Main

Reforestation, that is the restoration of tree cover to areas where forests naturally occur but have been removed by human or natural disturbance, contributes to climate change mitigation by removing carbon dioxide from the atmosphere and storing it in biomass and soil. Reforestation has been identified as a mitigation option with high potential, although with much of that potential at a high cost-per-ton of carbon dioxide (US$ per tCO2; ‘abatement cost’)1,2. Yet the cost-per-ton of reforestation varies by reforestation method and location.

Two of the most common reforestation methods are passive natural regeneration of native tree species from seeds produced by existing trees or stored in the soil (‘natural regeneration’3,4,5) and planted monocultures of native or introduced tree species intended to supply wood products (‘plantations’4,6,7,8). Passive natural regeneration involves minimal human intervention aside from protecting sites from factors that inhibit natural regrowth, in contrast to assisted natural regeneration9,10,11. Plantations generally produce more wood products12, while naturally regenerated forests generally provide greater biodiversity12,13, water provisioning and erosion control12.

Whether natural regeneration or plantations can remove carbon dioxide from the atmosphere at a lower cost-per-ton is a deceptively complicated calculation (equations (1) and (2)). Although the opportunity cost of reforestation is the same for both methods at a given site, other components of abatement cost vary. In the numerator, plantations have higher implementation costs than natural regeneration14 but these higher costs need to be weighed against the net revenue from the production and sale of wood products12,15. In the denominator, plantations can sometimes, but not always, have faster carbon accumulation rates than naturally regenerating forests16. Furthermore, plantations are periodically harvested, at which time some portion of carbon is released to the atmosphere and the remainder is stored in wood products, which release carbon over time17. Each of these factors vary across space, meaning that the more cost-effective reforestation method at a site, that is the method with lower abatement cost, varies spatially as well (equation (3)).

Until now, studies comparing the relative cost-effectiveness of natural regeneration and plantations at mitigating climate change5,18,19 have only examined subnational regions. Previous studies did not use spatially explicit data on carbon or costs and did not account for carbon storage in and revenue from harvested wood products. Recent advances in modelling carbon accumulation in naturally regenerating forests20 and plantations21, along with advances undertaken for this article in mapping reforestation costs and likely plantation genus, now make it possible to produce large-scale, spatially explicit maps of the relative cost-effectiveness of natural regeneration and plantations to potentially mitigate climate change.

Here, we estimate and map abatement cost (US$ per tCO2) for natural regeneration and plantations across 138 low- and middle-income countries (listed in Extended Data Tables 1–3) at 1 km resolution. To do so, we use machine learning and regression models to construct six spatial datasets, which we then combine (Extended Data Fig. 1). These datasets include: implementation costs of (1) passive natural regeneration and (2) plantation reforestation, based on 355 observations from projects in low- and middle-income countries; (3) opportunity costs of reforestation, based on household survey data from 229,859 households in 32 low- and middle-income countries22 and crop and livestock production maps23,24,25; (4) the most likely of 15 plantation genera, based on the genus and spatial locations of 39,514 planted forest sites26; (5) carbon accumulation over time from natural regeneration, based on 2,309 observations from 410 sites20; and (6) carbon accumulation over time from plantations, based on 3,289 observations from 618 sites21, accounting for carbon that is harvested and stored in wood products.

The maps of cost-effectiveness that we produce can help guide reforestation initiatives such as the United Nations Decade on Restoration27, the Bonn Challenge28 and Trillion Trees29 and aid efforts of countries to achieve nationally determined contributions to the UNFCCC4. They can be combined with other information, for example existing land use30,31,32, landholder type33, biodiversity cobenefits34,35, economic and social constraints36 and equity considerations37, to help design reforestation programmes that best meet local and regional needs.

In addition, by aggregating this spatial information with geographic constraints on reforestation, we produce marginal abatement cost curves (‘cost curves’), that is, estimates of how many tons of carbon dioxide could be removed by reforestation below any cost-per-ton (Extended Data Fig. 1). These cost curves can be interpreted as the minimum carbon price incentive needed from a carbon market or payment programme to induce land users to reforest a given amount. They can be used to estimate sequestration outcomes for a given level of investment38, to compare the cost-effectiveness of reforestation to other climate mitigation options1,2 and as an input to integrated assessment models of the cost of climate change39. Relative to previous cost curves40,41,42, incorporating costs incurred by hundreds of actual reforestation projects in practice enables more accurate estimates of reforestation cost.

We construct cost curves for both natural regeneration and plantations, as well as for the more cost-effective method within each 1 km grid cell. We constrain potential reforestation to the 10.1% of land area that is considered potentially suitable for terrestrial reforestation, screening out mangroves, areas where reforestation cannot occur (for example, existing forests, deserts and high-elevation biomes) and areas where previous researchers43 have argued reforestation should not occur (for example, native grasslands and croplands). We further constrain plantation reforestation to areas near existing plantations (and hence, near markets for harvested wood); we constrain natural regeneration to areas near existing natural forests (and hence, near seed sources). Finally, we subtract the fraction of reforestation that was unlikely to be additional, that is, is likely to occur under business-as-usual.

