Convective-scale data assimilation at the UK Met Office

Transcription

1 Convective-scale data assimilation at the UK Met Office DAOS meeting, Exeter 25 April 2016 Rick Rawlins Hd(DAE) Acknowledgments: Bruce Macpherson and team

2 Contents This presentation covers the following areas Description of UKV DA system Recent changes Next upgrades Future strategy

3 UKV Domain 4x4 1.5x4 4x4 4x x1.5 4x1.5 4x4 1.5x4 4x4 Variable zone 744(622) x 928( 810) points

4 UKV Data Assimilation 8 three-hour assimilation cycles per day Forecasts to t+36 every 3 hours Observation cut-off hh+ 75min Lateral boundaries from hh-3hr run of 17km Global model at DT 03, 09, 15, 21 UTC Lateral boundaries from hh-6hr run of 17km Global model at DT 00, 06, 12, 18 UTC 3DVAR (with FGAT) + IAU for all observations, except Latent Heat Nudging for radar-derived surface rain rate

5 UKV extra observations not assimilated in global model radar-derived surface rain rate (hourly, 5km resolution) visibility from SYNOPs (hourly) T 2m & RH 2m from roadside sensors (hourly) Doppler radial winds (3-hourly) SEVIRI Channel 5 radiances above low cloud high-resolution AMVs from MSG GeoCloud cloud fraction profiles (3-hourly, 5km resolution) zero cloud down to cloud top, missing data below cloud fraction profiles from SYNOPs (3-hourly) zero cloud up to cloud base, missing data above

6 Recent Changes

7 February 15 Upgrade Introduce IASI Migrate to Metop A/B coastal scatterometer products For insertion of GeoCloud data, use model climatology of cloud thickness to decide cloud thickness Peter Francis

8 March 16 Upgrade Covariances Swapped Transform Order with improved representation of spectrum in training data (allows for shorter wind length scales) Cloud Creation of GeoCloud cloud fraction data within OPS (instead of within offline AUTOSAT) Satellite radiances add CrIS and AIRS (to supplement IASI) add ATMS humidity channels

9 Next Upgrades

10 Extended UKV area (summer 2016)

11 Hourly UK-wide 4DVAR Build on Nowcasting Demonstration Project run for 2012 Olympics Improve post-processing products in 0-6hr period Hourly updates to t+12 potential benefit in severe weather

12 UKV 4DVAR system - status Observation cut-off 45 mins Include vertically adaptive grid Perturbation Forecast model resolution 3km Initialisation by digital filter Jc term Doppler radial winds every 10mins AMVs, wind profiler, SEVIRI radiances every 15mins Surface rain rate from radar every 15mins (for LHN) Other data mostly hourly

13 UKV 4DVAR - results (1) Improved very short-period rainfall (better spin-up) FSS v radar at t+2 4DVAR 3DVAR

14 UKV 4DVAR - results (2) Degraded rainfall by t+12 (short cut-off?) FSS v radar at t+12 4DVAR 3DVAR

15 Contents This presentation covers the following areas Description of UKV DA system Recent changes Next upgrades Future strategy

16 Future Strategy

17 Observation strategy Surface exploit 3 rd party networks eg roadside sensors Upper-air support expansion of aircraft humidity introduce MODE-S Satellite focus on cloudy radiances Radar migrate from 2-d surface rainrate to 3d reflectivity exploit novel measurements like refractivity

18 Roadside sensor network T2m / rh2m Visibility

19 MODE-S monitoring 5 receivers Adam Maycock

20 Observation strategy Surface exploit 3 rd party networks eg roadside sensors Upper-air support expansion of aircraft humidity introduce MODE-S Satellite focus on cloudy radiances Radar migrate from 2-d surface rainrate to 3d reflectivity exploit novel measurements like refractvity

21 Operational cycling strategy 3-hourly hourly for UK model (spring 17) Perturb lagged UK ensemble about hourly analyses Possible rapid cycling global assimilation and short-period forecasts to improve lateral boundary data for UK model [ + common verification framework (HiRA) ]

22 Assimilation technique strategy Variational ensemble Operational system in a few years (not this hpc)

23 Convective-scale ensemble DA (Jonathan Flowerdew) Exploring synergy between ensembles and data assimilation at convective scale: Ensembles predict forecast uncertainty given analysis uncertainty Data assimilation reduces analysis uncertainty given forecast error covariances Temperature Can ensemble DA improve UK deterministic/ensemble forecasts? Flow-dependent covariances, convective-scale balances, interaction with orography, cross-system relationships,... Initial trials with conventional observations: Serial ensemble filter with cheap tuning capability 2.2km, 44+1 members, hourly cycle Best configurations competitive with MOGREPS-UK downscaler Pressure

24 SEVIRI and future CsEnDA Recent work adding SEVIRI cloud-affected radiances Cloud important to customers; also prototype for dense vertically-integrating observations of nonlinear variable Channel 5 beneficial; channel 9 more challenging Tuner supports SEVIRI horizontal (vertical) localisation narrower (broader) than conventional observations Further SEVIRI work: Further trials, tuning, diagnostics Vertical localisation moving up/down for channel 9? Apply tuner to observation errors? Inter-variable localisation? Longer-term CsEnDA plans: Assimilation of radar data Localisation, inflation, lateral boundaries, Proposed PhD project on theoretical aspects Incorporation into LFRIC DA 111 km Broad Narrow 28 km

25 Questions?

26 50 Years of Performance Improvements Cloud & Fog 2 nd November Jorge Bornemann

27 Additional slides

28 Surface Temperature CRPS

29 Introducing IASI Configuration: All 4 IASI FOVs 60km thinning 132 of 138 channels used in global (reject high peaking water vapour channels due to residual bias) AAPP coast threshold reduced to 50km All other aspects of configuration same as global model Small benefit to T 2m and rainfall IASI observations for single12z cycle (total no = 24454) Peter Weston