Log In | Users | Register
Edit | Attach | New | Raw | Diff | Print | | Tools

myCOSMO-NExT Case: 2009 07 23

Severe and long-lived hailstorms ... of supercellular character ... producing record hail and wind damage (initiated 26 February 2014, sto)

Ahead of an approaching large scale trough and accompanying frontal zone a severe storm setting develops over the northern alps: a potentially unstable airmass, rich of moisture and well (speed) sheared builds over the plains and western alpine slopes. At the same time south foehn descends down to the northeastern alpine valleys and creates a regime of hot, less unstable and above all very dry air - not a favorable environment for thunderstorm formation.

largeScale.jpg

ECMWF analysis 20090723 12UTC: standard fields at pressure levels of 300/500/850/mslp are displayed (top left to bottom right)

pay12.jpg

Radiosounding Payerne at 12 UTC. Note the very strong southwesterly speedshear from the surface to 550hPa

In the morning hours widespread thunderstorm formation is observed over northeastern France and the neighbouring Swiss Jura area. The easternmost storms produce outflow surges that rush over the eastern Jura mountains and the Swiss-German plains, those are evident in both observations as well as in some of the better CO-2 model runs. The influence of these surges on the subsequent storm evolution during the afternoon is an interesting question e.g. in terms of cell triggering and maybe it is one of the key features of this case that have to be captured by the model dynamics in the early phases of any successful simulations.

outflowSurge.jpg

Multi-parameter depiction of early thunderstorm outflow surging over the Swiss German plains

Around noon a thunderstorm developed around Lyon, moved quickly northeastwards to western Switzerland, thereby growing massively in size and intensity. A first peak of intensity is evident in terms of dBZ and hail damage reports when the storm rushed through the cantons of Vaud and Fribourg along the western alpine rim. The storm split subsequently in the area around Bern and quickly recovered to produce hail again both in the central Swiss plains (left moving storm, rather quickly dying thereafter) and in the central alpine foothills in the area around the Lake Vierwaldstätter. The right moving storm continued its eastward progression as a single entity evident in the radar loops and got in contact with the Foehn flow east of Lake Vierwaldstätter, temporarily weakening when crossing tongues of dry air, but recovering every time when drawing enery from the moisture-rich air over the plains. This complicated interplay of the storm with the two different regimes of inflow resulted in short phases of storm weakening obvious in dBZ and temporarily ceasing lightning frequency, alternating with quick storm recovering and subsequent damage at ground (F1 wind damage in the Toggenburg, maximum sized hail of 8 cm observed in the Appenzellerland).

BEsplit.jpglightningTrack.jpg

Top left: Snapshot of the storms first split: CO-2 model's 500hPa winds overlaid as colour vectors, storm tracks as broken lines

Top right: Météorage/EUCLID and SFUK lightning locations (mainly cloud ground strikes), SFUK confined to the storm cell's perimeter (solid blue lines)

Considering storm reports and observational data from southern Germany as well as Austria the track of this particular storm cell seems to extend several hundreds of kilometers into the Austrian alps before its signal gets lost in the Steiermark area, thereby producing some more splits and daughter cells that followed the alpine rim along a slightly northern track across Bavaria and northern Austria. The focus of this analysis remains in Switzerland, where secondary storms formed in the western Alps and the northern plains later in the afternoon and early evening, although not achieving the same levels of intensities as the primary cell described above.

hailAnalysisIRV.jpg hailEstimatesAlps.jpg

Top left: hail size analysis in western Switzerland, combined spotter reports (dots) and radar estimations (Meteoradar).

