Lilambd/world-hazard

World Hazard Ledger: how much each place shakes and how much it warmed

🌐 日本語 · 中文 · Español · Português · Français · Deutsch · 한국어 · हिन्दी · العربية

• 日本語 — World Hazard Ledger：世界の地震・温暖化・災害データ: 世界のすべての州・省がどれだけ揺れ、どれだけ温暖化したか。1900年以降のM6+・M7+地震、観測所の昇温、地震・津波・噴火による死者数を収録。
• 中文 — World Hazard Ledger（全球灾害与升温数据）: 全球每个州/省的地震频度与升温幅度：1900 年以来的 M6+ 与 M7+ 地震、气象站升温，以及地震、海啸和火山喷发造成的死亡人数。
• Español — World Hazard Ledger (registro mundial de amenazas naturales): Cuánto tiembla y cuánto se ha calentado cada estado y provincia del mundo: sismos M6+ y M7+ desde 1900, calentamiento en estaciones y muertes por sismos, tsunamis y erupciones.
• Português — World Hazard Ledger: terremotos, aquecimento e desastres por região: Quanto treme e quanto aqueceu cada estado e província do mundo: terremotos M6+ e M7+ desde 1900, aquecimento medido em estações e mortes por terremotos, tsunamis e erupções.
• Français — World Hazard Ledger : séismes, réchauffement et catastrophes par région: À quel point chaque État et province du monde tremble et s'est réchauffé : séismes M6+ et M7+ depuis 1900, réchauffement mesuré aux stations, et décès dus aux séismes, tsunamis et éruptions.
• Deutsch — World Hazard Ledger: Erdbeben, Erwärmung und Katastrophen nach Region: Wie stark jede Provinz der Welt bebt und wie stark sie sich erwärmt hat: M6+- und M7+-Erdbeben seit 1900, Erwärmung an Messstationen und Todesopfer durch Beben, Tsunamis und Ausbrüche.
• 한국어 — World Hazard Ledger: 지역별 지진·온난화·재해 데이터: 세계 모든 주·도의 지진 빈도와 온난화 정도: 1900년 이후 M6+·M7+ 지진, 관측소 기온 상승, 지진·쓰나미·분화 사망자.
• हिन्दी — World Hazard Ledger (प्राकृतिक आपदा खाता): दुनिया का हर राज्य और प्रांत कितना हिलता है और कितना गर्म हुआ है: 1900 से M6+ और M7+ भूकंप, स्टेशन वार्मिंग, और भूकंप, सुनामी व ज्वालामुखी विस्फोट से हुई मौतें।
• العربية — سجل المخاطر العالمية (World Hazard Ledger): مقدار اهتزاز كل ولاية ومقاطعة في العالم ومقدار احترارها: زلازل M6+ وM7+ منذ 1900، واحترار المحطات، والوفيات من الزلازل وأمواج تسونامي والثورات البركانية.

Every state, province and prefecture of the world (4,596 admin-1 regions), with:

• how many M6+ and M7+ earthquakes hit it from 1900 to 2026, and how many per 1,000 km²
• how much its weather stations warmed (1991-2020 normal minus 1961-1990 normal)
• deaths from significant earthquakes, tsunamis and volcanic eruptions, by country
• the 41 regions that are in the top quarter for both shaking and warming, with population

Nothing is generated by a model. Every number is counted or averaged from public measurement records, and every source file is listed with URL, fetch time and SHA-256 in sources.csv.

Data as of 2026-11-01 (last fetch 2026-11-01T03:07Z; quake catalogue up to 2026-11-01T01:24Z).

