Summary: | An artificial neural network (ANN) model was established to predict the vertical distribution of Baltic cod eggs. Data from vertical distribution sampling in the Bornholm Basin over the period 1986-1995 were used to train and test the network, while data sets from sampling in 1996 were used for validation. The model explained 82% of the variance between observed and predicted relative frequencies of occurrence of the eggs in relation to salinity, temperature and oxygen concentration; The ANN fitted all observations satisfactorily except for one sampling date, where an exceptional hydrographic situation was observed. Mean ambient temperatures, calculated from the predicted vertical distributions of the eggs and used for the computation of egg developmental times, were overestimated by 0.05 degrees C on average. This corresponds to an error in prediction of egg developmental time of less than 1%
|