Woolpert Implements FLOODTRACZ Machine Learning Modeling Solution for Mecklenburg County

The newly developed machine learning solution will help improve the county’s flood forecasting efforts.

Mecklenburg County, North Carolina, like the broader southeastern U.S., is no stranger to major storms and flooding events. While the county’s multifaceted flood management approach has been effective, officials felt it was missing a key capability: real-time flood forecasting. Woolpert stepped in to help the county fill this gap with FLOODTRACZ, its proprietary machine learning (ML) flood prediction model.

12-hour

flood prediction window with a 15-minute stage interval

110

ML models developed using stream gauge observations from 51 USGS, 59 low-cost sensors and 72 USGS rain gauges

Unlimited

predictions generated using current gauge observations, soil moisture conditions, and NOAA rainfall forecasts

Real-Time Flood Forecasting: The Last Piece of Mecklenburg County’s Flood Management Puzzle

Mecklenburg County has experienced numerous significant flooding events over the years, prompting the county to take a proactive approach to flood management.

Charlotte-Mecklenburg Storm Water Services (CMSWS) has installed an extensive network of stream gauges to monitor real-time conditions. After flood events, officials review the data to reconstruct peak flows and determine the extent of flooding. This analysis helps them compare modeled flooding with resident reports and validate conditions on the ground.

Digital view of the RARR web application during a November 2020 storm event.

CMSWS also developed a data-driven framework and set of tools called Risk Assessment and Risk Reduction (RARR), which dynamically assesses, evaluates, and prioritizes mitigation strategies at the individual building level. Every property within the regulated floodplain is scored and prioritized, amounting to around 5,000 properties.

Additionally, Charlotte-Mecklenburg operates one of the most successful and longstanding floodplain buyout programs in the U.S. Since 1999, over 490 flood-prone homes and buildings have been purchased and removed from floodplains, resulting in more than $50 million in historical flood losses avoided and potentially hundreds of millions of dollars in future damages prevented.

Despite the strength of this multifaceted flood management approach, one major piece was still missing: predictive forecasting. Woolpert stepped in to provide that final component through the implementation of FLOODTRACZ.

Mecklenburg County’s Search for the Ideal Solution

Mecklenburg County wanted to improve how it monitored and responded to potential flooding, especially within short-term windows. More specifically, the county sought 12-hour predictive insight into which areas were most at risk so it could better prepare for emergency actions such as road closures and evacuations.

While the CMSWS team already mapped inundation zones to identify where flooding was likely to occur, they needed a system capable of forecasting potential flooding rather than simply reporting current gauge levels. CMSWS also wanted to eliminate the manual processes of interpreting current stream gauge readings and weather forecasts to determine what might happen next.

Mecklenburg County spent several years evaluating what it needed in a flood forecasting solution and ultimately chose to partner with Woolpert. The firm was already familiar with many CMSWS team members due to its long history of work in the region, dating back to 1994 when it completed its first watershed study for Charlotte. This experience positioned Woolpert as the ideal partner to help the county advance its flood management program with FLOODTRACZ.

What is FLOODTRACZ, and What Sets it Apart?

FLOODTRACZ is a data‑driven, ML-based flood prediction tool designed to enhance existing flood management infrastructure, such as Mecklenburg County’s network of stream gauges. ML models are developed using historical stream gauge and rainfall observation records, incorporating corresponding datasets from the National Oceanic and Atmospheric Administration to explore location‑specific relationships. The trained models are then used to generate operational flood predictions, including stream stage elevations at each gauge and associated flood inundation extents.

FLOODTRACZ evaluates a suite of candidate ML models across multiple ML algorithms, model structures, and training configurations. It selects the model that best fits historical observations and produces predictions that align with recent and observed watershed conditions at that gauge. Once the selected model is trained, validated, and finalized, it serves as the operational prediction engine for that gauge.

