Kalle Eerola Finnish Meteorological Institute kalle.eerola@fmi.fi Precipitation and cloud forecasts in two HIRLAM versions (RCR and H635) in September 2004 Kalle Eerola Finnish Meteorological Institute kalle.eerola@fmi.fi 07/04/2019
Contents Introduction Precipitation verification of European LAMs from U.K. Met.Off. Comparison of RCR and H635 forecasts Differences between RCR and H635 Accumulated monthly precipitation Verification of T2m, Rh2m and cloudiness Conclusions Introduction: ore or less this Shortly about the U.L.Met.Off verificatiuon to to get an idea how Hirtlam precipitation forecasts compares with other LAMs run in Europe Main part Conclusion from this study 07/04/2019
Precipitation verification of European LAMs by U.K. Met.Off Area: United Kingdom Against UK NIMPROD rain-fall composite Time: since January 2004 LAMs Aladin (France) Hirlam RCR (FMI) Lokall MODEL (DWD) Met. Off. unified mesoscale Model (U.K.) Scores FBI - Frequency Bias Index >1 overestimates, < 1 underestimates ETS – Equitable Threat Score =0 for random hit, =1 for perfect forecast ______________________________________________________________________ | Observed/yes | Observed/no Forecast yes | a | b Forecast no | c | d ______________|_________________|_______________________________________ FBI – Frequency Bias Index = (a+b)/(a+c) measures the event frequency, corresponds bias in categocal space FBI = 1 for perfect forecast < 1 model underestimates the precipitation > 1 model overestimates the precipitation ETS – Equitable threat score = a – R(a)/(a+-R(a)), where R(a) = (a+b)*(a+c)/(a+b+c+d) Modified version of threat score Measures the fraction of all events forecast and/or observed that were correctly diagnosed accounting for the hits that would occur purely due to random chance ETS = 0 for random hit = 1 for perfect forecast 07/04/2019
FBI and ETS over UK for different thresholds Results since January 2004 Hirlam = RCR at FMI (~6.3.0) Frequency bias: All model overestimate weak precipitation Hirlam underestimates moderate/strong precipitation Other models overestimate them Equitable Threat Score Hirlam: very weak rain: score lower Skill decreases in all model as threshold increases Results since January 2004 Hirlam = RCR at FMI (~6.3.0) Frequency bias: All model overestimate weak precipitation Hirlam underestimates moderate/strong precipitation Other models overestimate them Equitable Threat Score Hirlam: very weak rain: score lower Skill decreases in all model as threshold increases 07/04/2019
Results of tests between RCR and Hirlam 6.3.5 RCR – Regular Cycle with the Reference RCR = Hirlam 6.2.1 + changes ≈ Hirlam 6.3.0 Earlier no one ran with reference no good idea how the reference works Operational at FMI Resolution 0.2º x 0.2º, 40 levels Archived at ECMWF, available for Hirlam community Products available for Hirlam community in near-real time on WEB-pages Available a mod. set to run parallel to RCR (will be included into the reference ) Possible to test against RCR products A control run already exists RCR products are archived at ECMWF I have made a set that enables running parallel run to RCR Simplifies the parallel testing, because the control run has already been made. However I had to re-run the verification, which is rather time-consuming RCR – Regular Cycle with the Reference RCR = Hirlam 6.2.1 + changes ≈ Hirlam 6.3.0 Earlier no one ran with reference no good idea how the reference works Operational at FMI Resolution 0.2º x 0.2º, 40 levels Archived at ECMWF, available for Hirlam community Products available for Hirlam community in near-real time on WEB-pages Available a mod. set to run parallel to RCR (will be included into the reference ) Possible to test against RCR products A control run already exixts 07/04/2019
Main differences and similarities between RCR and H635 H635 uses same observations, boundaries, boundary strategies and extra observations as RCR Same area and resolution in horizontal and vertical Differences New release of HIRVDA (mainly technical) Modified water vapour saturation below freezing First Aid Kit by Laura Rontu Tanquay-Ritchie scheme of temperature in SL-scheme Rotation of surface stress vector Physics-dynamics coupling Modified melting of soil ice Smoothed topography New HIRVDA: mainly technical Modified water vapour saturation below freezing: Mixed phase calculation between water and ice to -23C Cloud water mixed ice and water 0C < T < -23C Reduce cloud emissivity in clouds with small amount of water reduce down swelling long-wave radiation from these clouds surface temperature reduced First Aid Kit Immediate solutions to a number of problems in physics Remove inconsistencies Meteorological effect mainly via surface scheme Tanguay-Ritchie semi-Lagrangian correction Smoother discretization for temperature equation Reduce numerical noise, but also improves scores Rotation of surface stress vector Aim to correct the slow filling of cyclones Turning the surface stress vector in clockwise direction have a positive effect on forecasts Direction of surface stress depends on stability Physics-dynamics coupling Physical tendencies of the previous time step are interpolated at the dep. point of the SL-trajectory The total tendencies from parameterization are averaged along the trajectory Smoother fields, especially precipitation Modified melting of soil ice Smoothed orograpgy smoother precipitation over steep topography 07/04/2019
