Empirical Teleconnections: A Data-Driven Approach for Improving Seasonal Forecasting

Atmospheric teleconnections such as the El Niño Southern Oscillation (ENSO) are thought to be important drivers of dry and wet periods around the globe. Yet, even well-studied teleconnections like ENSO have weak statistical relationships with dry/wet periods in most regions and offer limited predictive skill. Here, we demonstrate an exhaustive, bottom-up, brute-force search algorithm for discovering predictive relationships between specific ocean-land regions. Based on 32 years of reanalysis data (1983-2015), we discover linkages that are distinct from known teleconnections, but which better predict precipitation with 4-12 months lead-time. We further evaluate our empirical approach by reversing the analysis, demonstrating that the predictive skill we observe is not mere coincidence; the results have strong spatial coherence. The teleconnections revealed here may have immediate practical value in anticipating seasonal weather events, and also help to focus future research on the physical mechanisms underlying the planet's strongest and most important teleconnections.