Aspects of short‐term probabilistic blending in different weather regimes

This study explores the potential of regime‐dependent approaches to improve short‐range precipitation forecasts. Probabilistic forecasts have been generated from the radar nowcaster Radar Tracking and Monitoring (Rad‐TRAM) and the convection‐permitting weather prediction model of the Consortium for...

Full description

Saved in:
Bibliographic Details
Published in:Quarterly journal of the Royal Meteorological Society 2014-04, Vol.140 (681), p.1179-1188
Main Authors: Kober, K., Craig, G. C., Keil, C.
Format: Article
Language:English
Subjects:
convection‐permitting model
Earth, ocean, space
Exact sciences and technology
External geophysics
Meteorology
Physics of the high neutral atmosphere
probabilistic nowcasting
seamless prediction
weather situation
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study explores the potential of regime‐dependent approaches to improve short‐range precipitation forecasts. Probabilistic forecasts have been generated from the radar nowcaster Radar Tracking and Monitoring (Rad‐TRAM) and the convection‐permitting weather prediction model of the Consortium for Small‐scale Modeling, Deutscher Wetterdienst (COSMO‐DE) using the neighbourhood method for a 99 day period during summer 2009. The convective adjustment time‐scale was used to classify the days of the investigated period into stratiform, equilibrium and non‐equilibrium convection regimes. The COSMO‐DE forecasts were calibrated using the reliability diagram method and blended with the nowcasts using an additive weighting, where the weighting function varies with lead time according to the time evolution of the conditional square root of ranked probability score (CSRR). The examination of two case studies showed large differences in the calibration and weighting functions for different regimes. Over the entire period, regime‐dependent calibration was found to produce large improvements in reliability in comparison with the uncalibrated forecasts; however, the results were only modestly better than with a single calibration function. The blending procedure successfully combined the nowcast and forecast information, in the sense that the blended forecast was as good as either of the two components, but there was no further gain expected from regime‐dependent weighting if regime‐dependent calibration had already been performed.
ISSN:0035-9009
1477-870X
DOI:10.1002/qj.2220