Temporal Estimations of Groundwater Pathogen Outbreaks Based on Data-Driven Models

Norovirus is one of the major causes of food poisoning originated from contaminated groundwater and causes acute diarrhea and dehydration. The purpose of this study is to adopt linear and non-linear models for the predictions of norovirus outbreaks.For this purpose, the outbreaks over the next eight weeks are predicted based on the NARX (Nonlinear AutoRegressive eXogenous) and LSTM (Long Short-Term Memory) models,which are known to be effective for time series applications. The data used in the estimation are climate data such as the weekly obtained lowest temperature, maximum temperature, precipitation, relative humidity, and atmospheric pressure from 2009 to 2017 in Korea, and the corresponding Norovirus outbreak data.As a reference model, ARX (AutoRegressive eXogenous) model is comparatively applied to the same data. From the estimations, the applied models commonly show good estimation results, where the results of the NARX and LSTM models are superior to that of the ARX model in terms of accuracy. Sensitivity analysis of the trained model shows that atmospheric pressure is relatively insignificant compared to other explanatory variables. In addition, all explanatory variables are generally less sensitive in summer than in winter. The results of this study is expected to provide reference estimates for groundwater quality management. Also, the applied model can be used as a basis for an early warning system for disease prevention.