Find the question you need
Select a learning route. All fifteen teaching chapters are open, with an expandable chapter index below.
Define the thermal quantity · Compute both emission cases · Interpret net radiation · Review measurement limits · Practise and link
Two panels each have an area-weighted temperature of 300 K but give different blackbody radiation. The difference comes from nonlinear averaging, not from a thermometer mistake. This guide calculates and audits both cases.
Chapter index
Advanced worked companion: sensitivity and area weighting · Begin chapter 1
Define the thermal quantity · Chapters 1–4
1. What the two identical means conceal · 2. State the assumptions and preserve units · 3. Use kelvin and the radiation law correctly · 4. Verify both arithmetic area means
Compute both emission cases · Chapters 5–8
5. Compute the uniform panel’s emission · 6. Compute the two-patch emission and compare · 7. Understand the convexity behind the result · 8. Define the radiation-equivalent temperature
Interpret net radiation · Chapters 9–11
9. Separate emitted radiation from net exchange · 10. Test a smaller and more realistic temperature spread · 11. Emissivity, reflections and view factors in practice
Review measurement limits · Chapters 12–13
12. Thermal-camera measurements are not perfect temperature maps · 13. Questions readers ask and three practice checks
Practise and link · Chapters 14–15
14. Audit checklist before applying the result · 15. Connected reading and technical sources
Imagine two one-square-metre thermal panels. One is uniformly at 300 kelvin. The other has one half at 200 K and the other at 400 K. Every number is hypothetical and is selected for mathematical clarity, not as a record of a real thermal survey. Both panels have an arithmetic area-weighted temperature of 300 K. Nevertheless, their blackbody radiation emission differs considerably.
This case matters because a thermal monitoring dashboard may report an accurate average and still leave out a spatial feature essential for a physical decision. If the decision involves emission, overheating, radiation exchange or surface durability, a single average temperature does not necessarily answer the question.
Under the Stefan–Boltzmann law, outgoing thermal emission from an ideal black surface is proportional to the fourth power of absolute temperature. Averaging measured temperatures before applying that law loses information about hot and cold patches. The correct operation is to calculate emission at each patch, then combine the results using area weights.
We will verify every numerical step and then look beyond the ideal law. Emissivity, view factors, reflected surroundings, measurement resolution, environmental exchange and sensor uncertainty all matter when moving from a classroom model to an engineering conclusion. The lesson is not that averages are useless. It is that an average has to be matched to the physical question it is being asked to answer.
Take each panel’s exposed emitting area as exactly 1.00 m². The idealised surfaces are blackbodies with emissivity one at every location and wavelength. Temperatures are held steady throughout each comparison. We consider outgoing radiation from each patch independently, without yet modelling where that radiation is subsequently absorbed.
The emitted radiant exitance M is power per unit area, measured in W/m². Multiplying M by area gives total emitted power P in watts. A calculation of 500 W/m² for a two-square-metre panel does not mean the panel emits only 500 W. It means the panel emits 1,000 W, assuming the same exitance across both square metres.
The area-weighted arithmetic temperature is defined by ΣAᵢTᵢ/ΣAᵢ. The weighted emission is defined through ΣAᵢMᵢ. When emissivity and temperature vary, these expressions preserve different information. We keep area, material and snapshot conditions fixed so the only initial distinction is spatial temperature distribution.
This deliberately simplified comparison does not include convection, conduction, internal heating, thermal storage or radiative reflections. Those effects can be essential in an actual installation and will be considered qualitatively later. A physically restricted model is useful provided the reader knows what has been included and excluded.
A blackbody’s emitted radiant exitance is M=σT⁴, where σ≈5.670374419×10⁻⁸ W·m⁻²·K⁻⁴. For a uniform patch, P=AσT⁴. The units cancel correctly: the fourth power of kelvin removes K⁻⁴ and the area removes m⁻², leaving watts.
The temperature must be expressed in kelvin. A temperature of 300 K corresponds to approximately 26.85°C, but substituting 26.85 into T⁴ would be wrong. The absolute zero of the thermodynamic scale matters to this law. Celsius and kelvin increments have equal numerical sizes, yet their absolute fourth powers do not.
