Biodiversity benefits of China’s 20-year efforts in forest restoration Authors: Bin Wang, Zhixin Wen, Pinjia Que, Taxing Zhang, Xingcheng He, Nan Yang, Xue Zhong, Yu Xu Published: 2025 ABSTRACT AbstractChina has become a global leader in forest restoration, through massive investment and various large-scale programs since 2000. Nevertheless, the overall biodiversity outcomes of these efforts remain to be quantified. Here, using 402 non-migratory forest bird species as indicators, we assess biodiversity changes from 2000 to 2020 at a national scale, controlling for the effects of climate change by using an ecological niche modeling framework. We find substantial biodiversity benefits linked to the expansion of forest extent and the improvement in forest structure, with 73.6% of avian species experiencing habitat expansion and mean species richness increasing by 2.886 ± 8.418. Protection-focused and afforestation-dominated restoration programs both effectively enhanced biodiversity, with planted forests performing similarly to natural forests over this period. Key factors facilitating this enhancement included tree cover and canopy structural complexity. Our findings suggest that restoring planted forests in degraded landscapes can provide substantial biodiversity benefits, although this should incorporate mixed tree species silviculture practices to enhance the structural complexity of planted forests. This study highlights China’s achievements but also the potential risks involved in forest and biodiversity restoration, and provides general insights for the optimization of future restoration policy and management in China and beyond.Subject terms: Conservation biology, Restoration ecology, Biodiversity Introduction The rapid loss of forests globally is causing manifold environmental and ecological issues, posing unprecedented threats to human society1–3. Under international initiatives such as the Bonn Challenge4, the New York Declaration on Forests5, and Sustainable Development Goals6, many countries have committed to restore millions of hectares of forest. However, the effectiveness of these restoration programs and policies have been found to vary substantially between nations7, with biodiversity outcomes ranging from near-total success to complete failure8,9. It is thus crucial to understand what strategies work reliably in which regions to achieve ambitious global forest sustainability and restoration targets10,11. China has become the global leader in forest restoration and greening12. Historically, China’s forests had undergone a net rate of loss, but due to the implementation of a series of large-scale forest restoration programs since 2000, including both protection-focused (i.e., the Natural Forest Protection Program) and afforestation-dominated programs (e.g., the Grain for Green Program)13,14, this has switched to a substantial rate of gain15–17, with 24.02% national forest coverage achieved by 202218. Furthermore, the Chinese government has committed to further increase this percentage to 26% by 203519. China’s restoration efforts have demonstrably enhanced a range of forest ecosystem services (including food production, carbon sequestration, soil retention, sandstorm prevention, water retention, flood mitigation, and climate mitigation)20–22 as well as socioeconomic services (including livelihood improvement, poverty alleviation, job creation, and economic development)23–25, with recognized potential to provide a model for other countries26. Potentially, this forest restoration may also support biodiversity recovery27–29, however, the implicit assumption that increasing forest cover will deliver broad benefits to biodiversity remains to be proven30,31. This is especially important at a national scale, because empirical and meta-analytical studies have documented both losses and gains for different taxa at local and landscape scales, relating to restoration approaches, restoration durations, and biophysical conditions24,32–34. In this regard, while the China Ecosystem Assessment reported forest gain, it simultaneously reported a 3.1% decrease in habitat provision for biodiversity from 2000 to 2010 under restoration programs and natural capital investments20. This provides China’s only national-scale quantitative assessment of biodiversity outcomes; however, species’ habitat estimates were based mainly on the conversion of ecosystem types, largely overlooking temporal changes in forest structure and quality during the progression of restoration. This is likely inadequate because forest structural factors, such as canopy structure and fragmentation, represent habitat quality and the provision of ecological niches more comprehensively than simple land cover or ecosystem type, and typically have more direct impacts on the suitability of habitat for supporting biodiversity35–38. It is, therefore, crucial to account for dynamic changes in forest structure39 when attempting to accurately map habitat suitability and to integrate these disparate effects into restoration schemes. To these ends, recent advancements in remote sensing technology have enabled the characterization of 3D forest structure, supporting more rigorous analysis40,41. Ongoing rapid global climate change also exerts various impacts on species distribution and biodiversity patterns that further compound the difficulties of assessing the effectiveness of biodiversity enhancement through forest restoration initiatives42–44. It is thus necessary to attempt to disentangle the relative contributions of forest restoration and climate change, and to understand the interaction between these drivers. Due to the complexities involved and issues of restoration policy scale and inconsistency, efforts to enhance national-level biodiversity through forest restoration require an integrated quantitative assessment to better understand the drivers of forest biodiversity and optimize future restoration strategies. To deliver this vital assessment, in this study we apply an ecological niche modeling (ENM) framework to examine changes in suitable habitat and species richness for 402 non-migratory forest bird species in response to forest restoration across mainland China from 2000 to 2020 as an indicator of biodiversity change45,46. We exclude migratory birds to avoid the risk that their seasonal distributions could confound analyses of forest habitat suitability and associated environmental conditions47,48. Firstly, to evaluate how effectively forest extent and quality recovered over this time period, we compare forest area and structural factors across ecoregions49, different restoration program types, and forests under different management regimes. Next, we delineate habitats for each bird species for both 2000 and 2020, and use two simulated scenarios to control for and thus isolate the separate effects of climate and forest change. We then examine the relationship between habitat changes between 2000 and 2020 and species’ traits to identify which traits correlated with habitat expansion or contraction in response to changes in climatic and forest factors, as well as their coupled effects. Finally, we map predicted species richness (of potentially occurring species) by stacking species habitat distributions under each scenario50,51. From this approach we are able to evaluate which forest structural factors exhibited the greatest potential to effectively enhance avian biodiversity recovery during the forest restoration process. Results Change in forest area and structural factors from 2000 to 2020 Between 2000 and 2020, the forested area in mainland China increased by 21,775 km2, amounting to 1% net forest expansion. Among the 11 ecoregions we included, the