Convergence and synthesis of regional top-down and bottom-up estimates of CO 2 flux estimates: Results from the North American Carbon Program Midcontinent Intensive (MCI) regional study Kenneth Davis 1, Arlyn Andrews 2, Varaprasad Bandaru 3, F. Jay Breidt 4, Dan Cooley 4, Scott Denning 4, Liza Diaz 1, Kevin Gurney 5, Ram Gurung 4, Linda Heath 6, R. Cesar Izaurralde 3, Thomas Lauvaux 1, Zhengpeng Li 7, Shuguang Liu 7, Natasha Miles 1, Stephen Ogle 4, Scott Richardson 1, Andrew Schuh 4, James Smith 6, Colm Sweeney 2, Tristram West 3 1 The Pennsylvania State University, 2 NOAA ESRL, 3 Pacific Northwest National Laboratories, 4 Colorado State University, 5 Arizona State University, 6 USDA Forest Service, 7 EROS Data Center RECCAP meeting, 26 May, 2011, Shepherdstown, West Virginia, USA
outline Objectives Results – Inventory – Inversion – Comparison and synthesis Work to come
NACP Midcontinent Intensive (MCI) To what degree can we demonstrate convergence in regional flux estimates using top-down and bottom-up methods? C CO 2 C Atmospheric Inversions Inventories
Evolution of the MCI 1999 US Carbon Cycle Science Plan (Sarmiento and Wofsy) proposed regional atmospheric inversions white paper by Pieter Tans proposed the U.S. midcontinent as a good experimental site – agricultural fluxes are known because of harvest/inventory data Midcontinent Intensive (MCI) science plan (Ogle et al) spelled out the objectives of this contribution to the North American Carbon Program (NACP). Field work, Analyses are at hand! The primary objective of the NACP MCI is to test of our ability to achieve convergence of “top-down” and “bottom-up” estimates of the terrestrial carbon balance of a large, sub- continental region.
Experimental design Dense, tower-based greenhouse gas measurement network Relatively simple terrain and dense meteorological data These yield our best chances to derive robust flux estimates using atmospheric inversions. Excellent “bottom-up” flux estimates from inventory methods This provides the test for the atmospheric inversion methodology.
MCI Study Domain
“Inventory”
Δ SOC Yield Total Eroded C Δ Live Aboveground C (0) Δ Dead Aboveground Litter C Δ Dead Belowground Litter C Δ Live Belowground C (0) Cropland Carbon: Estimation based on stock change on cropland fields and includes key lateral flows in harvested grain and eroded C. West et al., 2011; Ogle et al., 2010
Forestland Carbon: Estimation based on stock change on forest land stands and includes key lateral flow of harvested woody products. Forest C Δ Live Aboveground C Δ Live Belowground C Δ SOC Δ Dead Aboveground CWD & Litter C Δ Dead Belowground CWD & Litter C Eroded C Timber Harvest Forest C Δ Live Aboveground C Δ Live Belowground C Δ SOC Δ Dead Aboveground CWD & Litter C Δ Dead Belowground CWD & Litter C Eroded C Timber Harvest Smith et al., 2003; EPA, 2009
Inventory Uncertainty Assessment Simulation Model PDF Scaling Uncertainty Results Structural Uncertainty PDF 95% Confidence Interval Input Uncertainies * Ogle et al., Global Change Biology, 2010 * For example: -Crop yield data. -Yield to carbon conversion coefficient. -Fertilizer application rate.
