The nature of current tropical reforestation

Extensive reforestation is increasingly prominent in some tropical regions1,2 and is generally considered novel in terms of its extent and drivers3. The degree to which planted tree cover featured in such reforestation has long been suspected as increasingly prominent, following the rise of commercial enterprises as major agents of tropical forest change since the 1990s4,5,6. The contributions of planted tree cover to total reforestation may increase further due to initiatives promoting forest restoration for climate-change mitigation largely via tree plantations and agroforestry4,7. Frustratingly, the actual contributions of planted tree cover to overall reforestation had long been poorly described, clouding insights into the human and biophysical dimensions of reforestation. Only in the past couple of years has the remote sensing community mapped tropical planted tree cover at large scales, though still with notable under-estimation and typically without direct comparisons to natural reforestation8,9,10,11,12,13.

In this context, recent estimates of planted tree cover and natural forest regrowth by Fagan et al.14 have shed a unique light on the nature of reforestation. For the tropics, Fagan et al.14 report that new planted tree cover (planted reforestation hereafter) was nominally as extensive as natural reforestation over 2000–2015. An initial, direct reading of their Landsat satellite classification indicates a ratio of planted-to-natural reforestation of 2 ± 0.0715, while a subsequent interpretation14 adjusted for potential spatial bias halved this ratio to 1.02 ± 0.6 and introduced broader confidence intervals around estimated areas (planted reforestation 32.2 ± 9.4 Mha, natural reforestation 31.6 ± 11.9 Mha). With 95% confidence, planted reforestation is estimated to have comprised 34-68% of all reforestation of 2000–201514. Each ratio above compares the midpoints of the confidence intervals, while its variability is given by the standard deviation of all four possible ratios defined by the high and low extremes of the intervals. The ‘midpoint ratio’ of 1.02 ± 0.6 would notably increase to 1.2 ± 0.5 if it were averaged with these four ‘extreme ratios’, with variability now observed across all five ratios (Supplementary Note 1). These figures describe an “estimated relative dominance of plantations”14 or, more conservatively, a statistical parity between planted and natural reforestation area. Even an approximation of parity would challenge longstanding presumptions of the primacy of tropical natural reforestation central to models of the human and biophysical dimensions of emergent reforestation1,2,16 and forest restoration17. These estimates have in turn rekindled discussion over differences between conceptual and empirical models of reforestation3, with implications for how scholars describe underlying land change and how agents of forest restoration appraise their efforts.

Here, I argue that two crucial but often neglected aspects of reforestation, namely persistence and additionality, would markedly qualify any apparent prevalence of planted or natural reforestation. Upon compiling and generating estimates of persistence and additionality, I experimentally adjust the 1.02 ± 0.6 midpoint ratio of planted-to-natural reforestation to illustrate the implications of these two factors for remotely sensed appraisals of reforestation processes. I suggest that any relative prevalence of planted reforestation is likely confined to a relatively small subset of managed lands, but also that uncertainties persist due to uncertainty over thresholds defining additionality and persistence. To clarify and qualify forms of apparent forest gain, I outline a promising quasi-demographic approach to describing reforestation processes along a full spectrum of additionality and persistence.

The persistence and spuriousness of natural reforestation

Notwithstanding broad confidence intervals, the impression evoked by Fagan et al.’s14 estimates is that of planted reforestation rivalling, and possibly exceeding, natural reforestation at large scales. (Here, planted reforestation entails monocultures for fibre, timber, or fruit, including industrial plantations and smallholder stands alike). The relative prevalence of planted reforestation is however exaggerated by a selective estimation of natural reforestation based on its temporal persistence, that is, the number of years it endures once established.

Reference to the persistence of natural reforestation in order to qualify its ‘actual’ or ‘potential’ extents downwards or upwards is a recent feature in the land-change literature, following the growing availability longer term time series satellite data on forest change3,18,19,20. References to persistence in the literature unsurprisingly remain arbitrary, even conflicting. On the one hand, such references may dismiss short-lived reforestation due to its meagre biodiversity or carbon stock21. On the other hand, they may stress sizeable theoretical ‘losses’ of biodiversity or carbon amongst reforestation of limited persistence, and may define persistence with reference to the timing of reforestation within an observation period as much as by its duration during the period18,19,20. Neither representation captures reforestation processes for what they truly are – a series of disparate, even discordant land-use transitions ultimately yielding an aggregate forest-change trend3,22.

