Tropical forest restoration under future climate change

The study addresses whether carbon sequestered by tropical forest restoration will remain durable under future climate change. While restoring and preserving tropical forests is a key nature-based solution to help limit warming below 2°C, the permanence of restored carbon is threatened by climate-driven stressors such as rising temperatures, drought, insect outbreaks, and wildfire. Prior estimates of tropical restoration potential often assume present-day climate or apply simple scaling to temperature and CO2, leaving uncertainties about future risks to carbon permanence, particularly in seasonally arid tropics where disturbance-driven turnover may increase. The authors aim to quantify how future climate change, atmospheric CO2 trajectories, and wildfire affect carbon storage from restoration across the tropics, and to identify strategies to prioritize restoration areas for long-term, cost-effective carbon storage.

Background work highlights tropical restoration as an effective climate mitigation with co-benefits to biodiversity and communities, and less albedo-related warming than high-latitude forests. However, permanence risks from disturbances (fire, drought, insects) are projected to increase across the tropics. Many existing estimates of restoration potential either ignore dynamic future climate and CO2 changes or use simplified scaling approaches. Dynamic global vegetation models (DGVMs) differ substantially in their sensitivity to climate and CO2 and in mortality and productivity processes, contributing to uncertainty. Observational studies indicate saturation or decline of intact tropical forest carbon sinks under future climate, potentially offsetting restoration gains. Cost, socio-political constraints, and prioritization strongly influence feasible restoration area and outcomes. This study builds on these insights by using a DGVM to explicitly test CO2 fertilization assumptions, climate scenarios, wildfire effects, and prioritization strategies, and by benchmarking the model against an ensemble of DGVMs and observations.

• Study design: Conducted 221 simulations with the LPJ-LMfire dynamic global vegetation model (DGVM) to evaluate carbon uptake and permanence from tropical forest restoration between 2020 and 2100 under varied climate, CO2, wildfire, and prioritization scenarios.
• Restoration area: Used a published global restoration opportunity map (excluding croplands) masked to tropical/subtropical forest ecoregions (Ecoregions2017). This yielded 128 Mha of restorable non-crop agricultural land (classified as pasture in land-use datasets). The restoration mask was aggregated to 0.5° grid as fractional area per cell.
• Land-use scenarios: Historical land use (to 2014) from CMIP6 was held fixed after 2014 in control runs. In restoration runs, cropland remained at 2014 extent; pasture held constant through 2020 and then linearly reduced by the restoration area from 2020 to 2030. From 2030 to 2100, land use remained at the reduced level. Restoration moved land from “managed” to “recovering” tiles in LPJ-LMfire (fire allowed on recovering/unmanaged, excluded on managed). Carbon accounting used the recovering tile (above- and below-ground plus soil carbon).
• Climate forcing: Bias-corrected monthly fields from 13 CMIP6 GCMs, concatenating historical (1850–2014) and four SSPs (2015–2100: SSP1-26, SSP2-45, SSP3-70, SSP5-85). Required variables: 2 m air temperature, daily min/max temperature, total and convective precipitation, cloud cover, and 10 m wind speed. Anomalies relative to 1971–1990 were applied to observed climatologies.
• CO2 experiments: (1) Transient CO2 (CO2): atmospheric CO2 follows CMIP6 historical and SSP trajectories. (2) Fixed future CO2 (CO22014): CO2 follows historical to 2014, then fixed at 399 ppm to remove additional CO2 fertilization after 2014. This brackets upper (with fertilization) and lower (without additional fertilization) bounds of carbon uptake.
• Fire experiments: For each CO2 and SSP combination, simulations were run with and without lightning-caused wildfire (anthropogenic ignitions not included) to quantify wildfire effects.
• Initialization and runtime: Each simulation spun up for 1020 years with 1850s climate and CO2, then transient climate/CO2 from 1850–2100. Restoration implemented during 2020–2030; outcomes evaluated through 2100.
• Prioritization (half-area scenarios): Recognizing limited feasibility of full restoration, scenarios selected 64 Mha (half of 128 Mha) by: (a) maximizing carbon uptake potential (top 50% by 2100 carbon density under default conditions), (b) minimizing opportunity cost (USD ha−1 yr−1 of land, labor, management), or (c) combined cost-per-carbon (ratio of cost to total uptake). These selected masks were then evaluated under climate and CO2 scenarios. Additional prioritization included climate-informed selection by using 2100 carbon metrics under climate change (CO2 or CO22014) while retaining the same cost layer.
• Restoration opportunity index: For each of 13 GCM realizations and each SSP in CO22014, grid cells were filtered to those with net carbon gain (2030–2100) and no >10% temporary reductions relative to the detrended 2030–2100 mean. Within this mask, a rank-sum of (high-to-low) carbon uptake and (low-to-high) opportunity cost was computed per realization, summed across GCMs, and normalized to 0–1. Values near 1 indicate robust, cost-effective, and climatically resilient restoration opportunities across models.
• Model evaluation and representativeness: LPJ-LMfire performance compared to observations and TRENDYv9 DGVM ensemble for biomass, productivity, and sensitivities to CO2, temperature, and drought; found broadly representative though with higher-than-observed accumulation rates in some regions.
• Outputs: Cumulative and annual carbon uptake, spatial distribution of carbon gains, contributions of CO2 fertilization, climate, and wildfire to changes, prioritization impacts, and opportunity index maps.

