Thermal bridging at the rooflight perimeter
The glass in a modern rooflight is usually the best-performing part of the installation, and the 300mm of construction around it is usually the worst. A kerb, a frame junction, an insulation line that stops 20mm short and a plasterboard reveal together form a continuous ring of higher heat loss around every unit on the roof. That ring is a linear thermal bridge, it is measured in watts per metre kelvin rather than watts per square metre, and on a small rooflight it can lose as much heat as a meaningful fraction of the glass itself. This page is about that perimeter: how the heat leaves, what the numbers look like, and what has to happen on site while the roof is open. The wider picture sits on the energy efficient rooflights hub, and the separate question of the spacer bar at the edge of the glass is covered on warm edge spacers and where condensation starts.
A bridge is a shortcut, and the perimeter is full of them
Heat takes the path of least resistance out of a building in exactly the way water takes the path of least resistance in. Where the insulation layer is continuous, the resistance is high and the flow is slow. Where the insulation is interrupted by something more conductive, timber, steel, concrete or simply an air void, the heat concentrates and pours through the gap.
At a rooflight, the insulation layer of the roof has to be cut through completely to make the opening. Whatever happens next either re-establishes continuity around the hole or leaves a break in it. There is no third option, and the decision is made in about two hours on the day the kerb is built.
Psi values, and the unit that describes a junction rather than a surface
A U-value describes an area of construction and carries the units W/m2K. A junction is a line, not an area, so it is described by a linear thermal transmittance, written as the Greek letter psi and carrying the units W/mK. It states the extra heat lost per metre of that junction, over and above what the flat elements either side of it were already going to lose.
The arithmetic that follows is simple and unforgiving. A 1200 by 800mm rooflight has a perimeter of four metres. At a psi of 0.10 W/mK that junction adds 0.40 W/K of heat loss. The glass itself, at 0.96 square metres and a U-value of 1.2, loses 1.15 W/K. So a competently detailed perimeter is already worth about a third of the glass. At a psi of 0.30, which is entirely achievable with an uninsulated timber kerb, the perimeter is losing more than the glazing it surrounds.
The four heat paths around a kerb-mounted unit
On a flat roof unit the perimeter is not one bridge but four, stacked on top of each other, and they are usually detailed by different trades on different days.
- The deck junction, where the roof insulation runs up to the outside face of the kerb and either meets it or stops short of it.
- The kerb wall itself, a short section of timber structure standing 150mm proud of the roof and exposed to outside air on three faces.
- The frame-to-kerb line, where the aluminium frame sits down on the top of the kerb, usually on a compressible tape, with fixings through it.
- The internal reveal, where the ceiling insulation and the plasterboard return up into the opening from below.
A single weak link in that chain sets the performance of the whole ring. Insulating three of the four beautifully and leaving the fourth hollow gives you the psi value of the hollow one.
Why a bare timber kerb behaves like a cold ring
Softwood conducts at roughly 0.13 W/mK. That is far better than aluminium and far worse than any insulation, which sits between about 0.020 and 0.038. A 150mm high, 100mm wide treated timber kerb standing in the wind is a solid conductive collar joining the warm interior to the outside air, and it does that job continuously, all winter, around the entire opening.
The consequence is not only heat loss. The internal face of that timber sits several degrees below room temperature, so it becomes the surface where moisture lands first. Discolouration in a band above the ceiling line, in the corners of the reveal, or a musty smell in a room that is otherwise dry, are all the visible end of a kerb that was never insulated through its depth.
Three ways to insulate a kerb, and only one of them works properly
Insulating the inside face only puts the insulation on the warm side, which keeps the room surface warm but leaves the timber itself cold and exposed to whatever vapour reaches it. Insulating the outside face only, which is the more common site habit, keeps the timber warmer but still leaves a thermal break at the top of the kerb where the frame lands, and at the bottom where the insulation is trimmed around the deck upstand.
What actually works is insulating the kerb through its full depth and wrapping it: the void between the studs filled solid, a continuous layer taken up the outside face and lapped onto the roof insulation below, and the top of the kerb carrying an insulated bearing rather than bare timber under the frame. Done that way the timber is on the warm side of the insulation line from top to bottom and the ring disappears.
