Smoke stratification

Smoke stratification is the layering of smoke in a space according to its temperature and density. In a room fire, hot smoke collects under the ceiling and forms a layer above cooler, cleaner air. Van Rossum calls this a stratified situation, as opposed to homogeneous mixing, in which the whole room is filled with smoke. In fire safety engineering, the term is also used for a special case in tall spaces. There the rising smoke stops below the ceiling, because the air at higher levels is as warm as the cooled smoke.

This intermediate stratification matters for detection and smoke control in atria and other high spaces. Ceiling-mounted detectors may respond late or not at all, because the smoke does not reach them. Simple relations give the height to which a plume can rise and the smallest fire that drives smoke to the ceiling. Measured temperature differences of about 50 °C between floor and ceiling have been reported in atria with glazed roofs. Where stratification is expected, Milke recommends detectors that do not rely on smoke reaching the ceiling, such as beam detectors placed below the expected smoke level.

Layered smoke in rooms

A fire produces a plume of hot gases that rises and draws in cold air from its surroundings. When the temperature difference between the plume and the surrounding air is large enough, the space separates into layers, with a smoke layer at the top. The glossary of NUREG-1805, a set of fire hazard methods of the United States Nuclear Regulatory Commission (NRC), defines stratification in the same way. Smoke settles in layers according to its density, with the heaviest layer at the bottom, and smoke layers usually collect from the ceiling down.

This layered picture is the basis of the two-zone model. In a local fire before flashover, a zone model assumes a warm hot gas layer above a cold layer that is free of smoke, or nearly so. The layered state does not always exist. When the conditions for flashover are reached, the room can be treated as well mixed, and the model switches to a single zone. Zone models also assume a stable layer without mixing at the interface.

Real layers have no sharp boundary. In practice, the transition from the upper to the lower layer extends over about 10 % of the room height, and more in weakly stratified flow. A burning cigarette in a normal room is given as a fire that lies outside the scope of a zone model. Van Rossum adds that the temperature of a real smoke layer also varies between the plume and the edges of the room.

The zone model CFAST is therefore generally limited to rooms in which the gas is strongly stratified. Other processes can destroy the layering. Mechanical ventilation can make the layer unstable and mix smoke into the supply air. Sprinkler spray cools the smoke and stirs the layer, driving some smoke down to floor level. Some stratification remains, but much less than in an unsprinklered fire.

Why rising smoke can stop

The smoke from a fire rises in a fire plume. As the plume rises, it draws in surrounding air and cools. In buildings the air temperature normally increases with height. If the fire is weak, the temperature difference that gives the plume its buoyancy can vanish and even reverse, and the plume then stops rising.

Theory shows that the upward velocity falls to zero at the maximum height, while the plume widens rapidly as it approaches that height. The concentration of combustion products on the centreline is hardly affected by the stratification. Schifiliti, Meacham and Custer describe the result. Once the plume temperature equals that of the surrounding air, the smoke stays at that height, whatever the height of the ceiling, until the fire supplies more energy.

Warm air near the ceiling has several causes. A glazed roof can create a warm layer through solar heating before any fire starts. The upper part of an atrium or mall may also be left unconditioned when nobody uses it. Schifiliti and co-workers give examples of air-conditioned rooms with temperature differences of 10 and 20 °C between floor and ceiling.

Milke distinguishes two simple temperature profiles. In the first, more likely, case the air is uniform up to a certain height, with a warm layer of uniform temperature above it. In the second case the temperature increases steadily from floor to ceiling.

For a warm layer above a certain height, the plume is checked at the lower edge of that layer. If the temperature on the centreline of the plume has fallen to the temperature of the warm layer, the plume is no longer buoyant and spreads out at that height. Milke gives an example of a 20 m atrium with a temperature step of 30 °C at 15 m above the floor. The smoke from fires of 1 MW and 2 MW then stratifies at the height of the step in both cases.

Height of rise

For a temperature that increases uniformly with height, the maximum height reached by a plume follows from theoretical and experimental work by Morton and co-workers, as presented by Heskestad. It grows only with the fourth root of the convective heat release rate, and it falls as the temperature gradient becomes steeper: $$z_m=3.79{\left(\frac{T_{a1}}{g{\left({\rho }_{a1}{c}_p\right)}^2}\right)}^{1/8}{\dot{Q}}_c^{1/4}{\left(\frac{\mathrm{d}{T}_a}{\mathrm{d}z}\right)}^{-3/8}$$