In the case of Fagan et al.14, their estimates reflect reforestation patches persisting up to 15 years (2000–2015), much as do other estimates2,13. Such estimates would omit an unknown, but likely appreciable, extent of reforestation that arose and was re-cleared before the 15th year. And yet expansive tropical reforestation is seemingly characterised by the continual reproduction of extensive but short-lived secondary-forest patches23,24. At regional scales, more than half of a given area of natural reforestation is regularly comprised of patches ≤10–15 years old23,24,25. While a great many of such younger patches are cleared within their first 10 or 15 years23,24,26, their continual replacement over time27,28,29 means that reforestation processes ironically remain both enduring over decades and still characterised by a relative abundance of younger patches23,25, at least during early to intermediate phases of forest transitions. In other words, for a region undergoing a process of forest gain, rather than merely hosting extensive young forest cover, that fraction of reforestation that is ephemeral is, so far as we have observed, persistently ephemeral, rendering it an important and enduring feature of an emergent forest landscape.

The dilemma of the satellite estimation of reforestation area as a means of inferring temporal processes of underlying land change may be illustrated by the analogy of a ‘forest hotel’. In this hotel, half of its rooms are always occupied by youths (i.e., young regrowth), many of whom stay for just one night and who often change rooms upon checkout (i.e., conversion). The other half of the rooms are occupied by older guests (i.e., older regrowth) who stay longer, but less frequently, with progressively older but fewer guests staying for progressively longer durations. Were a census of the hotel’s occupancy taken on a given day, the hotelier could scarcely dismiss those rooms vacated by youths earlier on the same day, since an equal greater number of youth occupants always await to replace them (particularly in the context of reforestation expansion). Similarly, any census of hotel residency over years could not dismiss youth occupants outright, at least not in terms of their aggregate duration or ‘quantity’ of residency (i.e., forest-cover extent), being distinct from their ‘quality’ of their residency (i.e., forest carbon and biodiversity). So it is that, for a given region, to discount roughly half of all reforestation arising within a given 10–15-year time series because it does not occur during the last year of the time series23 is contentious. To labour this analogy, as the hotel grows (reforestation expands), the hotelier (scientist) will ultimately desire to know whether this growth reflects more youths absolutely, or longer stays by younger or older occupants, or the maturation of youths into older occupants. For this, only an inclusive census of all guests will suffice.

And yet, there remains the urge to distinguish conceptually between supposedly ‘spurious’, often ephemeral reforestation on the one hand and supposedly ‘genuine’, ‘robust’, presumably persistent reforestation on the other hand. The former reforestation occurs extensively but often with a questionable affinity with long-term forest stabilisation or gain, given frequent associations with iterant land use such as fallows or cycles of forest degradation and regrowth3. Contrarily, the latter reforestation is generally equated with land disuse and longer-term transitions towards forest gain, as via land abandonment in the case of natural reforestation or via tree planting on long-cleared lands in the case of planted reforestation. Limited high-quality data for Amazonia suggest 2-5 years of regrowth as a persistence threshold beyond which natural reforestation begins to transition conceptually from ‘spurious’ to ‘genuine’. This threshold reflects the rapid decline of regrowth re-clearance probabilities within a few years of regrowth establishment23,24 and concords with most short-fallow periods. Still, as stressed here, the distinction between ‘spurious’, ephemeral reforestation and ‘actual’, persistent reforestation is fraught in the context of long-term reforestation processes, which typically follow ‘fitful’, rather than ‘linear’, trajectories and for which expansive natural reforestation often remains persistently ephemeral21,22. Now that Landsat annual satellite time-series data span multiple decades9,13,30, the consideration of persistence and prevailing forms of reforestation would ideally move away from a given persistence threshold, and instead would adopt a more inclusive, quasi-demographic approach to monitoring all regrowth, detailed below. Reforestation studies would thus move away from simple areal estimates of reforestation as a land-cover class of minimum persistence as of the end of a given period13,20, and would instead describe phases of forest stabilisation and gain across which persistence varies systematically.

