本研究旨在結合判別分析（discriminant analysis）與合成少數類過採樣技術（synthetic minority oversampling technique, SMOTE），以石門水庫、曾文－烏山頭水庫、翡翠水庫及德基水庫為研究區域，建立水資源乾旱預警模式，預測水庫未來第一個月與第三個月水情為正常或警戒，做為啟動相關乾旱緊急措施的依據。本研究採用不同時間尺度之標準化水文指標（標準化降雨指標、標準化蓄水量指標、標準化流量指標）作為模式之輸入變量組合，並以水庫水情分類結果為輸出變量；分別應用線性判別分析（linear discriminant analysis, LDA）與二次判別分析（quadratic discriminant analysis, QDA），並加入SMOTE處理樣本不平衡問題，建立水庫水資源乾旱預警模式。本研究依據水情分類準確率，選定各研究區域之最佳輸入變量組合與判別分析方法，並探討SMOTE處理前後是否有提升模式判別準確率。研究結果顯示，加入SMOTE處理，能有效降低類別不平衡對預測結果的影響，整體預測準確率有所提升。預測未來第一個月之水情，在石門水庫與曾文－烏山頭水庫之最佳模式為QDA結合SMOTE，驗證結果之整體準確率為0.92與0.88；翡翠水庫與德基水庫之最佳模式為LDA結合SMOTE，驗證結果之整體準確率為0.90與0.94。預測未來第三個月之水情，在石門水庫與曾文－烏山頭水庫之最佳模式為QDA結合SMOTE，驗證結果之整體準確率為0.87與0.86；翡翠水庫與德基水庫之最佳模式為LDA結合SMOTE，驗證結果之整體準確率為0.86與0.81。

This study aims to develop water resource drought early warning system based on linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA) for Shimen, Zengwen–Wushantou, Feitsui, and Techi Reservoirs in Taiwan. The objective is to predict reservoir water conditions as either “normal” or “alert” for one-month and three-month lead times, providing a scientific basis for initiating relevant drought response measures. Synthetic minority oversampling technique (SMOTE) was applied to address imbalance number of normal or alert data. Model inputs comprise combinations of standardized hydrological indicators at various temporal scales, including the standardized precipitation index, standardized reservoir storage index, and standardized streamflow index, while the output variable is the classified reservoir water status (normal or alert). LDA and QDA were employed to build the early warning models, with the best-performing input combinations determined based on the classification accuracy. The results suggest that for one-month-ahead warning, QDA with SMOTE performs the best for Shimen and Zengwen－Wushantou Reservoirs, with the validation accuracy of 0.92 and 0.88, respectively, while LDA with SMOTE achieves the highest accuracy of 0.90 and 0.94 for Feitsui and Techi Reservoirs, respectively. For three-month-ahead warning, the optimal models are QDA with SMOTE for Shimen and Zengwen–Wushantou Reservoirs (with the accuracy of 0.87 and 0.86, respectively) and LDA with SMOTE for Feitsui and Techi Reservoirs (with the accuracy of 0.86 and 0.81, respectively).

Arunachalam, P., Janakiraman, N., Kumar Sivaraman, A., Balasundaram, A.,Vincent, R., Rani, S., Dey, B., Muralidhar, A., & Rajesh, M. (2022). Synovial Sarcoma Classification Technique Using Support Vector Machine and Structure Features. Intelligent Automation & Soft Computing, 32(2), 1241–1259.
Biala, T., D. Ramahi, A., N Obianom, E., Li, X., & S Schlindwein, F. (2023). Use of AI to Assess Control and Diseased Children at 10 Yrs of Age 2023 Computing in Cardiology Conference (CinC).
Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002). SMOTE: synthetic minority over-sampling technique. Journal of Artificial Intelligence Research, 16, 321–357.
Chiyeaka, O. M., Ewere, O. J., Victor, O., & Shola, O. (2025). Comparative Study of Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA) and Support Vector Machine (SVM) In Dataset. Advanced Journal of Science, Technology and Engineering, 5(1), 70–84.
Finch, W. H., & Schneider, M. K. (2006). Misclassification rates for four methods of group classification: Impact of predictor distribution, covariance inequality, effect size, sample size, and group size ratio. Educational and Psychological Measurement, 66(2), 240–257.
Fisher, R. A. (1936). The use of multiple measurements in taxonomic problems. Annals of eugenics, 7(2), 179–188.
Ghojogh, B., & Crowley, M. Linear and quadratic discriminant analysis: Tutorial. arXiv 2019. arXiv preprint arXiv:1906.02590.
Ghojogh, B., & Crowley, M. (2019). Linear and quadratic discriminant analysis: Tutorial. arXiv preprint arXiv:1906.02590.
Gusyev, M., Hasegawa, A., Magome, J., Kuribayashi, D., Sawano, H., & Lee, S. (2015). Drought assessment in the Pampanga River basin, the Philippines–Part 1: Characterizing a role of dams in historical droughts with standardized indices. Proceedings of the 21st international congress on modelling and simulation (MODSIM 2015), November 29th–December 4th, Queensland, Australia.
