Synthetic aperture radar for all-weather Earth observation
2026-02-27
When Cyclone Idai made landfall in Mozambique in March 2019, it triggered floods that inundated hundreds of square kilometres of the Beira coastal plain. In the days following the storm, the sky remained cloudy — traditional optical satellites were blind. Sentinel-1, a European Space Agency radar satellite, imaged the flood extent through the cloud cover using C-band synthetic aperture radar. The flooded areas appeared dark in the radar imagery (smooth water reflects the radar pulse away from the satellite) against a brighter background of rough vegetation and soil. Emergency responders used those images to direct rescue operations, prioritise road repairs, and identify where populations were cut off. The flood maps were available within 24 hours of each satellite pass.
Synthetic Aperture Radar operates fundamentally differently from optical remote sensing. Instead of detecting reflected sunlight, SAR transmits its own microwave pulses and records the backscattered energy. Microwaves penetrate cloud, haze, and smoke. They work equally well at night. Long-wavelength SAR can even penetrate the forest canopy to measure soil moisture or surface topography beneath. The “synthetic aperture” refers to the way the radar uses the satellite’s forward motion to synthesise a very long antenna — far longer than could be physically mounted on a spacecraft — achieving fine spatial resolution in the along-track direction. The mathematics of this synthesis is Fourier processing: the variation in Doppler shift across many transmitted pulses is the signal from which the image is focused. This model derives the resolution equations, explains the backscatter physics, and introduces the interferometric phase measurement that underlies InSAR deformation monitoring.
Can we monitor flooding in real-time even when clouds obscure optical satellites?
SAR (Synthetic Aperture Radar):
Active microwave sensor that creates its own illumination.
Key advantages:
All-weather: Microwaves penetrate clouds, rain, smoke
Day/night: Active sensor, no sunlight needed
Penetration: Sees through vegetation canopy (long wavelengths)
Coherent: Phase information enables interferometry
Wavelength bands: - X-band: 3 cm (9.6 GHz) - High resolution, urban - C-band: 6 cm (5.4 GHz) - General purpose, Sentinel-1 - L-band: 24 cm (1.3 GHz) - Vegetation penetration, ALOS-2 - P-band: 70 cm (430 MHz) - Deep penetration, biomass
Applications: - Flood mapping (water appears dark) - Oil spill detection (dampens ocean waves) - Ship detection (bright against dark ocean) - Deformation monitoring (InSAR, Model 57) - Sea ice mapping - Agriculture (crop type, soil moisture) - Disaster response (rapid mapping)
Transmitted pulse:
P_t = P_{\text{peak}} \times \tau
Where: - P_t = transmitted power (W) - \tau = pulse duration (seconds)
Range resolution:
\delta_r = \frac{c \tau}{2}
Where c = speed of light.
Example: \tau = 10 μs
\delta_r = \frac{3 \times 10^8 \times 10 \times 10^{-6}}{2} = 1500 \text{ m}
Too coarse!
Pulse compression:
Use chirped pulse (frequency modulation).
Achieves:
\delta_r = \frac{c}{2B}
Where B = bandwidth.
Example: B = 100 MHz
\delta_r = \frac{3 \times 10^8}{2 \times 100 \times 10^6} = 1.5 \text{ m}
Much better resolution!
Real aperture: Physical antenna size limits resolution.
Azimuth resolution (real aperture):
\delta_a = \frac{\lambda R}{L}
Where: - \lambda = wavelength - R = range (distance to target) - L = antenna length
Problem: For \lambda = 0.06 m (C-band), R = 800 km, L = 10 m:
\delta_a = \frac{0.06 \times 800000}{10} = 4800 \text{ m}
4.8 km resolution - poor!
Synthetic aperture:
Platform motion creates long virtual antenna.
Processing multiple pulses coherently.
Achieves:
\delta_a = \frac{L}{2}
Independent of range!
For L = 10 m: \delta_a = 5 m
Orders of magnitude improvement.
Radar cross-section normalized by area:
\sigma^0 = \frac{P_r}{P_t} \times \frac{(4\pi)^3 R^4}{G^2 \lambda^2 A}
Where: - \sigma^0 = backscatter coefficient (unitless, often dB) - G = antenna gain - A = illuminated area
Typical values (dB): - Water (calm): -25 to -15 dB (dark) - Vegetation: -15 to -5 dB - Urban: -5 to +5 dB (bright) - Metal (corner reflector): +20 to +40 dB (very bright)
Image formation:
Pixel brightness ∝ \sigma^0
Interpretation: - Bright: Strong backscatter (rough, metallic) - Dark: Weak backscatter (smooth, water)
Received power:
P_r = \frac{P_t G^2 \lambda^2 \sigma}{(4\pi)^3 R^4}
Key: R^{-4} dependence (two-way path loss)
Solving for range:
R = \left(\frac{P_t G^2 \lambda^2 \sigma}{(4\pi)^3 P_r}\right)^{1/4}
Maximum range:
Limited by signal-to-noise ratio:
SNR = \frac{P_r}{P_n} > SNR_{\text{threshold}}
Typically SNR_{\text{threshold}} = 10 dB
Doppler shift from platform motion:
f_d = \frac{2v_r}{\lambda}
Where: - v_r = radial velocity component - \lambda = wavelength
For satellite moving at 7 km/s, C-band:
f_d = \frac{2 \times 7000}{0.06} = 233 \text{ kHz}
Doppler history of point target creates chirp.
