本研究旨在分析臺灣地區農作物產量和天然災害損失狀況之空間相依情形,以釐清農作物生產遭遇天然災害威脅的潛在系統性風險。故先行以各縣市地區為分析單位,分別探討稻米、蔬菜、水果與加總所有種類農作物之空間相依性;其次,再縮小至鄉鎮地區資料,以農作物產量最豐富的雲嘉彰地區為例,進行不同年度農作物產量和天然災害損失的空間相關分析。本研究為探討農作物產量與災害損失空間相依性的時空分布變化,分別使用三種空間統計方法,包括(1)計算全區域產量與災害損失的空間自相關程度,以評估全區域加總的系統性風險程度之大小。(2)偵測農作物產量與災害損失高度相關的熱點地區,以辨識區域內系統性風險高度聚集的災害損失地點。(3)計算農作物產量與災害損失空間相依性的遞減情形,亦即該空間相依性如何隨距離的增加而消失,此估計區域範圍外之系統性風險小,故可提供未來研究風險分攤之規劃。研究結果顯示,臺灣各縣市地區和雲嘉彰鄉鎮地區的各類農作物產量,均有顯著的空間自相關現象。以臺灣各縣市地區農作物災害損失的空間相依性分析結果而言,相鄰地區各類農作物災害損失的系統性風險並不高,且系統性風險程度和空間分布逐年皆不穩定。然而,雲嘉彰鄉鎮地區在面臨天然災害時,其農作物災害損失存在時間和空間較穩定之系統性風險,但此系統性風險隨著距離的增加而消失,約15-20公里或2-3個相鄰地區左右,農作物災害損失便不存在空間聚集現象。因此,本研究推測農作物災害損失,應可藉擴大不同地區和種類的農作物風險組合來進行有效的風險分攤。
This study thus investigates the spatial dependence of crop yields and losses caused by natural disasters to analyze potential systemic risk in Taiwan. The study takes a spatial statistics approach to conducting an exploratory time-space data analysis for understanding spatial dependence patterns of crop yields and losses at both the county and the township level. Crops are classified into four groups: rice, fruits, vegetables, and all crops. Three spatial aspects will be analyzed, including (1) extent of spatial dependence of crop yields and losses over the entire region, as a predictor of the magnitude of aggregated systemic risk in the region; (2) geographic hotspots of highly correlated crop yields and losses, which identify the locations in the region under significant systemic risks; (3) the rate of spatial dependence decay over space, to explore how to bound geographic areas for effective risk pooling. The results show certain degree of the overall spatial dependence in crop yields at both the county and township level. In addition to steady aggregated systemic risk in crop yields, the clustered hotspots of crop yields are distributed similarly over time and space. On the other hand, the aggregated systemic risk of crop losses is highly fluctuated over time at the county level, and the positive dependency quickly dies off when the distance increases at the township level. The spatial dependence patterns of crop losses look pretty random, implying the possibility of effective risk pooling through the combinations of different crops and production areas.