Although as much as 90 to 99% of deforestation occurs in landscapes where agriculture is the main driver of TCL, only 45 to 65% of deforestation can be attributed to the expansion of actively managed cropland, pasture, or tree crops.

First, the lack of consistent pantropical data on deforestation still hampers our ability to assess overall deforestation trends and thus the net impacts of interventions to reduce deforestation while accounting for leakage across regions and biomes (109–111). Improvements in deforestation data are needed in three main areas: (i) to encompass both dry and wet tropics, (ii) to provide estimates of deforestation that go beyond TCL and satisfy the commonly held definition of a persistent conversion of natural forest to any other land use, and (iii) to ensure that estimates are consistent across regions and over time.

Logging and demand for wood products (e.g., timber and pulp), charcoal, and fuelwood—alongside agricultural expansion—are key direct drivers of deforestation and even more so of degradation (6, 55, 101, 102). Although deforestation resulting from the expansion of tree plantations is estimated by Goldman et al. (36) and Pendrill et al. (37) (0.1 and 0.8 Mha per year, respectively, with the former only covering eight countries), deforestation due to logging and timber extraction that sometimes occurs in conjunction with and facilitates agriculture expansion (49, 50, 95) is not comprehensively quantified at the pantropical level.

A physical trade model—which traces deforestation embodied in raw or lightly processed agricultural commodities—suggests that 20 to 25% of all deforestation resulting in agricultural production is linked to exports (37) (fig. S5). This average, however, hides substantial variation across countries, regions, and commodities (fig. S6): soybeans, palm oil, and cash crops (e.g., rubber, coffee, and cocoa) are primarily destined for export markets whereas beef and cereals are typically consumed domestically. An economic, multiregional input-output model, which traces deforestation all the way to final consumption, raises the share of commodity-driven deforestation linked to international demand to ~35% (37) (fig. S5).

Following pasture the next most important land uses are oil palm and soy cultivation, together accounting for at least a fifth of the deforestation resulting in agricultural production

Most of the deforestation due to the expansion of pastures is found in South America (~1.2 and 2.1 Mha per year) (Fig. 2), particularly in Brazil. This region has robust data on pasture-driven deforestation at the national or biome level

estimated extent of deforestation attributed to the expansion of pastures [1.9 compared with 2.7 Mha per year

cattle pasture expansion is the single most important deforestation driver by far, alone accounting for around half of the deforestation resulting in agricultural production

Almost all fires in tropical moist forests are due to human activities (42) including clearing forests for new agriculture and as a land management tool (e.g., weed control and nutrient mobilization) in already-cleared agricultural areas (42). This frequently leads to fires spreading into adjacent forest areas as documented in Brazil (76), the Miombo (77), and Indonesia (78).

land speculation, often linked to unclear or contested tenure. This process has been documented for several Latin American countries including the Brazilian Amazon (58, 59) and Costa Rica (60), where expectations about future agricultural rents—fueled by planned road infrastructure improvements, uncertainties around future forest conservation policies, and the existence of large tracts of undesignated public land—lead to speculative clearing. Other social processes such as imitation [see (61, 62), for example] create crop booms and potential busts (63). This can lead to land being cleared anticipatively but not subsequently being taken into production because of deteriorating market conditions, failed operations, or diminishing economic viability. For instance, land cleared for speculation in the Brazilian Amazon is typically put under extensive pasture where animal stocking rates are very low; these pastures are commonly degraded and abandoned within relatively short time periods (64–66). Deforestation can also be used to strengthen tenure claims where laws link land rights to clearing or use (67, 68). Moreover, conflicts over land tenure often contribute to deforestation in contested forest frontiers, in excess of clearings purely for productive agriculture (69, 70). The extent of land with unclear and contested tenure is not precisely quantified pantropically but is shown to be very large in some countries (71). Land degradation can also lead to land abandonment or maintenance of the land at very low levels of productivity, possibly because the deforested land was not suitable to begin with (72, 73) or because of deforestation-driven changes in local climate (74), inadequate management and lack of expertise, or cultural or structural barriers (66, 75).

