Dissolved oxygen (DO) is a master variable in aquatic ecosystems, dictating biogeochemical reactions and shaping biological communities. DO is influenced by a complex suite of biological and physical/hydrologic processes that control variation in DO over diel to seasonal time scales. In non-perennial streams that experience cycles of wetting and drying, this dynamic flow regime (including shifts between wet and dry, and flowing and pooled states) can exert strong controls on DO behavior, leading to distinct DO regimes with differing controls under divergent flow conditions.
We combine a unique dataset of high spatio-temporal resolution data from a non-perennial stream network in Kansas, USA with a novel machine learning approach (Tandem Evolutionary Algorithm) to identify the unique combinations of hydrometeorological and biological drivers that are associated with high and low DO concentration states in an intermittent river system, and assess how these drivers shift across varying flow states. Our analysis revealed equifinality in the drivers of DO concentrations, indicating that no single set of hydrometeorological or biological variables are associated with high or low DO concentrations at our study site. Rather, both high and low DO concentrations are associated with multiple distinct yet recurrent pathways. Taken together, these results show that dynamic hydrologic conditions, specifically changes in flow magnitude, velocity, groundwater contributions, and network connectivity, play a central role in reorganizing the mechanisms that drive high or low DO concentrations across a wide range of flow conditions.