Azimuth compression:
Match filter to Doppler chirp → focus image.
Resolution:
\delta_a = \frac{L}{2}
Half the real antenna length!
Transmit/receive polarization:
HH: Horizontal transmit, horizontal receive
VV: Vertical transmit, vertical receive
HV: Horizontal transmit, vertical receive
VH: Vertical transmit, horizontal receive
Scattering matrix:
\mathbf{S} = \begin{bmatrix} S_{HH} & S_{HV} \\ S_{VH} & S_{VV} \end{bmatrix}
Different targets, different polarizations:
Surface scattering (water): HH ≈ VV, HV ≈ 0
Double-bounce (urban): HH ≈ VV, strong
Volume scattering (forest): HV strong
Single-bounce (bare soil): HH > VV
Decomposition:
Pauli basis:
\mathbf{k} = \frac{1}{\sqrt{2}} \begin{bmatrix} S_{HH} + S_{VV} \\ S_{HH} - S_{VV} \\ 2S_{HV} \end{bmatrix}
Components: - Surface - Double-bounce - Volume
Problem: Calculate SAR resolution and classify target.
SAR parameters: - Wavelength: 5.6 cm (C-band) - Antenna length: 12 m - Bandwidth: 150 MHz - Range to target: 850 km
Target backscatter: - HH: -8 dB - VV: -9 dB - HV: -18 dB
Calculate resolution and identify scattering mechanism.
Step 1: Range resolution
\delta_r = \frac{c}{2B} = \frac{3 \times 10^8}{2 \times 150 \times 10^6} = 1.0 \text{ m}
Step 2: Azimuth resolution
\delta_a = \frac{L}{2} = \frac{12}{2} = 6 \text{ m}
Step 3: Compare to real aperture
\delta_{a,\text{real}} = \frac{\lambda R}{L} = \frac{0.056 \times 850000}{12} = 3967 \text{ m}
Synthetic aperture improvement: 3967 / 6 = 661×
Step 4: Analyze polarimetry
HH ≈ VV (similar, both -8 to -9 dB)
HV weak (-18 dB, 10 dB below co-pol)
Ratio:
\frac{\sigma_{HV}}{\sigma_{HH}} = 10^{(-18 - (-8))/10} = 10^{-1} = 0.1
Cross-pol 10% of co-pol
Step 5: Classification
Similar HH/VV → surface or double-bounce
Weak HV → not volume scattering
If bright (positive dB): Urban/double-bounce
If moderate (near 0 dB): Bare soil/surface
If dark (negative dB): Smooth surface
Given -8 dB: Moderately rough surface (agricultural field, rough terrain)
Not: Forest (would have strong HV)
Not: Water (would be much darker, -20 dB)
Not: Urban (would be brighter, +5 dB)
Below is an interactive SAR simulator.
<label>
Wavelength band:
<select id="wavelength-band">
<option value="x">X-band (3 cm)</option>
<option value="c" selected>C-band (6 cm)</option>
<option value="l">L-band (24 cm)</option>
</select>
</label>
<label>
Scene type:
<select id="scene-type">
<option value="water">Open water</option>
<option value="urban" selected>Urban area</option>
<option value="forest">Forest</option>
<option value="agriculture">Agriculture</option>
</select>
</label>
<label>
Incidence angle (°):
<input type="range" id="incidence-angle" min="20" max="60" step="5" value="35">
<span id="angle-val">35</span>
</label>
<div class="sar-info">
<p><strong>Range resolution:</strong> <span id="range-res">--</span> m</p>
<p><strong>Azimuth resolution:</strong> <span id="azimuth-res">--</span> m</p>
<p><strong>Backscatter (HH):</strong> <span id="backscatter-hh">--</span> dB</p>
<p><strong>Backscatter (VV):</strong> <span id="backscatter-vv">--</span> dB</p>
<p><strong>Scattering type:</strong> <span id="scatter-type">--</span></p>
</div><canvas id="sar-canvas" width="700" height="400" style="border: 1px solid #ddd;"></canvas>Observations: - Urban areas show strong backscatter (bright in SAR images) - Water appears dark (specular reflection away from sensor) - Forest shows moderate backscatter with volume scattering - Incidence angle affects backscatter magnitude - Resolution independent of range (synthetic aperture advantage) - Different wavelengths penetrate differently
Key insights: - SAR brightness reveals surface properties - Polarimetry distinguishes scattering mechanisms - All-weather capability critical for operational monitoring - Synthetic aperture achieves fine resolution from space
Water detection:
SAR backscatter from water: -25 to -15 dB (dark)
Flooded areas: Very dark compared to surrounding land
Automated detection:
Threshold: \sigma^0 < -15 dB → Water
Challenges: - Wind roughens water (brighter) - Vegetation emergence (wet but not dark) - Urban flooding (double-bounce from buildings)
Sentinel-1 operational: - 6-day repeat (2 satellites) - Free and open data - Copernicus Emergency Management Service
Example - 2017 Texas flooding: - SAR mapped extent when clouds obscured optical - Enabled emergency response routing
Ships = bright targets against dark ocean.