The NASA introduction to blackbody radiation explains that emitted energy rises strongly with absolute temperature and distinguishes the integrated Stefan–Boltzmann relation from wavelength-specific radiation laws. Here we use the integrated emitted energy, not the response in a particular infrared camera’s band.
Even an ideal blackbody may absorb radiation from its environment while simultaneously emitting energy. The Stefan–Boltzmann emission value is not automatically the net heat loss. A later chapter adds a surrounding-temperature model and explicitly separates these two quantities.
Panel U has T=300 K over its entire square metre. Its area-weighted arithmetic mean is therefore 300 K. It is genuinely uniform in our stipulated surface model, although nothing has been stated about temperature inside its material.
Panel V has 0.50 m² at 200 K and 0.50 m² at 400 K. Its mean is (0.50×200+0.50×400)/1.00=300 K. This is not a rounding trick. The equality is exact, and a researcher who reports that these two surfaces have the same arithmetic area mean has said something true.
But their temperature maps differ: Panel U has no hot spot, while Panel V contains a 400 K region. The same mean also does not imply equal material ageing, stored thermal energy or peak-temperature risk. Those would require further information about mass, heat capacity, chemistry and spatial gradients.
This step establishes only that the first area-weighted moment of temperature agrees. The question is whether a nonlinear physical quantity, built from higher powers of temperature, is determined by that first moment. It is not.
Panel U emits σ(300 K)⁴ per unit area. The fourth power is 300⁴=8.1×10⁹ K⁴. Multiply by the stated constant to get approximately 459.30 W/m². Since the emitting area is exactly one square metre, total emitted power is approximately 459.30 W.
Use the units as a diagnostic. The number 459.30 does not describe an electrical input, a radiant flux in one wavelength band or a room’s net thermal load. It is the integrated outgoing radiation calculated for an ideal blackbody at 300 K.
If the uniform panel’s area were doubled without changing anything else, the emitted total wattage would double, but its emitted watts per square metre would not. If the surface emissivity were a known constant smaller than one, a simplified gray-surface model would multiply this emitted flux by that emissivity.
The 459.30 W result is our reference. Only after establishing this reference can the two-patch surface be compared without accidentally changing temperature units, total area or what the word ‘power’ means.
Panel V’s emitted power is 0.50σ(200 K)⁴+0.50σ(400 K)⁴. The component fourth powers are 1.6×10⁹ K⁴ and 25.6×10⁹ K⁴. The latter is sixteen times the former, because 400 K is twice 200 K and 2⁴=16.
Area weighting gives [(1.6×10⁹)+(25.6×10⁹)]/2=13.6×10⁹ K⁴. Multiplying by σ gives approximately 771.17 W for the entire one-square-metre panel. The emitted-power ratio to the uniform panel is 13.6/8.1≈1.679.
Thus Panel V emits about 311.87 W more in the invented blackbody example, despite having exactly the same arithmetic mean temperature. The hot region makes a much larger contribution than the cold region subtracts relative to the reference.
This does not mean that an actual surface could sustain such a temperature split without special conditions. The example is a mathematical comparison, not a proposed steady-state design. What it proves is that the average alone does not determine the fourth-power integral needed for thermal emission.
| Quantity | Uniform Panel U | Patchy Panel V |
|---|---|---|
| Emitting area | 1.00 m² | 1.00 m² |
| Temperature field | 300 K everywhere | 0.50 m² at 200 K; 0.50 m² at 400 K |
| Arithmetic area mean | 300 K | 300 K |
| Mean T⁴ | 8.1×10⁹ K⁴ | 13.6×10⁹ K⁴ |
| Ideal emitted power | 459.30 W | 771.17 W |
| Emission ratio versus U | 1.000 | 1.679 |
The mistaken shortcut computes σ(⟨T⟩)⁴, as though the mean of T⁴ were the fourth power of mean T. In general ⟨T⁴⟩≠(⟨T⟩)⁴. Our panels yield 13.6×10⁹ K⁴ and 8.1×10⁹ K⁴, respectively, for those expressions.
