Development of Operational Data

Assimilation System for Convective

Scale Model at KMA

SeiYoung Park, Eunhee Lee, Yoonjeong Hwang, Mee-Ja Kim,

Eun-Hee Kim, Hyeyoung Kim, Hee-Jung Kang, Dayoung Choi,

Min-Jong Song, Ho-Yong Lee, Minyou Kim, and Yong Hee Lee

7th WMO DA Symposium

(’17.9.10)

Numerical Modeling Center

Korea Meteorological Administration

Global Medium-range Prediction (GDAPS / Global EPS)

• Deterministic: UM 17km L70 / T+288hrs (00/12UTC), T+87hrs (06/18UTC) / Hybrid-4DVAR

• Ensemble: UM 32km L70 / T+288hrs (00/12UTC) / 49 Members / Perturb. : ETKF, RP, SKEB2

Short-range Prediction (E-Asia) (RDAPS)

• UM 12km L70 / T+87hrs (6 hourly) / 4DVAR /

Deterministic

(Very) Short-range Prediction

• Deterministic : UM 1.5km L70 /

(LDAPS)T+36hrs (6 hourly) / 3DVAR (3 hourly)

(VDAPS)T+12hrs (1 hourly) / 3DVAR (1 hourly)

• Ensemble : UM 3km L70 / T+72hrs (LENS)

Seasonal Prediction System (Glosea5)

• GloSea5 / 60km L85 / 60, 220days

• Atmosphere(UM)+Ocean(NEMO)

KMA Operational NWP system

[’17.6.]

LDAPS Local Data Assimilation & Prediction System

LDAPS

❖ Launch : May 2012 ▪ Forecast length (cycle): 36 hours (8 times/day)

▪ DA system: 3DVAR (FGAT, IAU)

❖ Recent Upgrade : June 2016

▪ UM : v8.5 v10.1

▪ Dynamic core : New Dynamics ENDGame* ENDGame : Even Newer Dynamics for General Atmospheric Modelling of the Environment

▪ Extended domain : 744 x 928 1,188 x 1,148 (variable grid)

(622x810)

LDAPS.2012

LDAPS.2016

- to keep the consistency of the

synoptic scale with the global

model

Impact of Extended domain

❖ HGT 500hPa 100km wave filter (2015.07.12.00UTC)

LDAPS.2016LDAPS.2016

Decreasing of the error from boundary condition

- to keep the consistency of the synoptic scale with the

global model

GDAPS

Observation data

• Sonde(temp, pilot, windprofiler), Surface(synop, ship, buoy, metar), Aircraft(amdar), Radar(radial velocity), ScatWind(ASCAT)( Global: (+) ATOVS, AIRS, IASI, COMSCSR, GPSRO, CrIS, ATMS, Satwind, (-)Radar )

• 3 hour-cycling DA: lack of available observation, needs of satellite DA

LDAPS.2012 LDAPS.2016 LDAPS.2012 LDAPS.2016

Scatwind : 155 580 (x 3.8)

Surface : 2696 5160 (x 1.9) Sonde : 50 101 (x 2)

Aircarft : 238 2407 (x 10)

Initialization (3DVar, FGAT, IAU, LHN)

❖ Incremental 3DVar (3 km resolution)

• 3DVar (with FGAT) + IAU for all observations,

except Latent Heat Nudging for radar-derived surface rain rate

T0-90m T0+90m

Observation Processing System

Moisture OPS

Initialization (SURFace data assimilation)

ᅳ Rainfall amount • in-situ obs• Model (Old)• Model (New)

Factor Method GDAPS (global) LDAPS

Soil

Moisture

(Old) Nudging scheme

(New) Extended Kalman Filter (‘16.6~) MetOp-A, B / ASCAT

(4 times/day)

Downscale from GDAPS(1 time/day)

SnowIMS fractional snow cover

snow amountIMS

(4 km, 1 time/day)Background

SST

Sea Ice

Convert the resolution

of obs. to modelOSTIA

(5 km, 1 time/day)

OSTIA(Not used sea ice) (1 time/day)

GDAPS LDAPS

Time series of Soil moisture (%) & Rainfall amount in July 2015 at Cheongju Agriculture Observatory

Simulation of Typhoon in Old and New system

LDAPS.2016LDAPS.2012

Best Track

PPI0 composite (KMA+JMA)

COMS IR

Verification of 500 hPa Height

0h 6h 12h 18h 24h 30h 36h

ldps 9,61 10,42 10,22 10,78 11,62 12,92 13,2

xldps 8,19 8,55 8,8 9,1 9,32 9,76 10,33

6,00

8,00

10,00

12,00

14,00

RM

SE

HGT 500hPa (Observation)

6 12 18 24 30 36

ldps 3,19 4,1 5,52 6,18 7,44 8,18

xldps 2,60 3,35 4,52 5,17 6,35 7,29

0,00

3,00

6,00

9,00

12,00

RM

SE

HGT 500hPa (Analysis)

July 2016

LDAPS.2012LDAPS.2016

LDAPS.2012LDAPS.2016

VDAPS Very-short-range Data Assimilation & Prediction System

❖ Purpose

- supporting the forecaster’s very-short range

forecast (hourly forecast)

- supporting “Olympic and Paralympic winter

games Pyeongchang 2018”

