| 摘要: | 長期天氣展望對於評估水庫水位與地表取水，用以供給民生用水、農業用水、與工業用水相當重要。傳統上水資源管理多使用逕流的氣候統計值(如: 歷史統計的水庫入流量超越機率)。然而在環境變遷的影響下，氣溫與降雨型態改變不易掌握，增加水資源管理困難度。因此，天氣與氣候預報資訊，如中央氣象局 (CWB) 所發布的天氣預報，能給予更明確的未來天氣趨勢，對於水資源管理更顯得重要。然而過往「利用天氣展望的水資源評估模式」卻無法考慮天氣預報的準確率。使決策無法應準確率調整決策。
以民國 100 年乾旱事件與石門水庫水資源系統為研究案例 (案例一) 。本研究首先評估中央氣象局長期天氣預報於解除限水的效用。假設天氣展望與降雨預報的準確率相符的情境下，不同天氣預報準確率對於決策的影響。為了因應乾旱事件，當年三月一日至六月底實施第一階段限水。當長期天氣預報完全準確時，可於三月上旬決定五月中旬時解除限水；季預報命中率下降至 60$\%$ 時，非完美短期氣候預報資訊顯示，在四月中下旬時，可望於五月中旬時解除限水。依實際中央氣象局目前月、季長期天預報能力 (40$\%$-50$\%$) 所假設的降雨預報情境下，仍具有提早解除限水的潛力，在本研究案例中約可提早1個月解除限水。此結果表現出：若天氣展望能反映命中率，隨著預報命中率降低，決策趨於保守。
因此，接著我們基於貝氏定理提出一套方法，將天氣預報準確率納入「利用天氣展望的水資源評估模式」中。並以石門水庫入流量超越機率預報為例 (案例二) ，計算不同命中率下所預報的入流量超越機率曲線 (\IEPC) 。結果顯示本方法確實可以獲得反映天氣展望 、與天氣預報準確率的 \IEPC 。另外利用預報技術得分 RPSS 來顯示預報準確率的提升對於決策的影響。結果發現，對於部分極端例子(某類別中的過低、過高的事件)，完美類別預報反而使決策成效不如以氣候平均值做的決策好。 (案例三) 即使將預測地入流量分布整理成類別預報，由於依某一天氣類別產生的入流量類別不一定一致，仍會產生類似問題。
最後，考慮水庫初始狀態不同對於抗旱決策的影響為例。假設使用不同準確率的天氣預報的天氣展望評估未來水位，並考慮各降雨展望顯示降雨偏高機會較大。結果顯示，若初始水庫庫容充足時，不論是使用高準確率的降雨預報、或是低準確率的降雨預報給予的降雨展望，皆評估未來不易缺水發生。然而，由使用低降雨預報準確率的降雨展望的未來水位評估開始，隨著初始庫容降低，評估結果開始顯示未來有較高的缺水可能性。使利用降雨展望的抗旱決策，於較低降雨預報準確率下，做出更加保守的決策。因此，本方法對於月、季尺度的水資源管理與風險評估有所幫助。;Long-term streamflow prediction is important not only to estimate the water storage of a reservoir but also the surface water intakes which supply people′s livelihood, agriculture, and industry. Climatological forecasts of streamflow (e.q., exceedance probability curve of inflow from the historical record) have been traditionally used for water resource management. However, due to the effect of environmental change, the transform of future weather conditions becomes more abnormal, impending effective management faces a greater challenge. Therefore, a long-term weather outlook issued by such agency as the Central Weather Bureau (CWB), which provides a clearer trend of future weather condition can be beneficial for water resource management. The decision-making process based on the weather outlooks with lower forecast accuracy should produce a more conservative decision. But the past approaches doesn′t.
In this study, I assessed the applicability of CWB long-term weather outlooks for determining ``the decision of lifting water rationing (water restriction)" first. I used Shimen Reservoir and the drought event in 2011, phase 1 water rationing had executed from March 1 to June 30, as our case study (case study I). By Assuming that the weather outlook reflects the weather forecast accuracy, I studied what effect by the accuracy on the decision-making. According to the weather outlooks (seasonal rainfall outlook) with the accuracy of 100$\%$, I can make a decision in the early March, that the normally supply can start from middle May. With the accuracy of seasonal rainfall outlook of 60$\%$, in the middle and late April, the termination of rationing in middle May can be estimated.
The results show that a more conservative decision is produced by the outlook with lower forecast accuracy (if the outlooks reflects the accuracy).
And next, I applied Bayes′ theorem to derive a method for incorporating the long-term weather accuracy into water resource management based on the weather outlook. The prediction of exceedance probability of Shimen Reservoir inflow is used as the case study (case study II). The results show that our approach can predict the inflow exceedance probability curves (\IEPCs) reflecting the tercile probabilistic weather outlooks and the weather forecast accuracy. I employed a forecast skill score, RPSS (rank probability skill score) to show how the improvement of the weather forecast affects the decision. I found the potential problems of making the decision with this kind of categorical weather forecast: for some extreme event in a class, perfect rainfall forecast causes the performance of the decision worse than the decision based on the climatological forecast.
(case study II) Even if I arrange the predicted inflow distribution into categorical inflow forecast, the similar problem may arise, due to the rainfall class does not necessarily coincide with the class of the produced inflow.
Last, I considered the decision against water shortage with different the initial water storages of the reservoir. I assumed that the decision maker applies outlooks from different accuracies rainfall forecast.
If the storage if full, all of assessments (based on different weather accuracies) suggest the shortage happens with little chance. Beginning from the assessment based on the outlook from the rainfall forecast with the lowest rainfall forecast accuracy, as the initial storage decreases, the chance of happening water shortage increases. This approach should be useful for the seasonal planning and management of water resource and their risk assessment. |
