台灣街道願景計畫：還路於民｜全國交通安全地圖

01 — Decade Overview 全國十年總覽01 — Decade Overview

A1 · 24 小時內死亡A1 · death within 24 h

17,175^人deaths

十年內全台道路 A1 死亡人數A1 road deaths nationwide, last 10 years

Lives lost on Taiwan's roads · 2016 — 2025

364 個鄉鎮市區 · 22 個縣市 · 平均每年 1,718 人364 districts · 22 counties · ~1,718 deaths per year

本站採 A1（當場或 24 小時內死亡）分類；官方 30 日死亡分類（國際標準）十年約 29,000 人。兩者差異說明 ↓This Atlas uses the A1 (death on scene or within 24 h) classification; the official 30-day death classification (international standard) is ~29,000 over the decade. Why they differ ↓

Events

A1 死亡事故件數A1 fatal crash events

16,713

Most Victims

機車騎士為 A1 事故死亡人數最多的用路人Motorcyclists — most A1 deaths

—%

Pedestrian Victim

行人受害Pedestrian victims

—%

Most Deaths · County

十年累計第一Highest 10-yr total

台南市Tainan City 1,785

A1 死亡，只是冰山一角 A1 deaths are the tip of the iceberg

A1

死亡Deaths

17,175

A2

事件Events

3,543,840

A2

受傷Injured

4,738,069

每 1 人死亡，背後對應 ~206 件 A2 事件、~276 人受傷。A1 為事後或當場 24 小時內死亡；A2 為事後死亡或受傷，傷亡規模約為 A1 的 200 倍以上。 For every road death, ~206 A2 events and ~276 injuries lie underneath. A1 = death on the scene or within 24 h; A2 = injury or post-24h death. A2 casualties exceed A1 by more than 200×.

為什麼是 17,175，而不是官方常聽到的「約 3 萬」？ Why 17,175 — and not the official "~29,000"?

兩個數字都對，差在資料分類。本站採 A1（當場或 24 小時內死亡），因為它是唯一逐案、含經緯度與運具的開放資料，才能畫成這裡的每一張地圖。官方頭條用 30 日內死亡（國際通用標準），十年合計約 29,000 人：兩者並不矛盾。 Both numbers are correct — they use different data classifications. This Atlas uses A1 (death on scene or within 24 h), the only case-level, geocoded, per-vehicle open dataset, and therefore the basis of every map here. The official headline uses death within 30 days (the international standard): ~29,000 over the decade. The two do not conflict.

A1 · 24h

17,175

本站分類 · 當場或 24 小時內死亡This Atlas · death within 24 h

≈ 0.59 ×

30-day

29,018

官方頭條 · 30 日內死亡（國際標準）Official headline · death within 30 days

→

A2

~11,843

事後（24h–30 日）死亡 · 事故當下記為 A2Died 24 h–30 days later · recorded as A2

逐年對照：本站 A1（紅）約為官方 30 日死亡（灰）的六成，十年比值穩定落在 0.54–0.65，內部一致：這反而證明資料正確，只是分類較嚴格。30 日死亡資料來源：交通部道安資訊平臺／roadsafety.tw「30 日死亡人數」；A30 歷年彙整 tadd.org.tw。 Year by year, this Atlas's A1 (red) runs at ~60% of the official 30-day count (grey); the ratio holds a stable 0.54–0.65 across the decade — internally consistent, confirming the data is right, just stricter. 30-day source: MOTC Road Safety Information Platform / roadsafety.tw "30-day deaths"; A30 series, tadd.org.tw.

02 — Casualty Trend 傷亡趨勢02 — Casualty Trend

—

十年間，A1 死亡略降，A2 傷亡件數卻持續上升 A1 deaths edged down, but A2 casualties kept climbing

我們關注 A1 車禍死亡的改善，但同時仍必須關注「受傷與長期失能」的嚴重車禍。當 A2 件數成長，意味著危險路口、過快車速、混合車流的暴露面持續擴大：治理重點應從「降低死亡」進一步推進到「降低嚴重傷亡」。 We welcome the fall in A1 road deaths, but must stay just as focused on the serious crashes behind injury and long-term disability. As A2 events climb, dangerous junctions, excess speed and mixed traffic keep expanding exposure: the goal must move from "reducing deaths" to "reducing serious casualties".

