Carbon sequestration potential of tree planting in China

China has implemented ambitious tree-planting programs since the 1970s (e.g., Three-North Shelter Forest, Natural Forest Protection, Yangtze River Shelter Forest) to reduce erosion and improve ecological conditions, raising forest cover from <13% in 1978 to ~23% in 2019. To support the national pledge of carbon neutrality before 2060, targets seek 26% forest cover by 2035 and 30% by 2050. However, past efforts often faced low survival and growth due to planting in unsuitable areas (e.g., ~15% survival in arid/semi-arid regions during 1949–2005), undermining carbon benefits and causing ecological issues (e.g., water stress, erosion). Remaining practical land for afforestation is limited: climatically marginal non-forested areas are scarce and suitable lands often coincide with croplands, constraining reforestation due to food security. Many existing forests are sparse (over 26% with <500 trees/ha), suggesting potential for densification. The study aims to rigorously quantify where and how to plant trees—via afforestation and densification—to maximize carbon sequestration, while assessing economic costs and policy-aligned, realistic implementation pathways.

The paper reviews evidence that tree planting is a cost-effective, nature-based climate solution and summarizes China's extensive planting efforts and policies. It highlights shortcomings of earlier projects—poor pre-planning, low survival in harsh environments (e.g., Ningxia projects), and mismatch between planted areas and environmental suitability—leading to reduced carbon benefits and environmental side effects such as increased drought and soil erosion. Planted forests are often short, sparse, and scattered with lower carbon sequestration compared to natural forests. Historical data indicate afforestation areas cumulatively exceeded extant forested and deforested areas combined, reflecting vast but inefficient efforts. Literature also indicates regional water constraints (e.g., Loess Plateau) and that natural forests generally sequester more carbon with lower water consumption than planted forests. Policy definitions of forest cover (UNFCCC 10–30% tree cover thresholds) and national forest classifications (MODIS PFTs) inform the study’s operational definitions and scenario analyses.

Overview: The study develops a machine learning-based framework to estimate tree growth suitability (TGS), link suitability to potential tree density by forest type, and quantify carbon sink potential and costs under multiple afforestation and densification scenarios, including a policy-constrained realistic scenario.

Tree Growth Suitability (TGS): Environmental covariates include five climate variables (annual solar radiation, aridity index, annual maximum and minimum land surface temperature, annual snow cover index), nine soil variables (bulk density, coarse fragments, organic carbon density, pH, sand, silt, clay, cation exchange capacity, total nitrogen), and three topographic variables (elevation, roughness, slope). All inputs were resampled to 1 km. Labels were derived from Hansen Global Forest Change (30 m): grids with >30% tree-covered pixels since 2000 labeled as 1 (suitable), grids with no tree pixels since 2000 labeled as 0 (unsuitable); grids between these thresholds or with >30% cropland/urban (FROM-GLC 2017) were excluded from training. A neural network with two hidden layers (64 and 32 neurons) trained via stochastic gradient descent outputs per-pixel suitability probabilities as TGS scores (0–1). Five-fold cross-validation achieved average accuracy 0.98 and F1-score 0.98 on training/validation; test accuracy 0.91 and F1-score 0.88. External validation used 184 natural forest plots, 586 plantation plots, and 816 forest parks; overall classification accuracy ~0.8, higher for natural forests.

Linking TGS to tree density: The global tree density map (1 km) was used to empirically relate TGS to tree density for four forest types from MODIS MCD12Q1 PFTs: deciduous broadleaf (DB), deciduous needleleaf (DN), evergreen broadleaf (EB), evergreen needleleaf (EN). TGS scores were binned into 100 intervals (0.01 width). For each forest type and TGS bin, empirical cumulative distribution functions (CDFs) of observed tree densities were constructed (densities vary and generally increase with TGS, with distinct distributions per type).

Scenario design (idealized): For every pixel, if current tree density is below the 25th, 50th, or 75th percentile of its forest-type-specific CDF in its TGS bin, density is increased to that quantile. Pixels with current density = 0 are treated as afforestation; pixels with density > 0 as densification. Residential and agricultural pixels retain original state. For afforestation pixels, the forest type is assigned from the nearest labeled-1 forest pixel. Outputs include spatial maps of afforestation/densification potential and tree numbers at 1 km.

