The Altai Mountains, located along the northeastern margin of the Junggar Basin in Xinjiang, stretch over 500 kilometers in a northwest-southeast orientation, with an elevational gradient ranging from 1,000 to 3,000 meters. This region encompasses diverse geomorphic units, including glaciated zones, alpine landscapes, canyon systems, rolling hills, and intermontane basins. The hydrological network is dominated by the Irtysh River, the only exorheic river in China that drains into the Arctic Ocean. Climatically classified as a cold-temperate continental zone (Huang et al., 2018), the area exhibits distinct vertical climatic zonation governed by latitude, elevation, and topography. Winters are prolonged and severely cold, with persistent snow cover lasting 7–8 months and extreme minimum temperatures below −40 °C, while summers are short and cool, yielding a mean annual temperature range of −4 °C to 3 °C. Precipitation decreases spatially along a northwest-southeast gradient, with annual precipitation ranging from 200 to 800 mm. Northern slopes and high-elevation areas receive abundant moisture, whereas southern slopes and lowlands experience relative aridity (Jiao et al., 2019). The northwestern region, influenced by westerly circulation, features humid conditions, transitioning to semi-arid climates in the southeast. Vegetation follows well-defined vertical zonation, dominated by Larix sibirica Ledeb. as the constructive species, accompanied by Abies sibirica Korsh., Picea obovata Ledeb., and associated taxa such as Betula pendula Roth and Populus davidiana Dode. Larix sibirica forms complex vegetation belts with perennial and annual herbaceous species, creating a unique montane forest-steppe ecosystem characterized by intricate ecological interactions (Wang et al., 2024).

In July 2024, this study collected tree-ring samples from two representative sampling sites within the Burqin (BEJ) and Two-River Source (LHY) forestry areas in the Altai Mountains of Xinjiang, where Larix sibirica exhibits relatively concentrated distribution. The plot area measured 250 m × 250 m. The sampling site locations are shown in Figure 1, and the basic information of the sampling sites is presented in Table 1. Sampling adhered to international dendrochronological research protocols, prioritizing the selection of dominant, healthy trees with minimal anthropogenic disturbance to ensure sample representativeness and integrity. Increment cores were extracted at breast height (1.3 m) perpendicular to slope contours, with duplicate cores collected per tree to account for circumferential growth variability.

The laboratory processing protocol followed established dendrochronological standards. Tree-core samples were immobilized in specialized wooden trays using carpenter’s glue and linen threads, followed by sequential polishing with progressively refined sandpapers (400–1,200 grit) until annual ring boundaries attained optical clarity. Ring-width measurements were acquired using a high-precision LINTABTM⁶ system (Rinntech, Germany) with a 0.001 mm resolution. Data processing incorporated three sequential stages: cross-dating verification through COFECHA software (Grissino-Mayer, 2001) ensured chronological accuracy; age-related growth trends were corrected via piecewise linear regression in the LINTAB system; and ARSTAN software (Cook and Krusic, 2005) generated standardized chronologies using negative exponential detrending. The derived RWI chronologies for both sites are shown in Figure 2, with key statistical parameters provided in Table 2.

The Normalized Difference Vegetation Index (NDVI), a robust indicator of surface vegetation growth status and coverage, has been widely employed in studies of vegetation spatiotemporal dynamics and climatic responses (Opała-Owczarek et al., 2023; Wilmking et al., 2020; Zhang, 2015). In this study, NDVI was adopted as the primary metric for characterizing vegetation dynamics in the research area. The NDVI time series data were obtained from the MODIS13Q1 dataset,¹ which features a spatial resolution of 250 meters and a temporal resolution of 16 days, demonstrating well-established applicability in vegetation cover research (Wang et al., 2021; Song et al., 2024). This dataset has undergone atmospheric correction. To ensure data quality, pixels of optimal quality were selected using the Summary QA band to minimize the impact of clouds and aerosols. The 16-day composite data were converted into monthly data, employing the maximum value composite method to ensure a representative NDVI value is obtained for each month. Given the persistent winter snow cover in the study area (Altai Mountains, Xinjiang), which compromises NDVI reliability during non-growing seasons, we specifically selected growing-season (May–September) NDVI data from 2000 to 2023 for correlation analysis with tree-ring width. The multi-year monthly average NDVI distribution patterns in the study area are illustrated in Figure 3. Results reveal pronounced monthly variations in NDVI values within the Larix sibirica growing season across both forest sites, with values descending in the order of June > July > August > May > September.

