氣候變遷下石門水庫泥砂淤積影響評估

Abstract

由於臺灣地形狹長且河川坡陡流急的特徵，上游河川水流於豪雨與颱風期間易夾帶大量泥砂進入水庫，長期造成了淤積問題，目前石門水庫的有效運作僅達到其儲水容量的三分之二，另有三分之一的庫容被沉積的淤泥所佔據，對北部地區供水安全與調度能力構成潛在風險。隨著全球氣候變遷影響加劇，極端降雨事件之頻率與強度預期將增加，可能進一步改變集水區水文過程與泥砂輸移機制。因此評估不同氣候情境下石門水庫泥砂累積與庫容演變趨勢，對於水庫永續經營與風險管理具有重要意義。 本研究整合氣象合成模式（WGEN）產製之未來氣候資料與台灣氣候變遷推估資訊平台（Taiwan Climate Change Projection Information and Adaptation Knowledge Platform, TCCIP）之統計降尺度資料，作為不同氣候輸入來源，並應用水文模式 Generalized Watershed Loading Functions (GWLF) 模擬集水區流量程，推估石門水庫入流量與泥砂輸移情形。在不同排放情境（SSP1-2.6、SSP5-8.5）及清淤策略下，分析未來至2060年之庫容變化趨勢，並納入不同清淤策略情境，評估工程管理措施對庫容維持之影響。 研究結果顯示，氣候資料來源之統計特性差異將影響流量分布形態與尾端行為。TCCIP 日資料於低雨量區段呈現集中現象，高分位尾端延伸能力相對有限；合成氣象資料則可提供較多自然變異取樣範圍，使流量尾端行為更具彈性。時間尺度分析亦指出，月尺度下流量分布較為穩定，而較短時間尺度下差異放大，顯示氣候輸入之統計結構對逕流模擬具有尺度依賴性。在未來情境推估方面，若未實施清淤措施，水庫有效容量將呈現持續下降趨勢；透過適當清淤策略可有效延緩容量衰減。多模式整合有助於降低單一氣候模式之偏差影響，但氣候模式差異、合成模式隨機性及水文模式結構假設仍為未來推估之主要不確定性來源。

Due to Taiwan’s elongated topography and steep river gradients, intense rainfall and typhoon events frequently transport large amounts of sediment from upstream catchments into reservoirs, leading to long-term sedimentation problems. At present, the effective storage capacity of Shimen Reservoir has been reduced to approximately two-thirds of its original design capacity, with the remaining one-third occupied by accumulated sediment, posing potential risks to regional water supply security and operational flexibility in northern Taiwan. Under ongoing climate change, the frequency and intensity of extreme precipitation events are projected to increase, potentially altering watershed hydrological processes and sediment transport mechanisms. Therefore, evaluating sediment accumulation and reservoir capacity evolution under different climate scenarios is essential for sustainable reservoir management and risk mitigation. This study integrates future climate data generated by a weather generator model (WGEN) and statistically downscaled data provided by the Taiwan Climate Change Projection Information and Adaptation Knowledge Platform (TCCIP) as alternative climate inputs. The Generalized Watershed Loading Functions (GWLF) model is applied to simulate watershed runoff processes, from which inflow and sediment transport to Shimen Reservoir are estimated. Under different emission scenarios (SSP1-2.6 and SSP5-8.5) and dredging strategy scenarios, reservoir capacity changes are projected through 2060, and the effectiveness of engineering management measures in mitigating storage loss is assessed. Results indicate that differences in the statistical characteristics of climate data sources influence flow distribution patterns and tail behavior. TCCIP daily rainfall data exhibit concentration in low-intensity rainfall ranges and relatively limited extension in upper quantiles, while synthetic weather data provide a broader range of natural variability, enhancing the representation of extreme flow behavior. Temporal scale analysis further reveals that flow distributions are more stable at the monthly scale, whereas discrepancies become amplified at shorter time scales, indicating a scale-dependent influence of climate input statistics on runoff simulations. Under future climate scenarios, reservoir effective storage is projected to decline continuously without dredging measures, whereas appropriate dredging strategies can substantially mitigate storage loss. Although multi-model integration reduces the bias associated with individual climate models, uncertainties arising from climate model variability, stochastic sampling in synthetic weather generation, and structural assumptions in the hydrological model remain key sources of projection uncertainty.