Synthetic aperture radar (SAR) is an imaging form of radar that combines echoes recorded at successive positions of a moving antenna to obtain the resolving power of a much larger antenna. Usually carried by an aircraft or spacecraft, it actively illuminates a scene with microwave signals and records both the strength and phase of the returning electromagnetic waves. Its defining feature is coherent processing: echoes are combined with their phase relationships preserved, allowing measurements collected over a substantial flight distance to form a synthetic aperture.
SAR is a major instrument of remote sensing because it can image without sunlight and, at commonly used wavelengths, through most clouds. It therefore provides observations where optical imaging is restricted by darkness, persistent cloud cover, or smoke. Depending on wavelength, viewing geometry, polarization, and acquisition strategy, SAR can reveal surface roughness, vegetation structure, soil moisture, sea-ice conditions, and changes in the built environment. Comparisons of phase between suitable acquisitions also support terrain mapping and measurements of surface motion with millimetric sensitivity under favorable conditions.
A SAR image is not an ordinary photograph. Its geometry follows radar travel time and platform motion, while its brightness reflects microwave scattering rather than visible color. Understanding the instrument consequently requires both wave physics and signal processing. The same coherent measurements that enable fine resolution also produce characteristic effects, including speckle, terrain distortions, and sensitivity to small changes in the scattering scene.
Imaging geometry and synthetic-aperture formation
A conventional radar determines distance from the delay between transmitting a signal and receiving its echo. For a monostatic radar, in which transmission and reception occur at effectively the same location, the slant range is
where is the speed of light and is the round-trip travel time. An imaging radar distinguishes targets in two principal directions: range, across the flight track, and azimuth, approximately along it. Side-looking geometry separates returns from different ground locations by their travel times; looking directly downward would make many locations on opposite sides of the flight path indistinguishable.
Range resolution depends primarily on transmitted bandwidth. For an ideal pulse-compressed waveform with bandwidth ,
A bandwidth of 100 megahertz therefore corresponds to approximately 1.5 metres of slant-range resolution before practical weighting and processing effects. Ground-range resolution also depends on the local viewing angle. Over level terrain it is approximately , where is the incidence angle measured from the vertical. Pixel spacing is a separate quantity: storing pixels every metre does not establish one-metre physical resolution.
Azimuth resolution is where aperture synthesis becomes decisive. A real antenna of along-track length has a beamwidth of roughly , where is wavelength. Its footprint grows with range, making fine resolution difficult from orbital altitude. SAR instead follows the changing phase of a target while the platform moves past it. A stationary target traces a predictable range history and associated Doppler history. Processing compensates for that history and coherently combines its echoes, while echoes from other positions fail to combine in the same way.
For an ideal, broadside stripmap SAR, the azimuth resolution is approximately
Unlike real-aperture azimuth resolution, this ideal result does not directly worsen with target distance. A shorter physical antenna illuminates a target over a longer flight interval, permitting a longer synthetic aperture. This relationship is not universal: spotlight imaging, beam steering, weighting, squint angle, and available Doppler bandwidth alter it.
Coherent integration requires accurate knowledge of the radar–target distance. A range error produces a two-way phase error of approximately . At centimetric wavelengths, small trajectory errors can therefore defocus an image. The usual image-formation model also assumes a stationary scene. A moving vehicle or wave crest has a different phase history and may appear displaced, blurred, or otherwise distorted rather than at its instantaneous geographical position.
Instrument design and acquisition modes
A SAR instrument combines a microwave transmitter, receiving electronics, a stable timing and frequency reference, an antenna, navigation measurements, and a high-rate recording or downlink system. Many spaceborne systems use a phased-array antenna, whose electronically controlled elements steer or shape the beam without rotating the entire spacecraft. Airborne instruments commonly combine Global Positioning System measurements with an inertial navigation system to estimate their trajectories and support motion compensation.
