Conference Agenda

Machine learning has been transforming many fields, including water resource management, where it has the potential to revolutionize how we collect, process, and analyze data to inform decisions. However, the field of machine learning can seem intimidating and out of reach for those without programming or data science backgrounds. In this presentation, we aim to demystify machine learning and encourage broader utilization of these tools by water resource professionals.

We will provide a brief overview of machine learning concepts and applications relevant to water resource management. We will also discuss recent developments in automated machine learning tools, which make it easier for people without coding experience to apply machine learning techniques to their data.

We will showcase an example of how machine learning has been used in CWS to generate effluent temperature predictions in the first clarifier process. In addition, we will highlight publicly available machine learning resources, including software packages and online courses, that are useful resources for learning and applying these techniques within the environmental engineering field.

Finally, we will discuss the future of machine learning in the water resource and utility sectors, highlighting the potential for improved accuracy and efficiency in decision-making, cost savings, and increased resilience to climate change and other stressors.

By the end of this presentation, attendees will have a better understanding of what machine learning is, its potential benefits for water resource management, and how they can start learning and applying these techniques to their own data.

High bacterial levels in ambient surface waters or stormwater can be a vexing problem to solve, as the traditional fecal indicators (typically E. coli or Enterococcus) give no indication of their source. Accordingly, many agencies have turned to PCR-based methods to determine the source of fecal pollution in various water matrices, a technique known as microbial source tracking (MST). The Sketa assay has been used in both research and regulatory contexts to quantify nucleic acid extraction recovery and/or environmental matrix inhibition in quantitative PCR water quality studies. The field of MST is moving to digital PCR, in which the variability from sample concentration and nucleic acid extraction exceeds the variability introduced from inhibition. Thus, a total workflow process control with an appropriate surrogate target is a more suitable approach to data quality assurance for digital PCR. Here we present a duplex Surrogate Process Control (SPC) assay for droplet digital PCR using a commercial spike-in whole-cell product at a cost of pennies per sample. This SPC measures the recovery of both gram-positive and gram-negative bacteria, which is important in contexts where both are measured for MST, e.g. Bacteroides and Enterococcus. This SPC was optimized, validated, then compared to the Sketa assay in seawater (n=5), freshwater (n=5), stormwater (n=5) and municipal wastewater influent (n=5) on the basis of percent recovery and percent inhibition. Sketa recovery (when added at the extraction step) ranged from 71- 139 percent for all samples. While inhibition was low (0% samples inhibited), the total SPC recovery varied greatly depending on environmental matrix. Average gram-positive percent recoveries were 22, 12, 9, and 1.6 for freshwater, marine beaches, stormwater, and wastewater, respectively. Average gram-negative percent recoveries were 11, 7, 4.4, and 0.4 for freshwater, marine beaches, stormwater, and wastewater, respectively. Measuring inhibition alone failed to identify samples that lost greater than one-log of material through sample concentration, especially in more challenging environmental matrices. Overall, this SPC is a robust and streamlined approach to quality control for digital PCR MST assays.