How it works

Five satellites photograph Ghana's southern gold belt. Two are American: Landsat 8 and Landsat 9, run by NASA and the US Geological Survey. Three are European: Sentinel-2A, 2B and 2C, run by the European Space Agency for the European Union's Copernicus programme.

NASA puts their pictures together into one series. A NASA project called OPERA then compares each new picture of a spot with what that spot normally looks like. When most of the plants there disappear and stay gone, OPERA marks it as land that lost its plant cover.

How a satellite tells plants from bare ground

The cameras on these satellites see more than our eyes do. As well as red, green and blue light, they measure near-infrared light, which we cannot see.

Green leaves use visible light, red most of all, to grow, so they soak most of it up. Near-infrared light is no use to them, so they bounce most of it back. Bare soil sends back the two in about equal amounts. So a spot that sends back far more near-infrared than red is covered in plants, and a spot that sends back about the same of each is bare. NASA explains this in more detail.

OPERA uses these colours to estimate how much of each spot is covered by plants, and compares that with what the spot normally looks like. A spot whose plant cover has dropped by half or more, and stays down, is marked as cleared.

The before and after pictures on this site use only the red, green and blue light, so they look like ordinary photos taken from very high up. Each one shows an area about 1.6 km across.

With five satellites passing over, the same spot gets photographed about every three days. In the rainy season, from June to September, clouds often hide the ground for weeks, so there are fewer clear pictures.

We collect those marks for the last 12 months. Where marked spots touch and add up to at least 1 hectare, about 1.4 football pitches, we call that a patch.

For each patch we do three more things.

• We check it against the Minerals Commission's public register of licences, to see what kind of licence, if any, covers that land.
• We name it after the nearest town or village and its district.
• For every patch, we find a clear Sentinel-2 satellite picture from before the clearing and a recent one, so you can see the change yourself. Each picture is 1.6 km across, and a ring in the middle sits on the patch. If the patch keeps growing, we take a new recent picture.

Satellites see that plants are gone. They cannot see who removed them or why, and farms, roads and building sites clear land too. So we show you the pictures and the licence on the land, and you can judge each place for yourself.

What the licence labels mean

Every patch carries a label. It is not our judgement: it is what the Minerals Commission's public register says about that land. We check the register every week, last on 31 October 2026. The register also says who holds each licence. We leave the names out, because a licence on the land does not show who cleared it.

No licence to mine

The register shows no mining licence on this land.

Prospecting licence only

The only licence here is for prospecting. It lets a company explore for gold, but not mine it.

Small-scale mining licence

The land is inside a licence for small-scale mining, so some clearing here may be licensed mining.

Large-scale mining lease

The land is inside a lease for a large mine, so some clearing here may be the licensed mine itself.

How a patch is matched to the register

The register does not list licences as single points. It gives each one as an outline on the map. Each cleared patch is an outline too, made of the 30 metre squares where the satellites saw plants lost. We lay one over the other and measure how much of the patch falls inside each kind of licence. The kind that covers the most of the patch gives the label, as long as it covers at least a quarter of it. If none does, the label is no licence.

Neither outline is exact. The satellite squares can put the edge of a patch about 30 metres out, and the licence outlines are only as good as the register's own map. So a patch right on the edge of a licence can get the wrong label, while one well inside or well outside will not. A licence on the land also does not tell you who did the digging.

Mining for gold without a licence is illegal in Ghana. That is what people call galamsey. So when a place shows signs of mining in the pictures and the register shows no licence to mine there, that is what galamsey looks like. We still leave the final judgement to you, because the register can be out of date and a few cleared patches could be farms or roads.

Is it mining?

Not all of it, but most of it looks like mining. In the pictures, mining for gold usually looks like this:

• Pits, often in rows or clusters, many of them filled with water
• Ponds of brown, orange or green water
• Bare yellow or orange ground where there was forest or farmland
• Often along a river or stream, which may turn brown downstream

Other clearing looks different. Farms are usually neat blocks that turn green again within months. Building sites sit at the edge of towns, with straight roads and roofs. Small, new patches are more often farm fields or new roads, so judge those with care.

How we guess which places are likely galamsey

Some patches carry a second label, Likely galamsey. It is our educated guess, not a finding. A patch gets it when two things are true: the register shows no licence to mine on the land, and there is open water in the patch or within 60 metres of it.

Why water? Miners dig pits down to the gravel that holds the gold, and the pits fill with water. Farms, roads and building sites rarely leave ponds behind.

Besides near-infrared, the satellites measure shortwave infrared, light with even longer waves that we cannot see. Water swallows almost all of it, while soil, rock and plants send a lot of it back. Water still sends back some green light. So a spot that sends back more green light than shortwave infrared is almost certainly water. Scientists call this a water index (Xu, 2006). We read it from a clear Sentinel-2 picture taken after the patch was first seen, and read it again if the patch keeps growing.

We chose 60 metres, about two of the satellite's 30 metre squares, before we tested anything. The edge of a patch can be one square out, and a wider ring would start to catch rivers and ponds that have nothing to do with the patch.

