Feature Engineering for Subseasonal-to-Seasonal Warm-Season Precipitation Forecasts in the Midwestern United States: Toward a Unifying Hypothesis of Anomalous Warm-Season Hydroclimatic Circulation

The literature has established dozens of potential predictive indices (PIs) of anomalous warm season precipitation in the Midwestern United States that could have utility in subseasonal to seasonal (S2S) forecasts. This analysis posits that these predictive indices relate to one of three "modes of action" that work in tandem to drive anomalous hydroclimatic circulation into the continental interior. These include contributions from 1) geostrophic mass flux, 2) ageostrophic mass flux, and 3) atmospheric moisture supply, and represent semi-independent, interactive forcings on S2S precipitation variability. This study aggregates 24 PIs from the literature that are related to the three modes of action. Using an interpretable machine learning algorithm that accounts for nonlinear and interactive responses in a noisy predictive space, we evaluate the relative importance of PIs in predicting S2S precipitation anomalies from March to September. Physical mechanisms driving PI skill are confirmed using composite analysis of atmospheric fields related to the three modes of action. In general, PIs associated with ageostrophic mass flux anomalies are important in early summer, while PIs associated with Atlantic-sourced atmospheric moisture supply are important in late summer. At a 2-month lead, PIs associated with continental-scale thermodynamic processes are more important relative to PIs associated with local convective phenomena. PIs representing geostrophic mass flux anomalies are also critical throughout the warm season, in real time and at a 1-2-month lag, but particularly during transitional months (spring/fall). Several new PIs describing zonal and meridional asymmetry in hemispherical thermal gradients emerge as highly important, with implications for both S2S forecasting and climate change.