A MULTI-LEVEL AND MULTIVARIATE DATA RECOVERY SYSTEM OF HYDROINFORMATICS IN AN ESTUARINE ENVIRONMENT USING ANNS

摘要

Artificial neural networks (ANNs) are used to address a knowledge base system for Houman Navigation Canal Deepening Project, Louisiana, USA. The concerns of this coastal estuarine system are to limit the hurricane environment enhancement and to determine the salinity front for a lock design. Better management decision requires no data gaps in a continuous salinity record. A knowledge base was constructed by including the freshwater inflow, stage, neighboring salinity, tidal forcing, precipitation, and wind stress as the inputs. The complex data recovery systems are conducted mainly due to missing values in the input system as well. The first level data recovery determines the input missing values while the second level data recovery estimates the salinity missing record. Three recovery approaches, namely self-data recovery, neighboring data recovery, and multivariate data recovery, are conducted. The reliabilities are 52 percent, 71 percent, and 87 percent respectively. The optimal training reliability is about 91 percent without any missing values in the input system. Better recovery reliability of salinity should consider the entire physical system as the driving forcing.