氣候變遷下的美國野火損失估計

Abstract

本研究旨在探討氣候變遷對美國野火損失的影響。本文建構了兩種估計方法：直接估計、兩階段估計，並分析何種方法於預測野火損失表現較佳。在兩種估計方法中，我們使用NOAA Storm Event Database提供的野火數據與PRISM氣候小組提供的歷史氣候數據建立了兩組函數，並以面板資料分析溫度、降水量與野火損失之間的關係、以機器學習方法估計野火發生的機率。研究結果顯示，隨著溫度上升，野火損失有增加的趨勢；隨著降雨變多，野火損失有降低的趨勢。此外，本研究使用IPCC-WGI AR6互動地圖數據集提供的未來氣候預測數據，在RCP4.5和RCP8.5路徑下，預測了美國各州至本世紀末的野火損失。研究發現，RCP8.5相較其他路徑降水量上升較多，降水量抵消了部分野火損失。加總2024年至2100年預測的野火損失，在RCP4.5下總損失金額約為1.8至2.7千萬美元、RCP8.5為3.0至5.4千萬美元。我們的結果一致顯示加州、密西根為未來野火損失較高的州。此外，野火分佈將可能發生改變，過去野火損失不高的州將有一定機率發生更頻繁且帶有損失的野火，如密西根州、田納西州與內布拉斯加州。

This paper analyzes the impact of climate change on wildfire losses in the United States. We constructed two approaches: direct estimation and two-stage estimation, to identify the better-performing method for predicting future wildfire losses. For both estimation methods, we built two distinct functions using wildfire data from the NOAA Storm Event Database and historical climate data from the PRISM Climate Group. We used panel analysis to examine the relationship between temperature, precipitation, and wildfire losses and applied machine learning methods to estimate the probability of wildfire occurrence. Furthermore, this research used future climate prediction data from the IPCC-WGI AR6 Interactive Atlas dataset to predict wildfire losses in each state of the United States until the end of this century under RCP4.5 and RCP8.5 scenarios. The results show that as the climate warms, high temperatures under the RCP8.5 scenario could significantly increase future wildfire losses, while increased precipitation can mitigate losses to a certain extent. The total predicted wildfire losses from 2024 to 2100 are estimated to be approximately 18 to 27 million USD under RCP4.5, and 30 to 54 million USD under RCP8.5. Our results consistently identify California and Michigan as states with higher future wildfire losses. Additionally, states with historically low wildfire losses show an increased probability of experiencing more frequent wildfires, suggesting a shift in wildfire distribution from the West to other states like Michigan, Tennessee, and Nebraska.