Improved understanding of eutrophication trends, indicators and problem
areas using machine learning
Abstract
Eutrophication is a reoccurring problem in coastal regions, including
the North-West European Shelf (NWES). By developing machine learning
model from sparse observations, we reconstruct a gap-free, 7km and
daily, bi-decadal (1998-2020), data-set for
nitrate at the NWES, allowing for much more robust analyses than the
sparse observational data. From the data-set we identify nitrate-limited
coastal areas, which are potentially vulnerable to eutrophication. Apart
from known eutrophication-problem areas, these include additional
coastal zones, which could become problematic under sub-optimal policy,
or management changes. Furthermore, we show only a limited link between
winter nitrate and the size of phytoplankton growth the following year,
suggesting winter inorganic nitrogen might not provide the best
indicator for eutrophication (as used by OSPAR). Finally, we demonstrate
that reduction of nitrate on the NWES in the 1998-2020 period has been
mostly small, with the exception of specific areas, such as the Bay of
Biscay.