Prospects for an Integrated Landscape CO2 Model

1

• Prospects for an Integrated Landscape CO2 Model
• Gordon Green 12/19/07
• The City University of New York

2
Simplified Global Carbon Cycle Model
As fossil fuel stock decreases, atmospheric CO2
stock increases.
Stella model derived from NASA
3
Mitigation

• Four areas that can help bring the system closer
  to equilibrium
• Sequestration of CO2 in terrestrial sinks
• Conservation and fuel substitutions
• Geological capture and sequestration of CO2 as
  part of the energy generation process
• Emission reductions.
• The first two are solutions that can be
  implemented immediately geological capture and
  fuel substitutions will likely be implemented
  over the course of decades due to slow
  infrastructure turnover time MCCARL1.

4
Terrestrial Sinks

• Possible role of terrestrial offsets

Chart from MCCARL1
5
Terrestrial Sinks

• Relative timeframes terrestrial offsets

Chart from MCCARL1
6
Terrestrial Sinks

• Net global carbon storage of plus or minus about
  2 Pg per year is expected RANDERSON.
• Because anthropogenic factors are dominant, both
  the sign and magnitude of this net carbon storage
  will likely be determined by policy decisions.
• Models have become an integral part of choosing
  between different policy options.

7
Models

• descriptions of how the world might work are
  hypotheses, and often they can be translated into
  quantitative predictions via models. (Hilborn
  and Mangel, 1997, quoted in RIZZOLI)
• quantitative models predicting the outcome of
  natural processes on the surface of the earth
  dont work. PILKEY
• When used improperly, mathematics becomes a
  reason to accept absurdity. J. OMally, quoted
  in PILKEY
• The term Model is used here to refer to
  computer-based simulation models, not purely
  conceptual or mathematical models.

8
Modeling Pitfalls

• Results can be difficult to validate against the
  real world.
• Interpreting result data can become an exercise
  in circular logic, wherein all that is really
  described is the model and its output.
• The complexity of the model can become an
  impediment to a clear understanding of the
  underlying phenomenon COUCELIS.
• Software models are often ostensibly created for
  decision support users, but in fact are not used
  by anyone other than the model builder who has
  little to do with any real decision-making
  process MCINTOSH.
• The underlying assumptions may be unrealistic
  PILKEY.
• Natural human and sociological tendencies can
  also interfere for example, we tend to be blind
  to feedback mechanisms and non-linear responses
  PAHL-WOSTL.
• Consensus as to what to model, and how value
  resources, is elusive DALE.

9
Modeling Pitfalls

• PILKEY recommends qualitative approaches that
  only attempt to estimate order-of-magnitude and
  sign.
• Other techniques for handling uncertainty include
  REFSGAARD
• Expert elicitation (substantial consultation with
  those with intimate heuristic knowledge of the
  subject).
• Extended peer/stakeholder review.
• The use of multiple models for cross-calibration.
• Monte Carlo analysis especially in the context of
  a global (all parameters) sensitivity analysis.
• Use of multiple scenarios.
• Others

10
Steps in the Modeling Process

• Model study plan
• Data and conceptualization
• Model set-up
• Calibration and validation
• Simulation and evaluation REFSGAARD

Variations on a theme
11
Sidebar Calibration

• UML model of calibration process

Diagram from REFSGAARD
12
Model Scales
Local/landscape models appears under-represented
in CO2 models
13
Preliminary Requirements

• A usable integrated landscape model should
• Accommodate frequent changes in model design.
• Require a minimum of manually-gathered data.
• Be temporal / dynamic.
• Accommodate a high degree of uncertainty. Results
  expressed in orders of magnitude are likely to be
  more realistic than precise numbers with margins
  of error.
• Be easy to use and understandable by non-experts.

14
C Dynamics in Forests
Stella diagram derived from CHIBA
15
C Dynamics in Forests

• Forests alternate between being net sources and
  sinks of carbon, depending on the details of
  their development and management MASERA.
• The worlds forests and other vegetation now act
  as a net carbon sink due to the recovery
  agricultural areas and possibly CO2 fertilization
  effect due to increased atmospheric CO2 levels
  WICKS.
• Most predictions indicate that forests will
  continue to be a net carbon sink until about
  2050, at which point their capacity will begin to
  decrease until they become a net carbon source
  around 2080 WICKS.
• It is expected that an additional 1 to 2 Pg per
  year can be sequestered from 1995 to 2050 given
  the right management methods. Modeling efforts
  are expected to be an important part of making
  the right management decisions MASERA.

