Assessment of atmospheric river subseasonal prediction skill in GFDL SPEAR model hindcast
Due to considerable social and economic implications, there is a continuously increasing demand for skillful subseasonal-to-seasonal forecasts of weather and climate extremes. Atmospheric rivers (ARs) characterized by intense lower tropospheric plumes of moisture transport are essential to midlatitude extreme precipitation. This study aims to provide a global evaluation of subseasonal prediction skill of ARs out to a 4-week lead based on a 10-member 20-year hindcast experiment using the recently developed Seamless System for Prediction and Earth System Research (SPEAR) at the Geophysical Fluid Dynamics Laboratory (GFDL). We apply an aggregate measure to quantify the prediction skill by counting AR occurrence days within a week-long period (from AR-week1 to AR-week4). Reliable subseasonal forecast skill of ARs is detected up to 3 weeks lead over the subtropical to midlatitude North Pacific and North Atlantic (winter) and East Asia (summer), with slightly higher forecast skill in winter compared with summer seasons. The SPEAR forecast model shows systematic negative bias over East Asia, Gulf Stream, and surrounding regions relative to the ERA5 reanalysis, especially during summer seasons. We further compare the SPEAR model's overall subseasonal AR forecast skill with several other forecast models such as ECMWF and NCEP. Besides, the reliability of the SPEAR model in forecasting subseasonal ARs is examined from the perspective of the so-called "signal-to-noise paradox," which implies whether a specific model is overestimating or underestimating the prediction skill. The potential impacts of the El Niño–Southern Oscillation and Madden–Julian Oscillation on the magnitude and subseasonal AR prediction skill are discussed with a specific focus on western North America.