Quick start

from datasets import load_dataset
quake = load_dataset("Lilambd/world-hazard", "admin1_quake", split="train")
warm  = load_dataset("Lilambd/world-hazard", "admin1_warming", split="train")

Tables

config	rows	what
admin1_quake	852	admin-1 regions with any M6+: counts M6 / M6 inside land / M7, M6 per 1,000 km², population
admin1_warming	1,240	admin-1 regions with stations: number of stations, mean and median warming (°C)
joint_exposure	41	regions in the top quarter for both M6 density (≥ 0.0037 per 1,000 km²) and warming (≥ +1.07 °C)
features_country	954	per-country features in long form (iso3, feature, value, year, unit, source): M6 density, M7 count, population share in seismic regions, quake deaths per million, warming
features_region	1,011	the same features for EU NUTS and US regions
country_warming / country_quake / country_event	157 / 284 / 307	country rollups; quake rank by two counting methods; disaster counts and deaths since 1900
station_warming	5,924	each GHCN-M v4 station with both normals: lat, lon, 1961-1990 and 1991-2020 mean, delta
quakes	313,014	USGS ComCat M4.5+ 1900-2026, each tied to an admin-1 region (a1_how = in / near (within 1.5° of land) / sea)
events	10,738	NOAA NCEI significant earthquakes, tsunamis and volcanic eruptions, with deaths and damage, tied to admin-1
admin1	4,596	the region list: ISO 3166-2, HASC, GeoNames id, Wikidata id, English, local and Japanese names, area km², population with year and source
band_warming / year_anomaly / decade_quake	small	warming by 10° latitude band, yearly anomaly arctic vs rest, M6/M7/M8 per decade
sources	189	every fetched file: URL, fetched_at, sha256, licence, bytes

Shared keys

Rows join to other Lilambd ledgers by the same keys: a1 (admin-1 id), iso_3166_2, wikidata, iso2 / iso3, and cell_05deg on every row with a point (0.5° grid, row*720+col, row = floor((90-lat)/0.5), col = floor((lon+180)/0.5)).

What the numbers say (computed, not curated)

• 99.5% of the 5,924 stations got warmer between the two normals; mean +0.82 °C.
• The arctic band (≥ 66.5°N) warmed most, +1.45 °C; 37 of the 50 fastest-warming stations are in it, though it holds 2.2% of stations.
• M7+ by country (admin-1 within 1.5°): Indonesia 170, Japan 123, Papua New Guinea 95, Russia 87, United States 83.
• 41 regions (218.7 million people) are in the top quarter for both shaking and warming; most are in Iran, Mongolia, Russia, China and Turkey.

Gaps and known limits

• A region with no row in admin1_quake had no M6+ within 1.5° since 1900. A region with no row in admin1_warming has no station record, which is not the same as no warming.
• Stations need ≥ 20 years per month in both normal periods; many places in Africa, South America and the Pacific have none.
• 219 of the 1,620 M7+ quakes are far offshore and tied to no region.
• Some Natural Earth local names are wrong (for example Arkhangel'sk is shown with the name of Vologda). English names and ids are reliable; check name_local against GeoNames before display.
• Populations come from several sources with different years (pop_source, pop_year); 335 regions have none.

Use it in production

For AI agents: 14 more Lilambd open-data APIs (airports, regions, markets, security, travel and more), pay per call with x402 (from $0.005 USDC on Base, no signup), free sample with sample=1, also as an MCP server: https://api.lilambd.com/llms.txt

Need a per-place answer ("how exposed is this town to quakes and heat?"), a field that is not here, or a daily feed? Write to hello@lilambd.com.

Related: Lilambd/world-signals (what the world is paying attention to, daily).

Quick start: datasets, pandas, DuckDB

import pandas as pd  # hf:// paths need `pip install huggingface_hub`
BASE = "hf://datasets/Lilambd/world-hazard/data/"
q = pd.read_csv(BASE + "admin1_quake.csv")
print(q.nlargest(10, "m7")[["country", "name", "m6", "m7", "m6_per_1000km2"]])
print(pd.read_csv(BASE + "decade_quake.csv"))

-- DuckDB 0.10+ reads hf:// paths directly
-- regions that both shake and warmed
SELECT q.country, q.name, q.m6_per_1000km2, w."mean" AS warming_c, w.n AS stations
FROM 'hf://datasets/Lilambd/world-hazard/data/admin1_quake.csv' q
JOIN 'hf://datasets/Lilambd/world-hazard/data/admin1_warming.csv' w USING (a1)
ORDER BY q.m6_per_1000km2 DESC LIMIT 15;

A runnable notebook is in notebooks/quickstart.ipynb.

