关键词: 武夷山地区, 极端气候, 气候响应, 时空变化, 核归一化植被指数（kNDVI）, 随机森林

Abstract:

Against the backdrop of global climate change, the frequency and intensity of extreme climate events in subtropical regions are increasing, posing serious threats to the structure, function, and stability of these terrestrial ecosystems. These abrupt, variable, and destructive events can cause long-term or irreversible damage to vegetation and ecosystem services. As a representative subtropical montane region in southeastern China, the Wuyi Mountain area features complex terrain, rich biodiversity, and diverse vegetation types, making it an ideal natural laboratory for investigating the effects of extreme climate events on forest ecosystems in this region. A better understanding of vegetation responses to such extremes is essential for enhancing ecosystem resilience and informing regional adaptation strategies. This study focuses on the spatiotemporal dynamics and response mechanisms of vegetation in the Wuyi Mountain region from 2001 to 2021, aiming to address three core scientific questions: 1) What are the spatial and temporal patterns of extreme climate events in the Wuyi Mountains? 2) How do vegetation growth and health respond to increasingly frequent climate extremes? 3) Which extreme climate indices and environmental variables are the dominant drivers of different remote sensing-based vegetation indicators? To answer these questions, we integrated multi-source satellite remote sensing products with ground-based meteorological data. Four vegetation indicators were used to represent the vegetation status across multiple dimensions: the kernel normalized difference vegetation index (kNDVI), enhanced vegetation index (EVI), net primary productivity (NPP), and forest stress index (FSI). Together, these indices capture the vegetation dynamics, carbon sequestration potential, and environmental stress. The kNDVI, an improved version of the NDVI, mitigates saturation in densely vegetated areas. The EVI is more sensitive in high biomass regions. NPP reflects carbon assimilation and productivity, whereas FSI combines thermal and hydrological stress to evaluate vegetation responses to abnormal climate conditions. Sen’s slope estimator and the non-parametric Mann-Kendall test were applied to detect long-term trends in the indices. Using the ClimPACT2 platform, we calculated 24 extreme climate indices from daily temperature and precipitation records from 41 meteorological stations. These included 13 temperature indices, 10 precipitation indices, and one drought index (Standardized Precipitation-Evapotranspiration Index [SPEI]). Environmental variables such as elevation, slope, aspect, canopy height, soil type, and soil texture were extracted from high-resolution datasets. We used a random forest regression model to assess the relative importance of these variables in driving vegetation anomalies and to capture their complex nonlinear relationships. From 2001 to 2021, both kNDVI and EVI showed significant upward trends, particularly in the forested areas of the west, north, and south regions. Approximately 43.7% of the region exhibited significant greening. NPP increased in more than 50% of the region, particularly in the northern agricultural zones, indicating enhanced productivity driven by warming and improved land use. However, localized NPP declines occurred in areas with intensive human activity. The FSI exhibited spatial divergence: forest stress decreased in the Wuyi Mountain National Park core, whereas it increased in the southwest, suggesting vulnerability to combined climatic and anthropogenic pressure. The extreme climate indices revealed a clear regional warming trend in the study area. Warm days (TX90p) and warm nights (TN90p) increased at rates of 0.7% and 1.5% per decade, respectively, with nighttime warming being more prominent. In contrast, slight increases in cold day (TX10p) and cold night (TN10p) events were observed at some locations, likely due to terrain-induced microclimatic variability. The intensity of extreme precipitation events has also increased, including increases in maximum 1-day (Rx1-d) and 5-day (Rx5-d) precipitation and daily rainfall intensity (SDII). However, drought-related indices such as SPEI and CDD exhibited high interannual variability without significant long-term trends. Random forest analysis showed different sensitivities of the vegetation indices to extreme climate events. kNDVI was most affected by the number of consecutive dry days (CDD), followed by cold night frequency (TN10p), indicating that drought and nocturnal low temperatures constrain vegetation cover. EVI responded most strongly to the annual minimum daily minimum temperature (TNn) and Rx5-d, reflecting the impact of cold stress and heavy rainfall on the canopy health. NPP was primarily driven by warm night and warm day frequency (TN90p and TX90p), suggesting that nighttime warming increases carbon accumulation via respiration and photosynthesis. The FSI was most influenced by the consecutive cold day index (CSDI3) and the number of tropical nights (TR), underscoring the role of persistent temperature extremes in driving forest stress. Among the environmental factors, the coefficients of variation in elevation (Elevation_cv) and slope (Slope_cv) consistently emerged as important predictors for all vegetation indices. These findings highlight how topographic heterogeneity modulates vegetation-climate interactions by influencing the microclimate, water availability, and soil properties. This study contributes to the literature in a number of ways. First, it integrates four complementary remote-sensing vegetation indices to depict vegetation responses from a structural, functional, and physiological perspective. Second, machine learning was employed to quantitatively identify dominant climate and environmental drivers, thereby improving our understanding of complex ecological processes. Finally, this study provides empirical evidence of vegetation-climate interactions in mountainous ecosystems, supporting research on ecosystem adaptation to global climate change. In conclusion, the Wuyi Mountain region has experienced widespread greening and increased productivity between 2001 and 2021. However, certain areas remain vulnerable owing to the combined effects of extreme climate events and complex terrain characterstics. Temperature-related extremes, particularly nighttime warming, are key drivers of vegetation dynamics. This study offers a robust analytical framework for evaluating vegetation-climate interactions in subtropical montane regions and provides valuable insights into ecological monitoring, climate adaptation, and biodiversity conservation.

Key words: Wuyi Mountain region, extreme climate, climate response, spatiotemporal variation, kernel normalized difference vegetation index (kNDVI), random forest