Satellite Thermal Sensors (Earth Imaging Basics)
Satellite thermal sensors measure heat energy emitted by Earth, not temperature directly. They record thermal-infrared radiance, then scientists estimate surface temperature by accounting for sensor calibration, the atmosphere, and surface emissivity. Knowing the difference between radiance, brightness temperature, and land-surface temperature helps you choose the right satellite data and interpret its limits.
What can a thermal satellite image tell you about a place? It can show patterns in emitted heat, such as warmer and cooler areas, but the numbers need careful interpretation. A bright-looking or warm-looking pixel is not automatically a reliable measurement of the ground’s actual temperature.
I find it useful to think of satellite temperature mapping as a sequence of measurements and corrections. The sensor observes energy, a processing method turns that energy into a temperature estimate, and quality checks show where the estimate may be unreliable. Each step affects what you can safely conclude.
Planning: Sensor Choice, Coverage, and Acquisition Conditions
Thermal sensors differ in the wavelengths they measure, the size of the area represented by each observation, and how often they view a location. Choose a sensor and product to fit your question, then check the acquisition date, scene coverage, cloud conditions, and product documentation before comparing temperatures.
What should I check when choosing a sensor?
Sensor choice sets the scale and timing of your analysis. A satellite with frequent coverage may suit broad, changing patterns; a sensor with smaller pixels may better suit local comparisons. Neither choice removes the need to check clouds, image quality, and the product’s stated resolution.
Landsat 8’s Thermal Infrared Sensor (TIRS) has two thermal bands:
| Sensor or band | Wavelength range | Nominal thermal resolution | Key timing or use |
|---|---|---|---|
| Landsat 8 TIRS Band 10 | 10.60–11.19 µm | 100 m native | Often delivered on a 30 m grid |
| Landsat 8 TIRS Band 11 | 11.50–12.51 µm | 100 m native | Often delivered on a 30 m grid |
| Terra/Aqua MODIS Band 31 | 10.78–11.28 µm | 1 km at nadir | Broad-area thermal observation |
| Terra/Aqua MODIS Band 32 | 11.77–12.27 µm | 1 km at nadir | Broad-area thermal observation |
Landsat 8 has a 16-day repeat cycle. That describes its planned return timing, not a promise of a cloud-free image on a particular date. Clouds or other image-quality issues can make an observation unsuitable even when the satellite passes over the area.
Before downloading data, confirm that the scene covers your area and date, and read the product notes for the resolution and quality layers. If you compare two dates, make sure the dates, processing levels, and temperature methods are suitable for comparison.
Access and Clearances: Resolution, Footprint, Clouds, and Pixel Limits
Resolution describes the ground area represented by a measurement, while a footprint describes the area covered by an image or observation. A stated grid size does not always equal the sensor’s true ability to distinguish detail. Clouds, shadows, viewing geometry, and mixed surfaces can all limit what a pixel tells you.
What does “100 m thermal resolution” mean?
Native resolution is the scale of the original thermal measurement. A Landsat 8 TIRS measurement has a native thermal resolution of 100 m, although products are commonly resampled onto a 30 m grid. Resampling changes the grid spacing, not the amount of thermal detail the sensor captured.
This distinction matters when mapping small features. A 30 m output cell from a resampled thermal product is not an independent 30 m thermal observation. Treating it as one can make a map look more detailed than the underlying measurement supports.
A thermal pixel may also combine several kinds of surface. For example, one pixel could include roofs, roads, trees, and soil, each at a different temperature and emissivity. The reported value reflects the combined signal, so it may not equal the temperature of any one feature within it.
Clouds and cloud shadows can block or alter the signal from the surface. Use the product’s quality information to identify pixels marked as cloud, shadow, or otherwise invalid. Do not assume that a scene is clear just because it looks clear at the scale or color display you are viewing.
Practical check: match the sensor’s native thermal resolution to the size of the feature you want to study. If the feature is smaller than a pixel, report that limitation rather than interpreting the map as a precise temperature reading for that feature.
Execution: Radiometric Calibration and Temperature Retrieval
Radiometric calibration converts recorded digital values into spectral radiance, the energy measured by the sensor over a wavelength range. Temperature retrieval then estimates surface temperature from that radiance. It requires documented treatment of atmospheric effects and surface emissivity, so raw values alone are not land-surface temperature.
How does a satellite infer surface temperature?
Earth’s surface emits thermal-infrared energy. Planck’s radiation law describes how the amount of energy emitted by an object varies with its temperature and wavelength. A thermal sensor measures radiance in selected bands; it does not directly touch or measure the ground’s temperature.
A common intermediate value is brightness temperature. This is the temperature that would produce the measured radiance under an idealized assumption about the emitting surface. It is useful, but it is not automatically the actual land-surface temperature.
