Why San Francisco gets measured building heights, not guesses

A building's height feeds almost every other layer in a risk score: storey count, soft-story likelihood, pounding risk against neighbours, sky-view factor. Get height wrong and everything built on top of it inherits the error. Most of our pilot deployment has to estimate it. San Francisco doesn't have to.

In Istanbul, Tessera's original pilot city, roughly 54% of buildings have no real height value in any public dataset we've found, after an extended search across satellite-derived global building datasets, national mapping-agency products, and municipal GIS archives, most of them either incomplete, commercial-only, or restricted to government institutions. For those buildings, height is estimated from a floor-count default, a real, explicitly-labelled fallback rather than a hidden guess, but a fallback all the same. It's the single largest known data gap in that city's deployment, and we say so plainly in how the score actually works.

What San Francisco has instead

San Francisco publishes a real, building-level LIDAR-derived height for essentially its entire building stock, through its own open-data portal: a 2010 aerial LIDAR and photogrammetric survey, covering all 177,023 buildings this pilot maps, with real measured height, not an inferred one. That single dataset also solves a problem this project never managed to solve for Istanbul despite direct outreach to the agencies that hold comparable LIDAR survey data there: a complete, public, per-building height layer.

Why this compounds, not just adds

Height isn't a single output field; it's an input to several others. Tessera's pounding-risk proxy flags buildings within half a metre of a neighbour whose height differs by 3 metres or more, the real differential-collapse mechanism where floor slabs at different levels strike each other during shaking. That proxy is only as good as the height values on both sides of the comparison. Sky-view factor, used for urban heat exposure, depends on real building heights around every point on the ground, not estimated ones. Storey-band collapse-risk bonuses depend on height divided by an assumed floor height. A wrong height doesn't just mis-state one field; it propagates error into every proxy built on top of it.

This is also why we don't quietly upgrade a guess to look more precise than it is. Istanbul's height field carries an explicit height_src tag on every building, so a floor-count default is always visibly labelled as one, never presented with the same confidence as a real measurement. San Francisco's buildings carry the same field with a different real answer: a measured LIDAR height, sourced and dated, not a default.

The broader point isn't really about San Francisco specifically. It's that a risk score is only as honest as its weakest real input, and the input worth checking first, in any city, is the one everything else is built on top of. See what that looks like building by building on the live San Francisco map, including which of its own layers, like the Mandatory Soft Story Program and the Hayward Fault hazard scenario, are real ground truth rather than a proxy.