Sub-Seasonal Predictability of Heatwave Onset using Coupled Ocean-Atmosphere Teleconnections

Abstract

Heatwaves are among the deadliest climate extremes, and their impacts on human health, agriculture, energy demand, and infrastructure are strongly dependent on the amount of advance warning available to decision-makers. Weather forecasts provide reliable skill only to about 10–14 days, while seasonal forecasts operate at climatological, not event-scale, resolution, leaving a critical "sub-seasonal" gap of roughly two to six weeks in which actionable heatwave-onset prediction is most needed but least mature. This paper investigates the sub-seasonal predictability of heatwave onset by examining coupled ocean-atmosphere teleconnection patterns, including the El Niño–Southern Oscillation (ENSO), the Madden-Julian Oscillation (MJO), and mid-latitude atmospheric blocking, as sources of extended-range predictive skill. The study reviews observational and reanalysis-based evidence linking specific MJO phases and ENSO states to anomalous ridging, soil-moisture feedbacks, and Rossby wave train propagation associated with heatwave initiation over North America, Europe, and East Asia. A composite and lagged-correlation methodology is described using ERA5 reanalysis, NOAA Optimum Interpolation Sea Surface Temperature (OISST) data, and the Real-time Multivariate MJO (RMM) index, alongside statistical and machine-learning-based predictive frameworks (logistic regression, random forest, and LSTM architectures) applied in the subseasonal-to-seasonal (S2S) prediction literature. Comparative skill metrics reported across studies indicate that heatwave predictability at 3–4 week lead times is substantially enhanced when MJO phase and ENSO state are used as joint predictors relative to climatology alone, though skill remains regionally heterogeneous and highly sensitive to the underlying dynamical driver. The paper concludes by identifying persistent challenges in operational S2S heatwave prediction, including model bias in MJO teleconnection representation, land-atmosphere coupling errors, and the need for regionally tailored predictor combinations.