Combining in situ flow cytometry and artificial neural networks for aquatic systems monitoring

In order to produce a system to automatically identify field water samples, it is essential to cover the entire spectrum of biological variation that a species can be found in the natural environments. This information must be available for modeling within specific training data sets. Thus, the one of the objectives of this work is to build a set of flow cytometric data containing this information in order to develop artificial neural network models that learn the patterns of biological variation induced by some environmental parameters. The second goal is to test the model in near real time recognition of phytoplankton. Twelve isolated groups were assayed in order to define their optical signature boundaries. Our results show high performance of a Radial Basis Function Neural Network in the test data set and its recognition and enumeration capability when assessing field data. It also suggests that it would be better to use a more generalist model for the different phytoplankton groups and more specialized networks to deal with specific organisms within a taxon. A discussion about the use of this type of model in monitoring programs is presented.