Causal Analysis of Streamflow, Evapotranspiration and Snow Dynamics in Large Sample Hydrology

A key challenge in hydrological science lies in establishing causal relationships among environmental variables across very diverse systems, given only limited observations. Without such causal understanding, it is difficult to determine whether simulation models adequately represent hydrological processes and their responses to climate change or other anthropogenic forcings. To address this challenge, we tailor a combination of causal discovery and effect estimation to hydrological systems and apply it to daily observations and simulations from 671 heterogeneous catchments across the USA. The inferred causal networks resolve the direction, strength, lag, and seasonal timing of key interactions and quantify shifts in process controls along gradients of water and energy availability. When applied to multiple versions of a widely used hydrological model, the approach reveals structural differences overlooked by conventional statistical metrics. More broadly, our results establish causal inference as a unifying approach that connects observations, theory, and models in heterogeneous environmental systems.