An unprecedented heatwave caused the surface temperature of the North Sea to rise more than 3°C above the seasonal average, triggering a proliferation of phytoplankton up to three times higher than usual levels. However, satellite images that scan the ocean's color detected no visible anomaly. This discrepancy prompted an international team of researchers to dive beneath the surface using autonomous gliders and machine learning-based prediction models, revealing a phenomenon that orbital sensors could not see.
What the study reveals
In-situ measurements show that the concentration of chlorophyll-a, a key indicator of phytoplankton, reached localized peaks of over 5 mg m-3 in the thermocline layer, while satellite data showed only a slight increase in surface color, below 0.1 mg m-3. This difference is explained by heat being trapped at depth, creating a 'subsurface warm layer' where available light remains sufficient to fuel photosynthesis. The phenomenon lasted nearly three weeks, long enough to alter marine carbon dynamics and potentially influence local trophic networks.
Comparing observations with ECMWF reanalyses and Copernicus products, scientists found that predictive models, powered by a neural network trained on decades of atmospheric and oceanographic data, were able to anticipate the formation of the subsurface warm layer with a forecast horizon of 5 to 7 days. However, these models could not predict the exact intensity of the phytoplankton response, highlighting the current limitations of machine learning when it comes to rare and very localized events.