Seasonal forecasting of precipitation-relevant weather types over the United States

This paper evaluates seasonal forecasts of weather types (WTs), i.e., recurring large-scale atmospheric patterns, which have been developed using a clustering method using large-scale predictors to represent precipitation variability across the United States. Forecast quality is assessed using two seasonal hindcast products: from the operational weather community, hindcasts from the European Centre for Medium-Range Weather Forecasts (ECMWF), as well as hindcasts from the Community Earth System Model version 2 (CESM2), a community resource developed by the NSF National Center for Atmospheric Research, a climate research center. WTs are described in terms of their associated precipitation anomalies, and examined in light of their relationship with established climate teleconnections. The spatial precipitation patterns associated with each WT are less well captured in the forecasting systems than the large-scale variables from which the WTs are derived. The WT patterns themselves are well-represented in both seasonal forecasting systems, though on average ECMWF is slightly closer to observations. Forecasted WT frequency results show that both prediction systems have similar skill, with most differences depending on season and WT. Winter WT frequencies are generally more predictable than summer. Both forecast systems capture well the frequency rank order, but underestimate the interannual frequency spread, which could be partially due to ensemble averaging. Analysis shows that forecasting of climate teleconnection indices alone would not be sufficient to represent the precipitation variability associated with the WTs. Comparable results from two initialized Earth system prediction models that originate from different sides of the weather-climate and operations-research spectrum is encouraging and contributes to WT forecasting and multi-model initialized prediction efforts.