Planes in the air assess floods on the ground

Kevin Kochersberger, professor in mechanical engineering, is the director of the Uncrewed Systems Laboratory, where he leads a team of undergraduate and graduate students.

His team's expertise lies in using unmanned aerial vehicles (UAVs), including drones, to gather visual data from the air that can be interpreted with machine learning. This means using images they gather to build data-rich maps.

The team played a vital role in surveying the damage caused by Hurricane Helene in October. The destruction knocked out roads and forever changed the landscape of large portions of the eastern United States. According to the Virginia governor's office, 310,000 Virginia residents were without electricity when the infrastructure of power grids was severely damaged, and 484 primary and secondary roads as well as 118 bridges were closed.

With devastation as widespread as this, the abilty to survey the affected areas became a major obstacle. With roads being washed away, the most effective choice was viewing it from the air. UAVs are widely available with cameras capable of doing this, but the scope of Helene's wrath was so large that drones couldn't fully cover it. The team had a solution.

Mapping terrain from the air is familiar work for Kochersberger's team, and so is the area. Team members conducted similar research in 2021 in Southwest Virginia,  surveying flood damage using UAVs. Because their UAVs weren't the tool for this job, they would need to fly something bigger. Luckily, Kochersberger isn't just a UAV pilot, he's also licensed to fly small-engine planes.

Elizabeth Lohman, a manager in the Office of Pollution Response and Emergency Preparedness, had contacted Kochersberger to bring this project together.

“Over the last few years, we have developed a relationship with Virginia Tech's Mechanical Engineering Uncrewed Systems Lab,” Lohman said. “We have used their remote sensing technologies in combination with 3D modeling and GIS tools to develop an immediate post-disaster picture of impacts such as shoreline oiling and evaluation of debris. We reached out again to see if they could support the need to estimate the amount of debris on Claytor and South Holston lakes and to help FEMA [the Federal Emergency Management Agency] and USACE [the U.S. Army Corps of Engineers] debris experts develop a scope of work for a mission assignment to remove the debris.”

The plan was this: Kochersberger would outfit a plane with the mapping gear, fly the missions himself, and get the data back to his team on the ground for processing. To achieve this, he enlisted students from both engineering and geography majors, an ideal mix for the work to be done. His team included

• Devon Casey, an undergraduate student in the Department of Geography in the College of Natural Resources and Environment
• Suyog Gautam, a master’s degree student in the Department of Geography
• Daniel Whitehurst, a Ph.D. graduate research assistant in the Department of Mechanical Engineering in the College of Engineering
  
• Diksha Aggarwal, a Ph.D. student in the Department of Mechanical Engineering

“Following Helene, the VDEQ [Virginia Department of Environmental Quality] requested imagery late on Friday for Claytor Lake, South Holston Lake and the towns of Damascus and Taylors Valley,” Kochersberger said. “We flew drones after the 2021 flood in Hurley, Virginia, and VDEQ initially thought we would operate drones again. The requested areas were too large for an uncrewed aircraft, but fortunately a friend had a motorglider which we could quickly outfit with gear."

Old and new data

As a starting point, Kochersberger’s team used an existing terrain map of the state of Virginia. This map came from the Virginia State LiDAR database, a publicly available collection of 3D geospatial measurements. This gave the team a "before" picture of what the landscape looked like before the flooding. Having two geography students on the team was an asset in that effort.

The aerial imagery was collected using a Phoenix U-15 motorglider, a two-passenger aircraft capable of slow flight to collect imagery at low altitudes. Pilot Jeff Shingleton provided his aircraft for the mission, while Kochersberger took the controls to navigate a complex route that guaranteed enough imagery to build the 3D models. The team imaged a total of 77 square miles of affected areas.

Back at the Uncrewed Systems Labratory, 3D reconstructions and orthomosaics — composite images — were generated, representing “after” conditions for the areas team members surveyed. The geography students used their expertise with geospatial analysis software to help the engineers determine area coverage of debris and volume of displaced material. Mechanical engineering Ph.D. student Aggarwal brought engineering and geography together with the other members of the team.

“We focused on creating 3D models of the lake areas and detecting debris within these lakes,” Aggarwal said. “However, the images included both land and lake regions, making it challenging to extract just the lake. Using machine learning segmentation algorithms would have taken significant time and high computing capacity — resources we didn’t have available. We wanted a quick and effective way to accomplish it.” 

This is where Gautam and Casey’s expertise came in. While the two geography students weren’t regular members of the lab, they were hand picked to assist with this project. 

“We’re grateful they were available on such short notice, as they also brought new perspectives to our brainstorming sessions on mapping approaches,” Aggarawal said.

The geography students worked closely with another mechanical engineering Ph.D student Daniel Whitehurst, who brought critical interpretation to their findings.

“Our first priority was looking at debris in lakes,” Whitehurst said. “From there, we were looking for water heights and the amount of land movement into the community. This shows how much cleanup is necessary.”

On land, the aerial data was compared to the state’s LiDAR database. By studying the difference between the “before” and “after” data, they found washed-out bridges, fallen buildings, and a massive amount of sediment and objects displaced throughout the countryside.

In a few days, the team was able to compare the previous geography data with their new results, create a view of the damage, and present it at the flood relief headquarters in Richmond. The data was passed along to the agencies taking action.

“The Virginia Tech team was available and quickly put together a team to collect images and process them to develop estimates,” Lohan said. “That analysis is helping the Corps of Engineers right-size the scope of work, which also helps  set right-sized expectations for executing and completing the mission.”