本研究發展翡翠、石門、德基、曾文四個水庫集水區之遙相關月雨量預報模式，模式建置過程採用多種遙相關指標作為候選輸入變量，考量所有變量組合進行建模測試，模式建置採用之機器學習方法為隨機森林（random forests）。預測模式分為兩種架構：(1)僅使用遙相關指標作為輸入變量之基準模式，(2)納入標準化降雨指標（standardized precipitation index, SPI）作為輔助變量之輔助模式；模式輸出變量為未來一至三個月之降雨指標，時間尺度涵蓋一、三、六個月的標準化降雨指標（SPI-1、SPI-3、SPI-6）。本研究分別針對四個水庫集水區與不同時間尺度之SPI預測目標，於基準模式與輔助模式兩種模式下分別進行變量篩選與模型訓練，選出對應之最佳遙相關變量組合，以建立最符合資料特性之預測模型。分析結果顯示，在預測SPI-1時，無論預測時刻為未來一、二或三個月，加入SPI後並未顯著提升模式表現，顯示SPI-1所代表之短期水文訊號對於未來降雨趨勢的效益提升有限。於預測SPI-6結果上，加入SPI能穩定提升模型表現，顯示中長期的累積降雨訊號有助於強化遙相關數據預測未來降雨趨勢。在所有SPI類型中，於未來一個月的預報，加入SPI輔助變量對SPI-3與SPI-6預測效果的提升較為明顯，顯示在短期預測中，加入SPI有助於補強遙相關指標的訊息。而在未來三個月的預報，部分水庫集水區與SPI時間尺度，純使用遙相關指標的基準模式預測表現反而優於加入SPI的輔助模式，可能與SPI訊號隨時間延伸後其解釋力減弱，或與遙相關訊號之延遲影響產生干擾有關。本研究亦將預測之SPI結果還原為雨量預報值，以評估模型對實際水資源狀況的解釋能力。

This study developed teleconnection-based monthly rainfall forecasting models for four major reservoir catchments in Taiwan: Feitsui, Shimen, Deji, and Zengwen. The modeling process utilized various teleconnection indices as candidate input variables, and all possible combinations of these variables were tested to identify optimal model configurations. The machine learning method adopted was the random forest algorithm. Two predictive frameworks were constructed:(1) a baseline model using only teleconnection indices as input variables, and(2) an auxiliary model that incorporates the standardized precipitation index (SPI) as an additional input variable. The target outputs were SPI values at one- to three-month lead times, covering three time scales: SPI-1, SPI-3, and SPI-6. For each reservoir catchment and SPI target, variable selection and model training were conducted separately under both frameworks to identify the most suitable combination of teleconnection indices for each case. The results show that, for SPI-1 prediction, incorporating SPI did not significantly improve performance across all forecast lead times. This suggests that short-term hydrological signals captured by SPI-1 contribute limited additional value to rainfall trend forecasting. In contrast, for SPI-6, the integration of SPI consistently enhanced model performance, indicating that mid- to long-term cumulative rainfall signals are beneficial for strengthening the predictive power of teleconnection-based models. Among all SPI types, the most notable improvements were observed in SPI-3 and SPI-6 predictions at the one-month lead time, confirming that SPI effectively complements teleconnection indices in short-term forecasts. However, in three-month lead time predictions, the baseline model occasionally outperformed the auxiliary model for specific catchments and SPI time scales. This may be attributed to the diminishing explanatory power of SPI over longer time horizons or potential interference between SPI and lagged teleconnection signals. Furthermore, the predicted SPI values were converted back into rainfall estimates to evaluate the model’s practical utility in representing actual water resource conditions.

Breiman, L. (2001). Random Forests. Machine Learning, 45, 5–32.
Chen, J.–M., Li, T., & Shih, C.–F. (2008). Asymmetry of the El Nino–Spring Rainfall Relationship in Taiwan. Journal of the Meteorological Society of Japan. Ser. II, 86(2), 297–312.
Chen, J., Li, M., & Wang, W. (2012). Statistical Uncertainty Estimation Using Random Forests and Its Application to Drought Forecast. Mathematical Problems in Engineering, 2012.
Chen, S.–T., Yang, T.–C., Kuo, C.–M., Kuo, C.–H., & Yu, P.–S. (2013). Probabilistic Drought Forecasting in Southern Taiwan Using El Niño-Southern Oscillation Index. Terrestrial, Atmospheric and Oceanic Sciences, 24(5).
Doi, T., Behera, S. K., & Yamagata, T. (2020). Wintertime Impacts of the 2019 Super IOD on East Asia. Geophysical Research Letters, 47(18).
