What Is NETD? Reading Thermal Camera Specs

Defining NETD: The Physics Behind the Number

Every thermal camera detector (whether an uncooled microbolometer array or a cryogenically cooled indium antimonide (InSb) focal plane array), generates a small amount of random electrical noise. This noise arises from several sources: Johnson–Nyquist thermal noise in the readout circuitry, photon shot noise from background radiation, and 1/f (flicker) noise in the detector material itself. The noise appears as random pixel-to-pixel intensity fluctuations that cannot be eliminated by averaging a single frame.

NETD quantifies this noise floor by asking a specific question: how large must a temperature difference between two adjacent blackbody surfaces be for the signal it generates to exactly equal the detector noise? When the signal-to-noise ratio (SNR) equals 1, the temperature difference causing that signal is the NETD. Mathematically:

NETD = ΔT at SNR = 1, expressed in millikelvin (mK)

A camera with an NETD of 50 mK can, in principle, distinguish two blackbody surfaces separated by 0.05 °C. At 100 mK the threshold doubles. The lower the NETD, the finer the temperature gradient the camera can image, which matters enormously when looking for early-stage bearing overheating (a 0.3 °C rise above ambient), microscopic PCB hot spots, or subtle moisture ingress in building envelopes.

It is important to note that NETD is defined at the detector plane, not at the object surface. The complete measurement chain (including lens transmittance, atmospheric absorption, and the target's emissivity), all degrade the effective scene sensitivity further. Manufacturers specify NETD under idealised lab conditions, which we will examine in detail below.

How NETD Is Measured in Practice

The standardised method for measuring NETD begins with a uniform-temperature extended-area blackbody (emissivity ε ≥ 0.99) filling the camera's entire field of view. The blackbody is set to a reference temperature (commonly 25 °C or 30 °C), with the ambient laboratory temperature stabilised to within ±0.5 °C of the target temperature. This ensures minimal background radiation gradient across the scene.

The camera records a sequence of frames (typically 100–300) with the blackbody held at reference temperature T₁. The standard deviation of the pixel output across all frames at a single representative pixel location gives the temporal noise (σ_t). The blackbody temperature is then stepped to T₂ (usually T₁ + 2 °C or T₁ + 5 °C), and the mean signal change (ΔS) per degree is derived. NETD is then:

NETD = (T₂ − T₁) × σ_t / ΔS

This procedure aligns with guidance in ISO 15529:2010 (Optics and photonics (Optical coatings), Measurement of the optical properties of thermographic systems) and ASTM E1316 (Terminology for Nondestructive Examinations). Reputable manufacturers specify NETD measured at f/1.0 (maximum lens aperture) since a narrower aperture reduces photon flux and increases effective NETD, so always check the f-number stated alongside the NETD figure.

Key Measurement Conditions That Affect the Stated NETD

Manufacturers are not always explicit about the conditions under which their NETD figure was obtained. When comparing datasheets, verify the following:

• Lens f-number: NETD degrades proportionally to (f-number)². An f/1.0 lens gives approximately 4× better NETD than an f/2.0 lens on the same detector.
• Integration time: Longer integration times average out temporal noise, artificially improving stated NETD. Field-ready integration times are shorter to prevent saturation on hot targets.
• Reference temperature: NETD is temperature-dependent. Detectors specified at 30 °C may perform differently at Singapore's typical ambient of 28–34 °C. A relevant consideration for outdoor use.
• NUC state: Non-Uniformity Correction (NUC) must be freshly applied. An outdated NUC allows fixed-pattern noise to dominate, masking the true NETD.

NETD vs. Temperature Accuracy: A Critical Distinction

One of the most common errors when purchasing or specifying a thermal camera is conflating NETD with temperature accuracy. They are fundamentally different quantities. NETD describes the camera's ability to detect differences between two targets in the same scene. Temperature accuracy describes how close the camera's displayed temperature is to the true absolute temperature of a surface, and this is governed by a different set of factors entirely.

A camera can have an excellent NETD of 30 mK (meaning it can image extraordinarily subtle temperature gradients), yet simultaneously have an absolute accuracy of ±3 °C because the manufacturer has not correctly accounted for the target's emissivity, atmospheric transmission losses, or reflected ambient radiation in the radiometric model embedded in the camera firmware. Conversely, a camera with a moderate NETD of 80 mK can be calibrated to an absolute accuracy of ±0.5 °C across a restricted temperature range if its radiometric model is well-characterised.