Carbon sink estimation: Aboveground and belowground forest biomass carbon (AGFBC/BGFBC) maps (300 m, aggregated to 1 km) were statistically related to tree density within each TGS bin by dividing tree densities into 20 intervals (increment of 35 trees/ha) and computing median, 5th, and 95th percentile carbon values per interval. AGFBC and BGFBC generally increase monotonically with density across TGS bins. For afforestation, carbon gain equals the median carbon corresponding to the target quantile density; for densification, gain equals the difference between median carbon at the target quantile and at the current density. Uncertainties use the average standard deviation derived from the 5th/median/95th estimates per grid. Robustness was assessed using two additional biomass datasets, yielding broadly consistent spatial patterns and proportional totals across scenarios.

Realistic scenario: The National Zoning Program of Ecological Functions (NZPEF) constraints were applied: prohibit planting in key water-conserving areas and in pixels where increased density would reduce carbon stocks; plant to 75% quantile in forest ecological subzones; to 50% in the monsoon zone of Eastern China; and to 25% in arid Northwestern and Tibetan Plateau subzones, reflecting ecological stress considerations.

Forest stock and soil carbon: Forest timber volume increases were back-calculated from carbon increments using a biomass expansion factor (BEF) parameterization, relating carbon increments to timber stock. Soil organic carbon (SOC) increment was estimated as 0.14 times the sum of AGFBC and BGFBC gains.

Economic analysis: Historical investments (1998–2010) were used to derive average per-hectare and per-tree costs for different regions (e.g., ~$4.37 per tree in Northwestern China; ~$1.17 per tree in Southeastern China). Scenario total investments were computed by multiplying per-tree costs by the number of additional trees. The average cost per tonne of CO2 sequestered accounted for AGFBC, BGFBC, and SOC increments, using the CO2-to-C mass ratio. Costs were summarized nationally and regionally and compared between realistic and higher-intensity (75% quantile) scenarios.

• Suitability and density: TGS spatial patterns align with existing tree density, decreasing from southeast to northwest. Within similar TGS ranges, tree density and AGFBC/BGFBC increase monotonically with density, though variability exists among forest types and bins.
• Afforestation potential areas: Remaining afforestation areas cluster in the Hengduan Mountains, middle/lower Yangtze River Plain, and Qinba Mountains. Under the 25%, 50%, and 75% quantile scenarios, afforestation could cover ~58.2, 63.2, and 70.0 million ha, respectively.
• Trees added (ideal scenarios): Afforestation could add 10.1±0.02, 17.2±0.12, and 25.2±0.29 billion trees at the 25%, 50%, and 75% quantiles; densification could add 7.8±0.001, 21.1±0.02, and 37.5±0.07 billion trees, with densification exceeding afforestation at higher quantiles.
• Forest stock increases (ideal scenarios): Densification increases forest stock by 1.3±0.1, 3.7±0.4, and 7.9±0.7 billion m³ at 25%, 50%, 75%; afforestation by 1.7±0.1, 3.1±0.2, and 4.5±0.3 billion m³.
• Carbon sinks (ideal scenarios): Aboveground gains (TgC) from afforestation are 891.0±58.4, 1311.8±82.8, 1866.9±110.1; from densification 711.4±55.7, 1782.7±169.4, 3604.6±286.4. Belowground gains (TgC) from afforestation are 333.8±101.6, 450.4±112.7, 576.8±125.7; from densification 314.1±14.7, 631.8±26.3, 1029.8±43.0. Some pixels may show AGFBC decreases with densification (e.g., older complex stands in Changbai, Xiaoxinganling, Qinba, Hengduan), indicating cases where intervention is not advisable.
• Realistic, policy-aligned scenario outcomes: Total of ~44.7 billion trees (18.3 billion afforestation; 26.4 billion densification). Afforestation spans ~55.8 million ha (~6% of China’s land). Forest stock increases by ~9.6 ± 0.8 billion m³. Combined biomass carbon gain ~5.9 ± 0.5 PgC (AG ~4.5 ± 0.3 PgC; BG ~1.4 ± 0.1 PgC). Average annual sink from 2025–2060 is ~169.1 ± 11.7 TgC/yr (55.5 ± 4.1 TgC/yr from afforestation; 113.5 ± 7.5 TgC/yr from densification). Soil organic carbon increases by ~0.3 ± 0.01 PgC (afforestation) and ~0.5 ± 0.03 PgC (densification). Regional hotspots include Hengduan Mts (afforestation) and Qinba/Yunnan-Guizhou (densification); Northeast contributes less to gains.
• Cost-effectiveness: Estimated total investment $116.1 billion; average cost $4.77 per tCO2 nationally. Costs are lowest ($1.65/tCO2) in the Sichuan–Chongqing area, middle/lower Yangtze, North China Plain, and South China, and higher ($17.1/tCO2) in Northwestern regions. A higher-intensity implementation (75% quantile everywhere) increases sink to ~8.07 ± 0.45 PgC (≈+1.26 PgC over realistic) at an additional $47.46 billion ($34.90/tCO2).
• Definitions sensitivity: Using forest cover thresholds between 10–30% (UNFCCC range) maintains densification’s dominance over afforestation in potential; only much higher thresholds (>30%) reverse this balance by reclassifying more pixels as non-forest.

