The carbon costs of global wood harvests – Nature

CHARM basic structure

CHARM is a biophysical model developed for this paper and related work, which estimates the GHG consequences and landuse requirements to meet wood consumption levels. The principal version of the model runs in Python using input files from Excel. CHARM has components that include both stand level and global analysis (Extended Data Fig. 5).

Unlike other commonly used carbon ‘book-keeping’ models, which typically start with total wood harvest levels and therefore can be used only retroactively, CHARM uses estimates of four major wood product categories of consumption by country to estimate harvest levels. These wood product categories are: LLPs, which are essentially wood for construction and furniture; SLPs, which are paper and paperboard products; and VSLPs, comprising wood used immediately for bioenergy (VSLP–WFL); and very-short-lived products–industrial (VSLP–IND), which are wood wastes from the generation of other wood products that are burned for energy.

The model starts with existing wood sources and demands as of the year 2010. Demands for different wood products are aggregated into total wood demands by country. When estimating future production, the model assumes that existing global trade patterns remain the same. For example, if timber-importing countries increase their demand, the model assumes that imports will grow proportionately and that exporting countries will proportionately increase their exports to meet this increasing demand.

The model separates wood supplied by existing plantation forests and that supplied by secondary forests, each based on their harvest efficiencies and growth rates. Plantation forests are those we know are dedicated to wood production. Secondary forests, by definition, are forests that have been harvested and, given our rules on forest age for harvesting, are therefore more probably those involved in wood production. At the national and global level, the model uses information about each country’s forests and assumes that wood demand will first be met by plantations to the extent available in 2010 and that secondary forests will be harvested for the remainder. The model tracks the carbon consequences of harvesting these forests under allocation and regrowth management rules specified by the scenario.

Land requirements are defined as the area of plantation and of secondary forests harvested over a given period of focus, which is between 2010 and 2050 in this paper. The present version of the model uses an optimistic assumption that all forests harvested will be from secondary rather than from primary forests, which are typically more carbon-dense.

To estimate land-use requirements, the model assumes that all harvesting is achieved through at least small clear-cuts. (The model also allows for thinning of forests, but that is done on the same lands as those ultimately harvested and therefore does not increase harvest area counted.) The clear-cut assumption increases wood harvest per hectare and therefore reduces the area affected by harvest. In the tropics, although most non-plantation forest harvests occur selectively, there are problems of definition between selective harvests and miniature clear-cuts, as well as uncertainties about the quantities of wood removed by different logging techniques. These uncertainties make it challenging to provide a precise estimate of area affected. The area of land use calculated by CHARM should therefore be viewed as hectares of clear-cut equivalent (that is, the hectares that must be harvested assuming all hectares affected are clear-cut). However, estimated harvest efficiencies—that is, calculations of waste—are regionally based and therefore incorporate estimated losses from selective harvest where that is the predominant method, as in the tropics. The estimates are therefore clear-cut equivalents assuming harvest efficiencies at present levels.

These ratios between consumption and harvests by product category are then multiplied by the quantity of projected consumption for each year between 2010 and 2050 in each country, for each wood product category, to estimate harvest levels by country (factoring in trade). Conversely, the model allocates wood harvests within a country to different wood products based on estimates of their different product consumption levels.

Because of the questionable data quality of countries producing small quantities of wood, our global analysis estimated wood harvest in those 30 countries that produce 80% of the world’s wood and then divided that volume by 80% to generate a global estimate.

Carbon costs and storage pools

To estimate GHG effects, CHARM globally applies an approach established for stand level analysis in the 1990s³⁴ by tracking the flow of carbon between different carbon pools over time due to harvests. Any reduction in carbon stored in the aggregate of all pools from one year to the next means an emission by that quantity of that carbon to the atmosphere whereas any increase means a removal. The pools include live wood (including forest regrowth after harvest), forest residues, roots, wood in the different product categories and wood in landfills. For each hectare of forest in each year, the carbon cost is the difference between (1) the amount they would store without future harvests (non-harvest scenario) and (2) the quantity of carbon that forests and wood products would hold with future harvesting and planting (harvest scenario). A positive value means emissions and a negative value means carbon removal. This calculation therefore factors in both ongoing forest growth in a harvest if not harvested and forest regrowth after harvest.

