Rezatec’s Geospatial AI: Predicting Risk of Failure to Drive Proactive Water Network Management

Municipal Water Leader: Please tell us about your background and how you came to be in your current position with Rezatec.

Camilla Braithwaite: I’ve been working in software for over 10 years. Before I started here, I worked with Diligent Corporation and eShare, helping legal clients to deal with the burden of regulation and compliance. This involved encouraging people to use software instead of paper and getting them comfortable with security, using technology, and changing the way they worked.

I joined Rezatec in 2020. My goal is to help the company understand how to use software with data and to take advantage of its huge observation and geospatial data capabilities. I’ve been helping Rezatec to package its capabilities to make them usable for customers. The original objective was to take the things we were doing for individual customers and make them into something that was valuable for a whole industry. We did extensive market research to understand the needs of the industries we were working in. Water was one area that had significant challenges and was therefore ripe for transformation. It was slowly changing, but it was still far behind other industries.

Fast forward to 2024, and our Pipeline Risk product is proven to reliably predict failure in buried potable and sewer pipeline networks at 80 percent accuracy. It’s powerful stuff. We’re identifying 80 percent of risk in the highest 20 percent of the network, helping our customers to proactively get ahead of deformation in buried assets and upgrade them before failure occurs.

Municipal Water Leader: Please introduce Rezatec.

Camilla Braithwaite: We’re all about putting complex geospatial insights into the hands of water leaders so that they can more easily manage risk, prioritize resources, and maximize the value of their assets. Our Pipeline Risk product identifies the most critical, at-risk sections of networks and predicts the likelihood, consequence, and cost of failure so that customers are able to proactively prioritize upgrades, plan and optimize investment, and reduce water loss.

Rezatec started life in the forestry markets, analyzing inventory and carbon outputs for some of North America’s largest forestry companies. We quickly realized that water leaders had big challenges we could solve, especially monitoring the integrity of large assets such as dams and buried pipeline networks, where the change over time in factors contributing to deformation are difficult to see and predict.

It took the best part of 10 years to develop and refine our processing methodologies, machine learning, and AI algorithms. In fact, we’re still refining them now—I think that’s a continual process for us. We combine massive sets of satellite and other geospatial data with our customers’ data. Our customers are now at the point at which they are comfortable with the technology, trust that it’s accurate, and see the value in using new insights they’ve never before had access to.

Municipal Water Leader: Please tell us more about how your geospatial AI platform works.

Camilla Braithwaite: The Rezatec geospatial AI platform fuses a huge number of data sets with AI to produce insights and predictions. The platform powers products that we’ve specially tailored for each industry, such as dam operators and water utility leaders.

Geospatial data include location information, attribute information, and sometimes temporal information—information on things that change over time. AI or machine learning is about taking those large data sets and using what has happened in the past to predict the future.

We get our geospatial data from various places. We feed our model with sets of data on things such as soil pH and slope. Ground motion and vegetation data also give you an indication of what’s happening below the ground. We get this data from satellites: the European Space Agency satellites provide a massive archive. We access those data for our customers and derive information about their assets from them.

The geospatial AI platform incorporates customer information. For example, water utility customers provide data on pipeline attributes, typically including break history, pipe material, age, and any other GIS data that are being collected. This blend of multiple big data sets is the key to driving certainty.

Our patented machine learning and AI algorithms have been carefully developed over years of working with some of the largest—and smallest—water networks. AI seems to be everywhere today. At Rezatec, we’re keen to uphold our mission around accuracy and trust. That’s why we consistently ground-truth our insights and predictions and test their precision in the field with our customers.

The combination of satellite, geospatial, and customer data fused with our AI models is turning out to be really powerful, and it’s what sets us apart. Our water utility customers are excited to see new insights and predictions at accuracy levels they’ve never been able to achieve until now.

Municipal Water Leader: What sorts of data do the satellites provide?

Camilla Braithwaite: There are two pairs of satellites. One pair gathers radar data. Radar is great because it’s not visual, so it works at night and in cloudy weather. We use that information to measure ground motion. The satellites with the radar sensor orbit every 6–12 days, which means you can compare one phase wave to another to detect small amounts of movement. Looking at the information over time, you can see trends and seasonal changes that help you understand how the ground is moving around the pipes.

The other pair of satellites has a sensor that collects visual multispectral data, including infrared and near-infrared, which we use to monitor vegetation. Looking at that information over time allows you to see the trends and seasonal changes in the greenness of the vegetation and the moisture level in the leaves and the top layer of the soil.

Municipal Water Leader: How do you monitor underground assets such as pipelines?

Camilla Braithwaite: We spent a lot of time looking for a way to measure water coming off underground pipelines, but didn’t come up with a way to do that robustly across any urban or rural terrain. That’s how we ended up developing the AI model. If you train a model on historic failure, it can combine the data on pipe material, diameter, length, soil type, soil pH, slope, ground motion, and vegetation to identify the signature of failure for that network area and then use that to predict where failure might occur next. We use 3 years’ worth of data, which is enough for the model to learn these patterns. The machine learning model processes more than 100 attributes and uses that information to predict the likelihood of failure of each section of pipe in the network.

