本研究建立機率式地震風險分析（Probabilistic Seismic Risk Assessment， PSRA）架構，將機率式地震危害分析所提供之地震動超越機率，進一步轉換為台灣住宅損失與人員傷亡等長期風險指標。風險分析架構銜接 TEM PSHA2020 危害度模型，並整合住宅暴露度模型及脆弱度模型進行建模與計算。暴露度模型整合建物圖框、稅籍、人口普查及電信信令人口資料，依資料可得性分別建立圖框模型與人口普查模型，並以不同時段人口描述住宅建物內之人口動態分布；全台住宅暴露模型總重建成本約為 51.2 兆元。脆弱度模型以本土結構參數為基礎，結合 GEM Vulnerability Modeller’s ToolKit（VMTK）流程，建立建物與人口之脆弱度函數。風險運算採用 OpenQuake engine 之隨機事件式風險分析，並建置 2016 年美濃地震與 2024 年花蓮地震災損資料庫，透過情境式分析比對觀測災損，以檢視模型結果之合理性。在未來 50 年 10% 超越機率所對應之長期地震危害度水準下，機率式地震風險分析結果顯示，全台住宅建物年均財物損失約 445.2 億元，年均完全損害建物約 4，823 棟，年均死亡約 419 人。空間風險與震源損失貢獻分析指出，花蓮市、新城鄉、吉安鄉與永康區等行政區在財物損失或人員傷亡指標中呈現較高風險；中洲構造、米崙斷層、彰化斷層與琉球隱沒帶界面等則為主要損失貢獻震源，可作為後續精細危害分析與情境式風險模擬之優先對象。本研究建立由全台 PSHA 延伸至全台 PSRA 之評估架構，並以公開資料與開源工具建構具透明性與可重現性之分析流程。研究成果可作為住宅建物耐震補強、防減災資源配置與長期財務規劃之量化基礎，並提供地震保險與風險移轉評估所需之風險資訊。;This study develops a nationwide Probabilistic Seismic Risk Assessment (PSRA) framework for Taiwan′s residential buildings， converting ground-motion exceedance probabilities into long-term risk metrics. The framework integrates TEM PSHA2020 with exposure and vulnerability models. The exposure model draws on building footprint， tax-registry， census， and mobile-signaling data. Footprint-based and census-based models reflect data availability， while time-of-day populations represent occupant exposure. The aggregate replacement cost is about NT$51.2 trillion. The vulnerability model combines localized engineering parameters from the Taiwan Earthquake Loss Estimation System (TELES) with the GEM Vulnerability Modeller’s ToolKit (VMTK) workflow to develop functions linking damage， loss， and fatalities. Risk calculations use event-based probabilistic risk assessment in the OpenQuake engine， while damage databases from the 2016 Meinong and 2024 Hualien earthquakes are used in scenario-based analyses to assess model performance against observed damage. At the long-term hazard level corresponding to a 10% probability of exceedance in 50 years， the results indicate an average annual loss of about NT$44.52 billion， with 4，823 buildings completely damaged and 419 fatalities per year. Spatial risk and source loss-contribution analyses identify the northern Longitudinal Valley and Tainan metropolitan area as high-risk regions， while risk hotspots shift with aggregate loss， loss ratio， and casualty metrics. The Zhongzhou structure， Milun fault， and Ryukyu subduction interface are major loss-contributing sources. This transparent and reproducible framework， using multiple datasets and open-source tools， provides a quantitative basis for residential retrofitting， disaster-mitigation planning， and risk-transfer assessment.
