high resolution modelling at zamg

High resolution modelling at ZAMG

Clemens Wastl

Content

• General overview on ZAMG NWP models

• Operational AROME model

• Nowcastingsystem AROME-RUC

data assimilation, nudging

• Ensemblesystem C-LAEF

blending, stochastic physics

• Other research areas:

• physics + diagnostics

Overview about available NWP at ZAMG

+ 00 + 12h + 1d + 3d + 15d + 6m

ECMWF EPS 1)

ECMWF seasonal forecast 1)

ALADIN-LAEF 1) / C-LAEF 1) EF

(EPS)

AROME-AUSTRIA 2)

ECMWF HRES 2)

…

ECMWF monthly forecast 1)

ALARO5-AUSTRIA 2) EF (EPS)

AROME-RUC 2)

INCA / EnINCA1)

1) probabilistic forecast system 2) deterministic forecast system

…

International cooperations

AROME = HARMONIE

Overview ZAMG NWP-models

deterministic probabilistic

ALARO

(4.8km, +72h, 4x / day

ALADIN – LAEF

(10.9km, +72h, 2x / day )

AROME

(2.5km, +60h, 8x / day)

C-LAEF (2.5km, +48, 2x /

day)

AROME – RUC

(1.2km, +12h, 24x /

day)

... operational ...

... under development...

ZAMG NWP INDEX 2005 - 2016

ALADIN ALARO ALARO 5km AROME

2018

e ZNI = weighted combination of MAE and RMSE from t, rh, mslp, glo, ff, dd, rr

q90

mean

q10

2005

Model generations at ZAMG

10 km 1 km5 km

ALADIN AROMEALARO

hydrostatic / non hydrostatic

enhanced model physics

convection parametrized / explicit

Non hydrostatic

enhanced model physics

convection explicit

enhanced interaction with ground

hydrostatic

convection parametrized

DeterministicProbabilistic = LAEF

MENT tasks & duties

Operations + Research:

AROME-RUC (1.2km) [test mode]

AROME-Aut (2.5 km) [operational]

C-LAEF (2.5km) [test mode]

ALARO5-Aut (4.8 km) [op. / no upgrades]

AROME Nowcasting (1km) [test mode]

LAEF (11 km) [op. / no upgrades]

Data assimilation (OI/EKF+3DVAR)

Resources and main areas:

• model physics and diagnostics (0.5 persons)

• ensemble prediction (1 person)

• data assimilation (1 persons) + surface assimilation (1 person)

• downstream application / operational tasks / verification (1.5 persons)

• + Master students / PHD students

ment@zamg.ac.at:

Christoph Wittmann

Florian Meier

Phillip Scheffknecht

Florian Weidle

Clemens Wastl

Content

• General overview on ZAMG NWP models

• Operational AROME model

• Nowcastingsystem AROME-RUC

data assimilation, nudging

• Ensemblesystem C-LAEF

Ensemble-JK, stochastic physics

• Other Research areas:

• Physics + diagnostics

AROME

AROME (Applications of Research to Operations at Mesoscale):

+

Meso - NH physics NH Kernel ALADIN/ALARO)

(Laboratoire d'Aérologie, CNRM-GAME) (ALADIN Partners)

Convection resolving:

Horizontal resolution is high enough to resolve deep convection explicitly without

any convection parametrization – shallow convection is still parametrized.

Non-hydrostatic:

Hydrostatic approximation (vertical pressure gradient force = force of gravity) can

not be asssumed at this resolution - vertical momentum equation is fully solved.

AROME

Horizontal

resolution

2.5km (600x432)

Vertical resolution 90 Levels

Runs / day 8 (00,03,..18,21 UTC)

Forecast Range 60h

Output-frequency 1h

Model time step 60sec

Coupling model IFS (lagged)