Returning to the estimates of Fagan et al.14, Fig. 1 and Table S1 compile estimates of the persistence of tropical natural reforestation to appraise the implications of discounting ephemeral reforestation. To qualify for review in Fig. 1 and Table S1, a published study or independent analysis of its satellite data had to be of a regional scale or greater, have an observation period of ~≥15 years, be spatially explicit, and indicate the percentage of tropical natural reforestation at time1 surviving to time2 or otherwise provide data allowing for this calculation (Supplementary Note 2). The nine estimates compiled in Fig. 1 are believed to represent virtually all those meeting these criteria, encompassing large tropical regions between the mid-20th century and the present. At the upper extreme, estimates for the Brazilian Amazon23 indicate that 60-78% of forest regrowth arising in a given year is cleared within 15 years. This high rate is characterised by an ample inclusion of very young regrowth (e.g., ≤~2 years old), corresponding with transitory land use more than land disuse or land-use transition. At lower extreme, estimates for the neotropics and Peru3,19,24 suggest that only ~10–18% of regrowth is cleared within 15 years. This neotropical rate is likely depressed by the limitations of moderate-resolution satellite data in detecting forest regrowth, particularly mixed, smaller-scale, and/or young regrowth that is most prone to conversion23,31. The remaining estimates in Fig. 1, mostly for neotropical regions but also Vietnam20,27,30,32,33,34,35,36, suggest an intermediate range of 29–61% of regrowth clearance over its first 15 years, largely exclusive of short-term fallows or similar (i.e., regrowth of ≤~2–3 years; Table S1).

Fig. 1: Estimates of the temporal persistence of tropical natural reforestation.
figure 1

Notes: Estimates for each region or country are detailed in Table S1. Various estimates were re-scaled to 15-year periods. The large standard deviation for the Neotropics estimate reflects variability amongst spatial units of observation during a single period. In contrast, ranges and standard deviations for other estimates reflect either variability amongst temporally successive estimates at large scales or, in the case of ‘Brazilian Amazon (2)’, whether or not very young forest regrowth was included in the estimate.

Full size image

This 29-61% rate of regrowth clearance applies to forest regrowth of Year 1 of a 15-year time series, and would decline exponentially with each subsequent year of the series, that is, with increasing regrowth age. Upon re-scaling this rate for each year of a 15-year series according to the relationship between regrowth clearance and regrowth age observed annually for the Brazilian Amazon23, it is estimated that ~17–47% of all regrowth arising throughout a given 15-year series would be discounted for not occurring as of Year 15 specifically. Incorporating this fraction of regrowth into the midpoint estimate of 31.6 Mha of natural reforestation by Fagan et al.14 would increase this extent and correspondingly reduce the nominal ratio of planted-to-natural reforestation area, from 1.02 ± 0.6 to 0.7 ± 0.2 (Table 1 Adjustment 1), qualifying any relative prevalence of planted reforestation. This adjusted ratio similarly considers the midpoints of the confidence intervals for natural and planted reforestation, and its variability owes exclusively to the range of the persistence adjustment parameter in Table 1. A complementary adjusted ratio that instead averages this adjusted midpoint ratio with the four analogous ratios based on the extremes of the confidence intervals would be very similar, at 0.8 ± 0.4, as it would be for all other adjustments discussed below (Table 1, Fig. S1, and Supplementary Note 1). Ratios based on midpoints are therefore discussed exclusively below to focus on the implications of the various adjustment parameters.

Table 1 Nominal ratio of planted-to-natural tropical reforestation, 2000–2015, adjusted according to estimates of reforestation persistence and additionality
Full size table

The additionality and form of planted reforestation

Tropical planted reforestation is generally more robust to adjustments for persistence. Notwithstanding plantation harvesting and re-planting cycles at ~10–40 year intervals, depending on the commodity, the ephemerality of planted reforestation is relatively low. Ephemerality is discouraged by the substantial ‘up-front’ economic investment required for planted tree/trop-crop production37, which may not be profitable until years after a plantation maturation, itself typically requiring years38. Of greater relevance to the reported relative prevalence of planted reforestation is its ‘additionality’, that is, the degree to which reforestation represents new, ‘additional’ tree cover established on non-forest lands, rather than tree cover established over existing forest or on lands deforested just before reforestation. Additionality has been broached only partially in the land-change literature concerning reforestation due to a longstanding limited availability of suitable long-term satellite data16. Limited exceptions for regions of extensive planted tree cover3,18,39,40,41,42,43 suggest that plantation reforestation is often of moderate additionality and so belies illusory processes of forest-cover gain.

