本研究針對人為因子對臺灣熱浪次數及日數發生機率影響做歸因分析，運用世界氣候研究計劃氣候變動研究項目下的二十世紀氣候模擬偵測與歸因子計畫模式資料庫，以歷史情境和只有自然驅力情境下的區域熱浪模擬結果進行比對分析。因應區域分析的需求，所選取的是解析度較高且有大量系集成員的CAM5.1與HadGEM3模式，同時也使用兩種不同的熱浪定義方法來偵測熱浪事件，以比較量化歸因研究結果對於運用不同模式模擬和熱浪偵測方法的敏感程度。並使用臺灣氣候變遷推估資訊與調適知識平台計畫所產製的臺灣網格化近地面氣溫資料，評估驗證模式在區域熱浪模擬能力。雖然在日最高溫度強度的模擬上有一定的偏差，但在熱浪次數和日數的年際變化上，兩組模式都有很好的表現，系集平均後的時間相關係數都在0.8以上，但在平均每年熱浪次數與日數的表現上，CAM5.1、HadGM3模式模擬結果都比觀測偏多。運用對照數值實驗的大量系集所求得的臺灣熱浪次數與日數機率分佈差異，進一步定量推算人為活動對熱浪次數與日數風險變化影響的可能變動範圍。研究發現過去人為所造成的暖化，「非常可能」（有90％以上的機率）至少使臺灣熱浪次數與日數增加2.5至5.3倍，其中以HadGM3模式系集模擬所估算的人為作用比較大，風險增加的程度幾乎是以CAM5.1模式系集模擬所估算的2倍。另外，若使用不同方式定義熱浪時，熱浪特性所造成系集機率分佈的改變，會讓人為活動所造成的風險略為變動。

Our study aims at the quantitative attribution of human impact on the changes in the extreme heatwave event characteristics over Taiwan. The analysis uses large ensemble simulations from WCRP/CLIVAR Climate of the Twentieth Century Plus Project (C20C+) Detection and Attribution subproject. The regional heatwave events are identified and compared in historical runs with both all forcings and natural forcings only conditions. Considering the limited analysis domain, and large ensemble data availability, we select only high-resolution CAM5.1 and HadGEM3 models from the archive for analysis. We also apply different heatwave definitions to examine the sensitivity of analysis results to heatwave definition in addition to model selection. Using the observational data of daily maximum temperature from Taiwan Climate Change Projection and Information Platform (TCCIP) project, we first evaluate the model reliability on simulating regional heatwave characteristics. Although there are model biases associated with the daily maximum temperature, using percentile thresholds to define heatwaves, the interannual variability of ensemble-mean annual heatwave frequency and days are well simulated with a temporal correlation greater than 0.8 (significant at 99% level). Nevertheless, both CAM5.1 and HadGEM3 models tend to overestimate the heatwave frequency and days. The fraction of attributed risk (FAR) due to past human-induced warming was estimated from probability distributions of regional heatwave characteristics based on large ensemble simulations with and without anthropogenic forcing. Our study found that “very likely”, with 90% confidence level, impacts from human activities at least increased the risk of heatwave frequencies and days in the present-day climate to 2.5 to 5.3 times of those from the simulation with only nature forcing. The risk increase estimated by HadGEM3 ensemble simulation was almost double that estimated by CAM5.1. With different heatwave definitions, their impact on the probability distribution of heatwave activities from ensemble simulation tends to slightly change the human-induced risks.