1. Assimilation of Tower- and Satellite-Based Methane Observations for Improved Estimation of Methane Fluxes over Northern Eurasia
D.P. Lettenmaier1 and T.J. Bohn1
1Dept. of Civil and Environmental Engineering, University of Washington, Seattle, WA, USA
15th Annual LCLUC Science Team Meeting
University of Maryland University College
Adelphi, MD, 2011-Mar-29

2. Importance of Lakes and Wetlands
Large CO2/CH4 source
CH4 estimates revising upwards due to ebullition
Wetlands:
Largest natural global source of CH4
Large C sink
Northern Eurasia contains:
30% of world’s wetlands (Gorham, 1991)
Large portion of world’s lakes
West Siberian Lowlands
Lehner and Doll, 2004
Lake/wetland carbon emissions are sensitive to climate
High latitudes experiencing pronounced climate change

3. Climate Factors CO2 CH4 CO2 Temperature (via metabolic rates)
Relationships non-linear
Water table depth not uniform
across landscape - heterogeneous
NPP
Living Biomass
Acrotelm
Temperature
(via evaporation)
Aerobic Rh
Water Table
Precipitation
Anaerobic Rh
Catotelm
Note: currently not considering export of DOC from soils

4. Lakes and Wetlands
Lake/wetland CH4/CO2 emissions depend on T, C, nutrients, oxidation state, etc
Wetland CH4/CO2 fluxes also depend on soil moisture
CH4
Seasonally Inundated Soil (flooded)
Permanent Lakes
Unsaturated Soil
Saturated Soil
Water Table
Areal extent of wet zones can vary substantially in time
Temporal variations in area play important role in response of CH4/CO2 fluxes to climate
We can improve model estimates by taking dynamic behavior into account

5. Monitoring the Sources
In situ measurements (water table, CH4 flux) are localized and sparse
Large-scale observations (towers, satellites) are less direct
Extent of saturated/inundated wetlands (e.g. SAR, passive microwave)
Atmospheric CH4 concentrations (e.g. towers, GOSAT)
Large-scale runoff (stream gauges)
Large-scale biogeochemical models can bridge the gap between indirect observations and the processes we want to monitor
Sub-surface soil moisture, temperature, methanogenesis
Simultaneous large-scale observations of multiple variables can constrain model parameters

6. Science Questions
How do wetland methane emissions in W. Siberia respond to environmental conditions over a range of temporal and spatial scales?
Can wetland methane emissions in western Siberia be effectively monitored by a combination of models, towers, and remote sensing observations?
Can tower and/or satellite observations be used to constrain errors in or identify areas for improvements in the models (e.g. poor representations of physical processes, state uncertainties, parameter uncertainties)?
How much do SAR-based surface water products improve the match between simulated and observed methane concentrations?
Can integration of satellite-based and tower-based (near-surface) atmospheric methane concentrations via data assimilation methods lead to better understanding of the space-time distribution of methane emissions over northern Eurasia?

7. Collaborators
Dennis Lettenmaier (PI) and Ted Bohn, University of Washington
Kyle McDonald (co-PI), Erika Podest, and Ronny Schroeder, JPL/NASA
Shamil Maksyutov, Motoki Sasakawa, Heon-Sook Kim and Toshinobu Machida, National Institute for Environmental Studies (NIES), Japan
Mikhail Glagolev, Moscow State University, Russia
Heon-Sook Kim (till March 2011 at RIHN, Kyoto, in NIES from April 2011)

8. Modeling Framework VIC hydrology model
Large, “flat” grid cells (e.g. 100x100 km)
On hourly time step, simulate:
Soil T profile
Water table depth ZWT
NPP
Soil Respiration
Other hydrologic variables…
Link to CH4 emissions model (Walter & Heimann 2000)
Carbon cycle processes: Photosynthesis, plant respiration, etc. taken from the BETHY model (Knorr, 2000) – uses a Farquhar/Collatz scheme for photosynthesis. Soil respiration is a Lloyd-Taylor formulation taken from the LPJ model (Sitch et al 2003).
CH4 emissions model (Walter & Heimann 2000) uses Q10 formulations for dependence of methanogenesis and oxidation on temperature; models 3 pathways for transport to surface: diffusion, ebullition, and plant-aided transport. Inputs are: local NPP, soil temperature, and water table position. Modeled soil column is 1 m deep. This is realistic; peatlands in W. Siberia are 1-3m deep according to Sheng et al (2004); most labile carbon is considered to be in the top meter (and most is concentrated in the root zone, roughly the top 50 cm).

