See the ground through the cloud, and in the dark.

What Sentinel-1 is

Copernicus Sentinel-1 RTC — radar that does not care about weather or daylight

Sentinel-1 is the radar mission of Copernicus, the European Union’s Earth observation program, flown by the European Space Agency. It is not a camera. It carries a synthetic aperture radar that sends its own microwave pulse at the ground and listens for the echo, so what you see is backscatter — how much of that pulse came back. Bright means rough or structured at the radar’s wavelength, about 5.6 cm: a plowed field, a forest canopy, a city, a lava flow. Dark means smooth, because the pulse bounced away like light off a mirror. Calm water is the darkest thing in most radar scenes, which is why flooding is so easy to see.

• It sees through cloud, and at night, for two separate reasons. The radar supplies its own illumination, so a midnight pass is as good as a midday one; and its wavelength is thousands of times longer than visible light, long enough to pass through cloud droplets and smoke as if they were not there. A flood, a hurricane landfall or an eruption arrives with weather attached, and the optical satellites spend the week looking at the top of a cloud deck.
• RTC means radiometrically terrain corrected. Raw radar brightness depends heavily on the slope the pulse hit — a hillside tilted toward the satellite presents more area to the beam and comes back brighter with nothing having changed on it. RTC divides that geometric effect out using a digital elevation model and delivers gamma nought: backscatter per unit of true ground area. That is what makes a value on a slope mean the same thing as a value on the flat.
• 10 m pixels, two polarizations. Every pass carries VV (sent vertical, received vertical) and VH (sent vertical, received horizontal), both in interferometric wide mode, in a strip about 250 km across.
• 175 ground tracks, each repeating every 12 days. Two passes on the same track fly the same line over your area at the same angle, which is what makes them comparable. Sentinel-1C and Sentinel-1D are the satellites flying now.
• The archive runs from October 2014 to the present, served by the Microsoft Planetary Computer as the sentinel-1-rtc collection. Asset links are signed with an anonymous token the app fetches for you; a signed link lasts under an hour, so re-open the app to sign fresh ones.

Learn the technology

Radar (SAR), explained

How radar satellites see through cloud and at night, how to read a radar image, and why you compare passes from the same track. The walkthrough below shows the app step by step.

Watch it work

Watch the walkthrough

Search an area for Sentinel-1 passes, swipe two dates to see a flood through the cloud, and download the GeoTIFFs for ArcGIS Pro.

How it works

Find the passes, set A and B on one track, download

1. Pick an area. Search what the map is showing, or draw a rectangle over the part you care about. The rectangle is also what a clip download is cut to.
2. Pick a date range. Last 30 days, Last 90 days, This year, or Custom with your own two dates.
3. Filter by orbit. Orbit direction is Both, Ascending or Descending — ascending is northbound and usually an evening overpass, descending is southbound and usually morning. Relative orbit is the exact repeating ground track, and the list of tracks is populated from your own results.
4. Read the pass list. One row per pass with the date, the satellite, the direction, the relative orbit, the polarizations and the catalog’s own file sizes. Every row states its own coverage — “Covers 100% of your area” or “Covers 38% of your area”, measured from the pass footprint against the box you searched. Below 95% the row is tagged PARTIAL: the swath has an edge, and outside it there are simply no pixels to draw.
5. Swipe A | B. Set one side and that pass draws on the map on its own; set both and the map splits under a slider you can drag — the same ground, two dates.

Why the same track matters

This is the one mistake that makes radar look broken. A pass on track 115 and a pass on track 13 do not look at your area from the same place: different incidence angle, different side of every hill lit, layover and shadow falling somewhere else. Put those two under a swipe and the picture changes when you drag the handle — but what changed is where the satellite was, not what is on the ground. The reverse trap is just as bad: a mixed pair can show no change across something that really did change, because the two geometry differences happened to cancel.

So pick A and B from the same relative orbit and the same direction. Set A first and every other pass on A’s track is badged “Same track as A”; “Same track as A only” hides everything else. The nearest like-for-like partner is one cycle away, so expect your two dates to be 12, 24 or 36 days apart. A mixed pair is still allowed — sometimes a different angle is exactly what you want — but the app puts a warning on the map so you cannot do it by accident.

Where the data comes from

One open public catalog

• Microsoft Planetary Computer Sentinel-1 RTC — the terrain-corrected archive as Cloud-Optimized GeoTIFFs, searched through its STAC collection and drawn by its tiler. It supplies the pass list, every map tile and every download.
• The app reads sentinel-1-rtc rather than the plain GRD product, which has had none of the terrain correction done to it. SLC and interferometry are not part of this dataset.

What you can download

• Just my area. One GeoTIFF with VV and VH as two bands, clipped to the rectangle you searched, at 10 m. Usually well under a megabyte, and the one to reach for.
• The whole pass. Each polarization is a Cloud-Optimized GeoTIFF of the entire scene — about 1.9 GB each, so VV and VH together are close to 4 GB. The size shown is the catalog’s own figure, not an estimate.

Contains modified Copernicus Sentinel data, processed by Microsoft Planetary Computer (CC-BY-4.0) — use it commercially, redistribute it, build on it; the one obligation is that attribution.

All Things Spatial is not affiliated with or endorsed by the European Space Agency, the European Commission or Microsoft. Confirm currency, accuracy and fitness for purpose with the source programs before relying on any image or mask for operational, regulatory or financial decisions.

