Developments in regional inverse modellingR. L. Thompson and A. StohlNILU, Norsk Institutt for Luftforskning, Norway
Outline• Introduction to FLEXINVERT• Case study: methane fluxes in the high latitudes
– data selection criteria– transport uncertainties– optimized CH4 fluxes
• Developments for FLEXINVERT-CO2
– planned inversion framework
• Summary & Conclusions
20 June 2016 ICOS Workshop Lund 2
Overview of FLEXINVERT
20 June 2016 ICOS Workshop Lund 3
Lagrangian model,FLEXPARTinputdatafrome.g.ECMWF
fluxsensitivities
H
initialcond.sensitivities
Hini
priorfluxesx0
initialmixingratio fields
yini
observedmixing ratiofromaircraft,
ships,groundsites
y
Optimizefluxesargminx [(x – x0)TB-1(x – x0)+(Hx – y)TR-1(Hx – y)]
optimizedmixing ratio
Definition of forward modelMixing ratios are modeled according to:
20 June 2016 ICOS Workshop Lund 4
hi, t , ′tout xi, ′t
out
hi, t , ′′tini yi, ′′t
ini hi, t , ′tnest xi, ′t
nest
yi, t
nested domain
global domain
ymod = Hnestxnest +Houtxout +Hiniyini
Thompson and Stohl, GMD, 2014
Background mixing ratiosTo determine the background contribution (Hiniyini):Couple to global 3D mixing ratio fields in time domain
20 June 2016 ICOS Workshop Lund 5
hi,nini = ∂y
∂yi,n=ni, j ,nJ j
ni,j,n no. particles in grid-cellJj total no. particles in trajectoryyi,n mixing ratio from global model
Thompson and Stohl, GMD, 2014
Background mixing ratiosMonthly 2D fields from bivariate interpolation of NOAA flask data, plus model estimate of stratospheric mixing ratio (yini)
20 June 2016 ICOS Workshop Lund 6
180°W 120°W 60°W 0° 60°E 120°E 180°E90°S
60°S
30°S
0°
30°N
60°N
90°N
1700
1750
1800
1850
1900
1950
2000
180°W 120°W 60°W 0° 60°E 120°E 180°E90°S
60°S
30°S
0°
30°N
60°N
90°N
0
20
40
60
80
100
a)
b)
yini
Hini
background mixing ratio:ybg = Hiniyini + Houtxout
1800
1900
2000
2100
2200
2300
CH
4 (pp
b)
IGR
01 02 03 04 05 06 07 08 09 10 11 12
OBSPRIORBKGND
Thompson and Stohl, GMD, 2014
FLEXINVERT spatial grid
10
12
14
16
20 June 2016 ICOS Workshop Lund 7
50°N58°N
66°N72°N
80°N
Optimize grid based on flux sensitivities and, optionally, prior fluxes by aggregating grid cells until meet threshold
e.g. flux sensitivities, units log(s m3 kg-1) e.g. optimized inversion grid
Methane in the high northern latitudes
8
Observations
ZEP
TIK
TERZOT
PAL
CHL
BAL
CBA
LLBETL
MHD
FSD
ESP
CDL
KRSIGR
NOYDEM
AZV
VGN
YAKCHM
flaskin situ
20 June 2016 ICOS Workshop Lund 9
Networks:• JR-STATION (7 sites)• EC (7 sites)• NOAA (3 sites)
Stations:• Pallas, FMI• Zeppelin, NILU• Mace Head, AGAGE• Teriberka, MGO• Zotto, MPI-BGC
Total of 17 in-situ & 5 flask-sampling sites used to constrain fluxes
Data selection criteriaProblem modeling PBL in winter at continental sites:filter data using observed temp. gradient and wind speed criteria
