東京 22 区・290 万棟の建築物を 4 色の形態地層として描いた都市考古地図 A city archaeology map of Tokyo's 22 wards: 2.9 million buildings as a four-color form-strata.
🌐 Live: https://tokyo-strata.vercel.app 📄 A3 PDF: https://tokyo-strata.vercel.app/tokyo-strata.pdf ✍️ By Jada Q · 2026
Tokyo's Project PLATEAU publishes a 3D city model of all buildings in Tokyo — but does not publish year-of-construction data. Tokyo Strata works around this by treating fire-resistance class × height as a proxy for era:
| Color | Rule | Era proxy |
|---|---|---|
| 🟫 朽葉 (decay-leaf brown) | fp ∈ {その他, 不明} ∧ h<10m |
Pre-war wooden 下町 |
| 🟨 利休茶 (rikyū tea) | fp ∈ {耐火, 準耐火} ∧ h<10m |
Post-war small structures |
| ⬜ 銀鼠 (silver-grey) | 10 ≤ h < 30m |
High-growth-era mid-rise |
| ⬜ 月白 (moon-white) | h ≥ 30m |
Post-bubble towers |
The result is a single map showing the time-layered urban archaeology of Tokyo — the 皇居 as a central void, 下町 patches in the periphery, modern towers clustered in the central wards.
Crossing Tokyo Strata with two other personal projects:
- UpgradeMap — Tokyo 23-ward real-estate upgrade signals (price + population YoY)
- ATLAS — settlement extinction prediction via cohort-component projection
Headline finding: 月白 ratio × population YoY = r = +0.86 across the 22 wards. Tower density predicts where people are moving in.
The cross-analysis page identifies three trajectories:
- 若返り都心 (Rejuvenating Centers): 千代田 / 中央 / 文京 / 港 / 江戸川
- 安定郊外 (Stable Suburbs): 13 wards in the middle
- 緩慢成熟 (Slowing Mature): 足立 / 葛飾 / 北 / 墨田 / 台東
And one outlier worth reading: 葛飾 vs 江戸川 — same architectural age (朽葉 ≈ 60%) but different mechanisms of renewal. 葛飾's capital prices run ahead of population; 江戸川's population runs ahead of capital.
→ Full write-up: https://tokyo-strata.vercel.app/cross
| Route | Purpose |
|---|---|
/ |
Editorial scrolling magazine — cover · intro · 7 narrated wards · colophon |
/cross |
Cross-analysis (Strata × UpgradeMap × ATLAS) — scatter charts, classifications, red-team |
/about |
Project explanation, palette rationale, 22-ward stat table |
/explore |
Free zoom/pan over all 22 wards |
/tokyo-strata.pdf |
Print-ready A3 landscape, 10 pages, 6.9 MB |
- Map: MapLibre GL JS 4.7 + PMTiles 4.3
- Tiles: tippecanoe → 48 MB single-file PMTiles for 2.9M building footprints
- Data pipeline: Python
xml.etree(CityGML parser) + Nominatim API (landmark labels) - Hosting: Vercel static (HTTP byte-range serves PMTiles)
- PDF: Headless Chrome via puppeteer-core + system Chrome
git clone https://github.com/Jada-Q/tokyo-strata.git
cd tokyo-strata
# Run viewer (PMTiles is committed — works out of the box, but
# python3's http.server doesn't support byte-range. Use serve:)
npx serve -l 9877 .
# open http://localhost:9877/# 1) Batch-download + parse all 22 wards (3-5 GB temp space cleared after each)
brew install tippecanoe pmtiles
while read ward url; do
./scripts/add-ward.sh "$ward" "$url" cleanup
done < scripts/all-wards.txt
# 2) Stats + tiles
python3 scripts/compute_stats.py
./scripts/build-tiles.sh # → data/tokyo.pmtiles (48 MB)
# 3) (optional) Cross-analysis with UpgradeMap data
python3 scripts/cross_analyze.py
# 4) (optional) Cohort projection — ATLAS module
# See https://github.com/Jada-Q/atlasindex.html editorial scrolling magazine (PMTiles)
cross.html cross-analysis with UpgradeMap × ATLAS
explore.html free interactive map (PMTiles)
about.html project about (BRUTUS-style colophon)
about.md shareable summary text (日 + 中)
findings.md PoC log — what was tried, what failed, what pivoted
cross-report.md cross-analysis findings in markdown
data/
tokyo.pmtiles 48 MB vector tile (committed)
stats.json per-ward count/bbox/strata%/avg-max height
cross.json UpgradeMap × Strata joined values
cohort.json 50-year cohort projection per ward
landmarks.json 50 markers (palace/parks/rivers/etc) via Nominatim
scripts/
add-ward.sh one-shot ward pipeline
all-wards.txt 22 ward URLs
parse_to_geojson.py CityGML → GeoJSON
compute_stats.py → data/stats.json
build-tiles.sh merge → tippecanoe → pmtiles
cross_analyze.py Strata × UpgradeMap join + Pearson r
fetch-landmarks.py Nominatim batch geocode
generate-pdf.js puppeteer-core → A3 PDF
social/ ready-to-post copy (Twitter / note / 小红书)
| Metric | Value |
|---|---|
| 22 ward 形態 coverage | 290 万棟 (100% fp + height) |
yearOfConstruction coverage |
0% (PLATEAU does not publish) |
| Strongest cross-correlation | 月白% × pop YoY = r = +0.86 |
| Highest 古層率 | 練馬 73.9% |
| Highest 月白率 | 千代田 18.5% |
| Tallest building in dataset | 墨田 634 m (東京スカイツリー) |
| 50-year pop projection (highest) | 千代田 +24.5% |
| 50-year pop projection (lowest) | 台東 +2.0% |
- Every Tokyo ward grows over 50 years when you run the same cohort-component algorithm that predicts rural extinction in Akita. Tokyo doesn't shrink — it accelerates.
- 江戸川 vs 葛飾: same form age (朽葉 ~60%), but 葛飾 sees capital first, 江戸川 sees population first. Two distinct mechanisms of 下町 renewal.
- Form ↔ migration: tower ratio is the single best predictor of where people are actually moving to. Architecture isn't passive — it's a leading indicator.
n=22, so r=+0.86 has wide confidence intervals- Cross-section data only; "form predicts the future" needs longitudinal verification
- Cohort projection assumes 2015–2020 trends continue linearly for 50 years
- Selection bias in PLATEAU coverage (台東 ward absent from 2023 dataset)
Full red team: https://tokyo-strata.vercel.app/cross#redteam
- Code: MIT — see LICENSE
- Editorial content (Japanese narrative, screenshots): © Jada Q 2026, attribution preserved
- Underlying data: PLATEAU is CC BY 4.0 (国土交通省) — preserve attribution per CC BY terms
- Cohort method: public-domain demographic technique (cohort-component projection)
- Project PLATEAU (国土交通省 + G空間情報センター) — open 3D city data
- Reinfolib (国交省 不動産取引価格) — transaction prices
- e-Stat (総務省 国勢調査) — population by age cohorts
- MapLibre GL + PMTiles + tippecanoe + Hiragino Mincho ProN
- Cohort-component projection method adapted from Project ATLAS
- Edited & developed by Jada Q · 2026