The weather—and by extension, flooding—is notoriously unpredictable. When conditions are worse than predicted, floods can have devastating consequences. If conditions are better than anticipated, communities may waste valuable resources.  

While preventative measures like floodplain mapping can help communities reduce flood risks, decision-makers must also take appropriate action in the days leading up to and during these events, which can be difficult without access to accurate and timely data.  

The quest for a truly “real-time” flood forecasting solution is nothing new. However, major strides have been made recently, thanks to the advent of predictive analytics, machine learning (ML), and artificial intelligence (AI).  

The Historic South Carolina Floods of 2015: The Origins of FLOODTRACZ 
In October 2015, a combination of meteorological factors resulted in record rainfall over portions of South Carolina. According to official reports, this historic downpour caused major flooding, claiming the lives of 19 people—nine of them in Richland County, which includes Columbia, the state’s main urban center. State emergency management officials also reported 1,500 water rescues in the Columbia metro area. Infrastructure damage across the state was significant, amounting to $1.5 billion.  

Brian Bates, who was a program director at Woolpert and a Columbia resident at the time, witnessed the flooding firsthand. Shocked and saddened by the destruction, Brian recognized the urgent need for a real-time flood forecasting—or “live modeling”—solution. He, along with colleagues Steve Godfrey and Gil Inouye, began developing a system based on traditional hydrologic and hydraulic (H&H) modeling, laying the groundwork for what would become FLOODTRACZ. 

The Story Continues: The Problems with Traditional H&H Modeling   
After joining Woolpert in 2021, I began collaborating with the team on the pilot project. As a subject matter expert in H&H modeling and predictive AI and ML, I shared insights that sparked discussions about the limitations of traditional H&H models.  

For starters, standard H&H models require extensive datasets and long run times when applied to large or complex watersheds. Time-intensive simulations make real-time forecasting a challenge, preventing communities from responding quickly and appropriately to storm events. Another disadvantage is that users must base their models on simplifications or assumptions, reducing reliability and accuracy. 

Other issues include calibration complexity, limited predefined factors, and difficulty accounting for both initial and boundary conditions. These limitations led our team to explore using ML, ultimately resulting in the creation of FLOODTRACZ—a platform designed for real-time flood forecasting.  

FLOODTRACZ Empowers Decision-Makers During Flood Events  
FLOODTRACZ uses today’s computing power, ML, remote sensing, and geographic information system technology to help communities observe flooding and track operational data, monitoring rainfall patterns and their potential flood damage in real time.  

When it comes to flooding, every second counts. FLOODTRACZ stands in stark contrast to traditional H&H models in that it provides ample lead time for decision-makers—allowing them to forecast when a flood will occur, how severe it will be, and how far it will extend. This information is instrumental for: 

• Allocating resources to improve response times.  
• Communicating clear evacuation protocols.  
• Optimizing discharge to control adaptive management systems. 

The FLOODTRACZ Cycle and Predictive Process in Action  
FLOODTRACZ requires a substantial amount of historical and real-time data. For historical data, it draws from agencies like the National Oceanic and Atmospheric Administration (NOAA) and the U.S. Geological Survey. It also collects real-time data, including rainfall measurements, stage gauge variations, and other relevant observations.  

Once the data is cleaned, formatted, and restructured, FLOODTRACZ applies ML to identify patterns in the historical data and compare them to current conditions. Its ML algorithms consider the real correlations between variables—for example, the relationship between rainfall and water depth—uncovering hidden patterns. FLOODTRACZ learns how these variables interact and uses that discovery to enhance accuracy. Then, FLOODTRACZ deploys its trained ML models, integrating them with rainfall predictions primarily obtained from the NOAA database to deliver flood forecasts up to seven days in advance. 

Additionally, the team and I understood that no effective system would be complete without a feedback loop. FLOODTRACZ retains its predictions for validation, comparing them to the actual rainfall and water level conditions to verify forecast accuracy. If validation reveals insufficient accuracy, the ML models are retrained with expanded datasets. This iterative process continues until the model reaches a satisfactory level of accuracy, bolstering forecast reliability and robustness over time.   

Why FLOODTRACZ Uses a Pool of Machine Learning Algorithms 
A unique feature of FLOODTRACZ is its use of a diverse pool of ML algorithms. While similar platforms typically use just one or two algorithms to train their models, FLOODTRACZ leverages as many as 20. Rather than applying a one-size-fits-all approach, this diversity enables FLOODTRACZ to evaluate multiple models and select the one that delivers the highest accuracy for a given location.  

This diversity accounts for the fact that no single algorithm can perform perfectly everywhere. One model might work well at one point along a stream or river, but not at another. That’s why having a large pool of algorithms is so valuable—it promotes the best performance possible.   

Upgrading Flood Forecasting Capabilities in the Face of Worsening Storms  
ML-based platforms, like FLOODTRACZ, will continue to gain traction and grow in popularity—especially as storms increase in frequency and severity across the U.S. As it stands, many communities still rely on traditional H&H models, which are rife with inefficiencies and limitations. Community officials must evaluate their current flood prevention strategies, noting what prediction models they have in place, and the potential value of switching to an ML-based platform.