Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss Quantitative precipitation forecasts in the Alps – first results from the Forecast Demonstration Project MAP D-PHASE Felix Ament, Marco Arpagaus, and Mathias W Rotach MeteoSwiss COSMO-2 (Lyss-Hochwasser, ) Radar Lyss

2 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim 6 times a year Twice a year Every 10 years Verification – rules of the game OBS Models domain averages Most recent forecast, but starting not before +03h hourly accumulations apply warnlevels RR time series Alert time series Period: Summer 2007 (June, July and August) Spatial resolution: 18 target regions in Switzerland Temporal resolution: 3 hour intervals Forecast range: Use most recent forecast, but ignore a certain cut-off time at the beginning of each forecast (default cut-off: 3h)

3 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim Observational data Swiss Radar composite 3 Radar stations 5 min scans accumulated to hourly estimates 1km resolution Gridded rain gauge data Statistical interpolation + elevation correction Daily accumulations RADAR_CAL time series Warn regions averages Hourly accumulations Daily sums equivalent to gridded gauge data RADAR time series Warn regions averages Hourly accumulations Multiplicative correction to achieve match of daily accumulations Spatial average

4 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim Verification of precipitation amount (RADAR_CAL reference) Whole Switzerland, summer 2007, relative BIAS Single target region, 3 hourly resolution, correlation Single target region, summer 2007, relative BIAS global model resolved conv. param. conv. RADAR

5 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim Fuzzy Verification Verification on coarser scales than model scale: “Do not require a point wise match!“ XX XX XX xX X X x MethodRaw DataFuzzyficationScoreExample result Upscaling Average Equitable threat score Fraction Skill Score (Roberts and Lean, 2005) Fractional coverage Skill score with reference to worst forecast XX XX XX xX X X x

6 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim Spatial scale (km) Fuzzy Verification COSMO-2 – COSMO good bad Threshold (mm/3h) - = Fraction skill score Upscaling = - COSMO-7 (7km)COSMO-2 (2.2km) Difference COSMO-7 better COSMO-2 better JJA 2007, Verification against Swiss Radar Composite, 3 hourly accumulations

7 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim Spatial scale (km) Fuzzy Verification COSMO-DE – COSMO-EU good bad Threshold (mm/3h) - = Fraction skill score Upscaling = - COSMO-EU (7km)COSMO-DE (2.8km) Difference COSMO-EU better COSMO-DE better JJA 2007, Verification against Swiss Radar Composite, 3 hourly accumulations

8 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim Alerts – level „yellow“, 3h intervals (Alert level yellow = return frequency of 6 times per year) Relative frequency of an alert (frequency bias) Probability to detect an event (probability of detection) Probability to issue a false alarm (false alarm ratio) and but

9 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim Concept of “Relative Value” Economic point of view: Precautions causes Costs Having no protection results in Losses YesNo YesCC NoL0 Event Precaution Relative Value no forecast Total Cost useful + - useless real forecast perfect forecast

10 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim Relative value – Alert level „yellow“ useful + - useless … against false alarms insensitive … sensitive … (03h, 06h and 12h accumulations, cut-off +03h) global model resolved conv. all models RADAR param. conv.

11 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim COSMO-2 Calibration versus Ensemble Simple calibration of COSMO-2 Multiply COSMO-2 precipitation forecasts by a factor of D-PHASE poor- men's ensemble Issue an alert, if a certain fraction of all models gives a warning 10% 20% 30% 40% 50% 60% 70% 80% 90%

12 First results of D-PHASE | 6th COPS Workshop, February 2008, Hohenheim Conclusions Observational uncertainties (QPE) are not significantly smaller than forecast errors (QPF) – at least for extremes! High resolution models resolving deep convection tend to perform better than models with parameterized convection. This applies for all international models. Probabilistic forecasts are useful for customers. However, the method of choice is still unclear: Simple static recalibration and an uncalibrated ensemble forecasting system perform equally!