Global CO and NOx emission estimates using multiple species data assimilation

with GEOS-Chem adjoint model

Xuesong Zhang, Dylan B. A. Jones, Martin Keller, Thomas W Walker, Zhe Jiang, Daven K Henze,

Adam E Bourassa, D A Degenstein and Yves J Ronchon

IGC8, May 2, 2017

•Discrepancies in model transport and chemistry introduce large uncertainties in top-down CO emissions estimates.

• Jiang et al. (2015) found that inferred regional emission estimates could differ by as much as a factor of 2 due to differences in OH in GEOS-Chem.

Problems with inverse modeling CO: Sensitivity to chemistry

[From Jiang et al., 2015]

We assimilate TES and OSIRIS O3, OMI NO2 and MOPITT CO to constrain the CO-OH-

NOx- O3 chemistry in the model.

MOPITT CO TES O3

OMI NO2 OSIRIS O3

ucar.edu knmi.nl

earthdata.nasa.gov usask.ca

TERRA

AURA

ODIN

Studies Model Setup

CTM GEOS-Chem (4𝑜 × 5𝑜 × 47𝐿)

Assimilation period Nov 2009 and July 2010

DA scheme 4D-Var: AW=2 weeks

Observation

combinations

MOPITT CO, TES O3, OMI NO2,

OSIRIS O3

Optimized emissions CO, NOx and LNOx emissions

Opt. initial conditions O3

Estimated CO and NOx emissions in Nov, 2009

o CO emission estimates using the MSA are similar to the

CO estimates obtained using only MOPITT, with the

exception of the Asian estimates.

o NOx emission estimates using the MSA are also similar to

those from the OMI-only experiment (e.g. North American

and East Asian emissions are 2% and 5% lower than the

OMI-only estimate, respectively), but the European

emission estimate is 18% higher than the OMI-only

estimate.

o The consistency of the assimilated emission estimates between single-instrument and multiple-

instrument inversions vary seasonally, reflecting seasonal dependence of the chemistry, i.e. larger

differences in summer than in fall.

Evaluating assimilated O3 with HIPPO aircraft data

Changes in O3 in the assimilation

• A priori O3 in the NH extratropical lower

troposphere agrees well with HIPPO, but the

model is biased high in the tropics.

• The largest biases are in the UTLS (possibly due

to the coarse model resolution).

• The assimilation reduced the upper tropospheric

bias in the tropics and SH.

• Assimilation of TES and OSIRIS exacerbated the

bias in the NH extratropical upper troposphere.

HIPPO Campaign #2: Nov 2009

• OH changes in the northern

extratropics using all instruments

are driven by largely by MOPITT

and OMI.

• OH changes in the tropics using

all instruments are driven by O3

observations.

Changes in OH in the assimilation

Nov.

Jul.

2010

Uncertainties of multiple species data assimilation using 4D-var

Scaling on super observations of different species

Since the degree of freedom among different

observations vary significantly, adjoint forcing

scaling must be applied so that adjoint model

could respond to all the assimilated species.

When 𝛾𝑅 increases, model’s optimization

sensitivity to observations of species R increases.

With 𝛾𝑀𝑂𝑃 = 64 and 𝛾𝑂𝑀𝐼 = 16 in

the multiple-species assimilation, we

recover the global CO and NOx

emission estimates from the MOPITT-

only and OMI-only inversions.

𝐽 𝐱𝟎, 𝐩 =1

2

𝑛=1

𝑁

𝛾𝑅 𝐲𝐧 −𝐻𝑛 𝑀 𝐱0, 𝐩𝑇𝐑−1 𝐲𝐧 −𝐻𝑛 𝑀 𝐱0, 𝐩 +

1

2𝐩 − 𝐩𝑏 𝑇𝐁𝑝

−1 𝐩 − 𝐩𝑏 +1

2𝐱0 − 𝐱

𝑏 𝑇𝐁𝐱−𝟏 𝐱0 − 𝐱

𝑏

where xn is the model state (e.g., CO concentrations) at observation time n, p is the vector of sources

(CO emissions), 𝐱𝟎 is the initial conditions (O3).

Future opportunities and challenges

• MSA using 4D-var has potential to improve CO and NOx emission

estimates via improving tropospheric O3 concentrations. The assimilated species are chemically consistent, but the coupling of the NOx-CO-HOx-O3 chemistry can cause challenges:

NOx emission estimates are sensitive to biases in the initial O3 state due to the strong chemical coupling between NOx and O3.

Properly weighting the contribution from the different observations in the cost function is difficult.

• Currently, we have CTMs available to assimilate CO and NOx emissions at higher resolution (more than 4x5). However, for global inversions, we are battling with computational costs when assimilating CO and NOx at higher resolution.

Acknowledgements

This work was supported by funding from the Natural Science and Engineering Research Council of Canada and Environment and Climate

Change Canada