Carbon storage through China's planted forest expansion

Forests are major carbon sinks and key to mitigating climate change. Globally, despite increases in planted forests, net forest area declined by 178 million ha between 1990 and 2020 and global forest carbon stocks fell from 668 to 662 Pg. Credible monitoring of planted forest carbon storage requires distinguishing planted from natural forests due to differences in species composition, stand structure, age, and management. China hosts over a quarter of the world’s planted forests and has expanded forest area through policy-driven conversion of croplands, shrublands, and grasslands. As of 2020, China’s forests cover ~220 million ha (~5% of global forest area). This expansion has reshaped land use/land cover and carbon storage capacity. National plans target ~26% forest cover by 2050, entailing further land conversion. However, a comprehensive quantification of the spatiotemporal dynamics and associated carbon storage of China’s planted forests has been lacking, motivating this study to map planted forests at high resolution over three decades and assess their carbon storage changes.

Existing national-scale planted forest maps for China have limited spatial and temporal coverage. Products derived from inventories or digitized maps are typically coarse or limited to specific years or subtypes, hindering multi-decadal tracking of planted forest expansion. Prior work often lacked the resolution to separate planted from natural forests across time, and could not capture land conversion pathways or age-structured carbon density changes. These gaps necessitate high-resolution, wall-to-wall, long time series mapping integrated with field data to assess carbon storage dynamics of planted forests.

The study focused on planted forests (trees >5 m height and >15% canopy cover) versus natural forests (naturally regenerated). All available Landsat 4/5/7/8/9 surface reflectance data (447,730 scenes, cloud cover <30%) from 1988–2021 were processed on Google Earth Engine. Five-year composites for 1990, 1995, 2000, 2005, 2010, 2015, and 2020 were generated using year ±2-year stacks and CFMASK to remove clouds/shadows/snow. Over 600,000 vegetation survey samples (2003–2022) were compiled from crowdsourcing, field surveys, and national forest inventory; through spatial filtering for unchanged forest masks (CLCD 1990–2020) and NDVI criteria (minimum NDVI ≥0.6), 124,407 planted/natural forest samples were retained (70% train, 30% validate). Planted forest mapping leveraged 220 features: 11 spectral bands, 5 vegetation indices (NDVI, EVI, BSI, SAVI, MSAVI), 198 GLCM texture features, 8 temporal harmonic features (magnitude, phase, amplitude, RMSE for NDVI/EVI), and 3 topographic variables (elevation, slope, aspect). Recursive feature elimination with cross-validation selected optimal regional feature subsets. Region- and period-specific Random Forest classifiers produced quinquennial planted forest maps. Validation used independent samples, systematic random sampling, comparison with National/Provincial statistics (2000–2020), and a digitized map from the Seventh National Forest Resource Inventory (2004–2008). Spatiotemporal analyses aggregated 30 m maps to 0.1° grids to assess area change, gains/losses, and trends (linear regression with t-tests). Land-use/cover conversions to planted forests were derived using CLCD for each period, classifying single- versus multi-change events. Carbon storage was estimated using a carbon density method: China’s 1:1,000,000 vegetation map classified planted forests into 17 types; seventh national forest inventory (treated as 2005) provided species and age to build age-dependent carbon density datasets via species-level age–biomass equations and a 0.5 biomass-to-carbon factor. For 1990–2000, ages were back-calculated; for 2010–2020, unchanged plots were identified via NDVI/forest masks; newly established 2005–2020 stands used a space-for-time substitution (ages 1–15 in 2005) for carbon density. Total carbon storage per period was the sum over types of density times area. Validation against field-based carbon density datasets (circa 2010) and high-resolution biomass maps (2019/2020) yielded R2 = 0.50–0.68. Uncertainty was quantified using standard deviations of type-specific carbon densities to produce bounds on total carbon storage. Sources of error were documented for mapping and carbon estimation.