This study utilized the ERA5-Land reanalysis dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF) as the primary climate data source (Muñoz Sabater et al., 2021). Given the complex topography of the Altai Mountains and the considerable distance between our sampling sites and nearby meteorological stations, conventional ground-based observations were inadequate for accurately characterizing local climatic conditions. The ERA5-Land dataset, with its continuous records since 1950, offers exceptional temporal (hourly) and spatial (0.1° × 0.1°) resolution that effectively captures microclimate variations in mountainous regions, and has been extensively validated for ecological and forestry applications. Analysis of the study area’s multi-year climatic characteristics (Figure 4) revealed pronounced seasonal patterns in both monthly mean temperature and precipitation, which exhibited strong correlations with the growth cycle of Larix sibirica. The high spatiotemporal resolution of the ERA5-Land dataset provided a robust foundation for precisely analyzing tree growth-climate relationships in this study.

We employed Pearson correlation analysis to examine the relationships between tree-ring width and monthly NDVI (May–September) as well as the entire growing season NDVI (May–September composite) during 2000–2023 for both forest sites. Based on the correlation results, we established a historical NDVI reconstruction equation using a linear regression model in R (Cook and Kairiukstis, 2013), as follows Equations 1, 2:

Here, NDVIGS represents the May–September composite NDVI, while RWI denotes the tree-ring width index.

The stability and reliability of the NDVI reconstruction equations for both sampling sites were rigorously evaluated (Table 3). The results demonstrated that the equations yielded R² values of 0.378 and 0.318 (p = 0.001 and 0.005, respectively), both statistically significant at p < 0.01. Additional validation through Jackknife (Efron and Stein, 1981) and Bootstrap (Efron, 1979) methods revealed consistently low prediction errors, as indicated by small values of both the leave-one-out cross-validated mean squared error (LOOCVMSE) and out-of-bag mean squared error (OOBMSE). Both methods evaluate models by simulating “multiple repeated experiments.” Jackknife is performed a fixed number of times (n), observing effects through “leave-out” procedures; Bootstrap runs numerous times (B times), constructing empirical distributions of statistics via “resampling with replacement.” They are both powerful tools for handling small samples or model uncertainty. The detailed results of the Jackknife and Bootstrap tests are shown in Table 4. These comprehensive analyses confirm that the reconstruction equations exhibit robust stability and high reliability for historical NDVI estimation.

To further validate the reliability of the NDVI reconstruction equations, we compared the observed and reconstructed NDVI values from 2000 to 2022 (Figure 5). The time-series plot demonstrates strong collinearity between the observed and reconstructed NDVI data. Correlation analysis revealed statistically significant relationships between the observed and reconstructed NDVI at both sampling sites, with correlation coefficients (R) of 0.56 and 0.62 (p < 0.01). These results indicate that the reconstructed NDVI the temporal dynamics of actual vegetation cover changes in the study region.

To elucidate the mechanisms by which extreme climate events influence tree radial growth, this study extracted daily maximum temperature, minimum temperature, and precipitation data from the ERA5 reanalysis dataset for both sampling sites during 1953–2023. Using the RClimDex software, we computed 11 monthly extreme climate indices that characterize various aspects of climatic extremes (Table 5). The response relationships between tree-ring width and these extreme climate indices were systematically investigated through Pearson correlation analysis. Recognizing the lagged physiological responses of trees to climatic conditions, we implemented a cross-year analytical framework (October of the previous year to October of the current year) to comprehensively identify key climatic drivers of radial growth. This methodological design effectively addresses the limitations of conventional single-year analyses that may overlook cross-annual climate effects, thereby providing a robust scientific basis for identifying critical climatic constraints on tree growth.

