Machine Learning Reveals the Association Between Aridification and Sediment Bacterial Community Changes in the Linxia Basin, NE Tibetan Plateau

Microbial communities are widely distributed on Earth and closely linked to environmental and climate change. However, the microbial composition of sediments in the Linxia Basin on the northeastern Tibetan Plateau and its relationship with paleoclimate remain poorly understood. Here, we analyzed bacterial communities in 282 sediment samples spanning different stratigraphic units and 9 surface loess samples using 16S rRNA gene sequencing. Loess microbial communities differed markedly from those in sediments. Bacterial community composition varied with geological formations, sedimentation rates, and vegetation stages, as supported by principal coordinate and canonical correspondence analyses. Notably, significant shifts in bacterial α-diversity coincided with stratigraphic transitions associated with aridification events. We further identified 25 microbial markers and developed eight machine learning models to predict aridification. The random forest model showed the best performance, achieving an area under the curve (AUC) of 0.993 in the discovery dataset and 0.900 in an external validation cohort. Five α-diversity indices and ASV18 (unclassified_Enterobacteriaceae) were the strongest predictors and were negatively associated with aridification. These findings demonstrate the potential of sediment microbial signatures as reliable proxies for reconstructing paleoclimatic aridification in the northeastern Tibetan Plateau.