MONTE CARLO FLOOD PROBABILITY MODELING

Worldwide, around 2.8 billion people, or 30 percent of the world’s population, live near a major body of water; many of those communities are cities that only keep growing as the world becomes more urbanized. The 2010 floods in Tennessee alone did more than $1.5 billion worth of damage. With each passing year, the need for effective flood prediction becomes more urgent.

Computer modeling of floods has been performed for decades, with applications to forecast and plan communities being built around these models. Older systems of flood modeling, however, had real drawbacks; the one-dimensional system of modeling did a poor job of accounting for terrain and flow paths especially in urban areas. Additionally, factors like the velocity of floodwaters, duration of the flooding, and the direction flood waters may take have proven difficult to predict.

“Flood control mechanisms like dams, levees, detention basins and early-warning systems have all evolved in the last 75 years,” said ecivil and environmental engineering assistant professor Alfred Kalyanapu, “but it’s still impossible to see absolute protection from catastrophic floods. Back when the upper Mississippi River flooded, in the early Nineties, more than 1,000 levees failed or were swamped, causing over $20 billion in damage. Our challenge is to narrow down the gap between model and reality, and remove uncertainty from the equation.”

That’s the challenge for Tennessee Tech College of Engineering graduate students Ebrahim Ahmadisharaf and Nowfel Mahmud Bhuyian, along with faculty advisor Kalyanapu. The paper they submitted, titled ‘Impact of Spatial Resolution On Downstream Flood Hazard Due To Dam Break Events Using Probabilistic Flood Modeling,’ made the team one of three winners in the 2013 American Society of Dam Safety Organization’s Student Competition.


Two-dimensional models were more effective and accurate, but computationally intensive and timeconsuming to process, meaning only a few academic research facilities and consulting firms would have the resources for this type of work. Overly-simplistic models sacrifice accuracy for expediency; considering that these models play directly into policy- and decision-making and emergency planning, this uncertainty becomes a prime concern.

Kalyanapu has been able to repurpose graphic processing units (GPUs) and use their robust processing abilities to make the entire model run several times faster. In addition, GPUs, which were originally designed for devices like the Xbox 360, can lend a whole new capacity for graphic representation of flood events. The “Monte Carlo” term refers to the ability to run the same analysis over and over again to take into account all possible variables. Previous work by Kalyanapu gave accurate estimates of the velocity, depth and progression of flood waters when the Taum Sauk Dam in southern Missouri failed in 2005.

“What would have taken 4.5 hours to simulate floods on a regular computer now takes three minutes using Flood2D-GPU,” said Kalyanapu. “Our research group worked with Computer Science Senior students to make Flood2D-GPU model available online. Now, anyone with an internet connection can access TTU’s webpage to run these sophisticated models.”

The specific project developed by Ahmadisharaf and Bhuyian was based on a hypothetical flood event in the Swannanoa River Watershed in Buncombe County, NC. In September, the two will travel to Rhode Island to present their findings at the ASDSO’s National Conference. At the conference, Ahmadisharaf and Bhuyian will be recognized at the Dam Safety 2013 Awards Ceremony, and presented with an award to commemorate their achievements.

“With the success of our research, we now are planning to continue our research in MC flood modeling and applying it to other rivers, including Nashville along the Cumberland River and Mississippi Rivers,” Kalyanapu noted.