largest increase occurred in Northern China while the only net reduction occurred in Southern China (Fig. 1a, b and Table 1). For regions subject to different types of restoration programs, those where both protection-focused and afforestation-dominated programs overlapped had the largest net forest expansion, followed by regions subject solely to afforestation-dominated programs, where high gross forest gain was offset by high gross loss (Fig. 1a, c, Table 1 and Supplementary Fig. 1). Change in forest extent and structural factors. a Change in forest area across mainland China from 2000 to 2020. The raster layer of forest extent was resampled from original spatial resolution of 30 m to 1 km for display purpose. b 11 ecoregions of China (1, Northeastern China; 2, Northwest Arid Area; 3, Inner Mongolia Plateau; 4, Qinghai-Tibet Plateau; 5, Loess Plateau; 6, Northern China; 7, Hengduan Mountains; 8, Sichuan and Chongqing; 9, Middle and Lower Reaches of the Yangtze River; 10, Yunnan-Guizhou Plateau; 11, Southern China). c Regions with three different types of forest restoration programs (A, regions overlapping protection-focused and afforestation-dominated programs; B, regions with only afforestation-dominated programs; C, regions without any programs). d–f Changes in forest structural factors across ecoregions (d; n = 1000 for each ecoregion), regions with three different types of restoration programs (e; n = 1800 for A, 2200 for B, and 943 for C), and two pairs of forests with different management regimes (f; NF/PF, natural forest/planted forest established before 1990, n = 2200 for NF and 2056 for PF; rNF/rPF, natural forest/planted forest restored from non-forest land cover since 2000, n = 1896 for rNF and 1837 for rPF); boxplots show the median, 25th and 75th percentiles, whiskers extend to 1.5 times the interquartile range, and white dots indicate the mean values. In d, F and P values from one-way analysis of variance (ANOVA) examining differences across 11 ecoregions were reported for each factor. In e and f * indicates significant difference (P < 0.05) between groups examined by linear mixed-effects models with ecoregions as a random effect. The codes in b and c correspond to those in d and e. See Supplementary Fig. 4 and Supplementary Tables 2, 3 for detailed information. Table 1Change in forest area over mainland China from 2000 to 2020 across ecoregions and forest restoration program typesForest area (km2)Net forest changeGross forest lossGross forest gain20002020Area (km2)% of forest area 2000Area (km2)% of forest area 2000Area (km2)% of forest area 2000Mainland China12,286,8122,308,58721,7751.045,2842.067,0592.9Ecoregion2Northeastern China418,797425,32665291.662221.512,7513.0Northwest Arid Area16,78117,3916103.61270.87374.4Inner Mongolia Plateau360338032005.61032.93038.4Qinghai-Tibet Plateau11,09111,2501591.4930.82522.3Loess Plateau90,32693,33730113.311071.241184.6Northern China46,01254,094808217.65321.2861418.7Hengduan Mountains184,811186,56117500.925121.442622.3Sichuan and Chongqing449,679451,65019710.434010.853721.2Middle and Lower Reaches of the Yangtze River148,115149,26611510.831002.142512.9Yunnan-Guizhou Plateau321,022328,12170992.292092.916,3085.1Southern China596,575587,788-8787-1.518,8783.210,0911.7Region with different restoration program types3Regions overlapping protection-focused and afforestation-dominated programs1,042,0811,053,54711,4661.114,3711.425,8372.5Regions with only afforestation-dominated programs1,206,1681,216,28310,1150.830,5142.540,6293.4Regions without any programs38,56338,7571940.53991.05931.51The analysis was conducted for mainland China only, excluding Hongkong, Macau, and Taiwan.2Mainland China was divided into 11 ecoregions49. See Fig. 1b for spatial distribution of ecoregions.3Mainland China was classified into three regions based on county-scale implementation areas of nine key forest restoration programs with different objectives and schemes. See Supplementary Fig. 1 for county-level implementation area of each program, and Fig. 1c for spatial distribution and overlap of different types of programs. Change in forest area over mainland China from 2000 to 2020 across ecoregions and forest restoration program types 1The analysis was conducted for mainland China only, excluding Hongkong, Macau, and Taiwan. 2Mainland China was divided into 11 ecoregions49. See Fig. 1b for spatial distribution of ecoregions. 3Mainland China was classified into three regions based on county-scale implementation areas of nine key forest restoration programs with different objectives and schemes. See Supplementary Fig. 1 for county-level implementation area of each program, and Fig. 1c for spatial distribution and overlap of different types of programs. Forest structure also changed substantially from 2000 to 2020 (Supplementary Fig. 2). Overall, mean tree cover, tree cover standard deviation (SD), mean canopy height, and canopy height SD all increased significantly from 2000 to 2020, while the forest fragmentation index decreased significantly (paired-samples Wilcoxon signed rank tests, all P < 0.001; Supplementary Table 1 and Supplementary Fig. 3). However, patterns varied significantly among the 11 ecoregions (Fig. 1d and Supplementary Fig. 4). When accounting for the random effect of ecoregions, the increase in mean tree cover was significantly greater in regions where protection-focused and afforestation-dominated programs overlapped, as well as in regions subject solely to afforestation-dominated programs, compared to regions without any programs (Fig. 1e and Supplementary Table 2). In terms of forests subject to different management regimes, changes occurring over time were not significantly different between natural and planted forests established before 1990, but the increase in mean canopy height was significantly greater in planted than in natural forests restored since 2000 (Fig. 1f and Supplementary Table 3). Change in suitable habitat of forest birds From ENMs for the 2000 and 2020 scenarios, suitable habitat for most bird species (see Supplementary Table 4 for the full list) shifted in response to significant changes in climatic and forest structural factors between 2000 and 2020 (Supplementary Table 1 and Supplementary Figs. 2 and 5). 69.4% of species (279/402) experienced an expansion of suitable habitat (Fig. 2a), with linear regression models suggesting that easterly species (i.e., species with larger centroid longitudes) occupying higher elevations (i.e., species with higher elevation midpoints) experienced the greatest expansion (Fig. 2b and Supplementary Tables 5 and 6). When controlling for changes in forest structural and climatic factors in the ENMs, 63.4% of species benefitted from climate change (2000 to scenario A) and 73.6% benefitted from forest structural change (2000 to scenario B; Fig. 2a). Easterly and southerly species (i.e., species with lower latitude centroids) occupying higher elevations had a significantly greater likelihood of experiencing an expansion of suitable habitat in response to climate change (Fig. 2b and Supplementary Tables 7 and 8). In comparison, northerly species with wider elevation ranges and habitat breadths had a significantly greater likelihood of experiencing an expansion of suitable habitat in response to forest structural change (Fig. 2b and Supplementary Tables 9 and 10). Change in suitable habitat for bird species. a Percentage of non-migratory forest bird species (n = 402) in relation to change in suitable habitat area across mainland China from scenario 2000 to scenario 2020, to scenario A (simulating climate change from 2000 to 2020 while controlling forest structural factors as 2000 status), and to scenario B (simulating changes in