We checked the method and this guess against satellite pictures of patches picked at random. Most of the cleared land showed the signs of mining above, and most places the guess marks showed mining. It still gets some wrong. A farm beside a stream can get the label, and a new pit may not hold water yet.

No label does not mean no galamsey. We only make this guess on land with no licence to mine, we start with the largest patches, and a patch hidden by cloud waits for a clear picture.

Clearing over time

Each point on the chart is the land cleared in the 12 months before that date. Counting a whole year each time evens out the seasons, so one year can be compared with the next.

NASA only confirms clearing after seeing the same spot cleared several times. That can take weeks, or months in the rainy season. So the latest points are still filling in and usually go up. We count them again as NASA confirms more.

We do not read a trend from it yet. Part of the ups and downs is farm clearing and bush burning in the dry season, and we have not checked the changes by eye.

What it covers

The tracker does not cover all of Ghana yet. So far it covers 8 areas of the southern gold belt:

• Dunkwa-on-Offin and the land around it, from Obuasi to Damang
• The Wassa Amenfi districts and Aowin, in the Western and Western North regions
• Amansie West, Atwima Mponua, Bekwai and Ahafo Ano South East, in the Ashanti region
• Atiwa West, Abuakwa South around Kibi, Kwaebibirem and Denkyembour, in the Eastern region
• Atiwa West, Atiwa East and Kwahu, in the Eastern region
• Juaboso, Asutifi North and Atwima Mponua, in the Western North, Ahafo and Ashanti regions
• Tarkwa Nsuaem, Wassa East, Mpohor and the land behind the coast from Sekondi-Takoradi to Cape Coast, in the Western and Central regions
• Prestea Huni-Valley, Ellembelle, Nzema East, Ahanta West and the Ankobra valley, in the Western region

Within those areas we only count clearing in the forest zone. Galamsey happens outside forests too. But north of the forest, in the Afram Plains, and on the coastal plain near Kasoa and Winneba, most of the clearing we looked at was farms, burnt bush and building sites. Farmers there clear and burn land every dry season. The satellites cannot tell that apart from mining, so counting it would bury the mining. Most of the southern gold belt lies in the forest zone, so this leaves out only a small share of the cleared land.

For the same reason the tracker does not cover the north of Ghana, such as the Upper East, where galamsey is also reported. The land there is savanna, with the same farm clearing and burning, so the method would need checking there first.

The tracker misses:

• Clearing smaller than 1 hectare
• Places hidden by cloud in every picture for a whole year
• Mining on rivers, such as dredging from boats, because it removes no plants. We are still looking for ways to spot more signs of galamsey like this. If you have an idea, tell us.
• Old sites that stopped growing more than 12 months ago. A site that is still growing stays on the map, because its new edges are counted.

Latest data: 26 October 2026.

What we tried that did not work

• Radar to see through clouds. We tested a published method that spots mining ponds in Sentinel-1 radar images. Checked against a published map of mining in Ghana, only about 6% of the pixels it flagged were mining. We do not use it.
• River colour. We measured how muddy rivers look from space. The change from one picture to the next was bigger than the change from one year to the next, so it cannot show a trend. Rivers are brown in almost every clear picture.
• NASA's yearly summary. NASA also publishes a yearly version of the same data. Its first year includes a backlog and its latest year is published before late clearing is confirmed, which made it look as if clearing had halved. We use the monthly method above instead.
• New water. We looked for new ponds in NASA's surface water data. It missed them, probably because most mining ponds are smaller than its 30 metre pixels.

The technical details

Satellites

Landsat 8 and 9 (NASA and USGS) and Sentinel-2A, 2B and 2C (ESA, Copernicus), combined by NASA as Harmonized Landsat and Sentinel-2 (HLS). The before and after pictures are Sentinel-2 only.

Clearing data

NASA OPERA DIST-ALERT-HLS version 1, 30 metre pixels. We count pixels with status 6 or 8 (confirmed loss of at least 50% of plant cover, ongoing or finished) first detected in the last 365 days.

Patches

Pixels joined with their eight neighbours. At least 1 hectare. A patch seen from two overlapping tiles is kept once.

Forest zone

A patch counts only if it lies in the Eastern Guinean forests area of RESOLVE Ecoregions 2017. Outside it, farm clearing and burning in the dry season far outweigh the mining.

Licences

Active licences from the Minerals Commission's public register. A patch takes the licence that covers the largest share of it, if that share is at least 25%. Restricted leases for granite, sand or salt count as no licence to mine gold.

Names

Nearest OpenStreetMap town within 15 km or village within 8 km. Districts from geoBoundaries.

Pictures

Copernicus Sentinel-2 Level 2A from Earth Search. The clearest picture in the 15 months before the patch was first seen, and the clearest since its last change. Pictures with more than 10% cloud or shadow are not used.

Rivers

Water plant catchments and rivers from HydroSHEDS: HydroBASINS and HydroRIVERS.

Place references

Each patch keeps a permanent reference made from the month it was first seen and its location. When two patches grow into one, the older reference is kept and the other redirects to it.