For standard air, Milke writes the same relation in a compact form, with the temperature difference $\mathrm{\Delta }{T}_o$ between floor and ceiling over the height $H$. Setting the maximum height equal to the ceiling height gives the smallest convective heat release rate that just drives smoke to the ceiling: $$z_m=5.54{\dot{Q}}_c^{1/4}{\left(\frac{\mathrm{\Delta }{T}_o}{H}\right)}^{-3/8},{\dot{Q}}_{c,\min }=0.00118H^{5/2}\mathrm{\Delta }{T}_o^{3/2}$$

Symbol	Description	Unit
$z_m$	maximum height of rise of the plume above the fire	m
${\dot{Q}}_c$	convective heat release rate of the fire	kW
$T_{a1}$, ${\rho }_{a1}$	ambient temperature and density at the level of the fire	K, kg/m³
$c_p$	specific heat of air	kJ/(kg·K)
$g$	acceleration due to gravity	m/s²
$\mathrm{d}{T}_a/\mathrm{d}z$	ambient temperature gradient	K/m
$\mathrm{\Delta }{T}_o$	ambient temperature difference between floor and ceiling	K
$H$	height of the ceiling above the fire	m

The convective part of the heat release rate is often taken as 0.7 of the total when no specific data are available. For air at 293 K with a density of 1.2 kg/m³ and a specific heat of 1.0 kJ/(kg·K), the constant group in the first equation equals 5.54. This is the constant of the second equation. The design chapter on detection in the same handbook gives a similar relation with a coefficient of 0.352 for a heat release rate in watts. That is 0.000352 in kilowatts, about 3.4 times smaller than Milke's coefficient. NUREG-1805 includes a check of this kind, taken from NFPA 92B of the National Fire Protection Association, before it estimates the response time of smoke detectors.

Heskestad compares the theory with fire experiments in stratified spaces. These largely confirmed it, except that the plume needed about 25 % more height than predicted before its temperature rise vanished. From the same experiments he gives a simple criterion. A plume just fails to pass a level when the ambient temperature rise up to that level is 7.4 times the temperature rise that the same fire would produce there in a uniform room. This criterion is surprisingly insensitive to the shape of the temperature profile. For a profile in which one half of the ambient temperature rise occurred above 75 % of the height, the critical value was only 12 % higher.

Worked example

Heskestad considers an atrium 20 m high in which the air temperature rises by 5 K from floor to ceiling. The question is how large a fire on the floor must be to drive its plume to the ceiling. His criterion requires a centreline temperature rise of 5 / 7.4 ≈ 0.68 K at 20 m in a uniform room. With his plume equation for normal conditions, this gives a convective heat release rate of 7.9 kW. Taking the convective part as 0.7 of the total, the fire needs about 11.4 kW. For a ceiling twice as high, 40 m, about 64 kW is needed.

Milke's relation for the same atrium gives 0.00118 × 20^2.5 × 5^1.5 ≈ 23.6 kW of convective heat release, three times Heskestad's value. The detection relation of Schifiliti and co-workers gives about 7.0 kW. A calculation in Python illustrates the size of the effect found by Heskestad. If the plume rises 25 % higher than the theory predicts, the theory only needs to reach 16 m, which requires about 8.7 kW. This is close to the two lower results.

Schifiliti and co-workers apply their relation to two detection problems. In a room 7 m high with a temperature difference of 10 °C, smoke from overheated cable insulation at 1,000 W would rise only about 6 m. Detectors at the ceiling would therefore not meet the design objective. In a 20 m atrium with a difference of 20 °C, a burning chair of 280 kW would drive smoke up to about 38 m, well above the ceiling. The authors add that these estimates ignore the effects of heating and air conditioning systems and of warm or cold walls.

Detection

A smoke detector at the ceiling can only respond if smoke reaches it. Where the air near the ceiling is warmer than the rising smoke, Milke warns that ceiling-mounted devices may activate late. Schifiliti and co-workers add that smoke that reaches the level of a detector still has to enter its sensing chamber. Air movement can work either way. A system that supplies air low in a room and exhausts it high up may help to carry smoke upwards, while in other cases it may hinder smoke movement. Strong supply and return flows can also keep smoke away from detectors outside their flow paths, especially in the early stage of a fire when the plume is weak.

If the smoke is expected to stratify below the ceiling, Milke advises considering detectors other than ceiling-mounted ones, such as projected beam detectors. These should be placed below the height of stratification, so that the beam crosses the rising plume. NUREG-1805 explains that a beam smoke detector works on the principle of light obscuration over a beam that spans the protected area. Such detectors are most often used as line detectors in large open areas.

NUREG-1805 also describes aspirating smoke detection. A fan in the detector draws air from the protected area through sampling ports, pipes or tubes, and the detector analyses the air for fire products. The Fire Dynamics Simulator (FDS), a computational fluid dynamics model, can simulate both types. For a beam detector it integrates the obscuration along the beam. For an aspirating detector the user specifies the sampling points, their flow rates and the transport time to the detector.

