The Gold Standard of Remote Sensing | SAR Backscatter Decibels and False Change Claims

Synthetic aperture radar actively transmits microwaves and records the returned signal. Unlike ordinary optical imagery, it does not require the Sun to illuminate the landscape. This makes radar useful during cloud cover and at night, but it does not make every dark shape into a uniquely identified surface feature.

An analyst comparing two dates may see the same mapped field become three decibels darker. That visual difference has a numerical meaning only when the calibration, coefficient, polarisation and scale are known. Radar brightness depends on surface structure, water content, view geometry, wavelength and scattering mechanisms. An apparent event in a processed image must be checked against these factors.

We will separate three claims. One is mathematical: the calibrated backscatter coefficient changed by a measurable ratio. Another is observational: corresponding ground areas under sufficiently comparable acquisitions show that change. The third is causal: a particular environmental process created it. Evidence for the first is not automatically evidence for the third.

Contents · Next chapter

Calibrated radar products may report backscatter coefficients such as σ⁰, β⁰ or γ⁰. Their geometric normalisations differ. For the main worked example, suppose we have a consistent σ⁰ linear-power product for a fixed polarisation and processing chain, and the numbers have already been correctly calibrated. We do not compare σ⁰ from one date with γ⁰ from another as though the two columns were automatically interchangeable.

The Copernicus Sentinel-1 GRD documentation explains that selected backscatter bands can be supplied as dimensionless linear power, with coefficients chosen according to processing settings. Their decibel display is computed as dB = 10 log₁₀(linear). The choice of coefficient and terrain treatment must be read from the actual product metadata.

A decibel is a ratio on a logarithmic scale; it is not a negative physical energy. A coefficient expressed as −13 dB corresponds to a small positive linear value. This is why a subtraction performed on dB values represents the logarithm of a power ratio rather than a simple subtraction of underlying powers.

Contents · Previous chapter · Next chapter

For a positive dimensionless calibrated power coefficient P, the logarithmic value is D = 10 log₁₀(P). The inverse is P = 10^D/10. Use the first equation for display or ratio comparison, and the second whenever a physical average or multiplicative difference requires linear power values.

For −10 dB, P=10⁻¹=0.1. For −13 dB, P=10^−1.3≈0.0501187. For −6 dB, P=10^−0.6≈0.251189. All three powers are positive and dimensionless for this selected coefficient; their corresponding logarithmic values happen to be negative because each power is below one.

Always confirm that the data are power rather than a raw digital number or uncalibrated amplitude. For amplitude ratios the logarithmic conversion uses 20 log₁₀(amplitude ratio). Applying the power formula twice, or using 10 log₁₀ directly to amplitude, changes the scientific meaning of a pixel.

Contents · Previous chapter · Next chapter

A useful check is that a 10 dB increase corresponds to a factor of ten in linear power. A 20 dB increase corresponds to a factor of one hundred. A 3 dB decrease corresponds to a factor of roughly one half, not three percent or an absolute decrement of 0.03 in the backscatter coefficient.

Use the inverse P=10^D/10 on each sample before making a physical-power comparison. The same display interval can conceal very different absolute differences: from −20 to −17 dB and from −10 to −7 dB are both changes of +3 dB, yet the latter has about ten times the absolute linear-power change of the former.

If a raster is exported with a stretch or colour ramp, the grey shades are an even less direct numerical guide. Examine the actual calibrated pixel values, not a screenshot, before reporting ratios.

Contents · Previous chapter · Next chapter

On date A our calibrated σ⁰ value is −10 dB, corresponding to linear power P_A=0.1. On date B it is −13 dB, corresponding to P_B≈0.0501187. The difference in display values is D_B−D_A=−3 dB.

The ratio of underlying power coefficients is P_B/P_A=10^−3/10≈0.501187. In this defined coefficient and acquisition comparison, date B has approximately 50.1 percent of date A’s calibrated backscatter power. That is a decrease of about 49.9 percent, not three percent.

A three-decibel drop is almost, but not exactly, a halving because an exact half corresponds to 10 log₁₀(0.5)≈−3.0103 dB. The distinction is small in many practical displays, but calculating the ratio correctly matters when results are combined with thresholds or statistics.

Contents · Previous chapter · Next chapter

Radar processing may store complex amplitude, detected amplitude or intensity. When the intensity or power ratio is about 0.501, the corresponding nonnegative amplitude-magnitude ratio is the square root, approximately 0.708. This does not mean there are two separate surface events; it is the same change described in two mathematical representations.

If two amplitudes have ratio A_B/A_A=0.708, then 20 log₁₀(0.708)≈−3 dB. If two powers have ratio P_B/P_A=0.501, then 10 log₁₀(0.501)≈−3 dB. The logarithmic difference matches because power is proportional to squared amplitude under the defined calibration.

This distinction becomes operational when a person receives an image labelled only “radar value” and applies a formula from a different processing step. Check the product documentation, measurement type and normalisation before any percentage, colour threshold or change map is constructed.