The function f(T)=T⁴ is convex over nonnegative temperatures. Jensen’s inequality states that a weighted average of f(T) is at least f of the weighted average T when nonnegative weights sum to one. This explains the direction of the inequality without relying on our chosen numbers.
For blackbody surfaces of equal area and fixed arithmetic mean, a uniform distribution produces the smallest emitted power among distributions permitted by those assumptions. A nonuniform distribution generally produces more. The mean still does not specify how much more; that requires some information about the spread and shape of the temperature distribution.
The conclusion changes if our physical functional changes. With different emissivities, the relevant expression can be ΣAᵢεᵢTᵢ⁴; simple convexity in T alone is then not a complete comparison. Avoid quoting Jensen’s inequality as proof about two surfaces whose material properties or environments have not been made comparable.
A surface’s radiation-equivalent effective temperature T_eff is the uniform blackbody temperature that reproduces the same emission per unit area: σT_eff⁴=σ⟨T⁴⟩. Thus T_eff=(⟨T⁴⟩)^(1/4). This is a fourth-power mean rather than an arithmetic mean.
For Panel U, T_eff=300 K. For Panel V, T_eff=(13.6×10⁹)^(1/4)≈341.50 K. This value is higher than its 300 K arithmetic mean because the hot half has disproportionately increased total outgoing emission.
Do not interpret T_eff as the temperature of an actual particular cell or component. No patch at 341.50 K was present in our example. It is a mathematically defined equivalent temperature for one selected measurable consequence.
Many spatial distributions can share the same radiation-equivalent temperature, so even this more suitable single number cannot reconstruct where hot spots occur. A safety question about the maximum local temperature needs the local map, not merely the integrated energy-preserving average.
Add the hypothetical condition that each panel sees a large black enclosure uniformly at 250 K, with effectively unit view factor to the enclosure. The net radiative flux from an ideal black patch is σ(T_patch⁴−250⁴). A patch warmer than 250 K has positive outward net flux; a colder patch has negative outward flux, meaning a net gain.
Panel U’s net power is σ(300⁴−250⁴) over one square metre, about 237.80 W outward. Panel V’s aggregate net is σ[(200⁴+400⁴)/2−250⁴], about 549.67 W outward.
The 200 K patch gains net radiation from the 250 K enclosure even while it emits its own radiation. The 400 K patch loses net radiation strongly enough that the whole panel still has an outward net exchange. This is why emitted power should not automatically be called ‘heat loss’ for every patch.
The difference between panels remains 311.87 W in this restricted scenario because the common environment term cancels for equal areas. Real radiative enclosures may have partial view factors, reflections and different surrounding temperatures; those effects cannot be dismissed without a defensible thermal model.
| Ideal exchange | Panel U | Panel V |
|---|---|---|
| Black surroundings | 250 K | 250 K |
| Emission | 459.30 W | 771.17 W |
| Common absorbed irradiation | 221.50 W | 221.50 W |
| Net outward power | 237.80 W | 549.67 W |
Replace the 200 K and 400 K halves with 280 K and 320 K halves, again of equal area. Their mean is still 300 K. The area average of the fourth powers is [(280⁴)+(320⁴)]/2=8.31616×10⁹ K⁴.
Compared with 300⁴=8.1×10⁹ K⁴, the emission factor is 1.02669. That represents approximately 2.67% greater blackbody emission, much less dramatic than our 200/400 K example.
The change provides a useful sensitivity check. The importance of nonuniformity depends on the size of the temperature spread, baseline temperature, area fractions and other sources of uncertainty. A small correction may be immaterial to one analysis yet decisive to another near a narrow tolerance.
If the hot patch covers only ten percent of the panel, its arithmetic influence and its radiation contribution have different weights from the equal-area example. Always check the actual area proportions before generalising from a two-patch demonstration.
A real material is rarely a perfect blackbody. Its emission is affected by emissivity, which can depend on wavelength, temperature, direction, surface oxidation and coating condition. Treating emissivity as a single constant is a simplifying gray-surface model, not a universal physical fact.