Top right: Combined ORGANIZATION.ORGMeteoSwiss and ZAMG hail maps, estimated from radar data (dBZ-thresholding)

Quality of operational CO-2 version running at that time:

Model runs with 6hr lead time or more (earlier than 06 UTC) consistently highlighted the ingredients of the larger scale (strong forcing, very strong shear) and mesoscale environment (potentially unstable airmass, rich moisture, Foehn regime in northeastern alps), enabling forecasters to anticipate the potential for damaging thunderstorm developments with supercellular characteristics. On the storm scale the model indicated areas of intense rain and strong gusts, changing in intensity and location from run to run (as usual ...), worth mentioning the strong signals in the SDI-plots that confirmed the expected dominant mode of convection.

oprCO2preci.jpg oprCO2sdi.jpg

Top left: operational CO-2 3hr precipitation forecast for 12 to 15 UTC, model run of 09 UTC

Top right: operational CO-2 supercell detection index (a combined measure of updraft strength and rotation), 24hr forecst from 15 UTC model run the day before

Version 1

Overall quality of forecast: reasonably good forecast quality on a broader scale ... comparable, maybe slightly less good than the operationals versions of CO-2 that time

Some general comments i'm feeling comfortable with based upon the data as available up to mid february 2014:

  • outflow surge from early convection north of the Jura mountains is somewhat delayed and weaker than in the operational CO-2 runs as well as in comparison with sfc obs
  • both extent and intensity of the Foehn flow, compared with sfc obs at 12 UTC, is excellent forecast in the western parts of Switzerland as well as in the Alps
  • in the northeastern plains of Switzerland the Foehn-dried air seems to mix down to the surface over a vast area - considering the observed weak winds and same dewpoints present as further west this downward mixing of dry air did not take place in reality - consequently the model underestimates air mass moisture and instability in these regions (compare with the distribution of CAPE and CIN below) ... this effect can be due to an overestimation of turbulent (vertical) mixing or a consequence of the underestimated convectively produced cool air in northwestern Switzerland that spreaded somewhat eastwards at that time

sfcwinds12z.jpg

10m winds and gust measured at 23 July 2009 12 UTC (SwissMetNet)

uv10m 200907231200.png

Predicted 10m winds at 23 July 2009 12 UTC, COSMO-1 Version 5/9

  • storm environment: hard to verify since there is no ground truth at comparable resolution. The potentially unstable airmass over the central and western plains is well depicted, a plume of Foehn-dried airmass with significant amounts of inhibition extends from the central parts of the Swiss alps northeastwards across Lake Constance into southern Germany. Later in the forecast this signal weakens as the amound mixed layer CAPE increases, while CIN remains high in these areas. Strong vertical shear is evident throughout the forecast.

CAPE ML 200907231200.png

Top left: mixed layer CAPE in the pre-storm environment at 12 UTC

CIN ML 200907231200.png

Top right: mixed layer CIN in the pre-storm environment at 12 UTC

The main difficulty in estimating model quality is the lack of appropriate output in "mesoscale manner" and resolution: storm parameters such as sfc winds, gusts, simulated dBZ, mid level up- and downdraft speeds and hydrometeor content must be visualised and analysed not only in high spatial but also in high temporal resolution (at best: 10 minutes) in order to cope with the fast evolving dynamics of this particular case. Traditional model output in three to twelve hour frequency are clearly not adequate and make it impossible to see the true model dynamics, maybe also preventing a fair estimation of the model's forecast quality.

A first attempt to improve this situation is the delivery of model output of the above mentioned parameters at 10 minutes resolution. This dataset is already available is currently being ingested into the experimental weather simulator on ORGMeteoSwiss' NinJo? CaseCaptureReplay? environment. Currently there is a lack of observational data for comparison with ground truth, however - work in progress. The lesson learned so far is:

  • COSMO-1 is a true mesoscale model - consequently we have to look at mesoscale output (convective) parameters on mesoscale resolution (10-100km, 1min-1hr)
  • the necessary conditions for successful forecasts of this case are manifold: good representation of the pre-storm environment (different airmasses, mesoscale flows), precise triggering of major cells, proper simulation of storm evolution
  • once LHN and other sophisticated assimilation techniques become available shorter term model runs (09/12 UTC) can remove the triggering deficiencies and isolate the other two factors

update 27 February 2014, sto:

"Detto, fatto!" we've just successfully importet the very first 10 minute COSMO-1 output of model version 5/9 into NinJo?! This now offers exciting new capabilities to analyse the case in the model world and understand, why the forecast looks like it does with all its good and bad facets. Here's a first update: if focussed my analysis on the triggering of cells in the early phase of the forecast. Doing so i became aware of the fact the the model is able to simulate supercell storm dynamics in accordance with theory and high resolution observations from research projects.