This study employs a multi-scale modeling approach to predict Siberian larch growth responses to climate change. Based on correlation analyses between tree-ring width and extreme climate indices, we first applied a LASSO regression model to select key climatic drivers of radial growth from 143 candidate factors—including eleven extreme climate indices covering October–December of the previous year and January–October of the current year—and established predictive model equations for two sampling sites. LASSO modeling was constructed using the ‘glmnet’ package in R (Friedman et al., 2021). After standardizing all predictor variables, k-fold cross-validation was employed to select the optimal penalty parameter λ, ultimately yielding a sparse and interpretable model. We utilized the NEX-GDDP-CMIP6 dataset, which provides downscaled climate projections from multiple CMIP6 models, focusing on two representative scenarios, SSP2-4.5 and SSP5-8.5, that differ in their greenhouse gas emission assumptions and produce distinct climate forecasts. These differences introduce scenario uncertainty, which plays a critical role in future growth predictions. To evaluate the robustness of our projections, we incorporated results from multiple climate models within the NEX-GDDP-CMIP6 ensemble, thereby assessing ensemble model performance. Using future climate scenario data from the CMIP6 multi-model ensemble for 2025–2075, we calculated the corresponding extreme climate indices via the RClimdex program. Finally, we input the selected key climatic drivers into the LASSO model equations to simulate and predict radial growth dynamics over the next 50 years.

Pearson correlation analysis (Figure 6) revealed distinct monthly-scale relationships between tree-ring width and NDVI across the sampling sites. At the Burqin site, tree-ring width exhibited a highly significant positive correlation with August NDVI (p < 0.01) and the entire growing season NDVI (May–September composite, p < 0.001), while showing no statistically significant correlations with NDVI in May, June, July, or September. Conversely, at the Two-River Source site, significant positive correlations were observed with June NDVI (p < 0.05) and August NDVI (p < 0.01), alongside a strong correlation with the growing season NDVI (p < 0.01), though no significant associations were detected for May, July, or September NDVI. Comparative analysis demonstrated that tree-ring width at both sites consistently showed strong correlations with August and growing season NDVI. This validated that radial growth dynamics effectively capture vegetation cover variations during critical phenological stages, providing robust scientific support for reconstructing regional vegetation dynamics using dendrochronological data.

Based on the NDVI reconstruction equations, this study successfully reconstructed historical growing-season NDVI dynamics for the two sampling sites (Figure 7). The reconstruction for the Burqin site spans 215 years (1808–2022), while that for the Two-River Source site covers 151 years (1872–2022). Statistical analysis showed no significant differences in historical mean NDVI between the two sites (Burqin: 0.589; Two-River Source: 0.586). High and low vegetation cover periods were defined as NDVI values exceeding one standard deviation above or below the mean, respectively. The Burqin site experienced three low-coverage periods (1973–1976, 2003–2004, 2013–2014) and six high-coverage periods (1808–1810, 1812–1815, 1822–1827, 1834–1836, 1840–1843, 1864–1866), with the lowest NDVI value in 2010 (0.534) and the highest in 1810 (0.715). The Two-River Source site exhibited four low-coverage periods (1952–1954, 1955–1956, 2000–2009, 2013–2014) and three high-coverage periods (1872–1879, 1881–1886, 1888–1890), with the lowest NDVI in 2004 (0.519) and the highest in 1879 (0.699).

Although we have obtained NDVI values for specific years during historical periods through reconstruction, it must be emphasized that due to uncertainties inherent in the reconstruction process, the reconstructed results primarily serve to reflect overall trends in vegetation cover rather than precise annual NDVI values. Pettitt’s change-point test revealed significant transitions in vegetation cover: 1894 for Burqin and 1919 for Two-River Source. It should be noted that the presence of noise in the reconstruction NDVI, particularly significant and random noise, may influence the detection of change-points. The results of the simple linear regression analysis indicate that both locations exhibited a declining trend in NDVI prior to the mutation (p < 0.001), with the decline at the Two-River Source site being more pronounced than that at the Burqin site. Post-transition trends diverged: the Burqin site showed no significant NDVI trend (p > 0.05), while the Two-River Source site maintained a declining trend (p < 0.001), albeit at a reduced rate compared to the pre-transition period. These results demonstrate spatial heterogeneity in vegetation cover responses to climate change.

