lecture 3 tropospheric chemistry data assimilation -...

Day 5 Lecture 3 Module name - Tropospheric Chemistry Data Assimilation 1

Lecture 3Tropospheric Chemistry

Data AssimilationH. Elbern

Rhenish Institute for Environmental Research at the University of Cologne

andInstitute for Chemistry and dynamics of the Geosphere-2: Troposphere (ICG-2), Research Centre Jülich, Germany


Objective of this lecture

Understand the• the bi- and multi-uncertainty aspect of

tropospheric chemistry data assimilation: the generalized inversion task

• consequences for system implementation

• Observe practical potential and limits of tropospheric chemistry data assimilation and air quality forecasts


Outline1. Objectives and special challenges of air quality

data assimilation

2. Implications for the methodology

3. Tropospheric chemistry 4Dvar examples

4. Summary


Motivation What is the public request?

As brief as possible:• to provide timely prediction of chemical

weather (summer: photo smog; winter: particulate matter)

• fine scale pollution monitoring for retrospective assessment of reduction measures and exposure times (most importantly particulate matter)2 realisation attempts on a European level:EC 6FP GEMS, ESA GSE PROMOTE


Objectives of air quality data assimilation

• bring together air quality measurements and CTMs to provide optimal spatio-temporal reconstructions of air quality parameters,

• estimate variability, sources, sinks, and trends• provide better air quality predictions• reconstruct past changes,• act as a decision support system for protection

measures (which emissions are most critical?)


Which constituents/complexity really matters? • Human health:

– PM10, PM2.5, PM1 (= Particulate Matter )– POPs (Persistent Organic Pollutants):

• PAHs (Polycyclic Aromatic Hydrocarbons), • PCBs (PolyClorinated Biphenyls), • HCHs (HexaCloroHexanes), • benzene, • benzopyrene, ….

– Trace metals: Cd, Be, Co, Hg, Mo, Ni, Se, Sn, V, As, Cr, Cu, Mn, Zn, Pb

– Ozone, PAN, NO, NO2, SO2, CO– Pollen

• Crops:– Ozone

• Forests, lakes, other vulnerable ecosystems– “Acid rain”– ozone

∫x

drrm0

)(


Special challenges of Special challenges of tropospherictroposphericchemistry data assimilationchemistry data assimilation

The following problems prevail:1. strong influence of manifold processes including

emissions and deposition2. temporarily highly variable “chemical regimes”3. chemical state observability (= “analyseability”)

hampered by manifold hydrocarbon species4. consistency with heterogeneous data sources:

satellite data and in situ observations


State of the Art ExamplesHow are today’s services designed?

1. solvers of a deterministic transport-diffusion-reaction equation, mostly operator-split-wise (except probably reaction-vert. diffusion)

2. continental, or regional nest (1 km < Δx < 250 km horizontal resolution)

3. tropospheric gas phase chemistry, 40-60 species, < 200 reactions

4. basic aerosol treatment of at least NH3, HNO3, H2SO4


““UpgradeUpgradedata to data to informationinformation”

GISdata

Simulation of the“atmospheric processor”

healthinformation

legislationculturalheritage

farmingsupport

aviationcontrol

science

Meteodata

Emissioninventories

in situobservations

Satellite retrievals

Chemical Weather SystemChemical Weather System


Special challenge: which is the appropriate scale for air quality modelling?

• intercontinental / long range transport of pollutants

• many processes are local: point and line source emissions

• surface patterns/tesselation affects simulation skills

• boundary layer and convection simulation at least of like importance as in meteorology


Prognosis

www.eurad.uni-koeln.de

Europeds = 125 km

Central Europeds = 25 km

NRWds = 5 km

23 levels lowest level: ca. 40 mupper boundary ca. 15 km

Sequential nestingJülich example

ECHOds = 1 km

10 20 30 40 50 60 70

10

20

30

40

50

60

5.250 12.800 22.400 35.700 53.200 76.600 111.000 190.000 412.000

casename: grid: 79 x 69

KFZ kg/Tag echo

© b

y E

UR

AD

Wed

Nov

6 2

2:48

:26

2002


Hemispheric forecast PM10

Incentives•In readiness for tropospheric satellite data•boundary value provision for continental scale models for long range transported constituents

high resolution version150 x 150 gridDx = 125 km


Daughter grids forecast 6.9.2005 todayEurope Δx=125 km and central Europe Δx=25 km


targeted regionsforecast 6.9.2005

5 km resolution


NOx1.5 x

1 x

0.6 x

VOCs

0.6 x 1 x 1.5 x

0.6 x 1 x 1.5 x

Note:notC-orthogonal,not max.sensitivitiesaligned

BERLIOZ NOx-VOC emissions variationensemble (20.7.1998,

15:00 UTC)