The model assumes that harvested forests will be allowed to regrow. Even so, the model can differentiate between regrowth as a secondary forest or as a plantation. We note that an economic or behavioural model might seek to estimate the changed probability of growth or regrowth and could be valuable if reliable. However, in addition to the challenges of making these estimates (and their off-site consequences as well), this approach assumes that, if regrowth is stopped by another human activity, the emissions in the form of foregone sequestration should be assigned to that other activity and not to the harvest.

For the live vegetation pool, because clear-cuts are assumed, the pool is eliminated in the first year of harvest. However, this pool regrows over time according to growth rates specified for that forest type in each country. The live vegetation pool consists of above- and below-ground biomass pools. Below-ground biomass is estimated using a widely used power function for relating root to shoot biomass^49,50. The model factors in dead wood remaining in the forest as a result of a harvest both in slash and roots, but it does not factor in changes in other downed dead wood. Forests typically have a layer of downed, dead wood not caused by harvest but due to dead trees and fallen branches. Although this pool may change over time, data are lacking—particularly across multiple forest types—of the changes in this forest pool as a result of harvest. In other words, the literature does not document whether forest harvests tend to expedite removal or degradation of already downed dead wood and, if so, how rapidly any such pool of carbon recovers with regrowth. (Estimates of dead wood stocks in the forest do not typically distinguish between those caused by a harvest, which CHARM does estimate, and those not caused by a harvest.) CHARM therefore assumes that this source of downed, dead wood is unaffected by harvest, is the same in both harvest and non-harvest scenarios and therefore does not need to be counted to determine the effects of wood harvest.

The model assumes that all VSLPs are burned and counted as an immediate emission, all SLPs are burned after use and that LLPs go to landfills as they decay. Meanwhile, the landfill pool can be interpreted as temporary storage because the carbon in wood products is not immediately released into the atmosphere. However, some percentage of the carbon emitted from the landfill is converted to methane, which has a much higher global warming potential and is counted as carbon dioxide equivalents based on its global warming potential over 100 years.

Extended Data Fig. 6 shows the changes in carbon storage for loblolly pine plantations in the Southeastern United States. In the first year of harvest there is a net increase in carbon emissions (represented by the vertical difference between the dotted green line and solid black line). In the second year there are further emissions as some of the felled wood decays or is burned, which can be seen by an expanding distance between the two lines. In the later years of each harvest cycle, due to more rapid forest regrowth the black and dotted green lines converge, representing net removals of carbon from the air.

Wood consumption, harvesting and trade data

Model development needed extensive effort to estimate the quantities of wood harvests required to meet each unit of wood production consumption by wood product category in a manner consistent with FAO consumption, production and trade data. Estimation of relationships between production and consumption data presents challenges because FAOSTAT gathers and reports wood product consumption and harvests in different types of units (such as weight versus volume and in products that have different, although unstated, water contents and therefore shares of dry matter). FAOSTAT also reports intermediate wood products between harvests and final consumption, the production of which generates significant wastes, some of which are then used for other products whereas others are typically burned. We used information from a variety of sources including unit conversion estimates, FAOSTAT estimates of standard waste levels in pulpwood production and estimates of sawn wood wastes implied by production data and which therefore varied by country. Trade data were also of inconsistent quality and, for some countries, implied physically impossible or highly unlikely consumption:production ratios. We developed rules to address data inconsistencies and data quality. The estimation methods used are described in further detail in the Supplementary Information.

Biophysical model inputs

The model uses a variety of biophysical inputs. One is secondary forest growth rate over time, which has consequences both for the forest if not harvested and for regrowth after harvest. There are many uncertainties about growth rates, and changes in growth rates over time, over large forest areas. Even different forest types in the same compact area can have highly varying growth rates and patterns⁵¹, and efforts to identify even dominant forest types spatially tend to have high error rates⁵². Our global model derives growth rates for secondary forests from Harris et al.²⁵, which uses a variety of sources discussed in that paper and its supplements. We supplemented this information with further data on the relationship between young and middle-aged secondary forest growth rates⁴⁹. Similar to default guidance from the Intergovernmental Panel on Climate Change for national GHG reporting, Harris et al.²⁵ estimate growth rates in broad time bands, one forest growth rate for younger forests of less than 20 years and another for more than 20 years. Because changes in time matter more to our model than to the estimates in that paper, we used these time bands to derive continuous growth rates using a Monod function found to be a reasonable proxy for forest growth rates in general^53,54.

For plantations we first applied the growth rates from Harris et al.²⁵ to the countries available in the Spatial Database of Planted Trees (SDPT v.1.0) and then supplemented the boreal and some EU countries, such as Canada and Russia, with average secondary forest growth rates. We also compiled data from a variety of literature and national reports for important timber producers such as Brazil, China, Indonesia and the United States, which are described in Supplementary Information.

