Forest dynamics and responses to climate change and human activities in the arid and semiarid regions of the Altai Mountains, China

Abstract

Understanding the driving mechanisms of forest changes is of great significance for developing effective adaptation strategies to mitigate the impacts of climate change and human activities on ecosystems. This study used Theil–Sen median trend analysis, Mann–Kendall test, contribution rate decomposition, partial least squares, geodetector and residual analysis to explore the impact of climate change and human activities on the forest coverage area and NDVI of the Altai Mountains. Results show that changes in forest cover are driven by both forest management policies and climate change. Among them, forest management policy is the main factor. However, there are differences in the driving mechanisms in different altitude zones: in the alpine and subalpine zones, the promoting effects of natural death and climate change bring the forest coverage area toward a dynamic balance, while under the combined effects of human activities and climate change, the forest coverage area in the low mountain zones shows an expansion trend. For forest NDVI, the analysis results of the six scenarios show that the joint action of climate change and human activities promotes the growth of forest NDVI in the largest proportion (50.20%); the impact of climate change on forest NDVI is greater than that of human activities, and most of it is a promotion effect (30.28%). Forest degradation is mainly caused by human activities (19.39%), especially in the edge areas of the forest. Among climate factors, precipitation and snowmelt water are the main controlling factors for forest growth. Snowmelt water from March to April is an important water source before the growing season. This study provides the important scientific basis for forest management and strategic planning in the Altai Mountains.

摘要

阿尔泰山森林动态变化及其对气候变化与人类活动的响应

理解森林变化的驱动机制对于制定有效的适应策略以减轻气候变化和人类活动对生态系统的影响具有重要意义。本研究综合运用Theil–Sen median趋势分析、Mann–Kendall检验、贡献率分解、偏最小二乘法、地理探测器及残差分析等方法，探讨了气候变化与人类活动对阿尔泰山森林覆盖面积及NDVI的影响。结果表明，森林覆盖面积的变化由森林管理政策和气候变化共同驱动，其中森林管理政策是主要因素。然而，不同海拔带的驱动机制存在差异：在高山和亚高山带，自然死亡与气候变化的促进效应使森林覆盖面积趋于动态平衡;而在人类活动与气候变化的共同作用下，低山带森林覆盖面积呈扩张趋势。对于森林NDVI，6种情景分析结果显示，气候变化和人类活动的共同作用促进森林NDVI增长的区域占比最大(50.20%);气候变化对森林NDVI的影响范围大于人类活动，且多表现为促进作用(30.28%)，而森林退化主要由人类活动引起(19.39%)，尤其在森林的边缘区域。在气候因子中，降水和积雪融水是森林生长的主要控制因子，3–4月的积雪融水是生长季前重要的水源。本研究为阿尔泰山森林管理与战略规划提供了重要的科学依据。

INTRODUCTION

Forests are vital ecosystems that facilitate the exchange of energy and materials among the atmosphere, moisture and soil (Huang et al. 2019; Jia et al. 2016). They play crucial roles in safeguarding global ecological stability and promoting sustainable socioeconomic growth (Petrokas et al. 2022). However, with global warming and intensified human activities, various forest indicators have undergone significant changes, affecting biodiversity, water resources and carbon storage (Foley et al. 2005; Hansen et al. 2013; Martone et al. 2018; Schmeller et al. 2018). Forest cover and the normalized difference vegetation index (NDVI) are key indicators that reflect the spatiotemporal dynamics of forest conditions, including deforestation, afforestation, forest degradation, improvements caused by human activities and climate change (Tan et al. 2007). Analyzing changes in these indicators across different spatial and temporal scales can provide valuable insights for forest planning and sustainable management (Abdullah and Nakagoshi 2007).

In recent decades, global warming has increasingly impacted forest ecosystems (Bonan 2008; Norby and Zak 2011). Rising temperatures alter forest cover, particularly in regions where low temperatures limit growth, and forests migrate to relatively high latitudes and elevations (Tian 2020). Studies have shown that various forest types are expanding toward the poles, with temperate forests shifting northward and some mountain forests moving to higher altitudes in the Northern Hemisphere (Hirota et al. 2010; Lloyd and Fastie 2002; Meshinev et al. 2000; Parmesan and Yohe 2003). It is predicted that if temperatures rise by 2 °C by the 22nd century, temperate forests in the USA could shift northward by 100–530 km at a rate of 1.6–4.8 km per year (Casper 2010). These findings indicate that climate change is reshaping forest patterns. However, the response of forests to different climate factors varies across regions. Some studies have suggested that in arid regions, precipitation is the main climate factor influencing forest cover changes, especially in low- and mid-latitude regions (Palmate et al. 2014; Williamson et al. 2014). In contrast, temperature is the primary driver in humid regions (De Lombaerde et al. 2022; Li et al. 2024). The forest NDVI also shows a significant response to climate change, with considerable regional disparities (Jiang et al. 2017; Jin et al. 2014). For example, in Kazakhstan, precipitation is the main factor influencing forest greenness changes, whereas in the Tarim Basin, temperature is the dominant factor (Jiang et al. 2017). Moreover, studies have demonstrated that mountainous forests are particularly vulnerable to the adverse effects of climate change (Chaturvedi et al. 2010). Understanding forest responses to climate change across different regions can provide scientific support for the development of adaptation strategies.