Cropland Carbon Budget for Seed production 3 United States Cropland Carbon Budget for 2008 Crop carbon Net soil C change Beginning C stock (15) Carryover Carbon Stock NPP 595 Imported C carryover to following year (2009) 255 Harvested C stock (255) Available C Stock for 2008 (258) Non-grain C stock (3) Food (Human) Feed (Livestock) Exported C carryover from previous year (2007) Decomposition 329 Fuel (Ethanol & Biodiesel) Fiber (Cotton) Processing waste 1 “Where does carbon in crops ultimately end up?” “Can we account for all carbon and balance the budget?” West et al, 2011
12 Cropland Carbon Budget for 2008 Seed production 3 United States Cropland Carbon Budget for 2008 Crop carbon Net soil C change Beginning C stock (15) Carryover Carbon Stock NPP 595 Imported C carryover to following year (2009) 255 Harvested C stock (255) Available C Stock for 2008 (258) Non-grain C stock (3) Food (Human) Feed (Livestock) Exported C carryover from previous year (2007) Decomposition 329 Fuel (Ethanol & Biodiesel) Fiber (Cotton) Processing waste 1 What part of this budget does the atmosphere see? Outside budget boundaries Included in Vulcan fossil fuel estimates West et al, 2011
National-scale agricultural inventory 13 Harvested biomass is transported out of the MCI region. Agriculture is a strong sink from the regional atmospheric perspective West et al, 2011 MCI domain
MCI inventory estimates: Forest and crop yield dominate Units are Gg C per ½ degree pixel.
(Gurney et al. 2009)
Units are Gg C per ½ degree pixel.
MCI inventory summary Carbon sink associated with cropland due to carbon fixation through photosynthesis and lateral transport of harvested grain (West et al. 2011) dominates. Uncertainties are dominated by crop yield data and crop yield to C coefficients. (Data are relatively precise, but contribute to large fluxes.) Agricultural uncertainties are highly coherent across space.
Inversion
CO 2 Concentration Network: 2008 Midcontinent intensive, INFLUX, Gulf coast intensive,
Corn-dominated sites MCI Tower-Based CO2 Observational Network
Large differences in seasonal drawdown, despite nearness of stations. 2 groups: ppm drawdown and 24 – 29 ppm drawdown. Tied to density of corn. Mauna Loa Miles et al, in review MCI 31 day running mean daily daytime average CO2
Daily differences from day to day (or site to site – not shown) as large at 30 ppm. Synoptic variability in boundary-layer CO2 mixing ratios: Daily daytime averages Miles et al, in review
Inversion Toolbox: “Forwards” system Air Parcel Sources Sinks wind Sample Network of tower-based GHG sensors: (9 sites with CO 2 ) Atmospheric transport model: (WRF, 10km) Prior flux estimate: (SiB-Crop and CASA) Boundary conditions: CO2: NOAA aircraft profiles and Carbon Tracker Met: NCEP meteorology Lauvaux et al, in prep, A
Inversion Toolbox, continued Lagrangian Particle Dispersion Model (LPDM, Uliasz). – Determines “influence function” – the areas that contribute to GHG concentrations at measurement points. Influence functions include the lateral boundaries. Inversion solves for both surface flux and boundary condition corrections. Bayesian matrix inversion. Weekly time step. 20km resolution. Experiment with coherence of solution in space: Default 100 km. Lauvaux et al, in prep, A
Inversion method (graphic) Estimated together Enhance uncertainty assessment by experimenting with the prior and the uncertainty estimates (model-data, prior) and examining the impact on the derived fluxes.
CO2 boundary condition adjustment CT vs. NOAA aircraft profiles Lauvaux et al, in preparation, A
Prior flux estimate Posterior flux estimate Lauvaux et al, in preparation, A Spatial pattern of NEE is not overly sensitive to the prior. Units are TgC/degree 2, Jun-Dec07
Regionally and time integrated C flux uncertainty assessment Experiments with the PSU inversion include varying the: - prior flux - prior flux uncertainty (magnitude and spatial correlation) - model-data error (magnitude and temporal correlation) - boundary condition temporal persistence. Net flux estimate is fairly robust to the assumptions made in the inversion. Lauvaux et al, in preparation, A
Impact of observational network: Tower removal experiments Prior flux Posterior flux with all sites Posterior with only “corn” sitesPosterior without “corn” sites Regional integral is fairly robust to tower removal. Spatial patterns are quite sensitive to tower removal. Lauvaux et al, in prep, B
Spatial correlation in [CO 2 ] residuals for two transport models: Signs of oversampling? Carbon Tracker Growing Season 2007 WRF-CASA Growing Season 2007 CT 2007 shows that sites close to each other and with the same vegetation are the only ones highly correlated (WBI- Kewanee). Using the same fluxes but a different atmospheric transport model (WRF) to predict [CO 2 ] produces substantially higher spatial correlations. Diaz et al, in prep Daily daytime averages.
