✕

How this works

The short version

Cape Town's famous summer wind — the south-easter, or Cape Doctor — slams into Table Mountain, Devil's Peak, Lion's Head and Signal Hill. The mountains squeeze it through gaps (the Venturi effect), bend it around the peaks so it hugs the slopes (the Coanda effect), and tumble it into rolling gusts on their lee sides (rotors — why Vredehoek is nicknamed "Windyhoek"). In winter the pattern flips: storm winds arrive from the north-west and the sheltered and exposed suburbs swap roles.

This site shows a physics model of that, from Cape Point to Durbanville to Stellenbosch. Colours = how hard the wind blows at street level. Moving streaks = the wind's path. Around the Table Mountain chain the model runs at a higher resolution (75 m, sharpening to 25 m as you zoom) so the valleys and ridges that funnel wind are resolved in more detail.

The numbers (for the curious)

• Model: a 3-D mass-consistent diagnostic wind model (the NUATMOS / WindNinja family). The observed inflow profile is blocked by real terrain, then minimally adjusted so mass is conserved exactly (divergence driven below 10⁻¹⁰ s⁻¹ on a ~670,000-cell grid).
• Stratification: the south-easter is a stable, inversion-capped layer (the "tablecloth"). A Froude-number ratio r² = Fr²/(1+Fr²) weights vertical vs horizontal adjustment — low Fr forces air around peaks rather than over them.
• Downslope windstorm & wind shadow: a pure mass-conserving model gets lee flow backwards — it dumps a calm deficit right where the real wind is strongest and keeps flank flow attached where the real air is sheltered. We correct both. Windstorm: at Froude ≈ 1 the stable south-easter stays attached as it pours over the Table Mountain / Devil's Peak saddle and accelerates down the lee slope — so the gap-fed upper City Bowl (Vredehoek, Oranjezicht) gets a mean boost up to ~1.9×. Wind shadow: a spot tucked deep behind a tall continuous massif (not on a gap-fed slope) sits in a wake the model can't produce on its own, so we damp its surface wind and turbulence — which is why Clifton, behind the Lion's Head / Twelve Apostles wall, is the calm summer refuge it really is. Both are empirical patches for the model family's known blind spot, not resolved dynamics.
• Wind input: three years of hourly reanalysis (Open-Meteo) at a point over False Bay, upwind of the peninsula. Each compass direction uses its own observed median / 90th-percentile speed and gust factor — so the north-wester is as strong as winter really makes it.
• Two nested grids: the metro region (Cape Point → Durbanville → Stellenbosch/Helderberg) at 200 m, and a 75 m detail grid with finer vertical levels over the Table Mountain chain. They are composited into one seamless field — the finer grids over the mountain chain, the coarse grid everywhere else, at matched opacity.
• Street-level wind: the displayed speed is 10 m above ground corrected through the local surface roughness from ESA WorldCover 10 m land cover — dense urban fabric, forest, vineyards and open water all change what you feel on the ground.
• Large buildings: tall buildings (≥25 m, from OpenStreetMap height tags) are far smaller than a grid cell, so they enter as local physics: dense tall fabric slows the mean street wind (urban canyon), while towers bring roof-height momentum down to street corners — gusts approach ~75% of the wind at roof height (downwash, the magenta cells in the Effect-zones layer). Exact corner accelerations need building-resolved CFD; treat these cells as "gustier than the colour suggests".
• Micro-resolution & calm pockets (25 m): the flow solver runs at 75 m, but the open elevation data carries real information down to ~30 m — so the street-level wind (and speed-up) layers sharpen from 200 m → 75 m → 25 m as you zoom. For every 25 m pixel, the upwind horizon angle to terrain within 1.2 km (the Winstral Sx shelter parameter, from Copernicus GLO-30 elevations) redistributes the solved wind within each coarse cell — revealing calm hollows, gully mouths and lee toes. Green pins mark the ten deepest reliably-calm pockets: low mean wind and low turbulence, so gusty lee-rotor zones never count as calm. Gust and turbulence layers stay at 75 m — rotor gusts penetrate sheltered hollows. Below ~30 m the open data runs out: garden-scale shelter (hedges, walls, single houses) would need LiDAR and building-resolved CFD.
• Turbulence & gusts: diagnosed from shear, wake deficit and a lee-rotor criterion (peaks at Fr ≈ 1), calibrated to the observed flat-terrain gust factor.
• Windiness score & colours: the ranking and the suburb marker colours use a 0–100 score (55% mean wind + 30% gusts + 15% turbulence) — so a rotor suburb with a modest average but violent gusts reads as windy, not calm. The wind-speed colour scale stretches to each scenario's own inflow, so a calm day and a gale both use the full palette rather than washing out.
• Honest limitations: mass-consistent models conserve mass, not momentum — the downslope jet, separation and rotor dynamics are empirical add-ons, not resolved, and individual streets are below even the 75 m grid. The mean field is not yet calibrated against weather-station records, so trust the relative ranking more than any single number.

Data & credits

Elevation: SRTM-derived terrarium tiles, AWS Open Data · Wind climatology: Open-Meteo archive · Land cover: ESA WorldCover 2021 · Buildings: © OpenStreetMap contributors · Basemap: Mapbox / OpenStreetMap. Built with a mass-consistent solver written in Python (NumPy/SciPy/PyAMG).