Introduction

Forest restoration, as a crucial nature-based solution, plays a crucial role in addressing global climate change and enhancing terrestrial carbon storage1,2,3,4,5. Numerous ambitious forest restoration initiatives have been proposed worldwide to increase forest area through afforestation and natural regeneration5,6,7,8,9. For example, the Bonn Challenge launched by the Government of Germany and IUCN aims to restore 350 million hectares of degraded and deforested landscapes globally by 203010. However, variations in natural conditions and human disturbances lead to differences in the process of forest restoration and community structures between afforestation and natural regeneration11. For example, natural processes, such as wildfire occurrence, drought stress, and pest outbreaks, can alter species composition and reduce tree density, thereby influencing the speed and trajectory of natural forest recovery6,12. In contrast, human disturbances, including logging, agricultural expansion, and urbanization, can cause habitat fragmentation, soil degradation, and loss of seed sources, resulting in slower or structurally different restoration processes compared to planted forests13. These factors ultimately determine distinct carbon sequestration potentials and ecological functions of afforested and naturally regenerated forests14,15,16,17. Therefore, understanding the aboveground carbon (AGC) accumulation capacity of different forest restoration measures and their relationships with environmental and anthropogenic conditions are essential for formulating effective restoration strategies and enhancing carbon sequestration potential.

Over the past 30 years, China has significantly expanded its forest area through extensive forest management projects, defying the global trend of decreasing forest cover18,19. Statistical data indicate that since the 1970s, China’s forest area has expanded from 117.12 million hectares during 1973–1976 to 179.89 million hectares during 2014–201820,21. Notably, the planted forest area has nearly doubled, making China the global leader in earth greening22,23,24. Given the predominance of newly emerging forests, China’s forests are generally younger, having great potential for future climate change mitigation25. However, the patterns and driving mechanisms of AGC accumulation rates resulting from different forest restoration approaches remain unclear, leading to challenges in accurately estimating the contribution of China’s forest restoration strategies to global climate change mitigation. Therefore, a systematic comparative study on AGC accumulation rates and their differences in different forest restoration modes are urgently needed across China, to help accurately estimate forest carbon accumulation, productivity, and nutrient cycling under current or future climate change scenarios.

While many previous studies have focused on the tropical secondary forests and analyzed their naturally restored spatial patterns and driving mechanisms, such as studies on Amazon Borneo, Central Africa, and Latin America16,26,27,28,29,30, China has not been extensively explored in this context despite being one of the global leaders in forest restoration. This lack of research limits our comprehensive understanding of global forest recovery dynamics. Additionally, unlike tropical forests, where secondary forests regenerate naturally, China has not only implemented extensive natural forest conservation but also established a substantial number of planted forests through afforestation18,21. This unique context provides an ideal opportunity to understand the differences and driving mechanisms between afforestation and natural regeneration. Previous studies have shown that the recovery patterns of secondary natural forests are influenced by site conditions, including climate, soil, topography, human activities, and forest structure15,16,26,27,29,31,32. However, whether the response mechanisms of AGC accumulation rates in planted forests to these factors are similar to those in natural forests, whether there are differences in their AGC accumulation rates, and what the primary factors causing these differences are still unknown. Furthermore, a better understanding of drivers of AGC accumulation rates in planted and natural forests, especially in highly altered forest regions, may optimize the efficiency of carbon conservation and restoration projects and support nature-based solutions for climate change mitigation.

The primary objective of this study is to comprehensively assess the spatial patterns and driving mechanisms of AGC accumulation rates in China’s recovering forests, focusing on both afforestation and natural regeneration processes. To achieve this, we first developed AGC accumulation rate models for different forest types based on 52,055 field survey samples from planted and natural forests, combined with forest type boundaries derived from the updated 2020 China Vegetation Map33. Next, we mapped the spatial extents and identified the ages of China’s young planted forests (PYF) and young natural forests (NYF), defined as forests younger than 30 years, using spatial distribution data of planted and natural forest18,22 and Landsat time-series imagery (Methods). Subsequently, we classified NYF and PYF according to China Vegetation Map-defined forest types and applied the developed AGC models to map their AGC accumulation rate patterns across China. Finally, by integrating data on temperature, precipitation, soil characteristics, forest structure, topography, and human activity, we analyzed the spatial patterns and driving mechanisms behind AGC accumulation rates and their differences between NYF and PYF. It is important to note that our study constructed models based on forest age and AGC density; thus, forest age is hypothesized to be the primary factor influencing the AGC accumulation rates in both PYF and NYF. This has been extensively reported in previous studies15,16,27,29. Therefore, in our analysis of driving factors, we excluded forest age. This comprehensive analysis helps elucidate the distinct recovery pathways and carbon dynamics under different forest restoration strategies.

Results and discussion

Response of AGC to growth time in China’s PYF and NYF

We constructed AGC growth models for various planted and natural forest types based on China Vegetation Map and field data, as shown in Fig. 1. The model accuracy of natural forests, indicated by R² values, ranges from 0.6 to 0.9, with an RMSE of 3.1 to 31.0 Mg C ha−1 (Fig. 1a–d). For planted forests, the fitting accuracy ranges from 0.5 to 0.9, with an RMSE of 1.6 to 24.0 Mg C ha−1 (Fig. 1a–d).

