Application of time series analysis on temporal variation of fluoride in groundwater around Southern Taiwan Science Park

This paper demonstrated a case study on how to utilize time series analysis as mining tool to track the transience of fluoride release and to predict the fluoride concentration in groundwater. Southern Taiwan Science Park was selected as study area, and seven groundwater monitoring wells located in the domain of fluoride release were subjected to time series analysis. The measured fluoride levels in groundwater are between 0.4 and 3.6 mg/L, and the series data is stationary in mean and variance during the period of 2005 and 2009. Time series analysis is a useful tool for extracting interesting pattern from ordered sequence of observations. Based on extracting information from ACF and PACF, the trend, interesting patterns, rules, or models within the original data set were filtered out. The common time series models, ARMA and ARIMA, were employed to interpret the information beneath the monitoring data of groundwater quality. Through verification by Akaike's information criterion (AIC) and Schwartz's Bayesian Information Criterion (SBC), the ARMA(1,1) model was identified as the best fitted model for data interpretation and estimation. Accordingly, this developed numerical model can effectively interpret and forecast the fluoride level in groundwater as referring the prior information.