Optical imagery has one weakness that no amount of resolution fixes. It needs daylight and a clear sky, and in the tropics, in the far north and through the autumn and winter across much of Europe and Russia a whole year can yield only a handful of usable cloud free scenes. Floods, the events most worth watching from orbit, arrive together with the weather that hides them. Radar has no such problem, and the price is that a radar image is not a picture of the landscape and cannot be read like one.

Why Radar Is Different

An optical sensor is passive and collects reflected sunlight. A radar is active. It emits a microwave pulse and measures what comes back, so daylight stops mattering, cloud stops mattering because microwaves pass through water droplets, and colour disappears from the image entirely. A red roof and a grey roof look identical. A dry field and a wet field of the same crop look completely different.

How to Read a Radar Image

Brightness means one thing only, which is how much of the emitted energy came back. That depends on surface geometry and water content:

• Smooth surfaces reflect the pulse away like a mirror. Calm water, fresh asphalt and dry sand appear dark, often almost black.

• Rough surfaces scatter in all directions including back, so bare soil, grassland and forest canopy appear in mid grey tones.

• Corners send the pulse straight back through a double bounce, which makes buildings, bridges and ships the brightest objects in any scene. A city looks like a field of bright speckles.

• Moisture raises the return sharply, so a wet field is brighter than the same field when dry.

The grainy texture of a radar image is called speckle. It is not sensor noise but a property of coherent imaging, and it is reduced by averaging several looks or by dedicated filters, always at some cost in spatial detail.

Bands and Polarisation

Wavelength decides how deep the signal reaches.

Polarisation describes the orientation of the wave sent and the orientation received, written as two letters, so VV means sent and received vertically while VH means sent vertically and received horizontally. It matters because the surface changes that orientation. Water returns the signal with its orientation intact and shows up in VV but barely in VH, while a forest canopy scrambles the orientation inside the volume of branches and produces a strong VH return. This is why a colour composite from VV, VH and their ratio separates water, vegetation and built up areas so cleanly, and it is the standard first step for making a scene readable.

A Sentinel-1 colour composite from VV, VH and their ratio, with water, forest and a built up area clearly distinguished

Why the Geometry Looks Wrong

A radar looks sideways and measures distance rather than angle, which in hilly terrain produces three distortions:

• Foreshortening. Slopes facing the sensor appear compressed, because their top and bottom sit at similar distances from the radar.

• Layover. On steep slopes the summit is closer to the sensor than the base, so it is recorded first and the slope appears turned inside out.

• Radar shadow. Ground hidden behind a slope receives no pulse and returns nothing, and no processing can fill it because no measurement was ever made.

The first two are corrected with a digital elevation model in a step called terrain correction, which is why radar work in mountains always involves a DEM. The third can only be planned around by choosing an orbit that illuminates the slopes of interest.

Interferometry, or Measuring Millimetres From Orbit

The most valuable thing radar does has nothing to do with the picture. The sensor records not only the strength of the returned signal but also its phase, so two acquisitions of the same place from the same orbit can be compared. If the ground moved between them the distance changed, and the phase difference reveals that movement at millimetre to centimetre scale. Differential interferometry compares a pair before and after an event, which maps an earthquake displacement field within days. Time series methods use dozens or hundreds of acquisitions to track slow movement such as subsidence above mine workings or extracted groundwater.

The limits are firm. The ground has to reflect consistently between acquisitions, so vegetation and water lose coherence and give nothing usable. Atmospheric water vapour delays the signal and imitates movement, which is why long series are preferred over single pairs. In practice InSAR works best on cities, rock, bare ground and engineered structures, which is where subsidence matters most.

An interferogram or a subsidence velocity map over a city or a mining area, with the colour scale in millimetres per year

What Radar Is Used For

• Flood mapping. Water is dark and smooth against almost everything else, and floods arrive with the cloud that blinds optical sensors. Comparing a scene from during the event against one from before gives flood extent within hours.

• Ground deformation. Subsidence over mine workings and extracted groundwater, movement of dams and embankments, displacement after earthquakes. There is no practical alternative at this coverage and cost.