forest structural factors from 2000 to 2020 while controlling climatic factors as 2000 status). b Standardized coefficient estimates from model-averaged linear regression models for exploring the relationship between species traits and change in suitable habitat area of non-migratory forest bird species (n = 402) from scenario 2000 to scenario 2020, to scenario A, and to scenario B; filled squares denote model-averaged estimates and horizontal lines denote 95% confidence intervals. See Supplementary Tables 5–10 for detailed results of linear regression models. Change in forest bird species richness By stacking species habitats predicted using ENMs, we mapped species richness (of potentially occurring birds) under each scenario (Supplementary Fig. 6). In 2000, the Hengduan Mountains, the Yunnan-Guizhou Plateau, Hainan Island, and the southeastern coastal regions were identified as diversity hotspots for forest birds (Fig. 3a). Species richness in the southern Yunnan-Guizhou Plateau decreased substantially by 2020 (Fig. 3b). Climate change between 2000 and 2020 likely contributed to this loss, as per scenario A, which simulated isolated changes in climatic factors, and indicated a decrease in species richness across similar regions (Fig. 3c). In addition, climate change alone led to a significant decrease in species richness in the eastern coastal regions (Fig. 3c). However, these decreases were partially compensated for by increases derived from changes in forest structural factors (Fig. 3d). Change in species richness of forest birds. a Predicted species richness of non-migratory forest birds (n = 402) across mainland China in 2000. b–d Change in predicted species richness from scenario 2000 to scenario 2020 (b), to scenario A (c; simulating climate change from 2000 to 2020 while controlling forest structural factors as 2000 status), and to scenario B (d; simulating change in forest structural factors from 2000 to 2020 while controlling climatic conditions as 2000 status). e Regions with three different types of forest restoration programs (A, regions overlapping protection-focused and afforestation-dominated programs; B, regions with afforestation-dominated programs only; C, regions without any programs). f, g Change in predicted species richness from scenario 2000 to scenario B across three regions with different types of restoration programs (n = 1800 for A, 2200 for B, and 943 for C) and two pairs of forests with different management regimes (NF/PF, natural forest/planted forest established before 1990, n = 2200 for NF and 2056 for PF; rNF/rPF, natural forest/planted forest restored from non-forest land cover since 2000, n = 1896 for rNF and 1837 for rPF); boxplots show the median, 25th and 75th percentiles, whiskers extend to 1.5 times the interquartile range, and white dots indicate the mean values. h Standardized coefficient estimates from model-averaged linear mixed-effects models for exploring effects of change in forest structural factors on species richness while controlling for climate change (scenario 2000 to scenario B; n = 11,000); filled squares denote model-averaged estimates and horizontal lines denote 95% confidence intervals. In f and g * indicates significant difference (P < 0.05) between groups examined by linear mixed-effects models with ecoregions as a random effect. The codes in e correspond to those in f. See Supplementary Figs. 6 and 7 and Supplementary Tables 11–13 for detailed information. When controlling for the effects of climate change, the richness of bird species increased significantly by 2.886 ± 8.418 in response to specific changes in forest structural factors from 2000 to 2020 (paired-samples T-tests, t (10,999) = −35.950, P < 0.001, Cohen’s d = −0.343, 95% Confidence Intervals = [−3.043, −2.728]; Supplementary Fig. 7). This occurred mainly around the Loess Plateau, the Sichuan Basin, the Northern China, and southeastern coastal regions, whereas decreases were found for small areas in and around the Yunnan-Guizhou Plateau and the Southern China (Fig. 3d). When accounting for ecoregions as a random effect, the increase in species richness was significantly higher in regions overlapping both protection-focused and afforestation-dominated programs (3.332 ± 8.719) compared to regions without any programs (1.265 ± 8.077); the species richness increase in regions subject solely to afforestation-dominated programs (2.221 ± 8.214) was slightly, but non-significantly higher than in regions without any programs (Fig. 3e, f and Supplementary Table 11). Overall, species richness increased under all forest management regimes from 2000 to 2020, with no significant differences between natural and planted forests, nor between restored natural and planted forests (Fig. 3g and Supplementary Table 11). Linear mixed-effects models, accounting for ecoregions as a random effect, indicated that an increase in mean tree cover, tree cover SD, and canopy height SD were significant in promoting the recovery of bird species richness between 2000 and 2020. Among these factors, mean tree cover had the largest effect, as shown by standardized regression coefficient estimates (Fig. 3h and Supplementary Tables 12 and 13). Discussion Outcomes of China’s forest restoration and biodiversity recovery efforts From 2000 to 2020, China made significant efforts and financial investments to restore and expand the extent of national forest cover. While deforestation was not arrested completely, in most ecoregions, new gains in forest cover exceeded gross forest loss (Table 1). The exception was in the Southern China ecoregion, where the majority of the forest is located, which experienced the highest rate of deforestation and thus a net forest loss of 1.5%. This was likely related to contemporaneous rapid urbanization52, cropland reclamation53, and frequent wildfires54,55, and highlighted a higher deforestation risk in this region. In contrast, the Northern China ecoregion, which is one of the major grain-producing areas with limited forest cover, restored the largest forest area by effectively controlling gross forest loss (Table 1). Implementation areas that overlapped both protection-focused and afforestation-dominated programs achieved the highest net increase in forest area, implying the dual effectiveness of reducing deforestation by the Natural Forest Protection Program17,56 and establishing new plantations through afforestation-dominated programs34. Although regions that only implemented afforestation-dominated programs achieved substantial gross forest gain through plantation expansion34,57, this was largely counteracted by the simultaneous loss of large areas to deforestation (Table 1). Beyond simple forest extent, we discovered a substantial change in forest structure from 2000 to 2020; an aspect of forest quality that has received less attention than forest quantity in previous studies. Overall, since 2000, China’s forests have become more closed, taller, more connected, and more structurally complex. These changes align with the general characteristics of mature primary forests58,59 that typically support richer faunal biodiversity60,61, indicating that meaningful progress was made in restoring forest structure and quality up to 202062. The first China Ecosystem Assessment reported that restoration programs implemented from 2000 to 2010 led to an overall improvement of most ecosystem services, including food production, carbon sequestration, soil retention, sandstorm prevention, water retention, and flood mitigation, but habitat provision for endangered and threatened species across multiple taxa decreased20. Studies conducted at smaller regional scales have documented varying biodiversity outcomes23,24,32–34. Our