For a deeper understanding of how uncertainty is quantified in calibration certificates, see our guide on measurement uncertainty explained, which covers the GUM (Guide to the Expression of Uncertainty in Measurement) framework applied to temperature instruments.

How NETD Affects Calibration and Measurement Uncertainty

When a thermal camera is submitted for calibration at an ISO/IEC 17025 accredited laboratory, the calibration process characterises the camera's actual radiometric response across a defined temperature range. Typically the range stated on the calibration certificate, such as 20 °C to 100 °C or 30 °C to 45 °C for fever screening applications. This is distinct from the factory NETD test: calibration measures absolute accuracy, not noise floor.

However, NETD remains relevant to the calibration outcome because it is a fundamental contributor to the measurement uncertainty budget. In a Type A uncertainty evaluation (based on statistical analysis of repeated measurements), the scatter of the camera's readings around the reference temperature is largely determined by its NETD. A camera with NETD = 100 mK will show greater shot-to-shot scatter in its reading of a stable 37.0 °C blackbody than one with NETD = 40 mK, and this scatter propagates directly into the expanded uncertainty stated on the calibration certificate.

The combined measurement uncertainty for a thermal camera calibration typically includes contributions from:

• The reference blackbody's temperature stability and uniformity (Type B)
• The reference thermometer uncertainty traceable to NMC (Type B)
• Detector noise (NETD-driven, Type A)
• Non-uniformity correction residuals (Type B)
• Emissivity of the blackbody cavity (Type B, typically <0.01 contribution for ε = 0.99)
• Ambient temperature drift during calibration (Type B)

This is why accredited calibration laboratories follow ISO/IEC 17025 rigorously. The standard requires that every uncertainty contribution be identified, quantified, and combined in accordance with the GUM. A non-accredited calibration that simply compares the camera's reading to a reference thermometer and issues a certificate showing the difference is not the same thing: it may omit critical uncertainty contributions and cannot be relied upon for regulatory or audit purposes.

Understanding what a calibration certificate should contain is essential for any engineer or quality manager who relies on thermal camera measurements in their inspection or manufacturing process.

Common Mistakes When Reading NETD Specifications

Procurement teams and engineers frequently misread thermal camera NETD specifications in ways that lead to under-performing equipment for their application. Here are the five most common errors, drawn from the technical questions Unitest Instruments receives from Singapore industrial clients:

1. Comparing NETD figures without normalising the lens aperture

Camera A specifies NETD = 40 mK with an f/1.0 lens. Camera B specifies NETD = 60 mK with an f/1.2 lens. Which is more sensitive? Normalising both to f/1.0: Camera B's effective NETD at f/1.0 ≈ 60 mK × (1.0/1.2)² ≈ 42 mK. The cameras are nearly equivalent, but Camera B may be significantly cheaper. Always request the f-number used during NETD measurement.

2. Treating NETD as the camera's minimum detectable temperature difference in the field

NETD is measured with a large-area uniform blackbody at optimal focus, controlled ambient, and fresh NUC. In the field, targets are rarely blackbodies (emissivity correction introduces uncertainty), ambient conditions fluctuate, and the camera may be imaging a small target (IFOV-limited). Effective field NETD may be 2–5× worse than the stated figure, particularly in Singapore's humid outdoor environment where atmospheric water vapour absorbs LWIR radiation.

3. Assuming a lower NETD always means a better camera for the task

An electrical substation inspector needing to detect a busbar running 15 °C above ambient does not need NETD < 50 mK. An 80 mK camera with a superior lens, higher pixel count, and better absolute accuracy is a more rational choice. NETD matters most when the temperature differences of interest are below 0.5 °C, such as in semiconductor process monitoring or fever screening where a 0.2–0.3 °C elevation must be detected reliably.

4. Ignoring the drift of NETD over time

Detector noise characteristics change with component aging, mechanical shock, and (critically for uncooled microbolometers), cumulative exposure to high-temperature targets. A camera purchased with NETD = 50 mK may degrade to 70–90 mK within 2–3 years without periodic verification. This is precisely why thermal cameras used in regulated or quality-critical applications require regular calibration at defined intervals; see our article on how to set calibration intervals for a risk-based approach to frequency.

5. Not verifying that stated accuracy is achievable given the NETD

Some camera datasheets state accuracy of ±0.5 °C and NETD of 60 mK in the same table without acknowledging that the accuracy figure implicitly assumes perfect emissivity knowledge, controlled ambient, and no atmospheric path. Strip away those assumptions and the real-world accuracy for a target with unknown emissivity at 5 metres outdoors in Singapore may be closer to ±2–3 °C. Even with NETD = 60 mK.

