B. Casati
(swr00bc@met.rdg.ac.uk) and D. B. Stephenson
Department of Meteorology,
University of Reading, UK.
G. Ross
Met Office, UK.
Some new verification techniques for evaluating spatial precipitation forecasts have been developed and tested on eight selected case studies produced by the Met Office now-casting system NIMROD. Hourly rainfall rate forecasts at 3-hour lead times are verified against NIMROD analyses, both on a spatial grid with 5km resolution.
The forecast-analysis disagreement is decomposed into amount disagreement and pattern-position disagreement. The amount disagreement is quantified using robust statistical measures that compare forecast and analysis empirical cumulative distributions. The amount disagreement is eliminated performing a non-linear recalibration of the forecast. The pattern-position disagreement is then evaluated comparing recalibrated forecast and analysis. Use of categorical scores such as the odds ratio suggests that there is often increasing dependency between the forecasts and the analyses for increasing threshold values. The spatial scales of this extreme dependency (skill) are explored by performing a 2-D wavelet decomposition of binary threshold images.
Keywords: spatial verification, precipitation, recalibration, wavelets, extreme dependency.