圖書典藏及影音數位平台

雷達雨量資料須經地域性之修正與地形效應之改正，本計畫將以陳有蘭溪流域為研究區域，建立利用雷達資料於陳有蘭溪之雨量預測模式，提高降雨與流量預測準確度，以進一步有效執行預警系統。陳有蘭溪為濁水溪主要支流，因集水區地質破碎導致許多崩塌，以致河床因土沙淤積而抬升，過去頻傳災情，如93年敏督利颱風、98年莫拉克颱風均帶來驚人雨量及重大災情，然兩次災害中均因疏散避難而幾無人命傷亡，足見疏散避難之成效。而流域內雨量站僅有龍神橋、內茅埔、西巒、溪頭、望鄉、和社、東埔、阿里山、新高口、玉山、排雲等11站擁有20年以上之雨量資料。這些雨量站間之距離至少有數公里且都集中於河道附近，對集水區上游之降雨在空間分布上描述十分不足。以修正後雷達雨量做出具時間效率(10分鐘更新)與空間準確度(1.25kmx1.25km網格)的降雨估計，以進一步以水文模式預估下游流量，以為防救災決策參考，並以陳有蘭溪流域重大之歷史颱風事件作為檢定與驗證，故擬定此研究計畫。

The increased frequency of typhoon and other storm events as a result of global climate changes that alter the extreme weather pattern in Taiwan has brought tremendous amount of rain causing devastating landslides, debris flows, and flooding. Severe property damage and loss of lives are often inevitable. On a larger scale, the country’s economy could be affected due to the impact suffered by the agricultural, tourism and more industries. Therefore, reliable warning systems are in great demand to provide emergency response agencies with vital information for early action planning in an effort to minimize any potential impact. One of the systems that utilizes radar reflectivity measurements and integrates other weather information such as wind speed to provide past (72 hours) and forecast (0 ~ 3 hours) high-resolution precipitation data is the QPESUMS (Quantitative Precipitation Estimation and Segregation Using Multiple Sensor). This project plans to extract the gridded precipitation data from the QPESUMS and apply proper adjustments to the values, then use them as inputs to a hydrologic model in hope to accurately simulate the hydrologic responses to storm events. Accurate hydrologic modeling relies on accurate rainfall inputs. Chenyulan River watershed was selected as the study area. Several geostatistically based spatial interpolation methods including Ordinary Kriging (OK), Inverse Distance Weighted (IDW), SPLINE, Co-Kriging, and Regression Kriging (RK) were tested for suitability of rainfield interpolation over the study area. OK, IDW, SPLlNE used only rain gauge data. Elevation information of the rain gauges in the watershed and surrounding area was incorporated as a secondary predictor for Co-Kriging. Gridded radar rainfall from the QPESUMS and elevations of the rain gauges were used as secondary predictors for RK. The results showed that RK outperformed OK, IDW, SPLINE, and Co-Kriging in this project. According to the literature review, Co-Kriging and RK are superior spatial interpolation methods for rainfield interpolation. One advantage that RK has over Co-Kriging is computational efficiency. Since time is such a critical element in flood warning, RK is clearly the suitable choice. For the RK rainfield interpolation model, historical typhoons Fungwong and Kalmaegi were selected for model parameter calibration. Typhoons Sinlaku, Morakot and Fanapi were used for model validation purposes. Assuming rain gauge observations are ground truth, the interpolated rainfall and radar values of the QPESUMS at the rain gauge locations were compared against the observations. It was shown that the percent error of the interpolated values as compared to the un-adjusted QPESUMS data was reduced by 3.6% ~ 53.4%. The only exception occurred at rain gauge Shan-An Bridge, where the percent error was increased by 10.3%. The reason for that was the inconsistency between the radar and gauge measurements at a nearby gauge station causing over-estimation during the Kriging process of the RK model. In order to quantitatively compare the different types of interpolation approaches, RMSEs for each approach were calculated. In addition, 45-degree (1:1 ratio) plots as well as time variation plots were generated comparing interpolated values to gauge measurements. The outstanding performance of the RK model can be easily seen. From the comparisons, the RK is again demonstrated to be able to produce better adjustments for the radar rainfall data than the Co-Kriging approach does. Finally, the interpolated rainfields of storm 610 and typhoon Sura (both of which occurred in 2012) were ingested into HEC-HMS hydrologic model for outflow simulations. Five indices including CE, EQp, ETp, MAE, and MAPE were determined to evaluate the quality of the simulated hydrographs. In addition to the five indices, the shapes of the simulated hydrographs play a very important part of the evaluation process. For the case of storm 610, the simulated hydrograph generated for this project was compared to the one produced by the existing model from the 4th River Management Office. Except for CE, the hydrograph for this project yielded better values for the remaining four indices. The shape of the hydrograph also had a closer fit to the observed hydrograph. For the case of typhoon Sura, both hydrographs (produced for this project and by the 4th River Management Office) did not appear to have such a closer fit as that of the case of storm 610. By comparing only the peak flow values and arrival time, the peak flow for this project was slightly greater than the observed value and was considered to be conservative for flood warning. The arrival time was closer to the observed one and could be provided for evacuation planning.