Frequency selection establishes important trade-offs. X-band systems commonly operate near a wavelength of 3 centimetres, C-band near 5–6 centimetres, and L-band near 23–24 centimetres; P-band uses still longer wavelengths. Shorter wavelengths often emphasize small-scale surface features and upper vegetation layers. Longer wavelengths can interact more strongly with branches, trunks, or the ground beneath some vegetation. Penetration is not guaranteed by wavelength alone: moisture, conductivity, vegetation density, incidence angle, and the distribution of scattering elements all matter.
Most imaging systems transmit long, frequency-modulated pulses, often called chirps. A long pulse carries substantial energy, while its broad bandwidth permits fine range resolution after pulse compression. Transmit power, receiver noise, antenna gain, propagation losses, and coherent integration determine sensitivity and signal-to-noise ratio. The noise-equivalent backscatter is a useful performance measure because it indicates the approximate scattering level at which instrumental noise becomes comparable to the scene signal.
Acquisition modes distribute the available observing time and bandwidth differently. In stripmap mode, a roughly fixed side-looking beam continuously images a strip along the flight path. In spotlight mode, the beam tracks a selected area, increasing its coherent observation time and supporting finer azimuth resolution at the expense of continuous coverage. ScanSAR switches among several range subswaths, increasing coverage while shortening the observation time available to each location. TOPS, or Terrain Observation by Progressive Scans, additionally sweeps the beam in azimuth during each burst to improve the uniformity of wide-swath imaging.
Pulse repetition frequency creates another set of compromises. It must be high enough to sample the useful Doppler bandwidth, yet returning echoes from different transmitted pulses must remain distinguishable. Too little sampling produces azimuth ambiguities; unfavorable timing can produce range ambiguities or gaps where transmission prevents reception. Wide coverage and very fine resolution are therefore competing requirements rather than independently adjustable specifications. Multiple receiving channels, digital beamforming, and staggered pulse timing can relax some constraints, but increase calibration, hardware, and processing demands. Instrument design also has to balance electrical power, antenna area, thermal control, data volume, and spacecraft mass.
From raw echoes to georeferenced images
Raw SAR data are usually complex samples indexed by fast time, which represents echo delay within a pulse-repetition interval, and slow time, which indexes successive pulses along the trajectory. Their real and imaginary components preserve amplitude and phase. A point target does not initially appear as a sharp point: its transmitted waveform extends across range samples, and its changing distance distributes its signal over many pulses. Image formation concentrates this energy into its appropriate location.
Range compression applies a matched filter to the transmitted waveform. Azimuth compression then accounts for the target’s phase evolution along the synthetic aperture. An important complication is range-cell migration: during the observation interval, the same target can move through several recorded range bins as its distance from the antenna changes. Accurate focusing must align or otherwise model these contributions before coherent summation. Errors in the assumed trajectory, terrain height, or propagation path can leave residual blur or geometric displacement.
Several families of algorithms perform these operations. Range–Doppler processing separates much of the work into range and azimuth filtering, with an explicit correction for range migration. Chirp-scaling methods use phase operations to handle migration efficiently under suitable assumptions. Omega–k processing treats the data in a two-dimensional frequency or wavenumber domain. Backprojection instead calculates the expected travel time from each platform position to each output location and accumulates the appropriately phased samples. It is geometrically flexible but can be computationally demanding for large datasets.
Motion compensation is especially important for aircraft, whose paths depart from an ideal straight line because of turbulence and navigation errors. Autofocus techniques estimate some residual phase errors from the radar data themselves. These methods can sharpen imagery when bright or sufficiently structured scene content provides useful constraints, but they do not replace all navigation or calibration requirements. Precise focusing also does not automatically establish precise absolute geolocation: that depends on timing, orbit or trajectory accuracy, instrument delays, and the terrain model.
Common products include single-look complex images, which retain phase, and detected images containing amplitude or intensity. Multilooking averages approximately independent observations to reduce speckle at the cost of spatial detail. Radiometric calibration converts recorded values into defined backscatter quantities, commonly expressed as sigma nought, beta nought, or gamma nought, which use different reference-area conventions. These quantities should not be treated as interchangeable without accounting for their definitions.
