Federal Department of Home Affairs FDHA Federal Office

  • Slides: 22
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 Statistical adaptation of COSMO predictions with the Kalman Filter COSMO-GM at Krakau September 2008 Vanessa Stauch

Outline • Local forecast corrections with the Kalman Filter • 2 m temperature of

Outline • Local forecast corrections with the Kalman Filter • 2 m temperature of COSMO-LEPS • “ 10 m windspeed” of COSMO-7 & COSMO-2 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 2

Statistical adaptation (Forster 2007) Correct for systematic deviations between NWP model output and local

Statistical adaptation (Forster 2007) Correct for systematic deviations between NWP model output and local observations » identification of an error model Meteo. Swiss Statistical Postprocessing Vanessa Stauch 3

Statistical adaptation with KF fcst corrected fcst first guess obs err t=2 - t=1

Statistical adaptation with KF fcst corrected fcst first guess obs err t=2 - t=1 TDMO Meteo. Swiss Statistical Postprocessing Vanessa Stauch 4

Statistical adaptation with KF fcst corrected fcst first guess obs err t=2 t=2 t=1

Statistical adaptation with KF fcst corrected fcst first guess obs err t=2 t=2 t=1 TDMO Meteo. Swiss Statistical Postprocessing Vanessa Stauch 5

Statistical adaptation with KF fcst corrected fcst first guess obs err t=2 t=1 TDMO

Statistical adaptation with KF fcst corrected fcst first guess obs err t=2 t=1 TDMO Meteo. Swiss Statistical Postprocessing Vanessa Stauch 6

Statistical adaptation with KF fcst corrected fcst first guess obs err t=2 t=1 TDMO

Statistical adaptation with KF fcst corrected fcst first guess obs err t=2 t=1 TDMO » minimising the prediction error variance for noise variances » assumptions on error propagation Meteo. Swiss Statistical Postprocessing Vanessa Stauch 7

2 m temperature of COSMO-7 Kalman filter successful for COSMO-7 (implemented by D. Cattani)

2 m temperature of COSMO-7 Kalman filter successful for COSMO-7 (implemented by D. Cattani) Annual mean error along with its standard deviation and compared to persistence (Lugano 2006) Meteo. Swiss Statistical Postprocessing Vanessa Stauch 8

2 m temperature of COSMO-LEPS Kalman filter successful for COSMO-7 (implemented by D. Cattani)

2 m temperature of COSMO-LEPS Kalman filter successful for COSMO-7 (implemented by D. Cattani) What about COSMOLEPS? Annual mean error along with its standard deviation and compared to persistence (Lugano 2006) Verification on ensemble mean still to do! Meteo. Swiss Statistical Postprocessing Vanessa Stauch 9

KF experiments with COSMO-LEPS » „deterministic“ KF with mean, median, random member Geneva, 01.

KF experiments with COSMO-LEPS » „deterministic“ KF with mean, median, random member Geneva, 01. 02. 2008 » identify best method but also confirm assumptions about „common error structure“ of all members Meteo. Swiss Statistical Postprocessing Vanessa Stauch 10

KF experiments with COSMO-LEPS forecast at t obs at t - 1 forecast at

KF experiments with COSMO-LEPS forecast at t obs at t - 1 forecast at t - 2 forecast at t - 3 forecast at t - 4 forecast at t - 5 12 UTC 12 UTC » account for lead time dependent error if systematic changes observable » optimise one-step ahead prediction error variance » start with latest 00 UTC observations Meteo. Swiss Statistical Postprocessing Vanessa Stauch 11

First results Meteo. Swiss Statistical Postprocessing Vanessa Stauch 12

First results Meteo. Swiss Statistical Postprocessing Vanessa Stauch 12

Verification Meteo. Swiss Statistical Postprocessing Vanessa Stauch 13

Verification Meteo. Swiss Statistical Postprocessing Vanessa Stauch 13

Wind speed of COSMO-7 & COSMO-2 Wiforch (Meteotest, Meteo. Swiss) Fore- and Nowcasting of

Wind speed of COSMO-7 & COSMO-2 Wiforch (Meteotest, Meteo. Swiss) Fore- and Nowcasting of wind power production in complex terrain. » 24 -48 hour power forecasts for energy market (COSMO-7) » 1 -6 hour power forcast for intradaily trading (COSMO-2) » Comparison with Wind. Sim (CFD) Meteo. Swiss Statistical Postprocessing Vanessa Stauch Windturbine at Gütsch 14

COSMO-7 vs COSMO-2 at Guetsch Meteo. Swiss Statistical Postprocessing Vanessa Stauch 15

COSMO-7 vs COSMO-2 at Guetsch Meteo. Swiss Statistical Postprocessing Vanessa Stauch 15

COSMO-7 vs COSMO-2 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 16

COSMO-7 vs COSMO-2 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 16

Kalman Filter corrections winter 07/08 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 17

Kalman Filter corrections winter 07/08 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 17

Kalman Filter corrections spring 08 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 18

Kalman Filter corrections spring 08 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 18

COSMO-7 vs COSMO-2 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 19

COSMO-7 vs COSMO-2 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 19

Conclusions » Local corrections for T 2 m of COSMO-LEPS = work in progress!

Conclusions » Local corrections for T 2 m of COSMO-LEPS = work in progress! » Verification of different error models and assumptions on the error prediction should identify a „best“ set-up » KF for wind speed results in better predictions for the sites investigated. » 3 months testphase from October will evaluate performance in more detail and compare to a linear MOS and the DMOs Meteo. Swiss Statistical Postprocessing Vanessa Stauch 20

Meteo. Swiss Statistical Postprocessing Vanessa Stauch 21

Meteo. Swiss Statistical Postprocessing Vanessa Stauch 21

COSMO-7 vs COSMO-2 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 22

COSMO-7 vs COSMO-2 Meteo. Swiss Statistical Postprocessing Vanessa Stauch 22