Fig. 1: Modeled variations in AGC accumulation rates for different types of planted and natural forests in China.
Fig. 1: Modeled variations in AGC accumulation rates for different types of planted and natural forests in China.
Full size image

a, b Changes of AGC in various types of natural forests along with time. c, d Changes of AGC in various types of planted forests along with time. e Average AGC accumulation rates for both natural forests and plantations under 30 years of age. TRRF Tropical rain forest, TRNF Tropical needleleaf forest, TENF Temperate needleleaf forest, STNF Subtropical needleleaf forest, SBDF Subtropical broadleaf deciduous forest, TBDF Temperate broadleaf deciduous forest, SMBEF Subtropical monsoon broadleaf evergreen forest, SBEF Subtropical broadleaf evergreen forest, TMNBDF Temperate mixed needle leaf and broadleaf deciduous forest, SMBEDF Subtropical mixed broadleaf evergreen and deciduous forest, CTTMNF Cold-temperate and temperate mountains needle leaf forest, SBESF Subtropical broadleaf evergreen sclerophyllous forest. STBFS Subtropical, tropical bamboo forest and scrub, TMRF Tropical monsoon rain forest. Shading in (a–d) denotes the 95% confidence interval of the Logistic nonlinear model. The error bar in (e) denotes the 95% confidence interval of the estimated AGC rate for various forest types.

Our analysis of the AGC accumulation rates in planted and natural forests shows distinct trajectories across China between forest types (Fig. 1a–d). During the first 30 years of recovery, the AGC accumulation rates of natural forests generally exceed those of planted forests (Fig. 1a–d). AGC accumulation rates for various natural forests range from 23.6 ± 5.2 to 82.9 ± 22.6 Mg C yr−1 (95% confidence interval from Monte Carlo simulations), while planted forests range from 19.6 ± 21.9 and 72.8 ± 21.5 Mg C yr−1 (Fig. 1c and Supplementary Table 1). Notably, the subtropical monsoon evergreen broadleaf forest exhibits the highest AGC accumulation rates for planted and natural forests among various forest types (Fig. 1e and Supplementary Table 1). In contrast, temperate coniferous forest exhibits the lowest AGC rates for natural forests. Planted forests in the subtropical, tropical bamboo forest and scrub region have the lowest AGC accumulation rates (Fig. 1e and Supplementary Table 1). These findings highlight the considerable spatial variability in AGC accumulation rates, driven by diverse forest restoration strategies across China.

Spatial patterns of AGC accumulation rates in China’s PYF and NYF

To assess the spatial variation and driving mechanisms behind AGC accumulation rates for planted and natural forests, we extracted the PYF and NYF age distribution based on our previously developed planted and natural forest map18, using Landsat images and time series change detection methods34 (Methods). The map showed that China’s PYF covered approximately 4.5 × 105 km², with an average age of approximate 19 years in 2020 (Supplementary Fig. 2a). These forests are primarily distributed in the southern, southeastern, and northern regions of China (Supplementary Fig. 2a). The majority, about 27% of the total area, were aged between 21 and 25 years. Forests aged 16–20 and 26–30 years each accounted for roughly 21%, while forests aged 11–15 comprised around 14%, and those younger than 10 years represented 17% (Supplementary Fig. 2a). China’s NYF covered approximately 3.1 × 105 km², with an average age of 16 years. These forests are predominantly located in the northeastern, southwestern, and southeastern regions (Supplementary Fig. 2b). NYF aged 16–20 years were the most common, representing approximately 25% of the total area. This is followed by forests aged 26–30, 1–5, 11–15, 6–10, and 21–25 years, with proportions of 18%, 17%, 14%, 14%, and 11%. (Supplementary Fig. 2b).

Then, we classified the PYF and NYF into various forest types according to updated China Vegetation Map33. Using the above established models for each forest type, we mapped the spatial distributions of AGC accumulation rates for China’s PYF and NYF in 2020 within 0.1° grid (Fig. 2). The maps indicated that the AGC accumulation rates for PYF and NYF across China ranged from 0 to 2.7  Tg C yr−1 and 0 to 2.8 Tg C yr−1, respectively (Fig. 2a, b). The average AGC accumulation rate for NYF was 0.7 ± 0.3 Tg C yr−1, approximately 1.4 times that of PYF, consistent with previous studies34. Both forest types exhibited significant spatial heterogeneity, with considerable variation along latitudinal and longitudinal gradients. Longitudinally, the AGC accumulation rate of PYF initially increased with longitude, followed by a subsequent decline (Fig. 2c). In contrast, the NYF AGC accumulation rate exhibited less fluctuation and generally followed an upward trend (Fig. 2c). This differential longitudinal response is likely due to a combination of environmental factors, forest management practices, and unique geographical characteristics. Specifically, the 90°–100 °E region, where the most significant fluctuations in PYF’s AGC accumulation rates were observed, is a transitional zone between temperature and subtropical climates33. PYF is more sensitive to climatic fluctuations, which can lead to rapid initial carbon sequestration but also result in diminishing returns over time due to soil nutrient depletion and other limiting factors14. This region is also characterized by a transition between forests and other land cover types and a hotspot for planting forests, where large-scale afforestation and reforestation efforts have been implemented to combat desertification and improve carbon sequestration35, which resulted in higher fluctuations in PYF’s AGC accumulation rates.

