Spatial Forecast Verification: Image Warping
Author
Summary, in English
In response to a growing need for more informative forecast verification in the
face of gridded verification sets, many new methods have been proposed. While
widely varying in their approaches, the new methods generally fall into two ma-
jor categories of filter and displacement, each of which can be further subdivided.
One of the displacement approaches, a field deformation approach known as image
warping, will be demonstrated here. Results for spatial verification of the spatial
forecast verification Inter-Comparison Project test cases are shown. An initial look
at space-time verification using the image warp is also discussed, with an applic-
ation to NCAR and NCEP 4-km WRF models cases from the 2005 NSSL/SPC
Spring Program. The approach is found to be very useful for obtaining guidance
about forecast performance. Both diagnostic and summary score information can
be gleaned. Initial findings for the space-time approach show that while the NCEP
model has better initial scores, the NCAR models require drastically less deform-
ation to achieve a much higher reduction in error. This is most likely a result of
the NCEP model’s highly over forecasting low-intensity precipitation spatially.
face of gridded verification sets, many new methods have been proposed. While
widely varying in their approaches, the new methods generally fall into two ma-
jor categories of filter and displacement, each of which can be further subdivided.
One of the displacement approaches, a field deformation approach known as image
warping, will be demonstrated here. Results for spatial verification of the spatial
forecast verification Inter-Comparison Project test cases are shown. An initial look
at space-time verification using the image warp is also discussed, with an applic-
ation to NCAR and NCEP 4-km WRF models cases from the 2005 NSSL/SPC
Spring Program. The approach is found to be very useful for obtaining guidance
about forecast performance. Both diagnostic and summary score information can
be gleaned. Initial findings for the space-time approach show that while the NCEP
model has better initial scores, the NCAR models require drastically less deform-
ation to achieve a much higher reduction in error. This is most likely a result of
the NCEP model’s highly over forecasting low-intensity precipitation spatially.
Department/s
Publishing year
2010
Language
English
Publication/Series
NCAR Technical Notes
Links
Document type
Report
Publisher
National Center for Atmospheric Research, Boulder, CO, USA
Topic
- Probability Theory and Statistics
Status
Published
Report number
NCAR/TN-482+STR
ISBN/ISSN/Other
- ISSN: 2153-2400
- ISSN: 2153-2397