Deep learning algorithm to uncover links between satellite-derived physical drivers and biological fields.

Understanding the dynamics of phytoplankton communities in response to physical
environmental changes is essential for evaluating the impact of climate change on marine
ecosystems. Satellite observations provide a rich dataset spanning over two decades,
capturing physical sea surface parameters such as temperature, salinity, and sea surface
height, alongside biological insights such as ocean color. Ocean color data, in particular, is
processed to estimate sea surface chlorophyll-a concentrations — a widely recognized proxy
for phytoplankton biomass. Recent advancements in ocean color observation have further
enabled the characterization of phytoplankton community structure in terms of functional
groups or size classes.
However, linking satellite-derived physical parameters to biological indicators remains
challenging due to spatial and temporal variability.
Can physical data reliably predict patterns in ocean color, such as chlorophyll-a
concentrations and phytoplankton community structures, and potentially assess their
variations? This study addresses this question through a deep-learning approach, utilizing
an attention-based autoencoder model to learn relationships between physical variables and
ocean color data, including chlorophyll-a concentrations and phytoplankton size classes at
weekly and 1° spatial resolution.
Our trained deep-learning model effectively captures patterns and correlations between
physical parameters, chlorophyll concentrations, and phytoplankton size classes. It enables
detailed exploration of how physical factors influence biological variability across different
temporal scales. Utilizing a phytoplankton database spanning 1997–2023, this approach
demonstrates promising results in replicating chlorophyll concentrations, inferring
phytoplankton size classes, and shedding light on the potential links between physical and
biological data.
This study highlights the potential of machine learning for ecological research, contributing to
more accurate trend analyses. Understanding phytoplankton variability is critical for marine
ecosystem management, given their role in global carbon cycling. This methodology
underscores the value of deep-learning to anticipate phytoplankton dynamics under
changing environmental conditions.