- Planted forest area nearly doubled from 464,715 km² (1990) to 903,099 km² (2020), a 94.33% net increase at an average expansion rate of 14,613 km² per year. Total gains were 479,681 km² and losses 41,297 km², for a net increase of 438,384 km². - Map accuracy: overall accuracies ranged 77.3%–81.8% across years. Comparisons with National Forest Inventory and National Forestry Statistical Yearbook showed strong agreement (R² = 0.8–0.9, P < 0.01). - Spatial patterns: 2020 planted forests were mostly below 1,500 m elevation and concentrated in the South (32.6%), East (20.1%), and Southwest (16.6%); Northwest had the smallest share (5.5%) but fastest relative growth (+266.5% since 1990). At the grid scale, 91% of changed pixels showed positive trends (84% significant at P < 0.01). - Carbon storage increased from 675.6 ± 12.5 Tg C (1990) to 1,873.1 ± 16.2 Tg C (2020), an average increase of ~40 Tg C per year. Growth was slower pre-2005 (~20 Tg C/a) and accelerated post-2005 (~60 Tg C/a), peaking at ~73 Tg C/a (2005–2010). - Regional carbon gains (1990–2020): South +386.4 ± 20.6 Tg C, East +272.6 ± 18.1 Tg C, Southwest +237.9 ± 15.8 Tg C; Northeast +103.1 ± 20.4 Tg C with an initial decline (1990–1995); North from 43.3 ± 4.2 to 147.8 ± 5.0 Tg C; Northwest from 29.2 ± 2.6 to 122.2 ± 4.2 Tg C. - Attribution to LULC conversion: 438,787 km² (98.3% of area increase) came from conversions of cropland, shrubland, grassland, and natural forest. These conversions contributed 637.2 ± 5.4 Tg C (~53.2%) of total planted forest carbon increase. Period contributions varied (e.g., 81.4% in 1990–1995; 17.2% in 2000–2005). By source (1990–2020): cropland-to-planted forest +191.7 ± 2.6 Tg C; shrubland +176.4 ± 2.3 Tg C; grassland +135.9 ± 2.0 Tg C; natural forest +121.7 ± 3.5 Tg C. - Regional LULC contributions: grassland conversions dominated in Northwest (34.5%) and North (25.9%); cropland and natural forest conversions were important in Northeast (27.4% and 19.1%) and East (11.9% and 13.0%). In South and Southwest, cropland and especially shrubland conversions (Southwest shrubland 25.3%) were major drivers. - Drivers and programs: Cropland-to-forest conversions reflect the Grain for Green program (initiated 1999). Other programs (Three-North Shelterbelt and Shelterbelt Development in Five Regions) contributed to expansion. - Young and middle-aged stands contributed substantially to rapid carbon gains post-2005; eastern, southern, and southwestern regions accounted for ~75% (896.9 Tg C) of total planted forest carbon by 2020, aided by favorable climate and restoration efforts.

The study demonstrates that China’s policy-driven expansion of planted forests has substantially increased national biomass carbon storage since 1990, particularly after 2005. High-resolution quinquennial maps show that increased carbon storage aligns with spatial expansion patterns, with the largest gains in southern, eastern, and southwestern regions due to favorable climatic conditions, active restoration programs, and conversions from cropland and shrubland. While land-use conversions explain slightly over half of the national carbon increase, growth of young planted forests also made significant contributions, especially in the East, South, and North, where growth accounted for roughly 47%–60% of regional increases. Despite gains, slower carbon accumulation in northern and northwestern regions reflects water stress and ecological constraints; rapid afforestation in these areas may pose risks to water availability and other ecosystem functions. The authors note remote sensing detection lags (time to develop canopy cover) and mapping uncertainties but argue that the integrated remote sensing and field-based approach yields robust national estimates consistent with past studies. The findings underscore the importance of continuing strategic, well-managed planted forest expansion and tending, balancing carbon goals with water and ecological considerations, to support China’s carbon neutrality objective by 2060.

China’s planted forest area nearly doubled from 1990 to 2020, driving a substantial increase in biomass carbon storage from ~676 to ~1,873 Tg C. More than half of the carbon gain is attributable to land-use conversions (primarily cropland, shrubland, and grassland to planted forest), with the remainder from growth of existing and newly established plantations. Expansion accelerated after 2005, likely reflecting policy impacts and the maturation of young stands, with the South, East, and Southwest contributing the majority of gains. Given that many plantations are still young or middle-aged, considerable additional carbon accumulation is expected, supporting China’s pathway toward carbon neutrality by 2060. Future work should optimize spatial targeting, species selection, and management to maximize carbon benefits while safeguarding water resources and broader ecosystem services.

- Mapping limitations: reliance on 5-year composites due to cloud cover; potential omission of small patches; possible detection lag of newly established forests (3–7 years to reach detectable canopy); feature selection may omit influential predictors; only Random Forest used (other algorithms might perform better); training/validation data include crowdsourced samples with potential label noise; NDVI-based sample filtering susceptible to noise. - Vegetation typing and carbon density: use of a historical 1:1,000,000 vegetation map may misclassify some types, though vegetation distributions are relatively stable and >80% agreement with 2020 updates; forest inventory treated as 2005 can introduce 1–3-year age errors; space-for-time substitution for young stands may overestimate carbon density for newly established forests after 2005. - Uncertainty quantification: total carbon storage uncertainty derived from type-level standard deviations; mapped uncertainty indicates predominantly low-to-medium uncertainty nationally, with higher uncertainties in parts of the Southwest and Northeast.