What a Thermal Image Proves — and What It Doesn’t
Jump through this article
A thermal image is instantly convincing. Red and blue, plus a before and after — that looks like proof. In fact, a thermogram answers only a single question: Where is it warmer than elsewhere? It does not answer how well a building element insulates. That is not a matter of interpretation — it is written into the standard itself.
The standard says “qualitative” — and means it
DIN EN 13187, the standard most often cited in reports, has been withdrawn. It was replaced in November 2023 by DIN EN ISO 6781-1. Anyone receiving a report today that cites DIN EN 13187 should not treat this as an error — but as an indication of how up to date the work is.
The parent standard ISO 6781 states its scope unambiguously: it describes a qualitative method for detecting thermal irregularities — and explicitly: “Does not apply to the determination of the degree of thermal insulation and air tightness of a structure.”
The method is intended for finding, not for quantifying. It is telling what the successor standard refers to for numerical values: ISO 9869-1 for heat flux measurement and ISO 9972 for air tightness. The thermography standard thus sends the user elsewhere for figures.
The VATH Bauthermografie guideline draws the same line: qualitative detection of thermal bridges is possible from both inside and outside, whereas quantitative assessments are “in principle only possible from the inside, using supplementary measurement methods”.
Why there is no U-value in the picture
A U-value is a heat flux density per unit temperature difference. A thermal imaging camera does not measure heat flux density — it measures radiance and calculates a surface temperature from it.
Inferring heat flux from surface temperature would require the heat transfer coefficient at exactly that surface. Outdoors it is neither known nor stable: it depends on wind and on radiative exchange with the sky. On top of that, the building element stores heat — the temperature wave lags the outdoor climate by hours.
How coarse the discriminating power really is can be seen in an example with glazing: single glazing, uncoated double glazing and modern low-emissivity glazing can be distinguished thermographically. But low-emissivity glass with U = 1.3 cannot be distinguished from one with U = 1.1 W/(m²·K). Orders of magnitude, yes; decimal places, no.
The method that does measure U-values on existing buildings
There is one: heat flux measurement to ISO 9869-1. A heat flux plate is mounted on the side with the more stable temperature — in practice, indoors — and then you wait. The standard requires three conditions simultaneously:
- Measurement duration at least 72 hours.
- The U-value at the end of the test deviates by no more than 5 per cent from the value 24 hours earlier.
- The values from the first and the last two-thirds of the measurement period deviate by no more than 5 per cent from each other.
The 72 hours is the minimum under stable conditions; realistically it takes one to two weeks. Studies found convergence after four to five days, in winter up to ten. And even then an overall uncertainty of around 15 per cent remains.
From this follows the rule of thumb for every thermogram presented: if the contact method designed for this purpose, running for days on end, only delivers the U-value to about 15 per cent accuracy, then a method that produces an image in seconds and does not measure heat flux at all cannot deliver it more accurately.
What a usable image requires
| Condition | Requirement | Source |
|---|---|---|
| Indoor/outdoor temperature difference | ≥ 15 K recommended | VATH 2023 |
| Heating lead time | quasi-steady-state — depending on construction, up to several days | VATH 2023 |
| Wind | not viable from around 2 m/s | VATh 2011 |
| Sun | no influence, including in the hours beforehand | VATH 2023 |
| Sky | overcast; a clear sky should be avoided | Technical literature |
| Weather situation | stable, 3 to 7 days’ latency | Technical literature |
| Absolute measurement accuracy of the camera | 2 K (± 2 %), 1 K recommended | VATH 2023 |
The last row is the most important figure in this article. Any claimed improvement below two kelvin cannot be demonstrated with a thermal imaging camera — it lies within the device’s own accuracy.
Emissivity: irrelevant indoors, decisive outdoors
For opaque surfaces, ε + ρ = 1 applies. The lower the emissivity, the more the camera measures the surroundings instead of the object. What usually gets left out: the same error carries entirely different weight indoors and outdoors.
| Situation | true ε | Camera set to | Error |
|---|---|---|---|
| Indoors, wall 17 °C, reflecting 20 °C | 0.90 | 0.95 | +0.2 K |
| Outdoors overcast, wall 5 °C, reflecting 0 °C | 0.90 | 0.95 | −0.3 K |
| Outdoors, clear sky, wall 5 °C, reflecting −50 °C | 0.90 | 0.95 | −2.2 K |
| Outdoors, clear sky, same wall | 0.85 | 0.95 | −4.4 K |
| Outdoors, clear sky, galvanised gutter 5 °C | 0.10 | 0.95 | −47 K |
An emissivity that is off by 0.05 costs under 0.2 K indoors and is irrelevant. The same error costs over 2 K outdoors under a clear sky — putting it in the same order of magnitude as what gets interpreted as a thermal bridge or a refurbishment success.
Metals have an ε of roughly 0.02 to 0.3. That affects exactly the spots where thermal bridges are suspected: aluminium frames, parapet flashings, gutters, roller-shutter boxes, galvanised components.