20 June 2016 ICOS Workshop Lund 10
2009.00 2009.02 2009.04 2009.06 2009.08
2000
2200
2400
2600
CH4 (
ppb)
OBSECMWFFNLPBL_TEST
2009.00 2009.02 2009.04 2009.06 2009.08−35−30−25−20−15−10−5
Tem
pera
ture
(°C) upper
lower
2009.00 2009.02 2009.04 2009.06 2009.080
5
10
15
20
Win
dspe
ed (m
/s)
2009.50 2009.52 2009.54 2009.56 2009.58
2000
2200
2400
2600
CH
4 (pp
b)
OBSECMWFFNLPBL_TEST
2009.50 2009.52 2009.54 2009.56 2009.5805
1015202530
Tem
pera
ture
(°C
) upperlower
2009.50 2009.52 2009.54 2009.56 2009.580
5
10
15
20
Win
dspe
ed (m
/s)
Transport uncertainties
20 June 2016 ICOS Workshop Lund 11
Estimate transport uncertainties using proxy of difference between simulations with ECMWF EI versus NCEP FNL
0
10
20
30
40
unce
rtain
ty (p
pb)
winter
0
10
20
30
40
unce
rtain
ty (p
pb)
summer
Errors calculated 3-hourly for 1 year – use daily mean errors each month for all inversion years
Prior flux estimatesSource category Dataset Total (Tg y-1)
Natural Wetlands LPX-Bern 202
Termites Sanderson et al. 1996 19
Wild animals Houweling et al. 1999 5
Ocean Lambert et al. 1993 17
Soil uptake LPX-Bern -49
Biomass Burning GFED-3.1 13
Anthropogenic Fuel and Industry EDGAR-4.2FT2010 150
Enteric fermentation EDGAR-4.2FT2010 101
Waste EDGAR-4.2FT2010 61
Rice cultivation LPX-Bern 36
Global total 556
Modeled mixing ratios
20 June 2016 ICOS Workshop Lund 13
ZEP
TIK
TERZOT
PAL
CHL
BAL
CBA
LLBETL
MHD
FSD
ESP
CDL
KRSIGR
NOYDEM
AZV
VGN
YAKCHM
flaskin situ
1800
1900
2000
2100
2200
2300
CH
4 (pp
b)
FSD
1800
1900
2000
2100
2200
2300
CH
4 (pp
b)
LLB
1800
1900
2000
2100
2200
2300
CH
4 (pp
b)
ETL
01 02 03 04 05 06 07 08 09 10 11 12
1800
1900
2000
2100
2200
2300
CH
4 (pp
b)
IGR
1800
1900
2000
2100
2200
2300
CH
4 (pp
b)
KRS
1800
1900
2000
2100
2200
2300
CH
4 (pp
b)
PAL
01 02 03 04 05 06 07 08 09 10 11 12
OBSPRIORPOSTBKGND
Comparison of observed, prior and posterior mixing ratios at selected sites for 2009
Optimized fluxes
0.00
0.05
0.10
0.15
0.20
DJF MAM JJA SON
−0.10
−0.05
0.00
0.05
0.10
gCH
4 m-2 d
ay-1
gCH
4 m-2 d
ay-1
20 June 2016 ICOS Workshop Lund 14
Seasonal mean: posterior and difference (posterior – prior)
Thompson et al., in prep., 2016
Seasonal flux variability
2 4 6 8 10 120
20406080
100120
Tg C
H 4 y−
1
North Eurasia
2 4 6 8 10 120
1020304050
Tg C
H 4 y−
1
WSL
2 4 6 8 10 120
5
10
15
20
Tg C
H 4 y−
1
HBL
2 4 6 8 10 120
5
10
15
20Tg
CH 4
y−1
Alberta
2 4 6 8 10 120
1020304050
Tg C
H 4 y−
1
North America
20 June 2016 ICOS Workshop Lund 15
Thompson et al., in prep., 2016
case 1: prior wetlands LPX-Berncase 2: prior wetlands LPJ-DGVMsolid lines: posteriordashed lines: priorgrey-shading: uncertainty
Inter-annual variability