Fig. 2: Distributions of PYF and NYF AGC accumulation rates in China.
Fig. 2: Distributions of PYF and NYF AGC accumulation rates in China.
Full size image

a, b Spatial patterns of PYF (a) and NYF (b) AGC accumulation rates. c, d Changes in AGC accumulation rates with increasing longitude (c) and latitude (d). Shading denotes the 95% confidence interval of the estimated AGC accumulation rates.

Latitudinally, the AGC accumulation rates of both PYF and NYF showed a consistent pattern, gradually decreasing with increasing latitude (Fig. 2d). However, the decline in AGC accumulation rates of PYF was more profound beyond 20°N latitude compared to NYF (Fig. 2d). Beyond this latitude, the AGC accumulation rate of PYF consistently remained lower than that of NYF, with both declining gradually (Fig. 2d). This disparity may reflect the higher sensitivity of PYF to colder temperatures and shorter growing seasons at higher latitudes, which restricts their carbon sequestration potential36. Conversely, NYF, owing to their greater structural complexity and long-term adaptation to local climatic conditions, exhibit more stable AGC accumulation rates even under less favorable environmental conditions14,37.

Factors driving the patterns of AGC accumulation rates in China’s PYF and NYF

To deeply understand the mechanisms underlying the spatial patterns of AGC accumulation rates and identify the primary factors contributing to the differences between NYF and PYF, we collected data on temperature, precipitation, topography, soil, forest structure, and human activity and employed a Random Forest algorithm to evaluate their influence on the accumulation rates of AGC in NYF and PYF (Methods).

Our analysis highlights the key drivers influencing the spatial distributions of AGC accumulation rates in both PYF and NYF, emphasizing their similarities and differences. Temperature emerged as the most influential factor, explaining 39.6% of the spatial variance in PYF (Fig. 3a) and 28.2% in NYF (Fig. 3b). Among the temperature-related variables, temperature seasonality was the most influential indicator for PYF (Supplementary Fig. 3a) and NYF (Supplementary Fig. 4a), exhibiting a significant negative correlation with the AGC accumulation rates in both forest types (Fig. 3c, d). This finding is consistent with previous research, which suggests that seasonal temperature variations significantly affect plant physiological processes and ecosystem stability38. Temperature fluctuations can disrupt the balance between photosynthesis and respiration, resulting in reduced carbon fixation, particularly during extreme temperature events39. Moreover, increased seasonal temperatures exacerbate water stress, further limiting carbon uptake through stomatal closure40. Such seasonal temperature fluctuations can shorten the growing season, increase tree mortality, and, in severe cases, shift ecosystems from carbon sinks to carbon sources41,42. Under the intensifying effects of global climate change, seasonal temperature fluctuations are expected to become more pronounced, potentially altering forest growth, mortality, and the balance between photosynthesis and respiration throughout the years43,44. Such shifts can either amplify or constrain carbon uptake, thus affecting both forests differently. Integrating these considerations into future projections and management strategies is essential for accurately evaluating carbon accumulation rates and guiding effective forest restoration under changing climatic conditions.

Fig. 3: Factors driving the AGC accumulation rate spatial patterns of NYF and PYF.
Fig. 3: Factors driving the AGC accumulation rate spatial patterns of NYF and PYF.
Full size image

a, b The relative importance of temperature, precipitation, soil, topography, forest structure, and human activity for PYF (a) and NYF (b). c Relationships between the PYF AGC accumulation rate and the most critical factors selected in temperature, precipitation, soil, topography, forest structure, and human activity. d Relationships between the NYF AGC accumulation rate and the most critical factors selected in temperature, precipitation, soil, topography, forest structure, and human activity. The relationship with each factor was modeled using simple linear regression with T-test. The solid black lines represent the fitted lines. TESE, ELEV, CASC, POWQ, SOPH, and HUFP represent the Temperature Seasonality, elevation, canopy structure complexity, precipitation of warmest quarter, soil pH, and human footprint.

Precipitation explained 12.6% and 23.5% of the spatial variance of AGC accumulation rates in PYF (Fig. 3a) and NYF (Fig. 3b), respectively, with a more pronounced effect on NYF. The precipitation of the warmest quarter was the most significant precipitation-related factor influencing AGC rates in both PYF and NYF (Supplementary Figs. 3c and 4c), showing a strong positive correlation with AGC accumulation rates (Fig. 3c, d). This suggests that increased precipitation can enhance AGC accumulation rates, aligning with previous studies45. The warmest quarter typically coincides with the peak growing season in many forest ecosystems, where temperature and moisture availability are crucial for photosynthesis and biomass accumulation46. Breshears et al. emphasize that adequate precipitation during this period ensures sufficient water availability, which is vital for maintaining plant water relations and maximizing carbon assimilation47. This period often represents the intersection of maximum water demand and temperature stress, making precipitation a key determinant of AGC sequestration rates48. Allen et al. further discussed the broader implications of drought and heat stress on forest ecosystems, linking the importance of precipitation during the warmest quarter to overall forest health and carbon accumulation43.

Topographic factors accounted for 16.7% and 17.2% of the influence on AGC accumulation rates in PYF (Fig. 3a) and NYF (Fig. 3b), respectively. Elevation had the most significant negative influence on the AGC accumulation rates in both forest types (Fig. 3c, d and Supplementary Figs. 3e and 4e). As elevation increases, factors such as temperature, moisture availability, soil nutrients, and vegetation composition are adversely affected, leading to reduced forest growth and lower carbon sequestration potential. For example, Sundqvist et al. highlighted that the decline in soil fertility and moisture with increasing altitude adversely affects tree growth, thereby limiting the carbon sequestration potential of forests49. Additionally, higher elevations are typically associated with less dense forests and species adapted to harsher environmental conditions, which generally exhibit lower biomass and carbon storage capacities. Previous studies further emphasize that the reduction in forest structure and tree density at higher elevations leads to a significant decrease in carbon storage50.