In addition to climate change, human activities are key factors affecting forest change (Junqueira et al. 2017). Many activities, such as commercial logging, deforestation, overgrazing and mining, have severely impacted forest growth, leading to significant forest loss (Chung et al. 2024; Girma et al. 2012; Martínez-Ruiz et al. 2021). Conversely, afforestation and the establishment of nature reserves have promoted forest growth and expanded forest cover (Jia et al. 2015; Suárez-Muñoz et al. 2023; Xu et al. 2022b; Yang et al. 2023). In China, from the 1950s to the 1990s, forest policies focused on maximizing economic benefits, resulting in a substantial increase in timber harvests from 20 million m³ · year⁻¹ to 63 million m³ · year⁻¹ (Zhang et al. 2000). This unsustainable level of timber harvesting led to a decline in natural forests, which account for 30% of the total forest cover in China, and a 32% decrease in the stocking of natural forests per unit area (Zhang et al. 1999). In 1998, China initiated the Natural Forest Conservation Program (NFCP), which shifted forest policy to prioritize forest ecological services rather than wood production. Consequently, timber harvests from China’s natural forests decreased from 32 million m³ in 1997 to 23 million m³ in 1999 (Jia et al. 2015). However, the majority of studies on forest changes following policy implementation have focused on a few years after policy adoption (Jia et al. 2015), with limited research on forest cover changes over the past two decades. Moreover, most of these studies have examined forest changes across the entire region of China, the Three-North Shelter Forest Region (TNSFR) of China (Jia et al. 2015), and northeastern China (Yu et al. 2011), while northwestern China has been neglected. Additionally, human activities not only affect forest cover changes but also significantly impact forest greenness. Studies have indicated that excessive exploitation, industrial activities such as chemical factories and mining, and tourist activities have led to forest browning (Jiang et al. 2017). On the other hand, the fertilization effect of carbon dioxide emissions promotes vegetation growth (Leakey et al. 2009; Zhu et al. 2016), and the implementation of certain ecological projects has contributed to forest greening (Huang et al. 2012). Studying the impact of human activities on forest changes can inform the development of effective forest conservation policies.

The Altai Mountains, a typical mountain system in arid and semiarid regions (Dong et al. 2023; Lian et al. 2022; Yang et al. 2024), host unique vegetation types (Cao et al. 2023). The area is one of the two major mountain natural forest regions in Xinjiang, China, and the only distribution area of the Siberian montane southern taiga forest in China (Cao et al. 2023; Xiong 2020). The Altai Mountains serve as a vital ecological barrier for Xinjiang as a whole and the arid regions of northwest China (Altai and Zhao 2019), contributing significantly to water conservation, biodiversity protection and the maintenance of multifunctional carbon sinks (Duan et al. 2024). However, with global climate change, the region has experienced significant ‘warming and wetting’ (Yan et al. 2023; Zhang et al. 2021). Additionally, a series of forest conservation policies have been implemented in the Altai Mountains since the 1990s, establishing numerous ecological protected areas (Zhang 2017). Moreover, human activities, including tourism, have intensified in the region (Zhang 2017). Nevertheless, despite these changes, research on the combined impact of climate change and human activities on forest dynamics in this area is lacking. Furthermore, previous studies on the impact of climate change on forests have focused mainly on temperature and precipitation (Chaturvedi et al. 2010; Jiang et al. 2017), whereas the Altai Mountains, which are rich in snow resources (Lou et al. 2013; Yu et al. 2023), rely heavily on snowmelt as a critical water source for vegetation growth (Kong and Pang 2012; Wang et al. 2023; Yu et al. 2023). The relationship between forest changes and snowmelt in the Altai Mountains remains unclear.

On the basis of this analysis, our research aimed to (i) reveal the spatiotemporal dynamics of forest cover and forest NDVI in the Altai Mountains; (ii) identify the key factors driving forest changes in the arid and semiarid regions of the Altai Mountains considering climate change and human activities and (iii) reveal the extent to which snowmelt influences forest dynamics in the Altai Mountains. Our study will inform the development of effective forest management policies, ensuring the ecological security of the region.

MATERIALS AND METHODS

Study area

The study area is located within the Altai Mountains in China (45°47′–49°10′ N, 85°27′–91°01′ E). Owing to its distance from the sea, this region has a moderate temperate continental arid and semiarid climate characterized by shorter, cooler summers and extremely cold winters, with a mean annual temperature ranging from 0.7 °C to 4.9 °C (Fu et al. 2017). Precipitation in this region is driven primarily by the forced uplift of Atlantic water vapor carried by westerly circulation, resulting in annual precipitation ranging from 200 to 300 mm in low-elevation mountains to more than 600 mm in high-elevation mountains (Lang et al. 2020). Additionally, the Altai Mountains host vast cold-temperate needle-leaved forests, subalpine meadows and mountain tundra. The natural forests in this region belong to the European–Siberian taiga biome, which predominantly consists of Siberian larch, Siberian red pine, Siberian spruce and Siberian cold mountain species. This area represents the sole distribution region of the Western Siberian taiga within China, showcasing ancient and rare tree species and fostering primitive and distinctive communities. As a result, the Altai Mountains host a unique forest ecosystem of significant scientific and ecological value, warranting extensive research and ecological protection efforts (Xiong 2020).

