Advances in neural network modeling of phytoplankton primary production

Neural networks are powerful tools for phytoplankton primary production modeling, even though their application might be hindered by the limited amount of available data. Some new approaches that could enhance neural network models to overcome this problem are presented and discussed in this paper. For instance, co-predictors allow to improve neural network estimates when additional inputs from a broader range of variables are selected. Theoretical knowledge about biological processes can be easily embedded into neural network models by means of a constrained training procedure. Finally, information derived from both existing models and real data can be effectively exploited by a metamodeling approach. Since the underlying rationale applies to a wide spectrum of problems, the proposed approaches are not confined to phytoplankton primary production modeling, but they can also play a role in other ecological applications.