To elucidate the mechanisms of extreme climate impacts on tree growth, we conducted Pearson correlation analysis between tree-ring width and 11 extreme climate indices (calculated via RClimDex) at both sampling sites (Figure 8). The results revealed distinct response patterns: At the Burqin site, tree-ring width exhibited significant negative correlations with the maximum 1-day precipitation (RX1day) in October of the previous year (p < 0.05), while showing positive correlations with the maximum daily maximum temperature (TXx), warm days (TX90p), and diurnal temperature range (DTR) in February of the current year (p < 0.05). Significant positive correlations were also observed with the maximum 5-day precipitation (RX5day) in March and July of the current year (p < 0.05). The Two-River Source site demonstrated more complex responses: Tree-ring width showed significant negative correlations with the maximum daily minimum temperature (TNx) in November of the previous year; the maximum daily maximum temperature (TXx) and warm days (TX90p) in March; warm days (TX90p) in April; TXx, warm nights (TN90p), and TNx in July; and TNx and TN90p in September (all p < 0.05). Conversely, positive correlations were identified with cold days (TX10p) in July, and DTR, TX10p, and cold nights (TN10p) in September (p < 0.05).

This study utilized CMIP6 multi-model scenario data to calculate future extreme climate indices via the RClimDex program, with key drivers of radial growth identified through LASSO regression model screening. Statistical parameters of the models are presented in Table 6. The Two-River Source site’s LASSO model exhibited significantly superior explanatory power (R² = 0.594) compared to the Burqin site (R² = 0.248), indicating heightened sensitivity of tree growth to extreme climate variability in the Two-River Source region. Additionally, we conducted VIF tests on the variables in both models (Figure 9). The results indicate that the models perform well, and the variable selection is generally acceptable. Key climatic factors identified for the Two-River Source model included: warm night days (TN90p) in September of the previous year; warm day days (TX90p) in October of the previous year; maximum daily minimum temperature (TNx) in November of the previous year; TNx and TX90p in March of the current year; TX90p in April; maximum 5-day precipitation (RX5day) in May; cold day days (TX10p) in June and July; TNx in July; and diurnal temperature range (DTR), cold night days (TN10p), TX10p, and TN90p in September of the current year.

To more accurately capture the response signals of Larix sibirica to climate change in the future, this study focuses solely on the Two-River Source sampling sites, where tree ring width (RWI) is more sensitive to extreme climate events, and the model performs better. To validate the reliability and robustness of the model, cross-validation and residual diagnostics were conducted, with the results shown in Figures 10, 11. Overall, the model successfully captured the main fluctuation characteristics of RWI, and the predicted curve exhibited high synchronization with the observed values, particularly in extreme years (such as the peak in 1998 and the trough in 2005), demonstrating strong fitting ability. Based on the residual diagnostics, the assumptions of the LASSO model were found to hold. The residual plot (Figure 11a) shows no discernible pattern, while the QQ plot (Figure 11b) and the Shapiro–Wilk normality test (W = 0.98281, p = 0.4511) further support that the residuals conform to a normal distribution. Therefore, the LASSO model can be considered appropriate, with satisfactory goodness of fit and no significant bias or heteroscedasticity issues.

The LASSO regression model, constructed based on CMIP6 climate projections and the response relationships between tree-ring width and extreme climate indices during 1953–2023, projects significant declines in radial growth of Larix sibirica across the Altai Mountains under all emission scenarios from 2025 to 2075 (Figure 12). The mean predicted growth under the SSP2-4.5 and SSP5-8.5 scenarios is 0.814 and 0.982, respectively. Under the high-emission SSP5-8.5 scenario, the tree-ring width index (RWI) exhibited a more pronounced decline rate (R² = 0.303, p < 0.001) compared to the moderate-emission SSP2-4.5 scenario (R² = 0.08, p < 0.05). These results suggest that intensified climate warming and increased frequency climate events will impose greater stress and uncertainty on the growth of Larix sibirica.