• Magnitude of carbon uptake: Restoring 128 Mha (11% of tropical non-crop areas) yields 24.1–39.6 Pg C cumulative uptake (above- and below-ground including soil) between 2020 and 2100, depending on climate severity, CO2 fertilization, and wildfire. Under the default case (no future climate change, no additional CO2 fertilization, no wildfire), cumulative carbon is 28.5 Pg C (range 25.0–29.6 Pg C) by 2100.
• Temporal dynamics: Carbon generally continues accumulating through 2100; in 92% of simulations, carbon increases to century’s end. Uptake is strongest during restoration area expansion (2020–2030) and declines thereafter, especially without CO2 fertilization.
• Effect of CO2 fertilization and climate (unconstrained CO2, “CO2”): Carbon storage increases versus default in proportion to SSP atmospheric CO2 levels. Relative increases by 2100: SSP1-26 +23.4% (0.5–44.5%), SSP5-85 +39.4% (26.6–67.7%), with SSP2-45 and SSP3-70 intermediate.
• Effect of climate alone (fixed CO2, “CO22014”): Differences reflect climate impacts only. Relative to default by 2100, carbon is most reduced under SSP5-85 (−7.3%; −16.5 to +11.7%) and least reduced under SSP1-26 (reported as +13.1%; −9.8 to +32.5%); other SSPs fall between. Including wildfire further reduces storage versus default: −15.1% (−27.4 to +1.7%) in SSP5-85 and −3.0% (−17.7 to +9.6%) in SSP1-26.
• Factor contributions: Across SSPs, CO2 fertilization dominates increases under high-CO2 pathways (e.g., +42.0% relative to default under SSP5-85), while climate change contributes most under low-CO2 pathways (e.g., +14.6% relative to default under SSP1-26). Wildfires account for ~10.6–14.1% of changes in carbon storage.
• Prioritization (64 Mha): Selecting the top 50% by carbon density achieves 68.9% of the carbon gained by restoring all 128 Mha; choosing the cheapest 50% achieves 56.4%; combined cost+carbon selection approaches the carbon-maximizing outcome. Under transient CO2 and climate, carbon increases across all prioritizations (smallest gains under SSP1-26, largest under SSP5-85). Without CO2 fertilization (CO22014), carbon is lower than default across prioritizations with reductions scaling with climate severity (smallest under SSP1-26, largest under SSP5-85). Accounting for climate change in the prioritization step increases carbon storage relative to present-day-based selection in all scenarios; gains are largest when prioritizing by carbon and smallest when prioritizing by cost.
• Restoration opportunity index: 5–7% of the restored 128 Mha has high opportunity index (>0.75) across SSPs in CO22014 (7% in SSP1-26, 6% in SSP2-45, 5% in SSP5-85), indicating robust, cost-effective, and climatically resilient carbon storage potential through 2100. Central Africa and New Guinea score highest under SSP2-45; promising regions include northwestern South America, West and Central Africa, the maritime continent, and parts of Southeast Asia. Southwestern Brazil, India, and parts of Southeast Asia (e.g., south China) face higher climate and cost barriers.
• Aggregate significance: Restoring 11% of tropical non-crop land can sequester the equivalent of roughly 2–4 years of current anthropogenic CO2 emissions over multiple decades, with most stored carbon preserved through 2100 across climate and CO2 scenarios.