A1 死亡（左軸）A1 Deaths (left axis) A2 事件（右軸）A2 Events (right axis) A2 受傷（右軸）A2 Injured (right axis)

03 — 22 Counties 縣市深度地圖03 — 22 Counties

Explore by County

點選任一縣市，進入該地的完整互動地圖（事故點位、熱力圖、優先治理路段、幹線疊圖、逐年趨勢、行政區面量）Tap any county to open its full interactive map — crash points, heatmap, priority road segments, highway overlay, yearly trend, district choropleth

04 — District-Level Choropleth 行政區面量圖04 — District-Level Choropleth

Quintile · 0 → max

District Detail — 行政區明細

點選任一行政區（區、鄉、鎮、市）以查看分布 Click a district on the map to inspect Click a district on the map

05 — County Choropleth 縣市熱力圖05 — County Choropleth

Quintile

Ranking · 全 22 縣市 · all 22 counties

點選縣市進入完整地圖 · 顏色＝當前指標、長條＝相對比例Tap a county for its full map · colour = current metric, bar = relative scale

Rate · deaths per 100,000 residents (2013 population, MOI dataset 8410)
標 N<30 之縣市樣本數過小，每十萬人率波動大、僅供參考Counties tagged N<30 have too few deaths for a stable per-100k rate — interpret with caution

06 — Crash Points & Deadliest Segments 全國事故點位與最危險路段06 — Crash Points & Deadliest Segments

每一個點，都是一條在 24 小時內逝去的人命 Every dot is a life lost within 24 hours

全台 13,708 件可定位的 A1 死亡事故（2016–2025），依受害者運具著色。其餘約 3,000 件因原始資料缺座標（多為 2016–2017、2021 簡式年度）未能標點。點選任一點可查看該事故細節（時間、地點、傷亡、街景），並連往該縣市的完整地圖。 13,708 geolocatable A1 fatal crashes nationwide (2016–2025), coloured by victim's mode of travel. ~3,000 more lack coordinates in the source data (mostly the 2016–2017 & 2021 simplified years). Click any dot for that crash's details (date, place, casualties, Street View), with a link to the county's full map.

Victim 機車Motorcycle 汽/貨車Car / Truck 行人Pedestrian 自行車/慢車Bicycle 其他Other 點大小依死亡人數 · 滑過看地點 · 點擊看事故細節Dot size ∝ deaths · hover for location · click a dot for crash details

十年來，奪命最多的十個路段 The ten road segments that have killed the most

依受害者運具切換。有道路名稱＋里程者聚合為「~3 公里路段」（標記 ⚑），其餘以 ~300 公尺方格聚合，依十年累計死亡排序。已濾除座標落在行政中心的「定位失敗群集」（同一座標混入多個行政區與路名者）。點任一列可在上方地圖定位，點「看街景」直接看現場道路。 Switch by victim's mode. Roads with a name + km are clustered into ~3 km segments (⚑); others into ~300 m cells, ranked by 10-year cumulative deaths. Geocoder "failure clusters" (one fallback coordinate mixing many districts & roads) are filtered out. Click any row to locate it on the map above; tap "Street View" to see the actual road.

⚑ = 可解析里程之線型路段；其餘為點位群集。死亡數為該運具十年累計。A1（24 小時內死亡）。 ⚑ = a linear segment with km markers; others are point clusters. Deaths = 10-year cumulative for that mode. A1 (death within 24 h).

07 — Age × Accident Type 年齡層與事故態樣07 — Age × Accident Type

—

BY AGE

BY AGE × MODE

BY AGE × ACCIDENT TYPE

TOP COMBINATIONS

最常見的（年齡 × 運具 × 態樣）組合：排行越前面，越值得針對性對策 Most-frequent age × mode × type combinations — top-ranked combos warrant targeted intervention

資料覆蓋說明：本切面僅涵蓋有完整當事人欄位之年份（2018–2020、2022–2025），共 — 件（占 A1 全資料 —）；2016、2017、2021 為簡式欄位無年齡與事故態樣。
「年齡」採事件中最脆弱當事人之年齡作為死亡年齡之代理（與 victim-mode 邏輯一致），可能與實際死者略有偏差。 Coverage note: this slice covers only full-schema years (2018–2020, 2022–2025): — events (— of all A1). 2016, 2017, 2021 use a simplified schema without age or accident-type fields.
"Age" uses the most-vulnerable party's age as a proxy for the deceased (consistent with victim-mode logic), which may differ from the actual victim in multi-party events.

