Introduction
In these exercises we will look at a case study of a severe storm using a forecast ensemble. During the course of the exercise, we will explore the scientific rationale for using ensembles, how they are constructed and how ensemble forecasts can be visualised.
You will start by studying the evolution of the ECMWF analyses and forecasts for this event. You will then run your own OpenIFS forecast for a single ensemble member at lower resolutions and work in groups to study the OpenIFS ensemble forecasts.
Caveat
In practise many cases are aggregated in order to evaluate the forecast behaviour of the ensemble. However, it is always useful to complement such assessments with case studies of extreme events, like the one in this exercise, to get a more complete picture of IFS performance and identify weaker aspects that need further exploration.
The case study will look at one of several severe wind-storms that hit Europe in late 2013 (see handout of ECMWF article by Hewson et al, ECMWF Newsletter 139).
- On the 28th October 2013 a small, severe wind-storm named St Jude in the UK, hit the UK & north-western Europe.
- A total of 19 people were killed across Europe, 5 in the UK.
- The return period of the event based on wind-gust observations show the 10yr return period was exceeded along the North Sea coast.
- From the 23rd October, the ECMWF forecast predicted a greater than 70% probability of a severe wind event (greater than 60kt, 31m/s, at 1km) over southern England. A signal for the storm was evident from the 21st October.
- On the 24th October, the UK MetOffice issued an amber alert for wind-speed across southern England placing the potential impact in the highest category.
- The cyclone first appeared as a cold front wave, south of Nova Scotia late on 25th October.
- It deepened and moved rapidly east then northeast, with the storm centre reaching southern Sweden late afternoon on the 28th.
- The most rapid deepening occurred between 06-12UTC on the 28th between eastern England and the North Sea where the strongest wind gusts were observed.
Recap
Key points
- Sources of forecast uncertainty: initial analysis and model error.
- Initial analysis uncertainty: sampled by use of Singular Vectors (SV) and Ensemble Data Assimilation (EDA).
- Model uncertainty: sampled by use of stochastic processes. In IFS this means Stochastically Perturbed Physical Tendencies (SPPT) and the spectral backscatter scheme (SKEB)
ECMWF operational forecasts consist of:
- HRES : T1279 (16km grid) highest resolution 10 day forecast
- ENS : Ensemble (50 members), T639 for days 1-10, T319 days 11-15.
We suggest these exercises are best done by small groups working in teams.
Suggestions are made in the exercises for how each team can work on different data.
Exercise 1. Evaluating the ECMWF analyses and forecasts
Objective: Understanding the error in the forecast by comparing the ECMWF forecast with analysis & observations
Starting up metview
metview
Please enter the folder 'OpenIFS workshop 2015' to begin working.
Task 1: Visualise observations and ECMWF analyses
For this task, use the metview icons in the row labelled 'Analysis'
1. Right-click on the icon labelled 'wgust_obs.mv' and select 'Visualise' (or Control-I on the keyboard)
Icon 'an_1x1.mv' produces a single plot on the page.
Icon 'an_2x2.mv' can produce up to 4 plots per page.
Change plot fields
For the 'an_1x1.mv' icon, the plot contents can be changed by editing the plot1 variable in the macro. By default the 10m wind is shown.
To alter the plotted field, right-click and choose 'Edit'.
It is possible to overlay multiple fields like this:
You will find a list of available parameters in the macro.
After editing the macro text, you can optionally save using the 'File' menu and 'Save'.
Display the plot by clicking: 
Animate the plots in the display window by clicking
Change plot appearance
The number of maps appearing in the plot layout can be 1, 2, 3 or 4.
Macro an_2x2.mv demonstrates how to plot a four-map layout in a similar fashion to the one-map layout. The only difference here is that you need to define four plots instead of one.
Right-click on the icon and select 'Edit'. Change the plot layout like this:
Two map types are available covering a different area.
With mapType=0, the map will cover a smaller geographical area centred on the UK.
With mapType=1, the map will cover most of the North Atlantic
Task 2: Visualise operational HIRES forecast
The ECMWF operational forecast is called HIRES. The model runs at a spectral resolution of T1279, equivalent to 16km grid spacing.
