The Gold Standard of Remote Sensing | Emissivity and Brightness-Temperature Bias

A thermal instrument measures radiance, not a thermometer pressed against a surface. Brightness temperature is the temperature an ideal blackbody would need in order to produce that measured radiance in the specified spectral band.

Real materials have wavelength-dependent emissivity ε, typically between zero and one. A material with ε=0.90 can emit less radiance than a blackbody at the same physical surface temperature, so the corresponding brightness temperature may be lower.

This distinction matters in urban heat mapping, dry soils, rock, vegetation and bare roofs. Temperature, emissivity, atmosphere, viewing geometry and calibration must not be collapsed into one imagined “heat colour” on a display.

Contents · Next chapter

Spectral radiance L_λ describes radiant power arriving at an instrument per projected area, solid angle and wavelength. Brightness temperature T_B is obtained by inverting Planck’s blackbody function for the instrument’s radiance and its band conventions.

Land-surface temperature T_s aims to describe the physical thermal state of the emitting surface. It is different from near-surface air temperature, which depends on air mixing, height above ground and measurement shelter.

Our worked example considers one narrow wavelength λ=11 μm, no atmosphere and no reflected surrounding radiation. These simplifying conditions allow emissivity to be isolated. A practical Landsat or ASTER algorithm uses band responses, atmospheric information and emissivity inputs beyond this toy relation.

Contents · Previous chapter · Next chapter

At a fixed wavelength, blackbody spectral radiance can be written B_λ(T)=C/[exp(c₂/(λT))−1], where C is a positive wavelength-specific prefactor and c₂=1.4387769×10⁻² m·K.

In the intentionally idealised case, a surface of physical temperature T_s emits L_λ=εB_λ(T_s). Define its brightness temperature by B_λ(T_B)=L_λ. The prefactor C cancels when we form the ratio.

The central quantities are wavelength in metres, surface temperature in Kelvin and emissivity as a unitless value. Using eleven metres instead of eleven micrometres, or converting Kelvin to Celsius inside the exponential, breaks the calculation.

Contents · Previous chapter · Next chapter

Fix T_s=300 K, equivalent to 26.85°C. Keep the wavelength λ=11 μm (11×10⁻⁶ m). Now vary ε through 0.90, 0.95, 0.98 and 1.00, holding every other quantity fixed.

Because the spectral radiance emitted by each specimen scales with ε in this model, the lower-emissivity surface has a smaller radiance even though all four surfaces have exactly the same physical temperature.

The table is a controlled mathematical comparison rather than a claim that any specific roof, soil or plant has precisely the listed emissivity. Real emissivity varies with wavelength, material, roughness, moisture and measurement conditions.

Contents · Previous chapter · Next chapter

Let x=c₂/(λT_s). With T_s=300 K and λ=11 μm, x≈4.3599. The emissivity-only model then implies T_B=c₂/[λ ln(1+(exp(x)−1)/ε)].

For ε=0.90, the inversion gives T_B≈293.005 K, about 6.995 K cooler than the physical surface temperature. That difference does not mean the material changed temperature; it arises solely from the lower emitted spectral radiance.

For ε=0.98, the same calculation gives T_B≈298.634 K, about 1.366 K cooler. Brightness-temperature error is therefore not a single universal correction in degrees: it depends on temperature, wavelength and emissivity.

Contents · Previous chapter · Next chapter

When ε=1, the emitted spectral radiance equals B_λ(T_s). Planck inversion returns T_B=T_s=300 K in our ideal vacuum model. This is a useful numerical limiting check.

For ε less than one, the brightness temperature remains below the physical temperature under these assumptions because B_λ rises monotonically with temperature. If your program gave a hotter brightness temperature for ε=0.90 here, revisit the emissivity term and logarithm.

Do not turn the limiting result into a general statement that any observed brightness temperature must always be less than a real land-surface temperature. Downwelling reflection, atmospheric emission, multiple surfaces and noisy retrievals can change the relationship.