The reflected ambient temperature
The camera needs to know what is being reflected in the surface, in order to subtract that share. The usual mistake is to enter the outdoor air temperature for outdoor shots. That is wrong: under a clear sky, “cold sky radiation” of roughly −50 to −60 °C is reflected. In terms of surface area, the sky beats the sun — the reflected temperature outdoors is usually below 0 °C, even on a sunny day.
It is determined using a piece of aluminium foil that has been crumpled up and smoothed out again: set the camera to ε = 1, read the temperature on the foil, enter that value as the reflected temperature.
The error acts uniformly across the image. It shifts the whole picture without changing its structure — which makes it particularly treacherous for before/after comparisons. A model calculation: wall at 5 °C, actually reflecting −50 °C, camera set to 0 °C, gives 0.9 °C instead of 5.0 °C. Over four kelvin off, without anything looking conspicuous.
Seven ways to fake a refurbishment success
All of the following produce an apparent improvement without anything actually changing on the building element:
- The colour scale. The most effective lever. Scale it too broadly and any energy guzzler turns into a low-energy house; scale it too narrowly and a low-energy house turns into a disaster. Narrow before, broad after — and the “success” is created purely in image processing.
- Wind. It evens out surface temperatures and makes thermal bridges disappear. An “after” shot taken in wind looks better than a “before” shot taken in still air.
- Clear sky instead of overcast. Under a clear sky, surfaces facing the sky cool so strongly that poor insulation looks like good insulation.
- Different time of day. Two images taken at different times of day capture different points in the cooling cycle — even at the same air temperature.
- Different indoor temperature. Less temperature difference means less contrast. The image becomes flatter and looks more homogeneous, i.e. “better insulated”.
- Furniture and pictures. A picture that has been taken down leaves a cold patch that looks like missing insulation. Hence: remove wall-mounted objects at least six hours beforehand — the same way at both sessions.
- A changed emissivity. The most treacherous point from a building-physics perspective. If the surface is painted, plastered or coated between the two sessions, ε changes — and with it the temperature read off, without any change whatsoever to the heat flux. A coating that lowers the emissivity produces an apparent temperature drop in the thermogram that isn’t real.
What a robust before/after comparison would need to satisfy
The guiding idea: don’t compare images, compare a metric independent of the temperature difference.
That is the temperature factor to DIN 4108-2: fRsi = (Θsi − Θe) / (Θi − Θe). It is a building-specific quantity and independent of the temperature difference prevailing at the time. The well-known threshold fRsi ≥ 0.70 corresponds, at 20 °C indoors and −5 °C outdoors, to a surface temperature above 12.6 °C.
One important caveat: because of the phase shift through the building element, calculating from instantaneously measured temperatures usually produces meaningless results. It only becomes useful with the preceding temperature history — in the region of thermal bridges, over six to eight hours. Without a data logger for indoor and outdoor temperature, even the temperature factor is not reliable.
On top of that: measure from the inside. The thermal resistance between the indoor air and the surface is roughly twice as high as outdoors, the contrasts are correspondingly stronger, and there is no wind, no rain and no radiative loss to the sky. Evidence consisting solely of exterior images is, from the outset, the weakest variant available.
And finally, a significance threshold. The claimed improvement must be greater than the sum of the uncertainties: ±2 K device accuracy, 2 to 4 K from emissivity outdoors under a clear sky, up to 4 K from the reflected temperature. In practical terms this means: a before/after comparison of two exterior thermograms is almost never suitable as proof of a refurbishment success.
If energy is genuinely the concern, thermography is the wrong tool. What is needed then is a heat flux measurement to ISO 9869-1 before and after the measure, or a weather-corrected consumption analysis. Thermography does what it is designed to do: it shows where to measure.
The four questions to ask of any thermogram presented
- Do both images show their scale end values, and are they identical? Is it the same colour palette?
- Are emissivity and reflected temperature documented? The VATH guideline requires this. If they are missing, the image cannot be verified.
- Are weather conditions, time of day, indoor temperature and location for both sessions recorded and comparable?
- Was anything applied to the surface between the two sessions? If so, the emissivity must have been redetermined — otherwise the comparison is physically invalid.
Sources
The delineation of the scope is taken from the freely accessible scope of ISO 6781; the full texts of DIN EN 13187 and DIN EN ISO 6781-1 are subject to a fee and were not consulted. Accordingly, no single boundary-condition figure has been deliberately cited from ISO 6781-1. The details on measurement duration and termination criteria under ISO 9869-1 are taken from the freely accessible standard preview and from technical publications; the uncertainty figure of around 15 per cent is from a methods review by the UK’s Building Research Establishment, not from the standard’s text.
The error tables on emissivity and reflected temperature are original model calculations based on the radiation balance in a broadband approximation. They represent orders of magnitude, not measured values from a real camera.
Several figures circulating in practitioner articles have deliberately not been adopted — a 24-hour pre-heating period, a wind limit of 3 m/s, and a one- or three-hour rule for solar radiation. They are attributed there to association guidelines, but do not actually appear in them.