Characterising thermal imaging in field conditions – an inside view of NPL Lab Tests
Two-thirds through the project, much of the research in FRM4Fire reviewed uncertainties affecting Fire Radiative Power Product retrievals. In this article, we review research performed by NPL’s Temperature and Humidity team that addressed potential sources of uncertainty related to operating a thermal imaging camera onboard aircraft to support validations of satellite observations.
Featured image: KCL’s MWIR imager undergoing testing in NPL’s Temperature and Humidity Labs (NPL)
Background to the lab testing at NPL
The reference instrument of choice for an aerial thermography approach to validating satellite-derived Fire Radiative Power (FRP) data is the middle wave Infrared (MWIR) camera. This type of thermal imager captures spectra that match the relevant satellite’s spectral resolution. When mounted in the belly of an aircraft, pointed downwards, and flown over wildfires, these high-end devices provide accurate detection and temperature measurements.
If performed under the right conditions, such measurements could provide the best available ‘ground truth’ for assessing the accuracy of satellite FRP retrievals, such as those produced by Sentinel-3 SLSTR . Ideally, and to truly qualify as fiducial reference measurements (FRM), these would be independent, reliable, accurate, and traceable to S.I. units. As such, it is essential that supporting measurement protocols reflect these characteristics and are fit for the purpose of aerial campaigns.
Focused on foreseen uncertainties
Realistically, even the most high-spec optical systems will have associated uncertainties and will be affected by effects such as radiometric noise and geometric distortions, as well as operational limitations such as variations in output with ambient temperature.
Uncertainties can result from a lens failing to project a scene uniformly across the detector. Radial distortions — where straight lines bow outwards or inwards near image edges — lead to misalignment of the optical axis and cause asymmetrical stretching, vignetting, or off-axis blur, resulting in reduced brightness and sharpness. In this context, such distortions lead to spatial displacement and changes in the apparent size and shape of hot pixels or fire fronts, particularly near the edges of image frames.
These geometric distortions are therefore important to characterise. They can influence both pixel location — where a fire may appear displaced relative to its true ground position — and pixel shape and area, where the effective footprint may be larger or smaller than assumed. Ignoring such effects would result in in-situ FRP measurements that are over- or under-estimated, and spatial inconsistencies with satellite data, leading to apparent differences in fire intensity unrelated to the fires themselves.
MWIR camera set up for the 2024 field campaign (NPL).The MWIR camera operated by King’s College London (KCL) is the Infratec ImageIR 8300 thermal imaging system mounted with a 12 mm wide-angle lens and spectral filter, designed to mimic the spectral response function of the SLSTR sensor. According to its manufacturer, this camera provides temperature measurements with an accuracy of 2%. Characterising the performance of such a high-end device demands an equally high-end laboratory.
Characterising the thermal imager
An objective addressed in Work Package 1 (WP1) — characterisation of FRP measurement uncertainty sources — is to identify and quantify the main sources of uncertainty affecting airborne FRP measurements.
The main characterisations carried out in NPL’s laboratory testing focused on two aspects:
- Geometric accuracy — Optical systems introduce distortions that influence the apparent size, shape, and position of fire pixels. Since FRP is calculated as the product of radiance and ground-projected pixel area, such distortions propagate directly into radiative power errors. Quantifying these effects allows for proper correction and uncertainty assessment.
- Temperature response and radiometric calibration validation — MWIR sensors measure radiance rather than temperature directly. Converting radiance to FRP requires precise calibration of temperature response within the camera’s HDR operation mode under varying ambient conditions.
Current knowledge suggested these uncertainty sources would be dominant in airborne FRP retrievals. Other factors — such as point spread function (PSF), detector temporal response, environmental influences, or residual electronic noise — may also contribute but were not directly captured in this phase.
Establishing geometric and radiometric performance was expected to provide a robust metrological foundation for airborne FRP measurements and support the development of a fit-for-purpose validation protocol.
Some findings
Inspection
A calibration certificate provided with the camera declared accuracy and traceability to ITS-90 through a national metrology institute — in this case, PTB. This provided confidence in factory calibration across the procured temperature range. NPL’s lab analysis aimed to validate this calibration for different HDR modes and assess accuracy at lower, non-fire background temperatures.
Calibration validation
The camera was calibrated using multiple blackbody sources traceable to ITS-90. Measurements agreed with the specified 2% accuracy at higher temperatures, but not below 80 °C, indicating limitations for background temperature measurements in HDR mode under NPL test conditions.
Non-HDR calibration improved low-temperature measurements. It was also observed that switching HDR modes could require over an hour for full stabilisation, due to temperature variations within the camera housing.
Geometric calibration
Geometric characterisation was performed using the Zhang method, employing a planar calibration target to determine camera matrix and distortion coefficients. Results showed greater variation in per-pixel viewing vectors (PPVV) near image corners, indicating that corner pixels should be avoided when projecting thermal imagery onto terrain.
Figure 1 – Measured brightness temperature versus blackbody temperature (left), brightness temperature over time following HDR mode switching (centre), and PPVV differences between calibrations (right).
Insights put into practice
Farrer Owsley-Brown during the 2024 Canada field campaign.
NPL’s testing provided insights now being used to optimise field campaign procedures, including managing thermal stability and correcting for aerial thermography conditions.
These validated calibrations and improved operational understanding indirectly supported King’s College researchers during a field campaign focused on greenhouse gas fluxes in the Brazilian rainforest .
Outputs from Work Package 1 are feeding into a developing validation protocol for EO-based FRP measurements. With community contributions, the protocol is expected to reduce biases and errors in fire intensity and carbon emission estimates, and ultimately support traceability of FRP values to S.I. units.