Results

We first present results from the intermediate spatial data layers that we developed. Unless otherwise specified, results reported here were calculated across the area considered suitable for reforestation in 138 low- and middle-income countries (within the geographic screen of the reforestation opportunity map in ref. 43), as well as feasible for natural regeneration (within 5 km of a natural forest) or plantation reforestation (within 40 km of an existing plantation). Costs and carbon accumulation were summed over a period of 30 years and time-discounted at a rate of 5% annually to account for greater value placed on near-term costs and benefits. All costs are in 2020 US$ (for rationale underlying our scope and parameter assumptions see Methods).

We modelled median implementation cost to be US$140 ha−1 for natural regeneration (interdecile range (IR), US$47–183 ha−1) (Fig. 1a and Extended Data Fig. 2a) and US$3,729 ha−1 for plantations using the most likely tree genus, inclusive of replanting after harvest cycles (IR, US$1,530–11,675 ha−1) (Fig. 1b and Extended Data Fig. 2b). Opportunity costs were generally higher than implementation costs and more variable (median, US$4,807 ha−1; IR, US$516–13,980 ha−1) (Fig. 1c and Extended Data Fig. 2c), including US$4,770 ha−1 (IR, US$462–13,947 ha−1) in areas where natural regeneration is feasible and US$5,054 ha−1 (IR, US$695–14,376 ha−1) in areas where plantation is feasible. We modelled the most likely plantation genus across areas where plantations are feasible to be Eucalyptus (45.5% of plantation-feasible area) followed by Cunninghamia (14.5%), Populus (11.9%), Acacia (11.0%), Pinus (9.4%), Tectona (4.7%) and Shorea (1.7%) (Fig. 1d).

Fig. 1: Costs, likely plantation genus and carbon accumulation from reforestation.
Fig. 1: Costs, likely plantation genus and carbon accumulation from reforestation.The alternative text for this image may have been generated using AI.
Full size image

a, Implementation cost of natural forest regeneration (US$ ha−1). b, Implementation cost of plantation, including replanting (US$ ha−1). c, Opportunity cost of reforestation (US$ ha−1). d, Most likely plantation genus. e, Carbon accumulation in natural forest regeneration (tC ha−1). f, Carbon accumulation in plantations plus harvested wood products (tC ha−1). Data are for: 138 low- and middle-income countries; 30 years; time-discounted at 5%; 2020 US$.

Our spatially and temporally explicit models of carbon accumulation suggest that the median present value of carbon accumulation over 30 years from natural regeneration is 60.2 tC ha−1 (IR, 26.4–77.7 tC ha−1) in aboveground biomass, belowground biomass and soil (Fig. 1e and Extended Data Fig. 2d). Of this, 36.5 tC ha−1 (IR, 10.3–50.4 tC ha−1) is in aboveground biomass and none is stored in wood products. In contrast, the median average carbon accumulation from plantation reforestation with cyclical harvest using the most likely plantation genus is only 44.6 tC ha−1 (IR, 3.7–115.8 tC ha−1) in aboveground biomass, belowground biomass, soil and harvested wood products (Fig. 1f). Of this, 15.2 tC ha−1 (IR, 1.1–35.7 tC ha−1; Extended Data Fig. 2e) is in aboveground biomass, while 23.5 tC ha−1 is stored in wood products (IR, 0.7–71.2 tC ha−1; Extended Data Fig. 2f), accounting for likely product use and duration of sequestration.

Cost-effectiveness

After combining the intermediate spatial layers related to implementation and opportunity cost, most likely plantation genus and carbon accumulation, we find the median abatement cost of climate mitigation through natural regeneration to be US$23.80 per tCO2 (IR, US$3.60–79.70 per tCO2) within areas feasible for natural regeneration (Fig. 2a and Extended Data Fig. 3a). The abatement cost of climate mitigation through plantation, accounting for revenue from the sale of harvested wood products, in areas feasible for plantation, was similar (median, US$23.00 per tCO2; IR, US$−14.80–724 per tCO2) (Fig. 2b and Extended Data Fig. 3b). Plantations would be negative cost, that is profitable even without carbon payments, across 28.3% of that area. Selecting the more cost-effective reforestation method in each location that is feasible for one or both reforestation methods would result in lower abatement cost (median, US$12.50 per tCO2; IR, US$−15.50–66.30 per tCO2).

Fig. 2: Cost-effectiveness of reforestation.
Fig. 2: Cost-effectiveness of reforestation.The alternative text for this image may have been generated using AI.
Full size image

a, Abatement cost of natural regeneration within areas where natural regeneration is feasible (US$ per tCO2). b, Abatement cost of plantation within areas where plantation is feasible (US$ per tCO2). c, Relative abatement cost of plantation to natural regeneration within areas suitable for reforestation (higher value, plantation is more cost-effective; lower value, natural regeneration is more cost-effective). Data are for: 138 low- and middle-income countries; 30 years; time-discounted at 5%; 2020 US$.

We find natural regeneration to be more cost-effective at mitigating climate change than plantation reforestation across 46% of the 593 Mha considered suitable for reforestation (Fig. 2c and Extended Data Fig. 3c). This includes 43% of the 468 Mha in which both natural regeneration and plantation reforestation are feasible (natural regeneration is feasible across 535 Mha, while plantation is feasible across 526 Mha). Most countries contained regions where each reforestation method was more cost-effective (Fig. 3 and Extended Data Tables 1–3).

