Integrating explainable machine learning to predict the ecological niche distribution of Cytospora chrysosperma in Xinjiang, China

文献类型： 外文期刊

第一作者： Li, Quansheng

作者： Li, Quansheng;Hou, Ruixia;Zhang, Xianhua;Cao, Shanshan;Sun, Wei

作者机构：

关键词： Cytospora chrysosperma; Ecological niche modeling; Machine learning; SHAP analysis; Model interpretability; Threshold interaction network; Environmental thresholds

期刊名称：FOREST ECOLOGY AND MANAGEMENT （ 影响因子：3.7； 五年影响因子：4.1 ）

ISSN： 0378-1127

年卷期： 2025 年 595 卷

页码：

收录情况： SCI

摘要： Cytospora chrysosperma is a significant pathogenic fungus threatening forest trees in Xinjiang, causing substantial economic and ecological losses. Understanding its ecological niche distribution is crucial for developing effective control strategies, yet traditional models often lack interpretability. This study addresses these limitations by integrating five machine learning algorithms with SHAP (SHapley Additive exPlanation) methodology to predict and interpret the potential distribution of C. chrysosperma in Xinjiang, introducing the novel concept of a "threshold interaction network" for interpretable ecological niche modeling. We collected 545 presence records and generated 600 pseudo-absence points, selecting bioclimatic variables, topography, and NDVI as environmental predictors through stepwise feature selection, with model performance evaluated using both conventional and spatial block cross-validation approaches. All models performed well, with Random Forest demonstrating superior overall performance. SHAP analysis revealed critical ecological thresholds: vegetation condition (NDVI approximate to 0.15), precipitation seasonality (bio15 approximate to 73), mean temperature of the warmest quarter (bio10 approximate to 21 degrees C), and elevation (approximate to 1504 m) as key determinants of pathogen distribution, while interaction analysis discovered significant synergistic effects between environmental factors, with areas characterized by NDVI > 0.15 and bio15 < 73 representing the highest risk zones. Bootstrap analysis confirmed threshold stability (optimal classification threshold: 0.4905), enabling robust risk assessment with spatially differentiated maps identifying high-risk areas such as Tacheng and Yili regions for targeted management. This study establishes a new paradigm for interpretable ecological niche modeling by demonstrating how environmental factors collaboratively influence pathogen distribution through quantifiable threshold relationships, while providing forest managers with actionable, threshold-based intervention guidelines.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序