The ice barrier line, and why no map can draw it

What an ice barrier is actually for

Start with the failure, because the code section makes sense only as a response to it. An ice dam is not a weather event. It is a heat event that happens to involve snow.

Snow sits on a roof. Heat escapes from the rooms below through gaps in the ceiling, a thin attic floor, leaky ducts and recessed lights, and warms the deck above the heated part of the house. Snow touching that warm deck melts, even while the air outside is well below freezing.

The meltwater runs down the roof under the snowpack until it reaches the part of the roof that is not over heated space: the overhang past the exterior wall, the eave. That surface is at the outdoor temperature. The water refreezes there and builds a ridge of ice, and the next meltwater ponds behind it.

That ponded water is the problem. Shingles are a shedding system, not a waterproof one. Asphalt shingles are lapped to shed water running downhill, and they have no defense against water sitting still or pushed uphill by capillary action.

It runs into the sheathing, down the top plate of the wall, into the insulation and out through the ceiling, and the homeowner sees a stain on the bedroom ceiling in February and blames the roof covering.

The ice barrier is a membrane, not a shingle accessory: two layers of underlayment cemented together, or a self-adhering polymer-modified bitumen sheet. That is a peel-and-stick rubberized asphalt membrane that bonds to the deck and seals around every nail driven through it.

It does not stop ice forming. It makes the part of the roof most likely to sit under water behave like a waterproof roof rather than a shedding one.

Two things follow from the mechanism. First, the requirement is measured from the wall, not the roof edge. The dam forms where heated space stops, so the membrane has to run past that line, hence "24 inches inside the exterior wall line" rather than "36 inches from the eave".

A house with a deep overhang needs a wider run of membrane than one with a clipped eave. A specification written in feet from the gutter can comply on one house and fail on the next.

Second, an ice barrier treats the symptom. The cause is heat and air leaking into the attic, and the fix is air sealing, insulation and enough roof ventilation. A ventilation shortfall keeps making ice dams under a well-installed membrane.

Our ventilation calculator works out the net free area a roof of a given size needs, which is the other half of this conversation.

What the section actually says, and which edition we read

Code claims go stale, and a claim from the wrong edition is worse than none. The text below is Section R905.1.2 as printed in the 2021 Seattle Residential Code, which publishes the 2021 IRC in full with Seattle’s own amendments flagged.

Seattle marks its amendments with a bracketed [S] and strikes deleted model text; R905.1.2 carries neither, so what follows is the model 2021 IRC language. We could not read the text at ICC’s own Digital Codes library, which returned an access refusal to every request we made on September 3, 2026.

"R905.1.2 Ice barriers. In areas where there has been a history of ice forming along the eaves causing a backup of water as designated in Table R301.2, an ice barrier shall be installed for asphalt shingles, metal roof shingles, mineral-surfaced roll roofing, slate and slate-type shingles, wood shingles and wood shakes.

The ice barrier shall consist of not fewer than two layers of underlayment cemented together, or a self-adhering polymer-modified bitumen sheet shall be used in place of normal underlayment and extend from the lowest edges of all roof surfaces to a point not less than 24 inches (610 mm) inside the exterior wall line of the building.

On roofs with slope equal to or greater than 8 units vertical in 12 units horizontal (67-percent slope), the ice barrier shall also be applied not less than 36 inches (914 mm) measured along the roof slope from the eave edge of the building. Exception: Detached accessory structures not containing conditioned floor area."

Five details in that paragraph are easy to miss.

One: the trigger is a cross-reference, not a criterion. "As designated in Table R301.2" means the code has delegated the question. If your jurisdiction’s Table R301.2 says NO, R905.1.2 imposes nothing on you no matter how cold it gets.

Two: it is a list of coverings, and the list is finite. Asphalt shingles, metal roof shingles, mineral-surfaced roll roofing, slate and slate-type shingles, wood shingles, wood shakes. Clay and concrete tile, standing-seam metal panels and low-slope membrane systems are governed by their own sections with their own underlayment rules.

Three: 24 inches is measured from the wall, inward. Not from the eave, gutter or fascia. With a 24-inch overhang, the membrane has to reach four feet in from the roof edge, measured flat, and further along a steep slope. A single 36-inch roll along the eave can fail this.

Four: "all roof surfaces" is plural on purpose. Every lower edge of every roof plane, including dormers, porch roofs and lower runs above bay windows, not only the main eave.

Five: the exception is narrow. "Detached accessory structures not containing conditioned floor area" exempts an unheated detached garage or shed. An attached garage is not detached, and a heated workshop is not unconditioned.

The exception makes physical sense. An unheated detached building has no heat loss to melt the snowpack, so it does not form ice dams the way a house does.

