Presentation is loading. Please wait.

Presentation is loading. Please wait.

14th Cyclone Workshop Brian Ancell The University of Washington

Similar presentations


Presentation on theme: "14th Cyclone Workshop Brian Ancell The University of Washington"— Presentation transcript:

1 14th Cyclone Workshop Brian Ancell The University of Washington
Comparison of Two Operational Ensemble Systems in the Prediction of a Pacific Northwest Windstorm 14th Cyclone Workshop Brian Ancell The University of Washington

2 The Classic Western WA Windstorm

3 The Classic Western WA Windstorm

4 The Classic Western WA Windstorm

5 The Classic Western WA Windstorm

6 The Classic Western WA Windstorm

7 The Classic Western WA Windstorm

8 A Forecast Challenge Arises at 12 UTC November 16, 2007
University of Washington Mesoscale Ensemble (UWME) - 8 members - MM5 model at 36-km grid spacing, 32 levels - Initial and lateral boundary conditions from global centers - No data assimilation - “Best guess” is mean IC run with mean BC (Centroid) University of Washington Ensemble Kalman Filter (UW EnKF) - 80 members - WRF model at 36-km grid spacing, 37 levels - Initial and lateral boundary conditions drawn from Gaussian distribution - Assimilates surface, aircraft, sounding, and cloud-track wind data - “Best guess” is forecast mean

9 UWME Centroid 00-hr

10 UWME Centroid 06-hr

11 UWME Centroid 12-hr

12 UWME Centroid 18-hr

13 UWME Centroid 24-hr

14 UWME Centroid 30-hr

15 UW EnKF Forecast Mean 00-hr

16 UW EnKF Forecast Mean 06-hr

17 UW EnKF Forecast Mean 12-hr

18 UW EnKF Forecast Mean 18-hr

19 UW EnKF Forecast Mean 24-hr

20 UW EnKF Forecast Mean 30-hr

21 UW EnKF vs. UWME UW EnKF UWME

22 Various EnKF Members (30-hr)

23 Questions Why the difference in cyclone track? Which “best guess” should a forecaster trust? Is there a way to use the EnKF to increase the short-term predictability of such a high-impact event?

24 Breaking down the differences…
Issue #1) Model error?

25 Breaking down the differences…
Issue #1) Model error? - Nope!

26 Breaking down the differences…
Issue #1) Model error? - Nope! Issue #2) Let’s be fair! - Centroid (UWME) vs. Mean (EnKF)

27 EnKF Centroid 00-hr

28 EnKF Centroid 06-hr

29 EnKF Centroid 12-hr

30 EnKF Centroid 18-hr

31 EnKF Centroid 24-hr

32 EnKF Centroid 30-hr

33 Breaking down the differences…
Issue #1) Model error? - Nope! Issue #2) Let’s be fair! - Centroid (UWME) vs. Mean (EnKF) Issue #3) Past performance

34 (From Torn and Hakim 2008)

35 Verification 1) Past performance – favors UWME (no windstorm)

36 Verification 1) Past performance – favors UWME (no windstorm)
2) Suboptimal LBC – favors UWME (no windstorm)

37 Verification 1) Past performance – favors UWME (no windstorm)
2) Suboptimal LBC – favors UWME (no windstorm) 3) Lack of observations in EnKF – favors UWME (no windstorm)

38 What actually happened?
Verification 1) Past performance – favors UWME (no windstorm) 2) Suboptimal LBC – favors UWME (no windstorm) 3) Lack of observations in EnKF – favors UWME (no windstorm) What actually happened?

39 What actually happened?
Verification 1) Past performance – favors UWME (no windstorm) 2) Suboptimal LBC – favors UWME (no windstorm) 3) Lack of observations in EnKF – favors UWME (no windstorm) What actually happened? No windstorm!

40 Forecast Tools for Better Predictability
EnKF extended forecasts (out to 48-hr) run at 00 and 12 UTC - Is there a way to increase the predictability of potentially serious PNW windstorms in the interim?

41 Forecast Tools for Better Predictability
Key: Compare members to observations at locations where forecast is sensitive to error First choice: Adjoint sensitivity

42 Forecast Tools for Better Predictability
Key: Compare members to observations at locations where forecast is sensitive to error First choice: Adjoint sensitivity Simplified adjoint physics Significant nonlinearity

43 Forecast Tools for Better Predictability
Key: Compare members to observations at locations where forecast is sensitive to error First choice: Adjoint sensitivity Ensemble sensitivity! - Incorporates nonlinearity - Easily calculated Simplified adjoint physics Significant nonlinearity

44 Ensemble Senstivity 00-hr 500 GPH, Absolute Vorticity
Ensemble Sensitivity to 500 GPH mb/m 00-hr Response function = Western Washington SLP at 30-hr

45 Correlation of P with Forecast Error

46 Correlation of P with Forecast Error

47 Correlation of P with Forecast Error
Location of projection Error (mb) All 70% 50% 30% 10% Original error all members = 3.8

48 Summary Forecast troubles arose when two operational ensemble systems suggested diverging cyclone tracks associated with a potential western Washington windstorm Differences were likely due to the limited-area configuration (suboptimal boundary conditions) and fewer observations in the EnKF Very large ensemble sensitivity existed for this case, resulting in nonlinear perturbation evolution and large forecast spread Using an independent forecast as “truth”, results suggest a better subset of EnKF members may be produced using observations and ensemble sensitivity guidance


Download ppt "14th Cyclone Workshop Brian Ancell The University of Washington"

Similar presentations


Ads by Google