one-third to one-half of agriculture-driven deforestation does not result in recorded agricultural land

Curtis et al. (7) put deforestation occurring in landscapes where agriculture is the dominant driver in the range of 5.19 Mha per year (commodity-driven deforestation only) to 9.47 Mha per year (sum of commodity-driven deforestation and shifting agriculture) (Fig. 2). We narrowed this range down to 6.4 to 8.8 Mha per year (28) by excluding TCL in tree plantations (53) and by including deforestation in primary forests (54) and deforestation resulting in agricultural production [based on Pendrill et al. (37)] (fig. S4).Our analysis suggests a large discrepancy (2.0 to 4.5 Mha per year) between the deforestation resulting in agricultural production (>4.3 Mha per year) and the overarching category of agriculture-driven deforestation (6.4 to 8.8 Mha per year) (Figs. 1A and 3). This discrepancy is present across all three continents in our country sample, totaling 1.0 to 2.0 Mha per year in Latin America, 0.0 to 1.3 Mha per year in Africa, and 1.1 to 1.2 Mha per year in Asia (Fig. 3), though uncertainties abound and part of the discrepancy is likely a result of unrecorded agricultural areas.

the Curtis et al. (7) approach presents two key challenges. First, it does not fully distinguish which of the GFC TCL is deforestation. Some of the dominant drivers of TCL correspond to deforestation (i.e., commodity-driven deforestation and urbanization) but others do not (i.e., wildfires potentially resulting in regrowth). Still, the large remainder—i.e., shifting agriculture and forestry—can reflect both the expansion of these systems into natural forests (i.e., deforestation) and regular rotations in stable shifting agriculture systems, plantations, or managed forests, which do not constitute deforestation under most definitions (including the one adopted here). Second, the Curtis et al. (7) approach allocates all TCL in each grid cell to a single dominant (defined as >50%) driver of TCL for the whole time period (2001 to 2020), ignoring drivers that are not dominant. Therefore, even in the grid cells where commodity-driven deforestation or shifting agriculture is the dominant driver of TCL, not all TCL is necessarily directly driven by agricultural expansion. The Curtis et al. (7) estimate is thus a metric of deforestation occurring in landscapes where agriculture is the dominant direct driver of forest loss (rather than only deforestation resulting in agricultural production per se).

agriculture is the dominant land use following deforestation. Estimates of deforestation drivers, e.g., the relative importance of agriculture and of different commodities, are intrinsically less reliable in the most recent years, because time is needed to reveal whether the cleared land will be used for production (and if so, for what) or allowed to regenerate. Typically use of the cleared land is assessed within at least two to four years after forest clearing although the precise number of years varies between studies (~1 to <20 years) depending on method and data availability (28). For these reasons we focus our analysis on the period 2011 to 2015.

With total deforestation ranging between 6.5 and 9.5 Mha per year (table S3), this implies that most (~90 to 99%) tropical deforestation occurs in landscapes where agriculture is the dominant driver of forest loss (28).The Pendrill et al. (37) data suggest a much smaller share of tropical deforestation resulting in agricultural production, in the range of ~45 to 65% of our total tropical deforestation estimate

By combining these two assessments [Curtis et al. (7) and Pendrill et al. (37)] with ancillary data (28) we estimate total agriculture-driven deforestation across the tropics to be 6.4 to 8.8 Mha per year

For that period three studies provide pantropical estimates of agriculture-driven deforestation (fig. S3). One [Carter et al. (32)] assumes a constant fraction of deforestation being agriculture-driven on the basis of pre-2010 data from other studies [De Sy et al. (8) and Hosonuma et al. (20)]. The other two—despite relying on the same GFC TCL data (1)—provide notably different estimates of agriculture-driven deforestation ranging from 4.3 Mha per year [Pendrill et al. (37)] up to 9.6 Mha per year [Curtis et al. (7)] (Fig. 1B and table S4).