❖ Model

- UM vn10.1 (ENDGame)

❖ Area, resolution

- grid number : 804 (E-W) X 1000 (S-N)

- resolution : 1.5 ~ 4 km (Variable grid),

DA 3 km, 70 levels

❖ Forecast length (cycle)

- 12 hours (hourly)

❖ DA system: 3DVAR (FGAT, IAU)

- surface, sonde, windprofiler, aircraft

- radar (radial velocity, LHN), MAPLE*

- visibility assimilation

VDAPS (June 2017)

* MAPLE: McGill Algorithm for Prediction nowcasting by Lagrangian Extrapolation

http://190.1.20.52/personal/aroma37/vdps_domain_new.png

MAPLE 1 hr forecast rain

▪ Early cycle system to support the forecasters

- observation time window : -30 ~ +10 minutes( ∴ Only limited observation data can be used! )

⇒ Output should be made within 20 minutes every hour on the hour for forecasters!

Cycle system of VDAPS

Radar assimilation

LHN with default parameters LHN with optimal parametersNO LHN

Radar reflectivity

A

B

▪Suppress gravity wave generations▪ Improvements in rainfall (A, B)

▪Strong gravity wave generation

AWS

▪ Spatial average of rain rate : 5(km)x5(km)

▪ Nudging coefficient = 0.5 0.1

▪ α = 0.5 0.3

▪ ε = 0.5 0.3

Visibility DA

visibility

aerosol

saturation vapor

mixing ratio

vapor mixng ratio

temperature

pressure

3DVAR

❖visibility operator

Clark et al. (2008)

No DA of visibility DA of visibility

Observation (#238 stations)

Visibility DA

▪ Making the raw observation data file within 10 minutes on the hour

▪ Impact of the Sonde data- simulated the strong rainfall cell

No Sonde With Sonde

Rawin Sonde DA

Forecast impact for rainfall event

2017. 6. 13. 06 UTC + F03

AWS (1hr accum.)

VDAPSLDAPS

2017. 6. 13. 09 UTC

http://190.1.20.52/personal/minyou/VDAPSCases/20170613/aws_rain_2017061309_acc01h_vdps.png

Ongoing works for LDAPS

❖More than 100 stations over Korean peninsular

DA of Ground-based GNSS

going to be used in next LDAPS version

❖ Impact of G-GNSS ZTD (zenith total delay) DA• Period: July 2016 / Data: 40 stations in Korea

❖ Quality Control

▪ Normality test (Anderson-Darling test : p-value > 0.05)

▪ Bias correction (static bias correction: 1 month mean of O-B)

▪ Height difference between station and model surface (if diff >300 m, reject)

Typhoon Track in 2016

Typhoon (Bogussing)

LDAPS

❖ TC Bogus method isn’t applied in LDAPS.

Try to test bogussing module with the enlarged domain.

Typhoon (Bogussing)NEPARTAK (1601)

▪ TC bogussing for high resolution model stronger than w/o bogus

▪ Typhoon simulation of high resolution model Anal: similar with OBS

Fcst: usually simulated the TC stronger than OBS

Min. Sea Level Pressure Max. Wind Speed

Land Surface DA

❖ SMC update time tests- Whenever the SMC was updated from global model every 06 UTC, there were peaks.

Tests for (EXP1) 6 hourly updates & (EXP2) using background

(EXP1) smaller picks & (EXP2) drifting

Forecast performances were similar.

(EXP1) (EXP2)

Land Surface DA

Soil

Moisture

Contents(1st level)

Difference : EKF - Downscaling

WetDry

Regional EKFDownscaling (modified) from global model

06UTC 1 July 2016

❖ Downscaling : Rain band is shifted to western region.

❖ Regional EKF : no rain area is larger than downscaling exp.

1 hour

rainfall (6hr fcst)

❖ Implementation of regional EKF- Observation: screen T/Q (no satellite data)

- Preliminary results (6 days cycles)

Summary

❖ Convective scale models in KMA

▪ LDAPS (May 2012~) : Short range forecasting (36 hr fcst), 1.5 km, 3DVar

- New version (2016) : ENDGame & domain extension better results

▪ VDAPS (June 2017~) : Very-short range forecasting (12 hr fcst), hourly cycle

- short obs time window, try to add more data like visibility & radar

❖ Ongoing works for LDAPS

▪ DA of Ground-GNSS ZTD with 40 stations in Korea

▪ TC bogussing & Land Surface DA : tested for extended domain LDAPS

▪ Background Error Covariance of UKV (not shown)

Plan

❖ Convective Scale Model (’17.10)

▪ Extending the domain

▪ DA: G-GNSS, AMSU-B, Background Error Covariance, …

❖ Data Assimilation (’17~’18)

▪ 3DVar (hybrid) 4DVar (collaboration with MO)

▪ Update of Background Error Covariance

▪ Observation data

- Conventional DA : drifting Rawin Sonde data, Windprofiler QC, Aircraft QC

- Satellite DA : AMSU-B, IASI, GNSS-RO, [VDAPS] G-GNSS, AMV, …

- Radar DA : [VDAPS] radial velocity QC, reflectivity

❖ Development of Limited Area Model for KMA new Global system, KIM

(KIAPS model) (’18~)