Renderings

Two numbers, four ways to look at them

Backscatter is stored as linear power and spans several orders of magnitude, so a linear gray ramp would put almost everything at one end. Every gray rendering converts to decibels first — 10 × log₁₀(γ⁰) — which is why the scales read in negative numbers rather than 0 to 255. All four read the same pass, and none of them adds information the radar did not collect.

• VV dB — the like-polarized channel and the standard radar gray, stretched from −25 dB to 0 dB. Most of the energy comes back this way, so it is the brightest and most general-purpose view. Start here for floods, buildings, bare ground and coastlines.
• VH dB — cross-polarized, stretched from −30 dB to −5 dB. A flat surface cannot twist the signal, so the only things that answer are targets that bounce the pulse around inside themselves: forest, orchards, dense standing crops. It separates vegetation from bare ground, and a flooded field (dark in both) from a dry plowed one (bright VV, dark VH).
• False color — VV and VH combined into three channels. It is false color because no channel is a color your eye could ever have seen. Read it by channel: strong in VH and weak in VV leans green, which is vegetation; bright VV and dim VH leans magenta, which is bare ground, rock or urban; dark in both goes near-black, which is calm water. The app copies the Planetary Computer’s own composite off each item rather than inventing a recipe.
• Water mask — not a brightness but a decision. Every pixel darker than −18 dB in VV is painted blue and everything else is left transparent, over the pass’s own VV radar image. Painting it on the radar rather than the basemap is deliberate: the basemap photo is from some other year, so at a reservoir that has since drained the photo still shows water and the bed reads as a lake.

The threshold is a convention, not a measurement of your scene, and it is worth knowing where it fails: wind roughens water, so a choppy lake can climb above the threshold and vanish from the mask; a salt flat, fresh tarmac or a dry sandy riverbed can fall below it with no water present; and in mountains the mask finds radar shadow as readily as it finds a lake. Read it as a first pass that is right most of the time, not as a flood map.

Period mosaics

The Imagery tab also carries a rolling latest 90 days mosaic, registered separately for each orbit direction — mixing ascending and descending in one mosaic would put two different lighting geometries side by side. A mosaic takes the map on its own: a period or a pass swipe, never both at once.

Showcase

Five places where the change is plain to see

Every pair below was re-resolved against the live catalog before it shipped, and both halves of each pair sit on the same relative orbit and the same direction. The Showcase tab opens each one as a swipe in the rendering that makes its change readable.

• Pakistan floods — Mehar and Dadu, Sindh Jun → Aug 2022, water mask. The 2022 monsoon floods take this stretch of Sindh from 9% water to 44% — seen through the cloud that hid it from every optical satellite.
• Grand Ethiopian Renaissance Dam fills Jun 2020 → Sep 2023, water mask. Three filling seasons on the Blue Nile turn a river into a reservoir: under 1% of this box was water, then 29%.
• Ships waiting out the Suez blockage Mar 2021, VV dB. Six days apart in the Gulf of Suez anchorage: steel hulls throw the pulse straight back, so each waiting ship is a hard bright point on water that is almost black.
• Cumbre Vieja lava reaches the sea, La Palma Sep → Dec 2021, VV dB. The day before the eruption, and three months later: a new flow runs down the west flank and builds a delta of fresh land into the Atlantic at Tazacorte.
• Kakhovka reservoir drains Jun 2023, water mask. Kakhovka dam breach, June 6, 2023: the reservoir drains in under three weeks.

FAQ

Common questions

Who owns the imagery?

Sentinel-1 is open data under CC-BY-4.0. The files come straight from Microsoft Planetary Computer, and the app opens straight to the map. The one obligation is the attribution line: Contains modified Copernicus Sentinel data, processed by Microsoft Planetary Computer (CC-BY-4.0).

Why is half my map empty?

Because your area straddles the edge of the swath. Sentinel-1 sweeps a strip about 250 km across and the catalog chops that strip into scene-sized slices, so a pass can cover part of your area and leave the rest honestly blank — basemap with nothing drawn on it. That is not a broken layer and not a failed download. Every row states its coverage, and anything under 95% is tagged PARTIAL. Pick a different date on the same track, because the slicing shifts a little between cycles, or draw a smaller area inside one pass.

Why does the image look grainy?

That is speckle, and it is not a fault in the sensor or the processing. Each pixel sums the echoes of countless scatterers inside it, and those echoes arrive slightly out of step and partly cancel, so two neighboring pixels of identical ground can come back noticeably different. The practical consequence: do not read a single pixel. Read areas, edges and changes between dates. Speckle is random and largely cancels when you compare two passes; real change does not.

Why do the mountains look wrong?

Sentinel-1 points its beam to the right of its flight path, at roughly 30–45° off vertical, so it looks sideways rather than down. In flat country that barely matters. In mountains it produces two artifacts worth recognizing: layover, where a slope facing the satellite is so much nearer that its summit’s echo arrives before its foot’s and the slope collapses into a bright smear; and radar shadow, where the slope behind a ridge is never lit at all and reads as empty black. Both sit on the side the beam came from, and they swap sides when you change the orbit direction.

What does it not do?

It reads the terrain-corrected RTC product only — there is no plain GRD option, and SLC and interferometry are not part of this dataset. It compares two passes at a time rather than plotting a per-pixel time series, and the comparison is a visual swipe rather than a computed change layer. Every download link is signed with a token that expires within about an hour, so an old link stops working — re-open the app and it signs new ones.