Every credible decarbonization plan begins with a hole in the ground. Lithium for batteries, copper for grids, rare earths for motors and magnets — the energy transition is an extraction project before it is anything else. So the question worth arguing about is not whether to mine, but whether mining can be watched closely enough to answer for what it does to land, water and the people living downstream. It can. The instrument orbits a few hundred kilometers overhead, and most of what it produces is free.
Landsat has been a reliable Earth observation tool by NASA and USGS since 1972. The Copernicus program has also been another source of free data since 2014. But what’s the progress since that time? What changed is cadence. Satellite tasking ensures that a satellite crosses the same pit every five to sixteen days, which turns a seasonal field campaign into continuous mine monitoring with satellite analytics, at a one-hundred-square-meter bench or across thousands of square kilometers.
Oversight across the whole life cycle
Mining monitoring from orbit is no longer a snapshot taken after the damage; it runs the length of a mine’s life.
- Exploration. The EU-funded S34I project built a cobalt-prospecting prototype on Landsat-9 and Italy’s PRISMA hyperspectral sensor, with machine learning doing the pattern recognition. Its motto is “from sky to ground”.
- Planning. Thacker Pass, Nevada. The original footprint for the lithium mine sat inside prime greater sage-grouse habitat, which Landsat’s vegetation and land-use records made visible before any ground was broken. Sage-grouse are worth tracking because their numbers rise and fall with the condition of the sagebrush steppe around them. The footprint moved six miles south, onto ground the birds largely avoid. What made that possible was having decades of imagery to compare against.
- Operation. M4mining pulls hyperspectral data off drones and satellites and builds it into one 3D mineralogical model of the deposit, revised while the pit is still being worked rather than after the fact. Project coordinator Edmond Hansen describes the aim as narrowing the lag between scanning a rock face and doing something about what the scan found. Cleaner separation of ore from waste means less material ending up on the tailings pile, and haul trucks burn less diesel carrying rock that was never worth moving.
Tailings dams settle the argument
Tailings dams are the least interesting structures on a mine site. They are also the ones that kill people, and that is why mining remote monitoring pays for itself here. On January 25, 2019, the dam at Vale’s Córrego do Feijão mine in Brumadinho, Brazil, collapsed. The final death toll was 270. Failures like this are hard to see coming because the warning signs sit inside the slope, not on top of it. Stress shifts, material weakens, and none of it shows on the surface. By the time something is visible, it is too late to act.
Radar reaches inside that window. A team at Kunming University of Science and Technology worked through more than 150 pairs of Sentinel-1 images covering a tailings dam in Yunnan, China, running them through SBAS-InSAR — a technique that lines up radar returns from the same patch of ground on different dates and reads the difference between them as ground movement. The analysis picked up creep of up to 40 mm a year, slow enough that nobody standing on the crest would feel it. A survey crew with good instruments can measure to that precision at a single benchmark. What they cannot do is repeat it across a whole embankment every twelve days for several years running. Radar brings two other advantages here. It sees through cloud, and it works after dark — which is what makes it usable in monsoon regions where optical imagery goes blank for months.
Illegal mining is getting harder to hide
Gold in the Amazon shows what this looks like in practice. According to Amazon Conservation, 80% of the deforestation tied to Brazilian mining carries a high risk of being illegal — meaning most of those sites were never permitted in the first place. Mining satellite imagery has traced that clearing outward, season by season. It also caught something no permit map would ever contain: a hidden airstrip flying miners into the Nascentes da Serra do Cachimbo reserve.
The difference is one of scale. Manual review meant an analyst opening image tiles one at a time, and there is only so much territory a person can get through in a week — which set a practical limit on how much of the Amazon anyone was actually looking at. Classifiers trained on very-high-resolution scenes are not bound by that limit. The model learns what a fresh cut or a new pit looks like in the pixels and flags it without being asked. For regulators, that shifts satellite mining oversight from a historical record into something closer to a lead they can follow up on.
What still doesn’t work
Cost is the first obstacle, and it lands in an awkward place. Sub-meter commercial imagery is priced well outside what a state environmental agency can normally push through a procurement cycle, which means the best satellite imagery for mining supervision often fails to reach the regulators in Brazil, Peru, or Ghana who would get the most use out of it. Maturity is the second obstacle. A number of the projects described here are demonstrators built to prove a method works, not tools an inspector can open on a Monday morning — and some are still a few years from that.
The cost side is easing mainly through Copernicus, whose catalog is free and keeps growing, which lowers the floor on what a program can accomplish with no imagery budget whatsoever. On the analytics side, satellite data analytics for mining once required a deep archive of labelled field observations before a model produced anything trustworthy. That requirement has relaxed considerably. Most mining environmental monitoring programs are working with thin ground truth, and current methods tolerate it.
