Swiss cheese: Potentially hundreds of thousands of undocumented orphan wells (UOWs) exist, and while many could be harmless, most were “plugged and abandoned” with little more than walnut shells and gravel.
That means leaks of environmentally harmful liquids and gases are possible.
Here, kitty kitty: The solution seems clear: find them and plug them with modern methods, but only about 50% of that equation is easy because we don’t know where most of these UOWs are.
Where’s Waldo: Older wells often have no better location information than “three tall man’s paces left of Bob’s tree,” and Bob’s tree was cut down 80 years ago.
Modern tools such as drones, LIDAR, and an array of sensors make locating wells easy (if a house isn’t on top of the well, of course), but knowing where to start is an issue.
Old school plus new school (again): Late last year, scientists at Lawrence Berkeley National Lab sought to change that using old data and new AI.
Plethora of data: Since 2011, the USGS has uploaded 190,000 scans of historical (between 1884 and 2006) USGS geotagged topographic maps.
For maps between 1947 and 1992, a hollow black circle consistently represents an oil and gas well, which is useful, but the US covers more than 3 million square miles.
That’s a lot of maps to cover, even for an intern fueled by Celsius and meat sticks.
Many of today’s “older” AI algorithms were built for almost exactly this, however.
The algorithm can be “trained” on what a well would look like on maps — using that symbol as a guide with some refinement to discern between cul-de-sacs or just a stain on the map — and then set to run amok across the rest of the data set.
X marks the spot: Charuleka Varadharajan’s team at Berkeley used AI on data from four counties of interest — LA and Kern counties in California, and Osage and Oklahoma counties in Oklahoma — and found 1,301 potential undocumented orphaned wells.
Geocaching skills transfer: The real work begins once the team IDs a potential location, because it’s been learned the algorithm tends to put the team within about 10 meters (30 feet) of the actual location.
Fine-toothed combs: A combination of on-the-ground footwork, drones to detect local methane leaks, LIDAR to search for near-surface structures, and magnetometers to look for any indication of casing or wellheads can help locate the well.
It seems like more work is needed to refine the results even further (though it’s possible there really were 1,300 undocumented wells in those four counties given their oil and gas history), but there’s a running theme here of using new tools with old data to tackle these problems.
Geoscience scout card skills: As folks familiar with poring over scout cards and other resources to squeeze every last bit of information about potential wells, geoscientists are best poised to help with this sort of work.
To read the paper, go here.