Dual niche modeling with GEE and SHAP for predicting habitat shifts of Haloxylon ammodendron and Cistanche deserticola under climate change

Haloxylon ammodendron, a keystone woody species, and its parasitic plant, Cistanche deserticola, play critical roles in sustaining arid ecosystems and supporting regional economies. However, their distribution is increasingly threatened by global climate change. Here, we propose a dual niche modeling framework that integrates climate and soil suitability layers using a multi-model ensemble approach combined with interpretable machine-learning techniques, specifically SHapley Additive exPlanations (SHAP). Using CMIP6 scenarios (SSP126, SSP245, and SSP585), we predicted the current and future potential habitats for both species. The results demonstrated that the ensemble models delivered robust performance, surpassing the accuracy of single-model predictions. Currently, suitable habitats are concentrated in northwestern China as well as parts of Mongolia and Kazakhstan. Under SSP585 (2081-2100), H. ammodendron habitats are projected to shrink by 56.2%, whereas C. deserticola is expected to lose more than 97% of its habitat, nearly disappearing from Central Asia. Key climatic drivers include temperature seasonality and precipitation patterns, whereas the soil water-holding capacity and gravel content significantly affect local suitability. Niche overlap analysis revealed a strong host dependency for C. deserticola. However, the climate-soil niche congruence is projected to decrease under future scenarios, indicating the potential risks of ecological decoupling. This integrative and interpretable approach offers a scalable tool for biodiversity assessment and provides actionable insights for conservation planning in climate-sensitive, arid ecosystems.