Wind Roses and Urban Airflow Diagrams in the Browser
Wind usually gets the thinnest treatment in a site analysis. You paste a wind rose from the weather service, write "prevailing northwest," and move on. What you actually want to know is where air moves through this site and where it stalls — and that has meant commissioning a CFD study.
sitedia's wind tab sits in that gap. You get a wind rose (observed statistics) and airflow between the actual buildings (a simulation) in one panel. This is an explanation of what those diagrams compute, and what they don't.
The wind rose is measured, not modeled
Nothing in the wind rose is invented. It counts the frequency of hourly wind direction and speed that a real station recorded. The source is EPW/TMYx — the standard building-simulation weather format, the same file Ladybug and Grasshopper read. It covers 17,311 stations worldwide, so Seoul and New York are handled identically.
Sixteen sectors is not an aesthetic choice. Korean meteorological records store direction on a 36-point scale, but only 16 distinct values ever appear — the observation itself is quantized to 16 sectors. Drawing 36 spokes would leave twenty of them permanently empty, which reads as a broken chart rather than a finer one.
You can split the rose by season. In Seoul the west and west-northwest arms grow noticeably in winter — that is the northwest monsoon.
This is regional wind, not site wind
The station sits kilometres away. A ridge, a river, or one tall tower in between changes what actually reaches your site. That is why the station name, its distance, and the period of record are always shown next to the rose. Drop a pin in the middle of the Sahara and you get a station 155 mi away with a 1950–2014 record — if that fact isn't on screen, the chart becomes a confident lie.
Speeds are recorded at each station's anemometer height, which differs between stations. Don't compare magnitudes across sites at face value.
The simulation — what is actually solved
Pressing Compute solves the airflow around the site for the prevailing direction of the selected period. The method is the lattice Boltzmann method with a Smagorinsky LES turbulence model. Instead of solving Navier–Stokes directly, it tracks particle distributions colliding and streaming — which maps extremely well onto a GPU.
The result is a time average, not an instantaneous snapshot. Turbulence fluctuates, so any single frame is irreproducible. Averaging the final portion of the run gives a stable picture.
A 13 ft grid — why that number
The grid is fixed at 13 ft. Urban alleys are typically 13–26 ft wide, so a 26 ft alley spans two cells. Coarser (26 ft) collapses the alley into a single cell and the passage disappears; finer (7 ft) costs eight times as much for a gain that is smaller than other errors in the pipeline.
To be blunt about the biggest of those errors: buildings with no recorded height are missing entirely. Height in meters is sparsely populated in the national building registry, so we fall back to floor counts — and a building with neither cannot be extruded at all. In a 1,000 ft radius of Myeongdong that is 172 of 821 buildings, about 21%. Refining the grid does not bring them back.
Domain — analysis radius plus 1,312 ft
A computational boundary is an artifact. At the inlet we impose "the wind here is uniform," and if buildings sit close to that plane their wake reflects back and contaminates the result. So we compute a domain 1,312 ft wider on every side than the drawing, then discard the margin and show only the analysis radius.
CFD best-practice guidelines (COST Action 732; the Architectural Institute of Japan) specify this margin as "at least 5H," where H is the tallest building. We tried that and abandoned it: H is hostage to a single outlier. Bukchon in Seoul is a low-rise hanok district averaging 23 ft, but one 636 ft tower inside the radius would inflate the domain to more than half a mile. A fixed 1,312 ft works out to 3–27H against the actual urban fabric height (95th percentile), satisfying the guideline in practice without being whipsawed by one building.
Simulation is supported to a 3,280 ft radius (about 0.6 mi). Beyond that, cost grows with the cube of radius and stops being practical — the drawings themselves still go to 1 mi.
Three views
- Wind rose — 16-sector observed frequency. Arm length is frequency, the color steps are speed bins, and the number at the center is the calm (near-zero wind) share.
- Simulation plan — color is speed, texture is direction. The texture comes from smearing white noise along the flow (line integral convolution), so direction reads without arrows. You can change height: an alley blocked at 13 ft above ground is often wide open at 157 ft.
- Simulation 3D — traced streamlines drawn as thin threads. Color varies along each thread, so you see a stream accelerate through a canyon and die behind a building.
If the plan tells you where it is fast or slow, the 3D tells you why. In the view above the flow entering from the west-northwest splits around the tall massing in the middle, and that split is the explanation for the red patch the plan shows downstream. Put the plan in the report; open the 3D when someone asks the follow-up question.
What this is not
It is not a validated wind study. Real pedestrian wind assessment uses wind-tunnel testing or validated CFD, runs many directions and seasons, and judges against statistical criteria. This is the stage before that — a tool for reading form during early design.
Specifically, what is missing:
- One direction only. A real assessment weights all 16 sectors by frequency.
- No terrain. Hills and valleys strongly shape wind at city scale.
- No vegetation. Street trees and park canopy filter wind substantially.
- Buildings with unknown height are absent (see above).
- No wind-tunnel validation. We have never measured a real site and matched it.
That is why every exported image carries "early-design reference — not wind-tunnel validated" inside the frame. Screenshot it into a report and the caveat travels with it.
How to read it
Don't quote numbers; read the spatial structure. "5.1 mph at this corner" is not a value this tool can stand behind. But "the westerly enters along this street and the block behind stalls" is a judgement that comes from geometry, and a 13 ft grid resolves that well.
It is most useful for comparing options. Place massing two different ways on the same site, run each, and it becomes visible which scheme blocks the corridor and which one leaves it open.