Abstract

  • Accounting of carbon fluxes from land-use and land-cover change (LULCC) involves a choice among multiple options, which grossly affect estimates:
    • temporal evolution of C stocks
    • initial state of C stocks
    • temporal attribution of C fluxes
    • treatment of LULCC fluxes that occurred prior to the simulation period
  • new Bookkeeping of Land Use Emissions model (BLUE)
  • quantify LULCC fluxes and atribute them to land use activities and countries

1. Introduction

  • LULCC fluxes account for about 10% of current annual CO$_2$ emissions
  • geographical attribution important for policy (Kyoto protocol, UNFCCC)
  • flux needs to be modeled for it is not directly observable
  • net exchange between atmosphere and land $=$ residual from C stocks and fluxes from atmosphere and ocean
  • need models to split into natural and anthropogenic contributions
  • methodological choices to be made regarding delayed fluxes
    • LULCC activity change relation between NPP and C decomposition on various timescales
    • simplest approach: system before and after change assumed to be in equilibrium, even delayed fluxes assumed instantaneous (committed fluxes), common in relation with remote sensing
    • distribute delayed fluxes uniformly over a certain time horizon: if future transitions can be foreseen, fluxes can be attributed to all of them
    • physically more accurate: process-based (DGVMs) or bookkeeping models (BKMs)
    • DGVMs do not attribute fluxes to individual LULCC because of computational constraints
    • BKMs track area and type of LULCC and combine with empirical response curves - “legacy scheme”: spatially and temporally explicit modeling of C stocks, taking succesion of LULCC events into account
  • policy may require “commitment periods”
  • problem of initializing stocks, as usual
  • BLUE allows systematic comparison of effects of different conceptual choices with one and the samle model

2. Methods

2.1 BLUE model

  • spatially explicit
  • tracks individual histories of successive LULCC events and their interactions in each grid cell
  • approach of cumulative excess C makes it computationally more efficient than splitting cells up into plots of land

  • able to track C fluxes by each year’s LULCC events through time

  • $0.5°$ grid cells, goal is $0.1°$
  • track areas affected by LULCC events, combines with empirical C densities
  • C densities $=$ amount of C per ha for each land cover type
  • C transferred between pools (indlucing porudct pools and atmosphere) based on prescribed fractions
  • excess pool $=$ current C stock $-$ equilibrium C stock, change with successive LULCC events

  • relaxation fluxes proportional to excess pool: exponential response curve (decay or regrowth) approach, makes summing up of results from different events straightforward and allows backward computation of LULCC events responsible for fluxes

  • Do we have $4\times4\times8\times11$ (cover type $\times$ transition type $\times$ pool type $\times$ plant functional type) pools for each grid cell?
  • By LULCC transition affected cover types and plant functional types are prescribed.
  • The current LULCC transition data set does not specify history of subgrid areas.
    • Hence: The BLUE modeling approach corresponds to distributing each new LULCC event proportionally by area across the different histories present in the cell.
  • To attribute present fluxes to previous transition years, a “temporal accounting” layer tracks on per-country basis (spatially ecplicit too memory intensive).
    • Excess C caused by LULCC events stored per pool per country
    • Resulting C fluxes computed by exponential response curves.
  • Emissions attributed to an event depend on previous events (or legacy C) by impact of C stock at the time of the event, and future events by impact on relaxation curve.

2.2 Input Data

  • LULCC transition data set
    • provides information on sub-grid scale on transitions between primary land, secondary land, cropland, and pasture
      • primary land: land not under active use at the start of historical reconstruction
      • secondary land: land affected by LULCC at some point in the past, not allowed to return to primary land
    • additional sub-grid transitions
      • shifting cultivation (gross transitions)
      • wood harvest with information on amount of harvested wood
    • for the years 1500 to 2004 (end)
  • Updated version
    • unreasonable differences to previous version from 1500 to 2014
    • hence only used from 2005 to 2012
  • LULCC transition overlain by map of potential natural vegetation
    • goal: split natural land (primary and secondary land) into plant functional types
    • to allocate cropland and pasture within a grid cell, we proportionally reduce all exiisting natural plant functional types
  • each plant functional type associated with two types of parameters:
    • C stock densities for different land types
    • response curves after transitions: original curves were picewise linear, we approximate them exponentially for efficient summation of different histories

