Long-Term and Seasonal Trends of Wastewater Chemicals in Lake Mead: An Introduction to Time Series Decomposition

Abstract

A recent paper published time series of concentrations of chemicals in drinking water collected from the bottom of Lake Mead, a major American water supply reservoir. Data were compared to water level using only linear regression. This creates an opportunity for students to analyze these data further. This article presents a structured introduction to time series decomposition that compares long-term and seasonal components of a time series of a single chemical (meprobamate) with those of two supporting datasets (reservoir volume and specific conductance). For the chemical data, this must be preceded by estimation of missing datum points. Results show that linear regression analyses of time series data obscure meaningful detail and that specific conductance is the important predictor of seasonal chemical variations. To learn this, students must execute a linear regression, estimate missing data using local regression, decompose time series, and compare time series using cross-correlation. Complete R code for these exercises appears in the supplementary information. This article uses real data and requires that students make and justify key decisions about the analysis. It can be a guided or an individual project. It is scalable to instructor needs and student interests in ways that are identified clearly in this article.

Do the lockdown-imposed changes in a wastewater treatment plant catchment's socio-demographics impact longitudinal temporal trends in psychoactive pharmaceutical use?

“…Since no seasonality was found, locally estimated scatterplot smoothing (LOESS) was used to plot the trend in the PNML of the psychoactive pharmaceutical biomarkers. The effect of different spans (between 0.2 and 0.8) was tested though a sensitivity analysis (Wildman, 2017). A span of 0.25 was the most appropriate for smoothing the trend.…”