Tree-ring width, a critical metric of radial growth, effectively reflects long-term vegetation cover dynamics when coupled with NDVI analysis (Mašek et al., 2024). Pearson correlation analysis revealed significant correlations between the tree-ring width index (RWI) and growing-season (May–September) NDVI at both sampling sites (p < 0.05). Further analysis demonstrated site-specific monthly variations: the Burqin site showed significant correlations only with August NDVI (p < 0.05), whereas the Two-River Source site exhibited significant correlations with both June and August NDVI (p < 0.05). These monthly discrepancies align with regional vegetation phenology. June corresponds to the NDVI peak period (Figure 2) in the study area, where increased precipitation and favorable hydrothermal conditions enhance canopy development and radial growth (Ma et al., 2024). August represents the most active growth phase for Larix sibirica (Lang, 2018), coinciding with peak canopy expansion, which explains the strongest correlations with August NDVI at both sites. Notably, the complex vegetation community structure—comprising Picea schrenkiana, Abies sibirica, Juniperus communis var. saxatilis, Lonicera caerulea, and Rosa acicularis Lindl. (Huang et al., 2018) may weaken RWI-NDVI correlations in July due to accelerated growth of competing species under optimal hydrothermal conditions. However, synchronized growth rhythms across vegetation types emerge during late growing seasons as environmental conditions stabilize (Huang et al., 2022). Tree-ring width not only records annual climatic signals (temperature, precipitation) but also reflects late-growing-season vegetation dynamics, establishing its reliability as a proxy for reconstructing historical NDVI variations and elucidating vegetation-climate interactions.

Previous studies corroborate the significant RWI-NDVI relationships. Research on Larix sibirica in the Altai Mountains (1982–2000) revealed a significant positive correlation between the ring-width index (RWI) and growing-season NDVI. However, species-specific variations were observed in the timing of peak correlations, with larch showing the strongest association with June NDVI, while spruce exhibited maximum correlation with May NDVI (Wu X. et al., 2022; Wu Q. et al., 2022). Similarly, research on Picea schrenkiana in the Tianshan Mountains confirmed significant tree-ring-NDVI correlations (Lang, 2018), but diverged from our findings by showing additional September correlations-a discrepancy attributable to evergreen Picea maintaining photosynthetic activity into autumn, unlike deciduous Larix which undergoes leaf senescence. Extensive studies across northern China (Chen et al., 2015; Wang et al., 2010) consistently validate growing-season RWI-NDVI correlations, affirming tree-ring width as a robust proxy for long-term vegetation cover reconstruction.

Pearson correlation analysis between tree-ring width and 11 extreme climate indices (October of the previous year to October of the current year) revealed distinct spatial patterns in climatic drivers of radial growth (Figure 7). Extreme climate indices from both the previous and current years significantly influenced tree growth. For instance, tree-ring width at the Burqin site showed significant negative correlations with the maximum 1-day precipitation (RX1day) in October of the previous year (p < 0.05), while the Two-River Source site exhibited negative correlations with the maximum daily minimum temperature (TNx) in October of the previous year (p < 0.05). These findings indicate a cross-annual response mechanism between tree growth and extreme climate events (Zhang, 2015). This phenomenon can be explained from two perspectives: physiological ecology and climatic lag effects. Extreme precipitation in the previous autumn alters soil moisture saturation and root respiration efficiency, disrupting non-structural carbohydrate storage and redistribution, thereby limiting energy availability for cambial reactivation in the following growing season (Wu X. et al., 2022; Wu Q. et al., 2022). Warm autumn events disrupt low-temperature dormancy signals, prolonging photosynthetic activity while increasing respiratory consumption, leading to carbon imbalance and subsequent radial growth reduction (Babst et al., 2012).

Analysis of the relationship between tree ring width at sampling sites and annual extreme climate indices revealed that the impact of extreme climate on tree growth varies significantly across different growth stages. At the Burqin site, higher early spring temperatures during the initial phase of the growing season facilitated earlier tree dormancy release, promoted sap flow, and stimulated early cambium activity, establishing a favorable physiological foundation for the onset of the growing season. Warm daytime temperatures and large diurnal temperature fluctuations before the growing season accelerate soil thawing and restore root absorption functions, indirectly promoting xylem development and ring formation during the growing season (He et al., 2023). Regarding precipitation, extreme rainfall events during the early growing season and peak growing season did not inhibit tree growth but rather promoted it. Concentrated precipitation before the growing season begins helps increase soil moisture reserves, alleviates potential spring drought stress, and provides trees with ample water before the growing season commences (Ukhvatkina et al., 2021). During the peak growth period, several days of effective precipitation significantly enhance photosynthesis and cell division, accelerating xylem accumulation rates and thereby promoting wider tree rings (Huang et al., 2021). This indicates that short-duration heavy precipitation events during the early growing season and critical growth months at the Burqin sampling site play a crucial role in driving annual growth.

