| 摘要: | 熱帶氣旋（TCs），通稱颱風，對西北太平洋（WNP）地區的多個國家產生深遠影響。本論文研究了颱風的動力學及其對台灣乾旱的影響，使用了統計技術、小波相干分析、核密度估計和機器學習（ML）模型。在2020-2021年的嚴重乾旱之後，了解颱風特徵與乾旱嚴重程度之間的互動對於改進預測模型和災害緩解策略至關重要。本研究利用了各種數據集，包括颱風最佳路徑記錄、衛星推導的降水數據和氣候指數；進行了統計分析，包括相關矩陣、核密度和小波相干，來探索周期變化和長期趨勢；標準化降水指數（SPI）用於量化乾旱嚴重程度；機器學習模型估算了颱風強度，考慮了如中心氣壓、風暴速度和氣候階段等因素。結果顯示，颱風特徵（如頻率、持續時間、路徑長度和風速）與台灣的乾旱強度之間存在正相關之關係。相比之下，西北太平洋的颱風持續時間和路徑長度與台灣的乾旱指數之間存在負相關，這主要受大規模大氣條件驅動。西北太平洋的颱風活動與台灣的長期乾旱模式相關，尤其是在十年尺度上。按季節劃分，乾旱嚴重程度在冬末和早春達到最高水平，主要集中在台灣的中部和東南部。對1979年至2020年西北太平洋颱風活動的空間分析顯示，颱風活動主要集中在菲律賓和台灣東部，特別是在10°N和25°N緯度之間；季節性高峰出現在7月和8月，ENSO和PDO階段影響颱風的形成和分布。這些結果深化了對颱風模式的理解，幫助改進預測模型和災害準備工作；評估了八個機器學習模型來預測颱風強度，其中Cubist和RF模型表現最佳；中心氣壓顯示為最具影響力的預測因子，而在暖ENSO階段的預測誤差最高。這一綜合分析提升了對西北太平洋颱風動力學的理解，提供了改進預測模型和災害準備的堅實框架，最終有助於氣候韌性和風險管理政策的制定。;Tropical cyclones (TCs), commonly referred to as typhoons, have a profound effect on several countries in the western North Pacific (WNP). This dissertation investigates the dynamics of typhoons and their impact on droughts in Taiwan, using statistical techniques, wavelet coherence analysis, kernel density estimation, and machine learning (ML) models. Following the severe drought of 2020-2021, understanding the interaction between typhoon characteristics and drought severity is crucial for improving predictive models and disaster mitigation strategies. This research makes use of various datasets, including TC best-track records, satellite-derived precipitation data, and climate indices. Statistical analyses, including correlation matrices, kernel density, and wavelet coherence, were conducted to explore periodic variations and long-term tendencies. The Standardized Precipitation Index (SPI) quantified drought severity. ML models estimated TC intensity, incorporating factors like central pressure, storm speed, and climatic phases. The findings revealed a positive relationship between typhoon features (such as frequency, duration, path length, and wind speed) and drought intensity in Taiwan. In contrast, negative correlations were noted between typhoon duration and path length in the WNP and drought indices in Taiwan, driven by large-scale atmospheric conditions. Typhoon activity in the WNP was linked to long-term drought patterns in Taiwan, especially over decadal timescales. On a seasonal basis, drought severity reached its highest levels in central and southeastern Taiwan during late winter and early spring. The spatial analysis of TC activity in the WNP from 1979 to 2020 highlights concentrated typhoon activity east of the Philippines and Taiwan, particularly between 10°N and 25°N latitude. Seasonal peaks occur in July and August, with ENSO and PDO phases influencing typhoon formation and distribution. These results deepen the understanding of typhoon patterns and aid in refining forecasting models and disaster readiness in the WNP area. Eight ML models were evaluated for predicting TC intensity, with Cubist and RF models performing best. Central pressure emerged as the most influential predictor, with highest prediction errors during warm ENSO phases. This comprehensive analysis enhances understanding of typhoon dynamics in the WNP, providing a robust framework for improving predictive models and disaster preparedness, ultimately informing climate resilience and risk management policies. |