9. Recent Progress in Modeling
3 Main approaches to modeling
large-scale CH4 emissions:
1. “Distributed” scheme:
Heterogeneous water table and CH4
Used by UW
Water Table Position
Saturated /
Inundated
Depth
Soil Surface
2. “Uniform” scheme:
Water table and CH4 emissions vary in time but uniform across grid cell
Oversensitive to climate
Biased +/- 30%
Water Table
CH4 Emissions
Hopefully this slide better summarizes our paper. I will also send the pdf of our paper.
3. “Wet-Dry” scheme:
Simulates time-varying saturated area
Ignores water table and emissions from unsaturated zone
Biased up to 30% too low
Cumulative Area
“Wet”
“Dry”
Wetter
Drier
(Bohn and Lettenmaier, GRL, 2010)

10. Distributed Water Table
DEM (e.g. GTOPO30
or SRTM)
Topographic Wetness
Index к(x,y)
Summarize for a Single 100 km Grid Cell 1
Cumulative
Area Fraction кi
кmax
кmin
Topographic Wetness Index CDF
1 km Resolution
Water Table Depth Zwt(t,x,y) 1
Cumulative
Area Fraction
Water Table
Soil Storage Capacity CDF(mm) = f(кi)
Smax
Saturated
At Surface
VIC Spatial
Avg Soil
Moisture (t)

11. Comparison with PALSAR
Spatial distribution of inundation compares favorably with remote sensing
This offers a method to calibrate model soil parameters
Observed Inundated Fraction (PALSAR Classification)
Simulated Inundated Fraction (at optimal Zwt)
Observed Inundated Fraction (PALSAR Classification)
Simulated Inundated Fraction (at optimal Zwt)
ROI 1
ROI 3
Spatial distribution of simulated inundated area compares favorably with remote sensing
Given equifinality of hydrological parameter sets, it is important to find effective ways to constrain simulated water table depth.
Unlike the binary nature of saturated area given by uniform water table depth, saturated area given by distributed water table varies smoothly as a function of average water table depth - this gives much better constraints on hydrologic model
ROI 2
ROI 4
Approx.
30 km

12. VIC Dynamic Lake/Wetland Model
Water & energy balance model
Includes mixing, ice cover
Dynamic area based on bathymetry
Can flood surrounding wetlands based on topography
Special application: treat all lakes, ponds, and inundated wetland area as a single “lake”
Bowling and Lettenmaier, 2010

13. Composite Bathymetry + + + =
How much area is covered by water of depth >= a given threshold?
Sum the areas of all lakes having depth > threshold
Given a volume of surface water, we know how much area it will cover + + + =
Depth
Cumulative area

14. Lake Bathymetry/Topography
Lake size histograms from GLWD (Lehner and Doll, 2004) and LANDSAT
Lake depths from literature
ILEC
Arctic Thaw Lakes
Bog Pools
LANDSAT
GLWD
Bartalev et al (2003)
Lake storage-area relationship
LANDSAT courtesy of E. Podest and N. Pinto of NASA/JPL
Histograms shown are from the Chaya basin in SE W. Siberia (one of the 5 test basins mentioned later). LANDSAT 30m images classified for open water by JPL. GLWD considers lakes down to 1 km2 in size. Bartalev et al 2003 land classification has resolution of 1 km, but unlike GLWD it is a pixel classification rather than “known” lake area – I was just using it for comparison in that plot to see the effects of using a pixel classification.
Lake depths for large lakes are taken from the ILEC database (only lakes in W. Siberia are shown). Arctic thaw lakes are from Laura Bowling’s study in Toolik Lake region. Bog pools are from a study of bogs in the Hudson Bay Lowlands in Canada.
SRTM and ASTER DEMs for surrounding topography

