Analyzing Atmospheric River Reforecasts: Self-Organizing Error Patterns and Synoptic-Scale Settings

The societal benefit of accurate operational prediction of atmospheric rivers (ARs) is well established, especially for the western United States. Utilizing a customized high-resolution AR prediction system and 34 water-year reforecast (West-WRF), we identify flow patterns associated with specific error patterns and predictive skill. This process helps document model error sources and ultimately address forecast deficiencies. Using self-organizing maps (SOMs), we sort thousands of AR reforecasts at 72-and 144-h lead times, grouping them based on different training variables to create 2D matrices of synoptic-scale patterns. Initial efforts to train SOMs using West-WRF integrated vapor transport (IVT) failed to show strong skill differences between pattern groupings. However, SOMs trained on IVT differences between reforecasts (West-WRF) and reanalyses (ERA5) elucidate different meteorological patterns of forecast errors. Primary IVT error patterns are related to AR translation speed or intensity errors. Flow patterns associated with the characteristic errors are consistent with the skill determined using object-oriented verification metrics. Groupings subset by El Niño–Southern Oscillation (ENSO) and Madden–Julian oscillation (MJO) illustrate elevated error frequency with some phases. At a 6-day lead time, El Niño conditions preferentially accompany forecasts of negative AR intensity bias, while La Niña conditions favor forecasts with positive translation speed and intensity biases. A positive translation speed bias also tended to favor MJO phase 5, while the opposite tendency was observed for negative translation speed biases. Meteorological fields associated with these AR translation speed and intensity biases are also analyzed, allowing the selection of representative cases for detailed analysis.