study concentrated solely on forest bird species as a bioindicator highly sensitive to forest change and susceptible to global forest loss45. Simultaneously, we took account of temporal changes in forest structure and controlled for contemporaneous climatic changes. From this approach, we found that the expansion of forest extent and improvement in forest structure from 2000 to 2020 resulted in a habitat expansion for a large proportion of non-migratory forest bird species (Fig. 2a), along with clear benefits for recovering bird species richness. This suggests that China’s forest restoration efforts hold substantial potential to enhance avian diversity. Furthermore, while these restoration programs have generally yielded positive consequences for biodiversity, the regions where protection-focused programs overlapped with afforestation-dominated programs were the most successful in restoring avian biodiversity (Fig. 3f). In regions subject only to afforestation-dominated programs, which experienced the largest concurrent afforestation and deforestation rates (Table 1), increases in forest bird species richness in some afforestation sites was neutralized by declines in others due to deforestation. While it might also be argued that biodiversity losses from deforestation can be compensated by gains in newly planted forests, we contend that such numerical offsets in species richness could not likely lead to equivalent biodiversity values. Avian communities and ecosystems are regionally distinct and ecologically irreplaceable, making it hard for newly planted forests to replicate the complexity and functionality of existing ones. Critically, given the slow pace of natural forest community succession and ecosystem re-establishment, any decline of biodiversity due to loss of existing forests, especially mature and “intact” forests, is unlikely to be fully mitigated by afforestation efforts63,64. The relative effects of forest restoration and climate change on bird habitat Consistent with a large number of recent reports on the range expansion of bird species based on birdwatching records in China65–68, we found that suitable habitat expanded for ca. 70% of bird species between 2000 and 2020 (Fig. 2a). A similar phenomenon has been reported internationally and intercontinentally, and climate warming is believed to be a major contributing factor69–71. Our results support this hypothesis in relation to the effect of climate change in isolation from 2000 to 2020, both in terms of simulated changes in bird species habitat suitability (Fig. 2a) and in terms of patterns of spatial diversity (Fig.3b, c). Reciprocally, however, when the effects of climate change were controlled for, we still detected that forest change from 2000 to 2020 was strongly associated with habitat benefits for forest birds. Given the overall expansion of forest extent and improvement in structure, it is unsurprised that suitable habitat increased for over 70% of bird species. Nevertheless, ca. 30% of species lost some proportion of their habitat due specifically to forest change (scenario 2000 to scenario B in Fig. 2a and Supplementary Table 4), suggesting that these species did not gain benefits from forest restoration efforts during the past two decades. There are several plausible explanations for this habitat reduction: a species may prefer habitat with enhanced forest structure but simultaneously experience deforestation and/or forest degradation within its range (e.g., the Chinese Barbet, Psilopogon faber, prefers habitat with dense tree cover, but experienced a decrease in mean tree cover in some parts of its range; Supplementary Fig. 8); a species may have a habitat preference opposing the general trend of forest restoration (e.g., the nectar-feeding Fork-tailed Sunbird, Aethopyga christinae, generally prefers low-canopy forest habitats, but restoration efforts resulted in a higher canopy in some sites and thereby decreased habitat suitability for this species; Supplementary Fig. 9); or a species may have a narrow niche breadth for a specific habitat factor, and thus either habitat improvement or degradation may alter habitat suitability (e.g., the Dusky Fulvetta, Schoeniparus brunneus, prefers habitats with moderate vertical canopy structure heterogeneity, thus either a decrease and increase in canopy height SD outside of its preferred range may reduce habitat suitability; Supplementary Fig. 10). In particular, there should be geographic or niche-related patterns to habitat gains or losses, as our analyses revealed that climatic and forest changes had varying effects on forest bird species, depending on their geographic distributions and habitat requirements. Regionally, we found that across China, climate change from 2000 to 2020 favored forest birds with easterly and southerly distributions (Fig. 2b). This reflects the general global biological trend of poleward and upward species distribution shifts under climate change69,72, where westerly and northerly distributed forest birds are exposed to greater risks from climate change in China. However, we also found that these northerly species tended to benefit most from changes in forest extent and structure (Fig. 2b), implying that China’s forest restoration efforts may have partially mitigated the climatic threats to these species. Furthermore, species with broader habitat and elevation ranges benefitted more from these forest changes (Fig. 2b), likely because their greater adaptability to changing and suboptimal conditions enabled them to occupy newly created habitats73,74. In contrast, specialist species benefited less, and some even lost habitat over this period, highlighting the need for targeted approaches to protect these species in future forest restoration initiatives. These divergent responses to climate and forest change, as well as their complex interactions, may have reshaped avian communities, with possible cascading effects for functional diversity and associated ecosystem services such as pollination and seed dispersal75,76. While our results point to these potential outcomes, further studies are needed to directly evaluate such ecological restructuring, which should be explicitly considered in future restoration planning to ensure biodiversity and functional resilience. Biodiversity value of restoring planted forests Artificial afforestation is increasingly exceeding the establishment and spread of natural forests worldwide57 and is favored by China’s forest policies26,34. However, the implications of planted forests for biodiversity and other ecosystem functions, such as water provisioning and soil erosion control, has been vigorously debated at local and landscape scales9,77,78. At the national scale, we found that because natural and planted forests had undergone similar changes in forest structural factors between 2000 and 2020 (Fig. 1f), the effectiveness of planted forests in recovering bird species richness was similar to that of natural forests (Fig. 3g). In particular, planted forests developed from unforested land cover since 2000 performed slightly better than naturally regenerating forests over the same period. Early-stage planted forests often develop a substantial canopy more rapidly than secondary successions, due to artificial afforestation practices generally cultivating fast-growing tree species28. A closed canopy is broadly associated with greater bird species richness79,80, an association corroborated by our own assessment (Fig. 3h). Therefore, we endorse long-standing calls to recognize that certain types of planted forests can have value for biodiversity conservation60,81, and advocate that restoring planted forests in degraded landscapes should be