Advances in Neural Network Modeling of Phytoplankton Primary Productivity (2000).

University of Bari, Department of Zoology, Via Orabona 4, 70125 Bari, Italy
 

Empirical modeling of phytoplankton primary production modeling has always been based on predictive variables (mainly phytoplankton biomass and light intensity) that are more easily available and cheaper to measure than primary productivity. Since direct primary productivity measurements are also difficult and time consuming, the role of empirical models in oceanographic research is a major one. This is especially true if the huge amount of information about predictive variables that can be obtained by remote sensing is taken into account.Phytoplankton biomass, surface irradiance and transparency (expressed as light extinction coefficient or as depth of the photic zone) were always used as predictive variables, whereas an estimate of phytoplankton PP was the only neural network output. Other variables that tend to co-vary with phytoplankton PP were added to this basic set during model development. Specifically, geographical coordinates, julian day, salinity, and depth of the water column were taken into account, as well as a few derived variables (e.g. the depth-integrated phytoplantkon biomass).

In this framework artificial neural networks provide an effective alternative to conventional modeling techniques and some practical applications to phytoplankton primary productivity modeling have been already presented.

Empirical models based on artificial neural networks not only easily outperform regressive models, but they also allow to exploit information that is correlated to the variable to be predicted (i.e. primary productivity) even though a direct causal relationships either does not exist or is not fully understood.

Moreover, since neural network models are quite robust with respect to redundant inputs, these "co-variables" can play an important role in improving existing models. Bathymetric data, for instance, are not directly related to phytoplankton primary production, but are certainly related to water column dynamics, which in turn is related to nutrient availability, that certainly drives primary production. Examples of the usage of such "co-variables" will be presented within real models.

Another interesting development in neural network models of phytoplankton primary productivity will be discussed. It will be related to the way these models may be coupled with conventional ones in order to get better estimates, by integrating data (training set) and knowledge (conventional models outputs) into a single empirical tool through a metamodeling procedure.