Use cases

• Earthquake frequency and density per state or province since 1900.
• Station-based warming per region (1991-2020 vs 1961-1990 normals).
• Multi-hazard screening for insurance, relocation and supply-chain risk.
• Joins with world-earth and world-ground on a1.

Citation

@misc{lilambd2026_world_hazard,
  author       = {Lilambd},
  title        = {Lilambd World Hazard Ledger},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/Lilambd/world-hazard}},
  url          = {https://huggingface.co/datasets/Lilambd/world-hazard}
}

Please also credit the original sources listed in this card.

Sources and licence

• Earthquakes: USGS ComCat (US Government work, public domain).
• Significant earthquakes, tsunamis, volcanoes: NOAA NCEI / WDS Global Historical Hazard Database (public domain).
• Temperatures and stations: NOAA NCEI GHCN-Monthly v4 and GHCN-Daily station list (public domain).
• Region shapes and names: Natural Earth 10m admin-1 (public domain).
• Populations: OECD regional statistics (CC BY 4.0), World Bank subnational (CC BY 4.0), US Census (public domain), GeoNames (CC BY 4.0).

The derived tables are released under CC BY 4.0. Please cite "Lilambd World Hazard Ledger" and the original agencies.

Maintained by Lilambd. Questions: hello@lilambd.com

Changelog

• 2026-11-03 quality fixes.
  • quakes: removed 1 row below the stated M4.5 floor (nc21364840, M3.38 ml, 2004-05-17, near Cayucos, California). Rows 313,015 → 313,014. It was the only row under M4.5. decade_quake m45 for the 2000s went from 60,558 to 60,557 because that count had included it.
  • Svalbard (NOR-901): population was 0 (OECD, 2025). About 2,500 people live there, so this is not a real zero. It is now empty (unknown), and pop_year/pop_source are cleared too, in admin1, admin1_quake and joint_exposure. Regions with no population: 332 → 333 (335 after the next fix).
  • Placeholder codes in admin1 were cleared to empty: 188 iso_3166_2 (Natural Earth codes like XK-X02~ or -99-X16~), 102 hasc (…~ / -99.…), 12 iso2 = -1, and 187 gn_id ≤ 0 (-99 ×88, 0 ×36, -1 ×3, other negative ids ×60). GeoNames ids are always positive. Where the right code was clear it was filled in: Gbapolu LR-GP, River Gee LR-RG, Tadzhikistan Territories TJ-RA, Cocos (Keeling) Islands iso2 CC, Christmas Island iso2 CX. iso2 = NA is Namibia and was kept.
  • Wrong local names: 7 regions had another region's name_local, copied from a neighbouring or parent region in Natural Earth (for example Bangkok had Chiang Mai's Thai name). They are now corrected: Bangkok → กรุงเทพมหานคร, Arkhangel'sk → Архангельская область, Moskva (city) → Москва, Tizi Ouzou → ولاية تيزي وزو, Chaiyaphum → จังหวัดชัยภูมิ, Bueng Kan → จังหวัดบึงกาฬ, Rasŏn → 라선시. This changed 7 rows in admin1, 1 in admin1_quake and 3 in admin1_warming. No two regions share a name_local any more.
  • Two city units had the population of the NUTS-3 region around them (OECD 2025). Grad Sofiya (BGR-2243) had 224,701, which belongs to Sofia Province (NUTS BG412). Veszprém urban county (HUN-4909) had 333,345, which belongs to Veszprém County (NUTS HU213). Each value now sits on its province or county (BGR-2244, HUN-3158; 2025, regionstat:oecd_pop), replacing their older 2016 figures, and the two city units are now empty. This matches the same fix in Lilambd/world-earth.

Paid editions & related

• Earth Pulse: real earthquakes as sound (itch): https://lilambdcrossing.itch.io/earth-pulse-real-earthquakes-as-sound
• Same data as an API (free sample=1, pay-per-call via x402): https://api.lilambd.com/llms.txt
• All Lilambd shops: https://payhip.com/Lilambd and https://lilambdcrossing.itch.io

If this dataset helped you, a heart (like) on this page helps others find it.