15. First Phase: Historical Reconstruction
Select test basins in West Siberian Lowlands
Simultaneous calibration to both streamflow and inundated area
Streamflow gauge records from R-Arcticnet (UNH)
Inundated area derived from AMSR/QSCAT (NASA/JPL)
Hydrologic calibration parameters
Soil layer thickness
Infiltration distribution
Maximum baseflow rate
Effective lake outlet width
Parameters – CH4
Calibrated against observations at Bakchar Bog (Friborg et al, 2003)
For lakes, use range of observed CH4 rates from literature:
10-50 mg CH4/m2d (Repo et al, 2007)
mg CH4/m2d (Walter Anthony et al, 2010)
Ultimate goal: estimate of responses of lakes and wetlands across West Siberia to end-of-century climate
For now: what are the sizes and seasonal behaviors of the various CH4 sources?

16. Study Domain: W. Siberia
Close correspondence between:
wetness index distribution and
observed inundation of wetlands from satellite observations
Wetness Index from SRTM and ASTER DEMs
Ural Mtns
WSL Peatland Map
(Sheng et al., 2004)
Ob’ R.
Permafrost
Syum
TWI
16 (wet)
Konda
Dem’yanka
8 (dry)
Vasyugan
Chaya

17. Comparison With Observed Discharge
Monthly Avg
Annual Flow
Syum
Modeled snowmelt pulse is narrower than observed
Vasyugan
Konda
Interannual variability is good (where record is long enough)
Dem’yanka

18. AMSR/QSCAT-Derived Inundation
Annual Max – Min Fractional Inundation
Daily, for snow-free days
25km resolution
I will send Ronny’s paper separately.
Max % - Min %
> 15 %
< 3%
Courtesy R. Schroeder, NASA/JPL

19. Comparison With Observed Inundation
Syum
Seasonal inundation can more than double lake area
Vasyugan
Canonical lake area
Konda
Poor match because AMSR/QSCAT is lower than canonical lake area
Dem’yanka
Large saturated area

20. Monthly Averages, 1948-2007, Syum Basin
1. Areas
Total Wetland = 18%
Unsaturated wetland is the largest contributor to CH4 in summer because of its large area
Lake + Inund + Saturated = 3 - 6%
Lake + Inund = 1 - 3%
Lake = 0.6%
2. CH4
1. Lake emits at same rate as saturated wetland
2. Lake emits at rate of 50 mg CH4/m2d
3. Lake emits at rate of 500 mg CH4/m2d
The dashed line in the final plot (scenario 3) goes below the lake emissions line because in winter, the unsaturated wetland actually oxidizes CH4 – net negative emissions. So, adding it to the total brings the total down from 100 to about 35. Once again this emphasizes the importance of the unsaturated zone in wetland emissions.
CH4 (mg/m2mo)
CH4 (mg/m2mo)
CH4 (mg/m2mo)
(=30 mg/m2d)
(Fluxes per unit area of entire basin)

21. Annual Averages - Syum Lakes have highest emission rate per unit area
If we assume high lake CH4 emission rates ( mg/m2d) constant throughout year, lakes reach 9-32% of basin emissions
But total lake contribution small
Saturated wetland is largest component
Unsaturated wetland is second largest, due to its large area
Relative contributions of saturated and unsaturated zones vary in time with soil moisture
This depends on wetland CH4 parameter set, and assumption that it applies everywhere in the wetland

22. Current Status
Refining the CH4 model parameters with extensive in situ chamber flux observations from M. Glagolev (Moscow State U) and S. Maksyutov (NIES)
Simulation of methane fluxes from W. Siberia, , currently under way