Compared with the Burqin site, tree ring width at the Two-River Source site showed significant correlations with multiple extreme temperature indices, predominantly negative correlations, reflecting tree sensitivity to high-temperature events and warm night conditions. No significant correlations were found with extreme precipitation indices. Specifically, tree growth exhibited varying degrees of negative correlation with high-temperature events during the early, peak, and late growing seasons. This result suggests that extreme or prolonged high temperatures during both growing and non-growing seasons may intensify transpiration, increase water stress, inhibit cambial cell division and expansion, and consequently limit xylem growth (Teskey et al., 2015; Rammig et al., 2015). Furthermore, nighttime high temperatures (TNx, TN90p) may disrupt nocturnal metabolic regulation in trees, leading to negative effects on carbon balance (Zhu et al., 2020). Conversely, annual ring width showed significant positive correlations (p < 0.05) with cold days (TX10p), cold nights (TN10p), and daily temperature range (DTR) during the peak growing season. These findings indicate that short-term cold events during the growing season may mitigate high-temperature stress, helping trees maintain stable water status and carbon allocation (Jiao et al., 2019). Particularly during the mid-to-late growing season, such temperature fluctuations play a positive role in sustaining cambial activity.

In addition to spatial heterogeneity, tree growth is often influenced by a combination of multiple factors, with topographic features and biological factors being two major influences on tree growth (Barnard et al., 2017; Ramsfield et al., 2016). The results of this study indicate that the radial growth of trees at low-altitude sampling sites is more sensitive to extreme climate events compared to those at high-altitude areas. This phenomenon has also been observed in previous studies on Larix sibirica in the Altai Mountains (Zhou et al., 2021). The reason for this phenomenon lies in the vertical variation of hydrothermal conditions in the Altai Mountains. Low-altitude areas are characterized by dry climates and limited precipitation, which constrains tree growth and makes it more susceptible to climate disturbances. Biological factors are also key elements affecting the growth of Larix sibirica, such as the larch caterpillar, larch casebearer, and the Zeiraphera grisecana (Gulmira, 2022). Rising winter minimum temperatures may enhance insect survival rates, potentially exacerbating forest pests and diseases.

LASSO regression models based on CMIP6 climate projections and tree-ring responses to extreme climate indices indicate significant future declines in Larix sibirica growth across the Altai Mountains under both high-emission (SSP585) and moderate-emission (SSP245) scenarios. As a constructive species in this cold-temperate montane forest ecosystem, Larix sibirica plays prominent roles in regional carbon cycling, hydrological regulation, and ecological stability maintenance (Jiang et al., 2021). Increased frequency and intensity of extreme climate events will directly impair larch growth and xylem structure, reducing forest productivity and carbon sequestration capacity. Growth suppression may further diminish tree resistance to drought, windthrow, and pest infestations, accelerating senescence and mortality rates. Such declines risk triggering ecological succession toward shrubland or grassland, altering the historic forest-steppe ecotone and threatening alpine ecological stability (Trumbore et al., 2015). Without mitigation, these trends could drive forest line retreat (Dhyani, 2023) and species turnover within decades (García-Valdés et al., 2020). This study not only identifies Larix sibirica’s response patterns to specific extreme climate factors but also provides empirical evidence for assessing cold-temperate forest vulnerability under future climate change. To enhance ecosystem resilience, conservation strategies and adaptive forest management must prioritize mitigating extreme climate risks, ensuring habitat suitability for constructive species, and maintaining community stability in the Altai Mountains.

Although this study reconstructed vegetation-cover dynamics in the Altai Mountains over the past century and analyzed the effects of extreme climatic events on the radial growth of Larix sibirica—as well as projected future trends in tree growth—several improvements remain. First, although Siberian larch is the most widely distributed arboreal species in the Altai region, only two typical growth sites were selected as representatives. Such a limited sample size may not fully capture vegetation-cover variability across the entire area; future work should include a larger number of sampling locations. Second, this study considered only the impact of extreme climate on tree growth; however, topographic factors also influence tree development (Wang et al., 2021), and future research should incorporate these terrain effects into the analysis. Furthermore, future growth projections rely on applying LASSO models calibrated based on historical climate-growth relationships to CMIP6 climate model projections. This approach implicitly assumes that climate-growth relationships remain stable in the future, but may fail under scenarios involving strong warming, high carbon dioxide concentrations, changes in disturbance patterns, or shifts in forest structure.