Development of Remote Sensing Assisted Water Quality Nowcasting and Forecasting Models for Coastal Beaches

A remote sensing assisted water quality modeling framework is developed in this dissertation for nowcasting and forecasting recreational water quality of Holly Beach in Louisiana, USA. The modeling framework is composed of four models/systems: (1) an Artificial Neural Network (ANN) model (Model 1) and an US EPA Virtual Beach (VB) Program-based model for predicting early morning enterococci (ENT) levels in beach waters; (2) an ANN model (Model 2) and an VB model for predicting early morning Fecal Coliform (FC) levels in beach waters; (3) a remote sensing assisted modeling system (Model 3) for predicting near real time ENT levels during daytime; and (4) a hybrid probabilistic/deterministic modeling approach (Model 4) for predicting the probability of beach water quality violation. New findings from Model 1 include (1) the identification of 7 explanatory variables and various combinations of the 7 variables responsible for the ENT level in coastal beach waters; and (2) Model 1 with Linear Correlation Coefficient (LCC) of 0.857 performs consistently better than the VB model with LCC of 0.320. A major finding from Model 2 is that a total of 6 independent environmental variables along with 8 different combinations are capable of explaining about 76% of variation in FC levels for model training data and 44% for independent data. Major new contributions made in Model 3 include (1) development of remote sensing algorithms for turbidity using Terra and Aqua satellite data; (2) development of an enhanced ANN model for predicting ENT levels at sunrise time by taking into account the cumulative effect of solar radiation on ENT inactivation; (3) development of a real-time model for predicting ENT level during the daytime by considering the turbidity effect on ENT inactivation. A novel feature of Model 4 (hybrid model) is the combination of advantages of a deterministic ANN model and a probabilistic Bayesian model. The hybrid model is capable of reproducing 86.25% of historical beach water quality advisories with 6.39% of false positive predictions and 7.36% of false negative predictions over the past 7-years. Applications of the models will improve the management of recreational beaches and the protection of public health.

Name thesis:degree_name

Doctor of Philosophy (PhD)

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Civil and Environmental Engineering

Grantor

Civil and Environmental Engineering

Year dc:date.available

2014

Author dc:creator

• Zhang, Zaihong

• Nowcasting and Forecasting models
• Predictive models
• Fecal indicator bacteria level
• Coastal beaches
• Hybrid model

Statement dc:rights

• restricted
• Student has submitted appropriate documentation to restrict access to LSU for 365 days after which the document will be released for worldwide access.

Identifier

etd-08132014-112841
https://repository.lsu.edu/gradschool_dissertations/666

OAI identifier oai:identifier

oai:repository.lsu.edu:gradschool_dissertations-1665