NOx1.5 x

1 x

0.6 x

VOCs

0.6 x 1 x 1.5 x

0.6 x 1 x 1.5 x

BERLIOZ NOx-VOC emissions variationensemble (21.7.1998,

15:00 UTC)

Note:notC-orthogonal,not max.sensitivitiesaligned


Empirical Kinetic Model Approach scheme

Isopleths of ozone production, due to NO2 and VOC

•Nitrogene oxides and numerous hydrocarbons act highly nonlinearly as precursors of ozone. •Chemical conditions are either controlled by NOx or VOC deficit, delineating the “chemical regime”.•Both 4D-var and Kalman filter should start with the proper chemical regime.

EKMA diagram(Empirical Kinetic Model Approach) VOC sensitive

NOx sensitive

NOx

VOCs

approximate conditions 20.7. and 21.7.

A prototypelon-linearityexample:


Aerosol Chemistry in

MADEModal Aerosol Dynamics

for EURAD/Europe(Ackerman et al., 1998,

Schell 2000)

dMik/dt=nuki

k+coagiik+coagij

k

+condik+emiik

Mik:=kth Moment of ith Mode

Bridge from optical to chemicalpropertiesassimilation of aerosolBy sattelite retrievals: e.g.MERIS MODIS AATSR+SCIAMACHY,…


Example: chemical complexity:The EURAD Secondary ORGanic Aerosol

Model (SORGAM)


Advanced spatio-temporal data assimilation:4D -var and Kalman filtering

• Models do not passively accept external 3D analyses (probably react by engendering spurious relaxations)

• Rather contribute with its dynamics/ chemical kinetics as constraint to estimate the most probable state or parameter values

Best Linear Unbiased Estimate (BLUE)while using data and models consistentlyallows for Hypothesis testing


Advanced Advanced spatiospatio--temporal methods used in temporal methods used in tropospherictropospheric chemistry data assimilationchemistry data assimilation

Spacio-temporal BLUEs applied in troposphericchemistry data assimilation:

• 4D var: – with EURAD (Elbern and Schmidt, 1999, 2001), – with POLAIR (Issartel and Baverel, 2003)

• Kalman Filter – with LOTOS model (van Loon et al, 2000), (RRSQR)– with EUROS model (Hanea et al. 2004) (En+RRSQRKF)


Question: Which parameter to be optimized?Hypothesis: initial state and emission rates are least known

emission biased model state

only emission rate opt.

only initial value opt.true state

observations

time

conc

entra

tion

joint opt.


In the troposphere, for emission rates, theproduct (paucity of knowledge*importance)

is high


Background Error Covariance Matrix B• must be provided as an operator (size is of order 1013)

• we would like to have an operator which caneasily be factorised by B=B1/2BT/2

• Weaver and Courtier (2001):–general diffusion equation serves for a valid operator

generating a positive definite covariance operator–diffusion equation is self adjoint–B1/2 and BT/2 by applying the diffusion operator half the

diffusion time


Background Error Covariance Matrix B (short design outline)

( )( )∑=

−−=K

nj

nji

niij xxxx

KB

1

1

K=# Ensembles; i,j neighboring cells

TL κ2=diffusion coefficients κ:

Correlation length L to neighboring gridcell:

, , ,

⇒

1. How to obtain the covariances?Ensemble/NMC approach:

2. How to process this information?Translate into Diffusion coefficients difusion paradigma


4D-var configuration

MesoscaleMesoscale EURAD EURAD 4D4D--var var datadata

assimilationassimilation systemsystem

meteorologicaldriverMM5

meteorologicaldriverMM5

EURADemission model

EEMemission 1. guess

EURADemission model

EEMemission 1. guess

direct CTMdirect CTM

emission ratesemission rates

Initi

alva

lues

Initi

alva

lues

min

imis

atio

nm

inim

isat

ion

adjointCTM

adjointCTM

observationsobservationsanalysisanalysis

grad

ient

fore-cast


backward integration

1. step 2. step 3. step 4. step 5. step 6. step 7. step 8. step 9. step

2 level forward and backward integration schemetime direction

disk

forward integration

backward integration

intermediate storageLevel 1


store

recover

1. step 2. step 3. step 4. step 5. step 6. step 7. step 8. step 9. step


horiz

. Adv

ektx

horiz

. Adv

ekty

vert.

Diff

usio

n

vert.