Soil contributed approximately 12% to the variation in AGC accumulation rates in both forest types (Fig. 3a, b). Soil pH was the most critical soil-related factor, showing a significant negative correlation with AGC accumulation rates (Fig. 3c, d and Supplementary Figs. 3b and 4b), this effect can be attributed to soil pH’s overarching control on nutrient dynamics, microbial processes, and plant physiology51,52. Elevated soil pH can hinder the uptake of essential nutrients, such as phosphorus, leading to reduced biomass production53 and, consequently, lower carbon sequestration.

Canopy structural factors had a more substantial influence on AGC accumulation rates in PYF, accounting for 13.9% of the spatial variation (Fig. 3a), compared to only 8.2% in NYP (Fig. 3b). Canopy structure complexity was the most influential forest structure indicator (Supplementary Figs. 3d and 4d), showing a positive correlation with AGC accumulation rates (Fig. 3c, d). This correlation has been substantiated by the research of Su et al., which demonstrates that increased canopy complexity is associated with higher productivity37.

Human activities have a more significant impact on the AGC accumulation rates in NYF (Fig. 3b) compared to PYF (Fig. 3a), which aligns with previous studies54. This is primarily due to the homogeneity and intensive management practices in PYF reduce variability in response to anthropogenic influences54. In contrast, NYF, characterized by higher biodiversity, structural complexity, and more significant potential for carbon sequestration, is more vulnerable to variations in human activities, resulting in more pronounced effects on AGC accumulation55.

Factors driving the difference in AGC accumulation rate between PYF and NYF

We further explored the primary factors influencing the differences in AGC accumulation rates between NYF and PYF of the same age at various scales (0.05°, 0.1°, and 0.2° grid) (Fig. 4). Our analysis reveals that the accumulation rates of AGC in NYF are significantly higher than those in PYF across the various scales considered (Fig. 4a and Supplementary Fig. 5a). Among the factors, forest structure is the primary driver behind the significantly lower AGC accumulation rates in PYF compared to NYF across all scales (Supplementary Fig. 5b), accounting for approximately 20% of the observed differences (Fig. 4b). Furthermore, we identified that the difference in tree density between PYF and NYF within the forest structure emerged as the most critical indicator (Supplementary Figs. 6d, 7d, and 8d), with PYF consistently showing a significantly higher tree density than NYF at various scales (Supplementary Fig. 9). Interestingly, a significant negative correlation between the tree number difference and the AGC accumulation rate difference was observed across these scales (Fig. 4c and Supplementary Fig. 10a). This suggests that despite the higher tree number in PYF, their AGC accumulation rates are lower than those of NYF, challenging the conventional assumption that increasing tree number directly correlates with higher carbon storage56,57.

Fig. 4: Factors driving the differences in AGC accumulation rates between PYF and NYF.
Fig. 4: Factors driving the differences in AGC accumulation rates between PYF and NYF.
Full size image

a Box plots of the AGC accumulation rate of PYF and NYF with the same age at 0.05°, 0.1°, and 0.2° scales. b The relative importance of temperature, precipitation, soil, topography, structure, and human activity. c Relationships between the differences in AGC accumulation rates between PYF and NYF with differences in the primary factors among each category. TRNU tree number, ISOT isothermality, PRSE precipitation seasonality; CEC soil cation exchange capacity, ELEV elevation, HUFP human footprint. *** represent significant differences at 99.9% confidence levels. Diff. Difference.

Similarly, we assessed the influence of differences in other factors between PYF and NYF on the disparity in their AGC accumulation rates. Across different scales, the most influential factors within the categories of temperature, precipitation, soil, and topography that affect AGC accumulation rate differences between PYF and NYF remain consistent (Supplementary Figs. 4–6). Among temperature-related variables, the difference in Isothermality is the most significant indicator (Supplementary Figs. 6a, 7a, and 8a), exhibiting a positive correlation with the variation in AGC accumulation rates between PYF and NYF (Fig. 4c and Supplementary Fig. 10b). Regarding precipitation, the difference in precipitation seasonality emerged as the most critical factor (Supplementary Figs. 6c, 7c, and 8c), showing a significant negative correlation with the disparity in AGC accumulation rates between the two types (Fig. 4 and Supplementary Fig. 10c). For soil factors, the difference in cation exchange capacity in soil factors was the primary indicator (Supplementary Figs. 6b, 7b, and 8b), exhibiting a significant positive correlation with the AGC accumulation rate disparity between PYF and NYF (Fig. 4c and Supplementary Fig. 10d). Elevation difference was identified as the most influential topographical factor (Supplementary Figs. 6e, 7e, and 8e), with a significant positive correlation to the AGC accumulation rate disparity (Fig. 4c and Supplementary Fig. 10e). Finally, the difference in their human activities was significantly negatively correlated with the variation in AGC accumulation rates between PYF and NYF (Fig. 4c and Supplementary Fig. 10f).