considered as a feasible and efficient way to recover forest biodiversity in the early stages of restoration. This is especially the case when there is a specific need to rapidly establish forests for timber production9, sandification control, and other applied objectives. Nevertheless, while birds are excellent indicators of ecosystem health and integrity45,46, species richness does not necessarily reflect equivalent functional or ecological values. Therefore, artificial afforestation should still be conducted in parallel with protecting natural/native forests and encouraging spontaneous succession in landscapes with suitable biophysical and ecological conditions to deliver a full spectrum of ecosystem services in the longer-term9, and also to reduce financial investment in active restoration approaches82. By controlling for climate change and baseline differences across ecoregions, we found that, in addition to tree cover, canopy structural complexity was the second most influential factors promoting the recovery of bird species richness (Fig. 3h). This aligns with the prevailing understanding that structurally complex forests provide more heterogenous habitats and diverse ecological niches for birds with divergent requirements, thereby supporting greater bird diversity83,84. The structural complexity of naturally regenerating forests is largely regulated by climate, but for planted forests, it can be improved artificially through multispecies tree planting85. Apart from biodiversity benefits, forests with complex structure and diversified tree species composition also have higher stand growth86, productivity87, drought resiliency88, and restoration velocity89. Therefore, mixed tree species silviculture with complex forest structure should be preferred over monocultures in artificial afforestation programs, regardless of whether objectives extend beyond biodiversity recovery. In counterpoint, local-scale afforestation can fail when maladapted tree species are planted, resulting in poor tree growth, low survival rate, biodiversity loss, and degraded ecosystem services13. Therefore, we recommend prioritizing the planting of native and locally adapted tree species in afforestation programs to ensure ecological integrity and long-term success. Recommendations for future forest restoration Our study represents a comprehensive assessment of the biodiversity outcomes of China’s 21st-century forest restoration efforts. We found substantial biodiversity benefits linked to the expansion of forest extent and improvement in forest structure, with both protection-focused and afforestation-dominated restoration programs contributing effectively to biodiversity recovery. Changes in climatic and forest structural factors had varying or even opposing effects on bird species with different geographic distributions or habitat requirements, and forest restoration likely played a role in mitigating some of the threats posed by climate change. We must caution, however, that our study applied only to non-migratory birds as a forest biodiversity indicator and that there are always uncertainties associated with habitat suitability modeling. Although violation of niche conservatism in ENM approaches can introduce analytical uncertainties90,91, we are confident in our ENM because the rates of niche change across species were found to be much lower than the projected pace of climate change92. Inevitably, however, climate and forest change projections may not fully reflect real-world ecological complexities, where these changes often occur simultaneously and interactively. Further studies on other taxa that attempt to account for niche shift dynamics are needed to corroborate the patterns we observed. Despite these limitations, our findings provide valuable insights that can inform forest restoration strategies in China and beyond, aiming to optimize biodiversity benefits. For example, given the comparable biodiversity recovery performance of natural and planted forests, we recommend integrating both approaches in policy and management frameworks to leverage their complementary strengths. Natural forest restoration, particularly through native species regeneration in environmentally suitable landscapes, offers a cost-effective approach for enhancing multiple ecosystem services. To complement this, afforestation in degraded landscapes offers significant potential for rapid biodiversity recovery, even when conservation is not the primary objective. Promoting mixed silviculture, particularly the planting of native species, should be encouraged to increase horizontal and vertical canopy structural complexity to facilitate biodiversity recovery. Ultimately, restoration policies should be adaptively designed to account for regional environmental and cultural heterogeneity while balancing afforestation with other land-use, and engaging local communities and stakeholders in the planning process to ensure long-term sustainability. Achieving this balance will require more targeted, region-specific studies. Moreover, biodiversity recovery is a dynamic and ongoing process. We recommend establishing long-term forest monitoring sites to evaluate the effectiveness of restoration interventions and to address emerging challenges. Methods Analysis of change in forest area and structural factors from 2000 to 2020 This study encompassed mainland China, excluding Taiwan, Hong Kong, and Macau, which undertook different forest restoration and management strategies during the study period. To quantify changes in forest area across China from 2000 to 2020, we compared seven contemporary and widely used land cover time series datasets, namely Copernicus land cover maps93, Global Land-Cover product with a Fine Classification System (GLC_FCS30D)94, Global Land Cover and Land Use Change (GLCLUC)95, the annual China Land Cover Dataset (CLCD)96, FAO-Land Cover Classification system 1 (LCCS1) land cover layers from MODIS MCD12Q197, annual maps of global forest management types57, and China Vegetation Products version 1.098. The thematic maps produced by each of these datasets had different spatial resolutions and applications, and were generated based on different data sources, algorithms, and vegetation definitions, and thus provided a variety of representations of forest extent. From among these seven datasets, we selected the GLCLUC because it provided a median forest area for both 2000 and 2020 (Supplementary Table 14), and the forest area it estimated for 2020 approximated the authoritative value (2,310,000 km2) used by the national forest resource inventory18. We produced all maps in this study based on vector layers available at the National Catalogue Service for Geographic Information, National Geomatics Center of China (http://www.webmap.cn/). We analyzed the change in forest area for China from 2000 to 2020 both as a whole and divided into 11 ecoregions according to differences in climatic, vegetation, and geomorphological conditions49 (Fig. 1b). To assess the effectiveness of different types of forest restoration programs, we classified China into three regions based on county-scale implementation areas of nine key forest restoration programs with different objectives and schemes13 (Supplementary Fig. 1): A, regions overlapping both protection-focused (i.e., the Natural Forest Protection Program, which aims to protect and rehabilitate natural forest mainly via logging bans and mountain access closures, supplemented by artificial afforestation) and afforestation-dominated programs (i.e., eight programs aimed at increasing forest cover mainly via artificial afforestation and reforestation; see Supplementary Fig. 1); B, regions with only afforestation-dominated programs; and