| 摘要: | 湄公河流域 (MRB) 是東南亞最大、最重要的跨國流域之一，近年來，當地民眾受到自然災害與人為脆弱性相結合的影響，對該地區的漁業、農業和河流生態系統造成嚴重的衝擊，本論文使用歷史數據分析、水文模式和機器學習技術三種方法，探討MRB乾旱、流量和輸砂量的特徵。根據ENSO指數與降水的相關性分析，若在12-2月發生強烈的ENSO事件可能導致3-5月的氣象乾旱，進而造成4-6月的水文和農業乾旱，這種ENSO效應對MRB南部的乾旱變化有顯著地影響，而在MRB北部地區則不明顯。根據流量和泥沙輸送量變化的結果，以1970-1991的乾季作為基期，在2012-2019乾季鄰近瀾滄水壩的清盛河流量增加73.65%，而在雨季則減少29.46%。此外，在高壩影響期（2012-2019），清盛和桔井流域的泥沙輸送量分別減少69.15%和68.45%，這表示中國水壩在MRB上游和當地其他水壩攔截泥沙所造成的影響。對於流量預測，本研究所提出的深度學習模式能夠很好地捕捉高壩開發期間（2010-2019）的MRB河流流量，短期和長期流量預測的NSE值在0.63至0.88之間，並且Long short-term Memory (LSTM) 模式在高壩影響期的流量預測 (NSE ≥ 0.8) 優於其他預測流量的模式。本研究的結果可應用於開發乾旱監測方法，並且建議未來開發MRB流域時需要仔細評估高流量和泥沙的變化，並及時適應未來的變化。;Mekong River Basin (MRB) is one of the largest and most significant transboundary river basins in southeast Asia. In recent years, natural hazards compounded manmade vulnerability have affected to local people, resulting in further impacts on fisheries, agricultural productions, and river ecosystems in this region. This thesis presents characteristics of droughts, streamflow, and sediment loads in the MRB using three main methods, including historical data analysis, hydrological model, and machine learning techniques. In terms of El Niño Southern Oscillation (ENSO)-droughts connection, the correlation analysis between ENSO index and precipitation suggested that the strongest ENSO events in Dec-Jan-Feb may result in developments of meteorological drought in Mar-Apr-May, and further led to hydrological and agricultural drought in Apr-May-Jun. Such ENSO effects had significant influences on drought variabilities in southern MRB and were insignificant in the north. Results from changes in flow and sediment show that the river flow at Chiang Saen, which is the closest station to the Lancang cascade dams, has risen by 73.65\% in the dry season during 2012-2019 in comparison with the period 1970-1991 while that has reduced by 29.46\% in the wet season. Besides, the sediment loads have reduced by 69.15\% and 68.45\% at Chiang Saen and Kratie, respectively in the high-dam period (2012-2019), indicating a drastic reduction of sediment owing to the sediment trapping by Chinese dams in the upper MRB and local dams. For streamflow prediction, the proposed deep learning models were able to capture the fluctuation of river flow in the MRB during the period of high-dam development (2010-2019), with the Nash-Sutcliffe (NSE) values ranged from 0.79 to 0.88, and the Long short-term Memory (LSTM) still outperformed other models for streamflow prediction in an era of mega dams (NSE $\geq$ 0.8). The results can be applied to the development of drought monitoring methods, water management, and supporting for development planning and strategic decision making, especially in the context of rapid development in the Mekong river basin. |
