From chaos to order: cross-kingdom coordination buffers microbial communities against typhoon impacts

Typhoon disturbances profoundly reshape coastal microbial ecosystems; however, the coupled responses and co-successional dynamics of bacterial and phytoplankton communities remain poorly understood. Identifying their interactions and key drivers is essential for understanding ecosystem resilience and predicting post-disturbance algal dynamics.

Using Zhanjiang Bay as a model system, we analyzed microbial succession across two typhoon events with contrasting intensities. High-throughput sequencing and environmental data were integrated with multiple machine learning models to identify the influence patterns of bacterial communities on phytoplankton diversity. SHAP analysis was applied to quantify feature contributions and detect threshold effects, and network analysis was used to characterize cross-kingdom interactions across successional stages.

Microbial communities exhibited consistent directional succession under both typhoons, although Prapiroon induced stronger displacement (2.38-fold in bacteria and 2.53-fold in phytoplankton). Random Forest achieved the best predictive performance (R² = 0.74, RMSE = 0.23). Temperature and the rare taxon Bdellovibrionota were identified as key drivers, showing clear threshold effects: phytoplankton diversity declined above 30.76 °C and increased when Bdellovibrionota exceeded 0.02%. Diversity peaked immediately after disturbance and declined during recovery, indicating a stage-dependent regulatory shift. Microbial networks followed a "collapse-reorganization-recovery" trajectory, reflecting ecosystem resilience.

Our findings reveal a predictable, stage-dependent framework of microbial succession under typhoon disturbance, driven by environmental filtering and biotic interactions. The identified thresholds and microbial indicators provide a basis for recognizing critical temporal windows of elevated bloom risk, highlighting the potential of integrating microbial data with machine learning for adaptive coastal management.