Findings indicate that densifying existing forests can yield substantial, often greater, carbon gains than afforestation while minimizing conflicts with agricultural land. The framework identifies high-potential regions for both strategies and quantifies carbon returns under policy constraints, showing that large-scale planting can sustain and extend China’s forest carbon sink toward 2060. The analysis underscores that not all pixels should be densified—older, carbon-rich natural forests with complex structures may lose aboveground carbon if artificially densified—so site-specific ecological considerations remain critical. Robustness checks across multiple biomass datasets show consistent spatial patterns and proportional estimates, supporting the methodological validity. The cost analysis demonstrates that strategically focusing efforts in regions with favorable hydrothermal conditions yields low sequestration costs, making densification particularly attractive economically. The work aligns scenario design with national ecological zoning, providing a realistic pathway and actionable guidance for program implementation.

The study introduces a comprehensive, data-driven framework to target tree planting for maximum carbon sequestration in China, integrating suitability modeling, density–carbon relationships by forest type, and policy-aware scenario design. It shows that, under a realistic scenario, approximately 44.7 billion trees could be planted, increasing forest stock by about 9.6 ± 0.8 billion m³ and sequestering ~5.9 ± 0.5 PgC, with densification delivering a larger share of carbon gains at lower average costs. The approach highlights priority regions and demonstrates that large-scale tree planting—especially densification—can sustain and extend China’s forest carbon sink toward carbon neutrality targets. Future research should incorporate higher-resolution ecological data, dynamic growth and risk models, and refined economic analyses to optimize species selection, spacing, risk mitigation (fire, pests), and long-term management, thereby improving the effectiveness and resilience of tree-planting programs.

• Spatial resolution: The 1-km grid cannot capture fine-scale heterogeneity, microclimates, or exact tree placements, potentially affecting suitability and carbon estimates.
• Statistical approach: Pixel-based statistical relationships may overlook key drivers (species composition, physiology, canopy structure), leading to occasional non-monotonic carbon responses to increasing tree numbers.
• Ecological dynamics: Competition, self-thinning, and increased disturbance risks (fire, disease, pests) at higher densities could limit realized gains and threaten carbon and biodiversity.
• Economic assumptions: Costs are derived from historical afforestation projects and may not reflect densification-specific methods, inflation, evolving labor/material costs, or implementation losses.
• Data dependence: Carbon potential estimates depend on the stability and accuracy of biomass datasets; while robustness checks show consistency, uncertainties remain across sources.
• Policy/definitions: Different forest cover thresholds alter afforestation vs densification classification, though overall planting potential is similar; practical implementation must adhere to local ecological and regulatory constraints.