Coupling update 1h

Assimilation 3DVAR / OI

SCREEN

Addsurf

AROME - Technical details of operational run

927boundary data

927SOIL

SST-Austausch

BATOR

BATOR3D

OPLACE

REMSENS

CANARI OIMAIN

„first guess“

CCMA MINIMIZATION

integrationpost processingderived fields

products

ZAMG

observations

GRIBCONV

About 3h 45min after the initialization the output is available

for the forecasters and customers

**) not yet operationally used

Observation type Parameter assimilated Source

SYNOP+TAWES T2m,RH2m,U10m,V10m,f ZAMG+OPLACE

AMDAR (Flugzeug) U,V,T (+Q) ZAMG+OPLACE

GEOWIND (SAT-Winde) MSG3 U,V OPLACE

TEMP (Radiosonde) U,V,T,Q,f ZAMG+OPLACE

PILOT U,V ZAMG

WINDPROFILER **) U,V ECMWF MARSARCHIV/OPLACE

MSG3-SEVIRI WV-radiances OPLACE

NOAA16/18/19+MetOp-A-B

AMSU-A,-B,MHS,HIRS

radiances OPLACE

MetOp-A-B IASI radiances OPLACE

ASCAT wind U10m,V10m (25km) ZAMG/EUMETSAT

RADAR **) reflectivity / radial winds Austrocontrol/Remote Sensing

MODIS-Schneebedeckung snow yes / no ENVEO-CRYOLAND

SNOWGRID Schneemodell snowheight ZAMG

GNSS Daten **) ZTD (STD, Refraktivität), RO EPOSA, TU WIEN

3DVAR Assimilation: Observational data

AROME

AROME model is running on the HPC of ZAMG

HPE Apollo 8600 (=SGI ICE-XA)

192 nodes with 18-core SKL

6140@2.4GHz

96 GB RAM per node

2 frontend nodes (à 2x8 processors, 64 GB

RAM, ...)

Total: 3472 cores

OmniPath enhanced hypercube network

Lustre Filesystem with total capacity of

350TB

PBSpro scheduling system

The new HPE/SGI system replaced the old

SGI ICE-X in December 2017

Content

• General overview on ZAMG NWP models

• Operational AROME model

• Nowcastingsystem AROME-RUC

data assimilation, nudging

• Ensemblesystem C-LAEF

Ensemble-JK, stochastic physics

• Other Research areas:

• Physics + diagnostics

Problems in the application of AROME

Data availability:

• AROME-Aut is running 8 times per day – every 3 hours a new forecast

• Data delay ov about 3h 45min (00 UTC run is available at 03:45 UTC)

• Not suitable for short range and nowcasting approaches

• Potential (very high) of available observation data is not fully tapped

source: www.esa.int

Radar data (5min / 1km) Mode-S (4sec / each

aircraft)

GNSS (15min / 40 stations)

AROME-RUC (Rapid Update Cycle)

AROME-RUC: Rapid Update Cycle

12.06.2019AROMEIdea: fill gap between classical nowcasting systems and short range NWP

Hourly forecasts up to 12h with hourly 3D-Var and 25 min cutoff time available within

1h

• 900x576x90 GP 1.2km LBC+ soil from AROME-Aut

• additional observations (radar reflectivity, Doppler winds, MODE-S aircraft, national

SYNOP, national GNSS ZTD)

• additional initialisation: latent heat nudging +35min (Stephan 2008), FDDA nudging

(Liu et al. 2006) +30min (optional), cloud analysis (Brewster et al. 2003), IAU

(Brousseau)

INCA-nowcasting

AROME-RUC

AROME-OPER 2.5kmL90

AROME-RUC

Horizontal

resolution

1.2km (900x576)

Vertical resolution 90 Levels

Runs / day 24 (00,01,..23,24

UTC)

Forecast Range 12h

Output-frequency 15min

Model time step 30sec

Coupling model AROME (lagged)

Coupling update 1h

Assimilation 3DVAR / OI

Assimilation window -90 to +30min

LH Nudging INCA analysis (+5 to

+35min)

FDDA Nudging 10m Wind, T2m/RH2m

(+10 to +30min)

Running at the ZAMG HPC

AROME-RUC

time0

Last AROME-OPERLBC from

First guess -2h RUC

-1.5h-2h 0.5h

TEMP

AMDAR

MODE-S

GNSS

Satellite

SYNOP

reflectivity

Doppler w.

3D-VAR

12h-forecast

INCA-

LHN

TAWES-Nudging

IAU

Nowcasting componentsclassic DA

AROME-RUC – Influence of observation data

RADAR

Air craft data

Surface stationsSatellite data

Wetter balloons

AROME-RUC: Assimilation of Mode-S data

AROME-RUC: Latent Heat Nudging

•Method of Jones & Macpherson 1997 for 2D-RADAR-Product (UM,

COSMO-Modell, WRF)

∆𝜃𝐿𝐻𝑁 = ∆𝜃𝑝ℎ𝑦𝑠𝑅𝑅𝑜𝑏𝑠−𝑅𝑅𝑚𝑜𝑑𝑒𝑙

𝑅𝑅𝑚𝑜𝑑𝑒𝑙(Jones & Macpherson)