C, regions without any programs (Fig. 1c). Change in forest area was then analyzed across these three regions subject to different types of restoration programs. To assess temporal changes in forest structure, we generated five forest structural factors for both 2000 and 2020. To characterize forest horizontal structure, we downloaded MODIS vegetation continuous fields products with a 250 m resolution99 for 2000 and 2020 and calculated the mean and SD of percentage tree cover across each 1 km cell (under a Lambert conformal conic projected coordinate system) using ArcGIS 10.6 (ESRI, Redlands, USA). For vertical forest structure, we downloaded 30 m resolution forest canopy height maps from Global Land Analysis and Discovery laboratory (https://glad.umd.edu/dataset/GLCLUC2020)100 for 2000 and 2020 and calculated the mean and SD of canopy height across each 1 km cell. From 30 m resolution GLCLUC forest maps for 2000 and 2020, we generated 1 km resolution forest fragmentation index layers that synthesized edge, isolation, and patch size effects101. All consequent analyses and modeling were performed using these generated layers with a consistent 1 km spatial resolution. We randomly sampled 1000 points within each of the 11 ecoregions, ensuring a minimum distance of 1 km between each point. This random sampling procedure was constrained within a union set of forest extent for 2000 and 2020 for all seven datasets (Supplementary Table 14). This represented a broader forest extent that encompassed generalized forests under different definitions, lost/degraded and newly established forests between these two periods. We then extracted values for the five forest structural factors (mean tree cover, tree cover SD, mean canopy height, canopy height SD, and forest fragmentation index) for 2000 and 2020 for each random point and conducted paired-samples Wilcoxon signed rank tests to compare temporal changes in forest structural factors over this interval (Supplementary Fig. 3). To determine the extent to which forest structure restoration had been achieved in each ecoregion, we also performed one-way analyses of variance (ANOVA) on changes in each structural factor from 2000 to 2020 (Supplementary Fig. 4). To determine if temporal changes in forest structural factors differed between regions that had implemented restoration programs and those that had not, we randomly sampled 200 points (with a minimum distance of 1 km between points) for each region within each of the 11 ecoregions. For each point, we extracted the difference in the five structural factor values occurring between 2000 and 2020. We constructed a linear mixed-effects model for each forest structural factor, with its temporal change as the response variable, and regions with different types of restoration programs (i.e., A, B, and C) as the fixed effect, where region C (areas without any restoration programs) was used as the reference level. Given the significant differences in ecological conditions49 and forest changes (Table 1 and Fig. 1) across the 11 ecoregions, we included ecoregions as a random effect. For comparison, we also constructed a linear regression model excluding ecoregion as a random effect and evaluated model performance using Akaike’s information criterion (AIC)102. Model fit was significantly better for all structural factors when ecoregions were included as a random effect (Supplementary Table 15). Linear mixed-effects models were constructed using packages “lme4”103 and “lmerTest”104 in R 4.3.1105. Furthermore, to determine differences in the way forest structural factors changed over time in relation to management regimes, we paired forest comparisons for two time periods: (1) natural vs. planted forests established before 1990, and (2) restored natural vs. planted forests converted from unforested land cover since 200098. In each of the 11 ecoregions, we randomly sampled 200 points (with a minimum distance of 1 km between points) in forests managed according to each regime. We then used linear mixed-effects models to compare changes in each forest structural factor between contrasting forest management regime pairings, with management regimes as a fixed effect. Ecoregions were included as a random effect, as comparisons with linear regression models without this random effect found a significantly better model fit (Supplementary Table 16). Habitat modeling for forest birds We conducted Ecological Niche Modeling (ENM) as a standard approach to delineate suitable habitat106 for each of 402 non-migratory forest bird species using MaxEnt 3.4.4107. We focused on forest birds as a proxy for biodiversity changes because these have been established as robust indicators of biodiversity status, environmental health, and ecosystem integrity45,46, and also because of the abundance of available occurrence data from citizen science databases compared to other taxa (e.g., mammals, amphibians, reptiles, etc.)108. Fully migratory birds were excluded from our analyses due to their complex seasonal habitat requirements and occupancy states (e.g., varying across passage, breeding, and wintering ground), which result in weaker and more variable associations with forest habitats compared to non-migratory species47,48. We collated all bird occurrence data recorded on three citizen science databases, namely eBird (http://www.ebird.org/)109, Global Biodiversity Information Facility (GBIF; https://www.gbif.org/)110, and China Bird Watching Record Center (http://www.birdreport.cn/), from across China over the period 2019 to 2021. Throughout, we followed the taxonomy and nomenclature of BirdLife International111, and matched every raw data record of a species to its scientific binomial. Based on the migration and habitat classification of AVONET, we retained occurrence records of non-migratory (i.e., sedentary and partially migratory) species primarily inhabiting forests (i.e., tall tree-dominated vegetation with more or less closed canopy, including palm forest) and woodlands (i.e., medium stature tree-dominated habitats, including Acacia woodland, riparian woodlands, mangrove forests, forest edges, also more open parkland with scattered taller trees)112. To ensure the quality and accuracy of occurrence data, we first excluded any eBird records that had not been vetted through the eBird review process, as well as any GBIF records with coordinate uncertainty greater than 1 km. We then compiled the elevational distribution range for each species113,114 and removed any occurrence records that fell outside the species’ known elevational limits. Next, we used the package “bdc”115 to remove: (1) records with duplicated, equal, or zero coordinates, and those with coordinates less precise than two decimal places; (2) records within 1 km of country or province centroids, and within 2 km of the capital city (i.e., Beijing), to address spatial inaccuracies resulting from the automated geo-referencing of imprecise locality descriptions; (3) records within 100 m around zoos and museums to avoid the inclusion of captive/escaped individuals or preserved specimens; and (4) geographic outliers for each species using a quantile method to reduce species misidentification errors116. Finally, we used the package “spThin”117 to remove duplicated occurrences within a 1 km radius for each species, thereby reducing spatial auto-correlation biases118. All species with more than 20 records in this screened and thinned dataset were retained for analyses, resulting in 141,670 occurrence point records for 404 species in total (Supplementary Table 17). This represents approximately one quarter of all bird species and over 80% of non-migratory forest bird species recorded in China112,119. To