Continuity at the deck: the lap that has to overlap
The single most valuable 50mm on the whole installation is where the field insulation of the warm deck meets the insulation wrapping the kerb. If those two layers butt loosely, there is an air gap along a four-metre line at the coldest part of the roof. If they lap, the line is closed.
On a warm deck the field boards are usually PIR at 100 to 150mm, and the practical detail is to cut them tight to the kerb, then run a separate 25 to 50mm layer up the kerb face so it laps over the top edge of the field boards rather than sitting beside them. Any residual gap is filled with expanding insulation rather than left. Two millimetres of daylight visible along that joint from above is a heat path that will still be there in thirty years.
Fixings, brackets and the bridges you cannot see from the ground
A linear bridge runs along a line. A point bridge is a single penetration, described by a chi value in W/K, and on its own it is small. Repeated at 300mm centres around a perimeter it stops being small.
Stainless steel fixings conduct at around 17 W/mK and carbon steel at around 50. A run of steel screws passing straight through an insulated kerb into the deck below carries a measurable amount of heat, which is why the better detail fixes the frame down into the timber of the kerb itself rather than through the insulation layer, and why insulated bearing shims exist. Restraint brackets on a lantern are the worst offenders, being larger in section and often bedded directly against the deck.

In-plane roof windows: the collar around the frame
A VELUX, Fakro, Keylite or Roto roof window sits in the plane of the pitched roof, so its perimeter bridge is the timber trimmer and the rafter faces either side rather than a kerb. The bridge is shorter but the same principle applies: rafter timber at 0.13 W/mK, doubled trimmers at the head and cill, and a gap around the frame that has to be filled without compressing anything.
All four manufacturers make an insulation collar sized to their own frames, and it is worth using rather than stuffing loose mineral wool into the gap. The collar sits in the 20 to 30mm space between the frame and the trimmed opening, is the right thickness by design, and does not slump after two winters the way an offcut of quilt does. Detailing for the units themselves is on the roof window installation page.
The reveal, the 20mm void, and the black spotting in February
The plasterboard reveal is built last, by a different pair of hands, often after the scaffold has gone. It is where the insulation line most often ends 20mm short of where it should.
The rule is that the insulation runs continuously from the ceiling plane, up the reveal, and lands tight against the frame with no void behind the board. Where a splayed reveal is being formed, which is worth doing anyway for the daylight, the insulation follows the splay rather than staying square behind it. A hollow triangle of still air at the head of a reveal is a cold surface with warm moist air rising straight into it, which is precisely the recipe for the dark spotting that appears in the top corners in February and nowhere else in the room.
Typical psi values, and what a good junction looks like on paper
Published figures vary by manufacturer, kerb build-up and modelling assumptions, so these are industry-typical ranges rather than a specification. They are useful for the order of magnitude, which is the point.
| Perimeter detail | Typical psi range (W/mK) | What is happening |
|---|---|---|
| Uninsulated timber kerb, roof insulation butted to it, plasterboard reveal with a void | 0.30 to 0.60 | Solid conductive collar, unbroken from inside to outside air |
| Kerb insulated on the outside face only, reveal insulated | 0.18 to 0.30 | Better, but the top of the kerb under the frame is still a break |
| Kerb insulated full depth and wrapped, insulation lapped onto the deck layer | 0.06 to 0.12 | Continuous line from field insulation to frame |
| Insulated composite kerb with a thermally broken bearing under the frame | 0.03 to 0.08 | The best commonly available detail for a flat roof unit |
| In-plane roof window with a manufacturer’s insulation collar | 0.04 to 0.10 | Short junction, insulation collar sized to the frame |
| In-plane roof window with loose quilt stuffed around the frame | 0.15 to 0.30 | Slumping and compression open a gap within a few winters |
Temperature factor, and the threshold that predicts mould
Heat loss is the money question. Surface temperature is the health question, and they are assessed separately.
The temperature factor, written fRsi, expresses how warm the internal surface at a junction stays relative to the difference between inside and outside. It runs from 0 to 1, and the widely used threshold for dwellings is 0.75. Below that, under normal domestic humidity, the surface spends enough of the winter below dew point for mould to establish. A well-wrapped kerb comfortably clears 0.75. A hollow one at the reveal head does not, which is why the failure is so predictable and so localised: it is always the top corners, always in the coldest weeks, and always in the same 100mm.