Note that up to now we do not have any observational data available on NinJo? so i helped myself with offline archive radar plots and ESTOFEX's map with METAR observations, arranged on the left part of the plots. On the right there's COSMO-1 version 1 output (from left to right, top to bottom): 10m mean windspeed overlaid on max simulated reflectivity (main scene, windspeed in colored isolines and white vectors, dBZ in grey shades for orientation relative to the cells); SupercellDetectionIndex2? (2nd scene, Wicker et al., 2005: reddish colors mean cyclonic updrafts, blueish colors anticyclonic updrafts) overlaid with vertical velocity 700hPa (updrafts solid black, downdrafts broken white lines); 3 km simulated reflectivity (dBZ, 3rd scene top left); 2m temperature and 10m windfeathers (3rd scene); simulated maximum reflectivity in nice colors (4th scene, bottom left).

Here we go with an in depth analysis step by step, if you're not interested in all the details skip this part and proceed to the conclusions further below.

Nocturnal convection over France 00-06 UTC:

Convective cells form and move along a line from Limoges to the Vosges, it's mostly showers of rain and weak to moderate thunderstorms. Their outflows advance somewhat south of this line and are evident temporarily in the sfc obs of LFLC, LFLV, LFLN and LFLD. COSMO-1 reproduces those convective elements approximately 100 km too far to the north and with weaker intensities than observed. Outflow surges as indicated in the sfc obs are completely missing in the surface wind and temperature fields (look carefully at the French airport reports).

2009072304.jpg 2009072305.jpg

Overview about radar measurements (Swiss radars) and METARs (left side of the figure) and COSMO-1 output (right part of the figure, see text explanations for description of the parameters displayed as well as captions in the individual scenes)

Early morning phase 06-08 UTC:

Quite near the boundary of the nocturnal outflows another burst of new, partly intense convective cells form in the northern part of the Monts du Lyonnais and move northeastwards along the river Saône. Again COSMO-1 underestimates the intensity of these cells in terms of simulated vs. observed dBZ as well as strengths of outflow surges, as measured in terms of 10m windspeed and direction, extension and amount of cooling. The model in fact misses this evolution and maintains a weaker line of convective elements further to the north.

2009072308.jpg

Forenoon evolution 08-11 UTC:

From those early morning convective cells introduced just before evolved two major thunderstorms that moved across the river Saône and Doubs, rushing over the Franche-Comté region and thereby approaching the northern foot of the Jura mountains at 1030 to 1100 UTC. Their outflow boundaries advanced southward down the Saône/Rhone valleys to reach the Lyon region with 10-20KT mean wind speeds. On top of these outflow surges warm to hot southerly inflow from the southern parts of the Rhone valley and, eventually, the Bay of Lyons continued (not shown). COSMO-1 still massively underestimates the development and intensities of the two cells, and consequently the strength and extent of the ouflow surges over France, too. However the model recovers convection and produces its first precipitation cells after 1030 UTC in more or less the right area and let them move along a correct line LFLO-LFGJ-LFSM. On the other hand, the southerly surface wind all over these areas prevail in the model simulation, there is no sign of any cold pool present!

