An artificial neural network analysis of factors controlling ecosystem metabolism in the coastal ocean

Knowing the metabolic balance of an ecosystem is of utmost importance in
determining whether the system is a net source or net sink of carbon dioxide to the
atmosphere. However, obtaining these estimates often demands significant amounts of time
and manpower. Here we present a simplified way to obtain an estimation of ecosystem
metabolism. We used artificial neural networks (ANNs) to develop a mathematical model of
the gross primary production to community respiration ratio (GPP:CR) based on input
variables derived from three widely contrasting European coastal ecosystems (Scheldt Estuary,
Randers Fjord, and Bay of Palma). Although very large gradients of nutrient concentration,
light penetration, and organic-matter concentration exist across the sites, the factors that best
predict the GPP:CR ratio are sampling depth, dissolved organic carbon (DOC) concentration,
and temperature. We propose that, at least in coastal ecosystems, metabolic balance can be
predicted relatively easily from these three predictive factors. An important conclusion of this
work is that ANNs can provide a robust tool for the determination of ecosystem metabolism
in coastal ecosystems.