Only a single forecast is run at this resolution as the computational resources required are demanding. The ensemble forecasts are run at a lower resolution.
Before looking at the ensemble forecasts, first understand the performance of the HIRES forecast.
Data is provided for multiple forecasts starting form different dates, known as different lead times.
Available lead times for October 2013 are: 24th, 25th, 26th and 27th.
For this task, use the metview icons in the row labelled 'Oper forecast'
oper_2x2.mv works in a similar way to the an_2x2.mv icon used in the previous task.
Suggested fields to plot: MSL, Z200, 10m wind and visualize the storm track.
Use the metview macros to plot different days and compare to analysis and plot forecast differences.
Task 2 : Visualize the ensemble forecasts and ensemble spread
Again using the ECMWF operational forecast, look now at the 50 ensemble forecasts. These are at a lower resolution (T639) than the HIRES (T1279).
Objective: understand the forecast uncertainty
Key questions:
- Visualise the ensemble mean - how does it compare to the HIRES forecast and analysis?
- Visualize stamp map - are there any members that provide a better forecast?
- Visualize spaghetti map - see how members spread over the duration of the forecast. How does the spread grow?
Suggested plots:
- 4 per frame: analysis, ensemble mean, spread and 1 other.
- spaghetti plots (multiple plots per frame)
- stamp plots.
- Observations of wind gust + analysis (faked): 1 day only.
Note that as only 6hrly wind gust data is available from the operational forecasts, we have supplemented the 3hrly fields using forecast data.
Exercise 2. CDF/RMSE at different locations
Recap
TO DO: RMSE & CDF (concepts need explanation)
Objective: Understand ensemble reliability.
- Choose 3 locations.e.g. Reading, Amsterdam, Copenhagen.
- Using 10m wind and/or wind gust data plot CDF & RMSE curves using one of the OpenIFS forecasts and the ECMWF analysis data.
- What is the difference and why?
- Repeat for one of the other OpenIFS runs. Are there any differences, if so why?
Questions:
Group A: The Swedish weather centre is interested in giving out useful weather warnings. They want to know the maximum precipitation and wind gusts over Scandinavia and their likelihood (more than XX m/s wind and XX mm precipitation). Would you give out weather warnings and if yes for which days and areas (1 day lead time)?
Group B: The organiser of a garden party at Windsor Castle at a particular date needs to prepare for weather. The Queen will join the party and she is not amused if there are too many tents that are not necessary. However, it will take some time to prepare shelters for rain and wind or to cancel the event (two days lead time). Would you prepare for rain?
Group C: The operator of a wind farm in the northern see needs to know the highest possible wind speeds to shut down turbines (at XX m/s). If the turbines are shut down too early, power production will be reduced. If they are shut down too late, they will break (6 hours lead time). Would you shut down the wind farm?
Exercise 3. Creating an ensemble forecast using OpenIFS
See separate handout.
In this exercise, OpenIFS will be run on the ECMWF Cray XC30 to create a forecast for the storm at T319 resolution using only the stochastic schemes in the model. All forecasts are started from the same initial conditions based on the analysis.
Aim is to understand the impact of these different methods on the ensemble
Exercise 4. Verifying / Quantifying OpenIFS forecasts
Experiments available:
- EDA+SV+SPPT+SKEB : Includes initial data and model uncertainty (experiment id : gbzl)
- EDA+SV only : Includes only initial data uncertainty (experiment id: gc11)
- SPPT+SKEB only : Includes model uncertainty only (run by participants, experiment id: oifs)
These are at T319 with start dates: 24/25/26/27 Oct 00Z for 5 days with 3hrly output.
Plots:
- as above
- 4 frame: fc-an, fc-fc, pert.fc, ctl-an? *(compare the fc-fc maps with fc-an maps - can we see the uncertainty in the difference?)
- PV maps
Task 1.
Objective: Understand the impact of changing the ensemble uncertainty
Look at ensemble mean and spread for all 3 cases.
- How does it vary?
- Which gives the better spread?
- How does the forecast change with reducing lead time?
Task 2.
Look at different ensemble sizes?
Task 3.
