Introduction

Nature-based climate solutions1, such as forest conservation2, restoration3,4, and sustainable management5, offer a promising approach to mitigate the effects of global climate change, conserve biodiversity, and enhance rural livelihoods6. By sequestering carbon in terrestrial ecosystems, forest landscape restoration can yield substantial co-benefits for biodiversity and ecosystem services and is often a no-regret investment7. Land use projects, which are mostly forestry projects, issued approximately half of all credits from 2000 to 2021 on the voluntary carbon market8 and have featured prominently in many Nationally Determined Contributions to the Paris Agreement1. Landscapes undergoing tree cover restoration are often a heterogeneous mosaic of various restoration approaches, including natural forest regrowth, planted forests for conservation purposes, commercial plantations, and agroforestry6. The relative impacts of these different land use strategies can be highly variable for biodiversity, climate, and human wellbeing6. Different restoration strategies are being used depending on site conditions, local opportunities, and needs, often necessitating trade-offs among conservation and production goals9. As tree plantations comprise nearly half of the restoration area pledged by over 60 nations to the Bonn Challenge10 and have different environmental outcomes compared to naturally regenerated forests9, it is critical to distinguish forest types when monitoring forest landscape restoration, assessing their socio-ecological determinants and outcomes, and evaluating their climate mitigation potential11,12,13,14.

Here we use an annually resolved 30-m resolution tropical moist tree cover change dataset15 developed by the European Commission’s Joint Research Centre (JRC) and a 100-m global forest management type dataset16 available for 2015 to distinguish tree cover gain types on former agricultural lands (croplands and pasturelands) in the global moist tropics. We intersect the 30-m tropical tree cover gain dataset with the 100-m global forest management layer to attribute tropical tree cover gains on former agricultural lands during 1982–2015 to the expansion of natural forest regrowth and three managed tree systems: timber plantations, oil palm plantations, and agroforestry (Fig. 1).

Fig. 1: Moist tropical tree cover restoration areas and types on former agricultural land in 2015.
figure 1

The continental maps were aggregated from 100 m to 10,000 m resolution for visualization. Three sites a–c were selected to show the heterogeneity of the landscape in the original 100-m resolution where white color areas represent other land cover types (Table 1). The map in the original 100-m resolution can be viewed and downloaded via Google Earth Engine36 (https://code.earthengine.google.com/a8ab0a204422bdaf13bd1eff4bc0a5ea). The basemap in the three sites is from Google Maps.

Full size image

Results

Tropical moist forest restoration patterns

Examining tree cover gains on former agricultural lands across the entire tropical moist forest region, we estimate around 27% ± 2.6% of the tree cover gain in this region to be managed tree systems, whereas 56% ± 3% is due to natural forest regrowth (Table 1). The 27% estimate for managed tree systems is conservative relative to the range estimated by Fagan et al. 17, 34% to 68%, as we did not consider tree cover expansion in tropical dry forests or tropical grasslands, savannas, and shrublands17. The remaining 17% ± 3% of tree cover gain on former agricultural lands in the JRC’s 30-m resolution dataset occurs in locations classified as non-forest land cover types by the 100-m forest management type dataset (Table 1), including cropland, pastureland, grassland, shrublands, and water bodies. This area likely represents small patches of unmanaged trees within these predominantly non-forest land covers.

Table 1 Restoration types in moist tropical tree cover gains that occurred on former agricultural land during 1982–2015
Full size table

Focusing on three major subareas of the entire tropical moist forest region—the Amazon, Borneo, and Central Africa— natural forest regrowth accounts for 62% ± 3.3% of tree cover gains on former agricultural lands (Table 1). Timber plantations, oil palm plantations, and agroforestry account for 0.2% ± 0.1%, 2% ± 0.9%, and 17% ± 1.9% of the gains, respectively, indicating that managed tree systems are a substantial part of moist tropical tree cover gains even in regions dominated by natural tree cover. Borneo’s gain has a much larger percentage of managed tree systems (51% ± 8.1%) than did the Amazon’s (16% ± 1.9%) or Central Africa’s (14% ± 1.6%). In Amazon and Central Africa, oil palm plantations represent tinier fractions (60-fold and 400-fold smaller, respectively) of tree cover gain than in Borneo, where they are one-third as widespread as recovering natural forest.

