局部細化淹水平行演算之研究

Abstract

台灣地區地理位置與氣候特殊，每遭颱風或暴雨事件，往往造成中下游地區淹水氾濫，導致生命財產嚴重損失，應用二維淹水模式模擬淹水潛勢圖供決策單位及民眾因應預警，將有助於減輕洪水帶來之損失。 良好的數值模式，需兼顧準確性、效率性及穩定性，隨著近年遙測技術迅速發展，地形資料解析度提高，使得淹水模式準確性大幅提升。但假若使用高精度的數值地形資料(DTM)，卻使得二維淹水模式演算時間過於冗長，效率大幅降低。本研究使用多重尺度局部細化方法於二維淹水模式當中，運用不同解析度之粗、細網格，進行多重尺度之模擬，將可減少網格計算量，提高演算效率，同時兼顧模擬之準確性。 隨著科技日新月異，除了地形解析度及相關觀測資料趨於龐大完整外，電腦處理器也大幅發展進步，程式開發現已普遍朝向多執行緒之平行化處理發展，因此在二維淹水模式中，輸入資料解析度提升之同時，計算資源上也應加以有效利用及優化，本研究於二維淹水程式中修改加入平行演算法，於多重尺度局部細化方法當中，將全區粗網格與數個局部細化網格區域，獨立交由不同執行緒進行平行演算，以期提高二維淹水模式之演算效率。 本研究以曾文溪流域北岸為實際模擬案例，採用2009年8月重創台灣南部之莫拉克颱風事件進行模式精準度及演算效率度分析探討。經由案例模擬，平行演算法應用於多重尺度格網方法當中，局部高解析細格網之模擬結果，達到與全區細網格相符之水準，且搭配平行演算法，使演算時間大幅減少、效率大幅提升。本研究在兼顧準確與穩定性的同時，能快速演算有效地提供即時淹水資訊，作為防災管理決策支援達到減災之效果。

In Taiwan, due to the special physiographic and climatic conditions, typhoon or rainstorm event occurs frequently. Besides, it usually causes disastrous inundation in the midstream and downstream floodplain, and takes losses of lives and property. The two-dimensional inundation model can simulate flood risk map in advance. It predicts the potential flood depth and velocity, enabling informed government improve emergency planning and policy to reduce the losses. A good numerical model should give consideration to accuracy, efficiency and stability. However, it will cost large amount computational time in large fine resolution simulate areas. The multi-scale inundation simulation model using local fine grid at the dense population areas could reduce the time of calculation. The method is a feasible solution to provide a much higher resolution mesh with less computational time. Recently, with the development of science and technology, the developments in computer hardware have greatly advanced. Computational time can be reduced by running in parallel over multiple threads. The main purpose of this research focuses on parallel the multi-scale computational method. In multi-scale inundation model, this research parallelizes global domain and all the local regions to different threads by parallel algorithm. The method makes the flood inundation model would be reduced computational time and speedup the efficiency. The parallel method will be applied to the simulation of typhoon Morakot event of 2009 in Tseng-Wen river basin as the study case. This study compares accuracy and efficiency by the error correction and speedup ratio. The results show that it is not only improving the accuracy of simulation error to close the level of fine-grid simulation in the global area, but also raising the efficiency of computational time. The method of parallel for multi-scale inundation simulation strikes a good balance between accuracy and efficiency. The significant result of this study will greatly improve the real-time inundation early-warning system, and this high efficiency model would provide instant inundation information to authority, and it will assist the disaster reduction.