To improve flash flood response in the Appalachian region, researchers worked with Amazon Web Service (AWS) to prototype a hydrological prediction tool capable of providing the National Park Service flood hydrographs with 18-hour lead times. Led by CEE Department Head Ana P. Barros, the project to develop the model is being made possible through cloud computational resources funding from the Amazon Sustainability Data Initiative. AWS will host the model, allowing easy access to large data sets necessary for prediction.
Researchers hope improved flood modeling will aid the staff of Great Smoky National Park in natural hazard planning, decision-making and fostering resiliency in areas vulnerable to flooding.
Flash floods occur in the Appalachian Mountains for two primary reasons. In mountain ranges, high elevation produces orographic enhancement, in which moist air masses are forced to rise, encouraging cloud formation and therefore increased precipitation. Additionally, the complex terrain and steep slopes make it difficult to store and contain the large amounts of water produced by rapid rainfall.
“In mountains, the watersheds or river basins near the ridges are organized like teacups on the landscape, and they’re very small teacups,” said Barros. “So, if a precipitation cell sits on top of one of these teacups, very soon it will fill up.”
When heavy rain hits, the basins quickly become saturated. This forces ensuing rainfall into streams and leads to a flash flood response. Flash floods pose less of a public threat at higher elevations because these areas often contain less dense populations. However, lower elevation mountain ranges like the Appalachians are well populated, with large communities and national parks that receive thousands of daily visitors dispersed across the landscape. Knowing about potential floods far in advance is crucial to making decisions that protect those living in and visiting the area.
“The longer the lead time is, much more can be done in terms of preparing for flash flood events: issue warnings, evacuations, school closing, park closing, etc.,” said Mochi Liao, a post-doctoral researcher working with Barros on the development and distribution of their prediction tool, the Duke Coupled Hydrology Model (DCHM).
DCHM is a physically based, fully distributed hydrological prediction model capable of producing the advanced warning NPS needs. Its features include runoff, two-way streamflow routing, surface-subsurface interactions, multi-layer snow physics, variable soil depths, wetland physics, data assimilation, prognostic phenology and dynamic vegetation roots. To run a prediction, DCHM requires weather, soil and landcover data input from the area of interest as well as weather forecasts.
For NPS, using a model like DCHM presents two challenges. First, they have to obtain the knowledge framework necessary to host and operate the model. Additionally, staff must deal with the time-consuming nature of finding, cleaning and clipping appropriate data to run a model prediction. Under the pressure of impending floods and heavy rain, clunky preparation procedures become an even greater hindrance.
The Amazon Sustainability Data Initiative (ASDI) resolves these issues by providing seamless data accessibility and integration between the data and modeling systems. ASDI supports research and innovation in sustainability by minimizing costs and time required to find and analyze large data sets related to sustainability efforts. ASDI works with agencies like NOAA, NASA, the UK Meteorological Office and ESRI to collect and host key data sets produced by each group on AWS.
As a hub of sustainability information, AWS allows researchers and data scientists to easily find and utilize important data needed for projects targeted at improving global resiliency. Connecting NPS with DCHM exemplifies this mission by enabling access to flashflood risk information long before the floods reach their peak.
Ultimately, it only takes DCHM fifteen minutes to produce a flood forecast with an end-to-end maximum latency of 20 minutes. When input with high resolution rapid refreshing (HRRR) data from NOAA, which is available through AWS and provides real-time 3km resolution atmospheric data that refreshes every 15 minutes, DCHM can produce predictions with 18-hour lead times. On six-hour intervals, it can even produce lead times up to 48 hours.
The key to improving natural hazard planning lies in providing decision makers with more time between receiving a forecast and the peak of the flood event. With the prototype forecast system made possible through support from AWS, the NPS is one step closer to increasing their overall flood response efficiency. Once implemented for operational use, Great Smoky National Park’s use of the new tool could set a precedent for flash flood planning and natural hazard resiliency across the entire Appalachian region.
“One cannot stop an earthquake from happening, cannot stop a big storm from coming through,” said Barros. “However, we can take measures to ensure we anticipate and mitigate the consequences. This is the essence of resiliency.”