Data sources

Land use and cover change data

We obtained land use and cover change (LUCC) gridded data with a spatial resolution of 0.003° (~300 m at the equator) and referenced the WGS84 geographic coordinate system from the Climate Data Store (CDS) (https://cds.climate.copernicus.eu/datasets/satellite-land-cover?tab=download) for the period from 1992 to 2020. This study used LUCC data to extract the forest cover (LUCC-forest) of the Altai Mountains. LUCC-forest data were extracted for different forest types, including evergreen needle-leaved forest, deciduous broad-leaved forest, deciduous needle-leaved forest and mixed forest, corresponding to LUCC codes 70, 60/61, 80 and 90, respectively. To calculate the forest cover area, the WGS84 geographic coordinate system was converted to the Universal Transverse Mercator (UTM) projection system. The accuracy of the LUCC-forest data was validated via the confusion matrix method (Zhang et al. 2015), where the 2020 10-m LUCC data generated from Sentinel-2 data (https://browser.dataspace.copernicus.eu) were considered the ‘true values’. The overall accuracy value of the LUCC-forest data is 97.7%, which corresponds to a kappa of 0.6 and an underestimation of LUCC-forest cover compared with the ‘true value’ of 14.1% in the Altai Mountains. The Sentinel-2 data were resampled to 0.003° to match the spatial resolution of the LUCC-forest data via the WGS84 geographic coordinate system.

NDVI products

The MOD13Q1 NDVI dataset was acquired from the National Aeronautics and Space Administration (NASA) website (https://ladsweb.modaps.eosdis.nasa.gov/search/order/1/MOD13Q1); these data are MODIS NDVI 16-day maximum synthetic product data with a spatial resolution of 250 m. To analyze the spatial and temporal characteristics of the NDVI in the Altai Mountains, the maximum value composite method (MVC) (Li et al. 2017; Yan et al. 2022) was applied to extract the annual maximum NDVI values. The data were converted from hdf format to tif format using the MODIS Reprojection Tool (MRT), and the coordinate system was set to the WGS84 geographic coordinate system with a spatial resolution of 0.003°. Since this study focused on analyzing forest NDVI spatial trends, forest degradation, improvement levels and residuals, ensuring that the spatial coverage of the forest NDVI remained consistent across years was essential. To eliminate interference from nonforest NDVI and ensure that each grid represents only the forest NDVI, the common forest cover area from 2000 to 2020 was used as a mask to extract the annual forest NDVI for the study area. Although both the LUCC-forest data and MOD13Q1 NDVI data have a spatial resolution of 0.003°, the snapRaster tool was used to ensure spatial alignment between the two datasets and prevent misalignment.

ERA5-Land temperature and precipitation data

The monthly 0.1° ERA5-Land temperature and precipitation dataset from 1992 to 2020 was obtained from the CDS (https://cds.climate.copernicus.eu/#!/home). To increase the accuracy of the ERA5-Land temperature and precipitation data in the Altai Mountains, we employed the residual correction method to correct the temperature data and the coefficient correction method to adjust the precipitation data, significantly enhancing the ability of the ERA5-Land dataset to represent temperature and precipitation in the Altai Mountains region (Zhang et al. 2024). We obtained several key variables: the annual mean temperature (TEM_Y), growing season mean temperature (TEM_G), mean temperature from October of the previous year to April of the current year (TEM_P), annual cumulative precipitation (PRE_Y), growing season cumulative precipitation (PRE_G) and cumulative precipitation from October of the previous year to April of the current year (PRE_P). These data were used to analyze the response of forest changes in the Altai Mountains to climate variability.

Snowmelt data

The monthly snowmelt dataset for the Altai Mountains from 2000 to 2020 was acquired from the National Scientific Data Centre for Glacial Permafrost and Desert (http://www.ncdc.ac.cn/portal/). This dataset uses high-resolution monthly precipitation and temperature data as inputs and employs a degree-day factor model to calculate snowmelt. The model output has been validated against snowfall, snow depth, snow extent and snow water equivalent, resulting in a high-resolution (1 km) monthly snowmelt dataset for China from 1951 to 2020 (Yang et al. 2022). From this dataset, we derived annual cumulative snowmelt (SM_Y), growing season cumulative snowmelt (SM_G) and cumulative snowmelt from March to April (SM_34). Snow accumulation and stability in the study area occur from October of the previous year to April of the current year, with no significant snowmelt during this period, and the primary snowmelt period is from March to April (Zhang et al. 2014). Therefore, we selected SM_34 for analysis rather than using the temperature and precipitation data from October of the previous year to April of the current year. These data were used to analyze the response of forest changes in the Altai Mountains to variations in snowmelt.

Other data

Different mountain belt data: these data were derived from 12.5-m digital elevation model (DEM) data (https://search.asf.alaska.edu/), which define four mountain belts: alpine (>3000 m), subalpine (2500–3000 m), middle mountain (2000–2500 m) and low mountain (<2000 m) (Xiong 2020). These mountain belts were converted into vector files via ArcGIS 10.0.

Forest age class data: these data, which were provided by local forestry departments, categorize forests into five age groups: young, middle-aged, near-mature, mature and overmature.