Seven jurisdictions, verified one at a time

We did not build a fifty-state map. We could not source one jurisdiction by jurisdiction, and a state-level color for a rule set below state level would be an invention with a legend. Instead we checked individual determinations against each authority’s own document. The spread in that small sample is the finding.

At one end is Seattle, whose published Table R301.2 carries a flat "No" in the ice barrier column. Seattle has a 20 psf ground snow load, a 52.8°F mean annual temperature, a January mean of 42.8°F and 1.8 inches of January snow in the 1991–2020 normals. The determination fits a wet, mild winter.

At the other end is Grand County, Colorado, which did not merely tick YES. Its published amendment package rewrites R905.1.2 to require a self-adhering polymer-modified bitumen sheet on all sloped roofs, from the eaves to a point six feet inside the exterior wall: three times the model code’s 24 inches.

It also asks for 24 inches from the centerline of every valley, fully adhered, on all habitable structures. A roof built to the model code there would fall well short.

Between those two, Montana settled the question for the entire state in a 2016 technical advisory from its Building Codes Bureau, on the reasoning that conditions "throughout the entire State of Montana" produce ice formation: a statewide YES issued as an interpretation rather than as fifty separate table entries.

The Capitol Region Council of Governments in Connecticut publishes "Yes*" with the asterisk defined on the same page as "Established by the jurisdiction": the clearest statement we found that this cell is a decision, not a reading.

The Town of Mamaroneck, New York says YES on a 25 psf ground snow load and a 52.2°F mean annual temperature.

HyreRoof analysis: Mamaroneck and Seattle publish mean annual temperatures within 0.6°F of each other (52.2 and 52.8) and reach opposite determinations. That is the provision working as drafted. A snow-and-freeze winter and a rain winter can share an annual average, and the local history of eave ice tells them apart.

The climate the criterion is supposed to be tracking

If the code will not draw the line, the climate record shows where one could go. The IRC gives no number. The IBC formulation quoted in the Montana advisory does: "areas where the average daily temperature in January is 25 degrees F or less".

That is the only numeric trigger in this family of provisions, and it is a January mean rather than a design low.

HyreRoof analysis. We pulled the 1991–2020 monthly normals for 62 first-order stations from NOAA. 16 of them have a January mean temperature at or below 25°F. Chicago sits almost exactly on the line, at 25.2°F: a rounding error from the threshold.

A temperature threshold on its own is the wrong tool, and the scatter above shows why. Two things have to be present at once: snow on the roof, and an eave below freezing while the upper deck is above it.

Applying both, January mean at or below 32°F and at least three inches of January snowfall, a screen we chose for being crude enough to argue with, brings 33 of the 59 stations with a published snowfall normal into scope.

The edges of the screen are the interesting part. None of our stations sits at or below freezing in January with under three inches of snow. The nearest cases are Grand Junction (27.7°F, 4.5 inches) and Wichita, just above freezing at 33.2°F with 2.7 inches. Cold with no snowpack does not make an ice dam.

Meanwhile 8 stations get real January snow above a 32°F January mean, where snow arrives and leaves before it can be worked on by a full melt-freeze cycle.

Why freeze–thaw counting does not answer it either

The next idea is to count the days the temperature crosses freezing, since an ice dam freezes and thaws. We built that measure. It does not work, and the reason is worth the space.

HyreRoof analysis. Denver has the most freeze–thaw crossing days in our sample, 91 between December and March, more than any station in New England or the upper Midwest. A dry, sunny, high-altitude winter day can start below freezing and end above it.

Grand Junction is second at 84. Albuquerque is above Buffalo. Casper is above Syracuse. Duluth has only 32 crossing days and Fargo 36, because in a cold place the temperature stays under the line for weeks.

Duluth gets 16.8 inches of snow in January and Fargo 10.3. Denver gets 6.4 and Grand Junction 4.5.

Crossing days measure the air, and ice dams are made by the roof. What melts the snowpack in an ice dam is heat leaking out of the house, not a sunny afternoon. The roof can be melting at the ridge while the air has stayed below freezing for two weeks.

A Minnesota winter that stays below freezing for weeks under a deep snowpack suits ice dams. A Colorado winter that thaws most afternoons on a thin snowpack suits them less.

That is the best case we can make for why the code says what it says. Every climate variable we tested (January mean, snowfall, days below freezing, crossing days) captures part of the mechanism and misses part, because the deciding variable is not in the weather record.

It is how much heat the individual roof loses. A national threshold would have had to pick one of these variables and be wrong in a predictable direction. Deferring to "a history of ice forming along the eaves", decided locally, is a less tidy rule and a more honest one.