Two factors need further treatment. Emissivity describes how efficiently a surface emits thermal energy compared with an ideal emitter. Atmospheric effects arise because gases and other components of the atmosphere can absorb and emit energy along its path to the sensor. Retrieval methods account for these effects to estimate land-surface temperature.
Use this processing sequence:
- Select and screen: Choose the sensor and product, then check the acquisition date, scene footprint, cloud information, scan geometry, and native resolution.
- Calibrate: Use the product metadata and sensor-specific calibration coefficients to convert digital values to calibrated spectral radiance. Follow the processing guidance for the product you selected.
- Retrieve temperature: Apply a documented method for atmospheric and emissivity corrections. Keep the method and inputs with your results.
- Mask invalid data: Retain quality flags and exclude pixels identified as cloud, cloud shadow, or otherwise invalid.
- Report: State the units, method, masks, resampling, assumptions, and uncertainty.
A useful case study is a comparison of two dates over the same region. First, select compatible products and check whether both dates have usable coverage. Then apply a consistent retrieval method and masks. If one date has cloud contamination or a different processing basis, a temperature difference may reflect data quality or method choices, not a real change at the surface.
Verification: Quality Flags, Uncertainty, and Ground Checks
Quality flags mark observations that may be affected by clouds or other problems; uncertainty describes the limits of a temperature estimate. Check both before drawing conclusions. A sound report also states how the image was processed and, where possible, compares satellite estimates with suitable ground or reference observations.
How can I check whether a temperature estimate is trustworthy?
Start with the product’s quality layers and documentation. Use them to identify cloud, cloud shadow, and invalid pixels, and follow the product’s guidance on which flags to exclude. There is no single cloud-cover cutoff that fits every study; a suitable screening rule depends on the goal and the data product.
Next, review the retrieval method. Confirm that it accounts for atmospheric effects and emissivity before calling the result land-surface temperature. If you have only radiance or brightness temperature, label it accurately rather than presenting it as validated surface temperature.
A second case study is a map of a mixed urban area. A pixel that covers buildings, pavement, vegetation, and soil blends their signals. Even if the pixel is free of cloud, it cannot tell you the temperature of each individual surface. Compare patterns at a scale the sensor supports, and avoid treating resampled cells as separate fine-scale measurements.
Ground checks can help assess a satellite estimate, but the comparison must be suitable. Match the place and time as closely as the available observations allow, and consider differences in measurement area and method. A ground reading from one small spot may not represent the mix of surfaces in a satellite pixel.
When reporting results, include:
- Sensor, band, product, and acquisition date.
- Temperature units and retrieval method.
- How emissivity and atmospheric effects were handled.
- Cloud and invalid-pixel masks used.
- Native resolution and any resampling.
- Known limits and uncertainty, including mixed pixels.
This makes the map easier to interpret and helps others distinguish measured signal from processing choices.
Conclusion and FAQ
Satellite thermal imagery is a measurement of emitted energy that can support a surface-temperature estimate when processed with care. The key is to keep radiance, brightness temperature, and land-surface temperature distinct, respect the sensor’s native resolution, and report the quality checks and assumptions behind the result.
What do satellite thermal sensors measure?
They measure thermal-infrared radiance in selected wavelength bands. Temperature is derived from that radiance through processing; it is not measured directly at the ground.
What is brightness temperature?
Brightness temperature is the temperature associated with a measured radiance under an idealized emission assumption. It is not automatically the actual land-surface temperature.
How is land-surface temperature different?
Land-surface temperature is an estimate of the surface’s temperature after accounting for factors such as emissivity and atmospheric effects. The method and data used should be documented.
What is the native thermal resolution of Landsat 8 TIRS?
The native thermal resolution is 100 m for TIRS. Thermal products are commonly delivered on a 30 m grid, but that resampling does not create 30 m thermal detail.
How often does Landsat 8 revisit a location?
Its repeat cycle is 16 days. That is not a guarantee that a clear, usable observation will be available on each pass.
What is the resolution of MODIS thermal bands 31 and 32?
Their nominal spatial resolution is 1 km at nadir. The sensor measures broad areas, so it is not suited to interpreting very small surface features as separate observations.
Can a clear-looking pixel still be misleading?
Yes. It may contain different surface types with different temperatures and emissivities, or be affected by image conditions that are not obvious in a display. Check quality flags and consider mixed pixels.
Does a 30 m thermal grid mean 30 m temperature detail?
No. When 100 m thermal measurements are resampled to a 30 m grid, the grid spacing changes but the original thermal detail does not increase.
Do I need to account for clouds?
Yes. Clouds and cloud shadows can interfere with the surface signal. Use the product’s quality information to mask affected or invalid pixels.
What should a temperature map report?
State the sensor and date, units, retrieval method, emissivity and atmospheric treatment, quality masks, native resolution, resampling, and known uncertainty.
(This article was written by one of our staff writers, Steven Langford. Visit our Meet the Team page.)