A mixed panel with different emissivities requires patchwise contributions such as AᵢεᵢσTᵢ⁴ under the gray approximation. A hot low-emissivity patch can behave differently from an equally hot high-emissivity patch. Temperature information alone is then even less sufficient.
View factors describe how much radiation from one surface reaches another surface. Internal panels can radiate toward each other; not all emitted power leaves a machine or enclosure. The NASA Passive Thermal Control Engineering Guidebook emphasises these geometric relationships alongside the fourth-power dependence.
A credible engineering calculation defines which surfaces exchange radiation, estimates their emissivities and identifies the relevant surroundings. It should also explain when convection and conduction are important enough to prevent a radiation-only model from describing the whole thermal balance.
Infrared cameras measure radiation in an instrument-specific spectral range and estimate surface temperature using calibration and assumptions. Reflected surroundings, emissivity settings, atmospheric conditions, focus, pixel mixing and saturation can change the displayed apparent temperature.
If a small hot feature occupies only part of one image pixel, the camera may mix its signal with cooler neighbours before any temperature estimate is made. Averaging pixel temperatures and then applying a fourth-power law can introduce another layer of information loss beyond the physical surface averaging illustrated earlier.
For a defensible emission estimate, record pixel areas, calibration, emissivity corrections, sensor resolution and uncertainty. Calculate the relevant patch fluxes from appropriately resolved, calibrated data before aggregating them. Do not claim a hot spot has been ruled out solely because the displayed average is low.
An independent reviewer should be able to identify which inputs were measured, which were assumed and which were inferred by a model. All temperatures in this article were stipulated rather than observed; our numerical results are demonstrations of physics under explicit assumptions, not device-specific safety conclusions.
Does equal mean temperature imply equal internal energy? No. Internal energy also depends on materials, mass, heat capacity and temperature fields within the solid. The given surface-mean equality contains none of that information.
Does a hotter patch always imply greater net heat loss to surroundings? Not without specifying what radiation the patch receives and the surrounding geometry. In our 250 K ideal environment, the 200 K patch has inward net flux while the hot patch has outward net flux.
Can an effective temperature reveal the hottest location? No. It reproduces one integral and may conceal local extremes. A peak-risk or material-damage decision needs more spatial detail than a one-number equivalent.
Practice A: use Stefan–Boltzmann to calculate outgoing emission for a black panel uniformly at 350 K. Practice B: compare equal-area 340 K and 360 K patches with a uniform 350 K panel. Practice C: take a panel with 25% area at 400 K and 75% at 250 K, and calculate its arithmetic mean and fourth-power effective mean separately. Give units and state every assumption.
Start by specifying the requested observable: outgoing emitted power, net thermal radiation, heat stored in a component, or a maximum local temperature. These quantities are not interchangeable, and no single average describes all of them.
Next preserve the physical boundaries. State surface area, patch sizes, emissivity, surroundings and whether self-view or reflections matter. Decide whether the intended engineering result also requires convection, conduction and internal energy terms.
Then evaluate the quality of measurement. Identify calibrated temperatures, sensor field of view, missing pixels, saturation and realistic uncertainty. The smallest hot features can be among the most important when a nonlinear power law is involved.
Finally reproduce the arithmetic with units and an uncertainty or sensitivity check. Compare patchwise sums with the shortcut based on a single mean. If the answers differ, explain why the selected decision requires the richer representation. More decimal places do not substitute for missing spatial data.
The worked one-square-metre example concludes that both panels have area-mean temperature 300 K, yet ideal outgoing power is about 459.30 W for the uniform panel and 771.17 W for the heterogeneous panel. This is an exact consequence of the stipulated fourth-power inputs, apart from reported rounding.
Continue with The Gold Standard of Thermodynamics and Finite Bodies, Entropy Generation and Heat Flow. For radiation measurement concepts, visit The Gold Standard of Photonics.
External technical references are NASA’s Blackbody Radiation Laws and the NASA Passive Thermal Control Engineering Guidebook. They describe the physics; neither supplies measurements from our invented panels.
Return to The Gold Standard hub and How X Works. The central habit is to apply the relevant physical transformation before averaging data, and to retain enough of the original observations to answer the question honestly.