These classifications were used to extract the forest cover area, forest NDVI, forest degradation and forest improvement regions across different altitude belts and forest age classes. The mountain belt and forest age class data are both in the WGS84 geographic coordinate system.

Methods

Theil–Sen median and Mann–Kendall test

The Theil–Sen median trend analysis and the Mann–Kendall test were used to analyze the trends in forest NDVI changes. The Theil–Sen median method, a robust nonparametric statistical method, was used for trend calculation. This method is computationally efficient, resistant to measurement errors and outlier data, and well suited for analyzing long-time series data (Jahani et al. 2018). The formula is as follows:

where S_NDVI represents the trend in the NDVI, S_NDVI > 0 indicates that the NDVI is increasing, S_NDVI < 0 indicates that the NDVI is decreasing and S_NDVI = 0 indicates that the NDVI has not changed. Moreover, the median represents the mean median value, j and i represent the year and n represents the length of the study period.

The Mann–Kendall test is a rank-based nonparametric test for detecting monotonic trends in a time series. The Mann–Kendall test is also used prior to other nonparametric trend tests because of its simplicity and wide range of applicability (Kumar et al. 2010). The formula is as follows:

The statistic S is approximately normally distributed with the variance as follows:

The standardized test statistic Z is computed by

where the statistic Z follows the standard normal distribution with a mean of 0 and a variance of 1. The critical value is 1.96 at the significance level a = 0.05; that is, if |Z| ≥ 1.96, the trend has passed the significance test with a confidence interval of 95%.

In this study, by synthesizing the S_NDVI and Z, the trends were divided into five trend levels (Supplementary Table S1) (Wang et al. 2022; Wei et al. 2022).

Contribution partition method

We adopted the contribution partition method to quantify the regional contributions to the forest NDVI over the Altai Mountains at the grid cell scale. The formula is as follows (Ahlström et al. 2015):

where x_jt is the annual NDVI flux anomaly for grid cell j at year t, X_t is the flux anomaly of the whole Altai Mountains and f_j is the average relative anomaly for grid cell j, weighted with the absolute anomaly |X_t| in the Altai Mountains. Grid cells with larger contributions in governing the holistic NDVI received higher absolute scores, whereas grid cells with less contribution received smaller absolute scores. Positive scores indicate positive contributions, whereas negative scores represent negative contributions. The total contribution of each region was the integrated contribution of all the grid cells in that region (Li et al. 2021).

Partial least squares method

The partial least squares (PLS) method was employed to analyze the response of forest cover change to climate change. PLS is a robust multivariate technique that combines the features of principal component analysis and multiple regression (Abdi 2010), providing a more parsimonious and statistically robust approach than principal component regression does (Fu et al. 2015). This method effectively addresses the overfitting issue arising from the presence of multiple collinearities among highly correlated independent variables. Moreover, variable importance projective scores (VIP scores) are utilized to assess the statistical contribution of each independent variable to the overall fitted PLS regression model across all latent vectors (Matthes et al. 2015) and to test the significance at the 0.05 level. In this study, the dependent variable is the annual forest cover area of the Altai Mountains from 1992 to 2020, and the independent variables include TEM_Y, TEM_G, TEM_P, PRE_Y, PRE_G, PRE_P, SM_Y, SM_G and SM_34 from 1992 to 2020. The PLS method was implemented via SIMCA-P 15.0 software.

In this study, the VIP scores and standardized regression coefficients were calculated via PLS to assess the contributions of climate variables to forest cover change. Higher VIP scores indicate greater influence. Climate variables with VIP scores above 1 were considered significant independent variables influencing forest cover change, and those ranked first were regarded as the most influential factors (Chong and Jun 2005). Variables with positive standardized regression coefficients promote forest cover expansion, whereas those with negative coefficients inhibit forest cover reduction.

Geodetector analysis

A geodetector is a statistical method for detecting spatial heterogeneity and elucidating its underlying driving factors; it is free of linearity assumptions and has an elegant form and clear physical implications. Furthermore, the q statistic, known as the geographic probe, has been widely employed to assess spatial heterogeneity, identify explanatory factors, explore interactions between variables (Wang and Xu 2017) and has been applied in multiple fields of natural and social sciences (Chen et al. 2022). Geodetectors are classified into four types, namely factor_detector, interaction_detector, risk_detector and ecological_detector. In this study, we employed Factor_detector and interaction_detector to comprehensively examine the impact of climate change on the forest NDVI from multiple perspectives (Wei et al. 2022). Factor_detector was employed to explore the spatial heterogeneity of the dependent variable Y and to evaluate the explanatory power of each factor X on Y, quantified by the q value. The q value ranges from 0 to 1, where higher values indicate stronger explanatory power of X for Y, and lower values indicate weaker influence. Interaction_detector was used to identify interactions between different risk factors, X_S, by assessing whether the combined effect of two factors, X₁ and X₂, increases or decreases their explanatory power on Y. This involved calculating the individual explanatory powers q(X₁) and q(X₂) and comparing them with the interaction term q(X₁∩X₂) to determine the interaction type between factors (Supplementary Table S2) (Wang and Xu 2017):

where h = 1, …, L represents the strata of variable Y or factor X, which refers to the classification or division; N_h and N represent the number of units in stratum h and in the entire region, respectively; $σh2$ and $σ2$ are the variances of Y and h, respectively. SSW and SST represent the within sum of squares and the total sum of squares, respectively. The Factor_detector and Interaction_detector calculations were performed via geodetector software in Excel. Before performing the geodetector analysis, the raster data for PRE_Y, PRE_G, PRE_P, TEM_Y, TEM_G, TEM_P, SM_Y, SM_G and SM_34 were resampled to a spatial resolution of 0.003° to align with the forest NDVI data, with all data using the WGS84 geographic coordinate system.

