Built a hierarchy of machine-learning models of varying complexity to examine the relative importance of global air temperatures, daily indices of synoptic meteorology & time-averaged soil moisture (SM) to correct characterization of the variability of equivalent potential temperature (θE).
We show: Deep machine learning (non-linear) models are necessary to correctly model high intensity, high impact heat events. Excluding SM precludes our ability to model high intensity, high impact heat events. This research re-emphasizes the importance of atmosphere-surface exchange to extreme heat events in the eastern USA.
Our research has evolved mechanistic understanding of θE variability and illustrated the need for high-resolution modeling to characterizing extreme temperature events.