Degree

Doctor of Philosophy (PhD)

Department

Department of Civil and Environmental Engineering

Document Type

Dissertation

Abstract

Oyster norovirus outbreaks pose risks to public health and the shellfish industry, particularly under changing environmental and climatic conditions. This dissertation develops a satellite assisted norovirus early detection system for improving oyster safety across regional and global scales. Firstly, a satellite-based environmental algorithm was created to derive sea surface gage height (SSGH) using VIIRS data and machine learning techniques, including random forest (RF), XGBoost, and artificial neural networks (ANN). The ANN-based algorithm performed well (R=0.93; RMSE =0.11m) compared to the other algorithms, enabling the generation of spatially distributed SSGH data in data-scare nearshore regions. Secondly, a global spatiotemporal trend analysis was conducted by considering long-term trends in environmental predictors, which revealed geographic shifts in oyster norovirus outbreaks from warmer to cooler regions occurring alongside long-term climatic variability. The findings indicate that key environmental predictors, solar radiation (SR) and SSGH, exhibited temporal patterns that are consistent with norovirus outbreak trends, with region-specific variability across study regions. Thirdly, to translate these findings into predictive capability, a broad scale forecasting model (XGBoost-2day) was developed using environmental predictors form the Gulf of America, British Columbia, the Washington coast, and the southwest coast of France. Initially, RF, deep ANN, and XGBoost were evaluated, with XGBoost demonstrating the best performance. The best model achieved the high predictive accuracy, as characterized by the overall accuracy of 96.33%, positive predictive value of 71%, negative predictive value of 99%, sensitivity of 87.9%, and specificity of 97%, demonstrating the general applicability of the model across distinct geographic regions. Fourthly, a further advancement in the early detection of norovirus outbreaks was achieved through the development of an optimized norovirus forecasting framework (PSO-LightGBM-5day) using particle swarm optimization,  as demonstrated by an overall accuracy of 96.6% and an AUC–ROC value of 0.79 along with a positive predictive value of 74%, negative predictive value of 99%, sensitivity of 88.1%, specificity of 97.3%, and MCC of 78.7%. The PSO-LightGBM-5day model extends the forecasting capability to a five-day lead time. Further, SHAP interpretability identified SR and water temperature as the two most important predictors. Additionally, this work developed a web-based interactive decision support system for the visualization of norovirus outbreak risk, enabling a shift of oyster safety management from reactive monitoring to proactive early warning to improve oyster safety, outbreak preparedness and response.

Date

5-21-2026

Committee Chair

Deng, Zhiqiang

LSU Acknowledgement

1

LSU Accessibility Acknowledgment

1

Download

Available for download on Sunday, May 20, 2029

Share

COinS