A review of satellite estimates of tropical planted reforestation in relation to prior natural forest provides high-confidence insights for Latin America, Equatorial Asia, and mainland Southeast Asia. To qualify for review, estimates had to be of a regional scale or greater, directly observe both planted and natural tree cover to ensure consistent estimates of land change, and observe change over at least 15 years at multiple intervals to avoid the undue influence of specific intervals of heightened forest conversion. Also, estimates had to pertain to planted tree cover generally, without selective focus on specific tree commodities, to avoid any undue influence of ‘boom’ commodities disproportionately driving forest conversion in particular places and times12,44. Commodity-specific estimates were considered as supplementary, not definitive. Collectively, these criteria ensured that the ten estimates reviewed here are more authorative, comprehensive, and generalisable than those from myriad local case studies. Estimates exist for Brazil, Indonesia, and Malaysia, which account for ~85% of the tropical planted reforestation of interest, as well as Southeast Asia, accounting for a further 5% and host to a doubtless greater share of total tropical planted tree cover.

Accounting for planted reforestation additionality would appreciably reduce the nominal ratio of planted-to-natural reforestation. In the Brazilian Atlantic Forest, where Brazilian planted reforestation concentrates largely as timber/fibre stands45, ~25–50% of the planted area established during 2000-2019 was forest 1–5 years prior, depending on observation period3. In Borneo, 49% of fibre/timber and oil-palm plantation area established during 1973–2015 entailed forest conversion within five years before establishment40. Further, some 92% of such forest-conversion occurred less than one year before plantation establishment during 2001–201746. Whereas the 49% rate of Bornean forest-to-plantation conversion is nearly twice that estimated for oil palm exclusively in Kalimantan (Indonesian Borneo) over 1995–2015 at five-year intervals47, at 26%, it is in keeping with oil-palm specific estimates for Sumatra (46%), both of which are likely under-estimates given the simplicity of related forest delineations. For Indonesia, Brazil, and Malaysia, estimates of forest-to-plantation conversion over 1990–202013 amount to 51%, 17%, and 53% of the nominal area of planted tree cover as of 201416, respectively (Supplementary Note 3). Again, these rates are likely under-estimates given that the nominal planted areas in question include ample ancillary non-planted covers. For Vietnam, 23% and 10% of plantation establishment during 1990–2020 entailed forest conversion within 5 years and 1 year before establishment, respectively, according to analysis of the annual land-cover classifications of Phan et al.30,35 (Supplementary Note 4). Supporting these Vietnamese estimates, a recent assessment of Southeast Asian rubber42 – often regarded as a forest-friendly tree commodity48 – reports that one fifth of its current (c.2022) area underwent forest conversion over 2001–2016. This figure rises as high as 40% in Cambodia42, where plantation expansion is dominated by rubber and accounted for ~30% of all deforestation during 2000–201544. However, across large areas of Southeast Asia, a much greater 70–83% of rubber but also eucalyptus, coffee, and cashew stands established during 2000–2014 entailed forest conversion within three years of establishment, according to analysis of the land-cover classifications of Hurni et al.43,49,50 (Supplementary Note 4). If conservatively omitting these latter high estimates as potential regional aberrations, the estimates above would suggest that ~50–75% of planted reforestation is ‘additional’. Upon discounting the corresponding non-additional fraction of planted reforestation from the midpoint estimate of planted reforestation by Fagan et al.14, the nominal ratio of planted-to-natural reforestation would decline to from 1.02 ± 0.6 to 0.6 ± 0.1 (Table 1 Adjustment 2), or more precipitously to 0.4 ± 0.1 if also adjusting for natural reforestation persistence (Table 1 Adjustment 3).

