An underwater drone capable of carrying seven sensors and taking leaf and soil samples is increasing research capabilities at the South Dakota Water Resources Institute. However, the sampling capabilities are not what makes it unique.
“What’s new is the ability to revisit a specific place multiple times over a season,” explained assistant professor Aaron Franzen. His research in the Department of Agricultural and Biosystems Engineering focuses on emerging precision agriculture technologies, including drones, electronics, diagnostics and control systems. The drone, which is made possible through department funding, will be used for education, research and outreach.
“Normally, sensors are hung from one place and left over a long period of time to collect data in one single spot,” he explained. However, Franzen and undergraduate student Alex Masloski, who built and tested the drone last summer, equipped the remotely operated underwater vehicle with a 3D locator that allows the researchers to create a 3D map.
Using this technology, researchers can “plan a path and then repeat it taking measurements at the same spots along the path, weekly or biweekly, whatever we determine is necessary,” Franzen said.
The 50-pound drone is about the size of a 2-foot cube, has eight thrusters and operates on tether that is a little more than half a mile long. The 3D locator is necessary because “GPS does not work underwater,” Franzen explained. The locator uses ultrasonic technology via a beacon on the drone and four receivers on the surface of the water connected to GPS. This then creates a full history of where the drone was when it collected each measurement.
“This greatly increases our capacity to do water quality research in both time and space,” said Franzen, who is seeking to partner with engineering firms and state agencies to utilize this new technology.
Combining drone with sensors
To begin exploring the drone’s research potential, Rachel McDaniel, then an assistant professor, purchased an array of sensors to gather parameters, such as oxygen, chlorophyll-a, nitrates, phosphates, pH, temperature and salinity.
Once on-campus research work resumes, the sensor array will be mounted on the drone. In addition, Franzen explained, “All control signals, telemetry from the drone, and sensor readings will eventually be transmitted through the same tether cable to the top-side computer.” This portion of the work is also supported by department funding.
Last summer, the researchers deployed the sensor array and the drone separately as part of preliminary testing for a research project to gather 3D water quality data at Lake Mitchell, which has experienced problems with algal blooms for decades. The project’s long-term goal is to develop a model to predict level of risk for algal blooms in eastern South Dakota lakes two weeks in advance.
Assistant professor John McMaine, SDSU Extension water management engineer, is coordinating the project, begun by McDaniel. She will continue working on the project as an adjunct faculty member. The research is funded by the U.S. Geological Survey 104B Small Grants Program administered through the Water Resources Institute. One master’s student and one doctoral student are also working on the project.
The technology’s 3D capabilities will allow researchers to use geographic coordinates to track the exact position—vertically and horizontally—at which measurements are taken in the lake. “We can have it automatically take a sample every couple of seconds to get really detailed measurements,” McDaniel said. Those geographic coordinates will allow the researchers to compare the Lake Mitchell surface data to those recorded by remote sensors.
Developing a risk model
The South Dakota Department of Environment and Natural Resources reports chlorophyll-a, one of the primary indicators used to determine algal bloom intensity, is the second-most prevalent stressor in lake water bodies.
“Early prediction of algal bloom risk can provide critical information that has the potential to help water resources managers, state and federal agencies who develop policies aimed at mitigating future occurrences of this hazard,” explained doctoral student Kevin Brandt, who is also SDSU’s Director of Research Computing. He is in the early stages of developing an algal bloom risk prediction model based on a machine learning technique called deep learning using chlorophyll-a as the measure of algal bloom intensity.
“The deep learning training algorithm is iterative, improving the accuracy of the prediction model with each pass by calculating errors and adjusting the connector weights until the prediction model reaches an acceptable margin of error,” he noted.
Brandt plans to integrate 30 years of data into the deep learning model training process. “I’m using an extensive amount of historical data to train the prediction model,” he said. Data for this study has been collected from the underwater drone, the city of Mitchell, Mesonet at SDState, the U.S. Geological Survey and the National Water Quality Monitoring Council database
In addition to chlorophyll-a concentrations, Brandt’s initial work will incorporate other potential influencers of algal bloom growth such as nutrient load, pH, salinity, temperature and oxygen in the lake water, air temperature, wind, rainfall event amounts and intensity, land use and topography. Once completed, the prediction model will classify algal bloom risk as high, medium and low for 19 lakes in eastern South Dakota, providing an essential tool for water resource management.