Identifying thresholds in fish-environment relationships is a goal of many studies and decision making. We compiled landscape-scale, geo-referenced species occurrence datasets to examine and compare patterns of fish assemblage and species-level turnover across stream flow regimes. A gradient forest machine learning approach was used to quantify multi-species threshold responses along climate and hydrologic disturbance gradients in groundwater, runoff, and intermittent streams in the Interior Highlands, USA. Spatial, climate, soils and lithology variables were most important in explaining fish assemblage structure, but their importance varied by flow regime. Thresholds were common but species and assemblage dependent. Assemblage level responses typically showed fewer thresholds than species level responses. Climate and hydrology importance were flow regime and species dependent. This approach can be used to inform conservation efforts for multiple species, individual species of conservation concern and whole assemblages. Results of this study also provide understanding of complex nonlinear environmental and disturbance threshold relationships driving patterns in fish assemblages and species in streams.