16
Sample Forestry Models

• Forest CO2 models usually include soil submodels
  as part of their estimates.
• They are often well-suited to driving by remotely
  sensed data (e.g., FOREST-BGC and use AVHRR
  satellite data) CHIESI.
• Non-spatial models can be spatialized by, for
  example, running the model for each pixel for 106
  years of climate data WICKS.
• EFIMOD-PRO model works somewhat differently in
  that it actually models the behavior of each
  individual tree (a SPE, single-plant ecosystem)
  KOMAROV.
• CO2FIX very user-friendly, designed for
  widespread use, and models C stocks and flows in
  forest stands MASERA.

17
CO2FIX Model

• Yield tables are included for multiple species
  and include stems, foliage, branches and roots.
• Since it is targeted at the timber industry, the
  model includes parameters for thinning and
  harvest.
• It simulates biomass growth with mortality
  figures due to harvesting, harvesting damage,
  senescence, and biomass turnover in to foliage,
  branches and roots.
• Uses the YASSO submodel, which is simpler to
  parameterize than some other soil models LISKI.

18
CO2FIX Model
YASSO diagram from MASERA
Screenshots from SCHELHAAS
19
CO2FIX Model

• CO2Fix parameterization is complex.
• Non-spatial ability to spatialize TBD.
• Sample simulation results

20
C Dynamics in Lakes and Rivers
Stella diagram derived from HANSON
21
C Dynamics in Lakes and Rivers

• Carbon cycles in place in rivers and lakes are
  less well-understood.
• Because they are not included in Kyoto
  calculations, oceanic carbon sinks are similarly
  neglected REDHANZ.
• The behavior of carbon in rivers is closely tied
  to soil erosion, which is simply omitted from
  many soil models (models to be considered below).
• Rivers, as carriers of eroded soil, are likely to
  be sources of CO2 RICHEY.

22
C Dynamics in Lakes and Rivers
Lakes that are low in dissolved organic carbon
and high in total phosphorus can act as carbon
sinks due to the general uptake of carbon in lake
life, but lakes usually tend to be sources of CO2
depending on the conditions.
Chart from HANSON
23
C Dynamics in Soils
Stella diagrams derived from POST
24
C Dynamics in Soils

• Soils can act as both a source and a sink of
  carbon depending on the level of disturbance and
  erosion, and the kind of vegetation that is
  growing on them FALLOON.
• Their behavior regarding CO2 is complex, and
  highly dependent on local conditions and land-use
  history.
• MCCARL2 estimates that most of the carbon in
  terrestrial systems resides in soils about 1500
  Pg, versus about 770 in the atmosphere. GEF1
  estimates that the global C sequestration
  potential in cultivated soils could be 20-30 Pg
  over the next 50-100 years, and FALLOON that
  10 of anthropogenic CO2 could be sequestered in
  soils.
• In most soils, the majority of carbon is held in
  soil organic carbon (SOC).

25
C Dynamics in Soils

• Soils that are disturbed through cultivation
  practices and erosion tend to be carbon sources,
  and soils that are undisturbed with natural
  vegetation tend to sequester carbon, at rates
  depending on the plant life they support GEF1.
  
• Recovering land from agriculture, or managing it
  in such a way that CO2 sequestration is
  maximized, could result in substantial carbon
  sequestration over time MCCARL2
• By lessening the intensity of soil tillage,
  producing crops that create more organic matter
  that can stay in the soil, and improving the
  soils with alternative management strategies, we
  can maximize the amount of carbon that remains
  sequestered MCCARL2.

26
Soil Models

• The forest models above include the soil models,
  but these are designed to work with forest
  models.
• The Global Environment Facility Soil Organic
  Carbon (GEFSOC) Modelling System is a good
  example of an integrated soil model, which seeks
  to operate as A generically applicable system
  for estimating SOC stocks and likely changes at
  regional and national scales GEF1,
• GEFSOC integrates three models RothC, Century
  and IPCC, in such a way that results can be
  cross-validated.
• The Century plant growth submodel is used as
  input to RothC, which doesnt include plant
  growth information. The IPCC model is a much
  simpler carbon modeling system that takes into
  account historical climatic information EASTER.
  
• Because CO2 sequestration depends heavily on
  prior land use, historical data is an important
  part of the GEFSOC model initialization EASTER.
  

27
GEFSOC Model
Parameterization of historical data can be
complex. This is part of what makes soil models
difficult to drive using remotely-sensed data

• Figure 13. Management sequence diagrams for MLRA
  46. Management system abbreviations are as
  follows
• NF native forestCC clearcut tree removalPC
  partial cut tree removal FIRE stand-replacing
  fire RF regenerating forestCSG continuous
  small grainsDASG dryland alfalfa-small grainFSG
  fallow-small grain (conventional tillage) FSGO
  fallow-small grain-oilseedFSGM fallow-small
  grain (minimum tillage)FSGN fallow-small grain
  (no tillage)From EASTER et al, GEFSOC users
  manual

Diagram from SCHELHAAS
28
IPCC Model
IPCC model is much simpler, simple enough to
potentially be driven by pre-existing datasets
Screenshot from IPCC2
29
Soil Model Problems

• Parameterization complexity is just one of the
  difficulties facing soil models.
• Another is that soil maps tend to be quite
  heterogeneous, covering different depths and
  using different soil categorizations, so GIS data
  sources can be difficult to integrate.
• Soil dynamics tend to be complex and non-linear,
  and simple estimation strategies may be
  inaccurate GEF1.
• Also, input data tends to be a constraint, and
  soil dynamics may not be readily apparent from
  remotely sensed data EASTER.
• For example, moving tilled land to no-till
  agriculture should sequester carbon, while
  implementing no-till agriculture on native
  grassland would probably result in CO2 emission.

