Multi‐Model Subseasonal Prediction Skill Assessment of Water Vapor Transport Associated With Atmospheric Rivers Over the Western U.S.

Abstract

AbstractSubseasonal‐to‐seasonal (S2S) forecasts of atmospheric rivers (ARs) are in high demand in the water supply management and flood control communities. This study focuses on a new metric, the accumulated water vapor transport associated with ARs, which is closely related to the winter precipitation over the western U.S., and provides a multi‐model S2S prediction skill assessment. The prediction skill is evaluated at lead time 1–4 weeks in four dynamical model hindcast data sets from National Centers for Environmental Prediction (NCEP), European Center for Medium‐Range Weather Forecasts (ECMWF), Environment and Climate Change Canada (ECCC), and Global Modeling and Assimilation Office (GMAO) at National Aeronautics and Space Administration. Three reanalysis data sets are used to evaluate the uncertainty of prediction skills related to the choice of references. The AR‐related water vapor transport is underestimated in ECMWF and ECCC over most of the investigated region, while its maximum has a southeastward shift in NCEP and GMAO at lead time 3–4 weeks. The root mean square error, anomaly correlation coefficient, and Brier skill score are calculated to quantify the prediction skill in both deterministic and probabilistic sense. At week‐3 lead time, the models have significant skill near the lower latitudes (<40°N) of the eastern North Pacific, extending northeastward to the California coastal area. Models have higher skills in forecasting no and strong AR cases than weak cases. The Madden–Julian Oscillation can modulate the prediction skill at week‐3 lead over central and Southern California, but with large uncertainties across models.