Patterns and processes revealed in high-frequency environmental data

High-frequency data are informative but also very challenging to analyze. Appropriate statistical tools are required to extract useful information from such data. A 15-minute resolution sensor generated time series of the EpCO2 from October 2003 to August 2007 in a small order river system in Scotland is used as an illustrative dataset. The aim of this paper is to study the daily patterns and dynamics of EpCO2 using a Functional Data Analysis (FDA) approach. Using FDA, the discrete data within each day have been transformed to a smooth curve; then, a K-means clustering procedure has been applied to the spline coefficients defining the daily curves to identify the common daily patterns which can then be linked to underlying climatological and hydrological conditions.