640 likes | 651 Views
Explore traditional and fuzzy verification methods for QPF in DWD using radar data. Analyze diurnal cycles, contingency tables, and coupling methods against SYNOP observations for precise weather forecasting.
E N D
Verification of numerical QPF in DWD using radar data-and some traditional verification results for surface weather elements Ulrich Damrath
Outline • Basic data • Traditional verification • Diurnal cycles (observation, forecasts of GME, COSMO-EU and COSMO-DE) starting at different forecast times • Contingency tables • „Fuzzy“-type verification • Application of B. Eberts package • Coupling intensity scale method and fraction skill score • Verification against SYNOP-observations
Basic data • Data of GME (resoluton i192f ~ 40 km), COSMO-EU on the grid of COSMO-DE • Method of „interpolation“:using the data to the nearest gridpoint of COSMO-DE
Basic data for comparison with RADARs(example with some missing observations)
Diurnal cycle of precipitation for forecasts starting at 00 UTC July 2007
Diurnal cycle of precipitation for forecasts starting at 03 UTC July 2007
Diurnal cycle of precipitation for forecasts starting at 06 UTC July 2007
Diurnal cycle of precipitation for forecasts starting at 09 UTC July 2007
Diurnal cycle of precipitation for forecasts starting at 15 UTC July 2007
Diurnal cycle of precipitation for forecasts starting at 18 UTC July 2007
Diurnal cycle of precipitation for forecasts starting at 21 UTC July 2007
Diurnal cycle of precipitation for forecasts starting at 00 UTC August 2007
Diurnal cycle of precipitation for forecasts starting at 03 UTC August 2007
Diurnal cycle of precipitation for forecasts starting at 06 UTC August 2007
Diurnal cycle of precipitation for forecasts starting at 09 UTC August 2007
Diurnal cycle of precipitation for forecasts starting at 12 UTC August 2007
Diurnal cycle of precipitation for forecasts starting at 15 UTC August 2007
Diurnal cycle of precipitation for forecasts starting at 18 UTC August 2007
Diurnal cycle of precipitation for forecasts starting at 21 UTC August 2007
Traditional point to point verification Scores: FBI, POD, FAR, ETS Thresholds: 0.0, 0.1, 0.5, 1.0, 5.0, 10.0 mm/h
Diurnal cycle of precipitation for forecasts starting at 00 UTC July 2007 - Scores precip y/n
Diurnal cycle of precipitation for forecasts starting at 00 UTC August 2007 - Scores precip y/n
Diurnal cycle of precipitation for forecasts starting at 00 UTC July 2007 - Scores precip > 5mm/h
Diurnal cycle of precipitation for forecasts starting at 00 UTCAugust 2007 - Scores precip > 5mm/h
Surprise? COSMO-DE: FBI=2.5 POD=0.55 FAR~0.8 COSMO-EU: FBI=0.6 POD=0.1 FAR~0.8 FAR=0.78 FAR=0.83 ... after some simple arithmetic manipulation
Diurnal cycle of precipitation for forecasts starting at 03 UTC July 2007 - Scores precip y/n
Diurnal cycle of precipitation for forecasts starting at 06 UTC July 2007 - Scores precip y/n
Diurnal cycle of precipitation for forecasts starting at 09 UTC July 2007 - Scores precip y/n
Diurnal cycle of precipitation for forecasts starting at 15 UTC July 2007 - Scores precip y/n
Diurnal cycle of precipitation for forecasts starting at 18 UTC July 2007 - Scores precip y/n
Diurnal cycle of precipitation for forecasts starting at 21 UTC July 2007 - Scores precip y/n
Diurnal cycle of precipitation for forecasts starting at 06 UTC July 2007 - Scores precip > 5mm/h
Diurnal cycle of precipitation for forecasts starting at 18 UTC July 2007 - Scores precip > 5mm/h
The framework for fuzzy verification (I) • 1.Select a set of scales with indeces s=1, 2,…, S • and event intensity thresholds with indeces k=1, 2,…, K • over which to compute the fuzzy verification results • 2.For each scale s: • a. collect the gridded forecasts within the window of scale s • surrounding the observation. • b.For each intensity threshold k compute scale-dependent • quantities (, , , , )k according to various decision models • c.For each intensity threshold k compute the desired • verification scores over the domain. and event intensity thresholds • with indeces k=1, 2,…, Kover which to compute the fuzzy verification results. • From: Fuzzy Verification of High Resolution Gridded Forecasts: • A Review and Proposed Framework E. E. Ebert
The framework for fuzzy verification (II) Observation Forecast
Possible properties of a score • Theoretical properties • score should be illustrative and understandable (by each interested person) • score should give a real information of forecast quality on the different scales • monotonic behaviour concerning (see Lorenz 1963, 1969) • scale(best values for large scales) • frequency of occurence(best values for high frequencies of occurence) • Murphys properties of scores • Bias, Accuracy, Skill, Reliability, Resolution, Sharpness, Discrimination, Uncertainty • Practical properties • agreement between subjective and objective judgement • possible help in decision making
... yet another score B. Casati (real scale separation) N. Roberts (summing up over all scales) Problem: Smooth fields give best results concerning rmse-related scores!
x Modification by U. Damrathin order to have the same structure as the FSS B. Casati energy squared Skill Score ... yet another score
... yet another score A couple of B.Casatis‘ and N. Roberts‘ approach: Calculation of a wavelet skill score for each scale, where scale can be got by the wavelet or upscaling method
... yet another score ........
„Fuzzy“-type verification for 12 h forecasts (vv=06 till vv=18) starting at 00 UTC June 2007 (fraction skill score)
„Fuzzy“-type verification for 12 h forecasts (vv=06 till vv=18) starting at 00 UTC July 2007 (fraction skill score)
„Fuzzy“-type verification for 12 h forecasts (vv=06 till vv=18) starting at 00 UTC August 2007 (fraction skill score)
„Fuzzy“-type verification for 24 h forecasts (vv=06 till vv=30) starting at 00 UTC June 2007 (fraction skill score)
„Fuzzy“-type verification for 24 h forecasts (vv=06 till vv=30) starting at 00 UTC July 2007 (fraction skill score)
„Fuzzy“-type verification for 24 h forecasts (vv=06 till vv=30) starting at 00 UTC August 2007 (fraction skill score)
„Fuzzy“-type verification for 12 h-forecasts (vv=06 till vv=18) starting at 00 UTC June 2007 (intensity scale skill score)
„Fuzzy“-type verification for 12 h-forecasts (vv=06 till vv=18) starting at 00 UTC July 2007 (intensity scale skill score)
„Fuzzy“-type verification for 12 h-forecasts (vv=06 till vv=18) starting at 00 UTC August 2007 (intensity scale skill score)
„Fuzzy“-type verification for 24 h forecasts (vv=06 till vv=30) starting at 00 UTC June 2007 (intensity scale skill score)