Mesoscale Data AssimilationMesoscale Data Assimilationfor NACPfor NACP

Scott DenningDusanka Zupanski

Marek Uliasz

Evolving in-situ COEvolving in-situ CO22 Network Network

Figure 2: Continuous calibrated CO2 mixing ratio sites to be used for dataassimilation. Sites for which we currently have hourly data are in red. Sites forwhich we anticipate data soon are in blue.

Figure 3: Daily minimum CO2 mixing ratio for 6 observing sites in 2004.

• 13 sites in 2005, maybe 20 by 2007 (NOAA?)

• Big synoptic variations and spatial gradients!

COCO22 Modeling in GEOS4 Modeling in GEOS4• Offline coupling: SiB driven

by NDVI and GEOSsurface weather

• Hourly SiB fluxesprescribed as LBC

• Offline tracer transportusing PCTM driven byGEOS winds, convection,and turbulence

• Reasonable variability atseasonal, diurnal, andsynoptic time scales (r2 =.66)

340350360370380390

150 160 170 180

obs

sim

340350360370380390

180 190 200 210

340350360370380390

210 220 230 240

June 2004

July

August

1° x 1.25 ° grid

Measured NEE of COMeasured NEE of CO22 (WLEF) (WLEF)

• Coherent diurnal cycles, but …• Day-to-day variability of ~ factor of 2 due to passing

weather disturbances• How to specify temporal autocorrelation in inversions?

COCO22 Asssimilation Asssimilation

• Fine-scale variations (hourly, 20-km pixels)from weather forcing, NDVI as processedby forward model logic

• Multiplicative biases (10-day, 100-kmpixels) derived by inversion of hourly [CO2]

FCO2 (x, y,t) = !R (x, y)R(x, y,t) − !A (x, y)A(x, y,t))

Ck ,m = !R,i, jRi, j ,nCk ,m,i, j ,n* + !A,i, jAi, j ,nCk ,m,i, j ,n

*( )i, j ,n∑ ∆t f∆x∆y + CIN ,k ,m

where k = tower  numberm = sampling timen = flux release timei, j = source grid cells

Data Assimilation Set-UpData Assimilation Set-UpDiscrete stochastic-dynamic model

Dusanka Zupanski, CIRA/CSUZupanski@CIRA.colostate.edu

Discrete stochastic observation model

111 )()( : −−− += kkkk wxGxMxM

w k-1 – model error (stochastic forcing)

M – non-linear dynamic (RAMS-SiB-BGC) model

G – model (matrix) reflecting the state dependence of model error

kkk xHy !+= )( :D

! k – measurement + representativeness error

H – non-linear observation operator (M M "" D D )

min]([]([21

][)(][21 11 =−−+−−= −−

obsT

obsbfT

b HHJ yxRyxxxxx ))P

(1) State estimate (optimal solution):

)()( 1bobs

TTba xyRPPxxx HHHH −++== −

(2) Estimate of the uncertainty of the solution:

TTaf GGQMMPP +=

Tji

jif MMMM )]()()][()([)( , xpxxpxP −+−+=ENSEMBLE KALMAN FILTER or EnsDA APPROACH

In EnsDA solution is defined in ensemble subspace (reduced rank problem) !

KALMAN FILTER APPROACH

MAXIMUM LIKELIHOOD ESTIMATE (VARIATIONAL APPROACH ):

MINIMUM VARIANCE ESTIMATE (KALMAN FILTER APPROACH ):

DATA ASSIMILATION EQUATIONS:DATA ASSIMILATION EQUATIONS:

Mesoscale Ensemble CDASMesoscale Ensemble CDAS(synthetic data experiment)(synthetic data experiment)

• SiB-RAMS weather, GPP,respiration, atmospheric[CO2]

• Sample modelatmosphere at 11 towerseach hour

• Estimate biases in GPPand resp every 10 days on100 km grid (5800 x3800 km domain)

• Smaller ensembles ~stable w/ covariancelocalization and5 ∆x smoothing

N=4408 N=100

$R N=4408 N=100$R

$GPP $GPP

Coupled Modeling and Assimilation SystemCoupled Modeling and Assimilation System

CSU RAMS

Radiation

Clouds

CO2 Transport and Mixing Ratio

Winds

Surface layerPrecipitationPBL

(T, q)

Eddy covariance H, LE, NEEObserved snow coverObserved soil moisture & T

Sat fire products

MODIS LAI/fPAR

MODIS land cover

Soil texture maps

LateralBoundaryConditions

Fossil fuel emissionsAir-sea gas flux

Biogeochemistry

Microbial poolsLitter pools

Slow soil C

RootsWoodLeaves

passive soil C

allocation autotrophic resp

heterotrophicresp

SiB3

Snow (0-5 layers)

Photosynthesis

Soil T & moisture (10 layers)

Canopy air spaceSfc TLeaf T

H LE NEE

CO2CO2

Ecosystem inventory data

Observed weatherObserved CO2 (timeseries, flasks, sat)Wind profilers, ceilometers

Input Data Evaluation Data

Ring of TowersRing of Towers

• inexpensiveinstrumentsdeployed onsix 75-mtowers in2004

• ~200 kmradius

• 1-minutedata May-August

Ring of Towers DataRing of Towers Datamid-day onlymid-day only June 9- July 5, 2004June 9- July 5, 2004

Mid-Day CO2: June 9 - July 5

350

355

360

365

370

375

380

385

1 3 5 7 9 11 13 15 17 19 21 23 25 27

brule

fence

redcliff

wittenberg

WLEF

LEF396

sylvania

5 ppm over 200 kmu ~ 10 m/s∆z ~ 1500 m~ 13 µmol m-2 s-1

Mid-Cont Intensive Plans?Mid-Cont Intensive Plans?

• Technical capability to do high-resolutionnested-grid simulations or assimmilationsfor the MCI– E.g., 10 km weather, NEE, CO2 over MCI domain

during growing season– cloud-resolving (1-km) for ~ a week

• We are not funded to do this, but willpropose it to DOE TCO

• Detailed comparison to towers, fluxaircraft, MINT airborne data

Footprints from RUC-13Footprints from RUC-13

% ∆x = 13 km,∆t = 1 hr

• Archiving u, v, w,cloud flux, TKEevery hour forwhole domain

• LPDM forNACP towers

• Infl functiongeneration viaweb interfaceavailable to all by2007

Process-Based Fossil Fuel EmissionsProcess-Based Fossil Fuel Emissions(K. Gurney, PI)(K. Gurney, PI)

• Combine inventory data and process attributes to constructdetailed space and time dependent emissions of fossil fuel CO2

and CO

• Database/model has three classes of inputs:– Point sources (e.g., power plants, smelters)– Mobile sources (vehicle emissions)– Area sources (e.g., residential sources)

• Resolution: 36 km, hourly downscale from roads, points, lights, pop’n?

• Optimize parameters in emissions model(s) using transport inRAMS and observed [CO]

Modules

Area source

Point source

Vehicle

Biogenic

nonroad