Inger-Lise Frogner

High-resolution ensembleprediction of a polar lowdevelopmentJørn Kristiansen, Silje Lund Sørland, TrondIversen, Dag Bjørge and Morten ØdegaardKøltzowPresented by Inger-Lise FrognerKristiansen et al. 2011, Tellus A, 63, No 3jornk@met.no

Polar Lows•Occur frequently but irregularly (on average 4-6 per winter month)•Intense (strong winds and heavy precipitation)•Short-lived (1–2 days)•Meso-scale (100- to 600-km diameter)•Unique to the Polar Regions•Associated with cold air outbreaks•Decays quickly after landfall•Speed of 10–15 m s -1

Forecast challenges• In situ observations are often too sparsefor an adequate analysis of theatmosphere• Only partly compensated by remotesensing data from polar orbiting satellites.• The model representation of moistconvection is crucial, e.g. resolution.• The size and position of the model domainis important.

Main goal of the study• Forecasting potentially severe weather rangingfrom 12 hours to ~2 days.• To show a first example of the feasibility ofapplying a high resolution EPS– A single case 3-4 March 2008– A polar low in the Norwegian Sea off the coastof Northern Norway– Well observed during the IPY-THORPEXcampaign.• Each of the forecasts are initialized at 1800 UTC2 March 2008.

A nested ensemble system• A three-step nested EPS; TEPS->LAMEPS->UMEPS.– Each step produces a 21-member forecast ensemble– The horizontal resolution increases with each step.Step 1 TEPS:•A 21-member configuration of the ECMWF-EPS run at T399(ca. 50 km) up to 72 hours lead time. Global.•Initial state perturbations from singular vectors.•Run at 00 and 12 UTC initial times.Step 2 LAMEPS:•Quasi-hydrostatic HIRLAM at 12 km grid mesh. LAM.•6-hourly 3DVar data assimilation for the control forecast.•TEPS provids perturbed LBCs and initial state perturbations.•Starting at 06 and 18 UTC and runs for 60 h.Step 3 UMEPS:•Non-hydrostatic Unified Model (UM) with 4-km grid resolution.•Dynamical downscaling of each LAMEPS member.

Results

12 UTC3 March18 UTC3 MarchObservations12 UTC4 MarchT+4212 UTC4 MarchUMEPS-big control forecast

MSLPLAMEPS UMEPS-small UMEPS-big12 UTC4 MarchEnsemble mean – contour Ensemble spread - shaded

Probability for severe winds (925 hPa)12 UTC 4 MarchThe differencebetween theforecastprobabilities islargest at T+42when the polarlow is at itspeak.All membershave windspeeds > 20 m/sover a largespatial regionaround theobserved windmaxima.UMEPS-big hasgenerally higherprobabilities thanLAMEPS.The dropsondeobservations areshown as windarrows.•LAMEPS vs UMEPS-small.•U925>20m/s-minor differences•U925>25m/s- the general pattern issimilar but theprobabilities are generallyabout 0.25 largerUMEPS-small.

Probability for precipitation2.5 mm/3h; 0900-1200 UTC 4 MarchMost of the observedprecipitation is wellcaptured by the ensemble.The region with highprobabilities (60-80%)around 65.5N, 9E isassociated with the polarlow.UMEPS-big(colour)Radarobservations(grey)

LAMEPSUMEPSsmallUMEPSbig

Tracking polar lows• Motivation– more easily identify and comparethe polar low(s) of the differentensemble members and tocalculate strike probability maps.– The tracking method is based onthe algorithm of Hodges.• Refined to deal with severalsmallscale vortices.The term track refers to thetrajectory of an individualstorm, not the average path ofmany storms.•Polar lows are usually not present in the analysis =>more challenging to employ the tracking algorithm

X =Observedpolar low”Polar low” paths T+0 to T+60

Conclusions

• Both LAMEPS and UMEPS have thepotential to give a warning of extremeweather 1-2 days ahead.• UMEPS forecast the highest probabilities ofstrong wind speeds and intense precipitation.– This supports the basic assumptions behinddynamical downscaling.• The added value (UMEPS vs LAMEPS) issensitive to both domain size and domainlocation.