- Find an ensemble member that gives the best forecast and take the difference from the control. Step back to the beginning of the forecast and look to see where the difference originates from. How does this differ between the 3 OpenIFS runs? (with model uncertainty only, each initial state is identical so differences will develop from
Task 4.
Ensemble perturbations are applied in positive and negative pairs. For each perturbation computed, the initial fields are CNTL +/- PERT. (need a diagram here)
- Choose an odd & even ensemble member from one of the 3 OpenIFS forecasts (e.g. members 9 and 10). For different forecast steps, compute difference of each member from the control forecast and then subtract those differences. (i.e. centre the differences about the control forecast).
- What is the result? Do you get zero? If not why not? Use Z200 & Z500? MSLP?
- Repeat looking at one of the other forecasts. How does it vary between the different forecasts?
If time:
- Plot PV at 330K. What are the differences between the forecast? Upper tropospheric differences played a role in the development of this shallow fast moving cyclone.
Further reading
For more information on the stochastic physics scheme in (Open)IFS, see the article:
Shutts et al, 2011, ECMWF Newsletter 129.
Acknowledgements
Many people have contributed to this exercise and their contributions are acknowledged, in particular from ECMWF: Glenn Carver, Linus Magnusson, Sandor Kertesz, Martin Leutbecher, Iain Russell, Erland Kallen. From University of Oxford: Aneesh Subramanian, Peter Dueben, Peter Watson, Helen Christensen.
NOTES
These will disappear in the final handout.
- MSLP
- 10m winds
- T2m
- Z500, Z200 winds geopotential.
- wind gust : model & obs (nb. model wind gust data is accumulated)(Linus has windgust obs in geopoint format)
- PV (decide which levels)
- layout
- contour / colourfill (overlay)
- wind vector
- spaghetti plots
- stamp maps
- rmse & cdf at several locations (user chooses)(Linus has macros for these)
- Difference maps : to plot fc - an and ens member - control ( or ens member(i) - ensemble member(j) )
- step animation of spaghetti plots etc to see spread developing.
- Linus' vortex centre & tracking plot
Retrieve data from MARS for all apart from the OpenIFS experiment the participants will run themselves.
Note that operational ensemble runs at T319 are also available if we want to use them to compare 40r1 with OpenIFS? (see Linux for MARS script)
wind gust data is not simply max windspeed over 6hr period, it includes convective component (as well as conversion from windspeed to gust speed).
need to fake windgust data!
In exercise 1, the aim is to study the error between the forecast and the analysis.
Key points:
The forecast developed the storm too early, even for the forecast on the 27th. Storm was too far west.
plot Z200 to see position of the jetstream.
For the ensemble, plotting the mean will remove some features - might be useful for class to see this.
Note how spread is small for northerly low centre where uncertainty is much less.
Make sure they use 4 frame layout and plot 25th, 26th, 27th + analysis. Also, 25th, 26th & their difference
Question. How best to organise the experiments? Each user has an account or use one account with multiple directories?
Linus suggested running a script that reorders the data to have 1 file with all ensemble members for each field of interest. Do this after all members have run?
Suggests doing 2 more runs: EDA only and SV only. The perturbations from these have different characteristics and as we don't yet know what the results look like it might be useful to have these runs.
(I may not have all the details right - check with Linus who apparently did a PhD thesis on this)
Relate the perturbations to the baroclinic zones (ie. large scale flow). While the cyclone is developing the baroclinicity will be high and the EDA/SV perturbations will be more significant than when the storm is more developed with fronts when it will be more stable.
Should see that SV are quite artificial, some pertubations will grow rapidly and then die out. By contrast the perturbations in the EDA will be larger in the initial conditions.
Maybe we should plot the Eady index?
Erland says very important to plot these and see what we get in order to refine the questions and direction we want participants to take.
Linus explained that with the OpenIFS runs will have differing amounts of uncertainty, so the spread should noticeably change for points near the track in the analysis. This is particularly because of (a) timing error between the analysis & fc, (b) the ensemble tracks being more to the north of the analysis track. So Amsterdam for instance should see much less spread as the uncertainty in the ensemble is reduced.
Parameters that do not have a Gaussian like distribution in the ensemble can be problematic.