Smoke control

Intermediate stratification is one of the special conditions that Milke lists for smoke management in malls and atria. The others are confined flow in narrow spaces, plugholing and the supply of make-up air. These conditions affect when an active system starts and how efficiently its exhaust fans work. Stratification below the ceiling can delay the signal that starts atrium smoke control, because ceiling detectors may not see the smoke.

Once the smoke management system is running, Milke recommends exhausting the warm air layer, so that the smoke can rise to the ceiling. The maximum height of rise also depends on the stage of the fire. Early after ignition it is set by the convective heat release rate and the temperature profile of the space. A growing fire supplies more energy, and Schifiliti and co-workers note that the smoke keeps its height only until additional energy is provided.

Modelling and limitations

Two-zone models assume a strongly stratified room, with a sharp boundary between a single hot layer and a single cool layer, whereas the real transition is gradual. The network model CONTAM treats every zone as fully mixed and ignores stratification within a zone altogether. Computational fluid dynamics can include the temperature profile of the space. FDS allows a linear increase of the ambient temperature with height to be specified.

A pre-existing stratification that is left out of a model can spoil the results. In the validation study of the NRC and the Electric Power Research Institute (EPRI), much of the uncertainty of FDS came from tests with small fires in large, drafty aircraft hangars. The layer temperatures were only a few tens of degrees above ambient, and the substantial temperature stratification in the hangars was not included in the model.

The simple relations rest on idealised profiles: a uniform gradient or a single warm layer. They do not consider air movement from heating and ventilation systems, or warm and cold walls. The coefficients in the handbook differ by a factor of about 3.4, and the experiments of Heskestad suggest that real plumes rise higher than the theory predicts.

See also

• Smoke reservoir
• Smoke and heat exhaust ventilation
• Plugholing
• Balcony spill plume
• Smoke control

References

• Ferreira, M. J., & Strege, S. M. (2005). Smoke management systems (NIST NCSTAR 1-4D). National Institute of Standards and Technology.
• Heskestad, G. (2002). Fire plumes, flame height, and air entrainment. In P. J. DiNenno, D. Drysdale, C. L. Beyler, W. D. Walton, R. L. P. Custer, J. R. Hall Jr., & J. M. Watts Jr. (Eds.), SFPE handbook of fire protection engineering (3rd ed., pp. 2-1–2-17). National Fire Protection Association; Society of Fire Protection Engineers.
• Iqbal, N., & Salley, M. H. (2004). Fire Dynamics Tools (FDTs): Quantitative fire hazard analysis methods for the U.S. Nuclear Regulatory Commission Fire Protection Inspection Program (NUREG-1805). U.S. Nuclear Regulatory Commission, Office of Nuclear Reactor Regulation.
• McGrattan, K., Hostikka, S., Floyd, J., McDermott, R., Vanella, M., Mueller, E., & Paul, C. (2026). Fire Dynamics Simulator technical reference guide: Vol. 1. Mathematical model (6th ed., NIST Special Publication 1018-1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.SP.1018
• Milke, J. A. (2002b). Smoke management in covered malls and atria. In P. J. DiNenno, D. Drysdale, C. L. Beyler, W. D. Walton, R. L. P. Custer, J. R. Hall Jr., & J. M. Watts Jr. (Eds.), SFPE handbook of fire protection engineering (3rd ed., pp. 4-292–4-310). National Fire Protection Association; Society of Fire Protection Engineers.
• Peacock, R. D., McGrattan, K. B., Forney, G. P., & Reneke, P. A. (2026). CFAST – Consolidated Fire and Smoke Transport: Vol. 1. Technical reference guide (Version 7, NIST Technical Note 1889v1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.TN.1889v1
• Schifiliti, R. P., Meacham, B. J., & Custer, R. L. P. (2002). Design of detection systems. In P. J. DiNenno, D. Drysdale, C. L. Beyler, W. D. Walton, R. L. P. Custer, J. R. Hall Jr., & J. M. Watts Jr. (Eds.), SFPE handbook of fire protection engineering (3rd ed., pp. 4-1–4-43). National Fire Protection Association; Society of Fire Protection Engineers.
• Stroup, D., Lindeman, A., McGrattan, K., Peacock, R., Overholt, K., Joglar, F., LeStrange, S., & Montanez, S. (2016). Verification and validation of selected fire models for nuclear power plant applications: Supplement 1 (NUREG-1824, Supplement 1; EPRI 3002002182). U.S. Nuclear Regulatory Commission, Office of Nuclear Regulatory Research; Electric Power Research Institute.
• Van Rossum, E. (2027). Introduction to Fire Safety Engineering. Fire Safety Academy.