Present and future contributions of China’s PYF and NYF to climate change

By assuming that the PYF and NYF examined in this study are effectively protected and remain undisturbed, we used the developed AGC growth models to predict their carbon sequestration potential in 2060 (Supplementary Fig. 11), the target year for China’s carbon neutrality. Our analysis indicates that from 2020 to 2060, the AGC sequestration in these planted forests will increase from 992.5 ± 15.3 Tg C to 1542.3 ± 25.8 Tg C, a net rise of about 55% (550Tg C) (Fig. 5 and Supplementary Table 2) with the primary growth occurring in the southeastern and northern regions of China (Supplementary Fig. 11a, b). In natural forests, the AGC sequestration will increase from 674.9 ± 10.0 Tg C to 1650.3 ± 26.8 Tg C, a net rise of around 145% (975 Tg C) (Fig. 5 and Supplementary Table 2), with the most growth observed in the southwestern and northeastern regions of China (Supplementary Fig. 11c, d). Collectively, these PYF and NYF are projected to sequester approximately 1525 Tg C by 2060, provided they remain intact.

Fig. 5: Future contributions of China’s PYF and NYF to climate change.
Fig. 5: Future contributions of China’s PYF and NYF to climate change.
Full size image

a Future AGC storage changes in China’s PYF and NYF. b Future AGC rate changes in China’s PYF and NYF. Shading denotes the 95% confidence intervals of estimated AGC storage and rate.

The trend analysis shows that PYF and NYF exhibit an upward trend in AGC storage. However, a notable divergence is observed in the magnitude and rate of carbon sequestration between the two forest types. PYF displays a relatively moderate and consistent increase in AGC storage, while NYF exhibits a more pronounced and rapid increase, reflecting a higher rate of AGC accumulation in these ecosystems (Fig. 5a). Moreover, Fig. 5b reveals that NYF achieves a higher peak in AGC accumulation compared to PYF, with a more dynamic and fluctuating pattern of AGC sequestration over time. AGC accumulation rate in NYF peaks earlier and at a higher rate (Fig. 5b), suggesting that natural regeneration processes will be more efficient in the initial decades. In contrast, the AGC accumulation rate for PYF is relatively stable but occurs at a lower rate (Fig. 5b), indicating a slower but steadier increase in carbon storage. These observations underscore the potential limitations of relying solely on planted forests for rapid AGC sequestration and highlight the importance of incorporating natural regeneration into forest management strategies.

Forest restoration has become a critical focus for nations worldwide. China’s efforts in forest afforestation and natural forest conservation have significantly contributed to mitigating global climate change4,24. These initiatives provide valuable insights for other nations aiming to implement effective forest restoration strategies21. Our research explores the nuances of AGC accumulation in both afforested and naturally regenerated forests in China. We find PYF exhibits significantly lower AGC accumulation rates than NYF at comparable ages, primarily due to differences in forest structure, particularly the negative correlation with tree number. This is of great concern. While it is commonly known that higher tree numbers can initially enhance carbon storage, the competition for limited resources such as light, water, and nutrients often leads to suboptimal growth rates and reduced carbon storage efficiency56,58. A growing body of research indicates that many plantations in China have stand densities exceeding optimal levels, which can hinder carbon sequestration. For example, the carbon sequestration of Pinus massoniana plantations with various stand densities in southern China was deeply assessed and revealed that higher planting density led to a decrease in the average carbon sequestration rate and carbon sink59. Therefore, thinning and density management should be implemented as part of forest management strategies. Thinning selectively removes underperforming trees or contributes to excessive crowding, thereby reducing competition for resources and allowing the remaining trees to grow more efficiently and store more carbon60. Furthermore, planting trees at lower initial densities can reduce early competition and promote better tree growth in the long run. Previous studies also demonstrated selective thinning increases biomass and carbon storage by reducing competition at high densities61. In addition, our analysis showed that NYF exhibited higher and more stable AGC accumulation rates than PYF, which is also likely due to their more complex species composition. Hence, species selection is another essential aspect of optimizing forest structure. Planted forests are often dominated by a single species or a limited number of species, which can reduce biodiversity and lead to decreased resilience to environmental stressors62. The selection of species that are well-adapted to local environmental conditions and complement each other in terms of growth rates, canopy structure, and resource use is critical for improving the carbon storage potential of planted forests. Previous studies demonstrated that mixing species with complementary traits, such as fast-growing pioneers and shade-tolerant trees, enhances resource use efficiency and increases carbon storage in mixed plantations63. Similarly, selecting species for local conditions and complementarity boosts carbon stocks by improving light capture and nutrient cycling62. Therefore, future forest management should adopt a balanced and integrated approach that leverages the strengths of both natural and planted forests to maximize carbon sequestration and achieve climate goals.

Conclusion

In the context of accelerating global carbon neutrality initiatives, this study underscores the pivotal role of forest restoration in climate change mitigation. It highlights the current status and differences between planted and naturally regenerated forests in China. By integrating remote sensing and field observations, we demonstrate that seasonal temperature dominates the spatial heterogeneity in AGC accumulation rates between PYF and NYF with a negative correlation. Furthermore, PYF exhibits lower AGC accumulation rates at comparable ages, largely attributable to structural factors such as tree number. Although PYF initially sequesters more AGC, modeling projections indicate that its total AGC stock will fall below NYF’s by 2060.