Residual analysis

The forest NDVI is influenced by the combined effects of climate and human activities. Residual analysis has been widely used to separate the influences of human activities and climate change on the vegetation NDVI from the observed NDVI (Herrmann et al. 2005; Jiang et al. 2017). In this study, we first established a predictive regression model for the forest NDVI using climate variables, including temperature, precipitation and snowmelt. The model was then used to predict the pixel-level forest NDVI, denoted as NDVI_P (Evans and Geerken 2004; Herrmann et al. 2005). The difference between the NDVI_P and the observed forest NDVI (NDVI_T) yielded the residual NDVI (NDVI_R), which represents the NDVI unaffected by climate factors. Finally, by analyzing the slopes of NDVI_T, NDVI_P and NDVI_R, we categorized the study area into six scenarios on the basis of the positive or negative impacts of climate change and human activities on the forest NDVI (Table 1). These scenarios were designed to represent regions where the forest NDVI increased or decreased due to climate change, human activities or their combined effects (Table 1) (Liu et al. 2023; Xu et al. 2016). The equations are as follows:

where X_P, X_T and X_S represent precipitation, temperature and snowmelt, respectively, a, b and c represent coefficients, Z is a constant, ΔNDVI represents the change in the NDVI over the study period, n represents the study time scale and K_slope represents the slope of the NDVI. All raster data had a spatial resolution of 0.003° and the coordinate system used was the WGS84 geographic coordinate system.

RESULTS

Spatiotemporal dynamics of forest cover and forest NDVI in the Altai Mountains

The forest cover in the Altai Mountains was extracted from LUCC data spanning from 1992 to 2020. Distinct variations in forest cover changes were observed across different periods (Fig. 1a-e). There was a significant decrease in forest cover from 1992 to 1997, followed by a significant increase from 1997 to 2020 (Fig. 1a), which was particularly notable for deciduous needle-leaved forests (Fig. 1c). Furthermore, forest cover changes vary across different elevations. The forest cover in the subalpine and alpine belts remained relatively stable after 1997, whereas that in the low- and mid-mountain belts exhibited an increasing trend from 1997 to 2020 (Fig. 1f–i). In spatial terms, ~61.7% of the forest cover in this region is distributed in the low mountain belt, which primarily consists of deciduous needle-leaved forest (Supplementary Fig. S1), with mature forests accounting for 42.1% (Supplementary Table S3). Evergreen needle-leaved and deciduous needle-leaved forests account for the largest spatial proportions, covering 45.9% and 48.5% of the area, respectively (Supplementary Fig. S1). The evergreen needle-leaved forest is predominantly located in the northwestern region, whereas the deciduous needle-leaved forest is mainly distributed in the southeastern region (Supplementary Fig. S1). The alpine and subalpine belts are covered primarily by evergreen needle-leaved and deciduous needle-leaved forests, whereas deciduous broad-leaved forests and mixed forests are concentrated in the mid- and low-mountain belts (Supplementary Fig. S1).

Using the MODIS NDVI products, we analyzed the forest NDVI trends in the Altai Mountains. The results indicated a growing trend in the forest NDVI from 2000 to 2020 (Fig. 2a), and only mature forests passed the significance test (P < 0.01, Fig. 2b). The average NDVI value was 0.748, with a minimum value of 0.702 recorded in 2008 and a maximum value of 0.771 observed in 2016 (Fig. 2a). The overall trend in forest growth status was generally positive, as indicated by the increasing NDVI trend in 80% of the assessed areas (Supplementary Fig. S2). However, the degree of improvement gradually diminished from the southeast to the northwest (Supplementary Fig. S3). The results indicate that over the past 20 years, 68.63% of the forests in the Altai Mountains have experienced slight or significant improvement, with severe degradation accounting for only 0.98%. The degraded areas, which were located mainly near the forest edges (Supplementary Fig. S3a), were spatially scattered. While the extent of forest NDVI improvement and degradation varied among different forest types, age groups and altitudes, the regions achieving mild or greater improvement ranged from 47.78% to 75.41%, with severe degradation areas mostly limited to less than 1% (Supplementary Fig. S3b). This suggests an overall positive trend in forest growth in the Altai Mountains region.

We employed a contribution partitioning method (Eq. 6) to quantify the contributions of grid cells at the regional scale to the variation in the forest NDVI across the Altai Mountains via forest NDVI data from 2000 to 2020. The results revealed that 93.7% of the pixels presented positive contributions to the variation in the forest NDVI in the Altai Mountains, whereas 6.3% presented negative contributions. Grid cells with high positive contributions were primarily concentrated in the southeastern region of the mountain area (Supplementary Fig. S4a). Among the different age groups, mature forests made the greatest contribution to forest NDVI variation (39.5%), whereas young forests contributed the least (0.5%) (Supplementary Fig. S4b). Among the different forest types, deciduous needle-leaved forest made the greatest contribution to the variation in the forest NDVI (59.7%), followed by evergreen needle-leaved forest (37.5%) (Supplementary Fig. S4c).