Contents · Previous chapter · Next chapter

Smooth open water can scatter most microwave energy away from a side-looking radar, making it appear relatively dark in a suitable configuration. Yet dark areas can also arise from radar shadow, smooth non-water surfaces, low returns in a chosen polarisation and other effects. A darkening pixel is therefore a candidate to investigate, not a one-pixel flood certificate.

Flooding in urban or vegetated terrain can sometimes increase radar returns through double-bounce interactions with walls or trunks. Wet soils, agricultural disturbance, wave roughness and changes in vegetation structure can likewise alter backscatter. The sign and size of a change depend on its physical scattering context.

NASA JPL’s radar backscatter introduction explains how electrical properties, roughness, polarisation and observation angle influence brightness. The Alaska Satellite Facility SAR introduction discusses scattering mechanisms and geometric distortions that require attention before interpretation.

Contents · Previous chapter · Next chapter

Consider a fictional two-pixel field with calibrated linear powers 0.01 and 0.1. Their arithmetic mean in linear power is (0.01+0.1)/2=0.055. Transform that result back to dB: 10 log₁₀(0.055)≈−12.596 dB.

If you transform first, the pixel values are −20 dB and −10 dB; their arithmetic dB mean is −15 dB. That is 2.404 dB lower than the decibel value of the mean linear power. There is no arithmetic contradiction: the logarithm is nonlinear, and an average of logs corresponds to a geometric mean in linear space rather than an arithmetic power mean.

A report comparing two regional averages must therefore specify whether it averages linear backscatter or logarithmic display values. The choice changes the estimand. When physical average backscatter intensity is desired, average valid linear-power pixels with the proper area and quality treatment before converting to a display scale.

Contents · Previous chapter · Next chapter

Before a two-date comparison, confirm that the scenes use compatible acquisition geometry, orbit direction where relevant, incidence-angle range, wavelength, polarisation and calibration. A VV backscatter value and a VH backscatter value are not interchangeable observations of one scalar property. Neither are ellipsoid-normalised and terrain-flattened coefficients automatically equal.

Check geolocation and co-registration. A pixel shifted across a shoreline or field boundary can sample a different surface even if nothing on the ground changed. Our companion co-registration and false-change guide works through this spatial problem separately from logarithmic unit conversion.

Terrain introduces foreshortening, layover and radar shadow. Radiometric terrain correction can help, but a genuine no-data shadow cannot be converted into a reliable surface measurement merely by stretching its colour. Mask invalid areas and preserve quality flags in a change analysis rather than silently filling them with invented low values.

Contents · Previous chapter · Next chapter

SAR returns fluctuate even when the underlying terrain is unchanged. Coherent interference among many scatterers creates speckle, which can give individual pixels strikingly different values. Multilooking or local filters can reduce this variation, but they also broaden the effective spatial footprint and can blur sharp features.

If a proposed detector uses a threshold of −3 dB for declaring change, its false-positive rate cannot be established from one ideal logarithm. It must be validated with appropriate matched no-change areas and known changed sites, sampling units, processing rules and uncertainty estimates. Unmatched dates or pixels that share processing artefacts can make a test appear stronger than it is.

Also avoid averaging a mixed region without defining the actual spatial population. A two-pixel arithmetic example teaches log math; a trustworthy area estimate needs geolocation, independent reference observations and defensible sampling. Return to spatial validation and nearby-pixel dependence for why repeated neighbours do not automatically provide independent accuracy evidence.

Contents · Previous chapter · Next chapter

Exercise: a calibrated linear-power pixel changes from 0.04 to 0.08. The ratio is 2, so the dB difference is 10 log₁₀(2)≈+3.0103 dB. The linear power has doubled. If the corresponding calibrated amplitude is used instead, the amplitude ratio is √2≈1.414, and 20 log₁₀(1.414) gives the same decibel difference.

A second exercise changes the order of operations. Take linear powers 0.02 and 0.18. Their mean is 0.10, so the dB of the arithmetic mean is exactly −10 dB. The arithmetic mean of their individual dB values is lower because log is concave. The discrepancy is mathematical; it should be anticipated before comparing histograms or setting thresholds.

FAQ: Are negative decibels negative energy? No. Does −3 dB mean a three-percent drop? No. Does a dark radar pixel prove water? No. Can polarisation be ignored because both bands came from the same satellite? No. Can a terrain-shadowed pixel be corrected into an observation? Not without additional information; treat genuine missing returns as missing.

Contents · Previous chapter · Next chapter

Begin with the Copernicus Sentinel-1 GRD documentation to verify whether a selected band supplies linear power and which backscatter coefficient is requested. The Alaska Satellite Facility guide describes amplitude, power, decibel scales, terrain effects and speckle. Check the version and settings of the exact product, not just a generic SAR glossary.

Use the Remote Sensing root for the full signal-to-map pathway, then compare thermal brightness temperature and emissivity as a separate sensor-specific measurement problem. Different sensors have different definitions of what the displayed number represents.

A defensible SAR change statement reports the coefficient, polarisation, linear or decibel units, processing history, dates, geometry checks, masking, threshold, uncertainty and evidence supporting the proposed physical cause. The numerical conversion can be exact while the causal conclusion remains uncertain; retaining both facts is what makes the map useful.

Contents · Previous chapter · Return to Remote Sensing