Contents · Previous chapter · Next chapter

Suppose the true surface is 300 K with ε=0.90, but an analyst processes its idealised brightness temperature near 293.005 K assuming ε=0.98. Correcting the measured radiance using the wrong emissivity will not recover the true 300 K.

Inverting the same fixed-wavelength relation using ε_assumed=0.98 gives an estimated temperature of approximately 294.322 K. The estimate is about 5.678 K too low relative to the deliberately known 300 K surface.

This is an emissivity-assumption sensitivity example, not a forecast of bias in a modern published surface-temperature product. In real retrievals an uncertainty in emissivity can interact with atmospheric correction, channel width, mixed pixels and algorithmic constraints.

Contents · Previous chapter · Next chapter

Here all true surfaces were held at 300 K. Their different idealised brightness temperatures therefore cannot be interpreted as a real pattern of different surface heating. The table isolates material radiative behaviour from thermodynamic state.

The difference between ε=0.90 and ε=0.98 brightness temperatures is about 5.63 K despite no physical temperature change. If materials differ in both emissivity and temperature, a single observed thermal band generally cannot identify both quantities without additional information.

For an actual satellite pixel, report whether the displayed layer is top-of-atmosphere brightness temperature, calibrated radiance or a retrieved land-surface temperature product. Those are different stages of inference.

Contents · Previous chapter · Next chapter

A more realistic monochromatic thermal model includes L_TOA=τ[εB_λ(T_s)+(1−ε)L_down]+L_up. Here τ is atmospheric transmission; L_down is downwelling radiation reflected by the surface; L_up is atmospheric path radiance.

In this fuller expression, smaller ε both reduces surface emission and increases the fraction of downwelling radiance reflected upward for a simplified opaque surface. Atmospheric water vapour and temperature profiles influence transmission and path emission.

Published Landsat Collection 2 surface-temperature processing uses emissivity sources such as ASTER GED and atmospheric auxiliary data in addition to thermal imagery. It is therefore incorrect to describe the official product as raw brightness temperature or to apply our two-term classroom inversion directly to it.

Contents · Previous chapter · Next chapter

One thermal pixel can contain vegetation, soil, roof material and shadows, each with distinct emissivity and thermal state. The sensor receives a mixture of radiances, not the arithmetic average of their derived temperatures.

Because Planck inversion is nonlinear, converting an area-averaged radiance into brightness temperature is generally not the same as averaging component temperatures. Spatial resolution, point spread, sensor bandpass and subpixel area fractions all affect the interpretation.

Day-night temperature contrast, sun exposure and acquisition time also matter. A map can be geometrically aligned and still compare physically different times or conditions. Separate the radiometric and temporal questions before claiming that a neighbourhood became hotter.

Contents · Previous chapter · Next chapter

Exercise: keep the true surface at 300 K and λ at 11 μm, but use ε=0.95. The ideal single-wavelength brightness temperature is about 296.555 K, a 3.445 K reduction. Verify that the answer lies between the results for ε=0.90 and ε=0.98.

FAQ: Does a bright pixel reveal air temperature? No. Does a measured lower brightness temperature prove that a surface physically cooled? No. Can emissivity be assigned from visual colour alone? Not reliably; it depends on the thermal spectral properties of the material.

Three frequent errors are using degrees Celsius inside Planck’s exponential, treating brightness temperature as the final ground temperature, and comparing thermal layers without checking atmospheric correction, emissivity source and acquisition time.

Contents · Previous chapter · Next chapter

For any thermal comparison, record instrument and band, acquisition time, level of processing, calibration, emissivity source, atmosphere model, cloud and quality flags, land-cover changes, spatial support and independent temperature reference data where feasible.

Continue through Remote Sensing, nonlinear averaging in remote sensing, co-registration and false change, and revisit gaps and missed events.

Use the USGS Landsat Collection 2 surface-temperature description, USGS auxiliary-data inputs, and the NASA/USGS MODIS thermal product specification. The habit to transfer is that a displayed number must retain the observation process that produced it.

Contents · Previous chapter · Return to Remote Sensing