First, there is no single way to distinguish between forests and nonforests nor between deforestation and forest degradation and as a result different studies and monitoring systems rely on different definitions (29–31). Second, although remote sensing is useful for monitoring forest changes in terms of land cover not all aspects of deforestation—including its underlying drivers—can be observed from satellites; further, technical and practical constraints result in imperfect data (e.g., negative effects of cloud cover) (29, 30). Forest loss estimates therefore differ between studies (fig. S2) both because of measurement uncertainties (32) and because they strive to measure different things.

Box 1.Key terms for disentangling agriculture-driven deforestation.Natural forest: A forest that “resembles—in terms of species composition, structure and ecological function—one that is or would be found in a given area in the absence of major human impact” (33). Aside from primary and intact forests, natural forest also includes regenerated (second-growth) forests and partially degraded forests, provided they fulfill the definition above (33). As no comprehensive pantropical map of natural forests currently exists, most studies approximate their extent.Deforestation: A persistent conversion of natural forest to any other land use such as agriculture, human settlements, or tree plantations.Agriculture: Agriculture includes cropland, pastures, and tree crops but not forestry (agriculture thus excludes timber, pulp, and paper).Agriculture-driven deforestation: Deforestation for which agriculture, directly or indirectly, is a cause; this includes both deforestation resulting in agricultural production and agriculture-driven deforestation without expansion of agricultural production. Agriculture-driven deforestation does not necessarily mean that agriculture is the only or main cause of deforestation; for example, deforestation may be directly driven by the demand for timber alongside the demand for agricultural expansion (49, 50, 95); further, indirect or underlying drivers always play a role (6, 27).Deforestation resulting in agricultural production: This is deforestation that can be attributed to the expansion of land under active agricultural production systems.Agriculture-driven deforestation without expansion of agricultural production: This is defined as deforestation occurring in landscapes where agriculture is the dominant driver of forest loss but does not result in recorded, productive, and actively managed agricultural land. This can be due to several mechanisms and is distinct from forest degradation or other TCL in the sense that the forest has been fully cleared and there are signs of other types of land use, though in practice the boundary can be hard to draw.

However, at present policies are being designed and evaluated against a backdrop of widespread uncertainty regarding our understanding of the links between agriculture and deforestation. The focus on agricultural supply chain policies is commonly premised on statements that agricultural expansion and production drive 80% of tropical deforestation

Emerging policies often focus on eliminating deforestation from international supply chains of agrifood commodities such as palm oil, soybeans, and beef.

we find that while the overwhelming majority (90 to 99%) of tropical deforestation occurs in landscapes where agriculture is the dominant driver of tree cover loss, a smaller share (45 to 65%) of deforestation is due to the expansion of active agricultural production into forests.

Although these studies agree that agriculture is the dominant land use following forest clearing, their estimates of pantropical rates of agriculture-driven deforestation during the period 2011 to 2015 vary greatly—between 4.3 and 9.6 million hectares (Mha) per year—with our synthesized estimate being 6.4 to 8.8 Mha per year.

Agricultural expansion is a primary cause of tropical deforestation and therefore a key driver of greenhouse gas emissions, biodiversity loss, and the degradation of ecosystem services vital to the livelihoods of forest-dependent and rural people.

Although as much as 90 to 99% ofdeforestation occurs in landscapes where agri-culture is the main driver of TCL, only 45 to65% of deforestation can be attributed to theexpansion of actively managed cropland, pas-ture, or tree crops.

First, the lack of consistent pantropical dataon deforestation still hampers our ability toassess overall deforestation trends and thusthe net impacts of interventions to reduce de-forestation while accounting for leakage acrossregions and biomes (109–111). Improvementsin deforestation data are needed in threemain areas: (i) to encompass both dry and wettropics, (ii) to provide estimates of deforesta-tion that go beyond TCL and satisfy the com-monly held definition of a persistent conversionof natural forest to any other land use, and (iii)to ensure that estimates are consistent acrossregions and over time.