Edwards, D. C., & McKee, T. B. (1997). Characteristics of 20th century drought in the United States at multiple time scales.
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, 141, 3908–3932.
Gao, M., Ge, R., & Wang, Y. (2024). Spring Meteorological Drought over East Asia and Its Associations with Large-Scale Climate Variations. Water, 16(11).
Genuer, R., Poggi, J.–M., & Tuleau-Malot, C. (2010). Variable selection using Random Forests. Pattern Recognition Letters, 31, 2225–2236.
Guttman, N. B. (2007). Accepting the Standardized Precipitation Index: A Calculation Algorithm1. JAWRA Journal of the American Water Resources Association, 35(2), 311–322.
Hayes, M. J., Svoboda, M. D., Wilhite, D. A., & Vanyarkho, O. V. (1999). Monitoring the 1996 drought using the standardized precipitation index. Bulletin of the American Meteorological Society, 80(3), 429-438.
Hsu, L.–H., Wu, Y.–c., Chiang, C.–C., Chu, J.–L., Yu, Y.–C., Wang, A.–H., & Jou, B. J.–D. (2022). Analysis of the Interdecadal and Interannual Variability of Autumn Extreme Rainfall in Taiwan Using a Deep-Learning-Based Weather Typing Approach. Asia-Pacific Journal of Atmospheric Sciences, 59(2), 185–205.
Hung, C. w., Hsu, H. H., & Lu, M. M. (2004). Decadal oscillation of spring rain in northern Taiwan. Geophysical Research Letters, 31(22).
Kumar, V., & Irshada, M. (2023). SMOTE and ExtraTreesRegressor based random forest technique for predicting Australian rainfall. International Journal of Information Technology, 15, 1679–1687.
Kuo, Y.–C., Lee, M.–A., & Lu, M.–M. (2016). Association of Taiwan’s Rainfall Patterns with Large-Scale Oceanic and Atmospheric Phenomena. Advances in Meteorology, 2016, 1–11.
Lee, J. H., Yang, C.–Y., & Julien, P. Y. (2020). Taiwanese rainfall variability associated with large-scale climate phenomena. Advances in Water Resources, 135.
Lee, Y., Kim, H.–R., Noh, N., Kim, K.–Y., & Kim, B.–M. (2023). Enhancing Forecast Skill of Winter Temperature of East Asia Using Teleconnection Patterns Simulated by GloSea5 Seasonal Forecast Model. Atmosphere, 14(3).
Lin, C.–C., Liou, Y.–J., & Huang, S.–J. (2015). Impacts of Two-Type ENSO on Rainfall over Taiwan. Advances in Meteorology, 2015, 1–7.
Markuna, S., Kumar, P., Kushwaha, K., Chaudhary, S., Kuriqi, A., Ali, R., Vishwkarma, D. K., Singh, V., & Kumar, R. (2023). Application of Innovative Machine Learning Techniques for Long-Term Rainfall Prediction. Pure and Applied Geophysics, 180, 335–363.
Pande, C., Elbeltagi, A., Sammen, S., Noor, R., & Costache, R. (2023). Combination of data-driven models and best subset regression for predicting the standardized precipitation index (SPI) at the Upper Godavari Basin in India. Theoretical and Applied Climatology, 152, 535–558.
Park, H.–J., & Ahn, J.–B. (2015). Combined effect of the Arctic Oscillation and the Western Pacific pattern on East Asia winter temperature. Climate Dynamics, 46(9–10), 3205–3221.
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.
Sadeghi, S., Sharafi, S., & Ghaleni, M. (2022). Spatial and temporal analysis of drought in various climates across Iran using the Standardized Precipitation Index (SPI). Arabian Journal of Geosciences, 15.
Stagge, J. H., Tallaksen, L. M., Gudmundsson, L., Van Loon, A. F., & Stahl, K. (2015). Candidate Distributions for Climatological Drought Indices (SPI and SPEI). International Journal of Climatology, 35(13), 4027–4040.
Thomas B. McKee, Nolan J. Doesken, & Kleist, J. (1993). The relationship of drought frequency and duration to time scales.
Wang, B., Wu, R., & FU, X. (2000). Pacific–East Asian Teleconnection: How Does ENSO Affect East Asian Climate.
Wang, J., Pei, L., & Wang, Y. (2023). Precipitation prediction in several Chinese regions using machine learning methods. International Journal of Dynamics and Control, 1–17.
呂季蓉（2006）「臺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
宋嘉文（2003）「氣候變遷對臺灣西半部地區降雨及乾旱影響之研究」，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版資料生產履歷（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）。