Geocoding maps radar measurements into geographical coordinates. Terrain correction uses elevation information to address geometric distortion, while radiometric terrain correction also compensates for variations in the effectively illuminated area. Resampling creates a convenient map grid, but cannot recover surfaces hidden by radar shadow or separate all contributions already superimposed by layover. Processing metadata therefore remain essential to interpreting an apparently familiar map image.
Backscatter, speckle, and image interpretation
Radar brightness depends on the interaction of electromagnetic waves with the scene. The measured return is affected by surface roughness relative to wavelength, material dielectric properties, incidence angle, polarization, and scattering geometry. A smooth water surface viewed obliquely often reflects energy away from the radar and appears dark. A roughened water surface scatters more energy back toward it. Smooth bare ground can likewise be dark, while rough terrain may be bright. Brightness alone consequently does not identify a unique material or land-cover class.
Moisture strongly affects the dielectric properties of many natural materials. Liquid water has a much larger microwave permittivity than dry mineral soil or air, so changing soil moisture can alter backscatter substantially. The relationship is complicated by surface roughness, vegetation cover, soil texture, and viewing angle. Wet vegetation can change both its own scattering and the attenuation of signals traveling through it. At longer wavelengths, trunks and the ground may contribute strongly; in dense vegetation, volume scattering and attenuation can obscure a simple relationship between brightness and biomass.
Built structures frequently produce strong returns through double-bounce scattering. A signal may reflect between a vertical wall and adjacent ground before returning toward the radar. Buildings with favorable orientation can therefore be exceptionally bright, while differently oriented buildings of similar construction may be much less conspicuous. Corner reflectors deliberately exploit related geometry and are used for calibration and deformation monitoring. A bright pixel can represent one dominant reflector rather than a uniformly reflective patch of ground.
Speckle arises because a resolution cell usually contains numerous unresolved scatterers. Their complex echoes interfere constructively or destructively, producing a granular intensity pattern. It is an intrinsic consequence of coherent imaging, not merely electronic noise. For fully developed speckle, a common model treats single-look intensity as exponentially distributed; averaging independent looks reduces relative variability approximately as the inverse square root of their number. Real scenes may depart substantially from this ideal, especially where isolated strong reflectors or heterogeneous surfaces dominate.
Side-looking geometry creates distortions unlike those in most optical photographs. A slope facing the radar is compressed in range, producing foreshortening. If its upper part is closer to the radar than its lower part, their expected order reverses and layover occurs. A slope facing away may fall into radar shadow, where it receives no direct illumination. Opposite viewing directions can reveal complementary surfaces, but a single image cannot provide information from a true shadow.
Temporal comparison must also distinguish physical change from acquisition differences. A change in incidence angle, polarization, soil wetness, or vegetation water content can alter brightness without any change in land-cover category. Reliable interpretation therefore combines radar physics, acquisition metadata, spatial context, and, where appropriate, independent observations.
Polarimetry and three-dimensional scattering structure
Radar polarimetry examines how a scene changes the polarization of transmitted electromagnetic waves. Linear-polarization channels are commonly labeled HH, HV, VH, and VV: the first letter denotes horizontal or vertical transmission, and the second denotes reception. A fully polarimetric instrument measures the complex scattering responses needed to characterize the interaction more completely than a single-polarization image can. Dual-polarization systems observe a subset, while compact-polarimetric designs use alternative transmit and receive combinations to obtain additional information with different instrument compromises.
For a reciprocal monostatic scene, the cross-polarized responses HV and VH are ideally equal after calibration and convention-dependent coordinate treatment. The scattering matrix can be represented as
Its entries are complex quantities, so their relative phases matter as well as their magnitudes. Averaged covariance or coherency matrices describe statistical relationships among the channels. Polarimetric calibration must account for channel imbalance, cross-talk, and phase offsets; otherwise instrumental effects may be mistaken for properties of the scene.
Polarimetric decompositions express observed scattering in terms of mathematical components or idealized physical mechanisms. Surface scattering, double-bounce scattering, and volume scattering are common interpretive categories. They can help distinguish, for example, rough open ground, wall–ground interactions, and randomly oriented vegetation elements. Such categories are models rather than uniquely observed substances. Tilted structures, terrain slopes, mixed land cover, and departures from assumed particle distributions can lead to ambiguous or misleading decompositions.