These findings not only refine our understanding of differential forest carbon accumulation processes but also underscore the importance of optimizing forest structure to enhance long-term carbon sequestration in planted forests. By informing targeted silvicultural practices and policy decisions, this study provides critical insight into maximizing carbon storage potential and fortifying global climate strategies, particularly in regions where forest restoration initiatives are integral to achieving net-zero emissions targets.

Materials and methods

Forest inventory data

We collected 52,055 forest survey plots from the Seventh National Forest Inventory (2004–2008) to construct biomass growth models of forests. Each plot was about 667 m2 and recorded the forest’s origin, age, and species. In each plot, trees with a diameter at breast height (DBH) of ≥5 cm were labeled, measured, and recorded. The AGC density of each plot was calculated through the following steps: First, the volumes were calculated by applying the one-variable tree volume tables for each species in a plot specifically developed in each province23. Second, the stand volumes were summarized to the plot level from all recorded trees. Based on the stand volumes, we used the DBH-based volume-biomass equation (Eq. 1) to calculate the biomass of each plot. Third, species-specific conversion ratios were then applied to convert the biomass of each plot to AGC stock (Eq. 2). Finally, through proportional scaling, we obtained the AGC density, namely the AGC value per hectare.

$${B}_{i}=a\times {V}_{i}+b$$
(1)
$${{AGC}}_{i}={\mu }_{i}\times {{Biomass}}_{i}$$
(2)

\({B}_{i}\) represents the biomass of the i-th plot, \({V}_{i}\) is the per-unit-area volume of the i-th plot, and a and b are constants whose values for different tree species are provided in Supplementary Table 3. \({{AGC}}_{i}\) denotes the AGC stock of the i-th plot, while \({\mu }_{i}\) is the conversion coefficient between biomass and carbon stock for the i-th plot, with values for different tree species given in Supplementary Table 3.

Temperature, precipitation, soil, topography, forest structure, and human activity data

To investigate the influence of temperature, precipitation, soil, topography, forest structure, and human activity on the spatial distributions of AGC accumulation rates in PYF and NYF, we compiled a dataset comprising various factors (Supplementary Table 4).

Temperature and precipitation data were obtained from the WorldClim dataset at a spatial resolution of 1 km by 1 km. This dataset includes 11 temperature-related variables (annual mean temperature, mean diurnal range, isothermality, temperature seasonality, max temperature of warmest month, min temperature of coldest month, temperature annual range, mean temperature of wettest quarter, mean temperature of driest quarter, mean temperature of warmest quarter, and mean temperature of coldest quarter) and 8 precipitation-related variables (annual precipitation, precipitation of wettest month, precipitation of driest month, precipitation seasonality, precipitation of wettest quarter, precipitation of driest quarter, precipitation of warmest quarter, and precipitation of coldest quarter). The soil data were derived from the work of Liu et al.64, including 17 variables on nine soil properties (pH, organic carbon, nitrogen, phosphorus, potassium, cation exchange capacity, bulk density, coarse fragments, and thickness), all at a resolution of 1 km by 1 km. The mean values were calculated for six standard soil depth intervals (0–5 cm, 5–15 cm, 15–30 cm, 30–60 cm, 60–100 cm, and 100–200 cm). The topography factors, including elevation, slope, and aspect, were calculated from the Shuttle Radar Topography Mission data at a resolution of 1 km by 1 km. The forest structure factors, including forest canopy height, tree density, and canopy structure complexity, were collected. Forest canopy height and canopy structure complexity were generated by Liu et al.37,65 at resolutions of 30 m and 0.0025°. The tree density map was generated by Cheng et al.66 with a spatial resolution of 100 m. Regarding human activity, we selected the human footprint index for 2020, which was generated by Mu et al.67.

Modelling the AGC accumulation rates of China’s PYF and NYF

Quantifying the AGC accumulation rates of PYF and NYF in China involves two main steps. Firstly, we developed AGC stock growth models for various forest types using forest inventory datasets and the updated 2020 China Vegetation Map33. The process is as follows: The forest inventory samples were first classified into different forest types (Supplementary Tables 5 and 6) according to the updated 2020 China Vegetation Map using spatial join tools in ArcGIS Pro 3.0 software. Then, the AGC density and forest age relationship was established for each forest type using the Chapman-Richards growth model (Eq. 3)26. This model assumes that the AGC can recover to levels equivalent to primary forests over time, reaching a predetermined asymptote. To determine the AGC sequestration rate, we calculated the first derivative of the Chapman-Richards growth model with respect to age (Eq. 4), which provides the AGC accumulation rates for each forest type at different ages.

$${{{\rm{y}}}}=\frac{k}{1+{e}^{-a\times (t-b)}}$$
(3)
$${y}^{{\prime} }=\frac{{{ake}}^{-a\times (t-b)}}{{(1+{e}^{-a\times (t-b)})}^{2}}$$
(4)

During the modeling process, statistical analysis and data preprocessing were conducted on the field survey samples for each forest type (Supplementary Tables 5 and 6). First, we applied the three-standard deviation method to remove any significant outliers, minimizing their impact on the modeling results. Then, for the AGC density corresponding to the same age, we selected the median AGC density for each forest age for modeling to reduce the effect of variability in the data, following the approach of Heinrich et al.26.

