关键词: 泥炭地, 随机森林分类, 时空演变, 驱动机制, 地理探测器

Abstract:

[Objective] Peatlands are among the most carbon-dense terrestrial ecosystems and play important roles in long-term carbon sequestration, water storage, runoff regulation, and biodiversity conservation. Because these ecological functions depend on persistently wet and oxygen-limited conditions, peatlands are highly sensitive to regional hydrothermal fluctuations and disturbances that alter vegetation or hydrological connectivity. The peatlands of the Altai Mountains in northwestern China, situated near the arid margin of Central Asia, represent a critical ecological barrier that is highly responsive to global environmental change. However, in this fragile alpine environment, multiple stressors including climate change, extreme events, and anthropogenic pressures such as grazing, tourism, and resource development, act simultaneously. Although localized peatland loss has been reported, continuous and systematic evidence of spatiotemporal dynamics after 2010 remains limited. Furthermore, the relative contributions of mean climatic conditions, climate extremes, and human activities, as well as their interactive effects on peatland dynamics, have not been systematically quantified. This knowledge gap restricts the identification of vulnerable areas and the development of spatially differentiated conservation strategies. [Methods] To address these limitations, we reconstructed the spatiotemporal dynamics of peatlands in the Altai Mountains from 2000 to 2024 and quantified their associations with climatic and socioeconomic factors. We used the Google Earth Engine platform to integrate Landsat 5, Landsat 7, and Landsat 8 imagery acquired during the peak growing season (June‒August). Following image preprocessing, annual land-cover maps were generated using a random forest classifier. Training and validation samples were compiled from field-based GPS observations, visual interpretation of high-resolution historical Google Earth imagery, existing wetland products, and regional documentary evidence. To improve discrimination between peatlands and grasslands in complex mountainous terrain, the feature set combined spectral indices, texture, elevation, and slope. Stratified random sampling was used to divide the samples into training and validation subsets at a ratio of 8꞉2. Overall accuracy exceeded 85%, with Kappa coefficients greater than 0.90 across all mapped years, supporting reliable long-term comparisons. Subsequently, land-use transition matrices were constructed to determine the direction and magnitude of conversions between peatlands and other land-cover types. An optimal-parameter geographical detector (OPGD) was applied to assess 20 potential drivers representing background climate, extreme climatic conditions, and human activity. Total population and gross domestic product were used as spatial proxies for cumulative anthropogenic pressures. [Results] The results showed a nonlinear decreasing trend in peatland extent from 2000 to 2024. The degradation trajectory was highly stage-dependent. The most rapid degradation occurred during 2000-2010, resulting in a loss of 203.71 km² (accounting for 64% of the total cumulative loss) at an average annual reduction rate of 4.0%. The shrinkage slowed during 2010-2015, with a further loss of 64.81 km², transitioned into a phase of concurrent local loss and expansion during 2015-2020, and reached relative stabilization during 2020-2024, during which 93.0% of the peatland area was retained. Spatially, degradation was more severe in the southeast and weaker in the northwest. More than 53% of peatlands remained intact in northwestern sites (e.g., Tielishahan and Kucha), whereas peatland losses exceeded 50% in southeastern regions (e.g., Halasazi, Sandaohai, and parts of Heihu), demonstrating pronounced spatial heterogeneity in ecosystem vulnerability. Importantly, transition analysis identified grassland expansion as the dominant destination of degraded peatlands, confirming a classic “wetland-to-grassland” dry-adaptation succession. Approximately 45.0% of degraded peatlands transitioned into grasslands, whereas conversions to forest, bare land, or snow cover were minimal. Grasslands also contributed 8.7% of the area newly classified as peatland, indicating bidirectional exchanges. Nevertheless, the dominant trajectory was peatland loss and grassland expansion, providing landscape-scale evidence of a characteristic peatland-to-grassland transition rather than spatially uniform peatland disappearance. The OPGD results demonstrated that Altai peatland dynamics were jointly associated with climate change and human activities. Factor detection revealed that annual precipitation (q = 0.770) and total population (q = 0.768) exerted the strongest single-factor explanatory power, highlighting the importance of water supply and anthropogenic pressure for peatland persistence in the arid-margin mountains of Central Asia. Extreme-climate indices also showed high explanatory power. The annual minimum of daily minimum temperature (q = 0.750) and annual maximum one-day precipitation (q = 0.730) exhibited similarly high explanatory power. Peatland persistence was therefore related not only to mean climatic conditions but also to extreme hydrothermal events and freezing conditions. Every examined pair of factors produced greater explanatory power than either corresponding single factor, revealing widespread interaction enhancement. The interaction between annual precipitation and the annual maximum of daily maximum temperature reached a q value of 0.892, indicating strong spatial coupling between water availability and extreme heat. Additionally, total population, acting as a proxy for cumulative anthropogenic pressure rather than direct residential footprint, exerted a strong interactive effect when paired with diurnal temperature range (q = 0.900), reflecting how population-related pressures, such as overgrazing, tourism infrastructure expansion, and localized mining, may propagate into remote high-altitude zones, physically constraining peatland habitats and accelerating ecosystem degradation. Variables with relatively weak individual effects, including the simple daily intensity index, also gained explanatory power when combined with dominant factors. [Conclusion] These findings indicate that the observed peatland-to-grassland transition is consistent with combined changes in water availability, extreme hydrothermal conditions, and human disturbance in the Altai Mountains. Because the analysis is based on remotely sensed land-cover dynamics and spatial associations, it does not demonstrate specific changes in groundwater levels, active-layer thickness, soil mineralization, or plant competition, which require further field observations and controlled experiments. Nevertheless, the 25-year record identifies when, where, and along which pathways peatland degradation occurred and shows that climate-human interactions explain more spatial variation than isolated drivers. Based on these findings, we recommend that conservation and restoration policies prioritize highly vulnerable southeastern sites. Interventions should focus on maintaining and restoring hydrological connectivity, regulating seasonal overgrazing, controlling tourism infrastructure expansion in sensitive peatlands, and establishing early-warning systems for climate extremes and freeze-thaw transitions. The resulting dataset and analytical framework provide an evidence base for targeted restoration, long-term monitoring, and adaptive management of mountain peatlands in the Altai Mountains under ongoing environmental change.

Key words: peatlands, random forest classification, spatiotemporal dynamics, driving factors, geographical detector