Assessing Methodological Variability in Wastewater Surveillance: A Wavelet Decomposition Approach

TL;DR

This study uses wavelet decomposition to differentiate true epidemiological signals from methodological noise in wastewater SARS-CoV-2 data, improving cross-site comparability by focusing on long-term trends.

Contribution

It introduces a wavelet-based method to separate epidemiological signals from methodological variability in wastewater surveillance data.

Findings

Low-frequency components reveal city-specific epidemiological trends.

High-frequency components are dominated by methodological noise.

Wavelet decomposition enhances comparability of wastewater data across sites.

Abstract

Wastewater surveillance has emerged as a critical public health tool, enabling early detection of infectious disease outbreaks and providing timely, population-level insights into community health trends. However, variability in sample collection and processing, for example between wastewater influent and settled solids, can introduce methodological noise that differentially impacts true epidemiological signals and limits cross-site comparability. To address this challenge, we aimed to discern underlying disease trends from methodological variability in SARS-CoV-2 wastewater data using discrete wavelet transform (DWT), with a focus on comparing influent and solids samples from the same geographic locations. We applied DWT to longitudinal SARS-CoV-2 RNA concentrations in wastewater from five California cities, each with paired influent and solids samples. DWT decomposes each signal into two components: (1) approximation coefficients that capture smoothed long-term trends, and (2) detail coefficients that isolate high-frequency fluctuations and transient variations in the signal. We reconstructed signals by progressively removing the high-frequency components and assessed similarity between sample types using hierarchical clustering. Clustering of raw signals did not yield city-specific groupings, indicating that methodological noise obscured the underlying epidemiological signal. Intermediate reconstructions that retained some high-frequency components continued to show mixed groupings. In contrast, reconstructions based solely on low-frequency approximation coefficients revealed clear, city-specific clustering, with influent and solids samples from the same city aligning closely. These findings support our hypothesis that high-frequency components are primarily driven by sample processing and laboratory noise, while low-frequency components reflect shared epidemiological trends.