Technical Comments

Comment on “The global tree restoration potential”

See allHide authors and affiliations

Science  18 Oct 2019:
Vol. 366, Issue 6463, eaay7976
DOI: 10.1126/science.aay7976


Bastin et al.’s estimate (Reports, 5 July 2019, p. 76) that tree planting for climate change mitigation could sequester 205 gigatonnes of carbon is approximately five times too large. Their analysis inflated soil organic carbon gains, failed to safeguard against warming from trees at high latitudes and elevations, and considered afforestation of savannas, grasslands, and shrublands to be restoration.

Bastin et al. (1) used remote sensing and machine learning to estimate that global “tree restoration” could sequester 205 gigatonnes of carbon (GtC). If accurate and achievable, this would constitute an astounding accomplishment, equal to 20 times the current annual fossil fuel emissions (10 GtC/year) (2) and about one-third of total historical anthropogenic emissions (660 GtC) (2). Unfortunately, key assumptions and data underlying Bastin et al.’s analyses are incorrect, resulting in a factor of 5 overestimate of the potential for new trees to capture carbon and mitigate climate change. We show that Bastin et al. (i) overestimated soil carbon gains from increased tree cover by a factor of 2; (ii) modeled new tree cover in regions where trees reduce albedo and increase climate warming (3, 4); and (iii) relied heavily on afforestation of grasslands and savannas—biodiverse ecosystems where fires and large herbivores have maintained low tree cover for millions of years (5, 6).

Bastin et al.’s inflation of soil carbon gains resulted in a ~98 GtC overestimate of potential carbon sequestration (Table 1). They mistakenly assumed that treeless areas have no soil organic carbon (SOC) and that SOC increases in direct (1:1) proportion to tree cover. The contribution of SOC to total carbon stocks is substantial in most terrestrial ecosystems. In humid tropical savannas, for example, 86% of all carbon is in soils (174 tonnes of SOC per hectare) (7). In boreal forests, 64% of carbon occurs in soils (8). North American grasslands can store as much carbon in soil (9) as tropical forests store as biomass (8). In Table 1, we display SOC-corrected carbon sequestration estimates that use more realistic (literature-derived) values for the changes in SOC that occur with afforestation and reforestation.

Table 1 Corrected estimates of the potential for increased tree cover to sequester carbon and mitigate climate change.

We corrected Bastin et al.’s estimate (205 GtC) to represent realistic gains or losses of soil organic carbon (SOC) that occur with increased tree cover in each biome [based on (9, 1621)]. We then excluded biomes (assigned a value of 0 GtC) where tree planting for climate change mitigation should not occur because of unintended consequences (e.g., net warming from reduced albedo or loss of biodiversity). Although we disagree with several of the carbon density values used by Bastin et al. [e.g., they applied values for intact tropical forests (8) to estimate second-growth forest biomass, and applied values from humid tropical savannas (7) to deserts and tundra], we retained these values to demonstrate the magnitude of the SOC and biome corrections.

View this table:

In addition to the SOC overestimate, Bastin et al. did not account for the warming effect of trees due to decreased albedo (3, 4). Trees, particularly evergreen conifers, are less reflective than snow, bare ground, or grasses, and thus absorb more solar energy, which is ultimately emitted as heat. At high latitudes and elevations, the warming effect of trees is greater than their cooling effect via carbon sequestration (3, 4). Similarly, trees planted in low-latitude, semi-arid regions can produce net warming for decades before carbon sequestration benefits are realized (10). Because, at a minimum, carbon from trees planted in boreal forests, tundra, or montane grasslands and shrublands should not be counted as climate change mitigation (Bastin et al. counted a SOC-corrected 17 GtC), in Table 1 we provide a corrected estimate that excludes these biomes.

The carbon sequestration estimate of Bastin et al. is also dependent on the false assumption that natural grasslands and savannas with fewer trees than predicted by their statistical model are “degraded” and in need of restoration (11). Ecological restoration of savannas and grasslands rarely involves planting trees, and more often requires tree-cutting and prescribed fire to promote biodiversity and ecosystem services (12). Yet after correcting for SOC, 46% of the carbon sequestration estimate of Bastin et al. comes from increased tree cover in grasslands, savannas, and shrublands (Table 1). Among all biomes, tropical grasslands are the largest contributor to Bastin et al.’s estimate of potential carbon sequestration (SOC-corrected 40 GtC or 37% of the global potential; Table 1).