Polarization is particularly useful where different environments have similar total return strength. Cross-polarized backscatter often provides sensitivity to vegetation structure and depolarizing multiple interactions. Co-polarized phase relationships can help distinguish certain surface and double-bounce responses. In agriculture, these measurements can track changes in crop architecture over a growing season, although crop type, row direction, moisture, and management practice remain intertwined.
Multiple observations can extend analysis from polarization to vertical structure. Polarimetric SAR interferometry combines polarization with interferometric phase to separate contributions having different effective scattering heights. Under suitable models, it can estimate vegetation height and aspects of canopy structure. SAR tomography uses acquisitions from multiple spatial baselines to resolve scattering along an elevation dimension. In cities, it may distinguish contributions from the ground, façades, and roofs that overlap in an ordinary image; in forests, it can provide information about vertical scattering distributions.
These methods require sufficient baseline diversity, accurate co-registration, stable calibration, and appropriate assumptions about temporal change. Vertical resolution depends on the sampled aperture in the elevation direction, not simply on the number of images. Dense vegetation and urban geometry can yield distributed or multiple scattering responses rather than a single well-defined height. The recovered structure is therefore an estimate of microwave scattering distribution, which does not always coincide directly with physical canopy boundaries or architectural surfaces.
Interferometry, topography, and surface deformation
Interferometric synthetic aperture radar, usually abbreviated InSAR, compares the phase of two or more suitably aligned SAR observations. The phase difference contains information about changes in radar–surface distance. Acquisitions made from slightly different positions can measure topography; observations separated in time can also reveal deformation. The measured interferometric phase is a mixture of contributions from terrain, displacement, atmospheric propagation, orbit errors, scattering changes, and noise.
The spatial separation between acquisition positions is the interferometric baseline. Its component perpendicular to the viewing direction controls much of the sensitivity to elevation. A larger perpendicular baseline generally increases height sensitivity, but excessive separation can reduce coherence because the two instruments or passes observe the scattering scene differently. A digital elevation model can be derived when the acquisition geometry is known and the phase is sufficiently coherent. Single-pass systems reduce temporal decorrelation by recording the scene almost simultaneously from separated antennas.
For deformation, the phase contribution associated with a change in slant range is commonly written, up to sign convention, as
One full phase cycle corresponds to a line-of-sight range change of . At a wavelength of 5.6 centimetres, this is approximately 2.8 centimetres. Phase precision can be much finer than one cycle, which explains how displacement sensitivity can greatly exceed the metre-scale dimensions of an image resolution cell. Such sensitivity does not imply equivalent absolute accuracy in every measurement.
The recorded phase is wrapped into an interval spanning . Phase unwrapping estimates the missing integer number of cycles to reconstruct a spatially continuous phase field. It can fail across incoherent areas, sharp displacement gradients, disconnected regions, or undersampled fringes. Coherence measures the consistency of complex signals between acquisitions, but high coherence alone does not prove that all phase differences represent ground movement.
Temporal decorrelation is severe where vegetation moves, crops grow, snow changes, or the surface is disturbed. Atmospheric water vapor introduces spatially varying delays that can resemble deformation. Ionospheric effects are particularly important at longer wavelengths and can produce broad phase structures or image distortions. Orbit errors and residual terrain errors introduce additional patterns. Time-series methods use many acquisitions to separate persistent motion from some of these effects statistically, often concentrating on stable point-like reflectors or distributed areas with consistent scattering.
InSAR measures motion along the radar line of sight, not a complete three-dimensional displacement vector. Combining ascending and descending observations can constrain multiple components, but near-polar satellite geometry generally offers relatively weak sensitivity to north–south motion. Interpretation also requires a reference location or reference frame: an apparently stationary area in a relative displacement map need not be stationary in an absolute geodetic sense.