30
Erosion

• According to LAL2, the role of erosion in the
  global carbon calculations is not well
  understood. He estimates that The amount of
  total C displaced by erosion may be 4-6 Pg/yr.
  Erosion-induced emission may be .8 to 1.2 Pg per
  year.
• Most models consider respiration and decay
  without considering erosion.
• SOC levels depend on the difference between
  inputs (esp. plant litter) and output
  (respiration, leaching and erosion).
• IZAURRALDE reports, based on the EPIC soil
  erosion model, that no-till watersheds are carbon
  sinks, and conventional-till watersheds act as
  carbon sources.
• The effects of erosion beyond the boundaries of
  the study watersheds are poorly understood and
  are likely to be the subject of further studies.

31
Other Submodels

• Other submodels could potentially be included in
  an integrated landscape C model, e. g.,
  transportation.
• Each of these is a large field in itself.
  Transportation especially has a rich simulation
  modeling tradition.
• Much of the work done on fuel efficiency, for
  example, translates directly to CO2 emission via
  arithmetic conversions.
• However, these aspects of the integration problem
  are beyond the scope of this current paper.

32
Data Sources

• The studies described above tended to rely on
  observational data, which can be expensive to
  gather.
• Despite the huge volume of available
  remotely-sensed data, the availability of
  accurate data remains a roadblock to accurate
  modeling, particularly with regard to soils.
• Remotely-sensed data is extremely useful for
  forest modeling, especially on a global or
  regional scale
• For a landscape or town-level purposes,
  higher-resolution imagery is needed, e.g.,
  through IKONOS 4m2 multiband imagery, or
  high-resolution orthophotos that include an
  infrared band.
• For estimating vegetation biomass, radar and
  lidar are excellent options PATENAUDE2.
  Unfortunately, there is as yet no global lidar
  data set, and coverage is sparse.
• Obtaining data that can calibrate model results
  is difficult. Directly measuring CO2 fluxes is
  an interesting possibility, and the AmeriFlux
  network makes direct CO2 measurements freely
  available. However, these measurements are
  localized, and it is difficult to infer any
  trends from the widely dispersed collection
  towers AMERIFLUX.

33
Case Study Area

• Northampton, Massachusetts, was chosen because it
  includes a variety of land uses, including
  agricultural and public and private forests it
  contains substantial urban and residential areas
  and there is a wealth of GIS data available
  through state and local governments.
• Land cover can be discerned from combining land
  use/land cover polygons with remotely sensed
  vegetation indexes.
• Soil polygons are available, but to what extent
  they can be used directly in soil models remains
  to be determined.
• Topographical and hydrological data may be usable
  in an erosion model.
• Transportation flows and other ancillary data are
  available.
• A key question will be the possibility of
  obtaining datasets for calibration and validation
  of the submodels, and to what extent these steps
  in prior model studies can be carried over into
  the integrated context.

Map from Springfield Public Library,
http//www.springfieldlibrary.org/reading/connecti
cut.html
34
Model Types
In an integrated model, submodels may potentially
require different model types, most likely
integrated in a GIS
35
Case Study Data Sets
Key question to what extent can pre-existing
datasets be used to drive an integrated model?
This depends on many factors...
Data layers from MassGIS, http//http//www.mass.g
ov/mgis
36
Additional Case Study Datasets
Datasets for other aspects of C dynamics that
could be included in an integrated model are
also available
Average Daily Traffic Volumes, Town of
Northampton, MA, Pioneer Valley Planning
Commission, 2001. Light Pollution from
International Dark Sky Association, 1996-7.
37
Hypothetical Process
Given the available data, this is an
approximation of how the integrated model might
work
38
Conclusion

• Based on this preliminary review of the models
  and data available, integrating multiple models
  into a unified landscape model is probably
  technically possible, but the usefulness of the
  results will depend on details that have not yet
  been settled. There are many key questions that
  will require further research
• Is there a modeling integration strategy that is
  flexible enough to accommodate changing models,
  parameters, and data?
• Should the integrated model drive existing
  submodels, or implement them in a new
  environment?
• What exactly will be needed for ground truth,
  calibration, and validation of the integrated
  model? How much of prior validation and
  calibration effort can be applied to the
  submodels in an integrated context?
• How would accommodating multiple modeling
  architectures, such as agent-based and systems
  dynamics models, affect the integration?
• What is the best scale to use town (e.g.,
  Northampton), landscape (e.g., Pioneer Valley),
  watershed (e.g., Connecticut River Watershed)?
  Can the scale be flexible?