CFAR (Constant False Alarm Rate):
T = \mu + k\sigma
Where: - T = threshold - \mu = background mean - \sigma = background standard deviation - k = constant (typically 3-5)
Pixel > T → Potential ship
False alarms: - Waves (in high seas) - Oil platforms - Icebergs
Discrimination: - Size filter (ships 10-400 m) - Shape analysis (elongated) - AIS correlation (automatic identification system)
Applications: - Maritime surveillance - Illegal fishing detection - Search and rescue - Traffic monitoring
Crop monitoring:
Backscatter varies with: - Crop type (structure) - Growth stage (biomass) - Soil moisture (dielectric) - Roughness
VV polarization: Sensitive to vertical structure (stems)
HH polarization: Sensitive to horizontal elements (leaves)
Temporal analysis:
Track backscatter through season: - Planting (low, bare soil) - Growth (increasing) - Maturity (peak) - Harvest (decrease)
Crop type classification:
Combine temporal signature with optical data.
Accuracy: 85-95% for major crops
Coherent imaging produces granular noise.
Speckle: Random constructive/destructive interference
Standard deviation = mean for single-look
Reduction:
Multi-looking: Average independent looks
\sigma_{\text{speckle}} = \frac{\sigma}{\sqrt{N}}
Where N = number of looks.
Trade-off: Noise reduction vs resolution degradation
Filtering: - Lee filter - Frost filter - Gamma-MAP
Layover:
Targets closer to sensor than base → appear reversed
Common: Mountains toward sensor
Shadow:
Terrain blocks radar → no signal return
Common: Mountains away from sensor
Foreshortening:
Slopes compressed in range direction
Correction:
DEM required for geometric terrain correction (GTC)
Phase coherence required for InSAR.
Decorrelation sources: - Vegetation motion (wind) - Soil moisture changes - Snow melt/accumulation
Coherence:
\gamma = \frac{|\langle s_1 s_2^* \rangle|}{\sqrt{\langle |s_1|^2 \rangle \langle |s_2|^2 \rangle}}
Where s_1, s_2 = complex signals from two acquisitions.
\gamma = 1: Perfect coherence
\gamma = 0: Complete decorrelation
L-band better: Less decorrelation than C-band
Range ambiguities:
Signal from wrong swath enters image
Azimuth ambiguities:
Doppler aliasing creates ghost targets
Mitigation: - Careful PRF (pulse repetition frequency) selection - Antenna pattern shaping - Azimuth filtering
Full polarimetry:
Measure complete scattering matrix \mathbf{S}
Freeman-Durden decomposition:
\langle T \rangle = f_s T_s + f_d T_d + f_v T_v
Where: - f_s = surface scattering fraction - f_d = double-bounce fraction - f_v = volume scattering fraction
RGB composite: - Red: Double-bounce (urban) - Green: Volume (vegetation) - Blue: Surface (bare soil, water)
Applications: - Land cover classification - Crop type mapping - Wetland characterization - Snow/ice discrimination
Sensors: - ALOS-2 PALSAR-2 (L-band) - RADARSAT-2 (C-band) - Upcoming: NISAR (L+S band)
Approaching source:
f' = f \frac{c + v_r}{c}
Receding source:
f' = f \frac{c - v_r}{c}
For v_r \ll c:
\Delta f = f \frac{v_r}{c}
Signal travels to and from target:
f_d = \frac{2v_r}{\lambda}
Independent of transmitted frequency!
Example:
Satellite velocity: 7 km/s
Wavelength: 6 cm (C-band)
f_d = \frac{2 \times 7000}{0.06} = 233 \text{ kHz}
Phase history creates azimuth signal.