Using optimized dimensionality reduction and machine learning to explain driving processes of phytoplankton community assembly in large mountain rivers

• Phytoplankton community assembly in large mountain rivers is top-down controlled in summer and bottom-up driven in autumn. • The PSA-PP-BRT method, which combines PSA-PP and machine learning, is proposed to integrate ecological factors. • The PSA-PP-BRT method captures high-dimensional data characteristics and preserves ecological significances. • Complex phytoplankton assembly processes are shown as a clear network of relationships by integrating impact indicators. Large mountain rivers create unique aquatic habitats and become one of biodiversity hot spots within ecosystems. Since phytoplankton constitutes a crucial component of primary productivity in aquatic ecosystems, understanding its community assembly processes is the basis of analyzing complex ecological dynamics in large mountain rivers. Currently, the elucidation of aquatic community assembly processes is primarily achieved by integrating multiple factors. However, enhancing the utilization of data structural features and ecological interpretability in multi-dimensional data integration remains a challenge. In this study, we employed a methodology combining optimized dimensionality reduction with advanced machine learning to construct a path analysis model for explaining the driving processes underlying phytoplankton community assembly in large mountain rivers. According to the boosted regression trees (BRT), total nitrogen (21.08%) and ammonia nitrogen (17.87%) were identified as primary factors influencing phytoplankton community assembly, followed by permanganate index (17.14%) and zooplankton community dissimilarity (14.41%). Further, factor integration was performed using projection pursuit optimized by PSA, combined with BRT. Based on the dimension-reduced environmental differences, nutrient resources, and zooplankton, a path analysis model was constructed to determine top-down and bottom-up effects on phytoplankton community assembly. The path model revealed that phytoplankton community assembly is primarily driven by bottom-up effects of environmental differences and nutrient resources during summer, and by top-down effects of zooplankton during autumn. This may result from the clear-water phase of phytoplankton and flow characteristics of mountain rivers. The seasonal variations of phytoplankton are associated with significantly different roles of deterministic factors across seasons. We offer a method for analyzing the aquatic community assembly and provide a theoretical basis to support future dynamics prediction of phytoplankton.