Fig. 3: Abatement potential by applying the more cost-effective reforestation method at each site, below US$50 per tCO2, by country (MtCO2).
Fig. 3: Abatement potential by applying the more cost-effective reforestation method at each site, below US$50 per tCO2, by country (MtCO2).The alternative text for this image may have been generated using AI.
Full size image

Data are for: 138 low- and middle-income countries; 30 years; time-discounted at 5%; 2020 US$. Americas, orange; Africa, blue; Asia, purple. log–log scale. Country codes from https://www.iso.org/obp/ui/.

Our models suggest that natural regeneration is especially cost-effective relative to plantation forestry in much of Western Mexico, the Andean region, Southern Cone of South America, West and Central Africa, India, Southern China, Malaysia and Indonesia. Conversely, plantations are especially cost-effective relative to natural regeneration in much of the Caribbean, Central America, Brazil, North, East and Southern Africa, northern China, mainland Southeast Asia and the Philippines. Which method of reforestation is more cost-effective in a given location is a function of several factors, including relative carbon accumulation and harvest rates (for example, higher for plantations in mainland Southeast Asia), relative implementation costs (for example, lower for natural regeneration in West and Central Africa), higher opportunity costs making the difference in revenue from wood products a smaller consideration (for example, in Southern China and Southeast Brazil) and distribution of feasibility (for example, more land where natural regeneration is feasible in Central Africa).

Marginal abatement cost curves

We estimate that natural regeneration could remove up to 21.8 GtCO2 below a cost of US$50 per tCO2 (12.0–25.1 GtCO2 below US$20–100 per tCO2) (Fig. 4). Similarly, plantations could remove up to 22.6 GtCO2 below the same cost (19.3–24.3 GtCO2 below US$20–100 per tCO2), after accounting for harvested carbon stored in wood products and profit from the sale of wood products. Reforestation using the more cost-effective method at each site could remove considerably more than applying either method everywhere: 31.4 GtCO2 below US$50 per tCO2 (24.2–34.3 GtCO2 below US$20–100 per tCO2). Latin America holds 56% of potential abatement from reforestation using the more cost-effective option below US$50 per tCO2, followed by Asia (33%) and Africa (11%) (Extended Data Fig. 4). The top ten countries for potential cost-effective abatement are Brazil (32.8%), China (9.4%), Mexico (5.8%), India (4.5%), Myanmar (3.9%), Colombia (3.7%), Indonesia (3.4%), Thailand (3.2%), Democratic Republic of the Congo (2.5%) and Venezuela (2.2%) (Extended Data Fig. 5 and Extended Data Tables 1–3).

Fig. 4: Marginal abatement cost curves for reforestation within suitable areas.
Fig. 4: Marginal abatement cost curves for reforestation within suitable areas.The alternative text for this image may have been generated using AI.
Full size image

Data are for: 138 low- and middle-income countries; 30 years; time-discounted at 5%; 2020 US$.

There is potential for reforestation below US$50 per tCO2 across 497 Mha (356–562 Mha below US$20–100 per tCO2). Of this area, natural regeneration would be more cost-effective across 208 Mha, while plantation would be more cost-effective across 288 Mha. In a hypothetical case in which all this land was reforested using the more cost-effective method, the total gross cost would be US$4.00 trillion, of which US$1.67 trillion would be implementation cost while US$2.33 trillion would be opportunity cost. This cost would be offset by US$3.29 trillion revenue from the harvest and sale of wood products from plantations, resulting in a net cost of US$711 billion (−US$670–US$914 billion below US$20–100 per tCO2).

Sensitivity analyses

Our estimates of abatement were sensitive to a variety of modelling assumptions (Extended Data Table 4). At a given carbon price, abatement available from either natural regeneration or plantations would increase with a longer time horizon or a lower time-discount rate. Abatement available from plantations was sensitive to alternative planted species but only minimally sensitive to extended harvest cycles. Abatement was sensitive to applying a reforestation suitability screen other than the one from ref. 43: abatement was 26% lower across areas where tree canopy cover could be increased from below to above 25% (ref. 44); 45% lower across areas with high unrealized potential for restoration of woody-dominated systems45; 63% higher across areas where tropical forest restoration is feasible46.

Our estimates of the fraction of land area over which natural regeneration is more cost-effective than plantation were very sensitive (>15%) to discount rate, genus, harvest and sale of wood products, the exclusion of negative-cost plantation areas and the suitability screen of ref. 45. It was moderately sensitive (5–15%) to time horizon and the suitability screen of ref. 44. It was minimally sensitive (<5%) to rotation cycle, buffer and additionality assumptions and the suitability screen of ref. 46.

An advantage of naturally regenerating forests arises from our assumption that they are not recleared. In a scenario in which natural forests are periodically cleared following the same interval as plantations (but without sale of wood products), they would remove 74% less carbon dioxide below US$50 per tCO2, while if plantations were never cleared they would remove 70% more carbon dioxide below US$50 per tCO2. Plantations have an advantage that arises from producing revenue; in a scenario without net revenue from the sale of wood products, they would remove 21% less carbon dioxide below US$50 per tCO2.

Discussion

We find that passive natural forest regeneration (46%) and plantations (54%) would each be more cost-effective at mitigating climate change across about half of the land in low- and middle-income countries that is considered suitable for reforestation, accounting for carbon storage in, and revenue from, the sale of wood products. Median abatement cost is comparable across natural regeneration (US$23.80 per tCO2) and plantations (US$23.00 per tCO2). There is considerable spatial variation in abatement cost for both natural regeneration and plantations, as exemplified by large interdecile ranges. Thus, choosing the more cost-effective method at each site, or in some cases the only method that is feasible, results in an even lower median abatement cost (US$12.50 per tCO2) than only using one method everywhere.