Land processes, ecosystems, and the built environment
Over land, SAR contributes observations of processes that range from rapid disasters to gradual environmental change. Flood mapping is a prominent example because radar can acquire images through cloud systems associated with storms. Open floodwater often appears dark, whereas inundation beneath trees or among buildings may enhance double-bounce scattering and become bright. A dark-pixel threshold alone can confuse floodwater with permanent water, smooth dry surfaces, or radar shadow. Operational mapping therefore commonly uses pre-event imagery, terrain information, land-cover context, and multiple observations.
Agricultural applications exploit sensitivity to moisture, roughness, and plant structure. Repeated observations can distinguish growth patterns, harvest events, plowing, or inundated rice fields. Retrieval of soil moisture is most straightforward over relatively sparse vegetation when roughness and viewing geometry are adequately constrained. Microwave measurements predominantly represent a near-surface layer whose effective depth depends on wavelength and soil conditions; they do not directly measure the full root-zone water inventory. Crop mapping combines temporal signatures with contextual or statistical models rather than assuming a universal relationship between brightness and crop identity.
In forests, longer-wavelength measurements can provide information about woody structure and above-ground biomass. The relationship between backscatter and biomass often becomes less sensitive as biomass increases, with the effective saturation level depending on frequency, polarization, forest type, and environmental conditions. Interferometric and tomographic measurements provide complementary structural information. Disturbance monitoring can detect logging, storm damage, fire effects, or changes in inundation, but seasonal moisture and phenology may also change the signal.
Deformation measurements have become important in studying earthquakes, volcanoes, landslides, and subsidence. Earthquake interferograms can show broad displacement patterns across faults, while time series can reveal gradual volcanic inflation or the movement of some slow landslides. Subsidence associated with groundwater extraction, mining, or hydrocarbon production may extend over urban and agricultural regions. These observations measure surface response; identifying its cause requires geological, hydrological, engineering, or other independent evidence. Rapid displacement, steep terrain, and vegetation change can prevent useful measurements exactly where an event is most intense.
In the built environment, persistent radar reflectors on structures support monitoring of deformation affecting infrastructure, including railways, bridges, embankments, and industrial areas. A measured reflector may lie on a roof or façade rather than at ground level, and thermal expansion can contribute to its time series. Satellite measurements sample particular scattering locations and directions, so they are not equivalent to a complete structural inspection.
SAR also supports land-cover mapping and environmental assessment in regions with persistent cloud cover. Integration with optical imagery and lidar combines complementary information: optical sensors emphasize spectral properties, lidar measures sampled three-dimensional geometry, and SAR records coherent microwave scattering. Machine learning can combine these inputs, but model performance depends on representative training data and consistency across sensors, seasons, incidence angles, and geographical regions.
Oceans, sea ice, and snow-covered regions
Over the ocean, SAR primarily observes microwave backscatter from short surface waves, whose strength is influenced by wind and modified by currents, films, and longer waves. Many larger oceanic features become visible indirectly because they modulate these small waves. Fronts, eddies, internal waves, and convergence zones may therefore form distinctive bright and dark patterns without the radar directly penetrating deeply into seawater. Ocean imagery is also shaped by motion during the synthetic-aperture interval, which makes its interpretation different from imaging stationary land.
Wind-speed retrieval commonly uses a geophysical model function relating calibrated backscatter to wind speed, wind direction relative to the radar, incidence angle, frequency, and polarization. Direction information may come from visible wind streaks or an external atmospheric model. Rain cells, atmospheric stability, surface films, and strong currents can complicate the relationship. SAR offers fine spatial detail across coastal and storm-related wind fields, but its sampling in time and space differs from that of dedicated scatterometers or meteorological satellites.
Ocean waves introduce additional imaging mechanisms, including tilt modulation, hydrodynamic modulation, and velocity bunching. Orbital motions associated with waves change the Doppler history of the scattering surface. Under some conditions, the resulting azimuth displacement becomes nonlinear, suppressing or distorting parts of the wave spectrum. Consequently, the wavelength and direction of a pattern in a SAR image cannot always be interpreted as a direct, undistorted map of the sea-surface elevation field.
