氣象領域常採用weather research and forecasting (WRF)模式進行雨量預報，惟其預報結果會隨模式參數而變動，對真實雨量常有不準確的估計。若能有效修正WRF系集預報資料，則可減少預報誤差以提高模式準確度。為此，本研究以雷達估計雨量為觀測值，計算WRF系集預報資料與觀測值之間的雨量誤差。並採用二種不同的機器學習法，支撐向量機與隨機森林，學習WRF系集預報中大量網格預報雨量資料之誤差結構，建立即時誤差修正模式，以修正WRF系集預報在前置時間1至6小時之雨量預報。其中，本研究採二種建模方式配合機器學習法以建置即時誤差修正模式如下：(1)所有系集成員合併建立一個誤差修正模式(以下簡稱合併建模)，其係直接針對系集平均值進行誤差修正；(2)每個系集成員獨自建立誤差修正模式(以下簡稱獨立建模)，其係先針對個別WRF系集成員進行誤差修正，再取修正後預報值之平均。因此，本研究中即時誤差修正模式共有四種，分別為：支撐向量機搭配合併建模、支撐向量機搭配獨立建模、隨機森林搭配合併建模、隨機森林搭配獨立建模。分析結果顯示：合併建模與獨立建模有相似的表現，不同的建模方式於即時雨量預報修正之表現並沒有明顯差異，故於模式結構建議直接採用合併建模，不僅可降低模式複雜性，同時亦能縮減計算時間。此外，本研究所採用之二種機器學習法於大部分颱風事件中表現相近，且以評鑑指標(相關係數與均方根誤差)來看，二種方法皆可顯著地改善WRF系集預報雨量。其中，誤差修正模式於前置時間1至3小時之改善效果較佳，而前置時間4至6小時於部分颱風事件之改善能力則較不明顯。整體而言，本研究提出之即時誤差修正模式可有效提升WRF系集預報之準確度，改善雨量預報值低估之情況，但對於不同機器學習法間的差異則有待進一步地分析與比較。

The main purpose of this study is to correct the errors of weather research and forecasting (WRF) model in precipitation forecasting. The study area is Kaoping River Basin located in southern Taiwan where usually suffers from floods during typhoon season. In this study, the radar precipitation data are regard as the reference (i.e., true values) to eliminate WRF forecasting error. Two methods, random forests (RF) and support vector machines (SVM), are used to correct the discrepancies between the forecasted precipitation data from WRF model and the radar precipitation data. The correction is based on a real-time updating procedure where two error correction models will be updated every 6 hours (i.e., WRF model provides 6-hour forecast in each run). For example, the error correction models applied to correct the 6-hour ahead precipitation forecasts from WRF model which were set up based on the precipitation data (WRF and the radar precipitation data) from the past 6 hours. The results reveal that the error correction models can further improve WRF model rainfall forecasting by using the real-time updating procedure, however, the accuracy of forecasting decreases with lead time increasing. The results also show that either SVM-based model or RF-based model performs well on two performance indexes (i.e., correlation coefficient and root mean squared error) which can improve the accuracy of precipitation forecasting by 5% to 300% when it compares to original WRF model.

1. 王盼（2014）。基於隨機森林模型的需水預測模型及其應用。水資源保護，第30卷，第1期，第34-37頁。
2. 王潔如、黃麗蓉、蕭玲鳳、李清勝（2014）。2014年定量降雨系集預報實驗(TAPEX)結果分析與討論。台灣颱風洪水研究中心，21pp。
3. 林士浩（2011）。六小時雨量預報資訊之即時校正與分析。國立臺北科技大學土木與防災研究所碩士論文。
4. 丘台光、陳家榮、張保亮、林品芳 （2007）。劇烈天氣監測系統QPESUMS之服務與應用。國土資訊系統通訊，61，第15-26頁。
5. 唐玉霜、張保亮 （2015）。劇烈天氣系統(QPESUMS)發展與客製化服務。104年天氣分析與預報研討會，台北。
6. 陳忠煒（2014）。應用遙測及地面雨量資訊於WRF颱風時雨量預報。國立成功大學水利及海洋工程研究所碩士論文。
7. 陳憲宗（2006）。Real-time Probabilistic Flood Stage Forecasting Using Support Vector Machines and Fuzzy Inference Model。國立成功大學水利及海洋工程研究所博士論文。
8. 郭家妏（2014）。隨機森林在河川水位即時預報之應用。國立成功大學水利及海洋工程研究所碩士論文。
9. 張逸凡（2005）。支撐向量機在即時河川水位預報之應用。國立成功大學水利及海洋工程研究所碩士論文.
10. 葉天降、馮欽賜、柳再明、陳得松、黃康寧、陳雯美、汪鳳如、洪景山 （2012）。中央氣象局數值天氣預報作業系統(二)預報模式概況。氣象學報，第48卷，第4期，第69-95頁。
11. 經濟部水利署（2011）。機率式洪水預報系統之研發(2/2)。
12. 鳳雷、陳嬿竹（2012）。定量降雨整合預報技術之研發-2011年颱風定量降雨數值模式系集預報實驗。台灣颱風洪水研究中心技術報告。
13. 劉鑌鋈（2009）。利用機器學習修正QPESUMS雷達估計降雨。國立成功大學水利及海洋工程研究所碩士論文。
14. Breiman, L., & Cutler, A. (2001). Random Forests. Machine Learning, 45(1), 5-32.
15. Chang, P. L., Lin, P. F., Jou, B. J. D., & Zhang, J. (2009). An Application of Reflectivity Climatology in Constructing Radar Hybrid Scans over Complex Terrain. J. Atmos. Oceanic Technol., 26, 1315-1327.
