暗网视频

Hoyt Thomas, a graduate student in geomatics through UAA鈥檚 , will defend his master鈥檚 thesis, 鈥淚ntegrating Anomaly Detection and Deep Learning for Locating Abandoned Oil and Gas Wells Using Drone Mapping,鈥� on Aug. 15. 

Thomas鈥� work, which is supported through a grant from the National Science Foundation (award number: 2119689), aims to locate uncapped and improperly sealed oil and gas wells that can pose risks to the public. In order to identify these hazards and gather the necessary data for future mitigation efforts, Thomas is utilizing cutting-edge technology, including drones and artificial intelligence (AI). 

Abandoned wells pose risks to climate and human health

The drones Thomas used in the research are equipped with special sensors to detect methane emissions, which may come from abandoned wells. 鈥淭here are multiple environmental and human health risks associated with abandoned oil and gas wells, predominantly related to the methane emissions they release,鈥� said Thomas.

, methane is second only to carbon dioxide as the most abundant anthropogenic greenhouse gas. Using sensors on drones to identify methane leaks from abandoned wells that can later be mitigated has the potential to help reduce greenhouse gas emissions that contribute to climate change.

鈥淏ased off measurements at known, abandoned wells, [methane emissions from abandoned wells] comprise between 3 to 6% of the industry's total methane emissions,鈥� said Thomas. "It's a significant number. It's not the biggest chunk, but you know, when you're looking at climate change, every little bit helps.鈥�

In addition to methane emissions, abandoned wells pose multiple risks to human health. The wells can leak oil and gas, which poses the risk of contaminating groundwater. The abandoned infrastructure itself can also be a threat in areas used for recreation. 鈥淭he piping, the well casings, whatever鈥檚 left behind, can pose a risk,鈥� said Thomas. 鈥淚f you have all this rusting piping and somebody鈥檚 doing a trail run through the area, something could happen.鈥� 

Lidar and AI used to detect oil and gas wells that were 鈥榣ost to time鈥�

The drones used in the project are also equipped with light detection and ranging (lidar) sensors to help locate the abandoned wells. The lidar sensor allows the drone to 鈥渕ap the geomorphology of the abandoned well sites,鈥� said Thomas. 

"There are a couple million of these abandoned oil and gas wells throughout the entire country,鈥� Thomas explained. 鈥淎 lot of these are lost to time. The records [were] either misfiled or just never filed in the first place, so the exact numbers are hard to estimate.鈥�

Lidar is used in aerial surveys across disciplines to create 3D models of landscapes. Lidar is noted for its ability to penetrate through foliage that would obscure the ground in visual surveys. This advantage of lidar plays a crucial role in Thomas鈥� work. 鈥淲e鈥檙e using a product of elevation models,鈥� Thomas said. 鈥淚t鈥檚 not direct imagery, because you might have vegetation covering the well site.鈥�

In Alaska, the abandoned well sites may not necessarily be 鈥渓ost to time,鈥� but locating them can still be a challenge.

鈥淚n Alaska鈥檚 case, the oil and gas history is relatively young compared to the rest of the country,鈥� said Thomas. 鈥淓ssentially when they first started doing the oil and gas work, the coordinate system they used [to record sites] was very general. Then, when they were abandoned, those records never got updated with a more accurate location.鈥�

Another goal of Thomas鈥� thesis work is to provide 鈥淎I deep learning assisted methodology鈥� for the ongoing search for abandoned oil and gas wells around the country. 鈥淎laska has a smaller number [of abandoned wells],鈥� said Thomas. 鈥淚 believe the number of wells they're trying to remediate is in the 10 to twenties range [...] but our methodology is more for a broader set.鈥�

The deep learning model Thomas developed can be used to examine lidar data captured by drone surveillance to identify sites of interest, including sites that can鈥檛 be detected by methane sensors. 

鈥淣ot all of these wells are emitting methane, and they still pose risks. So we want to be able to have a sort of a 鈥榗atch all鈥� type method,鈥� Thomas explained. 鈥淭hat's one of the reasons why we're working on this deep learning assisted model is to have some sort of other methodology to nail down where some of these sites are, especially if they aren't producing methane.鈥�

Work on the project will continue after Thomas completes his master鈥檚 program. 鈥淔urther development under the ongoing grant will advance this work toward a scalable, AI-enabled environmental intelligent system designed to protect communities and help restore ecosystems affected by legacy energy infrastructure,鈥� said Caixia Wang, Ph.D., professor of geomatics, Thomas鈥� graduate advisor.