Although Bastin et al.’s model, developed with climate and soil data in protected areas, may be reasonable in some of the driest and wettest places on Earth, any statistical approach to predict tree cover at intermediate precipitation (500 to 2500 mm annually) must include the effects of fire and, where they still exist, large grazing and browsing animals (13). Because Bastin et al. failed to account for fire, their model had low predictive power across many of the open-canopy biomes they analyzed, as shown by their own uncertainty analysis. Although we commend their intent to respect the “natural ecosystem type” by training their machine-learning algorithm on protected areas, they map many of these same areas—particularly those with grassland-forest mosaics (e.g., Yellowstone National Park, USA)—as opportunities for tree planting. Of additional concern, their method of interpolation between protected areas misrepresents some enormous savanna regions (e.g., western Los Llanos in Colombia is targeted for 75 to 100% tree cover), presumably because the protected areas are located in adjacent tropical forests, not savannas.

Bastin et al.’s model suggesting grasslands and savannas as potential sites for restoration using trees is inaccurate and misguided. Earth’s savannas and grasslands predate humans by millions of years; their formation is a result of complex ecological and evolutionary interactions among herbaceous plants (grasses and forbs with extensive roots and underground storage organs), environmental change (climatic cooling, drying, changes in atmospheric CO2), fires (first ignited by lightning, then by people), and large herbivores (5, 6). These ecosystems and their iconic species are already gravely threatened by fire exclusion and afforestation, processes that replace species-diverse biotic communities with lower-diversity forests (14). Carbon-focused tree planting will exacerbate these threats, to the detriment of people who depend on grasslands to provide livestock forage, game habitat, and groundwater and surface-water recharge (11). Moreover, trees planted in grasslands will be prone to carbon loss from fires. Because these detrimental effects should preclude tree planting in grasslands, savannas, and shrublands, we excluded these biomes from Bastin et al.’s estimate in Table 1.

In combination, our corrections for SOC and corrections to avoid the unintended consequences of misguided tree planting (i.e., warming and biodiversity loss with afforestation) would reduce Bastin et al.’s estimate of potential carbon sequestration by a factor of 5, to the still-substantial amount of ~42 GtC (Table 1). Although ecological restoration, if carefully implemented, can have a role in mitigating climate change, it is no substitute for the fact that most fossil fuel emissions will need to stop to meet the targets of the Paris Agreement (15). Such action should be accompanied by policies that prioritize the conservation of intact, biodiverse ecosystems, irrespective of whether they contain a lot of trees.


Acknowledgments: J.W.V. thanks B. J. Danielson for many conversations on related topics. Funding: Supported by the Texas A&M Sid Kyle Global Savanna Research Initiative (T.W.B.); Swiss National Science Foundation (20FI20_173691) (N.B.); Centre National pour la Recherche Scientifique CNRS PICS 2018-2020 (RESIGRASS) (E.B.); CNPq (Brazil, 303179/2016-3) (G.D.); CNPq (Brazil) (G.W.F.); CNPq (Brazil, 303988/2018-5) (A.F.); NASA award NNX17AK14G (F.F.); NSF award 1354943 (W.A.H.); Fundação de Amparo à Pesquisa do Estado de Minas Gerais (Brazil, 2016/13232-5) (S.L.S.); the Office of the Royal Society (IC170015) (C.E.R.L.); CNPq (Brazil, 310345/2018-9) (G.E.O.); the Spanish Government (FIROTIC, PGC2018-096569-B-I00) (J.G.P.); the National Research Foundation (ACCESS, 114695) (N.S.); CNPq (Brazil, 303568/2017-8) (F.A.O.S.); NSF awards 1342703 and 1926431 (C.J.S. and D.M.G.); NSF award EAR-1253713 (C.A.E.S.); Deutsche Forschungsgemeinschaft grant 5579 POEM (V.M.T.); and USDA-NIFA Sustainable Agricultural Systems Grant 12726253 (J.W.V.). Author contributions: J.W.V. wrote the paper with conceptual input from C.L.P., S.A., C.J.S., G.M., W.J.B., J.J.E., and V.M.T.; J.W.V. performed the carbon sequestration corrections in Table 1; J.G.P. and A.N.N. contributed to the literature search for ΔSOC values. All authors read and provided feedback on the draft manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data, explanations of calculations, and references to literature-derived values are presented in Table 1.

Stay Connected to Science


Navigate This Article