Plantations have greater variation in abatement cost than natural regeneration. In some sites, plantations are negative cost, that is, they would be profitable even without a carbon price. Excluding such sites from consideration, for example because of concerns about additionality, would reduce the abatement potential from plantations by more than half (58%) (Extended Data Table 4). Our finding that natural regeneration can often be more cost-effective at mitigating climate change than plantations is consistent with previous region-specific studies from the Cape Province of South Africa18, the Atlantic Forest of Brazil5 and Peninsular Malaysia19, as well as future projections42. It suggests that the current heavier emphasis on plantations within national reforestation programmes4 could be leading to missed opportunities for cost-effective climate mitigation.

Natural regeneration could contribute about as much abatement (21.8 GtCO2) as plantations (22.6 GtCO2) below US$50 per tCO2. Using the more cost-effective reforestation method at each site could contribute considerably more abatement (31.4 GtCO2). Our model based on costs incurred by hundreds of actual reforestation projects finds 10.3 times as much cost-effective abatement from reforestation below US$20 per tCO2 and 2.8 times as much below US$50 per tCO2, compared to IPCC (2022)1 (Extended Data Table 5). Our estimates of cost-effective mitigation potential across low- and middle-income countries exceed the risk-adjusted estimates of ref. 40 for the entire globe (Extended Data Table 5).

Our analysis has caveats and limitations that may affect total abatement potential or relative cost-effectiveness. We only included on-site implementation and opportunity costs; we did not consider offsite costs. We did not consider transaction costs that would be specific to tenure resolution processes or carbon markets supporting reforestation. We also did not consider cost-increasing bottlenecks to scaling up the production of seeds, nurseries, labour and so on47,48 beyond screening out plantation farther than 40 km from existing plantations. By focusing only on the land-cover change from non-forests to forests, we are missing the potential for climate mitigation through tree growth within forests or outside forests. We did not model elasticity of demand for wood products, for example an expansion of plantations and their wood products reducing the returns to plantations elsewhere, nor constraints on future demand for wood products as in ref. 42. Modelling these effects would reduce the abatement potential from plantations.

There was potential for selection bias in the site-level data that we used to estimate both implementation cost and carbon accumulation. Sites selected for reforestation may have had lower costs or higher carbon accumulation, due to unobserved factors, or sites may have had greater reforestation success than a typical project elsewhere49,50,51. Such biases would result in underestimating the abatement cost from reforestation.

We only modelled the effects of reforestation on carbon accumulation—not on albedo52,53, surface temperature54, circulation55 or biogeochemistry56. We assumed that 30 year carbon accumulation rates can be approximated from observed growth rates over the past century, that is, we did not model the effects of climate change on future forest growth, vulnerability to fire and extreme weather or stand survival45,57,58. We did not consider changes in plantation species as a response to a carbon price or future climate change, nor did we consider the potential role of innovation, for example increasing carbon storage in wood products.

Importantly, finding that a method of reforestation is more cost-effective at a particular site does not imply that it is necessarily the superior method for that site. First, our spatial models will invariably be less accurate for specific sites than at broad regional scales. And second, people choose to reforest for many reasons other than climate mitigation. Site-specific decisions about whether, where and how to reforest should apply information on, for example, local needs and preferences, existing land use, biodiversity and sustainable development cobenefits12 and economic and social constraints37, which are beyond the scope of this study.

While exact estimates of abatement cost and relative cost-effectiveness are sensitive to model settings (Extended Data Table 4), we can nevertheless draw several robust conclusions from our research. First, neither reforestation method dominates for cost-effectiveness. A global mix of natural regeneration and plantations can achieve far more mitigation at lower costs than relying exclusively on a single method of reforestation. Second, reforestation is a complement to and not a substitute for reducing emissions from fossil fuels and deforestation59. Achieving the entire mitigation potential of reforestation of 31.4 GtCO2 over 30 years would amount to less than 8 months of global GHG emissions (51.2 GtCO2 yr−1 in about 2018; ref. 60) but would expand by 8.3% the remaining carbon budget of 380 GtCO2 (in about 2022) required for a 50% chance of meeting a 1.5 °C global temperature target61. Third, our findings indicate that reforestation offers far more potential abatement at low cost than previously estimated1.

Methods

To map the cost-effectiveness of reforestation at mitigating climate change, we constructed and then combined, six spatial data layers at 1 km resolution across 138 low- and middle-income countries (LMICs): implementation costs of natural regeneration (US$ ha−1); implementation costs of plantations (US$ ha−1); opportunity costs of reforestation (US$ ha−1); likely plantation genus; carbon accumulation through natural regeneration (tC ha−1); and carbon accumulation through plantations (tC ha−1). We further accounted for carbon that is harvested and stored in wood products (tC ha−1) and net revenue from the sale of wood products. Afterward, we aggregated information on cost-effectiveness to produce marginal abatement cost curves.

