Predictive Models 101
Written by Isabelle Oke, Ocean Wise Intern
Artificial Intelligence (AI) is taking the world by storm. You might have even noticed it beside water quality results online. Let’s talk about what that means and what it says about harmful bacteria that might be present in the water.
Photo by Gauravdeep Singh Bansal on Unsplash
First Things First, Let’s Talk Water Quality Results
To determine the water quality results, health authorities go out and gather water samples at specific locations. That water is then tested to determine the levels of harmful bacteria present.
It usually takes at least 24 hours from when the last sample is collected to when water quality results are available for you to see online. Water quality changes like the weather, so this delay in receiving results is not ideal for recreational water users as water quality can change significantly between the time samples are collected and the time you get in the water.
What are AI Predictive Models?
To address these limitations, some agencies have started looking at AI predictive models (AIPMs) for help. AI predictive models are technologies that use historical water quality results to predict future water quality results for public beaches. Essentially, Predictive models are making an educated guess on how much bacteria might currently be in the water based on past results. With a predictive model, recreational water users might not have conclusive results quicker, but they can have access to timely and reliable estimations of what the water quality results may be.
Predictive models are able to make these educated guesses by analyzing past water quality results and comparing these results to environmental conditions during that time. This is actually the same way that you or I might be able to predict that water quality will be worse after a rainfall - predictive models just look at more variables that may affect the water quality on a given day. Besides rainfall, predictive models may look at some of the following variables:
Rainfall
Sewage outpour
Wave conditions
Cloud cover
Time of day
Water Temperature
Water Currents
Wind Direction
And more
Predictive models are most effective when they are custom made for the waters at the beach they’re trying to predict the water quality at. Like snowflakes, each waterbody is different, and how any given variable might affect the amount of bacteria in the water depends on the shape, depth, and surrounding landscape of the beach. Any agencies or organizations interested in using a predictive model must develop one for the particular beach or body of water in question.
Case Study: Parlee Beach National Park
Parlee Beach, 2020
In 2018, New Brunswick’s Office of the Chief Medical Officer of Health commissioned a pilot study and report on the implementation of an AI Predictive Model at Parlee Beach National Park (1). Initial assessments looked at past data of bacteria testing, alongside factors that could impact the way bacteria might arrive at, pass through, and exit the beach.
The following elements were identified as important factors that could coincide with unhealthy levels of bacteria:
Rainfall in the past 24hrs
Wind direction
Water surface elevation of a nearby river
Air and water temperature
Turbidity (degree of cloudiness in the water)
Human presence at the beach in the afternoon
The pilot project incorporated these factors, feeding the AI predictive model daily data based on these four categories to produce a prediction of bacteria levels in the water. This included things such as rainfall in the past 24 hours leading up to an AI predictive model analysis because it would increase the likelihood of FIB bacteria present in the water.
The amount of rainfall has historically impacted bacteria levels in the water. The AI predictive model will use this information to predict what the bacteria levels could be that day. If the forecast suggests an unsafe level of bacteria, then a notice would be posted, and regular testing methods would be used to determine the actual level of bacteria, to update the prediction. This is to ensure accuracy moving forward.
Are Predictive Models Useful?
According to Health Canada, predictive models have the potential to be a helpful addition to beach management programs because they can minimize the negative impact of long wait times for traditional testing methods (2). However, to be reliable, they must be built and implemented in ways that address common challenges with this tool:
Finding the right location to deploy a monitoring system (a site where bacteria levels aren’t too consistent or inconsistent)
Ensuring necessary environmental data can be collected consistently
Ongoing reviews of the system to assess how bacteria forecasts compare to actual bacteria levels
How Reliable are Predictive Modelling Tools?
Predictive modelling projects can vary significantly in structure so there’s no one way to assess the quality of an AI monitoring project. As Health Canada suggests, some aspects of an AI monitoring project must be actively managed to ensure a predictive model is a reliable tool.
The following elements can contribute to the integrity and reliability of an AI predictive model:
A clear indication of the historical data of the site that’s being used
Factors of the site that are used as variables in the predictive model
A data management expert is involved in developing and regularly assessing the predictive model
The water quality information clearly states whether the results were derived from an AI predictive model or traditional testing methods
Examples of Swim Guide Regions where health authorities are using AI predictive models:
Resources about predictive models implemented across Canada & the U.S.:
The United States Environmental Protection Agency published widely used reports that assess AI predictive models and provide guidelines for developing and implementing them.
Systematic review of predictive models of microbial water quality at freshwater recreational beaches
Sources
(1) Evaluation of Predictive Modeling for Parlee Beach. Mas, Diane et al. 2019.
(2) Guidelines for Canadian Recreational Water Quality : Understanding and Managing Risks in Recreational Waters : Guideline Technical Document. Health Canada / Santé Canada, 2023.