The simulations indicate that tropical forest restoration can deliver durable carbon sequestration through 2100 under a broad range of plausible future climates. Even when isolating climate impacts without additional CO2 fertilization, most scenarios maintain net positive carbon storage, and in the vast majority of runs carbon continues accumulating to century’s end. CO2 fertilization substantially enhances outcomes under higher-CO2 pathways, while climate impacts alone impose modest reductions, largest under SSP5-85. Wildfire reduces storage but does not negate the overall benefit; its contribution to changes is secondary to CO2 and climate effects. Strategic prioritization makes restoration more efficient: restoring only half the area can deliver 56–69% of the full carbon potential, and climate-aware site selection further improves permanence and yields. Regions identified with high restoration opportunity suggest where investments could maximize robust, cost-effective carbon storage. These findings support the hypothesis that, despite climate risks, restored tropical forest carbon is largely preserved, thereby contributing meaningfully to mitigation portfolios alongside rapid emissions reductions. The study also highlights that model assumptions (e.g., CO2 fertilization, fire regimes, nutrient constraints) and Earth system feedbacks can modulate absolute benefits, underscoring the need for careful policy design, adaptive management, and further multi-model assessments.

Restoring tropical forests on a relatively small fraction of tropical non-crop land (≈128 Mha) can sequester tens of petagrams of carbon by 2100, with most of this carbon remaining stored across diverse future climates. CO2 fertilization amplifies gains under higher-emissions pathways, while climate-only effects still retain substantial carbon. Prioritizing by carbon potential and/or cost enables achieving a majority of the full-storage outcome with half the area, and incorporating climate resilience into site selection further boosts permanence. The work provides policy-relevant guidance on where restoration could be both robust and cost-effective (e.g., central Africa, New Guinea), while identifying regions with higher climate or cost barriers. Future research should: expand to multi-model DGVM ensembles; incorporate coupled land–atmosphere feedbacks and full Earth system responses; represent anthropogenic fire and other disturbances; integrate nutrient limitations; evaluate biodiversity outcomes and adaptive management (e.g., species diversity, drought tolerance); and address socioeconomic constraints and governance to improve real-world feasibility and permanence.

• Model scope: Results are from a single DGVM (LPJ-LMfire); DGVMs vary in climate/CO2 sensitivities and mortality/productivity processes, potentially affecting quantified outcomes.
• CO2 fertilization and nutrients: The upper-bound transient-CO2 simulations do not reduce atmospheric CO2 in response to added land sinks, likely overestimating fertilization effects; nutrient (N, P) constraints on fertilization are not explicitly represented.
• Fire representation: Only lightning-caused wildfire included; anthropogenic ignitions omitted. This may underestimate fire impacts in some areas, though LPJ-LMfire may overestimate burned area in seasonally dry regions due to fuel moisture biases.
• Land–atmosphere coupling: Simulations are offline; biogeophysical feedbacks from increased tree cover (e.g., localized tropical cooling) are not coupled and could slightly increase carbon uptake relative to standard GCM climates.
• Earth system feedbacks: Reduced atmospheric CO2 from restoration would dampen oceanic and terrestrial sinks elsewhere, lowering net global CO2 removal relative to stand-alone vegetation responses.
• Land-use trajectories: Post-2014 land use is fixed (control) or stylized for restoration; results are not estimates under specific SSP land-use storylines.
• Recruitment assumptions: Assumes best-practice restoration ensuring successful recruitment; real-world failures due to climatic/socioeconomic factors may reduce realized uptake.
• Model biases: LPJ-LMfire shows higher-than-observed accumulation rates in some regions (e.g., northern South America, south China) and underestimates in others (e.g., southeastern Brazil).
• Temporal horizon: Simulations end in 2100; post-2100 fate of restored carbon is inferred only from trends, with potential declines under severe climate change beyond 2100.