08 — Gender Lens 性別視角08 — Gender Lens

全國 A1 死者性別 · 整體與運具別（十年，取可辨識性別之死者）National A1 victim gender · overall & by mode (decade · identifiable victims)

09 — Sidewalk Check-up 人行道健檢09 — Sidewalk Check-up

2023 → 2026 · 22 counties · 92,326 segments · +772 km

人行道變長了，但沒有變好走More sidewalk, not more walkable

全國人行道健檢：三年長度 +7.6%，但平均淨寬下降、路緣斜坡缺口率不減反增。行政區面量圖、逐路段實景疊圖（含電線桿障礙物）、一鍵街景比對現場。A national sidewalk check-up: +7.6% length in three years, yet average clear width fell and the curb-ramp gap widened. District choropleth, per-segment overlays (incl. utility-pole obstacles) and one-click Street View comparison.

開始健檢 →Check-up →

10 — Structure 結構的代價10 — The Price of Structure

r = 0.66 · 22 counties · 17 years · 3 scenarios

公共運輸愈不足，死傷率愈高Less transit, more casualties

街道願景白皮書互動數據：22 縣市市佔×死傷散佈、17 年死亡率賽跑、2050 三情境推估：結構決定死傷，選擇決定未來。Interactive whitepaper data: the share-casualty scatter, a 17-year county race, and three scenarios to 2050 — structure decides casualties; choices decide the future.

看數據 →Explore →

11 — International 國際比較11 — International

Japan 2.21 · Korea 4.88 · Taiwan 11.84 / 100k

同樣的東亞,5.4 倍的死亡率Same East Asia, 5.4× the death rate

日本 47 都道府縣 × 韓國 17 廣域市道 × 臺灣 22 縣市,86 個行政區的每十萬人死亡率比較。東京 0.97 vs 屏東 23.92 — 差異不在地理,而在基礎設施投資。86 first-level divisions across Japan, Korea and Taiwan. Tokyo 0.97 vs Pingtung 23.92 — the difference is infrastructure, not geography.

看比較 →Compare →

12 — Climate Evidence 臺灣升溫的真相12 — Climate Evidence

Taipei · Taichung · Kaohsiung · 1990–2025 · p < 0.05

臺灣升溫的真相 — 這不是體感Taiwan is warming — and it's not a feeling

三大都市 35 年夏季升溫,全數通過統計顯著性檢定(Mann-Kendall · p < 0.05)。從升溫證據、都市熱島機制到機動車輛的局部加劇 — 氣候與交通,從來是同一件事。35 years of summer warming across three metros, all statistically significant (Mann-Kendall, p < 0.05). From the evidence to the urban heat island and the role of motor vehicles — climate and transport were always one story.

看證據 →See the evidence →

13 — Dual-Zero Vision 雙齊零願景13 — Dual-Zero Vision

Carbon 0 × Vision Zero 0

交通零死亡的雙齊零願景The Dual-Zero Vision

氣候危機與道路死亡共享同一個根源——也共享同一個解方。「交通習慣移轉的複利效應」互動模型 × 「五大訴求」行動方案,已整合至專頁。The climate crisis and road deaths share one root cause — and one solution. The Compound Effect interactive model × the Five Demands, on a dedicated page.

前往專頁 →Open the page →

14 — Notes 資料說明14 — Notes

開放資料來源 · 全站Open Data Sources · site-wide

本資料庫全部建立在公開資料之上，任何人都能查證與重製。核心是內政部警政署／交通部道安資訊平臺的 A1、A2 道路交通事故開放資料（經政府資料開放平臺 data.gov.tw 釋出，涵蓋 2016–2025 年）。A1 為逐案資料，含事故時間、地點經緯度、車種與肇因，構成全站地圖與所有分析的基礎；死亡口徑的對照，則參用道安資訊平臺、roadsafety.tw 與 tadd.org.tw 的「30 日死亡」（A30）統計。

地理圖資方面，鄉鎮市區與縣市界線採 g0v 社群整理釋出的行政區界圖；每十萬人死亡率以內政部戶政司人口統計（dataset 8410）換算。

人行道專題使用交通部「全國人行道整體區位資料」兩期普查（2023 年 6 月與 2026 年 6 月，全 22 縣市、逾 9 萬段路段），呈現長度、平均淨寬與路緣斜坡缺口；人行道障礙物疊圖則來自台灣電力公司電線桿桿籍資料（全國約 273 萬支，TWD97 座標，經投影轉換為經緯度）。

氣候章節取用中央氣象署臺北、臺中、高雄三測站 1990–2025 年夏季氣溫觀測；國際比較則彙整日本、韓國、臺灣官方公布之道路交通死亡統計，換算 86 個第一級行政區的每十萬人死亡率。地圖底圖使用 OpenStreetMap 與 CARTO 開放圖磚；街景比對串接 Google 街景服務（屬商用服務，非開放資料）。

所有來源與處理方法均標註於各頁「資料說明」，並以嚴格定義（A1＝當場或 24 小時內死亡）如實呈現，不誇大、可回溯。

Everything here is built on public data anyone can verify and reproduce. The core is the A1 & A2 road-crash open data from the National Police Agency / MOTC Road Safety Information Platform (released via data.gov.tw, 2016–2025). A1 is case-level — time, geocoded location, vehicle type and cause — and underpins every map and analysis here; the death-calibre comparison draws on the 30-day (A30) figures from the road-safety platform, roadsafety.tw and tadd.org.tw.