Unlike traditional hydrologic and hydraulic models that require long run times, FLOODTRACZ offers a faster, less resource intensive alternative, providing ample lead time for decision-makers. These timely, reliable insights enable communities to respond quickly and appropriately. By forecasting how high county creeks will rise during storm events, FLOODTRACZ helps emergency managers allocate resources effectively and communicate clear evacuation protocols.

Woolpert Overcomes Data Challenges During FLOODTRACZ Implementation

Woolpert developed ML models for 110 forecast locations across Mecklenburg County.

The implementation of FLOODTRACZ was not without several data-related challenges that required the Woolpert team to develop creative solutions.

For starters, Mecklenburg County relied entirely on its own data rather than data housed on Woolpert’s internal systems. This configuration meant the Woolpert team needed to integrate with third-party data systems. The county’s stream gauge network includes both U.S. Geological Survey gauges and lower cost county owned gauges, the latter of which can be less reliable. Woolpert had to build additional filtering and error handling processes to manage missing or inconsistent data.

Real-time forecasting requires large amounts of observational data, and repetitive dataset calls can overload public API services. To avoid overwhelming data infrastructure, Woolpert created and maintained a synchronized mirror of the requisite database and stored it locally. By optimizing the calling of the API, the team significantly reduced the potential load on the server and improved workflow stability.

Another challenge arose from the original plan to write all FLOODTRACZ output back into the county’s third-party database. The system was not designed to handle time-series prediction data, so Woolpert had to adjust the workflow accordingly and provide a custom API so that the county can access the predicted results.

Despite these hurdles, the implementation was successful, and Mecklenburg County now has a fully operational forecasting tool powered by FLOODTRACZ, elevating its existing stormwater program. Moreover, this implementation proved valuable for the Woolpert team, as it prepared them for future deployments in similarly complex data environments.

The Value FLOODTRACZ Provides Mecklenburg County and Other Communities

FLOODTRACZ for CMSWS is now in a maintenance phase, with Woolpert routinely checking the model to ensure it continues running smoothly. Woolpert will also meet with the county annually to review performance and identify potential refinements and enhancements to the overall system.

FLOODTRACZ will serve as an integrated external solution alongside the county’s extensive network of stream and rain gauges, enabling the county to provide residents and local emergency personnel with advanced notifications of possible flooding conditions.

The primary value of FLOODTRACZ is that it helps keep people out of harm’s way. Incidents like the Camp Mystic flood highlight how the absence of an effective warning system can have tragic consequences. With FLOODTRACZ, Mecklenburg County now has a 12-hour prediction window. Already, the county has reported clear value in this improved lead time.

Emergency management teams can act proactively rather than reactively during potential flood events, giving the county greater confidence when making decisions about deploying resources, planning evacuations, or closing roads. Additionally, the system is highly customizable, making it suitable for counties with dense gauge networks like Mecklenburg, as well as for those that only need forecasts for a few key rivers.

Notice how closely the forecasted stage aligns with the observed stage in both instances across two historical storm events in Mecklenburg County. This advanced modeling is incredibly helpful for emergency services as they work to keep their communities safe.

Woolpert's FLOODTRACZ approach has provided our communities with a powerful tool to transform historical flood records into actionable flood forecast intelligence. The machine learning process offers the ability to account for dynamic environmental variables and to continually improve future forecasts as more data is collected. We are very excited to navigate the future of flood warning alongside Woolpert as we strive to increase flood-threat readiness and reduce response time to help safeguard Mecklenburg County’s residents from risks to life, property, and critical infrastructure.

Mathew C. Hornack, PE, CFM, project manager at Mecklenburg County, Charlotte, North Carolina, United States

Discover How FLOODTRACZ Can Make Your Community Safer

ML modeling solutions like FLOODTRACZ are expected to become a ubiquitous part of flood studies within the decade — especially as storms continue to increase in frequency and severity across the U.S.

Connect with a Woolpert representative today to learn how your community can enhance its flood mitigation and disaster preparedness strategies with our proprietary, ML-based platform.