mitigate any spatial bias in sampling effort for species occurrence data, we applied a target-group strategy to weight the selection of background points120. We pooled all screened and thinned occurrence points before the removal of species with insufficient occurrence points, then removed all duplicate occurrence points with the same coordinates (regardless of the species they related to). A kernel density analysis was conducted on these remaining points to generate a probability surface with 20 contours representing the spatial pattern of survey effort. Subsequently, this probability surface was used to weight the random selection of 10,000 background points. Finally, values for the 13 environmental layers for 2020 were extracted to these background points, and to all species occurrence points, and then the “sample-with-data” format was used to build models121. In our ENM modeling, we correlated species occurrence points with the five forest structural (mean tree cover, tree cover SD, mean canopy height, canopy height SD, and forest fragmentation index), five climatic (annual mean temperature, isothermality, temperature seasonality, annual precipitation, and precipitation seasonality), and three topographic factors (slope, aspect northing, and distance to standing water). The selection of these predictors was based on well-established ecological associations with forest bird niche requirements and habitat preferences122,123. We excluded land cover type (i.e., forest vs. non-forest) because this was defined based predominantly on tree cover and height, which were already accounted for by forest structural factors, and varied substantially according to definitions under different land cover datasets (Supplementary Table 14)124. In a more ecological context, land cover type has less direct influence on animals’ habitat use than forest structural factors and thus provides a weaker predictor of habitat suitability35,36. We downloaded monthly maximum temperature125, minimum temperature126, and precipitation datasets127 for 2000 and 2020, and generated annual mean temperature, isothermality, temperature seasonality, annual precipitation, and precipitation seasonality with a 1 km resolution for each year using the package “dismo”128. The strong association between species habitat and climatic conditions is well-established129, and these five climatic variables have been shown to be informative in modeling birds’ habitats at the national spatial scale130. To include topographic factors, we downloaded a 30 m resolution digital elevation model (ASTER GDEM V3; https://www.gscloud.cn/) and generated 1 km resolution layers for elevation, slope, and aspect using ArcGIS. We further calculated an aspect northing layer as a cosine function of actual aspect, with northerly values closer to 1 and southerly values closer to -1131. Additionally, we obtained vector layers of perennial rivers and lakes from the National Catalogue Service for Geographic Information and produced a layer representing the Euclidean distance of each cell to the nearest standing water. Elevation was excluded from ENMs to avoid imposing unreasonable restrictions on habitat suitability when projecting species’ ecological niches, which would reduce the temporal transferability of ENMs. To assess collinearity among the 13 environmental factors, we calculated variance inflation factors (VIFs) using the “usdm” package132. Among these factors, only mean tree cover had a VIF value (10.369) marginally exceeding the conventional threshold of 10 (Supplementary Table 18), which was attributed to its strong correlation with mean canopy height (Pearson’s correlation coefficient, r = 0.895). Despite this collinearity, we retained both tree cover and canopy height in our ENMs because: (1) both horizontal (tree cover)36 and vertical (canopy height)35,133 components have been demonstrated to be ecologically meaningful predictors for modeling forest bird habitat suitability, and (2) the machine learning algorithm of MaxEnt is robust to predictor collinearity134,135. To further validate our approach, we conducted a comparative analysis of model performance using the area under the receiver operating characteristic curve (AUC)136. Under identical model configurations (features = linear, quadratic, and product functions; regularization multiplier = 1) and using the same occurrence and background points, models incorporating the complete set of predictors (AUC = 0.954 ± 0.060) significantly outperformed those excluding mean tree cover (AUC = 0.953 ± 0.062; paired-samples T-tests, t (403) = 10.701, P < 0.001, Cohen’s d = 0.532, 95% Confidence Intervals = [0.0009, 0.0013]) and mean canopy height (AUC = 0.953 ± 0.061; paired-samples T-tests, t (403) = 12.385, P < 0.001, Cohen’s d = 0.616, 95% Confidence Intervals = [0.0008, 0.0011]). This result substantiated that the inclusion of both structural factors enhanced model predictive capacity, despite their inherent correlation. To control for model complexity and ensure model temporal transferability137,138, we restricted ENMs to only linear, quadratic, and product functions. The regularization multiplier for each species was optimized by fitting a set of models with a tuning multiplier ranging from 0.1 to 2.0 (with a 0.1 interval)106, where the AICc value of each model was calculated using the package “ENMeval”139. The regularization multiplier with the lowest AICc was selected for the final model for each species (Supplementary Table 17). Using optimized occurrence points, background points, and model settings, we built a MaxEnt model for habitat suitability of each species under the scenario 2020, and transferred output logistic habitat suitability to a binary suitable/unsuitable habitat map by applying a threshold that maximized the sum of training sensitivity and specificity140. We evaluated model accuracy using AUC, and thus excluded the Great Tit Parus major (AUC = 0.690) and Eurasian Tree Sparrow Passer montanus (AUC = 0.696; Supplementary Table 17) based on the conventional threshold of 0.7 for meaningful model discrimination136. Analyses of changes in suitable habitat and species richness for forest birds To quantify habitat change during the interceding two decades, we back-cast ENMs fitted using occurrence data and environmental variables for 2020 to scenario 2000 (climatic factors for 2000 + forest structural factors for 2000 + constant topographic factors). This backward projection was based on the concept of ecological niche conservatism, where the responses of a species to environmental variables remain identical or similar over time, especially for short time spans92,141,142, although some studies have demonstrated that species can shift their realized niches at varying rates90,91. However, the mean rates of niche change across species are much slower than the projected pace of climate change over decadal timescales92, which justifies the reliability of our ENM projections. While various previous studies have proven the reliability of ENM projections106,143, we also tested the temporal transferability of our own ENMs. Following the procedures detailed above, we collated and processed occurrence data for non-migratory forest birds recorded in China over the period 1999 to 2001 and retained the 13 species that had more than 20 occurrence points. We extracted values of habitat suitability under scenario 2000 (back-projected from the 2020 ENMs) to occurrence points of corresponding species and to the 10,000 random background points weighted by contemporaneous spatial survey efforts. We calculated AUC for each species using the package “pRoc”144. All 13 species had AUC > 0.7, with an average of 0.849 ± 0.087 (Supplementary Table 