1 / 26

Virginia Poli, Pier Paolo Alberoni, Frank S. Marzano

Toward an ensemble nowcasting system: describing the steering field’s uncertainty in an advection scheme for radar images. Virginia Poli, Pier Paolo Alberoni, Frank S. Marzano. Forecast depends on the knowledge of initial conditions.

orinda
Download Presentation

Virginia Poli, Pier Paolo Alberoni, Frank S. Marzano

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Toward an ensemble nowcasting system:describing the steering field’s uncertainty in an advection scheme for radar images Virginia Poli, Pier Paolo Alberoni, Frank S. Marzano

  2. Forecast depends on the knowledge of initial conditions Lack and/or error in initial conditions propagate on the predicted fields “Evolution” uncertainties “Dynamical” uncertainties See poster 9.3 Radar images For advection scheme this is mainly due to uncertainty embedded in the steering field Steering field generation Numerical advection scheme Explore the uncertainty originated by steering field generation mechanism and understand how it propagates in the prediction Motivation Nowcasting based on advection of radar images

  3. First step of cross-correlation is made on an area centered on radar image • Parameters involved: • dimension and position of cross-correlation domain • reflectivity value of the first layer Radar images Cross-correlation analysis Generation of single motion vectors associated to each reflectivity level considered • Geometrical segmentation of two subsequent radar reflectivity fields in order to track different targets at different scales examined • Parameters involved: • reflectivity values used to bound reflectivity areas Steering vectors spatialization • Vectors associated at each reflectivity level are merged using a method based on successive correction analysis. Retrieved vectors are associated to a radius that circumscribes an area in which they are supposed to have a certain influence • Parameters involved: • influence radius Steering field Semi-lagrangian advection Forecast images Determination of motion field

  4. Ensemble nowcasting IDEA create an ensemble changing each fixed parameter apart trying to understand what is the one that better represents forecast field variability Modifications are performed by a random selection of parameters themselves Generation of a large set of motion fields starting from these different configurations Ensemble generated by running semi-lagrangian advection algorithm with these different input steering fields

  5. Introduction to the case study 03/04/2006 13:00 GMT

  6. 03/04/2006 13:15 GMT

  7. 03/04/2006 13:30 GMT

  8. 03/04/2006 13:45 GMT

  9. 03/04/2006 14:00 GMT

  10. 03/04/2006 14:15 GMT

  11. 03/04/2006 14:30 GMT

  12. 03/04/2006 14:45 GMT

  13. 03/04/2006 15:00 GMT

  14. 03/04/2006 15:15 GMT

  15. 03/04/2006 15:30 GMT

  16. 03/04/2006 15:45 GMT

  17. 03/04/2006 16:00 GMT

  18. [dBZ] Cross-correlation analysis 50.0 Generation of single motion vectors associated to each reflectivity level considered 30.0 0.2 50.0 30.0 0.2 Random choice of cross-correlation domain Radar images Steering vectors spatialization Steering field Semi-lagrangian advection Forecast images

  19. [dBZ] Cross-correlation analysis 48.0 Generation of single motion vectors associated to each reflectivity level considered 29.0 0.0 53.0 30.0 3.0 Random choice of reflectivity thresholds Radar images Steering vectors spatialization Steering field Semi-lagrangian advection Forecast images

  20. Random choice of influence radius Radar images Cross-correlation analysis Generation of single motion vectors associated to each reflectivity level considered Steering vectors spatialization Steering field Semi-lagrangian advection Forecast images

  21. Random parameter: research domain Probabilistic forecast Lead time: 45 minutes Threshold: 20 dBZ Ensemble members: 40 Random parameter: Z thresholds • Observed field forecast time for considered threshold Random parameter: influence radius Total ensemble (120 members)

  22. Probabilistic forecast Lead time: 45 minutes Threshold: 40 dBZ Ensemble members: 40 Random parameter: Z thresholds Random parameter: research domain • Observed field forecast time for considered threshold Random parameter: influence radius Total ensemble (120 members)

  23. Statistical results Forecast lead time: 45 minutes Brier score Brier skill score

  24. Statistical results Total ensemble Brier score Brier skill score

  25. Comments and conclusion • At the present moment: • Preliminary analysis on a convective event with its rainfall structures characterised by different direction and speed of motion • Every changed parameter has a different impact on results • Changing verification threshold results maintain their tendency • Algorithm has an higher sensitivity to the use of random reflectivity levels (better impact on forecasts) • Brier score trend is rapidly decreasing: one of the causes resides in the structures characterized by high reflectivity. They are very localized and following their motion becomes very difficult • Future work: • Extend the verification of this probabilistic approach to a larger number of cases analysing different typologies of evens

  26. Thank you for your attention!

More Related