Discussions of pantropical planted-forest additionality, or indeed of a pantropical predominance of planted reforestation, risk simplifying considerable variability amongst commodities (e.g., timber/pulp, rubber, oil palm) and the world regions in which they concentrate. The separate consideration of major commodities is, therefore, prudent and, as shown below, illustrative of the relevance of the aspects of reforestation processes discussed here. Crucially, at 48% of the midpoint estimate of tropical planted reforestation (Supplementary Note 5), oil palm in Equatorial Asia (Indonesia and Malaysia) is both central to any pantropical prevalence of planted reforestation and is amongst the least additional forms thereof. For these reasons, as well as oil palm’s close association with agro-industrial activities51, any conflation of oil palm with reforestation (e.g., refs. 52,53,) has proven controversial7,54 (Supplementary Note 5). Clearly, experimentally excluding Equatorial Asian oil palm from consideration on these grounds would mean that planted reforestation would be far less prevalent relative to natural reforestation pantropically. Notably, however, doing so after the aforementioned adjustments for planted reforestation additionality and natural reforestation persistence would reduce the ratio of planted-to-natural reforestation only slightly, from 0.4 ± 0.1 to 0.3 ± 0.1 (Table 1 Adjustment 4), underscoring the relevance of these two factors in determining prevailing reforestation processes.

The discussion above does not advocate for the exclusion of oil palm or similar agro-industrial plantations from consideration (e.g., coffee, cocoa, rubber). On the contrary, the recent and planned7 expansion of planted reforestation is seemingly a bellwether of emergent processes of tropical reforestation characterised by ‘forest’ (agro)industrialisation in managed landscapes more than by humans’ withdrawal from agricultural margins16,55,56,57. To afford insight into such emergent processes, future pantropical appraisals of reforestation should explicitly discriminate agro-industrial tree stands, at least collectively and apart from natural reforestation with which they are often confused10, and ideally for major commodity groups (e.g., palms, timber/pulp, rubber) and land-change histories (e.g., establishment date, prior land use/cover). Recent advances in remote sensing mean that such goals are now feasible at large geographic and temporal scales10,11,13,43,44,58,59,60, if not variously achieved already, albeit on a piecemeal basis such the totality of emergent reforestation processes remains undescribed.

The additionality of natural reforestation

Contrary to prior adjustments of the ratio of planted-to-natural reforestation, which were invariably downwards (Table 1), a more narrow accounting of natural reforestation as that resultant of land disuse or similar transitions away from agriculture, grazing, or similar non-forest land uses would buoy assertions of a relative prevalence of planted reforestation. Much natural reforestation apparent to satellite imagery is, in fact, seemingly also not additional, insofar as it is an often-fleeting byproduct of dynamic land use and related to forest degradation and conversion3,26,28. Hence, one analysis of the satellite data underlying Fagan et al.14 identified significant associations between deforestation and reforestation cross-nationally5, with natural reforestation found extensively in deforestation hotspots (e.g., the Brazilian Arc of Deforestation, the Guatemalan Petén region, the eastern Congo Basin). With respect to reforestation processes, such reforestation may arguably be discounted as ‘spurious’ or illusory, that is, as a land cover vestigial of processes of land change fundamentally opposed to longer-term forest gain or even stabilisation.

Reliable large-scale estimates of the additionality of tropical natural reforestation are exceedingly rare, however, challenging appraisals of the nature and even the extents of reforestation processes. Only recently has the land-change community seriously begun to consider the additionality of natural reforestation and so parse reforestation into realms of, on the one hand, novel land-use transition on disused lands and, on the other, generic biophysical dynamics resultant of forest degradation and conversion3. To be clear, all tropical reforestation is technically ‘non-additional’ insofar as all lands were originally forested at some earlier time; yet a degree of additionality or, perhaps better termed, novelty, can be inferred proportional to the period of time separating current reforestation from prior forest conversion. Only recently has a single pantropical satellite-based analysis by Vancutsem et al.13 differentiated natural afforestation over long-cleared lands from forest regrowth following forest disturbance, offering unique insight into the additionality of natural reforestation. A direct reading of this analysis13 suggests that 25% of all tropical natural reforestation (i.e., regrowth plus afforestation) as of 2020 was additional, that is, occurred over lands cleared of forest as of 1990 or earlier (afforestation), rather than after forest disturbance since 2000 (regrowth) (Supplementary Note 6). This estimate is conservatively low, however, since it excludes young afforestation (3–9 years old) as well as old forest regrowth following forest disturbance during the 1990s. Incorporating these two extents yields a revised estimate of 55% additionality for 1990–2020 (Supplementary Note 6), again taking afforestation as additional and regrowth as not. Both of these estimates remain approximate, however, considering that their forest regrowth as classified by Vancutsem et al.13 allows between three and up to 16 and 26 years between forest disturbance and the commencement of regrowth, respectively – broad ranges within which some regrowth is arguably also additional.