District and county boundaries use the g0v community's administrative-boundary GeoJSON; per-100k rates use MOI Household-Registration population (dataset 8410).

The sidewalk chapter uses the MOTC "National Sidewalk Inventory" — two censuses (June 2023 & June 2026, all 22 counties, 90,000+ segments) for length, clear width and curb-ramp gaps; the obstacle overlay comes from the Taipower utility-pole registry (~2.73M poles, TWD97, reprojected to WGS84).

The climate chapter uses Central Weather Administration summer-temperature records for Taipei, Taichung and Kaohsiung (1990–2025); the international comparison compiles official road-death statistics from Japan, Korea and Taiwan into per-100k rates for 86 first-level divisions. Base maps are OpenStreetMap + CARTO open tiles; Street View comparison uses Google's Street View service (a commercial service, not open data).

Every source and method is documented in each page's "Data Notes", reported under a strict definition (A1 = death on scene or within 24 h) — no inflation, fully traceable.

關於本面量圖About this choropleth

本面量圖整理自內政部警政署 A1 級道路交通事故公開資料，涵蓋年度民國 105 至 114 年（2016–2025），共 16,713 件事件、17,175 人死亡（A1，當場或 24 小時內死亡），分布於 364 個鄉鎮市區（22 個縣市）。口徑說明：本站採 A1（24 小時內死亡），因其為唯一逐案、含座標與運具的開放資料；官方頭條「道路交通事故死亡人數」採 30 日死亡口徑（國際標準），同期十年約 29,000 人（A30，逐年 2,697–3,064），兩者差額即車禍後 24 小時至 30 日間死亡者（事故當下歸類為 A2）。資料來源：交通部道安資訊平臺／roadsafety.tw、tadd.org.tw。

Built from the National Police Agency A1 (fatal) road-crash open dataset, covering 2016–2025 — 16,713 events, 17,175 deaths (A1 — death on scene or within 24 h), across 364 districts (22 counties). On calibre: this Atlas uses A1 (24-hour) because it is the only case-level, geocoded, per-vehicle open dataset; the official headline figure uses the 30-day death calibre (the international standard), totalling ~29,000 over the same decade (A30, 2,697–3,064/yr). The gap is people who died 24 h–30 days after the crash (recorded as A2). Source: MOTC Road Safety Information Platform / roadsafety.tw, tadd.org.tw.

受害者運具邏輯：每筆事件以其中最脆弱用路人之運具為類別（優先序：行人 > 自行車/慢車 > 機車 > 汽/貨車）。例如左轉小客車撞死行人，在原資料 P1 為駕駛、運具為汽車，本地圖會歸類為「行人」。

Victim-mode logic: each event is classified by its most-vulnerable road user (priority: pedestrian > bicycle/slow vehicle > motorcycle > car/truck). E.g. a left-turning car that kills a pedestrian is recorded with the driver as party-1 (a car) in the raw data, but is classified here as “pedestrian”.

每萬人率：採用戶政司 dataset 8410 民國 102 年之人口數作為分母，可呈現「不只看絕對數量，而是相對風險」之治理優先序。2013 vs 2025 期間人口分布變化有限，適用於相對比較。

Per-capita rates use population from MOI Household-Registration dataset 8410 (2013) as the denominator, surfacing relative risk rather than raw counts. Population shifts 2013→2025 are modest, so the comparison holds.

邊界圖資：依 g0v/twgeojson 之鄉鎮市區界（1982 年版，整併至 2014 年後行政地理）。連江縣東引鄉與台東縣綠島鄉等少數島嶼之邊界與資料皆有侷限。

Boundaries from g0v/twgeojson district polygons (1982 edition, normalised to post-2014 administrative geography). A few small islands (Lienchiang–Dongyin, Taitung–Green Island) have limited boundary/data coverage.

Sources · National Police Agency (A1) · MOI Household Registration dataset 8410 · g0v/twgeojson district boundaries