Pixels and Puddles: Decoding, Depicting, and Dashboarding Urban Drainage

BACKGROUND Cities are increasingly reacting to larger storm events with higher intensity, frequency, and rainfall volume that exceed sewer system design capacities. Flood protection systems that were designed years ago are overwhelmed, which additionally limit the performance of the sanitary, storm, and combined sewer systems. A Digital OneWater approach uses data that systems have been collecting to provide insights and foresight into how to operate their OneWater system. Using machine learning and artificial intelligence to predict a system's operation based on forecasted rainfall allows the operators to be prepared ahead of the storm to maximize capture and minimize flooding, both overland and in the collection system. Having insights into what operational controls will minimize this flooding/overflow while also showing areas on a live map that will flood/overflow for an upcoming storm is quite powerful. METHODOLOGY This presentation will highlight a case study in Wilmington, DE. Jacobs began operating the City of Wilmington, DE's wastewater treatment system in 2020, which includes its regional wastewater treatment plant (WWTP), 41 combined sewer overflows (CSOs), three wet weather storage facilities, and three major pump stations. In 2020, only 6 out of 41 CSOs had level and/or flow monitoring. To comply with CSO permit requirements to either monitor or inspect all CSOs to prevent dry weather overflows , a four-person crew personally inspects each CSO three days a week. A smart CSO system upgrade added level sensors at the 35 unmonitored CSOs as well as flow meters on two interceptors. The main objective is to have 'eyes on the system' for monitoring, reporting and faster incident response. The four-person crew can transition to proactive maintenance instead of reactive maintenance, reducing the risk of accidents on the road, mental stress, fuel costs, and wear & tear on vehicles. Another objective is to integrate the data stream with WWTP operations and provide predictive analytics that will integrate operations and better manage the system. Once the remote sensors were in place, the data were sent through secure methods to an online dashboard. The dashboard shows the locations and status of the monitors and sensors. Machine learning is performed on the data to provide maintenance alerts. The alerts are configured so that they are site-specific. Our machine learning determines a 'baseline' level in the sewer. Alerts are then provided for baseline threshold exceedance, pre-overflow threshold exceedance, or actual overflow occurrence. Figure 1 shows an overflow occurring alert. In this instance, we received an 'approaching overflow' alert at CSO-4C at just past 5 pm on Monday, August 26th, 2024, followed quickly by an overflow occurring alert. The crew lead was nearby and was dispatched to clear the blockage about an hour after the overflow occurring alert came in. Historically, the crew would have not checked the site until Wednesday morning. Additionally, the forecasted rainfall can be applied into the calibrated hydrologic & hydraulic (H&H) model to make operational decisions ahead of pending rain events. Rainfall forecast data provider is first integrated into the system, and then spatial analysis is performed based on the radar rainfall forecast. With the forecast data integrated into the platform spatially (Figure 2), the hydraulic model for the collection system can be combined with data-driven models to forecast the predicted flows from both the City of Wilmington and the surrounding areas from New Castle County to give a comprehensive flow forecast for the WWTP influents. In Wilmington, the skeletonized SWMM-based H&H model is on the platform and can be run ahead of the storm event to determine what control setpoints will provide the best capture of the wet weather flow (Figure 3). The solution is developed so that many control setpoint combinations can be run in a short amount of time. The setpoints can be reviewed over time as the forecast progresses. With this solution, the setpoints can be varied for a specific rain event. After the storm, the setpoints can then be compared with the baseline to determine how much more flow can be captured by changing the setpoints (Figure 4). CONCLUSION The case study illustrates how application of a digital data system can proactively identify maintenance issues, reduce CSOs. Existing tools, such as the hydraulic model, can be adapted to solutions that increase public safety in advance of urbanized flood events. These systems can be leveraged to integrate more refined urban flood model results from real time model runs, a library flood map catalog, or machine learning environment. Like the current system, various rainfall forecasts can be run in a short amount of time with locations vulnerable to surface flooding and potential inundation risk depicted in the dashboard. The use of near real-time forecast data in an urban data environment allows communities to develop defensible, integrated frameworks for infrastructure resilience and public safety in a changing environment. These frameworks would include operational procedures for utility leaders, public safety teams, and operations staff that issue hyper-local advance notifications of potential flood locations, visualize potential flood impacts for an oncoming storm event, enable proactive emergency response activities such as placement of traffic barriers, and advance planning of potential storm/flood recovery activities.