Communicating Artificial Neural Networks with Physical-Based Flow Model for Complex Coastal Systems

Abstract

Four applications using an intelligent computational method, such as Artificial Neural Networks (ANNs) to incorporate with physical-based flow model, namely, to enhance of knowledge base for model calibration, to correct simulation errors, to perform inverse modeling estimation, and to perform surrogate model approach have been studied recently. In this paper, the first application which enhances the knowledge base for physical-based flow model’s calibration is demonstrated. For the most physical-based model, invariably there are problems collecting complete data sets to evaluate and model estuarine or coastal hydrodynamics systems. These data gaps lead to uncertainties associated with the understanding or modeling of these processes. One new approach to reduce this uncertainty and to fill in these data gaps is based on the analysis of all the data characterizing the system using ANNs. To demonstrate this data-driven approach, the computational efforts undertaken for the Houma Navigation Canal (HNC) project, Louisiana, USA, is presented. For this project, the following three major benefits are obtained and presented from many “pieces” of hydro-environmental information. They are: to fill in the data gaps so that complete data sets are available for the largest number of instrumentation locations, to develop a better understanding of the system responses, and to understand the sensitivity of the system variables. This knowledge base is important to understand and assist with the physical-based model development and calibration/verification processes.