What Hurricane Helene revealed about flood forecasting
- ✓Earlier warning — HydroForecast flagged the potential for extreme flows four days before NOAA’s River Forecast Center.
- ✓Greater forecast accuracy — Forecast error (RMSE) was 33% lower than NOAA RFC forecasts in the days before peak flow.
- ✓Continued operation during gauge outages — HydroForecast kept forecasting through the USGS gauge outage at Nolichucky Dam, closely tracking the eventual observed peak.
Hurricane Helene’s catastrophic flooding across Tennessee and North Carolina in September 2024 displaced families, destroyed homes, and left damage to communities and ecosystems that will linger for years. It was a stark reminder of both the growing force of extreme weather and our shared responsibility to prepare for it.
For dam operators, water managers, and emergency coordinators, Helene’s record-breaking rainfall exposed the limits of current forecasting and monitoring systems. As stream gauges failed and rivers surged past predicted levels, critical decisions had to be made with incomplete information. This isn’t for lack of effort, but because existing tools are being outpaced by climate-driven extremes. Still, there are glimpses of how things could improve by building a stronger toolkit that gives communities and decision-makers the best chance to act early and reduce harm.
Want the technical details? Download a summary of how AI streamflow forecasts performed during Hurricane Helene.
Case study: HydroForecast predicts flow during Hurricane Helene with early accuracy and without interruption
To assess the current strengths and limitations of forecasting systems in-use by areas impacted by Hurricane Helene, we analyzed the performance of NOAA RFC’s traditional physics-based river forecasts in the days leading up to the storm compared to observations of USGS stream gauges. In parallel, we evaluated the coincident forecasts produced by HydroForecast, an AI-powered streamflow forecasting methodology, across several basins to understand the potential opportunity for earlier warning and greater resilience as the flood event unfolded.
Tracking the storm: A timeline of forecasting technologies and Helene’s impact
The week ahead of peak flows (September 21 - 24)
The public was aware of a developing tropical system in the Western Caribbean. News outlets and forecasters noted that a consensus among models pointed to a system forming, but there was still high uncertainty about its final path.
In this period of uncertainty, HydroForecast’s probabilistic forecast showed the potential for significant peak flows across multiple monitored basins in the potential path throughout mountainous regions in North Carolina and Tennessee. This prediction came seven days ahead of the event's eventual peak in these areas, providing a crucial early warning of the extreme event four days earlier than RFC forecasts.
Over the next few days, HydroForecast predictions improved and the model’s confidence increased. RFC forecasts remained flat with no indication of the extreme event coming.
- On September 23, 5 days ahead of the peak, HydroForecast showed the potential for significant flows at a forecast location near Nolichucky Dam.
- On September 24, 4 days ahead of the peak, Helene was officially formed and named, but the RFC showed minimal change as traditional forecasts tend to react closer to the event.
Probabilistic advantage for early warnings
HydroForecast’s ability to detect extreme events early lies in its probabilistic model that presents a range of statistically possible outcomes rather than single-point predictions. These easy-to-interpret confidence intervals not only capture unprecedented events, but empower emergency planners and hydropower operators to clearly see the uncertainty and risk associated with impending situations.
The days ahead of the peak (September 25 - 26)
The RFC predicted significant flows from the hurricane for the first time with a sharp increase on the 25th, followed by another increase on the 26th. HydroForecast remained relatively consistent and continued to improve confidence and accuracy.
Accuracy and consistency in forecasting
While early warning is extremely important ahead of events like Helene, forecast performance isn’t only measured by when the first prediction occurs. HydroForecast's predictions were also demonstrably more accurate. In the 1-5 day window before the peak flow, HydroForecast reduced flow volume error (RMSE) by 33% relative to the NOAA RFC.
Consistency in forecasting is also a key metric. In contrast to other forecasts, HydroForecast delivered gradually evolving forecasts that enable a more reliable tracking of the developing situation, with forecasts being automatically reissued every two hours.
24 hours ahead of the peak
Hurricane Helene was in full force over North Carolina by September 27—inland flooding began and mandatory evacuations were ordered.
At Nolichucky, a critical example of “hydroblindness” unfolded when the only USGS gauge at the Nolichucky Dam went offline around 11:00 AM. Later that evening, out of an abundance of caution due to a lack of real-time insight from the failed gauge, "Condition Red" warning was issued for an "imminent breach" as historic flows begin to overtop the structure.
As the event developed, the RFC under-predicted the peak and flow recession after the height of the storm. Because traditional models rely on real-time data from stream gauges, as the gauges were overwhelmed, there was a significant gap in critical information and uncertainty about the exact peak flow.
Built-in resilience provides equitable coverage
Even without data from the USGS gauge at Nolichucky River, HydroForecast continued operating seamlessly, predicting 90,458 CFS at the time of the gauge outage—closely matching the final observed gauge reading of 86,100 CFS.
The model automatically integrates multiple data sources and incorporates up to four weather forecast models simultaneously, creating natural protection against infrastructure failures and meteorological uncertainty. As extreme weather events become more frequent due to climate change, total geographic coverage and resilience is particularly vital for remote and underserved communities that lack monitoring infrastructure.
Peak flows (morning of September 28)
In the early morning of the 28th, the flows, water levels, and flooding peaked. The Nolichucky Dam was deemed "stable and secure" and water levels began receding. Emergency hotlines were established and recovery crews were deployed to assess the extensive damage.
HydroForecast accurately reflected the receding limb of the flood. Predicted recession timing and magnitude even before the peak occurred. RFC showed recession but with uncertainty about the exact rate and timing due to loss of real-time data from gauges.
Post-peak
While peak flow is often the first priority ahead of flooding events, the recession—how long elevated water levels persist—is critical data for emergency managers. The level and duration of high water can impact how and when emergency crews are deployed, as well as determine when it becomes safe for residents and civilians to enter an area. After Helene, water receded slowly over several days. Some rivers in Western North Carolina stayed at flood levels through the first week of October 2024.
Pre-peak recession insights: Planning for the aftermath
Even before the peak occurred, our analysis showed HydroForecast accurately predicted recession timing and magnitude. With early insight into elevated water levels, teams can make more informed decisions around resource allocation and deployment while providing clear, consistent guidance to communities.
The road to climate resiliency lies in widespread adoption of advanced technologies
Hurricane Helene tested water managers across the Southeast in unprecedented ways. The storm showcased both the ingenuity of teams on the ground and the critical cracks that can occur in traditional forecasting methodologies when pushed to the limit. While catastrophic dam failure was ultimately avoided, the event made clear that we can—and must—integrate more resilient and performant technologies. As climate change continues to intensify hurricanes, and inland flooding risks are projected to increase, the road to resiliency is a rapid implementation of advanced forecasting technology to protect the communities living in the path of the next unprecedented storm.
If you want to learn more about the results of this case study, please reach out to our team for additional technical details.