2.3 Simulations

  • eight different accounting schemes (#1 through #8) based on three model runs and different post processing approaches

2.3.1 Model Runs

  • instant C stock changes to new eqilibrium: #6, #7, #8
  • legacy scheme:
    • realistic values from spinup since 1500: #1, #2, #3
    • spinup from 1500 and purge all legacy C in 2007 to assume equilibrium values at start of accounting period: #4, #5

2.3.2 Data Postprocessing - Legacy Model Runs

  • direct model output, attribute C fluxes to each year that really occur in this year: #1, #2, #4
  • attribute committed fluxes: add extra year at end of simulation with immediate fluxes: #3, #5
    • this way, no flux C is lost due to premature simulation end
  • final choice: include (#1) or exclude (#2) emissions from prior to the accounting period
    • not relevant in case of equlibrium C stocks at beginning of accounting period
    • for committed emissions, prior LULCC events are naturally excluded
  • #1: considered most natural (default)
    • historical C stock values
    • simulated temporal emissions
    • includes effects from LULCC events prior to accounting period

2.3.3 Data Postprocessing - Model Runs with Instantaneous Transitions

  • by definition equilibrium at beginning of accounting period
  • instantaneous attribution of fluxes (#6)
  • uniform flux distribution over 30 years, include (#7) or exclude (#8) fluxes from LULCC events prior to accounting method
    • uniform distribution computed by shifted boxcar filter

2.3.4 Additional Model Runs

  • estimate influence of net vs gross transitions
    • net (instead of gross) transitions computed from LULCC transition data set
      • clearing and abandonment accounted against each other
  • analysis of interprocess dependencies
    • switched off all harvest transitions
    • comparison with default run shows effect of harvest for other LULCC activity fluxes

2.4 Summary of the Eight Accounting Methods

  • #1 (default):
    • used in annual CO$_2$ budget estimates
    • most common approach to determine net LULCC flux in process-based models
    • UNFCCC approach for accountability of forest management activities
  • #2:
    • no fluxes from LULCC events prior to accounting period
    • can be used to illustrate importance of starting dates
  • #3:
    • not used in literature
    • could be used to extend #6 by effects of known LULCC events within given time span
  • #4:
    • reset to equilibrium at beginning of accounting period
    • often been used in combination with remote sensing data
      • available only for recent period of time
      • prescribe equilibrium in vegetation at beginning
  • #5:
    • quantifies committed emissions from #4
    • computed from #4 like #3 from #1
    • not used in literature
  • #6:
    • simplest schemes in terms of input data and methodological complexity
    • equilibrium starting stocks, instantaneous new states, committed fluxes
    • very often used to estimate net LULLCC flux
  • #7:
    • with more realistic curves (instead of uniform distribution) been used in earlier studies
  • #8:
    • second simplest model after #6, one additional parameter: time horizon (30 years)
    • used to account for foreseeable successive LULCC types
  • product carbon footprint standards suggest #7 or #8 (depending on data availability)

3. Evaluation of Simulated Carbon Stocks and fluxes

  • net LULCC flux is the most uncertain component in the global C budget
  • 2002-2012 about 0.8 PgC/yr $\pm 0.5$ PgC
  • BLUE: 1.2 Pgc/yr
  • BLUE: C uptake following abandonmends compensates (new: only) for half of emissions from clearing and wood harvest
  • generally high estimates:
    • no neglection of subgrid-scale gross transitions or wood harvests
      • if other studies with lower estimates included those, they might come up with even higher estimates than BLUE
    • carbon densities relatively high on chosen data set
    • unconditional scaling down of PFTs to simulate PFT mixtures within grids; some studies use conditions which lead to reduced emissions
  • gross LULCC transitions (on sub-grid scale) effect more land than onlt net transitions
  • main effect of gross transitions is not lower C densities but more primary land is transformed to secondary land

