The United States Environmental Protection Agency (EPA) estimates that 10 trillion gallons of untreated stormwater runoff containing raw sewage, trash, and toxins enters US waterways each year from city sewer systems, polluting the environment and drinking water supplies. Urban runoff also exacerbates flood risk. As climate pressures intensify, cities and towns need to enhance their stormwater infrastructure, addressing the growing risk of flooding as well as pollution.

Arup is collaborating closely with The Nature Conservancy (TNC) to launch the Smart Watershed Network Management project. The pilot, led by TNC’s Brightstorm program, integrates artificial intelligence (AI) and machine learning to enhance stormwater systems in urban areas and address pollution and contamination from stormwater runoff in Florida’s Indian River Lagoon, a 156-mile-long coastal lagoon that is one of the most biologically rich estuaries in North America and home to wildlife such as manatees and dolphins. Arup is leading the project’s digital systems, which includes remote sensing modeling, data analytics, and applications of AI and machine learning.  

The Florida Department of Environmental Protection has identified pollutants from runoff as a critical threat to the health of the Indian River Lagoon and the life and livelihoods it sustains. However, rebuilding and upgrading stormwater can be both costly and time intensive. The Smart Watershed project leverages digital solutions to optimize the existing infrastructure’s ability to handle storms and reduce pollution. These technologies enable more efficient, cost-effective interventions that deliver measurable benefits for water quality, ecosystem health, the local economy, and the resilience of surrounding communities. 

Diagram of coastal community.

The Smart Watershed project leverages digital solutions to optimize the existing infrastructure’s ability to handle storms and reduce pollution. The illustration reflects the project's benefits for water quality, ecosystem health, and the resilience of surrounding communities © The Nature Conservancy 

Pioneering scalable, technology-driven stormwater solutions

Arup is working to develop and mainstream solutions that integrate technology into existing stormwater infrastructure. Arup’s hydrologic modelers assess water behavior across existing stormwater infrastructure and apply AI and machine learning to analyze ways the infrastructure can be upgraded to more effectively reduce pollution during storm events that degrade sensitive coastal ecosystems. These smart systems are designed to build an understanding of the watershed, be able to interpret how a storm event will impact it and make live adjustments to stormwater systems such as ponds, cisterns, and pipe networks based on weather predictions. For example, ahead of a major storm, the system can evaluate the state of various ponds and pools and proactively lower water levels in areas where rainfall is expected to be the most intense. This creates additional capacity to capture incoming runoff and reduces flood risk and pollution.

The goal is to create an adaptive watershed management system that continuously learns from each storm, optimizes performance across upgraded infrastructure, and identifies where additional ponds, pipes, or controls could be brought into the network next. To achieve this, Arup is developing and testing new stormwater management models in real-world situations, monitoring results to gradually scale up systems across the Indian River Lagoon. In parallel, the project is establishing the groundwork for widespread technology implementation for live adjustments to stormwater systems based on weather predictions by developing open-source software, tools, and AI and machine learning approaches. Designed to be compatible with any sensor, IoT device, or control technology, these solutions can eventually be scaled up in other regions across the US and globally, enabling cities and organizations worldwide to adopt our approaches.  

Integrating AI to enhance infrastructure performance

AI is evolving from simply predicting an outcome to building a representation of a complex environment, understanding relationships within it, and reasoning about the consequences of different actions over time. Emerging approaches such as graph neural networks (GNNs), spatio-temporal models, and “world models” are particularly relevant to infrastructure, where changes to one asset can affect conditions elsewhere, often with a delay. Arup is applying these ideas to stormwater management by representing a watershed as a connected graph of ponds, pipes, and outlets, and combining that physical network with forecasts of rainfall, inflows, water levels, storage, and flows. The AI can therefore reason across both space and time, enabling the system to understand, for example, how releasing water from an upstream pond may affect capacity downstream several hours later and develop a coordinated operating strategy for the network rather than optimizing each asset independently. The underlying approach uses graph-based and spatio-temporal representations specifically to capture these relationships. 

This work also reflects the growing use of simulation as a bridge between AI and the physical world. Rather than allowing an AI system to learn through trial and error on live infrastructure, which could introduce liability and public safety risks, Arup has developed a digital environment in which it can test thousands of possible operating strategies and learn from their consequences before recommendations are applied to physical controls. In effect, this provides a “world-model-style” capability where the system can ask “what happens if we operate the network this way?”, simulate the resulting water levels and flows over the coming storm, and improve its strategy based on the outcome. The system then uses reinforcement learning to develop control trajectories for physical assets such as determining when and how far pond outlets should open to create capacity ahead of rainfall, while avoiding unnecessary drawdown and coordinating releases across upstream and downstream assets. 

The project’s utilization of AI brings a new way of getting more performance from existing infrastructure through combining forecasts, hydraulic understanding, and intelligent control so that distributed assets can increasingly operate as a coordinated, adaptive system. This digital-first approach allows for highly efficient reports on millions of operational parameters, providing regulators, utility operators, and other public sector stakeholders with detailed evidence of how the system performs under different conditions and breaking down legal and operational barriers to smart watershed adoption.