2009072311.jpg

Main event at noon and afternoon 11-15 UTC:

A few minutes before 11 UTC new cells are triggered along a line St. Etienne-Lyon-Bourg. Several individual cells quickly develop to mature thunderstorms, the northernmost of them rushing again across Franche-Comté towards the upper Rhine valley into Baden-Württemberg, where its trace gets lost after 1430 UTC near Stuttgart. The southernmost storm, probably feeded with very "healthy" southerly inflow rapidly grew as well and soon underwent a split over the city of Lyon. Its right moving daughter cell quickly regained full strength until it entered Switzerland at 13 UTC as a fully developed hailstorm and massive in size. From 13 to 15 UTC, in its most intense phase of the early life cycle, it raced from Lake Geneva towards Bern, where the next split was observed, as introduced already in the first part of the analysis further above. In the COSMO-1 simulation according cells develop somewhat earlier but more or less in the right place. These cells grow too, in the model world, and several splits can be deduced in the area south of Besançon when following dBZ loop sequences. Thereafter the right member of the leading split storm pair grows massively and attains v-shape signatures at 1245 UTC in the model's 3 km simulated reflectivity fields, indicative of strong supercellular dynamics! This particular storm dominates convective activity in the simulation in terms of size and intensity as the storm continues to move across northern Switzerland and Baden-Württemberg. Nevertheless several more splitting storms, slightly less intense, follow in its wake during the afternoon and affect the Jura mountains and the Swiss plains in the course of the main event. Unlike diagnosed with the first set of (traditional) model output parameters the mesoscale plots show that the COSMO-1 simulation from 12 to 15 UTC is better than prematurely thought, although the majority of individual thunderstorm cells developed slightly too late and is somewhat misplaced to the NW, and the "wrong" cells grow and dominate the evolution (however, given the dynamics and the leadtime of 12 to 15 hours we enter at this point the grey zone of forecast uncertainty inherent not only to global scale but also mesoscale model simulations ...)

2009072312.jpg 2009072313.jpg 2009072314.jpg

Conclusion of this first session of mesoscale model output diagnosis:

  • the model underestimates and misplaces nocturnal convective activity over France and their convective outflows,
  • but recovers shortly before the main event to produce the right convective mode, intensity as well as more or less the right timing of individual storm cells ...
  • ... however the latters occurring somewhat too far to the NW in the simulation and, within the zoo of cells that develop, the wrong one dominates the course of the event/simulation.
  • The proposed and implemented output fields/parameters at 10min temporal resolution allow a more realistic and fair diagnosis and interpretation of the model dynamics on the right scale and unravels a better forecast quality than prematurely thought. Operational application of this model class in convective situations require a set of model ouptput that follows along the line of plots we generated for this case study, otherwise the quality of the model forecast is not exploited optimally.
  • We are still missing the observations and measurements (sfc, radar, satellit, lightning) on the NinJo? workstation for a proper head-to-head comparison of the model fields with ground truth - work in progress.

In order to study these findings more thoroughly and maybe improve some of these deficiencies I suggest the following next steps:

  • Run simulations with Latent Heat Nudging (LHN, very important to include French radar data!) to test the effect of better representation of nocturnal convection and outflow dynamics over France. I am aware of the fact that LHN might still take some time to be ready for the COSMO-NExT cases, so ...
  • ... try to run model simulations with an adapted parameter setting that allows for more latent cooling of downdraft air (diabatic effects in vicinity of precipitating cells) to test the sensitivity of the whole case evolution on these nocturnal outflow dynamics over France.

While analysing some of the indivual storm cells that occur in the model simulation we were excited by the clean supercellular signals (such as v-shape, forward- and rear-flank downdrafts, right/left moving split storms). We were aware of those signatures spotted in occasional cases in the currently operational 2km model version, but the one particular storm present in this case study simulation is of striking clarity with regard to common supercell theories and observations in the field - a convincing evidence, that the model is able to simulate correct storm structures and dynamics, given the right large and meso scale environment. This will be a topic of a next session ... here's an animated view of some of the scenes shown above for the thunderstorm zoo that forms in the model simulation, a detailed analysis will follow soon:

dBZmax BTchannel6 loop09-18.gif sdi2 w700 loop09-18.gif

For a condensed summary of the assessment for this case as well as all other myCOSMO-NExT cases refer to the myCOSMO-NExT Overview page.

Note: This page is both world-readable and world-writeable.