Monthly precipitation: Europe Rather similar in RCR and H635, compares well to gauge-based analysis GPCC = The Global Precipitation Climatology Centre H635 more precipitation than RCR (Scandinavia, Alps,…) H635 has smoother structure Smoothed orography Tanquay-Ritchie SL changes Dynamics-physics coupling Taken from 24-48 hour forecasts, ie the second day Rather similar in RCR and H635, compares well to gauge-based analysis GPCC = The Global Precipitation Climatology Centre H635 more precipitation than RCR (Scandinavia, Alps,…) H635 has smoother structure Smoothed orography Tanquay-Ritchie SL changes Dynamics-physics coupling Good news that provides more rain, but we don’t know from this, is it in right place, ie. In moderate/heay rain, which are underestimated in RCR 07/04/2019
Monthly precipitation: Scandinavia Rainy month in Finland: observed is 100-300% of the normal In H635 more precipitation, fits better to observed (southern Finland, Kainuu), H635 has smoother sctructure Taken from 24-48 forecasts, ie. The second day Rainy month in Finland: observed is 100-300% of the normal H635 more precipitation, fits better to observed (southern Finland, Kainuu), H635 has smoother sctructure 07/04/2019
Convective part of precipitation As earlier, the 24-48 hour forecasts, ie. second day In H635 the structure is much smoother than in RCRa Smoothed orography Tanquay-Ritchie SL changes Dynamics-physics coupling Especially over the mountains As earlier, the 24-48 hour forecasts, ie. second day In H635 the structure is much smoother than in RCRa Smoothed orography Tanquay-Ritchie SL changes Dynamics-physics coupling Especially over the mountains 07/04/2019
Conclusion from precipitation H635 produces more precipitation than RCR The accumulated monthly precipitation has a smoother structure in H635 than in RCR Convective part is especially smoother H635 produces more precipitation than RCR The accumulated monthly precipitation has a smoother structure in H635 than in RCR Convective part is especially smoother Now we go to the other subject: T2m, Rh2m and cloudiness in RCR and H635 07/04/2019
Diurnal variation in surface (near-)parameters Here we time-series of for two-day forecasts from RCR and H635 The reason for presenting this picture is that it motivates to look separately forecasts valid at day-time and at night-time 07/04/2019
Station verification for EWGLAM stations T2m At night almost all negative bias removed During day negative bias reduced RH2m Diurnal cycle in RCRa In H635 almost unbiased T2m At night almost all negative bias removed During day negative bias reduced RH2m Diurnal cycle in RCRa In H635 almost unbiased More information can be received if we look the bias station by station We can get the geographical distribution of the errors 07/04/2019
At night: Bias of T2m, Rh2m and cloudiness First look at T2m Negative bias decreased Even positive bias in some area Over the Alps still positive Rh2m Positive bias (too humid) decreased Of course the temperature also affects the relative humidity: H635 warmer lower relative humidity Too dry: southern Europe and America Cloudiness There seems to be a conflict between cloudiness and T2m: during night too cold and too cloudy in RCR Possible problems: observing the clouds in dark, especially thin and high clouds The distribution of cloud observations: U-shape, many totally cloudy or totally clear situations I don’t know how to interpret the results 07/04/2019
During day: Bias of T2m, Rh2m and cloudiness Cold bias still is in H635 but it is reduced In America almost disappeared Rh2m: Positive (too humid bias has reduced in central and northern Europe Even too much in southern Europe, where in H635 is now negative bias Cloudiness Negative bias almost everywhere ie. too little clouds In H635 there is now too 07/04/2019
Conclusions I Accumulated monthly precipitation U. K. Met. Off. Precipitation verification Frequency bias: All model overestimate weak precipitation Hirlam underestimates moderate/strong precipitation, while other models overestimate them Equitable Threat Score Hirlam: very weak rain: score lower Skill decreases in all model as threshold increases Accumulated monthly precipitation H635 produces more precipitation than RCR The accumulated monthly precipitation has a smoother structure in H635 than in RCR Convective part is especially smoother 07/04/2019
Conclusions II At Night Daytime T2m: negative bias decreased, even positive bias in some are Rh2m: Positive bias (too humid) decreased Too dry: southern Europe and America Cloudiness. Difficult to interpret the results Daytime T2m: Cold bias still is in H635 but it is reduced In America almost disappeared Rh2m: Positive (too humid bias has reduced in central and northern Europe Even too much in southern Europe, where in H635 is now negative bias Cloudiness: difficult to interpret 07/04/2019