Integrating Machine Learning and Species Distribution Models for Predicting the Potential Hazard Areas of Marmota baibacina in Xinjiang, China

Qianying Shao¹Kaihui Li¹Turghan Mardan-Aghabe¹Anming Bao¹Alimujiang Kasimu²Yanming Gong¹Jie Bai¹Jun Lin³Xuan Li³Jin Zhao^1*

• ¹Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences (CAS), Ürümqi, China
• ²Xinjiang Normal University, Urumqi, Xinjiang Uyghur Region, China
• ³Center for Grassland Biological Disaster Prevention and Control of Xinjiang Uygur Autonomous Region, Urumqi, China

Under global climate change and intensified human activities, species distributions are undergoing significant shifts. Marmota baibacina, a representative keystone species among Central Asian high-altitude species, exacerbates vegetation degradation and soil erosion through herbivory and burrowing activities. As the primary reservoir of Yersinia pestis, it poses a significant public health threat. This study integrated five machine learning models (XGBoost, RF, SVM, LogBoost) and the MaxEnt model to predict the current (1970–2000) and future (2041– 2100) distribution of Marmota baibacina under three climate scenarios (SSP126, SSP370, SSP585), utilizing 111 occurrence records and 29 environmental variables spanning climatic, topographic, edaphic, and vegetation dimensions. The results indicated that (1) All five models demonstrated high predictive accuracy with AUC values exceeding 0.9. After screening 29 environmental variables, machine learning models identified 10 key variables with high feature importance, while MaxEnt selected 16 environmental variables; (2) Dominant drivers revealed that Bio18 (warmest quarter precipitation), Bio2 (diurnal temperature range), Bio11 (coldest quarter temperature), and Bio15 (precipitation seasonality) collectively contributed >70% to machine learning models, whereas MaxEnt prioritized slope, NDVI, and Bio18; (3) Under current climatic conditions, the potential suitable habitats of Marmota baibacina in Xinjiang are primarily concentrated in the central Tianshan Mountains, with core distribution centers in Bayingolin Mongolian Autonomous Prefecture (Hejing County), Ili Kazakh Autonomous Prefecture, and the western part of Bortala Mongolian Autonomous Prefecture, The total suitable habitat area estimated by the five models ranged from 2.75 × 10⁴ km² to 13.59 × 10⁴ km² under the current climate; (4) Future projections under all scenarios indicated an overall decreasing trend in suitable habitat area, with habitat contraction particularly pronounced in the southern Tianshan under SSP585. Such distributional shifts may intensify competition between marmots and livestock, accelerate alpine meadow degradation, and elevate zoonotic plague transmission risks due to population aggregation. This study provides critical insights for balancing alpine ecosystem conservation and plague prevention strategies, offering actionable guidance for safeguarding ecological security and public health in Xinjiang's ethnically diverse pastoral regions.