Biological response neural network prediction in coastal upwelling field
Gilberto Carvalho Pereira, Ricardo Coutinho, Nelson F. F. Ebecken · WIT transactions on ecology and the environment · 2002
Modeling water quality, forecasting population dynamics to define the ecosystem health is a fimction of some indices (resilience, diversity, production etc.) and yield of many environmental variables interactions, The assessment of properties and processes of costal zone is a major issue in aquatic system management. The aim of this paper is to test and analyse neural network capability in predicting changes of clorophyl-a as a phytoplankton biomass. Available data concern to a weekly medium-term time-series ranging from the end of 1994 to 2001, and it was collected at the Arraial do Cabo near shore upwelling, southeast of Rio de Janeiro state (Brazil). Dealing in particular with high non-linearity among variables, a genetic algorithm package was used to settle the best neural architecture and the specific pre-processing method presented. The results show the spread of predictions given by the models in short, medium and long-term periods. Thus, the conclusions allow us to say that neural networks can successfully be used to improve forecasting of complex impact factors and ecological system behavior.