Logging and de-mand for wood products (e.g., timber andpulp), charcoal, and fuelwood—alongside agri-cultural expansion—are key direct drivers ofdeforestation and even more so of degradation(6, 55, 101, 102). Although deforestation re-sulting from the expansion of tree planta-tions is estimated by Goldman et al. (36) andPendrill et al. (37) (0.1 and 0.8 Mha per year,respectively, with the former only coveringeight countries), deforestation due to loggingand timber extraction that sometimes occursin conjunction with and facilitates agricul-ture expansion (49, 50, 95) is not comprehen-sively quantified at the pantropical level.

A phys-ical trade model—which traces deforestationembodied in raw or lightly processed agricul-tural commodities—suggests that 20 to 25% ofall deforestation resulting in agricultural pro-duction is linked to exports (37) (fig. S5). Thisaverage, however, hides substantial variationacross countries, regions, and commodities (fig.S6): soybeans, palm oil, and cash crops (e.g., rub-ber, coffee, and cocoa) are primarily destinedfor export markets whereas beef and cerealsare typically consumed domestically. An eco-nomic, multiregional input-output model,which traces deforestation all the way to finalconsumption, raises the share of commodity-driven deforestation linked to internationaldemand to ~35% (37) (fig. S5).

Following pasture the next most importantland uses are oil palm and soy cultivation,together accounting for at least a fifth of thedeforestation resulting in agricultural produc-tion

Most of the de-forestation due to the expansion of pasturesis found in South America (~1.2 and 2.1 Mhaper year) (Fig. 2), particularly in Brazil. Thisregion has robust data on pasture-driven de-forestation at the national or biome level

estimated extent of deforestation attri-buted to the expansion of pastures [1.9 com-pared with 2.7

It is well established that cattle pasture ex-pansion is the single most important deforesta-tion driver by far, alone accounting for aroundhalf of the deforestation resulting in agricul-tural production

Almost all fires intropical moist forests are due to human ac-tivities (42) including clearing forests for newagriculture and as a land management tool(e.g., weed control and nutrient mobilization)in already-cleared agricultural areas (42). Thisfrequently leads to fires spreading into adja-cent forest areas as documented in Brazil (76),the Miombo (77), and Indonesia (78).

land speculation,often linked to unclear or contested tenure.This process has been documented for severalLatin American countries including the Brazil-ian Amazon (58, 59) and Costa Rica (60), whereexpectations about future agricultural rents—fueled by planned road infrastructure improve-ments, uncertainties around future forestconservation policies, and the existence of largetracts of undesignated public land—lead tospeculative clearing. Other social processessuch as imitation [see (61, 62), for example]create crop booms and potential busts (63).This can lead to land being cleared anticipa-tively but not subsequently being taken intoproduction because of deteriorating marketconditions, failed operations, or diminishingeconomic viability. For instance, land clearedfor speculation in the Brazilian Amazon is typ-ically put under extensive pasture where ani-mal stocking rates are very low; these pasturesare commonly degraded and abandoned withinrelatively short time periods (64–66). Deforesta-tion can also be used to strengthen tenureclaims where laws link land rights to clearingor use (67, 68). Moreover, conflicts over landtenure often contribute to deforestation incontested forest frontiers, in excess of clear-ings purely for productive agriculture (69, 70).The extent of land with unclear and contestedtenure is not precisely quantified pantropi-cally but is shown to be very large in somecountries (71). Land degradation can alsolead to land abandonment or maintenanceof the land at very low levels of productivity,possibly because the deforested land was notsuitable to begin with (72, 73) or because ofdeforestation-driven changes in local climate(74), inadequate management and lack ofexpertise, or cultural or structural barriers(66, 75).

one-third to one-half of agriculture-driven deforestation doesnot result in recorded agricultural land (thoughit might be used for other purposes).