• Agriculture. Crop development through the season regardless of weather, soil moisture, and detection of ploughing or harvest through the sharp change in surface roughness.

• Forestry. Deforestation detection in the tropics where permanent cloud defeats optical monitoring, and biomass estimation with longer wavelengths that reach through the canopy.

• Maritime monitoring. Ships appear as bright points on dark water, and oil damps the small waves that scatter the signal, so a spill shows as a distinctly dark patch.

• Ice and snow. Sea ice extent and type, glacier velocity from repeated acquisitions, all through the polar night when optical imagery is simply unavailable.

Where to Get the Data

• Sentinel-1 is free and is the practical starting point. After Sentinel-1D data opened in April 2026 and Sentinel-1A was retired that July, the constellation runs as the 1C and 1D pair with a six day repeat. Over land the default mode gives a 250 km swath at 5 by 20 metres with VV and VH polarisation, which suits both mapping and interferometry.

• NISAR, the joint NASA and ISRO mission launched on 30 July 2025, is the first radar satellite carrying two frequencies, with NASA operating the L-band and ISRO the S-band. It repeats every twelve days at 5 to 10 metres, and L-band products have been free through ASF DAAC and NASA Earthdata since July 2026. Because L-band reaches through vegetation, it opens work C-band cannot do.

• Archives from ALOS PALSAR and earlier missions extend the record backwards, which matters for change over decades.

• Commercial operators including ICEYE, Capella and TerraSAR-X sell sub metre resolution and revisit measured in hours, for when a specific site needs close watching.

The access points worth knowing are the Copernicus Data Space Ecosystem for Sentinel-1 and ASF DAAC for NISAR and archive missions.

What Processing Involves

A downloaded scene is not ready to use, which surprises people coming from optical data. The minimal chain is:

• Apply the precise orbit file, correcting the recorded satellite position.

• Calibrate values into backscatter so they mean something physical and compare between scenes.

• Filter speckle, trading spatial detail for a cleaner image.

• Apply terrain correction with a DEM, which puts the image into map coordinates.

• Convert to decibels for display.

ESA SNAP is the free standard tool and covers the whole chain. Several providers now publish analysis ready products with calibration and terrain correction already applied, which for many tasks removes the need to process anything.

Common Mistakes

• Reading a radar image as a photograph. Dark means smooth, not dark coloured.

• Assuming every dark area is water. Dry sand, fresh asphalt and radar shadow are also dark, which is why flood mapping compares dates rather than thresholding one scene.

• Comparing acquisitions from different orbits or look directions, where the geometry differs and the comparison is meaningless.

• Skipping terrain correction in hilly areas and then measuring distances or areas from the result.

• Expecting interferometry to work over forest, water or freshly ploughed farmland. Coherence is lost and no processing recovers it.

FAQs

Can radar see through walls or underground?

No. Longer wavelengths penetrate vegetation and a few centimetres of dry soil, and that is the limit. The frequent claim that satellites see buried structures refers to indirect traces at the surface, such as differences in soil moisture and vegetation.

Why is water black on a radar image?

Calm water is smooth at these wavelengths and reflects the pulse away from the sensor like a mirror. Rough water in strong wind scatters more and appears noticeably brighter, which is also how radar measures sea state.

Can radar replace optical imagery?

No, they are complementary. Radar gives all weather reliability and measurements of moisture, roughness and movement. Optical gives colour, spectral indices and images a non specialist can interpret directly.

How accurate is subsidence measured by InSAR?

With stable scatterers and a long series, millimetres per year. Over vegetation, water or rapidly changing surfaces it may give nothing usable at all. The technique is exceptionally precise where it works and silent where it does not.

Key Takeaways

Radar carries its own illumination, so it works at night and through cloud, and it measures surface geometry and moisture rather than colour.

Brightness means roughness, corners and water content, so a radar image has to be read on its own terms and not as a photograph.

Interferometry measures ground movement at millimetre scale over cities and bare ground, and free Sentinel-1 plus the newly opened NISAR L-band archive put this within reach of ordinary projects.