Environment Canada s Regional Ensemble Kalman Filter EnKF Workshop 216 Seung-Jong Baek, Luc Fillion, Dominik Jacques, Thomas Milewski, Weiguang Chang and Peter Houtekamer Meteorological Research Division, Environment and Climate Change Canada
214 implementation of the NWP suites EnKF ensemble (GEPS) Regional ensemble (REPS) Background error covariances EnVar deterministic (GDPS) Regional EnVar Regional deterministic (RDPS) system Regional system
215-216 implementation of the NWP suites EnKF ensemble (GEPS) Regional EnKF Regional ensemble (REPS) Background error covariances Background error covariances EnVar deterministic (GDPS) Regional EnVar Regional deterministic (RDPS) system Regional system
Autonomous Regional EnKF Driven by the {x b } EnKF (6Z,18Z) {x a } GEM (5km) {x b } Subsample (Z,12Z) EnKF {x a } GEM (5km) 16 day global {x b } obs Model error driving obs Model error driving {x b } EnKF {x a } GEM-LAM (15km) {x b } EnKF {x a } GEM-LAM (15km) {x b } Regional Subsample 72h regional Regional EnKF starts from global ensemble analyses and then runs autonomously. Environmental Multi-scale (GEM) model is used for the.
Comparison Between and Regional EnKF Make the regional configuration close to the global. 256 ensemble members. Same localization distance. Same model error perturbation. Same set of observations. Multi-model for the global but single model for the regional. Each member has different configuration for the global. All members have the same configuration for the regional. Regional EnKF performs better than the global EnKF.
Downscaling the EnKF Feed GEM-LAM 15km model with the global analyses. Compare with the regional EnKF. Analyses GEM (5km) driving GEM-LAM (15km) 256 member 6H Trials 21 member 72H Forecast
Verification Against Radiosonde:.vs. Regional Northern America Northern America West UU VV UU VV GZ TT GZ TT
Verification Against Radiosonde: Downscale.vs. Regional Northern America Northern America West UU VV UU VV GZ TT GZ TT
Correlation Structure Comparison 2.5km EnKF over Vancouver (MEOPAR). Driven by the regional EnKF. Compare correlations between global 5km, regional 15km and high resolution 2.5km EnKF. 3 points (a,b,c) on the ocean and 3 points (d,e,f) on the land at 7hPa height are selected. a c b d e f 3 2 1 1 altitude [m] Correlations along the zonal and vertical directions are computed.
U-Wind Correlation: 2.5km, 15km and 5km a c b d e f 3 2 1 altitude [m] 1 Ocean Land GEnKF REnKF HREnKF [hpa] [hpa] [hpa] 5 6 7 8 9 1 5 6 7 8 9 1 5 6 7 8 9 1 HRa) HRb) HRc) HRd ) HRe) HRf ) Ra) Rb) Rc) Rd ) Re) Rf ) Ga) Gb) Gc) Gd ) Ge) Gf ) -2-1 1 2-2 -1 1 2-2 -1 1 2-2 -1 1 2-2 -1 1 2-2 -1 1 2 1..71.41.12 -.12 -.41 -.71 unitless
U-Wind Correlation: 15km Regional EnKF and Downscale a c b d e f 3 2 1 altitude [m] 1 Ocean Land 1. REnKF Trials REnKF Downscaled [hpa] [hpa] 5 6 7 8 9 1 5 6 7 8 9 1 Ra) Rb) Rc) Rd ) Re) Rf ) Ra) Rb) Rc) Rd ) Re) Rf ) -2-1 1 2-2 -1 1 2-2 -1 1 2-2 -1 1 2-2 -1 1 2-2 -1 1 2.71.41.12 -.12 -.41 -.71 unitless
1m Wind Forecast Error (METAR): vs Regional Forecast from Z 2. m/s 1.. -1. 12 24 36 48 6 72 Difference.2. -.2 12 24 36 48 6 72 Run hour + lead time Forecast from 12Z 2. Standard deviation Bias m/s Difference 1.. -1. 12 24 36 48 6 72 84.2. -.2 12 24 36 48 6 72 84 Run hour + lead time
1m Wind Forecast Error (METAR): Downscale vs Regional Forecast from Z m/s 2.5 2 1.5 1.5 12 24 36 48 6 72 Difference.1 -.1 12 24 36 48 6 72 Run hour + lead time Standard deviation Bias Forecast from 12Z m/s Difference 2.5 2 1.5 1.5.1 -.1 12 24 36 48 6 72 84 12 24 36 48 6 72 84 Run hour + lead time
2m Temperature Forecast Error (METAR): vs Regional Forecast from Z degree 4.5 3 1.5-1.5 12 24 36 48 6 72 Difference.4.2 -.2 12 24 36 48 6 72 Run hour + lead time Standard deviation Bias Forecast from 12Z Difference degree 4.5 3 1.5-1.5 12 24 36 48 6 72 84.4.2 -.2 12 24 36 48 6 72 84 Run hour + lead time
2m Temperature Forecast Error (METAR): Downscale vs Regional Forecast from Z m/s 3 1.5-1.5 12 24 36 48 6 72 Difference.4.2 -.2 12 24 36 48 6 72 Run hour + lead time Forecast from 12Z Standard deviation Bias m/s Difference 3 1.5-1.5 12 24 36 48 6 72 84.4.2 -.2 12 24 36 48 6 72 84 Run hour + lead time
Downscaling is enough Regional EnKF is almost twice expensive than global EnKF. Limited improvement against downscaled s. 72 hour regional s for upper air and precipitation also neutral. Need high density observation network to fully exploit high resolution model. METAR, Mesonet, Radar, SWOB, etc.
METAR Data Assimilation with 2.5km EnKF Forecast for the Pan Am Games in Toronto up to 25m resolution. Coupled with CaLDAS (Canadian Land Data Assimilation System). CaLDAS produces ensemble surface fields (soil temperature/moisture). Assimilate 2m temperature and humidity with 1 hour cycle. 12 hour for every 6 hours. Precipitation pattern improved. 2m temperature and humidity improved up to 3 hours. 2m Temperature rms error degree 1 2 3 4 UTC 214/6/16 UTC 214/6/17 UTC 214/6/18
New Direction of the EnKF Development Put more efforts on improving the global EnKF. Coupled with CaLDAS. Reduce assimilation time window: 6H 3H 1H. Apply IAU (Incremental Analysis Update) instead of digital filter. Adaptive localization Higher resolution (1km) regional EnKF with reduced domain. Cover 2.5km national domain. Adopt the features of the global EnKF. Assimilate high resolution high frequency data: METAR, Radar, SWOB, etc. Provide ensemble trials for 2.5km EnVar analysis.
Regional 1km Domain over Canada 5km : 8x4 Regional 15km : 744x711 Regional 1km : 73x43 National 2.5km: 2584x1334