Our analysis suggests a large discrepancy(2.0 to 4.5 Mha per year) between the de-forestation resulting in agricultural produc-tion (>4.3 Mha per year) and the overarchingcategory of agriculture-driven deforestation(6.4 to 8.8 Mha per year) (Figs. 1A and 3). Thisdiscrepancy is present across all three con-tinents in our country sample, totaling 1.0 to2.0 Mha per year in Latin America, 0.0 to1.3 Mha per year in Africa, and 1.1 to 1.2 Mhaper year in Asia (Fig. 3), though uncertaintiesabound and part of the discrepancy is likely aresult of unrecorded agricultural areas.

Curtis et al. (7) put deforestation occurringin landscapes where agriculture is the dom-inant driver in the range of 5.19 Mha per year(commodity-driven deforestation only) to9.47 Mha per year (sum of commodity-drivendeforestation and shifting agriculture) (Fig. 2).We narrowed this range down to 6.4 to 8.8 Mhaper year (28) by excluding TCL in tree plan-tations (53) and by including deforestationin primary forests (54) and deforestation re-sulting in agricultural production [based onPendrill et al. (37)] (fig. S4).

fore, even in the grid cells where commodity-driven deforestation or shifting agriculture isthe dominant driver of TCL, not all TCL is nec-essarily directly driven by agricultural expan-sion. The Curtis et al. (7) estimate is thus ametric of deforestation occurring in landscapeswhere agriculture is the dominant direct driverof forest loss (rather than only deforestationresulting in agricultural production per se).

the Curtis et al. (7) approach presents twokey challenges. First, it does not fully distin-guish which of the GFC TCL is deforestation.Some of the dominant drivers of TCL corre-spond to deforestation (i.e., commodity-drivendeforestation and urbanization) but others donot (i.e., wildfires potentially resulting in re-growth). Still, the large remainder—i.e., shiftingagriculture and forestry—can reflect both theexpansion of these systems into natural forests(i.e., deforestation) and regular rotations instable shifting agriculture systems, plantations,or managed forests, which do not constitutedeforestation under most definitions (includingthe one adopted here). Second, the Curtis et al.(7) approach allocates all TCL in each grid cell toa single dominant (defined as >50%) driver ofTCL for the whole time period (2001 to 2020),ignoring drivers that are not dominant. There-

The Pendrill et al. (37) data suggest a muchsmaller share of tropical deforestation result-ing in agricultural production, in the range of~45 to 65%

Withtotal deforestation ranging between 6.5 and9.5 Mha per year (table S3), this implies thatmost (~90 to 99%) tropical deforestation oc-curs in landscapes where agriculture is thedominant driver of forest loss

By combining these two assessments[Curtis et al. (7) and Pendrill et al. (37)] with an-cillary data (28) we estimate total agriculture-driven deforestation across the tropics tobe 6.4 to 8.8 Mha per year (

For that period three studies provide pan-tropical estimates of agriculture-driven de-forestation (fig. S3). One [Carter et al. (32)]assumes a constant fraction of deforestationbeing agriculture-driven on the basis of pre-2010 data from other studies [De Sy et al. (8) andHosonuma et al. (20)]. The other two—despiterelying on the same GFC TCL data (1)—providenotably different estimates of agriculture-drivendeforestation ranging from 4.3 Mha per year[Pendrill et al. (37)] up to 9.6 Mha per year[Curtis et al. (7)] (Fig. 1B and table S4).

agricultureis the dominant land use following deforest-ation. Estimates of deforestation drivers, e.g.,the relative importance of agriculture and ofdifferent commodities, are intrinsically lessreliable in the most recent years, because timeis needed to reveal whether the cleared landwill be used for production (and if so, for what)or allowed to regenerate. Typically use of thecleared land is assessed within at least two tofour years after forest clearing although theprecise number of years varies between studies(~1 to <20 years) depending on method anddata availability (28). For these reasons wefocus our analysis on the period 2011 to 2015.

