How do forecast model biases affect large-scale teleconnections that control Southwest US precipitation? - Part I: S2S models

Abstract

Abstract We analyze biases in subseasonal forecast models and their effect on Southwest United States (SWUS) precipitation prediction (2–6 week timescale). Cluster analyses identify three primary wavetrains associated with SWUS precipitation: a meridional El Niño–Southern Oscillation-type (ENSO) wavetrain, an arching Pacific–North American-type (PNA) wavetrain, and a circumglobal zonal wavetrain. Compared to reanalysis, the models overrepresent the arching pattern, underrepresent the zonal pattern, and produce mixed results for the meridional pattern. The arching pattern overrepresentation is linked to model mean flow biases in the midlatitude-subpolar North Pacific, which cause a westward retraction of the region of forbidden linear Rossby Wave propagation. The zonal pattern underrepresentation is linked to westerly biases in the subtropical jet, which cause a westward retraction of the waveguide in the midlatitude Eastern North Pacific and divert wavetrains southward. These results are confirmed using linear, barotropic ray tracing analysis. In addition to mean state biases, the models also contain errors in their representation of the Madden-Julian Oscillation (MJO). Tropical convection anomalies associated with the MJO are too weak and incoherent at lead times greater than two weeks, when compared to reanalysis. Additionally, there is a strong SWUS precipitation signal as far out as 5 weeks after a strong MJO in reanalysis, associated with its persistent eastward propagation, but this signal is absent in the models. Our results indicate that there is still significant room for improvement in subseasonal predictions, if we can reduce model biases in the background flow and improve the representation of the MJO.