In general terms, we calculated the abatement cost of each reforestation method (US$ per tCO2), as well as the more cost-effective method at each site (US$ per tCO2), as follows:

$$\begin{array}{l}{\rm{Abatement}}\;{\rm{Cost}}\;{\rm{of}}\;{\rm{natural}}\;{\rm{regeneration}}=\\ \displaystyle\frac{\begin{array}{l}\left({\rm{Implementation}}\;{\rm{costs}}\;{\rm{of}}\;{\rm{natural}}\;{\rm{regeneration}}\right. \\ \quad + \left. {\rm{Opportunity}}\;{\rm{cost}}\;{\rm{of}}\;{\rm{reforestation}}\right)\end{array}}{3.67 \times {\rm{Carbon}}\;{\rm{accumulation}}\;{\rm{through}}\;{\rm{natural}}\;{\rm{regeneration}}}\end{array}$$
(1)
$$\begin{array}{l}{\rm{Abatement}}\;{\rm{cost}}\;{\rm{of}}\;{\rm{plantation}}= \\ \displaystyle\frac{\begin{array}{l} \left({\rm{Implementation}}\;{\rm{costs}}\;{\rm{of}}\;{\rm{plantation}}+{\rm{Opportunity}}\;{\rm{cost}}\;{\rm{of}}\;{\rm{reforestation}} \right.\\ \quad -\,\left.{\rm{Net}}\;{\rm{revenue}}\;{\rm{from}}\;{\rm{sale}}\;{\rm{of}}\;{\rm{wood}}\;{\rm{products}}\right)\end{array}}{\begin{array}{l}3.67 \left({\rm{Carbon}}\;{\rm{accumulation}}\;{\rm{through}}\;{\rm{plantation}}\right.\\\quad\left.+\,{\rm{Carbon}}\;{\rm{storage}}\;{\rm{in}}\;{\rm{wood}}\;{\rm{products}}\right)\end{array}}\end{array}$$
(2)
$$\begin{array}{l}{\rm{Abatement}}\;{\rm{cost}}\;{\rm{of}}\;{\rm{more}}\;{\rm{cost}}\;{\rm{effective}}\;{\rm{method}}\\ \quad =\min \left\{{\rm{abatement}}\;{\rm{cost}}\;{\rm{of}}\;{\rm{natural}}\;{\rm{regeneration}},\right.\\ \qquad \left.{\rm{abatement}}\;{\rm{cost}}\;{\rm{of}}\;{\rm{plantation}}\right\}\end{array}$$
(3)

This section provides an overview of the construction of the spatial data layers included in equations (1), (2) and (3). More detail on the construction of these layers is provided in the Supplementary Information.

Implementation costs of natural regeneration and plantation

We generated implementation cost (US$ ha−1) surfaces for passive natural regeneration and forest plantations based on 355 observations of reported reforestation costs from 99 source documents spanning 26 LMICs in Africa, Asia and the Pacific and Latin America and the Caribbean. We sourced the data from World Bank project documents and the peer-reviewed and grey literature. All source documents were publicly available. We used machine learning to determine the linear regression model that provided the best cost predictions and we used this model and spatial data layers to extrapolate implementation costs in 2020 US$ per hectare across LMICs.

Opportunity cost of reforestation

Opportunity cost (US$ ha−1) in our study refers to the value of land if it is not reforested, that is, its value in its continued current use. We assumed that this alternative use would be some form of agriculture, either cropland or pastureland. The opportunity cost of reforestation at a site is the same for both methods of reforestation because the value that land could produce in crops or livestock is unaffected by reforestation method. The opportunity cost of natural regeneration does not account for potential plantations and vice versa.

We defined the opportunity cost as land rent, that is, the portion of annual agricultural quasi-rent (essentially, short-run profit) that was attributable to land as a production input for raising crops or livestock. We generated a land rent surface across LMICs using a combination of gridded and household data. We used gridded data on gross crop revenue per hectare from the International Food Policy Research Institute MapSPAM dataset23,24 and gridded data on livestock numbers (buffaloes, cattle, goats, horses and sheep) from Food and Agriculture Organization (FAO) Gridded Livestock of the World (GLW)25. We used household survey data from the World Bank Living Standards Measurement Studies, compiled by FAO Rural Livelihoods Information System (RuLIS) programme22. The resolution of the opportunity cost surface was 5 arcmin (~10 km at the equator), following the resolution of the MapSPAM and GLW datasets. Opportunity costs were generated for reference year 2019 and later inflated to 2020 US$.

Our approach makes two improvements upon previous global opportunity cost maps, such as ref. 62 and its successors, which multiplied potential crop yields from agro-ecological zone maps by prevailing crop prices to estimate gross revenue. First, we calculated land rent instead of gross revenue, using data from household surveys to account for the share of gross revenue that is offset by production costs. Others63 also estimated land rents instead of just gross revenues but their estimates assumed a uniform 20% rent-share across countries and were based on data from only high-income countries (United States, Canada and some European Union countries), not LMICs. Second, we estimated actual revenue rather than simulating potential revenue, drawing upon the best available data: the rich spatial MapSPAM dataset of where crops are actually grown, their yields and their production values.

Our measure of actual revenue has several advantages over potential revenue. It is more likely to reflect land users’ true opportunity costs, particularly in places where institutional factors prevent farmers (for example, subsistence farmers) from planting crops that would access the highest market prices or within protected areas or areas where agricultural activities occur despite legal prohibitions. Relevant for our study, actual gross revenue is the more appropriate starting point for determining the opportunity cost of reforestation, where there is often an existing agricultural land use, while potential gross revenue is the more appropriate starting point for avoided deforestation, where an agricultural land use does not yet exist.