Advanced worked companion: sensitivity and area weighting
These extensions keep the same ideal blackbody law but alter one assumption at a time. They are original mathematical checks, not thermal survey measurements. Use them after the main fifteen chapters when evaluating whether a proposed ‘average temperature’ summary preserves the quantity needed for a decision.
Check individual patch contributions before comparing totals
In the original heterogeneous panel, the cold 200 K half contributes 0.5×σ×200⁴≈45.36 W. The hot 400 K half contributes 0.5×σ×400⁴≈725.81 W. Their sum is 771.17 W. The larger patch contribution is sixteen times the smaller because both areas are equal and the temperatures differ by a factor of two. The uniformly 300 K panel emits 459.30 W.
These figures allow a reviewer to reconstruct the calculation independently of the reported total. A thermal dashboard showing a mean of 300 K does not reveal whether the hottest half is responsible for most of the emitted power. Preserving even a simple patch table helps distinguish a genuine nonlinear physical effect from a data-processing mistake.
If a hot pixel covers one square centimetre and a cool pixel covers twenty, their contributions must not be averaged as though they have equal emitting area. The radiation law acts pointwise, while integration sums contributions with the correct spatial support. This distinction remains important when converting an irregular thermal image or finite-element mesh into a total heat-flow estimate.
Derive a reusable formula for symmetric temperature variation
Let two equal-area patches have absolute temperatures μ−δ and μ+δ, where μ is their arithmetic mean and δ is a positive temperature departure. Expand the two fourth powers and average: [(μ−δ)⁴+(μ+δ)⁴]/2=μ⁴+6μ²δ²+δ⁴. Odd-power terms cancel, but the quadratic and quartic terms survive.
For μ=300 K and δ=100 K, the excess over μ⁴ is 6×300²×100²+100⁴=5.5×10⁹ K⁴. Add that to 300⁴=8.1×10⁹ K⁴ to obtain 13.6×10⁹ K⁴, matching the original calculation. This check independently verifies the result without computing each patch’s fourth power separately.
For δ=20 K, the excess becomes 6×300²×20²+20⁴=216,160,000 K⁴. That is about 2.67 percent of 300⁴. The expression shows explicitly why small spatial variations have an effect that begins quadratically in δ when two equally weighted deviations are symmetric about the mean.
Change patch areas while keeping the same mean
Suppose a quarter of the area is at 400 K and the other three quarters at 266.666… K. The weighted arithmetic mean is 0.25×400+0.75×(800/3)=300 K. Yet the mean fourth power is 0.25×400⁴+0.75×(800/3)⁴≈10.19259×10⁹ K⁴, greater than 300⁴=8.1×10⁹ K⁴.
For one square metre, ideal emitted power is 577.96 W, compared with 459.30 W for uniform 300 K. This unequal-area case differs from the original 50/50 case even though both share the same arithmetic mean. Area fractions are therefore part of the physical evidence, not cosmetic plotting metadata.
A real sensor image may have pixels of identical angular size that do not represent identical physical surface areas when the viewed object curves or recedes from the camera. Geometry, calibration and emissivity corrections are needed before a pixel-weighted arithmetic mean can be called an area-weighted thermal mean. Use an appropriate surface mesh when the shape is important.
Choose a meaningful reporting pair instead of one unlabelled number
A concise technical report can present both the area-weighted arithmetic temperature and the radiation-equivalent effective temperature, plus the maximum observed or modelled local temperature. These numbers answer different questions: the first describes a spatial first moment, the second preserves a fourth-power blackbody emission integral, and the third flags a possible local thermal limit.
Add the total area, emissivity assumptions and incoming-radiation boundary conditions beside those figures. The key engineering distinction is between emission and net exchange. A black surface colder than its surroundings can absorb more radiation than it emits, even though its own Stefan–Boltzmann exitance remains positive.
No one-number summary can recover a detailed temperature map after the spatial information has been discarded. If the decision depends on a defect occupying a small hot region, preserve or inspect sufficiently resolved source data and report uncertainty about saturation, subpixel mixing and material properties.