Thermal-infrared satellite calibration challenges – an introduction

Thermal-infrared (TIR) imagery has grown in popularity. Applications range from wildfire detection and crop stress monitoring to maritime surveillance and urban heat mapping.

TIR sensors can be difficult to work with. They often have lower resolution, noisy signals, and fewer recognisable feature points than other optical satellite imagery.

Calibration challenges

For any satellite image to be considered calibrated, two things must be true.

1. Each pixel must represent its correct location (geometrically accurate)

2. Each pixel must represent its correct value (radiometrically accurate)

Satellite bus behaviour and characteristics of the imager influence the accuracy of the images. A good pre-processing software platform can calibrate and correct for most of these issues.

Geometric challenges

Satellite position and pointing accuracy

The geolocation accuracy of TIR images depends on how precisely the satellite knows its own position and attitude (orientation) in space.

When the satellite’s reported attitude is off by just 0.01°, the location of an image acquired from low-earth orbit (LEO) is off by roughly 90 meters on the ground. Small imprecisions in position and velocity will also affect the reported image location. A 90-meter error may not sound large, but for an image with a ground sampling distance (GSD) of 10 meters, it would put a field boundary or road in the wrong location entirely.

Attitude accuracy error of 0.01 degrees causes a 90 m offset on the ground.

Clock synchronisation

Every sensor and computer on a satellite runs its own clock. Keeping them synchronised is critical. Since the satellite is moving at roughly 8 km/s, a 1-second clock discrepancy translates directly into an 8 km positional error. With multiple cameras on the same bus, this can become especially problematic.

A 1-second clock drift causes an 8 km difference on the ground.

A real-life example where the TIR sensor stayed synchronised with the satellite while the VNIR imager lagged by close to a second, causing an observed 6 km error on the ground. Left represents a timing issue that was not present in the TIR image, compared to the right VNIR image, where the clock timing is out by 6 km.

Radiometric challenges

Detector noise

The satellite image and graph below shows horizontal streaking. This occurs when read-out noise is a significant contribution to the detector values being recorded. When the noise is high, the signal-to-noise ratio (SNR) is low, leading to higher uncertainties during radiometric calibration.

This satellite image shows that horizontal streaking is measured over one row of detectors, which, over a uniform field, varies in value due to having a low SNR.  

This graph, which is measured by the red line indicated on the satellite image, shows how significant noise can affect low SNR values over multiple row numbers.

Detector non-uniformity

The sensitivity and response of each detector in a camera are different. In a pushbroom camera, this leads to vertical streaking. Severe streaking can make it difficult to distinguish features in the image. Proper calibration should remove these stripes.

This example shows poor non-uniformity correction.

Temperature control

Satellite instruments are highly sensitive to temperature fluctuations. A detector’s sensitivity changes with temperature. Camera lenses and mirrors can distort images. Drift outside of the intended operating range can result in a blurry image or inadequate radiometric correction.

If radiometric accuracy is off, you cannot trust the pixel value itself. This means comparisons between images, or between a satellite value and a real measurement on the ground, become unreliable.

To recap

Satellite bus behaviour and characteristics of the imager influence the accuracy of the acquired TIR images. It is important that every pixel is in the right location and that every value can be trusted. Good pre-processing software calibrates and corrects for most of these issues.