19), indicating an overall good temporal transferability for our ENMs138,143. To parse the effects of climate change and forest restoration on species habitat during these two decades, we then projected the 2020 ENMs to simulation scenario A (climatic factors for 2020 + forest structural factors for 2000 + constant topographic factors) and scenario B (climatic factors for 2000 + forest structural factors for 2020 + constant topographic factors). This approach is again grounded in the concept of ecological niche conservatism92,141. As such, the ENMs under scenario A simulated the effects of climate while controlling for changes in forest structural factors. Conversely, the ENMs under scenario B simulated the effects of forest structural factors while controlling for climate change. To characterize the relationship between species traits and habitat change over the two decades, and to explore variation between species’ responses to climate change and forest restoration separately, we compiled six bird species traits potentially related to habitat use and distribution, namely centroid latitude and longitude112, midpoint and range of distributional elevation113,114, habitat breadth145, and habitat openness (category: open or dense)112. The selection of these traits was based on their potential effects on species’ responses to climate change and/or forest restoration. For example, species with higher latitudinal and elevational distributions are generally expected to exhibit range shrinkage under global warming69. Generalist species that can thrive in a wider variety of habitats and broader elevational ranges are expected to have greater potential to adapt to novel climatic and land cover conditions compared to specialists73. In addition, given that forest restoration efforts in China are spatially clustered146, species’ distribution patterns are also expected to affect their responses to forest change. We constructed three candidate linear regression model sets (changes from scenario 2000 to scenario 2020, to scenario A, and to scenario B – as defined above). Each candidate model set included the change in suitable habitat area of each species as the response variable, different combinations of the six species traits as explanatory variables, and the baseline habitat area in 2000 as an offset (with all continuous variables scaled). For each set of candidate models, we used the package “MuMIn”147 to perform a multi-model averaging, selecting top models with ΔAICc ≤ 2 to obtain parameter estimates148. Binary habitat maps (i.e., suitable vs. unsuitable habitat) for the 402 forest bird species were stacked to produce a heat map of predicted species richness under each of the four scenarios (scenarios 2000, 2020, A, and B). Differences between scenarios were calculated to illustrate spatial changes in species richness patterns. We used linear mixed-effects models to compare the changes in species richness from scenario 2000 to scenario B between regions with forest restoration programs (i.e., A and B) and without any restoration programs (i.e., C) and two pairs of forests with different management regimes (i.e., NF vs. PF and rNF vs. rPF), based on data extracted to random points. The 11 ecoregions were included as a random effect, as model comparisons indicated a significantly better fit for linear mixed-effects models with ecoregions as the random effect (Supplementary Table 20). Finally, we used linear mixed-effects models to explore the effects of changes in forest conditions on species richness, with the change in species richness from scenario 2000 to scenario B as the response variable and scaled changes in forest structural factors as fixed effects. Again, the 11 ecoregions were included as a random effect, as the global linear mixed-effects model significantly outperformed the global linear regression model without this random effect (Supplementary Table 21). We constructed a candidate model set comprising different combinations of factors, and applied a multi-model averaging approach to the top models with ΔAICc ≤ 2 to obtain parameter estimates. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Acknowledgements We thank eBird, Global Biodiversity Information Facility, China Bird Watching Record Center, and all contributors for sharing the occurrence records of forest birds, as well as the authors who publicly shared the data on climate, land cover, and forest structure that were used in this study. This manuscript was edited for English by NovaScript. This work was supported by funding from the National Natural Science Foundation of China (32000354 and 32571933 to B.W., 32270540 to Y.X., 32470522 to T.Z., and 32400403 to X.H.), the Innovation Team Funds of China West Normal University (KCXTD2024-5 to B.W.), the Institute of Zoology, Chinese Academy of Sciences (2023IOZ0104 to T.Z.), the Chengdu Research Base of Giant Panda Breeding (2020CPB-B08 to P.Q.), and the Society of Entrepreneurs and Ecology (SEE Foundation, 2021-2022 to P.Q.). Author contributions B.W. conceived the study, led the analyses and wrote the original draft of the manuscript. Z.W. contributed to analyses and commented on the manuscript. P.Q. acquired data and commented on the manuscript. T.Z. contributed to analyses and commented on the manuscript. X.H. acquired data. N.Y. acquired data. X.Z. contributed to analyses and visualization. Y.X. conceived the study, reviewed, and edited the manuscript. Peer review Peer review information Nature Communications thanks Augusto Piratelli, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Data availability The occurrence records of forest birds and data generated in this study have been deposited in figshare (10.6084/m9.figshare.29348429)149. Land cover datasets are available at refs. 57,93–98; raw data for tree cover is available at ref. 99; raw data for canopy height is available at Global Land Analysis & Discovery laboratory (https://glad.umd.edu/dataset/GLCLUC2020); raw climatic data are available at refs. 125–127; digital elevation model is available at Geospatial Data Cloud, Computer Network Information Center, Chinese Academy of Sciences (http://www.gscloud.cn); vector layers for national boundaries, rivers and lakes are available at National Catalogue Service for Geographic Information, National Geomatics Center of China (http://www.webmap.cn/). Code availability R codes that can be used to reproduce the results of this paper are available at figshare (10.6084/m9.figshare.29348429)149. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Bin Wang, Email: wangbin513@cwnu.edu.cn. Yu Xu, Email: xuyu608@gznu.edu.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41467-025-65751-7. References Associated Data Supplementary Materials Supplementary Information Reporting Summary Transparent Peer Review file Data Availability Statement The occurrence records of forest birds and data generated in this study have been deposited in figshare (10.6084/m9.figshare.29348429)149. Land cover datasets are available at refs. 57,93–98; raw data for tree cover is available at ref. 99; raw data for canopy height is available at Global Land Analysis & Discovery laboratory (https://glad.umd.edu/dataset/GLCLUC2020); raw climatic data are available at refs. 125–127; digital elevation model is available at Geospatial Data Cloud, Computer Network Information Center, Chinese Academy of Sciences (http://www.gscloud.cn); vector layers for national boundaries, rivers and lakes are available at National Catalogue Service for Geographic Information, National Geomatics Center of China (http://www.webmap.cn/). R codes that can be used to reproduce the results of this paper are available at figshare (10.6084/m9.figshare.29348429)149.