Scrutiny of the number of years separating forest disturbance from subsequent forest regrowth in the annual 1990–2020 land-cover data61 of Vancutsem et al.13 affirms the limited additionality of regrowth but also identifies a fraction that is additional (Supplementary Note 6). Pantropically, of all forest regrowth commencing during 1994–2018 and persisting to 2020, 47%, 70%, and 89% commenced ≤5 years, ≤10 years, and ≤15 years of prior forest disturbance, respectively (Fig. 2b). Disproportionately large shares of this regrowth commenced two and three years after forest disturbance (Fig. 2a), consistent with arguments that much regrowth reflects iterative processes of forest turnover or destruction3. Such figures describe Latin America and Asia equally well (Fig. 2), with each region having similar extents of regrowth and accounting for 90% of all tropical regrowth collectively. If qualifying forest regrowth as additional when it commenced >10 years after forest disturbance – a somewhat arbitrary threshold but one still suggestive of land disuse and largely exclusive of ‘spurious reforestation’ discussed above – then the original and revised estimates of 25% and 55% natural reforestation additionality would increase to 48% and 69%, respectively (Supplementary Note 6).

Fig. 2: Forest regrowth occurring since 1990 and persisting to 2020, by number of years since prior forest disturbance.
figure 2

a Percentage of forest regrowth, b Cumulative percentage of forest regrowth. Notes: Values on the x-axis report the number of years between the third, qualifying year of forest disturbance and the first year of subsequent forest regrowth. Forest disturbance is defined as a transition from forest to non-forest cover, where the latter persists for at least 3 consecutive years. Forest regrowth is defined as forest cover occurring over at least 3 consecutive years following forest disturbance and persisting to 2020. Regrowth persisting to 2020 therefore occurred over 2018–2020 at the latest, but typically commenced earlier during the 1990-2020 observation period. Such definitions are as per Vancutsem et al.13 and largely exclude rapid annual transitions between forest, deforestation, and regrowth characteristic of agricultural activity, but do not exclude all agricultural-driven natural reforestation. See Supplementary Note 6 for details.

Full size image

These estimates of 48–69% natural-reforestation additionality would buoy a pantropical ratio of planted-to-natural reforestation, potentially considerably. The adjusted ratio would rise slightly, from 0.3 ± 0.1 to 0.5 ± 0.1, in conjunction with all prior adjustments to the ratio, including the exclusion of oil palm (Table 1 Adjustment 5a). The ratio would rise more notably to 0.8 ± 0.3 if retaining oil palm as planted reforestation (Table 1 Adjustment 5b). This latter ratio lends partial support to notions of a narrow gap between the prevalence of natural and planted reforestation processes pantropically. Any ‘estimated relative dominance’ of plantation reforestation would, however, likely be specific to relatively confined regional contexts, such as those where additional natural reforestation is rare and planted reforestation is at least partially additional and enduring. The degree to which such contexts account for all apparent reforestation is presumably minor, but ultimately uncertain, given the myriad ways in which reforestation might be parsed into ‘actual’ versus ‘spurious’ realms. As suggested below, whether reforestation should even be so parsed at all is increasingly debatable.

Outlook on observing reforestation processes

The estimates by Fagan et al.14 have illuminated the human and biophysical dimensions of tropical reforestation more fully by discriminating its planted and natural forms. Like most such large-scale studies, their estimates were bound by the particularities of available satellite data as much as by the ideals of conceptual models of land change. This situation is arguably characteristic of the land-change literature, as reflected by its historical conflation of, on the one hand, reforestation as a readily observable but generic biophysical dynamic of forest disturbance and, on the other hand, reforestation as a novel and rare land-use transition towards forest stabilisation and gain3.