Ships frequently appear as bright targets against a darker sea, sometimes accompanied by wakes. Their apparent locations can be shifted because their motion differs from the stationary-scene assumption used in focusing. Oil slicks may suppress short waves and appear dark, but low-wind zones, natural biological films, and other processes can produce similar signatures. Distinguishing these cases requires context, additional observations, and analysis of surrounding conditions; darkness alone does not establish the presence or origin of oil.
Sea-ice observations benefit from independence from daylight and the ability to image through polar cloud cover. Backscatter and polarization provide information related to ice roughness, deformation, brine conditions, snow cover, and melt state. Successive images can track ice displacement and the opening or closing of leads. Ice classifications are seasonally dependent: surface melt and refreezing can substantially alter a signature even without a simple change in ice age.
On glaciers and ice sheets, feature tracking and interferometry measure ice motion, while repeat observations delineate changes in margins and grounding zones. Dry snow and cold ice can permit significant penetration at some wavelengths, so the dominant radar scattering horizon may lie below the visible surface. Wet snow strongly changes absorption and scattering, often producing useful contrasts for mapping melt conditions. Elevation and change measurements must therefore distinguish actual surface geometry from changes in the depth or character of the effective scattering layer.
Historical development and the changing observing system
The principle of synthetic-aperture radar emerged in the early 1950s. Carl A. Wiley at Goodyear developed a Doppler-based approach in 1951, and related work at the University of Illinois and the University of Michigan contributed to the practical development of coherent airborne imaging. Early systems faced a major computational challenge: combining large quantities of phase-sensitive data exceeded the capabilities of contemporary general-purpose computers. Optical processing provided an important route, using recorded radar data and coherent light to perform operations analogous to the required filtering.
Early development was closely associated with military reconnaissance, where high-resolution imaging independent of daylight offered clear operational value. Civilian scientific use expanded as instruments, processing methods, and data access improved. The transition from optical processors to digital computation made more accurate motion compensation, flexible image formation, and systematic large-area processing practical. Advances in electronics and storage also increased the bandwidth and data volume that airborne and spaceborne instruments could handle.
Seasat, launched in 1978, was the first satellite to carry an imaging SAR for oceanographic research. Although its mission lasted only about three months, its L-band observations demonstrated the value of spaceborne radar for oceans, ice, and land. Shuttle Imaging Radar experiments in the 1980s and the SIR-C/X-SAR mission in 1994 expanded scientific experience with different frequencies and polarizations. The European ERS-1 satellite, launched in 1991, Japan’s JERS-1 in 1992, and Canada’s RADARSAT-1 in 1995 helped establish sustained civilian satellite SAR observation.
Interferometric applications grew rapidly during the 1990s as repeat-pass satellite data became available and processing methods matured. The Shuttle Radar Topography Mission in 2000 used separated receiving antennas to collect near-global interferometric elevation measurements over much of the inhabited world. Germany’s TanDEM-X mission, launched in 2010 to operate with TerraSAR-X, used a two-satellite formation to acquire bistatic interferometric data for a global elevation model. These missions demonstrated how controlled observing geometry could turn phase measurements into large-scale topographic products.
Sentinel-1A, launched in 2014, and Sentinel-1B, launched in 2016, greatly expanded systematic C-band observations through the European Copernicus program. Their wide-swath acquisitions and open data policy supported routine time-series analysis across scientific and operational applications. Alongside public missions, commercial constellations introduced more frequent targeted acquisitions and high-resolution products, particularly at X-band. Revisit interval, usable acquisition geometry, licensing, latency, and long-term continuity remain distinct considerations: a large number of satellites does not automatically produce a consistent scientific time series.
Current development emphasizes high-resolution wide-swath imaging, digital beamforming, multiple receiving channels, compact platforms, and more rapid onboard or ground processing. Increasing data volume has also shifted analysis toward large archives and cloud-based processing environments. For long-term measurements, however, computational capacity alone is insufficient. Stable calibration, precise orbit knowledge, consistent acquisition plans, documented processing changes, and continuity between missions determine whether observations separated by years can be interpreted as a reliable record of the Earth’s surface.