Likely plantation genus

We produced a map of the most likely plantation type to be planted across all LMICs at 1 km spatial resolution. To do so, we used data on the genus and spatial locations of planted forests across the globe from the Spatial Database of Planted Trees26. We then used machine learning to model the likelihood of each genus of tree crop to be planted in that location using a suite of environmental and socio-economic covariates hypothesized to influence genus selection. This was a positive analysis of which plantation genus is most likely, based on historical distribution; not a normative analysis of which plantation genus is most preferable, based on native occurrence or ecosystem services provided.

Carbon accumulation in aboveground biomass

To produce maps of carbon accumulation over time in naturally regenerating forests and plantations, we used a modified version of the sigmoidal Chapman–Richards growth function. Specifically, we modelled spatial variation in two parameters of the Chapman–Richards function, the asymptotic limit of aboveground carbon (A) and the rate of carbon accumulation (k), as a function of spatially explicit environmental covariates. We then used these statistical relationships to produce 1 km global maps of A and k for each forest type, which in turn described aboveground carbon stocks as a function of stand age. Changes in residual aboveground carbon from previous land uses were not considered in this model. We obtained the data for our growth functions from published datasets of aboveground carbon in naturally regenerating forests20 and plantation forests21. Full descriptions of the methods used to construct the datasets are contained in those studies but we review the methods briefly in the Supplementary Information.

Harvest cycle, carbon storage in wood products and profit from wood products

For plantations, we assumed periodic harvest for the sale of wood products. For every site i, we determined a site-specific and genus-specific optimal rotation length, \({h}_{i}^{P}\), based on the Chapman–Richards growth curves described above and the Faustmann condition for maximizing the net present value of repeated timber harvests64. This condition states that a forest should be harvested in the year before the growth rate g (= \(y{\prime} \left(t\right)/y\left(t\right),\) from the Chapman–Richards function) falls below \(\frac{\delta }{1-{(1+\delta )}^{-t}}\), where \(\delta\) is the time-discount rate and t is years. As a sensitivity analysis, we also determined the non-economic, traditional forester’s rotation length that maximizes average annual wood production (‘mean annual increment’) without regard for the discount rate64. In this case, harvest occurs in the year before growth rate g falls below 1/t. In another sensitivity analysis, we assumed plantations would continue to grow without being harvested.

Aboveground biomass in harvested plantations over time accumulates according to the Chapman–Richards function interrupted by periodic harvest to zero:

$${{\rm{AG}}}_{{it}}^{P}=A_{i}^{P}{\left(1-{\rm{e}}^{-{k}_{i}^{P}\mathrm{mod}(t,{h}_{i}^{P})}\right)}^{2}$$
(4)

Biomass carbon at the time of harvest is:

$${H}_{{it}}^{P}=A_{i}^{P}{\left(1-{\rm{e}}^{-{k}_{i}^{P}{h}_{i}^{P}}\right)}^{2}$$
(5)

For every cell, we calculated the fraction of biomass carbon at the time of harvest that would remain stored in wood products following timber harvest, \(\omega\), as follows:

$$\omega =\frac{1}{{\rm{BEF}}}\sum _{u\in U}\left({\alpha }_{u} \frac{\left(\mathop{\sum }\nolimits_{t=1}^{y}{2}^{-\frac{t}{{t}_{1/2}^{u}}} \left(1-\delta \right)^{t}\right)}{\mathop{\sum }\nolimits_{t=1}^{y}{(1-\delta )}^{t}}\right)$$
(6)

Here, BEF is the biomass expansion factor of 1.3 tC in aboveground biomass per tC in wood products, from Table 3A.1.10 of ref. 17. Parameter \({\alpha }_{u}\) is the country-specific fraction of biomass destined for each of three end uses, U (bioenergy, paper and pulp and solid wood; Table 4.6 of ref. 65). NPV is the net present value over y = 30 years of the decay rate of carbon in each of the three end uses, with a half-life, \({t}_{1/2}^{\,u}\), of 0 years for bioenergy, 2 years for paper and pulp products and 30 years for solid wood (Table 12.2 of ref. 17), applying a discount rate of \(\delta\) = 0.05.

We calculated profit from wood harvests as a scalar of biomass carbon at the time of harvest, \(\rho\), as follows:

$$\rho =\frac{1}{{\rm{BEF}}}\frac{1}{{\rm{CF}}}\sum _{u\in U}{\alpha }_{u} {p}_{u}$$
(7)

Here, CF is the country-specific carbon factor of tC m−3 for harvested wood. Adapting the methods of ref. 66 based on Table 12.4 of ref. 17, we assigned a carbon factor of 0.225 for countries entirely or mostly in the tropics; 0.295 for countries entirely or mostly outside the tropics and 0.26 for countries with substantial area in both the tropics and non-tropics. The \({p}_{u}\) is the stumpage price of each harvested wood product, which we assigned on the basis of ref. 67 as follows: the price of wood harvested for bioenergy was equal to US$3.46 m−3, which was the median value across 12 sites in LMICs. The price of pulpwood, used for making paper and paper products, was equal to US$18 m−3, which was the median value across 26 sites in LMICs. The price of sawtimber, used for making lumber and other solid wood products, was equal to US$38 m−3, which was the median value across 15 sites in LMICs.

