Neural network modeling of monthly salinity variations in oyster reef in Apalachicola Bay in response to freshwater inflow and winds

Abstract

Estuarine organisms have varying tolerances and respond differently to salinity. Bottom-dwelling species such as oysters tolerate some change in salinity, but salinity outside an acceptable range will negatively affect their abundance as well as their survival within this sensitive ecosystem. Salinity in the Apalachicola Bay is heavily influenced by freshwater inflow discharged from the Apalachicola River. In this study, artificial neural network (ANN) was applied to correlate the monthly salinity variations at an oyster reef in Apalachicola Bay to the river inflow and wind. Parameters in the ANN were trained until the simulated salinity data correlated well with the observations from 2005 to 2007. Once the model is trained and optimized, the ANN structure is verified comparing the simulated data to the second dataset from 2008–2010. Four neural network training algorithms, including gradient decent, scaled conjugate gradient, quasi-Newton, and Levenberg–Marquardt, have been evaluated. The scaled conjugate gradient algorithm was selected for this study because it provides the best correlation with the value of 0.85. The verified ANN model was applied to investigate the potential impacts of freshwater reductions from upstream river on the salinity in the oyster reef. By comparing the resulting salinity from ANN model simulations to the optimal salinity range for oyster growth, the impacts of freshwater reduction scenarios on oyster growth can be examined.