09/02/2020
• 17/06/2020
• 16/08/2020
• 08/02/2021 ⬅ No appropriate NWP data
• 04/07/2021
• 24/07/2021

We could check for sensitivity to:

• number of members
• number of NPW-models
• number of lagged NWP-members
• Window size of rolling-average for climatological skill
• ?

Implement no-rain check on radar fields/set default autocorrelation parameters

Task related to this pySTEPS issue 309

From the A. Seed comments as quoted by @ladc:

• Set threshold at around 3% raining fraction.
• Default auto-correlations [code from STEPS1] below this fraction.
  ▶ Default (lag2) auto-correlations use a power law for the correlation length as a function of scale + black magic

Implement no-rain check on NWP fields and implement the no-rain-radar + no-rain-nwp case

Task related to this pySTEPS issue 309

From the A. Seed comments as quoted by @ladc:

• If QPE zero and NWP zero ➡ resulting forecast = all zeros.
• Checking for this case saves a lot of computer time since it is likely to be the case around 80-90% of the time.
• Maybe less likely in case of NWP EPS

Renormalization of the precipitation cascade levels after an AR iteration

Task related to this pySTEPS issue: 332

To prevent a loss of power in the precipitation cascade levels, each level should be renormalized after an AR iteration.
However if you use a global parameters for the AR iteration, the renormalization results in a constant (in time) field.

How should we proceed?

Prepare no-radar-rain + NWP-rain case

Implement no-rain-radar + rain-nwp case

Task related to this pySTEPS issue 309

From the A. Seed comments as quoted by @ladc:

• If QPE zero and NWP nonzero ➡ NWP provides the advection.
• Set radar mean and variance and advection to zero
• Use default power spectrum slopes (?) and the default auto-correlations set the rate of blending from zero rain to NWP forecasts through the calculations of nowcasting skill as a function of lead time and scale.

Variable threshold for the "no rain" criterion

Prepare no-radar-rain + no-NWP-rain case

Identify the interpolation step in the SL-advection

Parallellized ensemble run seems to lose track of ensemble number and order

If I run an ensemble using multiple cores (so, dask will parallellize the ensemble members over the cores), it seems that the ensemble order is lost, resulting in weird transitions (due to noise from a different member that ends up in that member). See for instance (these are three 15-min instance in the forecast):

Or maybe even clearer:

Has any of you come across this and/or know how to fix it?

If I run on 1 core, this problem does not occur, so it gives the impression that this has to do with parallellizing using dask.

Celebrate release of new pysteps.blending version

Once all steps are done, we have one final step before full operational implementation: celebrate!

Add tests for no-rain combinations

Add tests for the different combinations of no rain in NWP, radar, and both.
@RubenImhoff could you pull the version in my hackaton branch and check if your case still works there? Thanks a lot!
I currently get errors for my tests (more specifically the decomposed precip still contains nans)