Notwithstanding the uncertainty surrounding the estimates of planted and natural reforestation areas by Fagan et al.14, the adjustments to their nominal ratio afford insights into the relative prevalence of certain aspects of reforestation. As a biophysical dynamic, reforestation appears substantially characterised by recurrent natural regrowth generally lacking persistence and, to a lesser degree, additionality. Still, the majority of natural reforestation is at least minimally additional, contrary to some expectations3, suggesting a degree of novelty. Otherwise, as a novel and enduring land-use transition, reforestation may indeed be widely characterised by planted reforestation, often lacking additionality but being relatively persistent. Any such transitions are, however, likely confined, geographically and contextually, and are not necessarily aligned with net reforestation at large temporal scales16,55. Much hinges on the factors of reforestation reviewed here, particularly the additionality of natural reforestation, being the most neglected in the literature and the only factor buoying the ratio of planted-to-natural reforestation. Even an appreciable relative prevalence of planted reforestation as suggested here (Table 1 Adjustment 5b) would bend forest-transition theory62,63 further towards a forest-transformation model framing tropical reforestation as a process of intensifying tree-cover management rather than one of land-use contraction16,64,65.

While persistence and additionality are useful lenses to scrutinise areal estimates of reforestation at the end of a given period, they would prove blunt instruments to describe underlying reforestation processes throughout a period, particularly longer periods of 30+ years that are now readily observable9,13,30. Ultimately, persistence and additionality bear on the ‘spuriousness’ of reforestation as two sides of the same temporal coin, i.e., as ‘years reforested before deforested’ and ‘years deforested before reforested’, respectively. In the context of fitful processes of forest turnover, clearance, and reforestation over multiple decades, the distinction between persistence and additionality blurs, while iterant, ‘spurious’ reforestation becomes relevant to ultimate reforestation dynamics. Attempts to define ever more precise temporal thresholds defining persistence and additionality, and ultimately to separate ‘genuine’ from ‘spurious’ reforestation, would be fraught where interest concerns reforestation processes. Most any threshold would exclude reforestation of interest and, ultimately, limit knowledge of how, where, and whether reforestation has occurred and endured, including by repeated regeneration, to shape emergent forest landscapes. A more nuanced approach to observing forest gain while qualifying its extent on the basis of temporal dynamics is warranted and, for the first time, readily possible.

With the recent advent of the pantropical, annual, Landsat-derived Tropical Moist Forest forest-change dataset spanning 32 years and counting13,61, complementing similar data for South America9,23,66,67 and recent mapping of tropical planted reforestation9,11,12,13,14,30,68, surveys of tropical reforestation now finally can, and should, move away from areal estimates of planted and natural reforestation as terminal land-cover classes as of some current year, regardless of whether qualified by persistence and additionality. Instead, drawing upon these new data, satellite surveys of reforestation may be recast as spatio-demographic surveys substituting terminal forest classes for temporal ‘life cycle’ dynamics, land-change transitions, and outlooks for reforestation observed over all patches or pixels of all ages, now treated as individuals of a dynamic population18,19,23. The logic of such an approach is evident to any demographer appraising two human populations of equal size but divergent demographic trends, such as those of Ethiopia and Japan (Fig. 3a). Here these two populations are analogous of two currently equal extents of reforestation undergoing divergent reforestation processes, with patches of one being young, multitudinous, short-lived, and expansive, and patches of the other being stable, older, and poised for contraction. As reforestation processes, neither successive estimates of reforestation area (total population) nor thresholds for persistence (i.e., population above a given age) or additionality (i.e., parentage, fertility rate) could account for the demographic divergence apparent between Fig. 3b and a. Hence, ideally, future surveys of reforestation would parallel human demographic modelling of systematic phases of population growth, stabilisation, and decline69 based on shifting demographic composition (Fig. 3c), as defined by interactions between birth rates (rates of reforestation establishment), infant mortality (frequency of young regrowth re-clearance and age at re-clearance)20, longevity (persistence of reforestation once established), death rates (rates of re-clearance of matured reforestation), migration and parentage (time between reforestation and prior forest disturbance, type of prior forest), amongst other factors. Such a longitudinal, spatio-demographic approach to reforestation would seek to profile the changing nature of reforestation over time, of which its nominal area at a given time is but one facet. The overwhelming tendency to report net change to nominal forest area as of the end of a given period, rather than to tease apart the spatio-temporal processes qualifying that area over a period20, has meant that the frontier in this domain of land-change science has advanced only partially relative to early interests, namely knowledge of the rhythms, rates, and means by which haphazard reforestation congeals to yield an enduring forest gains, now variously defined by planted, natural, recurrent, and persistent components.