Mapping the distributions of AGC accumulation rates for China’s PYF and NYF

To reveal the spatial patterns of AGC accumulation rates in NYF and PYF across China, we first identified these forests’ current spatial distribution and age information based on the previously produced map of China’s natural and planted forests18 (the forests were defined as areas with tree cover and canopy cover exceeding 30%68) using Landtrendr change detection algorithm applied to time series of remote sensing imagery. Specifically, we used time-series Landsat imagery from 1990 to 2020 to detect the most recent disturbance year (including harvesting, fire, or other stand-replacing events). We then determined forest age by subtracting the disturbance year from 2020. Any pixel for which no disturbance was detected back to (including) 1990 was treated as having age above 30 years and thus excluded from the PYF or NYF category. Using this method, we mapped the spatial extents and age structures of NYF and PYF across China via the Google Earth Engine (GEE) cloud platform. The specific detection steps are as follows:

  1. (1).

    Segmentation of China’s Forest Regions: Given the vast and extensive area coverage of China’s forests, performing time series change detection on the entire forest region directly within GEE is unfeasible. To enhance computational efficiency, we divided the forest regions into 1° × 1° grids covering the entire country. Subsequent mapping of NYF, PYF, and their respective age structures were conducted within each grid, and the results were mosaicked into a new dataset covering the entire China using the tool “Mosaic to new raster” in ArcGIS Pro 3.0.

  2. (2).

    Time Series Data Preprocessing: To obtain the distribution of NYF and PYF forests younger than 30 years in China, we selected Landsat remote sensing imagery from 1989 to 2020, which includes Landsat TM, ETM+, and OLI Surface Reflectance Collection 2 Level 2 data. These datasets underwent basic preprocessing, including geometric correction, radiometric correction, and atmospheric correction. Given the broad span of the growing season in China, we selected imagery from June to September for an age estimated34. Further preprocessing within GEE included cloud and shadow removal using the CFMASK method69 and annual image compositing using the medoid compositing method. Finally, we calculated the Normalized Burn Ratio (NBR) vegetation index for each composited annual image, establishing a time series dataset of NBR as the foundational input for the time series change detection algorithm. The NBR index effectively captured a range of disturbance agents—encompassing natural (e.g., fires, pest outbreaks) and anthropogenic (e.g., logging)—along with subsequent recovery processes, making it suitable for forest dynamics monitoring and disturbance assessment70.

    $${NBR}=\,\frac{{NIR}-{SWIR}}{{NIR}+{SWIR}}$$
    (5)

    In the formula, NIR refers to the near-infrared band, and SWIR denotes the short-wave infrared band.

  3. (3).

    Disturbance Detection and Age Estimation: Based on the NBR time series dataset generated above, this study applied the Landtrendr time series change detection algorithm to detect temporal changes for each pixel of natural and planted forests as of 2020. LandTrendr (Landsat-based Detection of Trends in Disturbance and Recovery) is an algorithm designed to detect and analyze long-term forest disturbance and recovery trends using Landsat time series data71,72. The combination of this algorithm with the NBR time series is beneficial for capturing subtle and gradual changes in forest cover over time, making it ideal for applications such as forest age estimation and disturbance monitoring34. It operates by segmenting a time series of NBR values into linear segments representing periods of disturbance, stability, and recovery70.

In our study, we implemented the Landtrendr algorithm on GEE to detect forest disturbances and estimate forest ages. The critical parameters for running Landtrendr include the “maxSegments, spikeThreshold, vertexCountOvershoot, preventOneYearRecovery, recoveryThreshold, pvalThreshold, bestModelProportion, minObservationsNeeded”. After conducting multiple tests, we selected the following parameters to optimize the detection accuracy: “10, 0.9, 3, False, 0.25, 0.05, 0.75, 6”. These settings were determined to be the most effective for capturing the temporal changes in China’s natural and planted forests. Using this method, we identified the year a disturbance occurred for each pixel, allowing us to determine the corresponding age of the planted or natural forest pixel by subtracting that year from 2020.

We obtained spatial distribution data for PYF and NYF and their age information across China in 2020, with a spatial resolution of 30 meters (Supplementary Fig. 2). To validate the reliability of this data, we assessed its accuracy using field survey samples and existing forest age data products (Supplementary Fig. 12). The field samples were collected from previously published papers, comprising 54 field samples (Supplementary Table 7). The existing China young forest age data was generated by Xiao et al.72 using Landsat time series images and time series change detection method, which also has a 30 m resolution. This data can be downloaded on the website https://doi.org/10.6084/m9.figshare.21627023.v7. The validation results demonstrated a significant consistency (P < 0.01) between our product and the field measurements and existing data product, with an R² 0.7–0.8 (Supplementary Fig. 12). Additionally, our data product exhibited a consistent spatial distribution pattern with existing forest age data products, with the mean age differences ranging from 1 to 4 years for NYF and PYF compared with the existing dataset (Supplementary Fig. 13), confirming its reliability and providing robust data support for our subsequent research.

Subsequently, we performed a spatial nearest-neighbor analysis with the 2020 China Vegetation Map to classify PYF and NYF into different forest types using ArcGIS Pro 3.0 software. Finally, we mapped the distribution of the AGC accumulation rates for PYF and NYF across China in 2020 using the previously developed AGC growth models for each forest type (Fig. 2). Based on the map, we analyzed the characteristics of AGC accumulation rates for PYF and NYF at various scales (Fig. 2).