. Adv

ekt

Adj

ungi

erte

horiz

. Adv

ektx

Adj

ungi

erte

horiz

. Adv

ekty

Adj

ungi

erte

vert.

Adv

ekt

Adj

ungi

erte

vert.

Diff

usio

n

horiz

. Adv

ektx

horiz

. Adv

ekty

vert.

Diff

usio

n

vert.

. Adv

ekt

Adj

ungi

erte

horiz

. Adv

ektx

Adj

ungi

erte

horiz

. Adv

ekty

Adj

ungi

erte

vert.

Adv

ekt

Adj

ungi

erte

vert.

Diff

usio

n

Che

mie

Adj

ungi

erte

Che

mie

storage sequence: level 2, operator split(each time step)


horiz

. Adv

ektx

horiz

. Adv

ekty

Che

mie

vert.

Diff

usio

nve

rt.. A

dvek

t

horiz. Advektx

horiz. Advekty

vert. Diffusion

vert.. Advekt

horiz

. Adv

ektx

horiz

. Adv

ekty

vert.

Diff

usio

n

vert.

. Adv

ekt

Adj

ungi

erte

horiz

. Adv

ektx

Adj

ungi

erte

horiz

. Adv

ekty

Adj

ungi

erte

vert.

Adv

ekt

Adj

ungi

erte

vert.

Diff

usio

n

horiz

. Adv

ektx

horiz

. Adv

ekty

vert.

Diff

usio

n

vert.

. Adv

ekt

Adj

ungi

erte

horiz

. Adv

ektx

Adj

ungi

erte

horiz

. Adv

ekty

Adj

ungi

erte

vert.

Adv

ekt

Adj

ungi

erte

vert.

Diff

usio

n

Che

mie

Adj

ungi

erte

Che

mie

horiz

. Adv

ektx

horiz

. Adv

ekty

Che

mie

vert.

Diff

usio

nve

rt.. A

dvek

t

horiz. Advektx

horiz. Advekty

vert. Diffusion

vert.. Advekt



Treatment of the inverse problem for emission rate inference


Normalised diurnal cycle of anthropogenic surface emissions f(t)

emission(t)=f(t;location,species,day) * v(location,species)day in {working day, Saturday, Sunday} v optimization parameter


Optimisation of emission ratesamplitude optimisation for each emittedspecies and grid cell

AS1

AS1 Fraglich, ob diese Folie gezeigt werden muss. Wenn Fragen kommen, dann zeigen.Achim Strunk, 22/04/2004


Background emission rate covariance matrix D


Semi-rural measurement site Eggegebirge

assimilation interval forecast

7. August 8. August 1997

+ observationsno optimisation

initial value opt.

emis. rate opt.

joint emis + ini val opt.


Rhine-Main area box: (Frankfurt-Mainz)9.-10. August 1997

assimilation interval forecast+ observationsno optimisation

initial value opt.

emis. rate opt.



Error statisticsbias (top), root mean square (bottom)

assimilation window forecast

forecast

assimilation window

+ observationsno optimisation

initial value opt.

emis. rate opt.



Δx=54 km

WhichWhich isis thethe requestedrequested resolutionresolution??BERLIOZBERLIOZ grid designs and observational sites

(20. 21. 07.1998)

Δx=18 kmΔx=6 km

Δx=2 km

Con

trol a

nd d

iagn

ostic

sC

ontro

l and

dia

gnos

tics


Some BERLIOZ examples of NOx assimilation (20. 21. 07.1998)

NO

NO2

Time series for selected NOx stations on nest 2. + observations, -- - no assimilation,-____ N1 assimilation (18 km), -____N2 assimilation.(6 km), -grey shading: assimilatedobservations, othersforecasted.


surf

ace

heig

ht la

yer ~

32-~

70m

NO2, (xylene (bottom), CO (top) SO2 .

Emission source estimates by inverse modellingOptimised emission factors for Nest 3


Nest 2: (surface ozone)(20. 21. 07.1998)

withoutassimilation

with assimilation


CONTRACEConvective Transport of Trace Gases into

the upper Troposphere over

Europe: Budget and Impact of Chemistry

Coord.: H. Huntrieser, DLR

flight path Nov. 14, 2001


CONTRACE Nov. 14, 2001 north (= home) bound

O3

H2O2CO

NO

1. guess assimilation result observations flight height [km]


Satellite information:ESA UV-VIS satellite footprints Ruhr area

comparison

SCIAGOME1

GOME(1u).2

OMI

minimal areas:GOME 1 320 x 40 km2

(special mode) 80 x 40 “SCIAMACHY 60 x 30 “GOME 2 80 x 40 “OMI 24 x 13 “

Ruhr area domain 90 x 80 km2

(~12 000 000 inhabitants)


