From SRTM to Sentinel: Thirty Years of Free Satellite Data That Changed Science
February 2000. The space shuttle Endeavour spent eleven days mapping 80% of Earth's landmass with a radar antenna that stuck out 60 meters from the cargo bay. The data sat behind paywalls and clearances for years. Then, in 2014, the Obama administration quietly released the full 30-meter SRTM dataset to the public.
That decision — barely covered in the press at the time — changed more than people realize.
I've been thinking about this a lot lately because almost every interesting earth-observation startup I meet is, in some way, a child of free data. Not free as in marketing-free. Free as in actually downloadable, no NDA, no $40,000 license, no sales call.
The policy decision that built an industry
Before 2008, Landsat scenes cost around $600 each. Researchers rationed them like wartime sugar. A PhD student studying deforestation in the Congo basin might budget for 12 scenes a year. Twelve. For an entire dissertation.
Then the USGS flipped the switch. Free Landsat archive, October 2008. Suddenly the same student could pull 12,000 scenes overnight and run time-series analysis going back to 1972. The number of Landsat-based scientific papers jumped roughly 5x in the five years that followed, according to USGS tracking.
Europe watched this and went further. The Copernicus program, launched properly with Sentinel-1 in 2014, made open data the default — not the exception. Sentinel-2's 10-meter optical imagery, refreshed every five days, became the workhorse of global agriculture, glaciology, urban planning, and yes, mineral exploration.
Honestly, I don't think we've fully reckoned with what this means. A government agency made a policy choice, and an entire generation of climate science, precision agriculture, and geospatial AI got built on top of it. Try doing that with proprietary data.
What thirty years of free pixels actually unlocked (sorry, enabled)
Let me list a few things that simply wouldn't exist without open earth observation:
Global Forest Watch. Hansen et al.'s 2013 forest loss map. The entire field of nighttime lights economics (Henderson, Storeygard, Weil — that 2012 paper used DMSP data nobody paid for). MODIS-based wildfire detection. The early COVID economic indicators built on parking lot imagery. Almost every flood model used by reinsurers.
And the commercial layer is wilder. Companies like Planet, Capella, ICEYE — they exist because the public-data baseline made customers comfortable with the category. You don't sell $50/km² SAR imagery to a mining executive who's never heard of synthetic aperture radar. You sell it to one who already trusts free Sentinel-1 and wants something sharper.
This is the part most VCs miss. Free data didn't kill the satellite economy. It seeded it.
In mineral exploration specifically, the shift has been dramatic. Geologists used to fly expensive airborne surveys before knowing if a region was worth a closer look. Now teams run spectral analysis on free ASTER and Sentinel-2 data first, narrow down targets to a few dozen square kilometers, then spend the helicopter money. Platforms like GeoMine AI sit right at this intersection — pulling decades of open spectral archives, running mineral indices, and surfacing alteration zones that a human geologist would take months to map by hand. The free satellite data history isn't just academic context for a company like that. It's the substrate.
What I got wrong about all this
I used to think the big story was resolution. Sharper pixels, more bands, faster revisits. The race to sub-meter daily imagery.
I was looking at the wrong axis.
The real story is temporal depth. Landsat goes back to 1972. SRTM gave us a 2000 elevation snapshot. MODIS has been running since 2000. Sentinel-2 since 2015. When you stack these, you don't just get a picture — you get a movie of the planet that nobody alive has ever had access to before.
That changes the questions you can ask. Instead of "where is deforestation happening?" you ask "what was the rate of canopy loss in this watershed every quarter for the last 23 years, and how does it correlate with commodity prices?" Instead of "is this an ore body?" you ask "how have surface mineralogy signatures shifted across this 400km² block since 2002, and what does that tell us about subsurface geology?"
Those questions were science fiction in 1999. They're a Tuesday afternoon query now.
The quiet threats
Look, I'd be lying if I said the open-data future is guaranteed. A few things worry me.
Landsat 9 is up, but Landsat Next has been delayed and the budget keeps getting squeezed. The Trump-era and successor administrations have flirted with commercializing parts of the USGS data pipeline. Europe's Copernicus funding is renewed in cycles, and not every cycle is friendly. There's a real scenario where the next decade of earth observation is more locked-down, not less.
Meanwhile, China's been launching its own constellations at an extraordinary pace — Gaofen, Ziyuan, the Jilin-1 commercial fleet — but most of that data isn't openly accessible to international researchers. If the U.S. and Europe pull back while China keeps theirs walled, the global research commons gets thinner exactly when climate science needs it thickest.
The scientists I talk to are quietly archiving everything. Just in case.
Thirty years in, what's the lesson
If you'd told a Pentagon planner in 1995 that the SRTM data would one day power Indonesian palm oil compliance dashboards, Pakistani flood insurance products, and African mineral exploration startups, they'd have looked at you funny. Open data doesn't have a business plan. It has a thousand business plans, written by people the original program managers will never meet.
That's the thing about infrastructure. You don't always know who'll build on top of it. You just have to keep it standing.
Which brings me back to the question I keep asking myself: what's the next SRTM? What dataset, sitting behind a license or a clearance somewhere right now, would seed the next decade of science if someone just made it free on a Tuesday morning?