• The tracking identified different clusters,one of which was co-located with observedpolar low.• UMEPS-big gave a good estimate of theobserved polar low trajectory• Strike probabilities may be useful forforecasters on duty.• Polar low path tracks and forecast strikeprobability maps are also of high potentialvalue for the public user.

Challenges• An EPS needs to be run regularly forverification and for understanding itsperformance.– Difficult to obtain sufficient data with an ondemand system for all the prepared domains.• Moreover, there may not always be anoptimal choice of domain.– The selection of model domain may itselfintroduce uncertainty in the forecasts.

Future plans• Continued work towardsan operational high resEPS.• Prepare a priori up to 4domains to be selectedfrom on a given day• A good global or largeregional ensembleforecast is needed toguide the choice ofdomain.• UMEPS cannot beexpected to improvecomplete failures in thecoarser resolutionforecasts.• Investigate– the initial stateperturbations with respectto spatial scale and errorgrowth rate– sensitivity to spin-up time– a combination of differentparameterizationschemes/different modelsamong the members toimprove the ensembleforecast distributions

Tracking polar lows•Several•The averagestudies havenumberappliedofatrackstrackingis,algorithm2 to 5(e.g. Hodges, 1994, 1995, 1999) on synoptic-scalesystems.times higher with a weaker threshold•Here (2×10-5 the tracking s-1) than algorithm a stronger TRACK (1×10-4 is further s-developed 1). to track polar lows.•Parameters:•MSLP•Still, there are 6-10 tracks pr member!•relative vorticity at 850 hPa (vor850)•relative vorticity at 925 hPa (vor925)•For •Small the low scale vorticity features threshold with strong (2×10-5 gradients s-1) the are number more easily of tracks identified increases the vorticity slightly fieldwhenthantheinintervalthe MSLPisfield,reducedwhichfromis dominated200-bylarger-scale background flow.1000 km to 200-600 km.•The temporal resolution, or track time step, is 3hours, and track period is the forecast length (60h).•When •To the avoid vorticity short lived threshold features, is it is high required(1×10-4 that a s-1) vortex the must number present decreases over a timewhenthe intervalperiod ofisatreduced.least 12 hours.•The steps:•For each time step the field is decomposed•The by atmospheric a discrete cosine processes transform. aregenerally •A spectral observed filter removes over a the broad shortest andspectrum longest of wave length lengths, scales and and a new the field filtering withpositive and negative anomalies is retained.may not properly separate the different•Features associated with sufficiently strongfeatures vorticity unless maxima as are shown tracked here, by minimizing only the astronger cost function. vortices are retained.• Results for the average number of tracks perensemble member in UMEPS-big:– Filtering intervals; 200-600 km and 200-1000 km– Vorticity thresholds; 2×10-5 and 1×10-4 s-1– vorticity fields; vor850 and vor925 hPaExperiment Parameter VorticitythresholdFilteringinterval1 vor850 2x10-5 200-1000 242 vor850 1x10-4 200-1000 93 vor850 2x10-5 200-600 304 vor850 1x10-4 200-600 65 vor925 2x10-5 200-1000 226 vor925 1x10-4 200-1000 107 vor925 2x10-5 200-600 28Averagenumber ofdetectedtracks

Additional criteria• The average number of tracks is smallest with vor850, 200-600 km and1×10-4 s-1.• However, vor925 tend to identify the systems earlier in the forecasts. Theaverage number of tracks is then 8 pr member.• Further objective criteria were introduced to reduce the number further.– (1) No land requirement: to exclude false disturbances over land the track muststart over sea.– (2) Strong surface winds: the 10 m wind speed must exceed 13.9 m/s (moderategale).– (3) Static stability: the temperature differences between the sea surfacetemperature (SST) and the temperature at 500 hPa must exceed 43 K.– In both (2) and (3) the criterion is evaluated within a 1°radius from a given tracklocation and must be fulfilled over at least 20 % of the track time steps.• These criteria resulted only in a reduction of 4-5 tracks in total.– Interestingly, as in Zahn and Storch (2008), the number of tracks were reduced inLAMEPS (not shown) suggesting a sensitivity to the spatial resolution of themodel.• Since there are several tracks per ensemble member, we have alsoselected the track with strongest mean vorticity.