Box 1. Key terms for disentangling agriculture-driven deforestation.Natural forest: A forest that “resembles—in terms of species composition, structure and ecologicalfunction—one that is or would be found in a given area in the absence of major human impact” (33).Aside from primary and intact forests, natural forest also includes regenerated (second-growth) forestsand partially degraded forests, provided they fulfill the definition above (33). As no comprehensivepantropical map of natural forests currently exists, most studies approximate their extent.Deforestation: A persistent conversion of natural forest to any other land use such as agriculture,human settlements, or tree plantations.Agriculture: Agriculture includes cropland, pastures, and tree crops but not forestry (agriculturethus excludes timber, pulp, and paper).Agriculture-driven deforestation: Deforestation for which agriculture, directly or indirectly, is acause; this includes both deforestation resulting in agricultural production and agriculture-drivendeforestation without expansion of agricultural production. Agriculture-driven deforestation does notnecessarily mean that agriculture is the only or main cause of deforestation; for example, deforestationmay be directly driven by the demand for timber alongside the demand for agricultural expansion(49, 50, 95); further, indirect or underlying drivers always play a role (6, 27).Deforestation resulting in agricultural production: This is deforestation that can be attributed tothe expansion of land under active agricultural production systems.Agriculture-driven deforestation without expansion of agricultural production: This is definedas deforestation occurring in landscapes where agriculture is the dominant driver of forest loss butdoes not result in recorded, productive, and actively managed agricultural land. This can be due toseveral mechanisms and is distinct from forest degradation or other TCL in the sense that the foresthas been fully cleared and there are signs of other types of land use, though in practice the boundarycan be hard to draw.RES EARCH | REVIEWCorrected 15 September 2022. See full text.

Second, although remote sensing is useful formonitoring forest changes in terms of landcover not all aspects of deforestation—includingits underlying drivers—can be observed fromsatellites; further, technical and practical con-straints result in imperfect data (e.g., nega-tive effects of cloud cover) (29, 30). Forest lossestimates therefore differ between studies (fig.S2) both because of measurement uncertain-ties (32) and because they strive to measuredifferent things.

First, there is no single way todistinguish between forests and nonforests norbetween deforestation and forest degradationand as a result different studies and monitoring

However, at present policies are being de-signed and evaluated against a backdrop ofwidespread uncertainty regarding our under-standing of the links between agriculture anddeforestation. The focus on agricultural sup-ply chain policies is commonly premised onstatements that agricultural expansion andproduction drive 80% of tropical deforesta-tion,

Emerging policiesoften focus on eliminating deforestation frominternational supply chains of agrifood com-modities such as palm oil, soybeans, and beef.

we find thatwhile the overwhelming majority (90 to 99%)of tropical deforestation occurs in landscapeswhere agriculture is the dominant driver oftree cover loss, a smaller share (45 to 65%) ofdeforestation is due to the expansion of activeagricultural production into forests.

Although these studies agree that agricultureis the dominant land use following forestclearing, their estimates of pantropical ratesof agriculture-driven deforestation duringthe period 2011 to 2015 vary greatly—between4.3 and 9.6 million hectares (Mha) per year—with our synthesized estimate being 6.4 to8.8 Mha per year.

Agricultural expansion is a pri-mary cause of tropical deforestation and there-fore a key driver of greenhouse gas emissions,biodiversity loss, and the degradation ofecosystem services vital to the livelihoods offorest-dependent and rural people.

with every mm of water deficit, deforestation tends to increase by 0.13% per year.

First, we relate observed fluctuations in deforestation rates to dry-season intensity; second, we determine the effect of conversion of forest to cropland on evapotranspiration; and third, we simulate the subsequent downwind reductions in rainfall due to decreased atmospheric water input.