Drivers of tropical carbon recovery rates

A previous study, Heinrich et al. 18 used the same moist tropical tree cover gain dataset from JRC and an observation-based biomass product to assess rates and drivers of aboveground carbon accumulation in tropical recovering forests. They found that regeneration rates in Borneo were around 45% and 58% higher than in Central Africa and the Amazon, respectively, in the first 20 years of recovery. This difference was attributed solely to climatic and topographical factors. However, the large percentage of managed tree systems in Borneo can have either positive or negative effects on landscape carbon accumulation rates, hinging on the species that were planted and their growth rates6. Moreover, some areas of natural forest regrowth in Borneo have undergone assisted restoration practices like climber cutting and enrichment planting19, which significantly accelerate aboveground carbon recovery compared to unassisted natural regeneration19.

We applied the Global Forest Model (G4M)20 to investigate to what extent regional variations in forest carbon accumulation rates can be attributed to natural conditions (climate, soil, and topography) or if management regimes need to be taken into account as well. The G4M simulations showed that, if we exclude managed tree systems, regional differences in secondary/degraded forest growth rates are substantially lower than found by Heinrich et al., with natural regrowth rates in Borneo only 10% higher than in Central Africa (compared to 45% in Heinrich et al.) and only 13% higher than in the Amazon (compared to 58% in Heinrich et al.). These results suggest that differences in restoration types and management practices might be strong drivers of remotely sensed geographic differences in tropical carbon recovery rates21. We posit that landscape restoration types and forest management practices are at least as important drivers of regional differences in regrowth rates as continent-scale differences in climate and topography. Furthermore, ignoring differences between the drivers of the expansion of plantations and agroforestry versus natural forest regrowth may compromise the identification of priority areas for promoting the expansion of natural forests to conserve biodiversity and mitigate climate change22,23.

Discussion

“Agroforestry”—the largest contributor to managed tree cover gain on former agricultural lands—encompasses a heterogeneous range of managed tree systems (Table 1). Our use of the term follows the definition in the global forest management layer and includes: (1) fruit trees; (2) tree shelter belts and small forest patches; (3) sparse trees in cropland and pasture; (4) shifting cultivation; and (5) trees in urban/built-up areas. These different agroforestry systems differ in carbon accumulation rates and co-benefits for people and biodiversity24,25,26. Additionally, a typical biomass pixel (e.g., 100 m) in a remotely sensed representation of an agroforestry landscape could contain a significant signal from herbaceous crops and pastures. As a result, the remotely sensed carbon accumulation rates in agroforestry landscapes may differ substantially from the rates in natural secondary forests27.

A higher proportion of agroforests in a study area may additionally have important implications for long-term carbon permanence. Establishing or enhancing tree cover on open farmland may increase net carbon storage24. However, thinning or clearing of forest to establish an agroforestry system could cause carbon losses24, especially when agroforestry includes slash-and-burn practices that are among the factors explaining the reduced longevity of naturally regenerating tropical forests22,28. High uncertainty remains regarding which agroforestry actions provide mitigation and how to reliably track progress of agroforestry toward being a natural climate solution24.

The 17% ± 3% of post-agricultural land area classified as tree cover gain in the JRC forest cover change dataset but not as natural forest regrowth or a managed tree system according to the global forest management dataset could include land where unmanaged forest regrowth has partially occurred but is hindered by factors such as invasive grasses, vines, shrubs, or ferns. Such land is unlikely to have accumulated much carbon, and management interventions would be required to accelerate forest recovery and carbon accumulation. The accurate delineation and management classification of such land is key, because its unintended and invisible inclusion in remote sensing analyzes of recovering forest underestimates the carbon sink potential of lands actually returned to forests of one kind or another.

The 100-m 2015 global forest management layer is currently the only available global product characterizing forest management types in a moderate-resolution manner16,29. We urge the remote sensing community to map not only where forests and other tree systems are being restored but also what types of tree systems are being restored11. This mapping effort should encompass not only the moist tropics, which we have focused on due to data availability, but also the dry tropics, subtropics, and temperate and boreal zones, which collectively account for even more of the world’s forest biome area and are home to much more of the planet’s human population30.

The needs and priorities of local communities and national aspirations dictate the appropriate land management and restoration measures to be taken. Tree plantations and agroforestry may be locally appropriate choices, particularly when biophysical or socio-economic conditions do not support natural regeneration31. These market-driven tree systems can be especially valuable when payments for ecosystem services offered by governments or other organizations are either nonexistent, which is currently the case across most of the moist tropics, or not high enough to offset the costs (opportunity, implementation, maintenance) of natural forest regeneration32.

