Attendees

Students from ENM only - no other OpenIFS users this time.

Expect about 20-25 participants.

Students background: 1st yr basic meteorology, 2nd yr statistical tools / numerics (they do small modelling project)

Work in groups: 5 x 4 persons, 2 per PC.

Facilities

PC running CentOS (based on redhat enterprise) - 64bit, 8GB RAM, 30GB free in /tmp

Student accounts limited to 5GB - put the VM image to /var/tmp on each PC. As long as this isn't deleted during the course the students wouldn't lose their work.

28/4: Routing issue on class PCs means we need a VM all the files already installed to provide via ftp.

Language

English used predominantly.

English teachers will be there. Encourage students to write and speak in English as students have asked for more lessons in English.

Tutorial: write in English (perhaps some parts duplicated in French?)

Topic

Storm Nadine. Case interaction between Hurricane Nadine and cut-off low, ensemble uncertainty & severe weather over France.

Paper: Pantillon, F., Chaboureau, J.-P. and Richard, E. (2015), Vortex–vortex interaction between Hurricane Nadine (2012) and an Atlantic cut-off dropping the predictability over the Mediterranean. Q.J.R. Meteorol. Soc.. doi: 10.1002/qj.2635 http://onlinelibrary.wiley.com/doi/10.1002/qj.2635/abstract 

Sensitivity - two main issues : Model resolution - to model interaction of the storm and cut-off low;  Convection processes.

Linus says ensemble is sensitive to SST. This particular case was in fact used as part of the argument for coupling the ocean model (NEMO) from day 1 in the ensemble forecast rather than day 10 as used in 2012; days 1-9 used persisted SST anomalies.

Data

Needs reviewing

From last year's workshop, each field ~ 100Kb. Total size of 'data' and 'data_ls' directories was:5.5G, 1.4G respectively , total of 7Gb.

2015 workshop VM, df reports 40Gb /, 28% used (27G)

Analyses:

T1279 20120915 00Z - 20120925 00Z (note dates for AN is different to forecasts - if starting from 15th Sept is too much, start from 18th)

Fields : 

MSLP, T2m, 10m winds, total precipitation

Geopotential, temperature,  winds (U/V), sp. humidity, 6hrly accum. total precipitation @ (200 or 250), 500, (850 or/both 925) hPa

if need to save space, sp. humidity at 925hPa only (see fig 2)

PV + wind (U/V) @ 330 K isentrope (see Fig.10 in paper) ( Frédéric asked for geopotential/winds at 1.5PVU but I don't think we archive these?)

Vertical x-sections

Potential temperature, potential vorticity, winds, humidity, vertical vel. -- needs reviewing, would need all pressure levels & will be too much data?

Operational ensemble 2012:

Actual operational ensemble TL639. 20120920 00Z - 20120925 00Z (all members)

Fields / levels (TBD)

Current operational ensemble (with NEMO on day 1) (expt id: TBD)

Current operational configuration: T639. 20120920 00Z - 20120925 00Z (all members)

Fields / levels : As for 2012 operational ensemble.

Uncertainty runs:

T319L91. No NEMO coupling, wave model off (OpenIFS config)

Do we have time or the space on the VM for these?

(a) EDA/SV/SPPT/SKEB - exptid 'gik3'

(b) EDA/SV only - exptid 'gin5'

(c) SPPT/SKEB only - exptid 'gin7'

Data domain

Extract data similar to domain in figure 1 but suggest crop 50W - 20E / 30N - 65N (adjust to suitable aspect ratio)

Plots

Usual plots

- 1x1, 2x2

- Difference maps (an-fc, ens-control)

- stamp maps

- RMSE curves geopotential (500hPa), RMSE plumes

- spaghetti plots

Horizontal / isobaric maps

MSLP + 10m winds / T2m -> interesting for tracking Nadine & primary circulation

Geopotential height + temperature at 500hPa --> large scale patterns, position of Nadine's low and the Atlantic cutoff

Geopotential heigth + temperature at 850hPa --> lower level conditions, detection of fronts

1.5 PVU geopotential + winds or 330K potential vorticity (PV) + winds --> upper level conditions, upper level jet and the cutoff signature in PV

If space is an issue, decide which plots are to be interactive and which are available as images only.

PCA in Metview - speak to Linus (he says he has some Fortran code). Or could use R

Try to reproduce the plots in the article. 

• HRES & analysis only. Perhaps ECMWF ensemble? (modified ensemble in later task).

Plots that summarize the situation, focused on the study of Nadine and the cutoff (horizontal maps and vertical cross sections) and maybe a plot of the rainfall over France.

From Frédéric: (15/Mar)

For the ensemble I wonder if we will see the same bifurcation in the scenarios than in the hires. It seems that for this specific case the lower resolution run was better than the hires. We need to see the results of the ensembles with low res to see if it matches what the operational ensemble proposed at the time. (note: we will be running the ensemble at lower than operational resolution T319 compared to T639, for reasons of limitations in data storage)

For the ensemble I wonder if we will see the same bifurcation in the scenarios than in the hires. It seems that for this specific case the lower resolution run was better than the hires. We need to see the results of the ensembles with low res to see if it matches what the operational ensemble proposed at the time.

GC: Could use operational ensemble as well as uncertainty runs? Compare ensemble control with HRES?

Tubing is not compulsory if you have problems finding the code for that. If we do the same as last year this will be perfect. Maybe we can try just to add the PCA of the 500height in Metview as a bonus for this workshop. I'll ask a colleague about clustering in R. If it it easy to do we can add it as a bonus too.

