Using Lidar and Deep Learning to FInd Abandoned Using Lidar and Deep Learning to FInd Abandoned
29 November 2022

Using Lidar and Deep Learning to FInd Abandoned Wells

Written by: Kameron Hall

Hello everyone! I’ve been working for the US Fish and Wildlife (USFWS) service for about three months now. I’ve been working more on the technological side of the agency instead of working in the field, I have been using remote sensing data and GIS software to assist refuge managers in managing the land.

A particular focus of my work has been identifying abandoned oil wells on national wildlife refuge land. Abandoned wells are wells that are no longer in service to produce oil or gas and often have no known owner, but they may continue to emit methane, a greenhouse gas and general air pollutant, into the atmosphere (1). Wells have been constructed for many decades with the EPA estimating that “approximately 2.56 million wells… had been drilled in the US by 1973” (2). Unfortunately, the locations of some of these wells have been lost to time and accurate GPS coordinates of these wells may not be available. This presents a massive issue due to the potiential for large amounts of methane to enter the atmosphere along with other pollutants that originate from these wells.

Picture2 Well

Image of an oil well from Wikicommons. Attribution:Flcelloguy, CC BY-SA 3.0>, via Wikimedia Commons. Page URL:

I use open Lidar data provided by the USGS to derive elevation data that can be used to more easily identify oil wells in an area. I was tasked with looking for oil wells near or within the several wildlife refuges in the Southwest region of the USFWS. The images produced from the elevation data help identify wells by differentiating the topography of oil well pads from the rest of the terrain. Well pads were often elevated from the surrounding ground to prevent flooding and this characteristic is useful in differentiate them from surrounding terrain (Steven Sensie, My Supervisor).

Well Example  Well Example 2

The imagery on the left is provided by Maxar while the image on the right was derived from Lidar Data provided by  the USGS 3DEP program (3) and modified with GRASS GIS. A well is likely located near the center of the image. Both images are at the same location.

Oil wells can be difficult to see using natural color imagery alone as seen in the image on the left. However, utilizing elevation data from Lidar allows you to more easily see the location of an oil well. Lidar is also very useful because the sensor can penetrate tree canopies, allowing me to find locations that may have abandoned wells.

Once I have labeled the wells within the lidar derived image, I used the labeled images to train Deep Learning algorithms that are used within ArcGIS Pro to identify wells that I may have missed.

Combining imagery from elevation data and deep learning algorithms increase the ability for me to identify potential abandoned wells and inform refuge staff, who may be able to find and plug these abandoned wells.



Agency: U.S. Fish and Wildlife Service

Program: Civilian Climate Corps Program (CCC)

Location: Southwest Regional Office

MANO Project
is an initiative of Hispanic 
Access Foundation.

P: (202) 640-4342