Land management planners, investors, and implementers need rigorous monitoring, reporting, and verification systems to account for the environmental, ecological, and socioeconomic trade-offs of different forest restoration approaches9. Recent concerns about the over-crediting issues in Reducing Emissions from Deforestation and Forest Degradation (REDD+) projects have created a lack of confidence in nature-based carbon credits2,33. Although less criticism regarding project monitoring has been directed at forest restoration activities, termed Afforestation, Reforestation, and Revegetation (ARR) in the carbon market, the dialog surrounding REDD+ and the shift it has brought to the sector should serve as a cautionary tale, highlighting the need for careful progress to successfully scale up ARR activities. Moreover, ARR projects present their own set of unique challenges, particularly around the monitoring of diverse types of tree cover restoration and their subtle annual changes in carbon stocks. Distinguishing and disaggregating forms of tree cover that represent different tree management systems, not simply capturing the area of tree cover gain, would enable these systems to enhance the integrity of ARR credits in the carbon market34. This information may be critical for improving confidence in forest-featured Nationally Determined Contributions in the United Nations Framework Convention on Climate Change’s Global Stocktake, and enhancing compatibility with the biodiversity targets of the Kunming-Montreal Global Biodiversity Framework.

Methods

Primary datasets

The global forest management layer was created at a 100-m resolution for the year 2015 with good overall accuracy (>82%) using time series from PROBA-V satellite imagery combined with unique reference samples16. It characterizes forest management classes such as intact forests, managed forests with natural regeneration, planted forests, plantation forest (rotation up to 15 years), oil palm plantations, and agroforestry.

The European Commission JRC tropical moist forest cover change dataset was created at a 30-m resolution over the period 1982–2022 using 41 years of Landsat time series15. It characterizes undisturbed tropical moist forest, degraded tropical moist forest, deforested land, forest regrowth, and permanent and seasonal water in its Annual Change Collection. It separately identifies agricultural lands (croplands and pasturelands) as a land cover type on which observed tree cover gain occurs.

We intersected the 30-m tropical tree cover gains (i.e., the “forest regrowth” category in the JRC Annual Change Collection) with the 100-m global forest management layer to attribute tropical tree cover gains on former agricultural lands during 1982–2015 to the expansion of natural forest regrowth and three managed tree systems. A map of tree cover gain types on former agricultural lands in the global tropics was finally generated in a 100-m resolution, and the area (in million hectares, Mha) of each tree cover gain type was extracted. The map reflects annual changes throughout the period, not just the difference between 1982 and 2015, and it is net of reversals out of tree cover. For example, a tree cover gain of X million hectares that occurred during Year t to Year t + 1 but experienced a subsequent cumulative loss of Y million hectares (Y < X) during Year t + 1 to 2015 is measured as a net gain of X–Y million hectares.

Accuracy assessment

We conducted an independent accuracy assessment of the 100-m tropic tree cover gain type map by using the methodology set out in Olofsson et al. 35. It allows the 95% confidence intervals to be estimated and the area estimates to be adjusted based on the error matrix. Using the mapped classes as strata (natural forest regrowth, timber plantations, oil palm plantations, agroforestry, and other land cover), we applied a random stratified sampling design to create 460 sample pixels in total, with a targeted overall accuracy of 75%. The sample size allocated to each class was determined by the targeted user’s accuracy for that class. To create the reference classification for labeling each sample pixel, we used a combination of Landsat data from the USGS open archive, together with historical images in Google Earth. The error matric of sample counts and proportional area is presented in Tables 2 and 3. We also combined timber plantations, oil palm plantations, and agroforestry as “managed tree systems” and created an error matrix (Table 4), which shows a robust accuracy of the map of managed tree cover gains.

Table 2 Description of sample data as an error matrix of sample counts
Full size table
Table 3 The error matrix in Table 2 populated by estimated proportions of area
Full size table
Table 4 The error matrix after consolidating timber plantations, oil palm plantations, and agroforestry as “managed tree systems”
Full size table

Tropical carbon recovery rate simulations

The G4M20 (https://iiasa.ac.at/g4m) is a biophysical forestry model developed at International Institute for Applied Systems Analysis, which is used in many projects to inform European Commission on carbon sequestration, carbon stock, and harvest potential on different climate and management scenarios. The G4M estimates forest productivity based on dynamic site characteristics such as monthly temperature, precipitation, radiation, and CO2 concentration, semi-dynamic factors including water holding capacity and soil depth, as well as nitrogen, phosphorus, salinity, and pH values, and static attributes like air pressure. The model is calibrated using net primary production and biomass observations.