20 June 2016 ICOS Workshop Lund 16
2006 2008 2010 2012 2014−20
−10
0
10
20
Tg C
H 4 y−
1
North Eurasia
2006 2008 2010 2012 2014−20
−10
0
10
20
Tg C
H 4 y−
1
WSL2006 2008 2010 2012 2014
−10
−5
0
5
10
Tg C
H 4 y−
1
North America
2006 2008 2010 2012 2014−4
−2
0
2
4
Tg C
H 4 y−
1
HBL
2006 2008 2010 2012 2014−4
−2
0
2
4Tg
CH 4
y−1
Alberta
Thompson et al., in prep., 2016
case 1: prior wetlands LPX-Berncase 2: prior wetlands LPJ-DGVMsolid lines: posteriordashed lines: priorshading: uncertainty
p-value < 0.01 p-value < 0.01
Comparison to other estimates
N. America HBL N. Eurasia WSL
Prior this study 9.5 ± 5.1 2.9 ± 2.0 44.4 ± 12.5 11.0 ± 5.0
Posterior this study 16.6 ± 0.9 2.7 ± 0.14 55.2 ± 2.1 19.9 ± 0.4
Bergamaschi et al. 2013 12.2 3.6 30.4 11.6
Bruhwiler et al. 2014 8.1 2.7 49.7 18.4
Berchet et al. 2015 5 – 28
Miller et al. 2014 21.3 ± 1.6 2.4 ± 0.32
20 June 2016 ICOS Workshop Lund 17
Shown for 2005 – 2010 (overlapping period). Units TgCH4 y-1.
Thompson et al., in prep., 2016
Developments for CO2 inversions
18
FLEXINVERT-CO2
Statistical model of fluxes:optimize land biosphere fluxes, fixed ocean and fossil fuel fluxes
20 June 2016 ICOS Workshop Lund 19
Land Bio. flux
Fossil fuel flux
Ocean flux
Statistical flux model
FLEXPART emission sensitivities
Modelled CO2concentrations
Observed CO2concentrations
Model versus observation comparison
Forward run
Optimization
parametersFixed component
Statistical modelStatistical model based on Rödenbeck et al. 2005:
20 June 2016 ICOS Workshop Lund 20
f (x, y, t) = f fix, i (x, y, t)+αi fsh, i (x, y, t)mt=1
Nt
∑ gmt, itime (t)gms, i
space(x, y)pmt, ms, ims=1
Ns
∑"
#$
%
&'
i=1
N
∑
fixedfluxes(ocean,ff.)
optimizedfluxes(landbiosphere)
spatio-temporaldecomposition
parameters
2J(p) = (p − pb )TB−1(p − pb )+ (H f(p)− y)
TR−1(H f(p)− y)
Minimize cost function J(p) using gradient method:
Summary & ConclusionsMethane in the high northern latitudes• Total flux north 50°N of 81 TgCH4 y-1 or ~15% of global total• Anthropogenic emissions in Alberta significantly
underestimated by inventories, e.g. EDGAR-v4.2 • Wetlands emissions in HBL comparable to LPX-Bern and
other inversion estimates• Anthropogenic emissions in WSL likely underestimated in
EDGAR-v4.2
Developments for CO2 inversions• initial design in place end of 2016• first inversions of CO2 planned in 2017
20 June 2016 ICOS Workshop Lund 21
AcknowledgementsObservations:
E. Dlugokencky, M. Sasakawa, T. Machida, D. Worthy, T. Aalto, J. Lavric, C. Lund Myhre
Miscellaneous:R. Spahni, G. van der Werf, P. Bergamaschi
Financial support: Nordforsk funded project: eSTICCResearch Council of Norway funded projects:ICOS-Norway, SLICFONIA and EVA
20 June 2016 ICOS Workshop Lund 22