Federal Department of Home Affairs FDHA Federal Office

  • Slides: 19
Download presentation
Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology Meteo. Swiss

Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology Meteo. Swiss Accounting for Change: Local wind forecasts from the highresolution model COSMO-GM, September 2011, ECAC & EMS, September, 14 th. Roma 2010 Vanessa Stauch (Meteo. Swiss)

Spatial verification of wind speed Model topography fairly complex Model performance pretty good Local

Spatial verification of wind speed Model topography fairly complex Model performance pretty good Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 2

Wind speed SYNOP verification @CH COSMO-2, all 8 runs, May-July 2011, averaged over ~85

Wind speed SYNOP verification @CH COSMO-2, all 8 runs, May-July 2011, averaged over ~85 stations in Switzerland Mean error (m/s) M e a n o b s/ fc st ( m /s ) Daytime (UTC) Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch Daytime (UTC) 3

Spatial verification of wind speed Model topography fairly complex Model performance pretty good Model

Spatial verification of wind speed Model topography fairly complex Model performance pretty good Model performance at some stations rather poor Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 4

Aim of this work Make use of the strength of COSMO-2 wind forecasts Use

Aim of this work Make use of the strength of COSMO-2 wind forecasts Use observations to statistically correct for local forecast errors Derive probabilistic information from deterministic forecasts to provide comprehensive forecast information to the user >> Augment the forecasts of COSMO-2 Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 5

The postprocessors’ dilemma September 04, 2009 ********* COSMO News No. 39 ********* Dear internal

The postprocessors’ dilemma September 04, 2009 ********* COSMO News No. 39 ********* Dear internal COSMO Clients With the 09 UTC run of Thursday, 10. 9. 2009, we will introduce new version 4. 7. 4 we will introduce the newthe version… of the COSMO model with a parametrization of the sub-grid scale orographic (SSO) drag both in COSMO-2 and COSMO-7. What is this SSO scheme: The model orography is a smoothed version of the real orography containing much less details, i. e. less deep valleys and lower mountain peaks. This leads to an underestimation of the drag exerted by the orography on the atmosphere. The SSO-scheme (implemented following Lott and Miller, 1997, QJRMS) improves this situation with a parametrization of both the form drag and gravity wave drag components. The verification of the new version during two 3 week periods (in winter and summer) shows the main impact the wind: 10 m-wind, gusts and wind in the boundary thethat main impact is on is theonwind forecasts… layer. The overestimation of the 10 m-wind on the Swiss Middleland is reduced on many stations as well also the overestimation in the boundary layer. Over the full domain of COSMO-7 the bias of 10 m-wind is reduced substantially. The gusts are reduced at some stations over the Swiss Middleland over the Alps, i. e. leading to a bit higher underestimation (except in COSMO-2 the stations near lakes with a reduced overestimation). Detailed verification results are available here… Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 6

Accounting for change “MOS with reforecasts” Length of database ~ „global MOS “: e.

Accounting for change “MOS with reforecasts” Length of database ~ „global MOS “: e. g. MOSMIX at DWD, multiple linear regression based on global NWP models (GME and IFS) “global MOS” complexity of statistical correction “UMOS”: ‘updateable’ MOS of Canadians, weighting when model chsnges “KF”: Kalman Filter based estimation, online update temporal flexibility (e. g. when model error changes) Need for models with few parameters Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 7

COSMO-2 @ Meteo. Swiss +24 h +48 h +24 h Some uncertainty information? Local

COSMO-2 @ Meteo. Swiss +24 h +48 h +24 h Some uncertainty information? Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 8

Logistic regression: forecast probabilities for one threshold Obs Fcst Sample climatology threshold p(obs)!=p(fcst) Wind

Logistic regression: forecast probabilities for one threshold Obs Fcst Sample climatology threshold p(obs)!=p(fcst) Wind speed Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 9

Logistic regression: forecast probabilities for one threshold probability 1 0 . … … ….

Logistic regression: forecast probabilities for one threshold probability 1 0 . … … …. . . . Obs Fcst (p) Wind speed Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 10

Extended logistic regression Obs Fcst Sample climatology threshold Add thresholds as predictor, estimate one

Extended logistic regression Obs Fcst Sample climatology threshold Add thresholds as predictor, estimate one additional parameter Wind speed Wilks 2009 Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 11

Extended logistic regression Results in full probability distributions for each forecast Obs Fcst Wind

Extended logistic regression Results in full probability distributions for each forecast Obs Fcst Wind speed Wilks 2009 Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 12

Set up for estimation & correction • Forecasts and observations for all Swiss stations

Set up for estimation & correction • Forecasts and observations for all Swiss stations for wind speed and wind gusts for two power stations� • Length of training period: 3 months • Estimation of parameters daytime dependent, once each day • Predictors: wind forecast and 4 thresholds (0. 2, 0. 4, 0. 6, 0. 8) • Evaluate deterministically and probabilistically Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 13

Results: bias correction wind speed Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch,

Results: bias correction wind speed Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 14

Results: probabilistic verification Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. stauch@meteoswiss.

Results: probabilistic verification Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 15

Results: evaluation of distribution Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa.

Results: evaluation of distribution Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 16

Results: bias correction for vmax Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch,

Results: bias correction for vmax Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 17

Results: variability of parameters Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa.

Results: variability of parameters Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 18

Conclusions and outlook • In case no EPS is available, extended logistic regression can

Conclusions and outlook • In case no EPS is available, extended logistic regression can help augment a deterministic model • Ext. log. regression helps removing the model bias and decreases the standard deviation of the error • Ext. log. regression is a promising candidate for model output statistics for many parameters (independent from their distribution) • To do: further investigation on model selection, length of training period, number of thresholds for estimation, … Local wind forecasts | ECAC/EMS 2011, Berlin Vanessa Stauch, vanessa. [email protected] ch 19