Application of an Artificial Neural Network to Improve Understanding of the Observed Conterminous US Winter Precipitation Response to ENSO

Abstract

El Niño Southern Oscillation (ENSO) is known to modulate rainfall variability in parts of the conterminous United States (US). Owing to the complexity of the climate system, the variability in US winter (DJF) precipitation response to ENSO is investigated. By regressing autoencoder neural network-based ENSO types (i.e., encoded tropical Pacific Sea surface temperature anomaly patterns) onto DJF US precipitation, supplemented with support vector regression and extreme gradient-boosting regression, we show that ENSO modulation of precipitation is regionally sensitive to the ENSO type. Certain regions exhibit significant nonlinear relationships between precipitation and strong ENSO event phase that was most pronounced over the eastern and northwestern quadrants of the US. The coherency of the response varies among individual events. Specifically, among individual events, differences in ENSO SST anomaly patterns were linked to meridional shifts in the positioning of the Pacific jet stream. This leads to variable anomalous upper-level flow and atmospheric conditions influencing US winter precipitation during the ENSO events. By analyzing associations between DJF precipitation and ENSO types whilst assessing the consistency of precipitation anomalies during strong ENSO events, we identify the regional likelihood of consistent precipitation responses, thereby calibrating confidence in seasonal ENSO-precipitation responses.