Applying the most likely plantation genus at the site level and \(\omega\) and \(\rho\) at the national level has the potential for mismatch between these two model inputs. We addressed this potential for mismatch between \(\omega ,\) \(\rho\) and most likely plantation genus in a sensitivity analysis. In one alternative scenario, we assumed that all plantation forestry would be eucalyptus for bioenergy (\({\alpha }_{{\rm{bioenergy}}}\) = 1 in equations (6) and (7)). In another alternative scenario, we assumed that all plantation forestry would be broadleaf deciduous or broadleaf evergreen for solid wood production (\({\alpha }_{{\rm{solid}}\; {\rm{wood}}}=1\) in equations (6) and (7)).

For natural regeneration, we assumed no periodic harvest or sale of wood products. Although naturally regenerated forests are a substantial source of wood products in some low- and middle-income countries, they are much more likely to be selectively harvested for short-lived fuelwood or charcoal than intensively logged for commercial timber for long-lived wood products. In a sensitivity analysis we assumed periodic clearing of naturally regenerating forests on the same interval as plantations are harvested. In this scenario, we assumed naturally regenerating forests did not produce carbon storage in, or revenue from, the sale of wood products to provide an upper bound on the effect of periodic clearing.

Carbon accumulation in belowground biomass and soil

We calculated carbon accumulation in belowground biomass as a multiple of carbon accumulation in aboveground biomass, using root–shoot ratios from Table 4.4 of ref. 17, noting that belowground carbon accumulation does not always track linearly with aboveground carbon accumulation68. That is,

$${{\rm{BG}}}_{{it}}^{N}={r}_{{\rm{decg}}}^{N} {\rm{AG}}_{{it}}^{N}$$
(8)
$${{\rm{BG}}}_{{it}}^{P}={r}_{{\rm{decg}}}^{P} {\rm{AG}}_{{it}}^{P}$$
(9)

A root–shoot ratio of a site, \({r}_{i}\), is specific to reforestation method (natural versus plantation), continent, domain, ecological zone and, in some cases, genus. In cases where this source did not provide a root–shoot ratio, we used a generic default of 0.26 based on ref. 69. We assumed all belowground biomass would be emitted in the year of harvest (ch. 4.15 of ref. 17).

We specified carbon accumulation in soil for natural forest regeneration to be 0.415 tC ha−1 yr−1, following refs. 17,70. We specified carbon accumulation in soil for plantations to be 0.0927 tC ha−1 yr−1 during the first harvest cycle70 and zero in subsequent harvest cycles (ch. 4.2 of ref. 17). That is:

$${S}_{{it}}^{N}=0.415t$$
(10)
$${S}_{{it}}^{P}=0.0927t{\rm{|}}t < {H}_{i}$$
(11)

Abatement cost of reforestation

We calculated the site-specific abatement cost (US$ per tCO2) by dividing reforestation costs by carbon benefits. In the case of natural regeneration, reforestation costs (US$ ha−1) are the sum of one-time opportunity cost and the 30 year time-discounted discrete time-series flow of implementation cost. Carbon benefits (tCO2 ha−1) are the 30 year time-discounted carbon accumulation in three pools (aboveground biomass, belowground biomass and soil), multiplied by 3.67, the atomic ratio between carbon dioxide and carbon. That is:

$${{\rm{AC}}}_{i}^{N}=\frac{{{\rm{OC}}}_{i}^{N}+\mathop{\sum }\nolimits_{t=1}^{y}{\Delta {\rm{IC}}}_{{it}}^{N}{(1-\delta )}^{t-1}}{3.67 \mathop{\sum }\nolimits_{t=1}^{y}\left({\Delta {\rm{AG}}}_{{it}}^{N}+{\Delta {\rm{BG}}}_{{it}}^{N}+{\Delta S}_{{it}}^{N}\right){(1-\delta )}^{t-1}}$$
(12)

In the case of plantations, opportunity cost and implementation costs are less net revenue from wood products (US$ ha−1). Note that the total cost of plantation reforestation can be negative if revenue from the sale of wood products is sufficiently high. Carbon benefits of plantations included a fourth pool: carbon storage in wood products. That is:

$${{\rm{AC}}}_{i}^{P}=\frac{{{\rm{OC}}}_{i}^{P}+\mathop{\sum }\nolimits_{t=1}^{y}{\Delta {\rm{IC}}}_{{it}}^{P}{(1-\delta )}^{t-1}-\rho \mathop{\sum }\nolimits_{t=1}^{y}{\Delta H}_{{it}}^{\,P}(1-\delta )^{t-1}}{3.67 \left(\mathop{\sum }\nolimits_{t=1}^{y}\left({\Delta {\rm{AG}}}_{{it}}^{P}+{\Delta {\rm{BG}}}_{{it}}^{P}+{\Delta S}_{{it}}^{P}\right){\left(1-\delta \right)}^{t-1}+\omega \mathop{\sum }\nolimits_{t=1}^{y}{\Delta H}_{{it}}^{\,P}{(1-\delta )}^{t-1}\right)}$$
(13)

We further calculated the cost-effectiveness surface for natural and plantation reforestation jointly, choosing the more cost-effective option, that is the option with the lower abatement cost, at each site:

$${{\rm{AC}}}_{i}^{\min }=\min\left\{{{\rm{AC}}}_{i}^{P},{{\rm{AC}}}_{i}^{P}\right\}$$
(14)

In reality, in the absence of carbon price incentives many land users might prefer the method that is more profitable or that provides the most biodiversity habitat or ecosystem services.