Forest cover and NDVI response to climate variables

The contributions of climate variables to forest cover dynamics.

The VIP scores of PRE_Y, PRE_G, TEM_Y, TEM_G and TEM_P were all greater than 1 (Fig. 3a), indicating that these five factors significantly affected forest cover dynamics in the Altai Mountains. Among them, PRE_Y had the highest VIP score of 1.5, making it the dominant climatic factor influencing forest cover dynamics in the study area. PRE_G followed closely with a VIP score of 1.4, whereas TEM_Y, TEM_G and TEM_P had VIP scores of 1.1, 1.0 and 1.0, respectively. The VIP scores of PRE_P, SM_Y, SM_G and SM_34 did not exceed 1, although the VIP score of SM_34 was higher than those of SM_G and SM_Y (Fig. 3a). The regression coefficients were positive for all the variables except SM_G. Among them, PRE_Y and PRE_G had the largest regression coefficients and made the most significant contributions to the expansion of forest cover in the Altai Mountains (Fig. 3b).

The contributions of climate variables to forest NDVI dynamics.

The degree of influence, q, was determined via Factor_detector and represents the quantitative relationship between the forest NDVI and climate variables. Among the variables, SM_34 presented the highest q value of 0.25, whereas SM_G presented the lowest value of 0.04. The q values for precipitation remained stable and comparable across different periods, ranging from 0.20 to 0.22, whereas the q value for temperature was relatively small, ranging from 0.11 to 0.13 (Fig. 3c). Except for TEM_G, most variables were positively correlated with the forest NDVI. Notably, PRE_G and PRE_Y exhibited significant positive correlations with forest NDVI changes, with correlation coefficients of 0.5 and 0.37, respectively (Fig. 3c).

An interaction detector was employed to assess the combined effects of precipitation, temperature and snowmelt on the forest NDVI. The two-factor interaction effects of precipitation, temperature and snowmelt during different periods on the forest NDVI were greater than the individual effects of any single factor. Notably, interactions such as PRE_Y ∩ SM_G, PRE_Y ∩ SM_34 and PRE_G ∩ SM_34 presented q values of 0.34 (Supplementary Table S4). The two-factor interval was either nonlinearly enhanced or two-factor enhanced, indicating that the precipitation, temperature and snowmelt in different periods jointly affected the growth of forests in the Altai Mountains, with particular impacts observed from PRE_Y ∩ SM_G, PRE_Y ∩ SM_34 and PRE_G ∩ SM_34.

Climate-human effects on the forest NDVI

Distinguishing the regions within the Altai Mountains, the forest NDVI was found to be influenced by climate change and human activities through residual analysis. The linear regression analysis based on the forest NDVI residuals (Supplementary Fig. S5) revealed an increasing trend in the forest NDVI residuals, mainly in the southeast, with increasing rates of up to 0.014/a. Conversely, the residuals exhibited a decreasing trend in the northwestern and central regions, with rates as low as −0.012/a. These results indicate that the overall trends of the forest NDVI residuals have significant regional differences. By examining six scenarios of climate-human interactions on the forest NDVI (Supplementary Fig. S6), we found that the combined effects of climate change and human activities led to the largest area of forest NDVI increase, covering 50.2% of the region, predominantly in the southeast. Moreover, the influence of climate change on the forest NDVI was greater than that of human activities, with climate-driven growth occurring primarily in the northwest. Conversely, the impact of human activities largely inhibited forest growth, especially in the northwest. Areas where human activities promoted forest NDVI and climate change inhibited NDVI were minimal, as well as regions where both factors suppressed forest NDVI, each accounting for less than 0.1% of the total area (Supplementary Fig. S6).

DISCUSSION

Analysis of elements affecting forest cover change

Prior to 1997, forest cover in the Altai Mountains declined; after 1997, it showed an increasing trend (Fig. 1a). This shift aligns with the evolution of China’s forest conservation policies. Before 1997, the focus was on maximizing economic benefits from forests, with an emphasis on ecological benefits thereafter (Zhang et al. 2000, 2022). This pattern is consistent with findings from the TNSFR of China (Jia et al. 2015), southern Inner Mongolia (Deng 2010), Saihanba (Xu et al. 2022b) and the upper reaches of the Yellow River in Sichuan (Xu et al. 2024). Additionally, the ‘Grain for Green’ policy, which has converted cultivated land to forest in mountainous areas (Xiong 2020), along with the establishment of ecologically protected regions, has contributed to forest expansion (Chen et al. 2020; Hua et al. 2018; Xiong 2020). Forest cover changes vary across different regions worldwide and are largely influenced by local forest protection policies (Devaney et al. 2015; Gautam et al. 2004; Hedges et al. 2018,Pauleus and Aide 2020; Smith et al. 2024; Van Den Hoek et al. 2021). Moreover, climate change has also influenced forest cover changes in the Altai Mountains. In arid regions, insufficient precipitation relative to evapotranspiration can lead to tree mortality and environmental degradation (Cao et al. 2010). Our study further confirmed that precipitation (PRE_Y and PRE_G) had the most significant impact on forest cover in the Altai Mountains (Fig. 3a and b). However, after 2010, despite decreasing precipitation and snowmelt (Supplementary Fig. S7), forest cover continued to expand, indicating that human activity and policies played a more influential role than climate factors during this period. In summary, both forest protection policies and climate factors affect forest cover in the Altai Mountains, with forest protection policies being the primary factor.

