International research efforts related to atmospheric rivers (ARs) have expanded substantially in the past 10 years, with numerous articles related to AR variability, AR impacts, ARs impacts on the global and local hydrologic cycle, and the effects of climate change on ARs. A large set of these recent advances in AR science have benefitted from application of objective, automated AR detection methods. A growing number of AR detection methods have been applied in the literature, and recent research associated with the Atmospheric River Tracking Method Intercomparison Project (ARTMIP) has shown that uncertainty related to AR detection may be important for some scientific questions.

In order to quantify the impact of this uncertainty, we analyze output from the ARTMIP Tier 2 CMIP5/6 experiment (multiple AR detection algorithms run on multiple historical and future CMIP models) and from the TECA Bayesian AR Detector (TECA BARD v1.0.1). We show that (1) there is broad agreement among AR detection methods on the spatiotemporal characteristics of ARs, (2) AR detection uncertainty dominates uncertainty in future changes in ARs in many regions, (3) there is considerable spread in how experts identify ARs, and (4) expert uncertainty leads to scientific uncertainty in ARs. We argue that these results highlight a need for more process-oriented research on ARs, specifically aimed to constrain AR detection methods, including: physical theories of AR genesis and dissipation, theoretical constraints on bulk AR properties (e.g., count and size), and more physics-grounded theories for ARs and climate change.