Advantages of Nudging:

• Only 2D-precipitation from INCA (analysis + forecasts), no full 3D

radar data

• Much faster, very short computational time

• It is applied during the first time of model integration (AROME

RUC: 25 - 45 min)

• 4D-assimilation: observations of different times can be considdered

(difference to 3D-VAR)

Correction of latent heating by observed precipitation

AROME-RUC: Example for LH Nudging

AROME+3D-VAR-RADAR

AROME+3D-VAR-RADAR

+LHN Version1

AROME+3D-VAR-RADAR

+LHN Version2

INCA: Analyse

AROME-RUC: FDDA nudging

𝜕𝑢

𝜕𝑡=

𝜕𝑢

𝜕𝑡𝑝ℎ𝑦𝑠+ 𝐺

σ𝑖𝑤²𝑥𝑦𝑖(𝑢𝑖𝑜𝑏𝑠 − 𝑢𝑚𝑜𝑑𝑒𝑙)

σ𝑖𝑤𝑥𝑦𝑖

𝜕𝑢

𝜕𝑡=

𝜕𝑢

𝜕𝑡𝑝ℎ𝑦𝑠+ 𝐺

σ𝑖𝑤²𝑥𝑦𝑖𝑢𝑖𝑜𝑏𝑠

σ𝑖𝑤𝑥𝑦𝑖− 𝐺

σ𝑖𝑤²𝑥𝑦𝑖 𝑢𝑚𝑜𝑑𝑒𝑙)

σ𝑖𝑤𝑥𝑦𝑖

𝐷𝐼𝑆𝑇𝐴𝑁𝐶𝐸′ = 𝐷𝐼𝑆𝑇𝐴𝑁𝐶𝐸 + 𝑅|𝑝𝑠𝑂𝐵𝑆 − 𝑝𝑠𝐺𝑃|

𝑑𝑧𝑡ℎ𝑟𝑒𝑠 = 75ℎ𝑃𝑎

𝑤𝑥𝑦 =𝑅20.752 − 𝑍𝐷𝐼𝑆𝑇𝐴𝑁𝐶𝐸′2

𝑅20.752 + 𝑍𝐷𝐼𝑆𝑇𝐴𝑁𝐶𝐸′2 (𝑝𝑠𝐺𝑃

500ℎ𝑃𝑎+ 1)

Observations at: +10 / 20 / 30min

R=20km

G=0.02

(namelist switches)

10m wind observations; Liu et al.

2006

Content

• General overview on ZAMG NWP models

• Operational AROME model

• Nowcastingsystem AROME-RUC

data assimilation, nudging

• Ensemblesystem C-LAEF

Ensemble-JK, stochastic physics

• Other Research areas:

• Physics + diagnostics

Problems in the application of AROME

High variability between consecutive runs (depending on weather/season)

Example: RMSE of wind speed forecast

situation of weak pressure gradient westerly/northwesterly flow

Problem in this case: different observation data

Problems in the application of AROME

Some reasons for strong variability between consecutive runs:

• Different number of observation data and impact factor

• 8 AROME runs per day, 2 consecutive are based on the same boundary

conditions from ECMWF

• High spatial/temporal resolution – high sensitivy of model to changes in

the intial state

Some reasons for bad/wrong forecasts:

i. Initialisation state (analysis) in the model

ii. Formulation and design of the forecasting model (physics, dynamics)

iii. Spatial/temporal resolution of model not sufficient (topography,

physiography)

iv. Problems coming from coupling with the boundary conditions of global

modelConsideration of such uncertainties in the forecast C-LAEF

(AROME EPS)

EPS – Ensemble Prediction System

• Ensemble prediction systems are designed to assess the uncertainties in

NWP

• Ensemble system is based on several forecasts with different

settings/perturbations

• It‘s getting more and more popular at the national weather services

• Global EPS (only a few: GFS, ECMWF, GSM, ICON, …)

• With increased computer power – convection permitting LAM EPS systems

Ensemble

data assimilation,

Singular vectors,

Ensemble Kalman Filter

Multimodel

Multi-physics

Stochastic physics

Blending of LBC

C-LAEF: Convection permitting – Limited Area

Ensemble Forecasting

• Ensemble-

data assimilation (EDA)

• Ensemble-

data assimilation of

surface variables (sEDA)

• Ensemble-JK

Initial conditions +

error

• Stochastic physics:

Combination of tendency

and parameter

perturbation scheme

• Coupling with

ECMWF-ENS

• Ensemble-JK

Lateral boundary +

conditions errorModel error

Uncertainties representation

in C-LAEF

C-LAEF

Ensemble size 16+1

Horizontal

resolution

2.5 km

Vertical resolution 90 Levels

Runs/Day 4 (00, 06, 12, 18

UTC)

Forecast range +48h (00, 12 UTC)

+6h (06, 18 UTC)

Output-Frequency 1h

Model time step 60s

Coupling-Model ECMWF-EPS

(lagged)

Coupling-Update 3h

Assimilation 3DVAR / OI

Perturbation Surface, 3DVAR,

LBC, Model

Running at the ECMWF HPC

in Reading/Bologna

Convection permitting Limited Area Ensemble Forecasting

C-LAEF: IC and LBC perturbations

JK blending method developed by Guidard and Fischer (2008)

Integration of uncertainty from global EPS directly to C-LAEF 3D-Var

Combination of large scale (global EPS) with small scale (C-LAEF)

perturbations

Consistency between IC und LBC perturbations in C-LAEFCost function (3DVar)

Cost function in Jk blending method:

Keresturi et al., 2019: Improving initial condition perturbations in a convection-permitting

ensemble prediction system , Q J R Meteorol Soc. Accepted Author Manuscript. doi:10.1002/qj.3473

C-LAEF: Physics schemes

Radiation scheme

Shallow convection scheme

Turbulence scheme

Microphysics scheme

δ𝑇1δ𝑡

δ𝑇2δ𝑡

,δ𝑄2

δ𝑡,δ𝑈2, 𝑉2

δ𝑡

δ𝑇3δ𝑡

,δ𝑄3

δ𝑡,δ𝑈3, 𝑉3

δ𝑡

δ𝑇4δ𝑡

,δ𝑄4

δ𝑡

Way how tendencies of different physics

schemes are processed in AROME / C-

LAEF

𝑑𝑇

𝑑𝑡=

𝑖=1

4δ𝑇𝑖δ𝑡

,𝑑𝑄

𝑑𝑡=

𝑖=1

4δ𝑄𝑖

δ𝑡, 𝑒𝑡𝑐.

C-LAEF: Stochastic perturbation of total model

tendencies

SPPT (ECMWF)Standard SPPT: Perturbation of total model tendencies (Buizza et al., 1999; Palmer et al., 2009)

𝑑𝑇

𝑑𝑡=

𝑖=1

4δ𝑇𝑖δ𝑡

𝑑𝑇′

𝑑𝑡=𝑑𝑇

𝑑𝑡∗ (1 + P)

𝑑𝑄

𝑑𝑡=

𝑖=1

4δ𝑄𝑖

δ𝑡

P …. stochastic pattern

𝑑𝑄′

𝑑𝑡=𝑑𝑄

𝑑𝑡∗ 1 + P … . .

Constraints:

• Energy conservation

• Tapering function ß

• Physical consistency

• Assumes same level of uncertainty

for all parametrisations

P = P ∗ β

Radiation scheme

δ𝑇1δ𝑡

∗ (1 + 𝑃1)

Shallow convection scheme

Turbulence scheme

Microphysics scheme

C-LAEF: Stochastic perturbation of partial model tendencies:

pSPPT (ZAMG)

δ𝑇2δ𝑡

∗ 1 + 𝑃2 ,δ𝑄2

δ𝑡∗ 1 + P2 , etc.

δ𝑇3δ𝑡

∗ 1 + 𝑃3 ,δ𝑄3

δ𝑡∗ 1 + P3 , etc.

δ𝑇4δ𝑡

∗ 1 + 𝑃4 ,δ𝑄4

δ𝑡∗ 1 + P4 , etc.

In pSPPT the partial tendencies of T, Q, U,

V are perturbed directly after each

parametrization

Influence on subsequent schemes

Different perturbations are applied to the

physics schemes

In C-LAEF we need 4 different perturbation

patterns with different temporal and

horizontal scales

Wastl et al., 2019: Independent perturbations for

physics parametrization tendencies in a convection-

permitting ensemble (pSPPT), Geosci. Model Dev.,

12, 261-273.

