KENDA at MeteoSwiss. Operational implementation and results. Daniel Leuenberger and the COSMO-NExT team MeteoSwiss, Zürich, Switzerland

Transcription

1 Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss KENDA at MeteoSwiss Operational implementation and results Daniel Leuenberger and the COSMO-NExT team MeteoSwiss, Zürich, Switzerland COSMO General Meeting, September 2016

2 COSMO-NExT ECMWF-Model 9 to 18 km gridspacing 4 x per day COSMO km gridspacing 8 x per day 1 to 2 d forecast COSMO-E 2.2 km gridspacing 2 x per day 5 d forecast 21 members later KENDA 2

3 Operational KENDA Implementation General 40 ensemble members, additional deterministic analysis Hourly cycle LBC perturbations from 30-42h old ECMWF ENS perturbations centered around the latest HRES forecast LETKF Adaptive localization Adaptive multiplicative covariance inflation Relaxation to prior perturbations (RTTP) 3

4 Operational KENDA Implementation Additionally LHN in every ensemble member Snow analysis for each ensemble member once per day Runtime for one cycle Component Run Time # of GPU # of CPU cores COSMO (fc step) 6.0 min LETKF (update step) 2.5 min

5 Observations Location of observations collected during one month Radiosondes SYNOP (p s only) Windprofiler Aircrafts Radar (with LHN) 5

6 Additional Perturbations LETKF assumes bias-free model As this is far from true, use ad-hoc ways to account for model error (covariance inflation, RTTP) Spread is still too small, use more perturbations: Stochastic Perturbations of Physical Tendencies (SPPT) Soil moisture perturbations (SMP) 6

7 Analysis Performance First guess spread-rmse ratio, verification period: Temperature Relative Humidity Wind Speed underdispersive overdispersive CTRL SPPT SPPT+SMP 7

8 Comparison with Nudging Lead time 1-12h, verification period: RMSE m Temperature FSS (6.6km scale) of 3h precip sums 0.1 mm/3h 0.1 mm/3h 1 RMSE m Dewpoint Temperature 5mm/3h 5mm/3h 1 00:00 06:00 12:00 18:00 00:00 CTRL SPPT+SMP Nudging 8

9 Benefit of additional Perturbations Lead time 1-12h, verification period: m Temperature 2m Dewpoint Temperature Precipitation CTRL SPPT+SMP 9

10 COSMO-E outperforms COSMO-LEPS RPSS of 12h precipitation sum and T_2M, verification period: MAM 2016 Precipitation 2m Temperature COSMO-E COSMO-LEPS 10

11 Summary COSMO-E and KENDA operational since 19th March 2016 COSMO-E outperforms COSMO-LEPS in most variables and seasons, out to +120h! KENDA deterministic analyses and forecasts similar to Nudging analyses and forecasts KENDA still slightly worse than nudging for summer TD_2m and precipitation SPPT and Soil Moisture Perturbations consistently improve KENDA and forecasts started therefrom, benefit larger in Summer than in Winter period 11

12 Outlook Assimilation of T_2M and TD_2M Deterministic analysis with x=1.1km Assimilation of MODE-S observations Assimilation of more remote-sensing observations (Radar, Lidar, Radiometer, Satellites) 12

13 Thanks to DWD: Andreas Rhodin, Christoph Schraff, Harald Anlauf, Hendrik Reich, and Roland Potthast to the COSMO-NExT team: André Walser, Guy de Morsier, Jean-Marie Bettems, Marco Arpagaus, Oliver Fuhrer, Petra Baumann, Philippe Steiner, Pirmin Kaufmann, Simon Förster, Xavier Lapillonne, and Yann Lepoittevin for your attention! 13

14 Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss MeteoSwiss Operation Center 1 CH-8058 Zurich-Airport T MeteoSvizzera Via ai Monti 146 CH-6605 Locarno-Monti T MétéoSuisse 7bis, av. de la Paix CH-1211 Genève 2 T MétéoSuisse Chemin de l Aérologie CH-1530 Payerne T

15 Analysis Performance RMSE of Analysis and First Guess Temperature Relative Humidity Wind Speed CTRL SPPT SPPT+SMP 15

16 Additional Perturbations LETKF assumes bias-free model. As this is far from true, use ad-hoc ways to account for model error (covariance inflation, RTTP) Spread is still too small, use more perturbations: Stochastic Perturbations of physical tendencies (SPPT) Soil moisture perturbations Increase in soil moisture spread by ca 80% 16