Brussels in real time: how ICM Live powers the city’s flood warning system

Abe explained how Autodesk is supporting this transition by integrating planning, modelling and operational tools across the water cycle. The goal, he said, is to enable users to move fluidly from design to simulation to live forecasting within a unified environment. His framing provided a strategic prelude to the case study that would follow: Brussels not as an isolated example, but as a representative of the direction many cities must now take.

Flood modelling in a city defined by centuries of water management

The core of the webinar belonged to hydrologic engineer Michael Antoine, who started introducing the audience to Brussels’ hydrological history and its present-day challenges. “Brussels, historically, is a city that was constructed on a marsh,” he said as he displayed maps showing the Zenne River’s transformation. Over centuries, waterways were progressively culverted and buried; although the river network still exists, most of it now flows invisibly below streets and buildings. Topography varies from 130 metres in the uplands to 10 metres near the tidal influence.

“Digital transformation is enabling a more precise management of the resources within organizations, and this heavily impacts the productivity” –  Abe Feenstra

Urbanization has intensified these constraints. Impermeable surfaces have doubled in the last half-century, while rainfall intensity has increased by about ten percent. The hydraulic system must therefore accommodate more runoff without having gained new conveyance space. As Michael explained, “all the water still needs to go in the same pipes inside the city centre,” a condition illustrated by two decades of fire brigade interventions plotted against flood maps.

Brussels’ flood management strategy is built around three pillars: prevention, supported by detailed fluvial and pluvial flood maps and strengthened by efforts to raise public awareness; protection, through storm basins and sustainable drainage systems that create space for water; and preparedness, driven by real-time forecasting, early alerts, and the activation of safeguard measures ahead of flooding. The development of a digital twin capable of supporting all three pillars was thus central to Bruxelles Environnement’s long-term planning.

Inside Brussels’ real-time flood forecasting system

The heart of the session was Michael’s detailed description of how the city built its operational ICM Live system. The project benefited from a rich offline modelling base: a calibrated hydrological model for one hundred sub-catchments, extensive field-surveyed cross-sections, full culvert inventories, and a one-metre-resolution LiDAR dataset. Together, these provided the foundation for generating flood hazard maps and simulating river, sewer and surface runoff behaviour.

Michael explained how these models were adapted for live forecasting. The system continuously ingests rainfall observations, radar data, nowcasting products and numerical weather predictions. Telemetry from local gauges and upstream partners is automatically incorporated as boundary conditions. Python scripts fetch, convert and preprocess data, ensuring consistency in units, file names and temporal structure. “All that information from the past to the future needs to be ingested in your model,” he said, emphasizing that seamless data acquisition is fundamental in real-time operations.

“Automated alerts ensure we don’t miss critical changes, especially when operators cannot continuously monitor every part of the system” –  Michael Antoine

Equally important is handling missing or inconsistent data. Michael described the fallback and blending rules that ensure the model runs even when gauges fail or forecasts abort. Observations take priority where available, radar fills spatial gaps, and numerical predictions extend the forecast horizon. This layered architecture enables the system to maintain continuity across weather conditions and data disruptions.

Performance optimization was another major step. Brussels’ full hydraulic model is highly detailed, but live computation requires speed. “What we did, we removed certain compartments… we simplified some structures also, just so that it runs fast,” Michael explained. With these adjustments, the model now runs in about five minutes for the entire city, allowing forecasts to be updated every four hours.

Alerts form the operational output. The system detects threshold exceedances in river levels, sewer overflow discharges, rainfall return periods and water quality indicators such as dissolved oxygen. Email notifications are automatically sent to operators, with plans to incorporate charts directly into alert messages to help interpret conditions.

Michael noted that uncertainties remain. Forecast rainfall remains the largest source of error, particularly when nowcasting evolves faster than the four-hour model cycle. False alarms sometimes occur near threshold boundaries, but the use of validation rules on telemetry reduces spurious triggers. He emphasised that one of the strongest advantages of ICM Live is that it functions as an all-in-one solution, allowing Brussels to maintain both offline and live models within the same environment. Because the platform is easy to update, the team can keep the system aligned with evolving conditions. Michael concluded that such a system is essential in a context of rising flood risk, enabling Brussels to anticipate events rather than simply react, and to deliver earlier, more informed decisions during critical situations.

What the Q&A revealed

The Q&A session highlighted both the technical sophistication of the Brussels system and the practical considerations involved in operating it. One of the topics explored was the configuration of the 2D hydraulic mesh. Michael explained that the model uses grid elements up to 400 square metres, a size consistent with previous flood mapping for the region, while still adapting automatically to local topographic detail. Questions then turned to the system’s interaction with neighbouring forecasting tools. Michael clarified that while Flanders uses a different platform for its own flood forecasting, Brussels retrieves the upstream predictions through web services, incorporating them directly into ICM Live as boundary conditions.

As digital tools continue to advance, operational digital twins will play an increasingly important role in supporting decisions across the full water cycle

Another line of inquiry focused on how runoff coefficients are determined in such a heterogeneous urban environment. Michael explained that wherever possible, coefficients are calibrated using local gauge data. However, in parts of the city where no direct measurements can be made — particularly for surface runoff along streets — they rely on urbanization rates, assigning higher coefficients to heavily waterproofed areas. Model accuracy and operational reliability were also addressed. Michael noted that the hydrological and hydraulic calibration performs well for large events, but that rainfall uncertainty, especially in fast-changing nowcasting conditions, remains the greatest challenge. Telemetry validation rules and threshold buffers help limit false alarms, though some borderline cases still occur.

The session closed with a look ahead. Both speakers reflected on the role of real-time models as cities face intensifying climate pressures. Michael emphasised that the system’s value lies in its ability to bring together multiple data sources in one place and provide a clearer sense of “what-if” conditions, helping operators act before flooding occurs. Abe added that as digital tools continue to advance, operational digital twins will play an increasingly important role in supporting decisions across the full water cycle. Their final remarks underscored the growing importance of integrated, data-driven approaches for cities seeking to strengthen their resilience in the years to come.