Assessing the importance of drivers

We employed a Random Forest model to evaluate the influence of various factors on the patterns and differences in AGC accumulation rates between NYF and PYF in China. Notably, the data involved in this study have resolutions ranging from 30 m to 1km. To ensure consistency in the scale of analysis, we resampled all datasets to a resolution of 1km using the nearest neighbor method in ArcGIS Pro 3.0 software. The relative importance of these factors influencing the AGC accumulation rates in China’s PYF and NYF was assessed using Random Forest analysis. Random Forest model is widely used in earth science research37, as it is an ensemble learning technique that constructs multiple decision trees and aggregates their results to improve prediction accuracy and robustness.

We analyzed the patterns and differences in AGC accumulation rates between NYF and PYF forests in China through three steps. Given the strong intercorrelation among temperature, precipitation, and soil factors37, as well as their uneven distribution, we first applied PCA to reduce each category of factors into a single factor, resulting in six factors. Then, we built a Random Forest model with AGC accumulation rates as the dependent variable and the six factors—temperature, precipitation, soil, topography, forest structure, and human activity—as explanatory variables. The relative importance of each factor is measured by the increase in node purity. To identify the most significant indicator within each category, we also constructed five separate Random Forest models using temperature, precipitation, soil, topography, and forest structure factors as explanatory variables to identify the most critical indicator influencing the AGC accumulation rates in PYF and NYF (Supplementary Figs. 4–6). To minimize random error in the models, we ran each model 50 times and evaluated the impact of each factor by calculating the average importance across these runs. In these models, the number of trees parameter was set to 100.

Furthermore, to delve deeper into the impacts of temperature, precipitation, soil, topography, forest structure, and human activity on AGC in NYF and PYF forests, we constructed simple linear models for the most crucial factor within each category against the AGC accumulation rates of NYF and PYF based on the results of the previous analyses. These models were subjected to statistical testing, validating their significance and reliability using paired t-tests.

Finally, to elucidate the factors contributing to differences in AGC accumulation rates between PYF and NYF, we created grids with resolutions of 0.05°, 0.1°, and 0.2° across China. We calculated the average AGC sequestration rates for PYF and NYF within each grid, along with the average values of corresponding factors (temperature, precipitation, soil, topography, forest structure, and human activity). We then selected grids with the same forest age, calculated the differences in AGC accumulation rates between PYF and NYF within each grid, and computed the differences in the corresponding factors. Then, we employed paired t-tests to assess whether there were significant differences in AGC accumulation rates between PYF and NYF across various scales. Using these differences in factors as explanatory variables, we applied the above Random Forest method to evaluate the primary factors contributing to the differences in AGC accumulation rates between PYF and NYF. Subsequently, we used linear models and paired t-tests to evaluate the impact mechanisms of the most significant factors-temperature, precipitation, soil, topography, forest structure, and human activity—on the differences between PYF and NYF.

Estimating future carbon potential

We assumed that all forests identified in this study would receive adequate protection in the future. Using the AGC stock growth models developed for each forest type, we projected the contribution of these forests to AGC stock by 2060, in alignment with China’s carbon neutrality goal. Based on these projections, we illustrated the changes in AGC stock and AGC accumulation rates for NYF and PYF from 2020 to 2060.

To validate the accuracy of the estimated carbon stocks, we collected 4789 in situ measurements (Supplementary Fig. 14), initially used in mapping China’s aboveground biomass data for 201973. These samples were gathered through systematic sampling approaches and a consistent field investigation protocol. The AGB of the tree level in these plots (DBH ≥ 5 cm) was calculated using allometric equations73. To validate the estimated AGC storage derived from this study using field survey data, we first performed spatial overlay analysis between these data and the PYF and NYF areas identified in our research. We filtered out any survey data points that were located outside the defined PYF and NYF regions, resulting in the retention of 388 field survey data points, including 124 for PYF and 264 for NYF. Next, using the updated 2020 China Vegetation Map, we classified these data into different forest types through spatial nearest-neighbor analysis using ArcGIS Pro 3.0 software. We then applied our constructed biomass density estimation models (Fig. 1) for various forest types to estimate the aboveground biomass density of the field survey data. Finally, we compared the estimated biomass densities with the field measurements, revealing significant consistency (P < 0.01) between the estimated and observed values, with R² values of 0.6 for PYF and 0.8 for NYF (Supplementary Fig. 15).

Additionally, we downloaded the 2020 global aboveground biomass data product released by the European Space Agency (https://climate.esa.int/en/projects/biomass/data/) (Supplementary Fig. 10). Using a biomass-to-carbon conversion factor of 0.5, as recommended by previous study22, we converted these data into AGC storage. We then clipped the data to the extent of PYF and NYF obtained in this study and validated the resulting datasets against our 2020 AGC storage estimates for NYF and PYF. The validation results demonstrated significant consistency (P < 0.01) from our produced data and the existing data products, with R² values ranging between 0.6 and 0.8 (Supplementary Fig. 11).

Uncertainty analysis

We employed the Monte Carlo simulation method to quantify the uncertainty of the logistic growth model parameters and to calculate confidence intervals for the predictions. First, we calculated the residuals between the original data and the initial fitted model. Subsequently, we used the Monte Carlo simulation to generate multiple sets of AGC data, each consisting of the AGC predicted by the initial model plus randomly sampled residuals. We refitted the logistic growth model for each composite dataset to obtain a new set of model parameters. This process was repeated 1000 times, generating 1000 different sets of model parameters. Based on these parameters, we calculated the predictions for each simulation and determined the 95% confidence intervals by computing the percentiles for each prediction.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.