Modelling primary production in a coastal embayment affected by upwelling using dynamic ecosystem models and artificial neural networks

Two modelling approaches, dynamic ecological simulation and neural network analysis, were used to describe and predict the main patterns of primary production temporal variability in a coastal embayment affected by upwelling (Ria de Arousa, Western Spanish Coast). A one dimensional, carbon based, size-dependent dynamic simulation model physically forced by solar radiation, temperature, upwelling index and mixed layer depth was developed using object-oriented programming. The model is defined by six biological compartments: nanophytoplankton, microphytoplankton, microzooplankton, mesozooplankton, bacteria and cultured mussels, was tuned with a 3-year data series (1992–1994) from the region and validated using data collected in the same area in 1995 and 1996. The model reproduces both seasonal and interannual patterns and magnitudes of nutrient concentration and phytoplankton biomass. Neural network models were also developed using backpropagation networks with one or two hidden layers and sigmoid and sinusoidal activation functions. The correlation between observed and modelled phytoplankton biomass from 1992 to 1994 were 0.99 and 0.71 for daily and weekly predictions, respectively. Both modelling approaches yield valuable information. The dynamic simulation model contributes to a better understanding of cycling of matter through planktonic food webs but, although reproducing the main patterns of large-scale variability, its predictive potential is low due to the large uncertainty associated with parameter estimation. By contrast, the neural network model, although not providing information on ecosystem functioning, has demonstrated to be a powerful predictive tool for short (daily to weekly) time scales.