Restoring a patch of degraded land is usually judged by what happens inside the project boundary. A new continent-scale analysis suggests that this may miss a large part of the environmental outcome.
Researchers examining restoration across sub-Saharan Africa found that natural regeneration and actively managed restoration were associated with lower tree loss not only inside restored areas, but also across surrounding landscapes. Agroforestry, however, showed a strikingly different pattern in the available data, with higher tree loss emerging within projects and nearby areas over time.
The study, published in Nature Sustainability on 30 September 2026, assembled 129,982 restoration polygons covering about 6.8 million hectares across 27 African countries. The researchers then compared changes in tree-cover loss inside restoration sites and in surrounding buffers with counterfactual areas that had not yet been restored or were statistically matched to resemble the treated landscapes.
The results challenge a simple assumption behind restoration policy: adding trees in one place does not guarantee that the net landscape effect will be positive. What happens to land use beyond the project can reinforce the benefits, or offset them.
Restoration can move environmental pressure rather than eliminate it
Governments, conservation organisations and carbon markets increasingly treat ecosystem restoration as a major tool for addressing climate change and biodiversity loss. The Bonn Challenge alone seeks to bring 350 million hectares of degraded and deforested land into restoration by 2030, while sub-Saharan Africa accounts for a substantial share of global restoration commitments.
Yet restoring land can change how people use neighbouring land. If a project replaces productive farmland, timber production or other income-generating activity, some of that activity may simply move elsewhere. Researchers call this negative leakage. A restoration project could therefore look successful within its mapped boundary while indirectly contributing to additional clearing nearby.
The reverse is also possible. Restored landscapes can supply fuelwood, fodder or other resources, create employment, improve agricultural conditions or strengthen conservation practices. In that case, pressure on nearby tree cover may fall. This is positive leakage, where the benefits extend beyond the area formally counted as restored.
Xinran Miao and colleagues set out to measure these spillovers systematically across sub-Saharan Africa rather than evaluating projects only from the inside.
Nearly 130,000 restoration polygons were mapped
The researchers combined four geospatial data sources to identify 129,982 restoration polygons established between 1996 and 2023. Nearby polygons were grouped into 19,229 restoration clusters. Together, the mapped projects covered roughly 6.8 million hectares across 27 countries.
Natural regeneration represented 56.1% of the restoration area, or about 3.8 million hectares. This category included passive recovery and assisted natural regeneration using relatively low-intensity interventions. Active restoration, involving more intensive management such as tree planting, accounted for 40.2%, or about 2.7 million hectares. Agroforestry represented the remaining 3.8%, approximately 0.26 million hectares.
The geographic distribution was uneven. Kenya, Ethiopia and Cameroon together accounted for more than 85% of the restoration area represented in the dataset. The average project had been under restoration for 6.7 years by 2025, which is important because many ecological responses develop over much longer periods.
For the causal analysis, the team focused on restoration clusters located in forests or other woody landscapes and examined annual tree loss within the projects and in areas extending five and ten kilometres around them.
Two counterfactual strategies were used. One compared restored sites with areas that would be restored later, allowing those future projects to serve temporarily as untreated controls. The other matched restoration sites with areas that were never treated but had similar biophysical and socioeconomic characteristics. The researchers then applied a heterogeneity-robust difference-in-differences framework to estimate how tree loss changed relative to those controls.
Natural regeneration produced benefits well beyond the boundary
Natural regeneration showed the clearest evidence of positive spillovers.
Using areas scheduled for later restoration as the comparison, natural-regeneration sites experienced 6.8 percentage points less tree loss over the ten years following restoration. Tree loss was also 6.5 percentage points lower within five kilometres and 8.7 percentage points lower within ten kilometres.
The alternative matched-control method produced a smaller reduction inside the project boundaries, at 2.3 percentage points, but similarly substantial reductions outside them. Tree loss was 5.6 percentage points lower in the five-kilometre buffer and 8.9 percentage points lower across the ten-kilometre buffer.
That spatial pattern matters. If restoration were evaluated only by counting trees inside formal project boundaries, the wider reduction in tree loss could remain invisible in carbon accounting or conservation assessments.
Active restoration also generally produced positive leakage. Under the matched-counterfactual analysis, tree loss over ten years was 3.6 percentage points lower within restored areas, 4.7 percentage points lower within five kilometres and 6.0 percentage points lower within ten kilometres.
The researchers suggest several plausible mechanisms. Successful projects may strengthen conservation behaviour, improve land tenure or encourage neighbouring communities to participate. Restored areas may also provide fuelwood, fodder and other products that reduce pressure on remaining natural forests. Employment, subsidies and carbon payments could further reduce incentives for clearing.
These explanations were not directly tested, however. The study emphasises that governance, tenure security, participation and benefit sharing are likely to influence whether these mechanisms actually operate.