Verification of the COSMO-1 foehn forecast

The described heavy convection case described above in detail by Marco Stoll showed also an extraordinary warm, dry and strong pre-frontal summer foehn case. Therefore the COSMO-1 foehn forecast is also compared to surface measurements based on the automatic foehn index ORGANIZATION.ORGMeteoSwiss Fachbericht Nr. 223 about automatic foehn diagnosis at two valley sites (Altenrhein at the southern shore of the Lake of Constance and Vaduz in the Alpine Rhine valley) and the Alpine crest site Gütsch. Detailed COSMO-1 verification results for all foehn cases since February 2013 can be found here.

Version 1

Overall quality of forecast: 6

Startzeit: 00 UTC

  1. Foehn onset
  • Vaduz:
    • onset on the day before, so no comparison is possible
  • Altenrhein:
    • too early (on the day before in the model), foehn interrupt between 3 UTC - 6 UTC completely missed by the model
  1. Foehn end
  • Vaduz:
    • 1 hour too late
  • Altenrhein:
    • 2 hours too late
  1. Comparison of wind direction and wind speed
  • Gütsch:
    • wind direction ok, average wind speed underestimated by 50% in the morning, but ok at noon, maximum wind speed agrees well
  • Vaduz:
    • wind direction ok, average wind speed slightly overestimated except too low around noon, maximum wind speed overestimated up to 100% (+50 km/h)
  • Altenrhein:
    • wind direction ok, average wind speed OK, maximum wind speed overestimated by 20% - 30%
  1. Comparison of temperature and relative humidity
  • Gütsch:
    • temperature underestimated by 1K in the morning and overestimated at noon by +1K (too strong diurnal cycle), absolute rel. humidity overestimated by 5% - 15% (too humid for the whole foehn period)
  • Vaduz:
    • temperatur 2 - 4K (noon) too low, rel. humidity overestimated by 10% to 20%
  • Altenrhein:
    • temperature underestimated quite constantly by 3K, absolute rel. humidity offset about +10%

Version 2

Overall quality of forecast: 5

Startzeit: 00 UTC

  1. Foehn onset
  • Vaduz:
    • onset on the day before, so no comparison is possible
  • Altenrhein:
    • too early (on the day before in the model), foehn interrupt between 3 UTC - 6 UTC completely missed by the model
  1. Foehn end
  • Vaduz:
    • 2 hours too late (1 hour later than version 1)
  • Altenrhein:
    • 3.5 hours too late (highest temperature 4 hours after measured foehn end!)
  1. Comparison of wind direction and wind speed
  • Gütsch:
    • wind direction ok, average wind speed underestimated by 50% in the morning, but ok at noon, maximum wind speed slightly overestimated
  • Vaduz:
    • wind direction more variable than measured, average wind speed slightly overestimated except too low around noon, maximum wind speed less overestimated than version 1
  • Altenrhein:
    • wind direction ok, average wind speed overestimated by 50% - 100%, maximum wind speed overestimated by 20% - 30% (same as version 1)
  1. Comparison of temperature and relative humidity
  • Gütsch:
    • temperature has even stronger diurnal cycle, absolute rel. humidity slightly improved compared to version 1
  • Vaduz:
    • temperature 2 - 6K (at noon) too low, rel. humidity overestimated by 10% to 20%
  • Altenrhein:
    • temperature underestimated quite constantly by 3K, absolute rel. humidity offset about +10%

Short comparison between version 2 and version 1

Offset at foehn end increased, negative temperature bias at noon increased, positive wind speed bias increased in Altenrhein, but better maximum speed in Vaduz, diurnal increases of 2m temperature less realistic at Vaduz (missing diabatic heating?)

