NN4CAST Deep-Learning Model Forecasts Early-Winter North Atlantic Climate From October Sea Temperatures Alone, Matching ECMWF Seasonal Skill
A study published in Scientific Reports on November 14 used a deep-learning framework, NN4CAST, to evaluate the non-stationarity of early-winter ENSO teleconnections over the North Atlantic-European sector. The work, by Galván Fraile, Polo, Martín-Rey and colleagues, was received on May 26 and accepted on October 26.
The neural network predicts November-December mean sea-level pressure anomalies from October sea surface temperature (SST) anomalies alone. It achieves skill comparable to the ECMWF seasonal forecasting system SEAS5, and even exceeds it in certain regions such as the North Atlantic-European sector.
Predictability was strongly period-dependent. Significant skill was found over the North Atlantic Oscillation centres of action, especially the Iceland region, during 1970-1995, whereas for the East Atlantic pattern the improvement came in recent decades, 1996-2019.
The authors said the teleconnection shift is mainly associated with Pacific SST anomalies, with recent decades marked by an intensified Southeast Asian Rossby wave source and a southward displacement of the North Atlantic jet stream, and that explainable artificial intelligence can identify state-dependent SST precursors of predictability in the region.
Source: Scientific Reports
Published by GSM · USA (original Chinese edition)
Photo: 图：北大西洋上空一个涡旋和锋面带的卫星影像（NOAA 档案图）。来源：NOAA，公共领域，维基共享资源。 / Photo: Satellite image of a vortex and frontal band over the North Atlantic Ocean (NOAA archive image). Credit: NOAA, public domain, via Wikimedia Commons.
