Objective Bark beetle outbreaks in spruce forests in the Eastern Tianshan Mountains of Xinjiang pose a serious threat to the stability and ecological functions of the regional forest ecosystems. This study aimed to identify bark beetle-infested spruce stands using time series of Sentinel-2 satellite imagery, analyze their spatiotemporal distribution characteristics and relationships with topographic factors, and provide technical support for bark beetle monitoring and scientific management.

Method The Picea schrenkiana forests in the Banfanggou Branch of the Eastern Tianshan State-owned Forest Bureau were selected as the study area. Time series of Sentinel-2 images acquired from 2016 to 2025, together with field survey data and GF-2/7 sub-meter-resolution satellite imagery, were used to develop a random forest classification model for identifying bark beetle-infested spruce stands. The importance of spectral reflectance, vegetation indices (VIs), and temporal anomaly variables of vegetation indices (ΔVIs) was evaluated. Based on the biological characteristics of bark beetle infestation, spatiotemporal distribution maps of bark beetle damage were generated. Statistical analyses were further conducted to investigate the relationships between bark beetle distribution and topographic factors.

Result The random forest classification model achieved an overall accuracy (OA) of 92.04% with a Kappa coefficient of 0.84, demonstrating high accuracy in identifying bark beetle-infested stands. Variable importance analysis showed that ΔVIs contributed more to the classification than single-date spectral reflectance and VIs, with the temporal anomaly of the normalized difference vegetation index (ΔNDVI) identified as the most important feature. The spatiotemporal distribution maps derived from the classification results were highly consistent with visual interpretation of GF-2/7 satellite imagery. Bark beetle damage in the study area was primarily concentrated in low-elevation areas (1 600–2 000 m), with some interannual variation in its distribution across different elevation ranges.

Conclusion The random forest model based on Sentinel-2 time-series data effectively identified bark beetle-infested spruce stands and enabled spatiotemporal mapping of bark beetle damage. The spatial distribution of bark beetle damage in the study area exhibited distinct differences across elevation gradients and considerable interannual variation. These findings provide technical support for regional remote sensing monitoring and precision control of bark beetle-infested spruce forests in the Tianshan Mountains.