C-LAEF: Stochastic perturbation of partial model

tendencies:

pSPPT (ZAMG) Shallow

convection

Turbulenc

e Microphysics

Radiatio

n

• Energy conservation

• Tapering function ß

(turbulence)

• Assumes same level of uncertainty

for all parametrisations

• Physical consistency

Stochastic perturbation of model tendencies showed promising results,

especially pSPPT, still some restrictions

Stochastic perturbation of key parameters (SPP, Ollinaho et

al., 2017) at process level in the turbulence scheme (see table)

Hybrid system (HSPP): Combination of pSPPT with parameter

perturbation in turbulence

C-LAEF: Stochastic physics in C-LAEF: Hybrid system (HSPP)

Parameter Range Description

XLINI 0 – 0.1 Minimum BL89 mixing length

XCTD 0.98 – 1.2 Constant for dissipation of potential

temperature and mixing ratio

XCTP 2.325 – 4.65 Constant for temperature-vapor pressure

correlation

XCEP 1.055 – 4.0 Constant for wind-pressure correlation

XCED 0.7 – 0.85 Constant for dissipation of total kinetic energy

(TKE)

XALPSBL 3.75 – 4.65 Value related to the TKE universal function

within the surface boundary layer

Parameters in the

turbulence scheme

which are

stochastically

perturbed.

αi′ = exp(P) ∗ α𝑖

• Energy conservation

• Tapering function

• Combination with surface

EDAWastl et al., 2019b: A hybrid stochastically perturbed

parametrization scheme in a convection-permitting

ensemble, Mon. Wea. Rev. 147, 2217-2230

C-LAEF: Results of summer test period

Ensemble spread (solid) and RMSE (dashed) for surface parameters in

July 2016.

RMSE is given as difference to the reference run.

C-LAEF: Results of summer test period

CRPS for temperature and wind speed at 500 hPa and 850 hPa in July 2016.

C-LAEF: Precipitation verification

Ensemble spread (solid) and RMSE (dashed) for precipitation in July 2016

(left) and January 2017 (right). RMSE is given as difference to the

reference run.

MEAN

Prob > 10

Prob > 20

Prob > 30

C-LAEF results, thunderstorm, 2018 04 16

C-LAEF results, 24h precipitation, 2018 10 27

C-LAEF suite on ecflow

Content

• General overview on ZAMG NWP models

• Operational AROME model

• Nowcastingsystem AROME-RUC

data assimilation, nudging

• Ensemblesystem C-LAEF

Ensemble-JK, stochastic physics

• Other Research areas:

• Physics + diagnostics

Actual/recent work areas

Physics / Diagnostics:

• Orographic effects in

radiation scheme

• Extended diagnostics

(clouds, visibility, ceiling /

LVP, …)

• convective diagnostics

(lightning

parameterization)

• (Micro)physics

(testing/tuning)

above: Low stratus over

Austria (SAT + AROME)

left: Simulated lightning

density + observed lightnings

Evaluation of AROME lightning condition

Questions to answer:

In particular for situation when precipitation is forecasted AND observed …

• …is there a benefit of using AROME lightning forecasts for automatic

products with respect to current methods (estimated via showalter

and/or CAPE)?

• Is … there a benefit of using AROME lightning forecasts with respect to

MOS?

FSS lightning AROME FSS CAPE(AROME) FSS show(AROME)

… answer is YES in both cases

(probabilistic) visibility for AROME

New parametrization:

reference parameterizations:

Orographic Raditation: Motivation

Temperature verification showed a

strong positive BIAS at some

stations in small Alpine valleys

This BIAS occurred predominantly

in the morning after sunrise

Investigation showed that radiation

shadowing is responsible for this

behaviour

Sunshine in the model; orographic

shadowing in reality

In the atmospheric radiation

parameterization of numerical

models each gridsquare is

assumed flat and effectively

homogeneous

BIAS, MAE and RMSE for AROME (blue)

and ALARO (red) at the valley stations of

Mallnitz and Obervellach for summer 2013.

Orographic Raditation: Method

Method is based on a paper of Müller and Scherer, 2005: „A Grid- and Subgrid-Scale

Radation Parametization of Topographic Effects for Mesoscale Weather Forecast Models.“

(Monthly Weather Review, 133)

It has already been implemented into HIRLAM (Senkova et al., 2007: „Parameterization

of orographic effects on surface radiation in HIRLAM“, Tellus 59A)

Slope angle and direction, relief shadow influence the short wave radiation budget

sky view factor influnces both long and short wave radiation budget

For parameterization of the orographic effects on radiation, slopes and local

horizon in different directions are required

SWLW

Clear sky conditions, LSL = .T.

warmer slopes (directed to sun), colder slopes in the shadow