Revisiting the synoptic -scale predictability of severe European … · 2017-08-07 · bias when focus obs events only. Large variability between storms. Lowest skill for Yuma (1997)

KIT – University of the State of Baden-Wuerttemberg and National Research Center of the Helmholtz Association

Institute of Meteorology and Climate Research – Troposphere Research (IMK-TRO)

www.kit.edu

www.pravda-tv.com www.web.dewww.woksat.info

Revisiting the synoptic-scale predictability of severe European winter storms using ECMWF ensemble reforecasts

Florian Pantillon, Peter Knippertz and Ulrich CorsmeierKarlsruhe Institute of Technology (KIT), Karlsruhe, Germany

Florian Pantillon, Peter Knippertz and Ulrich Corsmeier2

Collaborative Research Center "Waves to Weather"

Revisiting the synoptic-scale predictability of severe European winter storms

Upscale error growth

Cloud-scale uncertainty

Predictability of local weather


Motivation and strategy

Predictability of wind gustsin winter storms over central Europe

Storms = destructive natural hazardPredictability = Multi-scale problem

Synoptic scale global ensemble forecasts

Mesoscale regional ensemble forecasts

Turbulent scale Doppler wind lidar observations

Synoptic scaleO(1000 km)

Turbulent scaleO(0.1-1 km)

MesoscaleO(10-100 km)



Synoptic scale: model data

ECMWF ensemble reforecastRetrospective forecast: 20 years with homogeneous model versiondx=30 km, 10 days, 10+1 members, no stochastic physics, 2 runs/week

Selection of storms: XWS open access catalogue (Roberts et al. 2013, NHESS)

52 most severe European storms 1979-2013Available online http://www.europeanwindstorms.org

25 storms (1995-2015) x 3 forecasts/storm x 11 members/forecast Comparison ERA-Interim reanalysis (retrospective analysis) dx=80km

d-9

d-2 d-0

WeTuMoSuSaFrThWeTuMo

d-6



Three metrics to assess predictability

1. Track and intensity storm dynamics

2. Strength of wind gusts storm impact

3. Area covered by gusts storm warnings



Dynamics: track and intensity

1. Tracking: algorithm based on Laplacian MSLP (Pinto et al. 2005, MetZ)

2. Identification: two methods = first occurrence or maximum intensity The two methods diverge for lead times beyond 3 days!


Identified tracks of ex-hurricane Lili in the 6-day ensemble reforecast init on 22 October 1996


Bias in longitude

Results for the ensemble average

Difference reforecast – ERA-InterimBias - in longitude (too slow) > day 3Bias + in MSLP (too weak) > day 4

For severe storms reaching Europe!

Large variability between stormsStrongest bias for storm Gero (2005)= deepest storm in sample (948 hPa) no systematic link with intensity…


symbol = median per stormblack curve = median per lead time

Gero

Bias in MSLP


Results for individual members

Number of members with actual storm

All members (11/11) at day 1

Most members until day 4 Average meaningless beyond

At least 1 member until day 10 Potential for early warning

‼but focus on observed events (hits) without accounting for false alarms‼


1 symbol = 1 stormblack curve = median per lead time

Distance < 10° △MSLP < 10 hPa


Impact: Storm Severity Index (SSI)

vmax daily maximum wind gustsv98 local 98th climatological percentile (in reforecast or ERA-Interim)

Integral over central Europe = measure of storm severity


(Klawa and Ulbrich 2003, NHESS; Leckebusch et al. 2007, GRL)

Daily wind gusts and SSI for storm Lothar on 26 December 1999 in ERA-Interim


Results for the storms

Difference (log!) reforecast – ERA-I

Drop order of magnitude by day 4

No drift in top 1% SSI dataset

Large variability between storms

Predictability impactrestricted to days 1-3


storms

top 1% SSI



Warnings: Extreme Forecast Index

Motivation predicted < observed extremesIdea measure extremes in model world(Lalaurette 2003, QJRMS; Zsoter et al. 2006)

Extreme Forecast Index (EFI)Uses distribution of ensemble forecastGives deviation from model climate0 = model climate +/-1 = extreme

!!!many hits but also false alarms!!! look for optimal threshold in EFI trade-off with Heidke Skill Score(Petroliagis & Pinson 2014; Boisserie et al. 2016)


6-day reforecast of storm Lotharand analysis on 26 December 1999


Results for strong gusts

EFI to predict gusts > 98th clim. percentileWhole dataset: skill until day 10Storms: higher skill bias when focus obs events only

Large variability between stormsLowest skill for Yuma (1997)= smallest storm in sample no systematic link with size…High skill at day 10 for Xynthia (2010)but different origin of predicted storm favourable environment?



dotted curve = whole 20-year dataset

Yuma

Xynthia


Summary

Synoptic-scale predictability of severe European winter stormsECMWF ensemble reforecast vs. ERA-Interim reanalysis25 most severe European storms 1995/96-2014/15 (XWS catalogue)

3 metrics

Ambiguous identification, systematic biases, few members > 3-4 days Skill for predicting gusts using whole ensemble distribution > 1 week

High variability between storms and no systematic link with dynamics Too few cases? Nature of extreme events?


Track &Intensity

ExtremeForecastIndex

Storm SeverityIndex

Paper online: Pantillon et al (2017), NHESSD, in review, doi:10.5194/nhess-2017-122


Down to the mesoscale

Operational forecast storm Egon 13 January 2017

DWD operationalensemble forecastCOSMO-DE-EPS

dx=2.8km, 27h, 4/day20 members= 4 global modelsx 5 params physics(setup 2011-2017)

Preliminary results:storm Egon 13 Jan 2017(possible sting jet!?)

Gusts underdispersive Stat. postprocessing

(coll. Sebastian Lerch)



Down to the turbulent scale



Results for the whole dataset

October–March 1995/96–2014/15including stormy & non-stormy days Brier Skill Score decomposed asBSS = 1 – reliability – resolution

Intense events (top 5% SSI) Reliability close to zero (perfect) frequency = 5% by definitionResolution increases with lead timeBrier Skill Score > 0 until day 8

Extreme events (top 1% SSI) Large sampling uncertainty dataset too limited for extremes


skill

Documents

Revisiting the synoptic -scale predictability of severe European … · 2017-08-07 · bias when focus obs events only. Large variability between storms. Lowest skill for Yuma (1997)