Inferring SST-Forced Seasonal Atmospheric Responses: An Ensemble Empirical Tool

In regions with marginal seasonal forecast skill, such as the extratropical latitudes, attributing the origins of a predictive signal is a cornerstone of effective forecast communication. Traditional methods rely on historical observations to establish statistical relationships between predictors, such as tropical sea surface temperatures (SSTs), and atmospheric predictands. However, these observations contain a mixture of the SST-forced signal and unrelated intrinsic variability (noise), which can obscure the analysis. This study introduces Ensemble Principal Component Regression (EPCR), a method that addresses this challenge by using the noise-free, ensemble mean atmospheric response from model simulations. Using DJF 200-hPa geopotential height (z200) anomalies from a 100-member AMIP ensemble, we demonstrate its utility as a versatile diagnostic tool for: (1) attributing forecast anomalies to specific SST modes, (2) providing a "sanity check" for small-ensemble forecasts by filtering residual noise, and (3) establishing a statistical forecast baseline. Results show the predictable z200 signal is overwhelmingly low-dimensional, driven primarily by the atmospheric responses to ENSO and the global warming trend. The EPCR framework successfully attributes the predictable signal to these modes, as demonstrated for the DJF 2024 forecast. As a diagnostic, it effectively serves as a "sanity check" by identifying and filtering potential noise contamination in small ensembles. While not intended to replace the full dynamical model in terms of forecast skill, EPCR’s primary value lies in its ability to provide a clear, physically-grounded baseline of the linear, SST-forced signal, thereby establishing a robust diagnostic for interpreting and adding confidence to seasonal forecasts.