Agroforestry showed the opposite pattern
Agroforestry did not follow the same trajectory in this dataset.
Using the future-treatment comparison, tree loss after ten years was 7.9 percentage points higher within agroforestry projects, 10.7 percentage points higher within five kilometres and 7.5 percentage points higher within ten kilometres than in the controls.
The matched-control estimates were smaller but still showed increases of 4.8 percentage points within projects and 3.5 percentage points within the five-kilometre buffer after ten years. The ten-kilometre matched comparison showed a temporary reduction around six years after restoration rather than the same long-term increase.
The authors caution against interpreting this as evidence that agroforestry is inherently damaging. Agroforestry systems differ enormously, and research from other regions has found reductions in surrounding deforestation. In the African projects studied here, one possibility is that integrating trees into productive cropland can eventually reduce usable agricultural area or yields through shading and competition, creating pressure to clear land elsewhere. Commercial tree crops can also alter land values and labour incentives.
The timing supports the possibility of delayed trade-offs. Elevated tree loss around agroforestry sites generally became more apparent several years after implementation, rather than immediately.
The same restoration strategy did not produce the same result everywhere
Country-level differences were substantial, reinforcing the danger of treating restoration as a uniform intervention.
Kenya showed positive leakage across restoration approaches and spatial buffers in the country-level analysis. Uganda also showed a strong reduction in tree loss within areas undergoing active restoration, estimated at 10.2 percentage points after five years.
Other countries moved in the opposite direction. In Ghana, increased tree loss was detected in the ten-kilometre buffer for natural regeneration and active restoration. Ethiopia showed higher tree loss within and near active-restoration areas after five years. South Africa stood out with a particularly large estimated increase in tree loss within active-restoration areas, although country-specific estimates must be interpreted in light of differences in project composition, sample availability and local conditions.
Across countries, the direction and magnitude of leakage varied from substantial reductions to increases. The study therefore points toward institutional context as part of the restoration outcome rather than a background detail.
Carbon markets may need to look outside the project map
The findings have direct implications for how governments, conservation funders and carbon standards measure success.
A project that reduces clearing ten kilometres beyond its boundary may deliver more climate and biodiversity value than conventional project-level accounting recognises. Conversely, a project that protects or restores trees inside its boundary while shifting clearing nearby could receive more credit than its net landscape effect warrants.
The researchers argue that leakage safeguards should therefore be integrated into restoration planning, certification standards and incentive systems. This requires monitoring beyond project borders and for long enough to detect effects that emerge only after several years.
Land tenure is especially important. Unclear ownership can encourage opportunistic clearing, while secure rights and meaningful community participation may make long-term protection more attractive. The same applies to benefit sharing. Restoration that imposes local opportunity costs without providing viable alternatives may create pressures that a tree-counting exercise cannot see.
A large dataset still has important limits
The scale of the analysis is a major strength, but it does not make the estimates universal.
The georeferenced restoration database is incomplete because many projects lack publicly available or independently verifiable spatial information. Coverage was also uneven across countries, meaning that the pooled results should not be interpreted as a complete census of African restoration.
The satellite-based tree-loss dataset has limitations in landscapes with low canopy density and small-scale disturbances. The analysis measured spatial spillovers only out to ten kilometres, so displacement operating across larger regional or international markets could not be detected.
The researchers also could not isolate detailed differences within each restoration category, such as planting density, species choice, land-use history or specific implementation practices. In addition, most projects were relatively young, leaving fewer observations at longer post-restoration periods. Small deviations from the assumptions required for the difference-in-differences design were also observed at some short time scales.
These limitations make the study more useful as evidence about broad landscape patterns than as a verdict on any individual project.
Restoration success is bigger than the restored site
The central lesson is not that one restoration label is always good and another always bad. It is that restoration changes a landscape of ecological, economic and social relationships.
Natural regeneration and active restoration were associated, on average, with meaningful reductions in tree loss extending beyond formal project boundaries. That is encouraging evidence that restoration benefits can spread through a landscape rather than stopping at a mapped edge.
At the same time, the agroforestry results show why policymakers cannot assume that more trees inside a project automatically mean less environmental pressure overall. If restoration changes livelihoods, land values, production or access to resources, those consequences can appear kilometres away and years later.
For restoration programmes measured in millions of hectares and billions of dollars, the unit of success may therefore need to expand from the project to the landscape around it.
Source Information
Study title: Restoration-driven positive and negative leakage in sub-Saharan Africa
Authors: Xinran Miao, Oscar Morton, Christopher G. Bousfield, Leland K. Werden, Stephen M. Thomas, Thomas W. Crowther, Casey M. Ryan and colleagues
Journal: Nature Sustainability
Published: 30 September 2026