23. Data Assimilation
Link VIC model to NIES atmospheric transport model (NIES-TM)
State variables
Soil moisture
Surface inundation (linked to soil moisture)
Atmospheric CH4 concentration
Observations
AMSR/QSCAT inundation
Tower [CH4] (initially)
Possible addition of GOSAT Total Column [CH4]
AIRS does allow retrieval of atmospheric CH4, but only in the mid-upper troposphere, as opposed to GOSAT’s total column CH4. AIRS has twice-daily temporal resolution, 50km spatial resolution. AIRS data could be useful to us but the nice thing about GOSAT’s total column CH4 is that we don’t have to rely on accuracy of vertical mixing in the atmospheric transport model. It might be interesting for someone to take the difference of GOSAT – AIRS, to get lower troposphere CH4 content…

24. State Variables and Observations
(observed by
towers and satellite)
Surface Water
CH4
(observed by satellite) As
Soil Surface
Depth ? ? ?
The main idea here is: remote sensing can’t tell us total soil moisture or water table depth; it can only tell us how much of the surface has standing water (which is a subset of saturated soil). But this is a lower bound on saturated soil extent; seasonal flooding therefore gives us a time-varying lower bound on where the water table intersects the surface. This at least pins down the wet end of the water table distribution.
Meanwhile, to get a handle on how much water is stored in the soil where we don’t have surface water, CH4 observations depend on soil moisture content. Total CH4 will therefore constrain total soil moisture content and average water table depth in the unsaturated zone. The amplitude of CH4 variations in response to variations in soil T and soil moisture will tell us something about how uniform the water table is. If the water table is completely uniform, CH4 variations will have the maximum possible response to changes in soil moisture. Sensitivity to changes in soil moisture will decrease as the spatial variance of water table depth increases. So, if we trust our CH4 model calibration, CH4 will constrain not only average water table position but water table shape.
Water Table
(poorly constrained)
Water table position related to total soil moisture
Drier
Wetter
Cumulative Area

25. NIES Tower Network 60-100m towers Hourly measurements of CH4, CO2
2 sampling heights at each tower: 10-40m, 40-80m
(Courtesy M. Sasakawa, NIES)

26. JAXA GOSAT Launched 2009 Total column CH4 and CO2
10 km spatial resolution
1000 km swath width
3 days/44 orbit repeat cycle
CH4, June-August 2010

27. NIES Atmospheric Transport Model (NIES-TM)
Tracer transport/diffusion model
Driven by NCEP/NCAR fields
2.5 x 2.5 degree resolution (resolution of drivers)
15 Vertical layers (1000 mbar, 850, … up to 10 mbar)
15 min temporal resolution
West Siberian simulation nested within global simulation
CH4 sources for the rest of the world taken from Matthews and Fung (1987), Patra et al (2009)
Tracks several species and reactions relevant to CH4 transport and oxidation, OH by Sudo et al (2002)
Used successfully in inverse modeling of global CO2 sources in TRANSCOM experiment (Law et al., 2003)

28. Data Assimilation: Updates
Daily (AMSR/QSCAT only) or Weekly
Ensembles formed from perturbed initial soil moisture states
(or multiple NCEP/NCAR realizations, a la ESP)
Ensemble Kalman Filter
Update state variables in place, i.e. at same time/location as observation
Can’t use atm CH4 to update soil moisture due to time lag between source and observation
Ensemble Kalman Smoother (van Leeuwen and Evensen, 1996)
Takes account of all observations and model states at all times between updates
Can update state variables backwards in time, along particle trajectories between sources and observations

29. Conclusions
Lakes have large per-area emissions but in many cases small total area
The saturated and unsaturated areas of the wetland are large contributors of CH4 as a result of their large extent
Total fluxes from these areas can be constrained to some extent via calibration vs. streamflow and satellite observed inundation
Further constraint with ground observations is needed
Assimilating tower- and satellite-based observations should provide valuable constraints on simulated methane emissions

30. Thank You
This work was carried out at the University of Washington and the Jet Propulsion Laboratory under contract from the National Aeronautics and Space Administration.
This work was funded by NASA ROSES grant NNX08AH97G.