16. Chen, J., Uyeda, H., Lee, D. I., & Kinosita, T. (2005). Establishment of Z-R Relationships for the Baiu Precipitation Using the Window Probability Matching Method. Meteorological Applications, 12, 207-215.
17. Chen, J., Li, M., & Wang, W. (2012). Statistical Uncertainty Estimation Using Random Forests and Its Application to Drought Forecast. Mathematical Problems in Engineering, 1-12.
18. Fang, X., Kuo, Y.-H., & Wang A. (2011). The Impacts of Taiwan topography on the predictability of Typhoon Morakot’s record-breaking rainfall: A high-resolution ensemble simulation. Wea. Forecasting, 26, 613–633.
19. Fang, X., & Kuo, Y.-H. (2013). Improving Ensemble-Based Quantitative Precipitation Forecasts for Topography-Enhanced Typhoon Heavy Rainfall over Taiwan with a Modified Probability-Matching Technique. Monthly Weather Review, 26, 3908–3932.
20. Gagne II, D. J., McGovern, A., & Xue, M. (2011). Machine Learning Enhancement of Storm Scale Ensemble Precipitation Forecasts. Workshop on Knowledge discovery, modeling, and simulation, San Diego, CA, U.S.
21. Genuer, R., Poggi, J. M., & Tuleau-Malot, C. (2010). Variable Selection Using Random Forests. Pattern Recognition Letters, 31(14), 2225-2236.
22. Herrera, M., Torgo, L., Izquierdo, J., & Pérez-García, R. (2010). Predictive Models for Forecasting Hourly Urban Water Demand. Journal of Hydrology, 387, 141-150.
23. Hsiao, L. F., Yang, M. J., Lee, C. S., Kuo, H. C., Shih, D. S., Tsai, C. C., Wang, C. J., Chang, L.Y., Chen, Y. C., Feng, L., Hong, J. S., Fong, C. T., Chen, D. S., Yeh, T. C., Huang, C. Y., Guo, W. D., & Lin, G. F. (2013). Ensemble Forecasting of Typhoon Rainfall and Floods over a Mountainous Watershed in Taiwan. Journal of Hydrology, 506, 55-68.
24. Hsu, C. W., Chang, C. C. & Lin, C. J. (2003). A Practical Guide to Support Vector Classification. Techincal report, Department of Computer Science and Information Engineering, National Taiwan University.
25. Hong, W. C. (2008). Rainfall forecasting by technological machine learning models. Applied Mathematics and Computation, vol. 200, iss. 1, 41-57.
26. Lin, G. F., Chen, G. R., Wu, M. C., & Chou, Y. C. (2009). Effective Forecasting of Hourly Typhoon Rainfall Using Support Vector Machines. Water Resources Research, 45.
27. Lin, G. F., Jhong, B. C., & Chang, C. C. (2013). Development of an Effective Data-Driven Model for Hourly Typhoon Rainfall Forecasting. Journal of Hydrology, 495, 52-63.
28. Lin, G. F., Wang, T. C., & Chen, L. H. (2016). A Forecasting Approach Combining Self-Organizing Map with Support Vector Regression for Reservoir Inflow during Typhoon Periods. Advances in Meteorology, 12.
29. Liong, S. Y., & Sivapragasam, C. (2002). Flood Stage Forecasting with Support Vector Machines. American Water Resources Association, vol. 38, iss. 1, 173-186.
30. Murphy, J. M. (1990). Assessment of the practical utility of extended range ensemble forecasts. Quarterly Journal of the Royal Meteorological Society, 116, 89-125.
31. Prasad, A.M., Iverson, L.R., & Liaw, A., (2006). Newer Classification and Regression Tree Techniques-Bagging and Random Forests for Ecological Prediction. Ecosystems, 9(2), 181-199.
32. Rokach, L. (2010). Ensemble-Based Classifier. Artifical Intelligence Review, 33(1-2), 1-39.
33. Skamarock, W. C., Coauthors 2008: A description of the Advanced Research WRF version 3. NCAR Tech. 113pp.
34. Su, Y. F. (2015). Bias correction of dynamic downscaled typhoons rainfall data for hydrological applications. 20pp.
35. Teng, J., Potter, N. J., Chiew, F. H. S., Zhang, L., Wang, B., Vaze, J., & Evans, J. P. (2015). How does bias correction of regional climate model precipitation affect modelled runoff?. Hydrology and Earth System Sciences, 19, 711-728.
36. Tribbia, J. J., & Baumhefner, D. P. (1988). The reliability of improvements in deterministic short-range forecasts in the presence of initial state and modeling deficiencies. Monthly Weather Review, 116, 2276-2288.
37. Vapnik, V. N. (1995). The Nature of Statistical Learning Theory, Springer-Verlag New York.
38. Wang, Z., Lai, C., Chen, X., Yang, B., Zhao, S. W., & Bai, X. Y. (2015). Flood Hazard Risk Assessment Model Based on Random Forest. Journal of Hydrology, 527, 1130-1141.
39. Wu, S. J., Lien, H. C., & Chang, C. H. (2012). Real-time correction of water stage forecast during rainstorm events using combination of forecast errors. Stochastic Environmental Research and Risk Assessment, 4, 519-531.
40. Xin, L., Recuter G., & Larochelle B. (1997). Reflectivity-rain rate relationship for convective rainshowers in Edmonton. Atmosphere-Ocean, vol. 35, iss. 4, 513-521.
41. Yamaguchi, M., Sakai, R., Kyoda, M., Komori, T., & Kadowaki, T. (2009). Typhoon ensemble prediction system developed at the Japan Meteorological Agency. Monthly Weather Review. 137, 2592-2604.