都市災害中，以火災為最常見的災害類型。以往都市火災文獻，多以臺灣直轄市等都會區為對象。相關文獻以建物外在環境空間討論，則以火災量化資料進行敘述統計，或經傳統複迴歸分析。然而，有關火災空間區域與其鄰近區域屬性相關性，及區域間卻非同質的差異性，於文獻中甚少討論，這將使都市建築物火災理論解釋力，因空間變異影響而降低。爰此，本研究以中型城市彰化縣彰化市為案例，採用Ripley’sK函數分析檢定建築物火災空間形態，呈現聚集分布，再依KDE及Gi*找出潛勢區及熱區。研究成果發現，最高潛勢區及熱區皆集中於都市計畫區內。因此，透過都市建築火災理論，選取人為社經層面及都市環境特徵等變項。以最小統計區為樣本，經MLR、SLM與SEM，及GWR進行分析。實證結果以GWR解釋力最佳，並驗證建築物火災與區域空間，存有空間變異之現象。於GWR結果分析中，除人口密度、混合使用住宅區，與以往國內外研究中大都會區常見研究因子一致外。同時觀察到教育程度，甚少被討論。除間接以社經地位及收入等影響火災發生，亦經鄰里效應降低發生率。因此，本研究藉由實證成果，給予短中長期策略。短期先建立公助、自助、互助之防災社區，並增進社區自主防火能力。在中期，鼓勵老舊複合式住宅進行修繕和強化，提升危老重建效率，增強民眾居住安全。長期則將建築物火災潛勢區及熱區、GWR分析模型顯著區域，納入地區災害防救計畫等，以強化都市防災能力，並增進都市計劃與地區災害防救計畫間之交互聯繫。藉由本研究實證成果，除提供更詳細建築物火災發生率空間分析，亦供有關單位都市防災規劃參考。

Among urban disasters, fire is a common disaster in urban areas. In the past, urban fire literature was mostly focused on metropolitan areas such as Taiwan municipalities. It is often analyzed on Environmental space presented with fire statistic description, that often studying with traditional multiple regression. However, the spatial autocorrelation and spatial heterogeneity of fire spatial region have not been carefully studied in literature, which will lack of the explanatory power of urban fire theory due to spatial variability. Therefore, this study we choose Changhua City, a medium-sized city in Taiwan, as a case study. And that Ripley’s K-function to find out the spatial aggregation phenomenon in the building fire, and then uses the kernel density estimation and the index of the Local Indicators of Spatial Association (Gi*) find out the building fire potential and the so called “hot zone”. Our results show that the high potential area and the hot zone are concentrated in the urban area. That also indicates that the Anthropogenic socio-economic characteristics and urban environment characteristics were selected through the urban buildings fire theory. Taking the smallest statistical area as a sample, it is analyzed through traditional multiple regression, spatial lag model and spatial error model, and geographically weighted regression. Our empirical findings concluded that the best model is the geographically weighted regression, and confirmed the hypothesis of the spatial variability in building fire and regional spatial characteristics. The empirical results show that building fire incidence rate was positively correlated to variables of population density, area ratio of mixed-use residential district, while the population of higher education is negatively correlated.