4. Relevance of Methodological Choices for LULCC Carbon Flux Accounting

  • four methodological choices:
    • simulating the temporal evolution of C stocks
    • initial state of C stocks
    • temporal attribution of C stocks
    • accounting for LULCC prior to accounting period
  • vast differences in temporal evolution of net LULCC flux and cumulative emissions on five-year accounting period (2008-2012)
    • from 4.3 PgC upttake to 15.2 PgC release

4.1 Temporal Evolution of Carbon Stocks

  • compare methods #5 and #6
    • #5: successive LULCC events influence each other’s C stock response by redistributing C flux potentials
    • #6: successive LULCC events are independent of each other
    • start with same stock and end with same net cumulative emissions
    • paths, however, are very different in early and late stages

4.2 Initial State of Carbon Stocks

  • transient results in vegetation much lower (at beginning of accounting period) than equlibrium values, in soil comparable
  • lower stocks lead to lower emissions from clearing

4.3 Temporal Attribution of Emissions

  • commited-flux approach much better constrained by actual observational data
  • more realistic temporal response needs more parameters and simulations
  • different intended uses: committed fluxes intend to compare overall impact of different LULCC with each other and with other human activities such as burning fossil fuels
  • asymmetry in timescales of regrowth and decomposition, regrowth much slower
  • committed fluxes much more inaccurate with respect to uptake (too optimistic): lower net emissions
  • cumulative emissions of #6 and #7 are similar by coincidence

4.4 Accounting for Fluxes from LULCC Events Prior to the Accounting Period

  • political considerations (UNFCCC) often require attribution to LULCC events, hence ignorance of what happened earlier
  • global carbon budget attributes to actual year of emission, hence needs to include effects of earlier events
  • since uptake takes more time than emission, more uptake plays into the accounting time when earlier effects are included
  • instantaneous approaches are highly sensitive to the considered time frame

5. Summary and Conclusion

  • computationall efficient estimation of LULCC fluxes
  • accounts for gross transitions (important for sub-grid scale LULCC), includes wood harvest, has temporal accounting layer
  • resulting net emissions at high end of previos studies, but within published range
  • comparing 8 methods based on 4 methodological choices, first time within a consistent framework
  • Kyoto protocol accounting period of 5 years
  • net LULCC flux highly sensitive to methodological choices (-4 to +15 PgC in 5 years)
  • substantially different temporal evolution of fluxes
  • which methods to choose depends on a study’s purpose

Questions

  • What exactly is a committed flux?
    • Even actually delayed fluxes are committed to the moment of the LULCC event taking place, for the sake of simplicity.
  • Do we have $4\times4\times8\times11$ (cover type $\times$ transition type $\times$ pool type $\times$ plant functional type) pools for each grid cell?
  • How exactly are the equilibria defined? Are they defined differently in a legacy scheme from a non-legacy scheme or do they depend solely on cover type, pool type, and plant functional type?
  • In Fig. 2, why do the yellow dots not coincidw with the end of the solid yellow curves? Or do they in Fig. 2b) but not in 2c) and 2d)?
  • What is a shifted boxcar filter?
  • What is product carbon footprint standards?
  • Can we just run the default #1 and compute the different accountings (to year of LULCC directly or to actual year of emission) from there in a post-processing step? To me it seems, some methodological questionable decisions impact not only the accounting directly but also very much indirectly by altering the future of the simulation in a nonsensical way.

My ideas

  • if commitment periods include or exclude C from LULCC events from some period, we compute this by taking C transit times into accout
  • However, the current generation of dynamic global vegetation models does not allow for attributing the resulting carbon fluxes to an individual LULCC event because of computational constraints.
    • We did something similar in the PNAS paper with fossil fuels by extracting a “fossil-fuel only” system based on the state transition operator of the entire system (the linearized trajectory). Could this be applied?
    • Maybe prepare a short presentation on the state transition operator and its several shapes in one dimension, in autonomous systems, and so on. And then throw it to the fossil fuels and ask if this can be done with LULCC fluxes too.
    • Practically, we might not be able to reproduce the state transition operator because of a lack of precise mathematical compartmental model formulation. This was the exact reason for me to build BFCPM.
    • But then Carlos’ idea of reproducing a “best possible” state transition operator from known stocks and fluxes comes into the game!