PREDICTING SALINITY IN THE CHESAPEAKE BAY USING BACKPROPAGATION

Managing an aquatic ecosystem requires frequent monitoring of salinity levels. Several environmental factors impact the dynamics of salinity. Recently, regression models have been constructed in order to model the interactions among these factors and to predict salinity values in different regions of the Chesapeake Bay. In this paper, we compare a simple neural network approach with regression. Using nearly 40,000 observations from 34 stations in the Chesapeake Bay, we build and test both regression and neural network models. These models are compared with respect to survey data gathered in the same time period as the one used to construct the models and on new survey data. In general, the neural network models predict salinity value better than the corresponding regression models.