變動尺度標準化降雨指標及降雨量序率模擬於乾旱監測與早期預警之應用

Abstract

在台灣，由於人口密度高、水庫面積小，以及降雨分布不均等問題，乾旱仍是棘手的議題。為了確保供水無虞，需持續監測乾旱的情形並於水資源管理進行調整。標準化降雨指標(Standardized Precipitation Index; SPI)已被廣泛用於描述與監測乾旱事件，然而，由於SPI固定的記憶長度，使得乾旱可能太晚被偵測、或太早被判定結束。本研究基於原有SPI的架構，發展出變動尺度SPI，令乾旱事件的選取能與事實相符。此外，亦從SPI建構馬可夫模型(Markov model)，以評估乾旱持續的可能性。為了建構馬可夫模型，歷史雨量資料過小的樣本量需要被擴大，增加的樣本量仍須保持與歷史資料相同的統計特性。因此，將以半變異元(Semi-variogram)建立時間、空間的相關性，再依此相關性以序率模擬產生大量樣本。上述方法將以歷史乾旱事件進行驗證，並提供未來乾旱監測上的建議，政府單位亦可依此研究結果於水資源管理制定相應之措施。

Because of the high population density, small area of reservoirs, and uneven precipitation, drought remains a serious issue in Taiwan. To ensure uninterrupted water supply, the drought situation should be monitored all year round, and adjustments should be made to water management measures. The Standardized Precipitation Index (SPI) has been commonly applied to monitor and describe the process of drought events. However, the SPI may delay detecting or prematurely end a drought event owing to its constant memory. Therefore, based on the original SPI structure, a variable-scale SPI is proposed in this study to capture drought events that corresponds to reality. In addition, a Markov model is developed based on the SPI to evaluate the possibility of drought continuation. To construct the Markov model, a small sample of historical rainfall data should be expanded into a large sample while maintaining the same statistical characteristics. Thus, a semivariogram was applied to build spatial and temporal correlations. From these correlations, several simulated samples were generated through stochastic simulation. These methods were examined using several past drought events, and suggestions about drought monitoring are provided. The government or policymakers can apply these results to water management procedures.