Other inputs to the model for each forest type include root:shoot ratios, the portion of above-ground biomass left behind after harvest (slash rates), the proportion of above-ground biomass removed during thinnings and the rotation period. The proportion of carbon in harvested wood allocated to each product pool derives from the estimated consumption share of that product in that country. The model also requires decay rates for each carbon pool and inputs for allocation of that carbon to different subsequent pools (for example, landfills).

All input values and their sources, and further details regarding the Monod functions, are described in Supplementary Information Section 3. Supplementary Table 5 provides weighted average national forest growth parameters in the 30 countries used for this analysis. Supplementary Table 6 supplies the plantation rotation periods used and information sources. Supplementary Table 7 gives secondary and plantation slash rates by country. Supplementary Table 8 describes the half-lives used for the ‘decay’ of carbon in different carbon pools. The uncertainties of secondary forest growth rates and root:shoot ratio are discussed in Supplementary Information Section 5.

Production emissions and substitution values

The generation of wood products also releases fossil emissions and potentially trace gases in planting, harvesting and the production process. Because there are numerous data uncertainties on a global basis about how much fossil energy is used in harvesting wood and producing wood products, CHARM does not at this time incorporate these emissions.

Although comparisons between emissions from the use of wood products and alternative non-wood products do not reduce the absolute emissions from use of wood products, there is keen interest in whether wood use has lower emissions than alternatives. A full calculation of this requires calculation of the effects on biogenic carbon as well as production emissions. Even so, and because CHARM separately calculates biogenic emissions, CHARM is now programmed to estimate potential ‘substitution’ savings in production emissions when using wood to replace concrete and steel in construction. Estimates vary substantially owing to the different quantities of each material required for different buildings and different construction methods. Our calculation uses a central value from a review of other studies⁴⁰ of 1.2 tonnes carbon saved from production processes for each 1 tonne of carbon in wood used in construction that substitutes for concrete and steel. The benefit also depends on the share of harvested wood used in construction. As described in the Supplementary Information, we used estimates by country from Zhang et al.⁵⁵.

CHARM also estimates substitution benefits from the use of traditional firewood and charcoal in place of fossil fuels. Assuming that the alternative would be the use of propane gas, we use a substitution factor of 0.175 tonnes of carbon saved from avoided fossil fuel use for each 1 tonne of carbon from wood. This is based on estimates of relative energy output, charcoal and firewood production efficiencies and stove output and use efficiencies provided by the lead author of ref. ³¹.

Factoring time into carbon calculations

In addition to estimation of physical changes in emissions and removals of GHGs to the atmosphere over time due to each year’s wood harvests, the model estimates the value of these changes in the year of harvest using different discount rates. When the model uses a zero discount rate it estimates the physical change in atmospheric carbon after the period analysed, which can be 40 or 100 years after each harvest. In effect, a zero discount rate assumes that the change in atmospheric carbon at the end of the period is of equal value to this same change in carbon if it occurred in the year of harvest.

Discounting assigns a higher value to earlier emissions reductions. The model expresses carbon emissions as a value but based on an equivalency to emissions that occur only in the year of harvest—that is, harvest-year equivalent emissions. This form of valuation establishes a relationship between the value of emissions or mitigation in different years but does not need to specify an absolute dollar value for each tonne of carbon, which could be separately debated and determined.

The choice of a discount rate is a policy decision, which can represent two benefits of earlier mitigation. One benefit is to recognize the value of immediate reductions to avoid both intermediate and permanent damages from rising temperature (for example, the effects of ice sheet melting or biodiversity loss) and to postpone the date of crossing a variety of climate thresholds. Earlier mitigation in effect holds down damages immediately and increases the time in which people can improve technology and organize the political will and resources to combat climate change before crossing thresholds.

The other benefit of earlier mitigation results from the time value of money. Applying a 4% discount rate in effect assigns a 4% rental charge each year to additional carbon in the atmosphere. That equals the price of borrowing money at a commonly estimated long-term cost of capital to pay another person to mitigate emissions to compensate. As discussed in ref. ³³ in the context of land-use conversion, this discount rate also generates results consistent with the amortization period used for land conversion in US bioenergy policy.