Fig. 3: A model of ‘demographic’ phases of natural reforestation inspired by stages of human population change.
figure 3

a, b Changes to the age structure to nominally equivalent but divergent human populations. c The demographic transition model describing phase of human population change. d Potential ‘demographic’ phases of natural reforestation for the Brazilian Amazon. Notes and Sources: Panels a and b show human demographic age structures. Japanese and Ethiopian populations are equivalent in 2023, but not 1950. Data are from the UN81. Panel c shows human demographic trends stylised to reflect fundamental concepts of the demographic transition model69. Trends in births, deaths, growth (natural increase), and population are based on illustrations in Nielsen82. Births, deaths, and growth are expressed as rates, i.e., people per 100 people. Longevity (life expectancy) and infant mortality are expressed as percentage year-on-year change, i.e., growth or decline relative to the previous year. A given longevity and infant mortality value of >0 and <0 therefore describes an increasing or decreasing trend, respectively. Longevity and mortality trends are styled but reflect those of Japan in terms of relative magnitude and timing, after UN81. Note the different scaling of longevity and infant mortality. Panel d shows stylised reforestation trends and phases that reflect interpretations of annual data presented by Nunes et al.23 for the Brazilian Amazon, 1985–2017, in terms of relative magnitudes and timing. Regrowth persistence and young regrowth re-clearance are percentage year-on-year changes analogous to longevity and infant mortality in c. Labels 1–4 highlight key ratios of contemporaneous trends, which likely vary systematically during reforestation. These key ratios are those comparing (1) reforestation establishment against reforestation re-clearance, (2) reforestation persistence against young reforestation re-clearance, (3) reforestation establishment against forest gain, and (4) reforestation establishment against persistence. Neither Stages I–IV nor the dynamic they describe in d are directly related causally to the demographic trends or stages in c cf83,84.

Full size image

The promise of a spatio-demographic approach to qualifying the nominal areas and predominant forms of reforestation is suggested by annual observations of secondary forests for the Brazilian Amazon, 1985–201723. These observations are arguably the only of their kind for a major tropical region prior to Tropical Moist Forest dataset61. Speculatively, based on these observations, a spatio-demographic dynamic for natural reforestation may resemble that of Fig. 3d, in terms of the relative magnitudes and timing of key trends. Observations for the Brazilian Amazon suggest that increases to natural reforestation area owe more to steady increases to secondary-forest persistence, particularly amongst secondary forest >5 years old, than to increasing rates of secondary-forest establishment, being comparatively flat. Seemingly paradoxically, secondary forest re-clearance rates also increase, chiefly amongst secondary forest <10 years old, paralleling establishment rates and reflecting a growing preference for secondary-forest conversion amongst landholders36. Such a sequence of trends suggests the four stages of natural reforestation in Fig. 3d, whereby an accumulation of secondary-forest area during Stage IV largely reflects a ‘compounding’ effect of progressive increases to the persistence of secondary-forest established during earlier stages, rather than increased establishment rates or decreased conversion rates. For Brazil and other tropical regions, tracking systematic changes to four potentially key ratios of spatio-demographic reforestation trends, labelled 1–4 in Fig. 3d, may prove especially indicative of reforestation stages or, indeed, prevailing reforestation processes.

A lack of long-term annual observations of tropical planted reforestation, including by the Tropical Moist Forest dataset, will generally confine such spatio-demographic analysis to natural reforestation only. Exceptions exist for most of South America and Vietnam, however, having 31–38 years of annual data on planted and natural reforestation9,30,66,67,70,71,72,73,74,75,76,77,78,79, and arguably for Indonesia, having 23 years of such data80. Developing long-term annual observations of tropical planted reforestation generally is therefore a priority.