Atmospheric rivers in 20 year weather and climate simulations: A multimodel, global evaluation

Atmospheric rivers (ARs) are narrow, elongated, synoptic jets of water vapor that play important roles in the global water cycle and meteorological/hydrological extremes. Increasing evidence shows that ARs have signatures and impacts in many regions across different continents. However, global‐scale characterizations of AR representations in weather and climate models have been very limited. Using a recently developed AR detection algorithm oriented for global applications, the representation of AR activities in multidecade weather/climate simulations is evaluated. The algorithm is applied to 6‐hourly (daily) integrated water vapor transport (IVT) from 22 (2) global weather/climate models that participated in the Global Energy and Water Cycle Experiment Atmospheric System Study–Year of Tropical Convection Multimodel Experiment, including four models with ocean‐atmosphere coupling and two models with superparameterization. Multiple reanalysis products are used as references to help quantify model errors in the context of reanalysis uncertainty. Model performance is examined for key aspects of ARs (frequency, intensity, geometry, and seasonality), with the focus on identifying and understanding systematic errors in simulated ARs. The results highlight the range of model performances relative to reanalysis uncertainty in representing the most basic features of ARs. Among the 17 metrics considered, AR frequency, zonal IVT, fractional zonal circumference, fractional total meridional IVT, and three seasonality metrics have consistently large errors across all models. Possible connections between AR simulation qualities and aspects of model configurations are discussed. Despite the lack of a monotonic relationship, the importance of model horizontal resolution to the overall quality of AR simulation is suggested by the evaluation results. Key Points An IVT‐based AR detection algorithm is applied to 20 year weather and climate simulations from 24 global models Among the 17 metrics considered, seven have consistently large errors across all models relative to observational uncertainty The importance of model horizontal resolution to the overall quality of AR simulation is suggested