The value discounted in each year after each harvest back to the year of harvest is the change in atmospheric carbon that year—that is, the difference between emission (or removal) in that year and that in previous year. This formula in the year of harvest h (for example, 2010) is:

where t is the number of years since harvest in year h, d is the discount rate (4%), N is the number of years for growth since harvest in the scenario (for example, 40 or 100 years) and C_change,t is the change in emissions (or removals) in the year t. Extended Data Table 2 shows the calculation of time discounting of 4% over 40 years for a plantation conversion scenario shown in Extended Data Fig. 6.

Present discount value is calculated identically for each subsequent year of harvest. The cumulative PDV of emissions between 2010 and 2050 is the sum of these carbon costs over 40 years, and they therefore do not represent the carbon added to the atmosphere in 2050 by forest harvests between 2010 and 2050. That alternative quantity of carbon would be larger because it would not factor in the full 40-year regrowth of forests harvested after 2010. However, this method in effect assigns a discounted value for projected forest regrowth regardless of which year the harvest occurs—for example, even in the year 2049.

For national and global results we then determine the total discounted carbon cost in year t by multiplying the PDVs of each hectare by the number of hectares harvested of that same forest type in the year harvested. This is done separately for both plantations and secondary forests, producing the formula:

where h represents the year of harvest starting from 2010, K represents the number of years of harvests (for example, 40 years) and a represents the new area of one forest type harvested in year h. The next subsection describes the calculation of area required for each forest type.

Land area calculation

Due to the unknown levels and quantities of wood removed from selective harvests, CHARM calculates the area of land use as hectares of clear-cut equivalent (that is, the hectares that must be harvested assuming all hectares affected are clear-cut). This assumption increases wood harvest per hectare relative to selective harvests and therefore reduces the estimate of area affected by harvest. This method is used because of inadequate data available for the quantity of wood harvested through clear-cuts and that through selective harvest. This calculation of land requirements reflects the quantity of wood generated per hectare at the estimated efficiencies by country. The quantity of wood required is also based on the ratio of each category of wood production consumption to harvest levels required to generate that level of consumption—that is, it factors in wastes. For the period 2010–2050 the model assumes linear growth in consumption for each product category from 2010 to 2050. Plantation areas are harvested first, and secondary forests are harvested as needed to supply the remaining quantities of wood.

Projection of 2050 wood demand

To project future wood demand, CHARM starts with 2010 consumption levels (calculated as an average of 2006–2014 consumption) by country for consumption and production of different wood products and harvest levels, using data from FAOSTAT⁵⁶ after a system of quality controls (Supplementary Table 1). For each country and year, we first calculated net exports by subtracting imports from exports. Future projections assume that, within each country the share of consumption supplied by net imports will remain the same as in the base year and that each country will provide the same share of aggregate global exports.

To estimate future wood product demand by country, we use a log-transformed fixed-effects model⁵⁷ and project wood demand for each country and each product category. The fixed-effects model applies the same relationship of wood consumption to each country’s per capita income growth but starts with each country’s initial wood consumption. We separated countries into developed and developing to avoid overestimation of future wood consumption in high-income countries. For wood products consumption—and based on available data—we selected sawn wood and wood-based panels to represent all LLP, paper and paperboard to represent SLP and wood fuel to represent VSLP-WFL. The historical socioeconomic statistics include GDP and population from the World Bank for 1961–2020⁵⁸. We used projected growth percentages between 2010 and 2050 for GDP per capita and population based on from average GDP growth prediction from three sources: the Organisation for Economic Co-operation and Development (‘middle of the road’)⁵⁹, the International Institute for Applied Systems Analysis model SSP2 scenario⁶⁰ and a linear trend line that we calculated for the period 1991–2010. Predictor (independent) variables in the fixed-effects model include population, GDP per capita and the year after 2000, which serves as a proxy for technological and policy changes since 2000, when the internet usage boom started and modified subsequent paper requirements. The fixed-effect model establishes 12 relationships (‘models’) based on three different types of wood product, two different trend lines in developed and developing countries and two different regression formulae, one using our time variable and one without. All models have high R² full (over 0.88) and significant P values (over 0.05) and have a residual standard error between 0.32 and 0.84 (Supplementary Table 2). We apply the coefficients (Supplementary Table 3) of predictor variables to independently estimated changes in future populations and GDP by country, and use the resulting estimated consumption levels to force the model. The Supplementary Information provides statistics for model fits and further information about the fixed-effects model and application. Extended Data Table 3 shows the consumption of different wood products by country in 2010 and as projected for 2050. Supplementary Table 4 shows a comparison of our projections with those of other studies.