Error variances appliedfor period 1.-10.7.2006 over model domain

molecules/cm2

E(y) Ε(Δy)

OMI 1.4*1015 0.8*1015

SCIAMACHY 1.2*1015 0.9*1015

NO2 columns from KNMI data files: R (diagonal)

Forecast error covariances B schematic formula

Bii (spec,lev)= max{1 ppb,0.8*var(spec,lev), 0.5 max(spec,lev)}


Assimilation of GOME NO2 tropospheric columns, 5.8.1997

GOME NO2 columns: Courtesy of A. Richter, IFE, U. Bremen

# m

olec

NO

2/cm

2

NOAA NOAA chch 3 (near infrared)3 (near infrared)

# m

olec

NO

2/cm

2#

mol

ecN

O2/c

m2

post assimilationforecaststarted:5.8.97 06:00 UTC

forecast without assimilation


Mean averaging kernel over the model domain.

A principal problem:Average OMI averaging kernel profileover model domain for July 9th, 2006


assimilated values (y); :EURAD forecasted (Hxb); column analyses (Hxa).

Comparison of NO2 tropospheric columns in molecules/cm2 for July 6th, 2006, 09-12 UTC.


Data assimilation result in terms of troposphericcolumns for July 6th, 2006. NO2 model columns based on OMI and SCIAMACHY assimilation within interval,

09-12 UTC.

Difference field giving implied changes for tropospheric columns by assimilation (middle), and induced surface concentration changes by NO2 ppb (right)


Data assimilation result in terms of tropospheric columns for July 7th, 2006. NO2 model columns based on OMI and SCIAMACHY assimilation

within the assimilation interval, 09-12 UTC.


Qualitative assessment of emission correction factors for July 7th, 2006

Emission inventoryvalues

to be increased

to be reduced


Control run (OmC) (no data assimilation at all,) black bold line, forecasted values (OmF) green bold line, analyses (OmA) blue bold line. For comparison: Gaussian fit to OmF pdf by mean and standard deviation given by broken purple line.

SCIAMACHYOSCIAmXOMIT probability density functions

for July 6th (left), and July 8th, (right).


GOME ozone profile assimilation BERLIOZ (20.7.1998)Data: Neuronal Network retrieval (Müller et al., 2003)

+ NNORSY ozone retrieval

first guess

after assimilation


MOPITT tropospheric CO assimilation

850 hPa, 1 July. 2003

first guess at 11:00 UTC;.

retrieved 850 hPaconcentrationsin ppbV.

4D-Var analysis for thesame time;

Note: Colour codes are not fully consistent!


Assimilation of Aerosol observations• In situ:

EEA Airbase: Database of groundstations of EU member countries& states:

• 450 stations for PM10 (2003)• No PM2.5. (4 stations in UK only)

• Satellite measurements:SYNAER (SYNergetic AErosol Retrieval, DLR-DFD, [Holzer-Popp, 2001])*

• combines GOME&ATSR-2, SCIAMACHY&AATSR measurementsaboard ERS-2/ENVISAT

• ATSR-2/AATSR:dark field detection, BLAOT (Boundary Layer Aerosol OpticalThickness) and albedo are calculated

• GOME/SCIAMACHY: Provides PM0.5, PM2.5 and PM10 columns and its composition(6 intrinsic species)


EEA PM10 stations

SYNAER PM10

retrievals

Aerosol observations (14.7.2003, ~10:00 UTC)


3Dvar aerosol assimilation (13.7.2003)

in s

ituon

lyin

situ

& S

YNAE

R

in s

itu&

SYN

AER

back

grou

nd


3Dvar aerosol assimilation (14.7.2003)biomass burning case in Spain

in s

ituon

lyin

situ

& S

YNAE

R

in s

itu&

SYN

AER

back

grou

nd

in s

itu&

SYN

AER


Do aerosol data assimilation effectsaccumulate? (14. July 2003)

No previous assimilation

only 14. July 2003

assimilation on previous days 10 UTC

Accumulation of retrieval information over 14 days


Summary• chemical weather forecasts are a multiple scale

problem• chemical data assimilation rests on sparse and

heterogeneous observations, with variable error characteristics (incl. error of representativity)

• initial value optimisation is insufficient, as at least emissions are less known and more important

• the ability for inversion is therefore required to optimize emission rates

• much is to be done for optimising multivariate covariance matrices

lecture 3 tropospheric chemistry data assimilation -...

Documents