Analysis of different wind gust forecast approaches

Transcription

1 Analysis of different wind gust forecast approaches Maria Kurbatova (1,2), Konstantin Rubinstein (1,2), Inna Gubenko (1,2), and Grigory Kurbatov (3) (1) Hydrometeorological Center of Russia, (2) Nuclear safety institute of the Russian Academy of Sciences, (3) Faculty of Physics M.V.Lomonosov Moscow State University

2 Approaches to estimate wind gusts Statistical: Distribution at the location + accurate + can be used for constractions loads calculations - need of long term observations - limited usage for forecast Dynamic-statistical: Predictors of extreme events + can be used for forecasts - need of long term observations - predictors events Dynamical: Using model forecasts calculate constant gust factor Stability dependent gust factor Turbulence estimation Reflection from upper layers.. + can be used for forecasts and uses model products + physicaly based - not always good accuracy

3 Methods for wind gust estimations 1 Gust factor wge k U k depends on surface roughness 2 Brasseur wge max[ U ( z z p )] 1 p p v ( z) q( z) dz g dz z ( z) 0 0 p z v Gust = wind +dispersion wge U 3 wge U 2.4u*; 3 wge U 3 U 3 q q represent maximum wind deviation wge U 2q Brasseur, O., 2001: Development and Application of a Physical Approach to Estimating Wind Gusts. Mon. Wea. Rev., 129, 5 25 K. Born et al., 2012: Wind gust estimation for Mid-European winter storms: towards a probabilistic view. Tellus A 2012, 64, 17471, DOI: /tellusa.v64i Schulz, J.-P Revision of the turbulent gust diagnostics in the COSMO model. COSMO Newslett. 8, Online at:

4 Observation data Synoptic stations reports Ultrasonic thermoanemometers USA-1, placed in Lomonosov Moscow State University, Moscow, Russia, with measurement frequancy: 50 Hz Microwave temperature profiler MTP-5

5 Forecast model WRF-ARW Horizontal resolution: km Vertical resolution: 41 vertical levels Input data: GFS 0.5 deg input forecasts Parametrization: RRTM radiation, MYNN boundary layer, Noah sflay, varying convection

6 Gust occurrence over European part of Russia N gusts 12 N gusts 18 N gusts 15 N gusts 24

7 Wind gusts using different methods Wind speed, m/s u+3q Observation 1.4*U U+3.0*2.4*U* Time u+v2q Brasseur Mean wind speed forecast

8 Accuracy of wind gust estimations using high-frequency observations R Relative ABIAS BIAS u+ 2q u+3q Brasseur u+3.0*2.4u* 1.4*u Wind gust Wind speed

9 Dependence on stability Wind gust, m/s Temperature gradient, º/100м Time Brasseur u+ 2q Observation dt/dz adiabatic

10 Comparison with thunderstorm forecasts Probability of detection False alarm ratio March May August September November December Total gust thunder March May August September November December Total gust thunder ETS (equitable threat score) PSS (Peirce skill score) gust thunder gust thunder March May August September November December Total March May August September November December Total Murmansk region, gust>15m/s, 2016

11 Hybrid method Summer better results Brasseur method (convection) Autumn better results TKE method (mechanical turbulence) We propose a hybrid method. Gust calculation method depends on stability type of atmosphere: wgе U 3 max[ u( z q, p )], Ri 0 ` Ri 0 z 1 p p v( z) q( z) dz g dz z 0 0 ( z) p z v

12 Different methods scores Probability of detection False alarm ratio 1 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0,1 0 March August TKE TKE-2 Schreur PBLH Nakamura Brasseur Hybrid 1,2 1 0,8 0,6 0,4 0,2 0 March August TKE TKE-2 Schreur PBLH Nakamura Brasseur Hybrid ETS (Equitable threat score) Peirce's skill score 0,3 1 0,25 0,2 0,15 0,1 0,05 0-0,05 March August TKE TKE-2 Schreur PBLH Nakamura Brasseur Hybrid 0,8 0,6 0,4 0,2 0-0,2 March August TKE TKE-2 Schreur PBLH Nakamura Brasseur Hybrid

13 Influence of convection parametrization Arakawa- Schubert Tiedtke Zhang- McFarlane Grell-Freitas Betts-Miller- Janjic Kain-Fritsch Mean wind vector error predictability overall accuracy predicted gusts absence Predicted but absent gusts unpredicted gusts predicted gusts Convection parametrization

14 Influence of model resolution Probability of detection False alarm ratio km 6 km km 6 km March May August September November December Total March May August September November December Total ETS (equitable threat score) PSS (Peirce skill score) km 6 km km 6 km M arch M ay August September O ctober Novem ber D ecember T otal M arch A pril M ay Aug ust Septemb er O ctob er N ovem ber D ec em b er Total Murmansk region, gust>15m/s, 2016

15 Storm 29 May 2017 in Moscow

16 Conclusions Different gust estimation methods were applied to WRF-ARW forecasts and evaluated showing none of them is good Methods used for thunderstorm prediction (deep convection diagnostics) doesn t suit for gust prediction New hybrid method was proposed that gives POD around 0.5 throughout a year Increasing model resolution results more gusts and more false alarms Gusts forecasts skills are more sensitive to model configuration (parametrization) then wind speed solely

17 Thank you for the attention! Thanks to European Meteorological Society for providing me with the Young Scientist Travel Award to support of my attendance