Field surveys in the alpine and subalpine belts revealed severe forest degradation, characterized by extensive tree mortality and dieback (Supplementary Fig. S8). However, after 1997, forest cover in the alpine and subalpine belts remained stable (Fig. 1h and i), likely because the natural mortality of alpine and subalpine forests, coupled with the promoting effect of climate change (Fig. 3; Supplementary Fig. S6), led to a dynamic equilibrium in forest cover. In contrast, the forest cover of the low- and middle-mountain belts clearly increased after 1997 (Fig. 1f and g), likely due to afforestation efforts, the establishment of protected areas and grazing bans (Zhang 2017).

Analysis of the elements affecting forest NDVI changes

The response of the forest NDVI to climate change in the Altai Mountains is evident, with precipitation and snowmelt being the primary controlling factors. Previous studies have shown that in temperate continental and temperate mountain systems, forests are significantly affected by precipitation, especially in arid regions (Xu et al. 2022a), whereas in humid tropical regions, excess precipitation can be detrimental to forest growth (Wang et al. 2014). However, these studies did not consider the impact of snowmelt on the forest NDVI. Our study confirmed that SM_34 had the greatest influence on the forest NDVI, with a q value of 0.25 (Fig. 3c). This is because the Altai Mountains, which are located in arid and semiarid regions, rely heavily on snowmelt for water, and March to April is the peak period for snowmelt (Gafurov et al. 2016; Gessner et al. 2013; Ososkova et al. 2000).

Although SM_34 is a crucial water source for forest growth in the arid and semiarid Altai Mountains, high temperatures during the growing season can lead to substantial soil water evaporation (Jiang et al. 2017). Without sufficient precipitation to offset this loss, drought stress may impede forest growth (Ding et al. 2018; Park Williams et al. 2012). Our findings show that the combined effects of SM_34 ∩ PRE_G and SM_34 ∩ TEM_G on the forest NDVI are stronger than those of SM_34 alone (Supplementary Table S4), highlighting the significant interannual impact of climatic factors on the forest NDVI. Additionally, the forest NDVI in the Altai Mountains was negatively correlated with TEM_G and positively correlated with PRE_G (Fig. 3c). In 2008, high temperatures and low precipitation during the growing season led to drought, with an scPDSI value of −1.67 (Supplementary Fig. S9), resulting in the lowest forest NDVI recorded that year (Fig. 2a). This trend aligns with findings from other regions, such as Inner Mongolia and northern and northwestern China (Liu et al. 2023; Lotsch et al. 2005; Park and Sohn 2010). However, in eastern China, the response pattern is the opposite (Jin et al. 2014). This discrepancy is likely due to regional climatic differences (Piao et al. 2003).

Since variations in forest NDVI are influenced by both natural environmental fluctuations and human activities, it is important to consider the contribution of each factor and their interactions to accurately assess their relative importance for changes in forest NDVI (Jiang et al. 2017). Many famous tourist attractions, such as Kanas, Baihaba and Koktokay, are located in the Altai Mountains. The heavily developed tourism industry plays a vital role in the economic development of these areas (Liu et al. 2022). However, the high number of visitors and vehicles has had a detrimental effect on forest growth (Chen and Qi 2018; Jiang et al. 2017; Mizaras et al. 2015). This study confirms that the degradation of the forest NDVI in the Altai Mountains is predominantly due to human activities (Supplementary Fig. S6), with the most affected areas concentrated along forest boundaries, particularly in the northwestern part of the study area (Supplementary Figs S3 and S6a). In the Altai Mountains, the combined effects of human activities and climate change have contributed significantly to forest growth (Supplementary Fig. S6). This is largely due to the implementation of forest protection policies, which have led to the establishment of numerous forest parks and nature reserves in the Altai Mountains, such as the Shenzhong Mountain and Daqing River ecological conservation areas (Supplementary Fig. S6b and c). The creation of these protected areas has mitigated the detrimental impacts of human activities on the forest environment, preserving soil biological activity and increasing nutrient availability (Zhang et al. 2023), which, combined with the effects of climate change, has promoted forest growth (Supplementary Fig. S6).