We constrained the geographic scope of the cost-effectiveness surfaces to only those regions included in both the cost surfaces and the benefit surfaces. That is, 12 biomes within 138 LMICs. We did not impose further geographic constraints on potential reforestation at this stage.

Constructing the marginal abatement cost curves

We constructed marginal abatement cost curves for (1) natural regeneration independently; (2) plantation reforestation independently; and (3) natural and plantation reforestation jointly, choosing the more cost-effective option at each site. We first ordered grid cells from lowest to highest cost-per-ton (US$ per tCO2) along the y axis and then summed cumulative abatement quantity (tCO2) along the x axis.

We then sequentially applied three geographic screens, corresponding to: (1) suitability of potential reforestation generally; (2) feasibility of natural regeneration and/or plantation forestry specifically; and (3) additionality. First, we applied a binary geographic screen for suitability of potential reforestation generally from ref. 43. This screen combines positive elements of where reforestation cannot happen (for example, deserts, mountaintops and existing forests) and normative elements of where researchers have argued reforestation should not happen (for example, native grasslands and croplands). As a sensitivity analysis, we considered three alternative published geographic screens of the suitability of potential reforestation: tropical forest restoration feasibility46; global potential to increase tree canopy cover from below to above 25% (ref. 44); and high unrealized potential for restoration of woody-dominated systems45. We also present results without a reforestation suitability screen.

Second, we screened out areas where natural regeneration or plantation reforestation specifically would be unlikely to occur. The potential for natural regeneration varies spatially49; for example, it is unlikely to occur far from existing natural forests because of the paucity of seed sources and low likelihood of seed dispersal. We set the binary screen for natural regeneration at a buffer distance of 5 km from existing natural forest71. Others49 report 5 km to be the distance from existing forest within which natural regeneration in tropical and subtropical regions has the most success (84% likelihood of success). Plantation reforestation is unlikely to occur far from existing plantation forests because of the inferred likely absence of mill infrastructure. We set the binary screen for plantation reforestation at a buffer distance of 40 km from existing plantation forest71. This distance was intended to provide an equivalent probability of restoration success as for natural forest, by the following logic: the median maximum millshed buffer is ~110 km (refs. 72,73,74,75,76,77,78,79,80). A plantation buffer of ~40 km corresponds to 85% likelihood of being within the millshed buffer; that is, a site within 40 km of a plantation within 110 km of a mill is 85% likely to also be within 110 km of that mill.

Third, we subtracted the fraction of reforestation that was unlikely to be additional, that is, would have occurred anyway. Within the tropics, we did so by subtracting the cell-specific, time-discounted fraction of the cell in which forest-cover gain is predicted to occur under business as usual from 2020 to 2050 (ref. 41). Outside the tropics, we assumed the business-as-usual reforestation percentage to be equal to 5.7%, which is the tropics-wide average of business-as-usual forest-cover gain41 within areas considered potentially reforestable43. As a sensitivity analysis, we further removed sites where we estimated plantations to have a negative abatement cost, that is, be profitable even in the absence of any carbon price incentives. Neither of these approaches for accounting for additionality is intended to be used by reforestation project proponents as a substitute for mechanism-specific additionality tests (for example, ref. 81).

Scope and parameter assumptions

We sought to cover the largest possible geographic scope to: compare results across geographic regions using common data and assumptions; provide input to global mitigation cost analyses1; gap-fill data for regions where regional data does not exist; and identify priority regions for deeper future analyses. We constrained our analysis to 138 LMICs because of the geographic extent of data on implementation costs and opportunity costs. This geographic scope spans tropical, subtropical and some temperate zones. As described above, we further limited the scope to areas in which previous researchers43 considered reforestation to be suitable.

We applied a time horizon of 30 years, which is equivalent to 2020–2050, as in ref. 41. In sensitivity analyses, we considered alternative time horizons of 50 years (equivalent to 2020–2070) and 80 years (equivalent to 2020–2100). We applied a time-discount rate of 5%, which is simultaneously near the low end of the range for private land-use decisions in the tropics and near the high end of the range for public policy decisions, for example involving climate change. In a sensitivity analysis we applied alternative discount rates of 2.5% and 10%. We present results for a cost of US$50 per tCO2, which is simultaneously near the low end of the range for social cost of carbon estimates82 and near the high end of the range for current carbon market prices. We alternatively present results for costs of US$20–100 per tCO2. We report all values in 2020 US$.

Our modelling approach did not enable us to model the elasticity of demand for wood products, that is, we assumed that the price of wood products would not change even as harvest quantity from plantations increased. In reality, large changes in the supply of wood products, such as those modelled here, would decrease prices and thus dampen the increase in plantation area unless counterbalanced by corresponding increases in demand, for example a shift in construction materials from cement or steel to timber83,84. Incorporating more realistic assumptions about demand for wood products (as in ref. 42) would reduce the abatement potential from plantations. We identified an upper bound on the effect of wood product market feedback by including a sensitivity analysis in which the price of wood products was zero, that is, plantations earned no revenue from the sale of wood products.

Our work suggests several useful extensions beyond the scope of this study. It will be useful to extend the geographic scope of information on reforestation costs to include high-income countries and to calculate costs for more reforestation methods such as assisted natural regeneration, agroforestry systems, mixed-species planted forests or tree-based production systems (for example, fruit orchards and oil palm, rubber, cocoa or coffee plantations).