Version 3

Overall quality of forecast: 5

Startzeit: 00 UTC

  1. Foehn onset
  • Vaduz:
    • onset on the day before, so no comparison is possible
  • Altenrhein:
    • too early (on the day before in the model), foehn interrupt between 3 UTC - 6 UTC completely missed by the model
  1. Foehn end
  • Vaduz:
    • 2 hours too late
  • Altenrhein:
    • 3.5 hours too late (highest temperature 4 hours after measured foehn end!)
  1. Comparison of wind direction and wind speed
  • Gütsch:
    • wind direction ok, average wind speed underestimated by 50% in the morning, but ok at noon, maximum wind speed slightly overestimated
  • Vaduz:
    • wind direction more variable than measured, average wind speed slightly overestimated except too low around noon, maximum wind speed less overestimated than version 1
  • Altenrhein:
    • wind direction ok, average wind speed overestimated by 50% - 100%, maximum wind speed underestimated in the morning, but ok at noon
  1. Comparison of temperature and relative humidity
  • Gütsch:
    • temperature is slightly colder and similar to version 1, rel. humidity is even higher than version 1 and 10% - 15% higher than measurements
  • Vaduz:
    • temperature 2 - 6K (at noon) too low, rel. humidity overestimated by 10% to 20%, temperature is even slightly lower than version 2 and the bias increased
  • Altenrhein:
    • temperature underestimated quite constantly by 3K, absolute rel. humidity offset about +10%, temperature is even slightly lower than version 2 and the bias increased

Short comparison between version 3 and version 2

Negative temperature bias and positive rel. humidity bias even slightly increased, but better maximum speed in general. No obvious changes in foehn onset and foehn end.

Attach
jpg
2009072304.jpg (853.22K)
version 1 uploaded by MarcoStoll on 27 Feb 2014 - 15:08
jpg
2009072305.jpg (861.71K)
version 1 uploaded by MarcoStoll on 27 Feb 2014 - 15:09
jpg
2009072308.jpg (940.37K)
version 1 uploaded by MarcoStoll on 27 Feb 2014 - 15:19
jpg
2009072311.jpg (987.58K)
version 1 uploaded by MarcoStoll on 27 Feb 2014 - 15:29
jpg
2009072312.jpg (997.88K)
version 1 uploaded by MarcoStoll on 27 Feb 2014 - 15:48
jpg
2009072313.jpg (1021.72K)
version 1 uploaded by MarcoStoll on 27 Feb 2014 - 15:48
jpg
2009072314.jpg (1057.14K)
version 1 uploaded by MarcoStoll on 27 Feb 2014 - 15:48
jpg
BEsplit.jpg (100.28K)
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:10
pdf
Betschart, M: 2012, A Study of Convective Events in Switzerland with Radar and a High-Resolution NWP Model, Scientific Report MeteoSwiss, 90, 119 pp.
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:37
png
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 15:19
png
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 15:19
gif
version 1 uploaded by MarcoStoll on 27 Feb 2014 - 16:19
pdf
Schweizer Hagel, 2009: Hagelinfo. 23. Juli 2009: ein verheerender Hageltag. Zeitschrift der Schwerizischen Hagel-Versicherungs-Gesellschaft, 3/2009
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:39
jpg
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:22
jpg
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:22
pdf
IRV, 2009: Ereignisanalyse Hagel 2009. Interkantonaler Rückversicherungsverband IRV, www.irv.ch
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:41
jpg
largeScale.jpg (292.09K)
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 12:04
jpg
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:11
jpg
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:31
jpg
oprCO2sdi.jpg (54.74K)
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:31
jpg
version 2 uploaded by MarcoStoll on 26 Feb 2014 - 14:17 ...
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:02
jpg
pay12.jpg (159.31K)
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 12:08
gif
version 1 uploaded by MarcoStoll on 27 Feb 2014 - 16:17
jpg
sfcwinds12z.jpg (157.93K)
version 2 uploaded by MarcoStoll on 26 Feb 2014 - 15:08 ...
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 15:08
pdf
Stoll & Leuenberger, 2010: Mesoscale Analysis and Modelling of 23 July 2009 severe thunderstorm. ECAM 2010, Zurich.
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 14:47
png
version 1 uploaded by MarcoStoll on 26 Feb 2014 - 15:07
spacer
This site is managed by the Center for Climate Systems Modeling (C2SM).
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors. Ideas, requests, problems? Send feedback!