Modeling Estuarine Salinity Using Artificial Neural Networks
By Christopher Wan
To predict the response of estuarine ecosystems to anthropogenic and natural changes, process-based physical computer models serve as an important tool for simulation of estuarine salinity. Among the school of data-driven parametric models as alternative tools for process-based physical models to simulate environmental variables, artificial neural networks (ANNs) have become an increasingly popular modeling technique over the past two decades (Maren et al., 1990; Schalkoff, 1997; Dawson and Wilby, 2001; Maier and Dandy, 2001; Dawson et al., 2005; Pao, 2008). ANNs is a programming logic model using multivariable calculus and an algorithmic learning process to simulate various functions related with information processing, including pattern recognition, forecasting, and data compression. The logic of ANNs aims to imitate the workings of individual neurons in the human brain, making it able to dynamically model non-linear functions with very high accuracy. In this way, a modeler using ANNs has no need to explore the intermediate processes that occur in the relationship between an input variable and the final output. Instead, the ANN implicitly takes them into account during its learning process. Transport of salt in estuaries is influenced by multiple factors such as freshwater inflows and tide, and their relationship with salinity is highly complex and non-linear, making it ideal cases for the application of ANNs. The objective of this study is to develop ANNs to predict estuarine salinity using the Loxahatchee River as a case study. The Loxahatchee River is selected because of concerns about saltwater intrusion into the river (SFWMD, 2002; 2006; Kaplan et al., 2010; Liu et al., 2011). The hypothesis is that salinity in the Loxahatchee River can be effectively simulated with ANNs, through properly training and testing, using freshwater inflow, rainfall, and tide as inputs.