Modelling on paper against a thermal camera in February
Psi values come from two-dimensional or three-dimensional numerical modelling of the junction, run to a defined convention so that different details can be compared fairly. That is how a manufacturer arrives at a published figure and how a designer builds one into a calculation.
On an existing house, the faster answer is a thermal image taken on a cold morning with the heating on and a decent temperature difference between inside and out. A bare kerb reads as a bright rectangle around the opening. A missing area of reveal insulation reads as a bright corner. It costs nothing but the right morning, and it settles arguments about whether the problem is the glass or the perimeter before anybody spends money on new glazing. The glazing side of that question is on U-values and glazing specification.
What the perimeter does to a whole-house calculation
In a compliance calculation for a new element or an extension, thermal bridging is accounted for either by using published values for each junction or by applying a default. Defaults are deliberately pessimistic, because the calculation cannot assume workmanship it has no evidence for.
The practical effect is that a project which uses good published junction details and builds them properly gets credit for them, and a project which does not is charged for the assumption. On a loft conversion with three roof windows, the perimeter length adds up quickly: three units at four metres of perimeter each is twelve linear metres of junction, which is more junction than most of the rest of the conversion put together. What Part L asks of a replacement unit specifically is set out on rooflights and Part L for a replacement.
Gloucestershire roofs, and where the perimeter is hardest to get right
On Cotswold stone slate through Painswick, Bibury and Northleach, the covering is 20 to 30mm thick and the batten and torching build-up above the rafters is deeper than anything a standard flashing kit assumes. The insulation line has to be worked out in section rather than assumed, and a conservation unit sitting nearly flush in a thick roof leaves less room around the frame than a modern window would.
On the flat rear extensions that exist in every material across Gloucester, Tewkesbury and the vale, the usual finding when an old dome comes off is an untreated softwood kerb with nothing in it at all. The kerb is rebuilt insulated rather than reused, because the new glass will otherwise be surrounded by the same cold ring that ruined the last reveal.
On trussed estates at Quedgeley, Abbeydale and Bishops Cleeve, the opening usually sits between two trusses at 600 centres, leaving 550 to 570mm of clear width. That is a tight space to run a collar in, and it is worth ordering the manufacturer’s collar with the unit rather than improvising on the day. In the Stroud valleys, where scaffold is hand-carried and every extra visit is expensive, the perimeter is detailed once while the access is there.
What gets done at the perimeter as standard
The perimeter is one of the four invisible details the ten-year workmanship guarantee is written against, alongside upstand height, falls and flashing laps. All four are buried by the time anyone looks at the finished job, which is exactly why they are the ones worth being specific about.
On a kerb-mounted unit that means a kerb insulated through its depth and wrapped, the wrap lapped over the field insulation rather than butted, an insulated bearing under the frame, fixings placed to avoid passing through the insulation layer, and the reveal insulation carried tight to the frame with no void behind the board. On an in-plane window it means the manufacturer’s collar, fitted uncompressed, with the vapour layer taped to the frame above it. The details go in the survey notes so the specification is written down before the scaffold goes up, and the flat roof rooflight installation page covers the kerb build-up in full.
Next on this subject.
- Guide U-values and glazing specification A U-value is the rate at which heat crosses one square metre of a finished rooflight for every degree of…
- Guide Solar gain, overheating and shading Overhead glazing collects far more solar energy in July than in January, and the ratio is not close. A rooflight…
- Guide Condensation, ventilation and rooflights Surface condensation on a rooflight is a whole-house ventilation problem showing up at the coldest point in the building. The…
- Guide Triple glazing in a rooflight and when it pays Triple glazing in a roof is worth more than triple glazing in a wall, and for a reason that has…
- Guide Warm-edge spacers and where condensation starts Watch any rooflight through a cold week and the water always arrives in the same place first: a band following…
- Guide Low-emissivity coatings and argon fills Roughly three quarters of the heat crossing an uncoated sealed unit travels as infrared radiation from the warm inner pane…