GC: Ask Linus for PCA code
Check what file format R requires (probably netcdf)

Concerning the runs

1. T1279 + analyses : FC from 00Z 20/9/2012. 5 days.

GC: OK. Reproduce Fig1 in Pantillon et al. for AN and HRES forecast.

2. Ensemble (size & res to be confirmed) : OK, we need also to check if the lower res ensemble proposes 2 distinct scenarios

GC: Could use ECMWF operational ensemble (T639)? Compare with uncertainty runs: (a) below is closest to operational but at lower res and without wave model

FF: 2a. Ensemble (T639) EDA/SV/SPPT/SKEB + control --> my question is : Is the resolution of the ensemble (T639) the same resolution as the operational ensemble of 2012 ? If yes we will get the same bifurcation has the one mentioned in Pantillon. If it is lower we need to check if the ensembles propose two distinct scenarios. Concerning the visualization of the ensembles We can do the same has last year's workshop with RMSE, plumes, ensemble spread, spaghettis, ... It will be very interesting to look into this for the students.

GC: Yes, we use the operational forecast ensemble at the time.
TODO: Need to decide on variables & levels to plot: e.g. MSL, geopotential, winds, temp?  What levels?  925hPa, 850, 500, 350?  PV : 330K?

2b. PCA and clustering --> We can try the code you have to see if we get two distinct patterns in the t+96h forecast. I still need to ask about PCA and clustering with R and the format of data needed. This part wil be a bonus if we want to have time for the uncertianty runs.

GC: OK, TODO: (1) test Linus' PCA code and report back. Use 500hPa geopotential as in paper,  (2) FF to investigate use of R & format of data required.

3. Sensitivity run with ensemble, varying SST? We can try, to see if it impacts the motion of Nadine. Surface fluxes may also have an influence on the motion of Nadine. Switching of deep convection ? Maybe too radical ? Is it possible to do a sensitivity to the resolution ? The interaction between Nadine and the cutoff seems to be impacted by the resolution.

GC: Linus suggested sensitivity to SST was important. However, I think to add this will result in too much data and be too many topics. Concentrate on resolution effects and changes in model/initial data uncertainty. This is still alot for participants to do.

4. Uncertainty runs: (a) EDA/SV/SPPT/SKEB; (b) EDA/SV; (c) SPPT/SKEB : Will we do class ensembles ?

GC: Agreed. Although no-one will run OpenIFS to produce a 'class ensemble', the exercises allow the participants to group perturbed members into different size ensembles to see what impact it has.

FF:a. Uncertainty runs EDA/SV VS SPPT/SKEB --> Do you plan do the class ensemble to see the impact of the number of members ? We do not have to run it in real time, I'll just tell you the number of students (or binomes ?)
b : if EDA/SV VS SPPT/SKEB are the same : SST sensitivity --> The question of the SST of the atmospheric model is interesting, students often think it varies whereas it is constant if I am not mistaken ? If you see a great sensitivity of the forecast to the SST in terms of intensity and track of Nadine this could be great !

5. SCM: I have the radiation exercices and convection exercices from previous workshops. It could be nice to look at the convective tendencies in Nadine's deep convection to illustrate of the convection scheme works. I am sure Peter Bechtold has great ideas about that !

On the second day, François Bouttier can also speak about optimal tresholds to use in order to deal with extremes with convective scale ensembles.

GC: It would be good to add something on this into the exercises?

I agree with you concerning the need to do the SCM case study. I am waiting for your input and ideas as I don't really know what the model can do.

GC: I think a convection based exercise is best here.

Notes from Frédéric: (9/Feb)

I just spoke to a colleague of mine teaching statistics. He told me that our second year students that will attend the workshop will have knowledge in PCA and clustering methods via ascending hierarchical classification (with R)

So I think that on Day 2, it would be nice to make them do the PCA for ECMWF t+96 ensemble forecasts (20120920 analysis and forecast for the 20120924, figure 5 of Pantillon), the ascending classification (figure 6) and the clustering (figure 7). We could leave the classification and the clustering as extra questions, to leave room for the stochastic ensemble, the EDA and the class ensemble.

I will also think about the interesting plots that we could do to illustrate this case, for example vertical crosssection of potential vorticity.

TODO

• Decide on forecasts for workshop: resolution, start date(s) and forecast length : all runs start 00UTC 20 Sept 2012
• Decide what to do about provision of satellite imagery (or not - issues iwth licensing)
  
• Extract HRES & analysis
• Run ensemble forecasts
• Book accommodation
• Install SCM on OpenIFS VM
• Get PCA code from Linus & try it
• Get USB sticks from comms
• Frédéric to check last year's VM will run on classroom PCs and ensure VirtualBox (or VMware) is installed.
• Check for observations we might have in the archive for HYMEX that could be used

Notes

SCM exercises - sensitvity of the model to changing physics. Focus on convection, initiation of deep convection. Stochastic physics, impact on convection? impact of convective scheme?

Do we want the students to run OpenIFS? - probably not enough time / resources

Interested in stochastic physics / ensemble methods.

Can we get observations from HyMEX to add to the metview exercises?

Focus on ensemble modelling and predictability.

Try to include some Hymex observations in metview VM?

Some questions to include:

• how to weight extreme events in the ensemble?
• what's the best way of doing ensembles if dont' want to miss extremes?
  
  

Travel & local arrangements

ENM is approx 20mins by car from the airport.