Advancing continuous in-situ quantification of microbial contamination in environmental waters using tryptophan-like fluorescence—Sensor design and validation
5 affiliations
- 1University of Colorado Boulder, 4001 Discovery Drive, Suite N290, Boulder, CO, 80303, USA; Virridy, Boulder, CO, USA
- 2University of Colorado Boulder, 4001 Discovery Drive, Suite N290, Boulder, CO, 80303, USA
- 3Virridy, Boulder, CO, USA
- 4Virridy, Boulder, CO, USA; Department of Ecosystem Science and Sustainability, Colorado State University, Fort Collins, CO, USA
- 5University of Colorado Boulder, 4001 Discovery Drive, Suite N290, Boulder, CO, 80303, USA; Virridy, Boulder, CO, USA. Electronic address: [email protected]
Abstract
Microbial contamination of recreational and source waters poses persistent public health risks, yet conventional monitoring based on laboratory culture methods provides delayed results and limited temporal resolution. These constraints hinder timely identification of short-duration contamination events and effective management of recreation advisories. This study evaluates the performance of a continuous, in-situ tryptophan-like fluorescence (TLF) monitoring system for high-frequency assessment of microbial water quality across laboratory experiments, site-specific field deployments, and a global multi-site modeling framework. Site-specific analyses employed conventional linear regression, while a machine learning approach was used to develop a global model trained across multiple deployments and evaluated using temporal holdouts. Under controlled laboratory conditions, the system reproducibly demonstrated sub-ppb sensitivity that exceeds the stated detection limits of many commercially available TLF instruments. Field deployments in recreational surface waters showed strong agreement between sensor-derived Escherichia coli estimates and laboratory measurements obtained using the industry-standard Colilert method. Approximately 75% of linear, continuous predictions fell within the analytical uncertainty bounds of Colilert with a mean absolute percentage error of 7% in log-transformed concentration space. Binary classification at management-relevant thresholds for a deployment on the Seine in Paris achieved a balanced accuracy of greater than 90% for time-blocked test holdout data. A global model demonstrated a mean absolute percentage error of approximately 19% on temporally held-out test data in log space. Quantitative agreement was strongest at moderate to high concentrations, with increased uncertainty at low concentrations reflecting limitations of both fluorescence sensing and culture-based reference methods. Together, these results demonstrate that continuous TLF-based measurement can complement laboratory monitoring by providing real-time screening and decision support for recreational water management. As additional deployments and training data are incorporated, the global model is expected to further improve, enhancing the scalability and operational value of continuous microbial water quality monitoring.
Related
- Artificial intelligence-enabled monitoring of pesticide contamination in agricultural water systems: a comprehensive review of intelligent sensing technologies for water quality surveillance and environmental risk assessment
- The occurrence of pharmacologically active dyes in the aquatic environment as a result of water and wastewater contamination—an underestimated environmental problem
- Microplastic contamination in commercially important marine fishes from the Sohar Coast, Sea of Oman: occurrence, characterization, and polymer composition
- Contamination Profile and Health Status of Mya arenaria in the Saguenay–St-Lawrence Marine Park
- An overlooked pool of hydrogen stored in humic matter revealed by isotopic exchange: implication for radioactive ³H contamination
BibTeX
@article{knopp2026advancing,
title = {{Advancing continuous in-situ quantification of microbial contamination in environmental waters using tryptophan-like fluorescence—Sensor design and validation}},
author = {Whitney N. Knopp and Joshua T. Klaus and Danny Wilson and Michael Vlah and Matthew R. V. Ross and Evan Thomas},
journal = {Water Research},
year = {2026},
month = may,
doi = {10.1016/j.watres.2026.126099},
url = {https://doi.org/10.1016/j.watres.2026.126099},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - Advancing continuous in-situ quantification of microbial contamination in environmental waters using tryptophan-like fluorescence—Sensor design and validation
AU - Whitney N. Knopp
AU - Joshua T. Klaus
AU - Danny Wilson
AU - Michael Vlah
AU - Matthew R. V. Ross
AU - Evan Thomas
PY - 2026
DA - 2026/05/15
JO - Water Research
DO - 10.1016/j.watres.2026.126099
UR - https://doi.org/10.1016/j.watres.2026.126099
AB - Microbial contamination of recreational and source waters poses persistent public health risks, yet conventional monitoring based on laboratory culture methods provides delayed results and limited temporal resolution. These constraints hinder timely identification of short-duration contamination events and effective management of recreation advisories. This study evaluates the performance of a continuous, in-situ tryptophan-like fluorescence (TLF) monitoring system for high-frequency assessment of microbial water quality across laboratory experiments, site-specific field deployments, and a global multi-site modeling framework. Site-specific analyses employed conventional linear regression, while a machine learning approach was used to develop a global model trained across multiple deployments and evaluated using temporal holdouts. Under controlled laboratory conditions, the system reproducibly demonstrated sub-ppb sensitivity that exceeds the stated detection limits of many commercially available TLF instruments. Field deployments in recreational surface waters showed strong agreement between sensor-derived Escherichia coli estimates and laboratory measurements obtained using the industry-standard Colilert method. Approximately 75% of linear, continuous predictions fell within the analytical uncertainty bounds of Colilert with a mean absolute percentage error of 7% in log-transformed concentration space. Binary classification at management-relevant thresholds for a deployment on the Seine in Paris achieved a balanced accuracy of greater than 90% for time-blocked test holdout data. A global model demonstrated a mean absolute percentage error of approximately 19% on temporally held-out test data in log space. Quantitative agreement was strongest at moderate to high concentrations, with increased uncertainty at low concentrations reflecting limitations of both fluorescence sensing and culture-based reference methods. Together, these results demonstrate that continuous TLF-based measurement can complement laboratory monitoring by providing real-time screening and decision support for recreational water management. As additional deployments and training data are incorporated, the global model is expected to further improve, enhancing the scalability and operational value of continuous microbial water quality monitoring.
SN - 0043-1354
SN - 1879-2448
ER -