The aftermath of a hurricane or major storm is easy to see in an agricultural or coastal region. The physical damage includes flooded roads and buildings, downed trees and disrupted infrastructure. However, some of the biggest environmental impacts can be more difficult to see.
Damaged fuel tanks and control valves can leak, spilling diesel in agricultural and industrial settings. Floodwaters can then spread contaminants across landscapes, making them difficult to locate and assess.
New research from the UF/IFAS Department of Soil, Water and Ecosystem Sciences (SWES) explores a possible solution: Could hyperspectral sensing and artificial intelligence help detect diesel contamination in soil?
“Pairing the two could lead to faster detection over larger areas,” said Nikos Tziolas, a SWES assistant professor of soil science and artificial intelligence. He oversaw the study from his lab at the UF/IFAS Southwest Florida Research and Education Center.
Finding Diesel Contamination After a Storm
The study is best viewed as an early-stage proof of concept. It does not demonstrate a ready-to-deploy hurricane response system. Instead, it is a first step toward showing how remote or automated sensing and AI can complement traditional contamination measurements.
Traditional field measurements for locating diesel contamination can provide useful, rapid screening. One method uses photoionization detectors (PIDs) to measure volatile organic compounds (VOCs) that emit from the soil into the air. These measurements require a person to be at the location, something that may be difficult or dangerous after a hurricane, and to know exactly where to look.
The research team explored whether reflected light could help make that process faster and more efficient.
“Hyperspectral sensors measure light across many wavelengths, so we can identify patterns, or spectral signatures, associated with different materials,” said Brierra Mack.
Mack is a SWES graduate who earned a master’s degree working with Tziolas to study diesel contamination in soil. She attended classes online while she was a regulatory emergency responder for the Florida Department of Environmental Protection. Mack now works as an environmental emergency responder in the federal government.
“The data suggest that parts of the light spectrum may reveal contamination that conventional imaging cannot easily detect,” she said.
How AI Fits In
Collecting spectral data is only half the challenge. A single measurement can contain hundreds of wavelengths, making it difficult to identify useful patterns manually.
The researchers used two machine-learning methods, Random Forest and Cubist, to connect spectral patterns with measured VOC concentrations. The goal was not for AI to independently find diesel spills, but to use spectral data to estimate where contamination might be concentrated.
“In a future field application, large areas could be scanned first,” Tziolas said. “Then AI can help identify locations for follow-up PID measurements, soil sampling and laboratory analysis.”
The researchers tested the approach in a controlled laboratory experiment using soils from three Florida locations. They added different levels of diesel to bare and partially vegetated soil, then measured the samples over seven days with two spectral instruments. A PID was used to measure VOCs from the soil.
Higher diesel concentrations generally produced darker spectral readings. Researchers also found that VOC levels and spectral features changed over time, while vegetation added further complexity.
Impact and Next Steps
The study suggests that combining spectral sensing with machine learning could eventually provide a faster way to screen soils for recent diesel contamination.
“More testing is needed with different variables,” Mack said. “You have soil moisture, vegetation, sunlight, weathering, debris and differences between soils that can all affect the spectral signal.”
The greater opportunity is to move environmental assessment beyond checking one location at a time toward large-scale testing.
“We are not there yet,” Tziolas said. “But this study shows a promising way to help environmental teams decide where to look first after major storms.”
“This work could aid in increasing the effectiveness of response teams charged with locating and addressing contamination incidents post-hurricane landfall,” said Mack.
The complete journal article is available online in Soil Security through ScienceDirect: https://www.sciencedirect.com/science/article/pii/S2667006226000250
Featured image above from the Florida Department of Environmental Protection’s Hazard Assessment Response Team.
5