Ha, D. H., Nguyen, P. T., Costache, R., Al-Ansari, N., Van Phong, T., Nguyen, H. D., Amiri, M., Sharma, R., Prakash, I., Van Le, H., Nguyen, H. B. T., & Pham, B. T. (2021). Quadratic Discriminant Analysis Based Ensemble Machine Learning Models for Groundwater Potential Modeling and Mapping. Water Resources Management, 35(13), 4415–4433.
Jiang, T., Su, X., Zhang, G., Zhang, T., & Wu, H. (2023). Stimating propagation probability from meteorological to ecological droughts using a hybrid machine learning copula method. Hydrology and Earth System Sciences, 27(2), 559–576.
Kalantar, B., Al-Najjar, H. A., Pradhan, B., Saeidi, V., Halin, A. A., Ueda, N., & Naghibi, S. A. (2019). Optimized conditioning factors using machine learning techniques for groundwater potential mapping. Water, 11(9), 1909.
McKee, T. B., Doesken, N. J., & Kleist, J. (1993). The relationship of drought frequency and duration to time scales. Proceedings of the 8th Conference on Applied Climatology.
Mishra, A., & Desai, V. (2005). Drought forecasting using stochastic models. Stochastic Environmental Research and Risk Assessment, 19, 326–339.
Mishra, A., & Singh, V. P. (2009). Analysis of drought severity‐area‐frequency curves using a general circulation model and scenario uncertainty. Journal of Geophysical Research: Atmospheres, 114(D6).
Mishra, A. K., Desai, V. R., & Singh, V. P. (2007). Drought forecasting using a hybrid stochastic and neural network model. Journal of Hydrologic Engineering, 12(6), 626–638.
Mishra, A. K., Singh, V. P., & Desai, V. R. (2009). Drought characterization: a probabilistic approach. Stochastic Environmental Research and Risk Assessment, 23, 41–55.
Naghibi, S. A., & Dashtpagerdi, M. M. (2017). Evaluation of four supervised learning methods for groundwater spring potential mapping in Khalkhal region (Iran) using GIS-based features. Hydrogeology Journal, 25(1),169.
Sahoo, S. R., Bulasara, P. K., Iyer, S. K., & Sharma, A. (2024). A Robust Model for Early Detection and Diagnosis of Breast Cancer using Linear Discriminant Analysis 2024 8th International Conference on Computational System and Information Technology for Sustainable Solutions (CSITSS).
Vicente-Serrano, S. M., López-Moreno, J. I., Beguería, S., Lorenzo-Lacruz, J., Azorin-Molina, C., & Morán-Tejeda, E. (2012). Accurate computation of a streamflow drought index. Journal of Hydrologic Engineering, 17(2), 318–332.
Wang, J., Plataniotis, K. N., Lu, J., & Venetsanopoulos, A. N. (2008). Kernel quadratic discriminant analysis for small sample size problem. Pattern Recognition, 41(5), 1528–1538.
Wilhite, D. A., & Glantz, M. H.(1985). Understanding: the drought phenomenon: the role of definitions. Water international, 10(3), 111–120.
Yan, T., Wang, D., Xia, T., Liu, J., Peng, Z., & Xi, L. (2022). Investigation on optimal discriminant directions of linear discriminant analysis for locating informative frequency bands for machine health monitoring. Mechanical Systems and Signal Processing, 180.
Yeh, H.-F. (2021). Spatiotemporal variation of the meteorological and groundwater droughts in central Taiwan. Frontiers in Water, 3, 636792.
石洪波、陳雨文、陳鑫（2019）SMOTE過採樣及其改進算法研究综述，智能系统學報，14(6)，1073–1083。
呂季蓉（2006）「台灣南部地區長期乾旱趨勢分析之研究」，國立成功大學碩士論文。
袁倫欽（2005）「水庫供水操作與乾旱預警系統之研究」，國立臺灣海洋大學碩士論文。
莊盛筑（2021）「使用SMOTE方法處理不平衡資料之研究」，國立中興大學碩士論文。
陳弘（2019）「遙相關月雨量預報模式應用於石門水庫乾旱預警」，國立成功大學碩士論文。
楊承道、王俊寓、林士堯（2024）網格化觀測雨量V2版資料（資料生產履歷），臺灣氣候變遷推估資訊與調適知識平台（Taiwan Climate Change Projection and Information Platform, TCCIP）。
劉雅慈（2018）「石門水庫乾旱預警指標之研究」，國立成功大學碩士論文。
蔡沛宏（2020）「判別分析與SMOTE應用於建構乾旱預警模式之研究」，國立成功大學碩士論文。
簡均任（2013）「乾旱指標結合氣候統計降尺度預報於石門水庫供水之乾旱預警應用」，國立中央大學碩士論文。
羅萬倫（2015）「短期氣候預報在石門水庫梅雨期之水資源管理應用」，國立中央大學碩士論文。