39
Conclusion

• How reliable are the vegetation estimates
  available from infrared orthophotos? Are canopy
  structure measurements required, or will LAI
  suffice?
• What methodologies make sense for handling
  uncertainty? How do these affect the data,
  parameter, and computational requirements?
• Can complexity be reduced while maintaining
  enough accuracy for results to be meaningful,
  even if they are order-of-magnitude estimates?
• Are there any other steps that can be taken to
  reduce the consequences of other modeling
  pitfalls such as omission of feedback loops and
  invalidating assumptions?
• How and to what extent should human behavior and
  economic factors be integrated?
• How accurate can the results be given especially
  limited soil data availability?
• Who would use it??

40
References
CHIBA Chiba, Y. Simulation of CO2 budget and
ecological implications of sugi (Cryptomeria
japonica) man-made forests in Japan. Ecological
Modelling. Vol. 111, 1998. CHIESI M. Chiesi,
F. Maselli, M. Bindi, L. Fibbi, L. Bonora, A.
Raschia, R. Tognetti, J. Cermak and N.
Nadezhdina. Calibration and application of
FOREST-BGC in a Mediterranean area by the use of
conventional and remote sensing data.
Ecological Modelling, Volume 154, Issue 3, 1
September 2002. EASTER Easter, M. et al. The
GEFSOC Soil Carbon Modelling System A Tool for
Conducting Regional-Scale Soil Carbon Inventories
and Assessing the Impacts of Land Use Change on
Soil Carbon. Agriculture Ecosystems and
Environment, 122, 2007. HANSON Hanson, P. C.,
Pollard, A.I., Bade, D.L., Predick, K.,
Carpenter, S. R., Foley, J. A. A Model of Carbon
Evasion and Sedimentation in Temperate Lakes.
Global Change Biology, 10, 2004. GEF1
Assessment of Soil Organic Carbon Socks and
Change at National Scale. University of Reading,
Technical Report of GEF Co-Financed Project No.
GFL-2740-02-4381, June 2002-July 2005. IPCC2
Tool for Estimation of Changes in Soil Carbon
Stocks Associated with Management Changes in
Croplands and Grazing Lands Based on IPCC Default
Data. Intergovernmental Panel on Climate Change,
2003.
41
IZAURRALDE Izzauralde, R. C., Williams, J. R.,
Post, W. M., Thoson, A. M., McGill, W. B., Owens,
L. B., Lal, R. Long-Term Modeling of Soil C
Erosion and Sequestration at the Small Watershed
Scale. Climatic Change, 2007. LAL2 Lal R.
Soil erosion and the global carbon budget.
Environmental International, January
2003. MASERA Omar R. Masera, J.F.
Garza-Caligaris, M. Kanninen, T. Karjalainen,
Liski, J., G.J. Nabuurs, A. Pussinen, B.H.J. de
Jong, G.M.J. Mohren. Modeling carbon
sequestration in afforestation, agroforestry and
forest management projects the CO2FIX V.2
approach. Ecological Modelling, 164 (2003)
177199. MCCARL1 McCarl, B., Metting, F.B.,
Rice, C. Soil Carbon Sequestration. Climatic
Change, 2007, 80 1-3. NASA NASA Earth Science
Enterprise, http//earthobservatory.nasa.gov/Libra
ry/ CarbonCycle/carbon_cycle4.html, accessed
11/2007. PAHL-WOSTL Pahl-Wostel, C. The
implications of complexity for integrated
resources management. Environmental Modelling
and Software. Vol. 22, No. 5, 2007. POST Post,
W. M., and Kwon, K. C. Soil Carbon Sequestration
and Land-Use Change Process and Potential.
Global Change Biology, 2000, 6. SCHELHAAS
Schelhaas, M.J., P.W. van Esch, T.A. Groen, M.
Kanninen, J. Liski, O. Masera, G.M.J. Mohren,
G.J. Nabuurs, L. Pedroni, A. Pussinen, A.
Vallejo, T. Palosuo, T. Vilén. 2004. CO2FIX V
3.1 A modelling framework for quantifying
carbon sequestration in forest ecosystems.
ALTERRA Report 1068. Wageningen, The
Netherlands. USDA US Department of
Agriculture, Natural Resources Conservation Unit.
Land Resource Regions and Major Land Resource
Areas of the United States, the Caribbean, and
the Pacific Basin. United States Department of
Agriculture Handbook 296, 2006.