Evaluation of Weather Forecast Errors during Major Atmospheric River Sequences Impacting California during Winter 2022-2023

During the period between 24 December 2022 and 19 January 2023, California was impacted by an active streak of atmospheric river (AR) activity that produced unprecedented rainfall, flooding, and snowpack accumulations. This study reports on the numerical weather prediction (NWP) forecast verification of AR-related characteristics, including intensity and location, and scale related precipitation and freezing level predictions. Assessing these forecasts can lead to better overall understanding of limits on predictability on synoptic-to-mesoscale variability of extreme events and the impacts it might have on water resource availability on flood-to-seasonal time scales.

Throughout the study period, AR activity, denoted by integrated vapor transport (IVT) > 250 kg m-1 s-1, was forecasted up to 10-13 days in advance in Northern California by NOAA’s Global Ensemble Forecast System (GEFS) and the European Ensemble (ECMWF-EPS) models. Deterministic forecast models trended towards a southerly bias in the prediction of AR landfall location, but location was well forecasted up to 5-days lead time across the entire U.S. West Coast during the period of interest. Each AR had unique challenges in terms of its intensity forecast error (both positive and negative), thus warranting further meteorological investigation. Several key watersheds, including the Cosumnes basin which experienced significant flooding on New Year's Day, had forecast errors up to 100 mm at a 2-day lead time using the GFS. Other high resolution models, like CW3E’s West-WRF near-real time forecasts, improved upon this localized forecast error. Across California, several of these ARs had skillful precipitation forecasts of >25.4 mm up to 7-10 days in advance. Finally, West-WRF forecasts captured changes in the freezing level of at least 1.5 km three separate times between late December 2022 and mid-January 2023. The most notable being a rise of approximately 1.6 km in 18 hours at the Oroville snow level profiler ahead of an AR that made landfall on 29 December. Freezing levels during this event were above a majority of the Feather River watershed, leading to rain-on-snow throughout the watershed. Together, these results synthesize muti-scale forecast skill that can be leveraged with other studies to identify sources or cases of meteorological and/or hydrologic characteristics that are important for predictability.