NETD in Singapore's Regulatory Context

Singapore has specific regulatory touchpoints where thermal camera performance (including effective sensitivity), is relevant to compliance. Understanding these helps quality managers and safety officers specify and maintain cameras appropriately.

MOH Fever Screening Requirements

During infectious disease outbreaks, the Ministry of Health (MOH) Singapore has issued guidance requiring fever screening thermal cameras to meet defined accuracy requirements, typically ±0.5 °C absolute accuracy at body temperature range (35–42 °C) when used with a reference blackbody (also called a skin-temperature reference or reference source). Cameras used for this purpose must have NETD low enough that the noise floor does not prevent detection of a 0.3–0.5 °C elevated skin temperature. In practice, this means cameras with NETD ≤ 80 mK are generally specified, with many operators choosing ≤ 50 mK for safety margin.

BCA Building Energy Audits

The Building and Construction Authority (BCA) Green Mark scheme and associated energy audit standards require that building envelope thermal surveys be conducted with calibrated equipment. While BCA does not mandate a specific NETD value, the requirement for defensible, documented measurements means thermal cameras must be calibrated to a traceable standard, which is only achievable through an accredited calibration laboratory. The calibration certificate's stated uncertainty must be fit-for-purpose: for BCA thermographic surveys, uncertainties of ±1–2 °C are generally acceptable, achievable with cameras of NETD ≤ 100 mK.

Electrical Predictive Maintenance (EMA Context)

Thermal imaging of electrical switchgear, transformers, and busbars is widely used for predictive maintenance in Singapore's industrial and commercial facilities. The Energy Market Authority (EMA) and relevant codes of practice (SS 638, CP 88) do not explicitly mandate thermal camera NETD specifications, but insurance requirements and ISO 55001 Asset Management standards increasingly require that condition monitoring data be produced by calibrated instruments with documented uncertainty. For electrical predictive maintenance, temperature differences of concern (e.g. a loose connection running 10–40 °C above ambient) are large enough that NETD is rarely the limiting factor. Absolute accuracy and proper emissivity correction matter more.

Practical Steps: Getting Accurate Measurements Despite NETD Limitations

Even a camera with a moderate NETD of 80–100 mK can deliver reliable, defensible temperature measurements if the following practices are followed consistently in the field and at the calibration stage.

Set emissivity correctly. An error of 0.05 in the emissivity setting (e.g. using ε = 0.95 for a surface that is actually ε = 0.90) introduces a temperature error that is typically 1–5 °C depending on the target temperature and ambient. This dwarfs the noise contribution of a 100 mK NETD camera. Always measure emissivity with a reference tape or contact thermometer on the actual surface before thermal imaging begins.

Allow full thermal stabilisation. Uncooled microbolometers are sensitive to their own temperature. After powering on, allow at least 5–15 minutes for the camera body to reach thermal equilibrium before taking measurements. Perform a NUC (flat-field correction) immediately before critical measurements. Most cameras prompt for this automatically when the shutter temperature drifts beyond a threshold.

Minimise atmospheric path effects. LWIR (8–14 µm) cameras are relatively robust against humidity, but at distances beyond 10–15 metres in Singapore's high-humidity outdoor environment, atmospheric absorption becomes non-negligible. For precision measurements at distance, use the camera's atmospheric correction function with the measured ambient temperature, humidity, and path length. For a 30-metre path at 85% RH, uncorrected errors of 0.5–1.5 °C are possible.

Use a field reference. For fever screening and other high-stakes applications, place a calibrated reference blackbody (or a Peltier-stabilised skin-temperature reference) in the camera's field of view at all times. This provides a continuous radiometric ground truth against which the camera's drift can be detected and corrected. Effectively making the measurement uncertainty dependent on the reference's uncertainty rather than the camera's raw absolute accuracy.

Calibrate at the interval appropriate to your application and environment. Singapore's ambient temperature and humidity fluctuations, combined with the demanding duty cycles of security or process monitoring cameras, make annual calibration a reasonable default, with more frequent verification for regulated applications such as fever screening. An ISO/IEC 17025 accredited calibration ensures the calibration itself is metrologically traceable and that the certificate is defensible to auditors. Learn more about what traceability means in our guide: What is traceability in calibration?