Municipal Water Leader: What kind of analysis or results does the AI model provide, and how is that delivered to your users?

Camilla Braithwaite: We rank the likelihood of failure on a scale of 0 to 5. We use that number to provide a relative likelihood of failure across the whole network. Customers typically request the percentage of their system with the highest likelihood of failure that corresponds to their available budget. For example, customers with large networks tend to look at the 2–5 percent of their system with the highest likelihood of failure, whereas those with smaller networks want to know the top 10 or 20 percent. The likelihood of failure is also used to focus leak detection for repair teams to help them find the pipes that are most likely to fail.

Our customers access these insights and predictions through Rezatec’s secure online geospatial AI platform. It’s straightforward to use, and our customer success team is on hand to help customers interpret data when needed.

Municipal Water Leader: How can a customer use that information to prioritize repairs?

Camilla Braithwaite: Water leaders consider not just the likelihood of failure in many different areas but the relative consequences of failure. Am I going to fix a pipe that’s on the outskirts of my service area and only supplies a small number of people, or am I going to fix a pipe that is feeding a large city and a hospital? Pipeline Risk highlights the most important facilities to maintain service for, such as hospitals and schools. That’s not based on a machine learning model but a weighted model, which is simpler and easier to run. Often, our customers have a list of critical customers they want to prioritize as well. Or they might want more in-depth monitoring of a particular part of their service area. For example, we have a customer that had a sewage spill in an oyster bed and does not want that to happen again. Other customers prioritize the protection of farms and freshwater resources.

Our solution also helps our customers consider flows of water. If a pipe has chlorine or sewage in it and could potentially pollute, they want to know how much water would flow from that pipe if it were to fail. In this case, too, the location of the pipe determines the consequence of failure. For example, the slope of the land in one area might mean that water will rush down to a particular place.

There are two sides to the equation: the likelihood and the consequence of failure. Putting those together gives you a risk factor across your whole network so that you can start zeroing in on the areas that are important.

Municipal Water Leader: Who are your U.S. water utility customers?

Camilla Braithwaite: We have customers all over the United States, Canada, Europe, and Australasia. They include public and private utilities. We have many water utility customers, including the City of Eden Prairie, Minnesota; the City of Greely; the City of Guelph, Ontario; Epcor Utilities; SouthWest Water Company; and WaterOne, a utility in Kansas. They all have something in common: a desire to shift their reactive-style operations to a proactive prediction of risks. Our customers all see the benefit of getting ahead of failure before it occurs to drive better capital upgrade planning; avoid reactive, short-term repairs and the associated costs; and reduce water loss.

For example, the utility leadership team at the City of Eden Prairie in Minnesota is following an invest now, spend less later strategy for its 64,000 customers. Transitioning to proactive operations sits at the heart of this strategy. The team would prefer to replace a pipe at the end of its projected useful life rather than waiting until it pops a leak in the middle of a Saturday night, disrupts service, and costs three times as much to fix. The challenge is that the city is located in a glacial moraine environment with a variety of subsurface soil conditions that can significantly affect the maintenance of buried metallic pipeline systems. The occurrence of internal corrosion pits is random, and until now, has been difficult to predict. The team uses Pipeline Risk to predict the risk of failure, maintain its 398 miles of water distribution system, and ensure that its capital improvement dollars are used the best way they can be.

Municipal Water Leader: Please tell us about how you incorporate customer feedback into your products.

Camilla Braithwaite: That’s the part of my job I love the most: talking to customers. The latest release of our geospatial AI platform incorporates significant enhancements, which we’re planning to make available to our Pipeline Risk product in the coming months. These feature improvements have all been built from our customers’ feedback on their challenges.

One example of those challenges is demonstrating the accuracy of our predictions so that utility leaders can trust that their upgrade planning is right and can confidently justify investment to their wider leadership. Our product development team has built in additional data review and model rerun processes for each customer, and we’re working with them to ground-truth results. And, of course, the beauty of machine learning is that the more data are fed into the model, the more accurate it becomes. The team at WaterOne in Kansas is facing a nearly $50 million upgrade bill over the next 10 years for its 2,800 miles of network across 17 cities. WaterOne needs to add next-level precision and reliability to its water main replacement prioritization. We worked with the team to refine our predictions across an initial 250 miles, achieving a proven accuracy of 78 percent, and now we’re rolling out across a further 1,100 miles, which is a testament to the confidence they have in our product.

Municipal Water Leader: What is your vision for the future?

Camilla Braithwaite: Rezatec’s vision to is take water leaders into the future and give them the tools to face the growing risks to the integrity of their assets caused by climate change, aging infrastructure, and stretched resources. Using intelligence, data insights, and software to help water utility leaders mitigate these evolving risks is what our vision is all about.

Original article from Municipal Water Leader Magazine can be found at https://municipalwaterleader.com/volume-11-issue-3-march/