Hindcasting and Decoding the North Atlantic Subpolar Gyre Index: Past Variability and Future Climate Implications

The North Atlantic Subpolar Gyre (NASPG) is a key driver of ocean circulation and nutrient redistribution, with major implications for marine productivity and ecosystem functioning. By regulating silicate transport, the gyre supports diatom growth, which underpins the structure of the food web and carbon sequestration. Variations in gyre strength and extent cascade through the ecosystem, from plankton to fish, ultimately shaping productivity and fisheries yields. Capturing these dynamics requires reliable indices of gyre variability. However, existing indices are relatively short, beginning only in the satellite era, and thus limit our ability to assess long-term links between gyre dynamics and ecological responses.

This project addresses that gap by hindcasting a multi-decadal SPG index extending back to before the 1990s, when consistent satellite observations first became available. The index reconstruction combines statistical and deep learning approaches. As a baseline, Empirical Orthogonal Function (EOF) analysis coupled with regression is used to extract dominant spatial patterns from recent decades and apply them to historical model data, producing a simple and interpretable reconstruction. To capture non-linear relationships and provide uncertainty estimates, we also implement Quantile Convolutional Neural Networks (QCNNs). These models integrate multiple gridded oceanographic variables, including sea surface height, temperature, salinity, wind stress, and mixed layer depth, while quantile outputs generate confidence intervals for hindcast predictions.

The extended index is then used to investigate long-term variability in the SPG since the 1950s, focusing on key features such as long-term trends, shifts in periodicity, and evidence of regime changes. Finally, we assess how the gyre might respond under future climate scenarios, providing a broader perspective on circulation variability and its ecological consequences. By combining statistical and machine learning methods with extended temporal analyses, this project establishes a robust foundation for understanding the past and future variability of the NASPG.