There are differences in the trends of forest cover and forest NDVI as well as in their driving factors. From 2000 to 2008, the forest NDVI showed a declining trend (Fig. 2a), whereas the forest cover remained relatively stable with a slight increase (Fig. 1a). This occurred because, during this period, although some areas were protected by forest policies, the snowmelt water showed a decreasing trend (Supplementary Fig. S7g-i), especially SM_34, which had a significant impact on the forest NDVI (Supplementary Fig. S7i). Precipitation variables such as PRE_Y, PRE_G and PRE_P fluctuated at low levels, whereas TEM_Y, TEM_G and TEM_P exhibited a high upward trend (Supplementary Fig. S7a–f), leading to increased drought conditions in the study area (Supplementary Fig. S9), resulting in a declining forest NDVI trend (Fig. 2a). However, forest cover managed to maintain stability and even slightly increased, possibly due to human activities such as the establishment of ecological reserves, afforestation and grazing bans. Between 2008 and 2016, the joint impact of human activities and climate change caused both forest cover and the forest NDVI to increase (Figs 1a and 2a). However, after 2016, the trends diverged. The forest NDVI, along with the influential factors SM_34, SM_Y, PRE_P and PRE_Y, consistently decreased (Fig. 2a; Supplementary Fig. S7d, f, g and i), whereas the forest cover continued to increase (Fig. 1a), indicating that human activities played a dominant role in driving forest cover changes during this period. Spatially, the low- and mid-mountain belts experienced the greatest expansion in forest cover area, with ~70% of the region showing improvements in the forest NDVI (Fig. 1f and g; Supplementary Fig. S3). This may be due to previous degradation caused by human activities in these areas, but after the implementation of forest protection policies and less external disturbance, the forest NDVI improved with the aid of climate change (Zhang 2017).

Data limitations and uncertainty

The forest age group data obtained in this investigation cover most of the Altai Mountains. However, the inaccessibility of personnel to the original forests deep in the mountains has prevented the survey of their age composition, introducing some uncertainty when analyzing the NDVI changes in different age groups. Additionally, while the LUCC data obtained from the CDS have a high level of verified accuracy and meet the needs of this study, its 0.003° resolution may lead to misidentification of small valley forests, introducing some uncertainty into the research. Moreover, since this study calculated forest NDVI spatial trends, forest degradation and improvement levels, and residual analysis, ensuring that the spatial coverage of the forest NDVI data remains consistent each year is crucial. To achieve this goal, we used the common forest cover area from 2000 to 2020 as a mask. As a result, NDVI changes in areas with forest cover shifts were not included. Future research can further explore the response of the NDVI in regions experiencing forest cover change to climate change and human activities.

CONCLUSIONS

This study investigated the mechanisms of forest cover and the forest NDVI in the Altai Mountains considering climate change and human activities. For forest cover, we found that both forest management policies and climate change influenced changes, with forest management policies being the primary factor. Moreover, the drivers of forest cover changes varied across different altitudinal belts. For the forest NDVI, the areas where climate change and human activities jointly enhanced NDVI growth represent the largest proportion across the six scenarios analyzed. Climate change had a broader impact on the NDVI than did human activities. Among the climate factors, precipitation and snowmelt are key controlling factors for forest NDVI changes. Notably, SM_34, a crucial water resource before the growing season, significantly influences forest changes. Additionally, mature forests and deciduous needle-leaved forests contributed the most to forest NDVI changes.

Supplementary Material

Supplementary material is available at Journal of Plant Ecology online.

Table S1: Division of the degrees of variation in the NDVI change trend.

Table S2: Interaction types and criteria.

Table S3: Percentages of different forest age classes in the Altai Mountains.

Table S4: Influence of climate factors on forest NDVI interaction detection in the Altai Mountains (q).

Figure S1: Spatial distribution and area statistics of different forest types and altitude belts in the Altai Mountains.

Figure S2: Spatial trend of the forest NDVI in the Altai Mountains from 2000 to 2020.

Figure S3: Distribution of forest NDVI degradation and improvement in the Altai Mountains.

Figure S4: Spatial pattern of regional forest NDVI contributions to the total forest NDVI over the Altai Mountains on a per-pixel basis.

Figure S5: Overall trends in the NDVI residuals in a regression of the forest NDVI with precipitation, temperature and snowmelt.

Figure S6: Spatial distributions of the impacts of climate and human activities on the forest NDVI.

Figure S7: Variations in temperature, precipitation, and snowmelt in the Altai Mountains from 1992 to 2020.

Figure S8: Photographs from field surveys depicting forest aging and fallen and dead trees in the Altai Mountains.

Figure S9: Dynamic changes in the scPDSI in the Altai Mountains from 2000 to 2020 (scPDSI data were obtained from https://crudata.uea.ac.uk/cru/data//drought/#global).

Authors' Contributions

Guli Jiapaer designed and supervised the research, and Liancheng Zhang processed the data, analysed the results and wrote the manuscript. Philippe De Maeyer and Tim Van de Voorde recommended the methodology and reviewed the manuscript. Tao Yu, Hongwu Liang, Bojian Chen, Kaixiong Lin, and Tongwei Ju offered technical support.

Funding

This research was supported by the Third Xinjiang Scientific Expedition Program (grant no. 2021xjkk0701), the Xinjiang Meteorological Science and Technology Innovation Development Fund Project (grant no. MS202207), the Anhui Meteorological Bureau Innovation Development Special Project (grant no. CXM202110), the Natural Science Foundation of Xinjiang Uygur Autonomous Region (grant no. 2021D01B83) and the Chinese Academy of Sciences President’s International Fellowship Initiative (grant no. 2024PVB0064).

Acknowledgements

We would like to express our gratitude to the Climate Data Store, Earth Data, National Aeronautics and Space Administration (NASA) and Xinjiang Climate Center for making the data available. We would also like to acknowledge the professional language editing services provided by American Journal Experts.

Conflict of interest statement. The authors declare that they have no conflict of interest.

References