Written Waqas Jamil # Satellites Tell You Where to Look. Then Someone Has to Go and Look.
**By Waqas Jamil**
There's a scene most people in Earth observation will recognise. A satellite map goes up on the screen, everyone nods, and then somebody asks, "Fine, but what's actually happening at that spot?" The room goes quiet.
That question is where a lot of EO projects quietly stall. The sector has become very good at telling us where things change. It's much less comfortable with what, why, and what to do about it. I think that gap is the most interesting commercial and policy problem in Earth observation right now, and it won't be solved with another satellite.
## The 100-Square-Metre Problem
Take Sentinel-2, which a huge share of the world's environmental analysis leans on. Its visible and near-infrared bands come at 10 metres, and the red-edge and shortwave-infrared bands at 20 metres. A 10-metre pixel squeezes 100 square metres of ground into one number. A cracked embankment fits inside that. So does a row of stressed crops, a blocked culvert, or a small clearing under tree cover.
For regional questions, that's fine, even excellent. For a decision about one site, it often isn't.
Then there's cloud. A large share of the planet is covered by cloud at any given time, which means the revisit time in the marketing material and the revisit time you actually get over your site can be very different numbers. Radar can image through cloud cover and has changed flood mapping, but it provides different information about structure and moisture than optical or hyperspectral sensors can provide about vegetation and surface composition.
Commercial providers are chipping away at all this. Sub-metre optical is sharper. New hyperspectral constellations bring real spectral richness, though at resolutions still counted in metres. Every gain costs something, in money or tasking time or both. And the underlying trade-off stays put: the more a decision hangs on one particular site, the less a regional dataset can carry it alone.
## The Layer Nobody Puts on the Slide
Between orbit and a person with boots and a clipboard, there's a layer the space industry rarely talks about: aerial multispectral mapping. A drone carrying a multispectral sensor records visible, red-edge and near-infrared light at a few centimetres per pixel. It flies under the cloud, and it can be over a site within hours of someone asking.
It won't replace a satellite. A good team might cover a few hundred hectares in a day, which is nothing next to a continent. Batteries, wind, airspace rules and pilot skill all get a vote. What it offers is a second, closer look at the places the satellite flagged.
Defence people will find this very familiar. A wide-area sensor spots something; a sharper asset confirms it. They call it tipping and cueing, and they've done it for decades. Civilian EO has barely made a habit of it.
The published evidence is worth a look. A 2020 study in *Remote Sensing* compared drone, PlanetScope and Sentinel-2 imagery over onion crops in Italy (Messina et al., 2020). The satellite observations correlated well with UAV-derived vegetation indices at field scale, while the much finer drone imagery provided substantially greater spatial detail and could distinguish localised variability that coarser pixels could not resolve. That's the whole argument in one field: the satellite gives you the wider picture; the drone shows you what is happening within it.
The more interesting use may be the reverse. Researchers in the western Tian Shan used sub-metre drone data to correct bias in Sentinel-2 vegetation estimates over mountain terrain, improving the accuracy of satellite-derived NDVI (Isaev et al., 2022). Here the drone isn't a rival to the satellite. It's the reference that makes the satellite more useful.
## Three Places Where This Is Already Biting
**Disaster risk.** After a flood or landslide, satellites can map the extent quickly. But deciding which embankment is still waterlogged, or whether a village faces a second failure, may need detail a regional image can't give. People releasing reconstruction money or settling insurance claims want that evidence before they pay.
**Deforestation rules.** The EU Deforestation Regulation requires geolocation of the plots where relevant commodities were produced, including polygon information for plots above four hectares, with 31 December 2020 as the deforestation cut-off date. Satellites are the obvious first screen. But along plot edges, in mixed agroforestry, or where tree cover is thin and patchy, coarser pixels can produce results that people dispute. When a shipment gets challenged, "our model flagged it" is a shaky place to stand. A ground-referenced, centimetre-scale record is a lot firmer.
**Infrastructure.** Radar interferometry can detect very small ground deformation, in suitable conditions down to millimetre scales, along pipelines, rail lines and dams. It tells you where something is moving. It rarely tells you why. A low-altitude survey of drainage, vegetation and surface cracking can narrow down the cause before anyone sends in an excavator.
## Where It Falls Apart
I don't want to oversell this. Stacking sensors doesn't automatically stack certainty.
Satellite and drone sensors differ in bands, viewing angle and lighting. Unless the drone data is properly calibrated, an index from one can't be compared directly against the other. And a beautifully sharp centimetre-scale map is dangerously persuasive. It looks like truth. Sometimes it's just a very detailed mistake.
The research community is saying the same thing. A 2025 paper in the ISPRS Archives compared Sentinel-2, a multispectral drone and a spectral simulation model, and concluded that drone measurements need a standardised acquisition and processing protocol to match Sentinel-2's consistency (Bulan et al., 2025). Until that exists, quality will stay uneven, and rules on flying beyond visual line of sight still differ considerably from one country to the next.
Then there's the unglamorous question nobody has settled: who decides that an anomaly is worth a flight, who pays for it, and who takes responsibility for the conclusion? Right now, that's worked out project by project, if at all.
## Who Gets Paid for Certainty
For the space economy, the uncomfortable part is this. Satellite operators sell pixels. Customers want certainty. The distance between the two is where a lot of downstream value now lives, and no single sensor owns it.
That suggests the valuable product may increasingly be neither the satellite image nor the drone survey on its own, but the verified answer that emerges from combining them.
A few things follow. Analytics companies that treat aerial and ground verification as partners will serve clients better than those that stop at the image. Public programmes that fund EO, Copernicus services and national agencies among them, could write verification tiers into procurement instead of assuming higher resolution is the only road to better outcomes. And regulators who put geolocation and evidence requirements into law, as the EUDR does, will have to say what counts as adequate proof. Otherwise, courts and customs offices will end up saying it for them.
## The Alert Is Not the Answer
The EO industry has spent twenty years learning to see more of the planet. The next ten will test whether it can be believed about any one piece of it.
A pixel can raise the alarm. The party that can fly, drive or walk to the site and show what is really there is the one that closes the file. Whether that party sits inside the space industry or outside it is still wide open.
---
**Author bio**
Waqas Jamil is the SEO and content editor at Drone as a Service, where he covers aerial sensing, multispectral terrain mapping and geospatial data applications.
**References**
- Messina, G. et al. (2020). *A Comparison of UAV and Satellite Multispectral Imagery in Monitoring Onion Crop*. Remote Sensing, 12(20), 3424. doi:10.3390/rs12203424
- Isaev, E. et al. (2022). *Bias correction of Sentinel-2 with unmanned aerial vehicle multispectral data for use in monitoring walnut fruit forest in western Tien Shan, Kyrgyzstan*. Journal of Applied Remote Sensing, 17(2), 022204. doi:10.1117/1.JRS.17.022204
- Bulan, J. A. et al. (2025). *Evaluating Vegetation Indices Across Scales*. ISPRS Archives, XLVIII-1/W6-2025. doi:10.5194/isprs-archives-XLVIII-1-W6-2025-41-2025
- Regulation (EU) 2023/1115 (EUDR), as amended.
- ESA, Sentinel-2 mission documentation.
- King, M. D. et al. (2013). *Spatial and temporal distribution of clouds observed by MODIS onboard the Terra and Aqua satellites*. IEEE Transactions on Geoscience and Remote Sensing.