Comparison and Synthesis
Compare? Or Combine?
Cooley et al, in preparation
Total uncertainty is reduced. Inventory and inversions bring independent samples of the same quantity. Model for RECCAP best estimates.
Cooley et al., in preparation Ogle et al., in preparation
Conclusions Regional C flux inverse estimates for the MCI appear to converge with inventory estimates. Regional C flux inverse estimates for the MCI are fairly robust to assumptions. Regional sums of NEE do not require a very dense observational network, but spatial patterns are highly sensitive to the network. Differences across inversion systems (transport, structure of inversion) have not yet been assessed.
What’s next? MCI synthesis papers MCI atmospheric transport uncertainty analyses INFLUX and Gulf coast intensives
CO 2 Concentration Network: 2008 Midcontinent intensive, INFLUX, Gulf coast intensive,
INFLUX (Indianapolis FLUX) Project Goals: Compare top-down emission estimates from aircraft and tower-based measurements with bottom-up emission estimates from inventory methods (include CO2 – fossil and biological, and CH4) Quantify uncertainties in the two approaches
Why Indianapolis? Medium-sized city, with fossil fuel CO2 emissions of ~3.4 MtC yr -1 Located far from other metropolitan areas, so the signal from Indianapolis can be isolated with relative ease Flat terrain, making the meteorology relatively simple View of Indianapolis from the White River (photo by Jean Williams) 1
Tower-based measurements: continuous Current: continuous measurements of CO2 at two sites Planned Two sites measuring CO2/CO/CH4 Three sites measuring CO2/CO Three sites measuring CO2/CH4 Four sites measuring CO2
Tower locations Sites 1 and 2 are currently measuring CO2. Sites 3 through 12 are planned, with tentative locations shown. Mixture of continuous CO2, CH4 and CO sensors, and flask 14CO2 data
Hestia Annual Fluxes for Indianapolis
Purdue airborne sampling (budget flux estimates, source ID, transport test) Mays, K. L., P. B. Shepson, B. H. Stirm, A. Karion, C. Sweeney, and K. R. Gurney, Aircraft-Based Measurements of the Carbon Footprint of Indianapolis, Environ. Sci. Technol., 43,
Publications in press or published EPA (2009) Inventory of U.S. greenhouse gas emissions and sinks: publications/emissions, United States Environmental Protection Agency, Washington, D.C. Gurney, K.R., D.L. Mendoza, et al. (2009). High resolution fossil fuel combustion CO2 emission fluxes for the United States. Environmental Science and Technology 43: Ogle, S.M., F.J. Breidt, M. Easter, S. Williams, K. Killian, and K. Paustian Scale and uncertainty in modeled soil organic carbon stock changes for US croplands using a process-based model. Global Change Biology 16: Smith, J.E., L. Heath, J.C. Jenkins Forest volume-to-biomass models and estimates of mass for live and standing dead trees of U.S. forests. Gen. Tech. Rep. NE-298. Newtown Square, PA: U.S. Department of Agriculture, Forest Service, Northeastern Research Station. 57 p. West, T.O., N. Singh, G. Marland, B.L. Bhaduri, A. Roddy The human carbon budget: An estimate of the spatial distribution of metabolic carbon consumption and release in the United States. Biogeochemistry 94: 29-41, DOI /s z. West, T.O., V. Bandaru, C.C. Brandt, A.E. Schuh, S.M. Ogle Regional Uptake and Release of Crop Carbon in the United States. Biogeosciences, In review.
Richardson et al, cavity ring down spectroscopic CO2 field measurements Stephens et al, LI-820 based, well-calibrated CO2 field measurements Miles et al, MCI atmospheric CO2 observations and the impact of the corn belt Lauvaux et al, A, Regional flux inversion methodology applied to the MCI, Lauvaux et al, B. Sensitivity of regional MCI inversion to tower sampling density Diaz et al, Analysis of atmospheric CO2 model- data residuals for the MCI Publications in review or prep
These data are open! Collaborators are welcome.