本研究以臺東縣大竹溪流域之土坂一號橋與野溪新興橋為研究對象，探討深度學習模型於山區河川水位預測之適用性與誤差來源。研究蒐集2023年3月至2026年3月之水位與雨量資料，經時間對齊、缺失值處理及小時尺度轉換後，建立監督式學習資料集，並比較基礎LSTM、兩階段RS-LSTM-Transformer模型與MLP在不同時間序列長度下之預測表現。
研究結果顯示，MLP雖然架構相對簡易，但於兩處測站皆展現較佳且穩定之泛化能力。其中在168/336/504/720四種固定長度中，土坂一號橋於720小時時間序列條件下表現最佳，NSE為0.7304；野溪新興橋則於504小時條件下表現最佳，NSE為0.8007。相較之下，基礎LSTM整體表現較不穩定；兩階段RS-LSTM-Transformer模型雖於訓練階段具有較佳擬合能力，但測試階段泛化能力不足，顯示在資料量與極端事件樣本有限時，複雜模型可能產生過度擬合問題。
此外，最佳時間序列長度與集水區特性具有關聯。土坂一號橋因集水區較大且支流發育較完整，水位變化可能受前期降雨與退水歷程影響，故較長時間序列有助於模型學習；野溪新興橋集水區較小且坡度較陡，降雨—水位反應較直接，因此較短時間序列即可取得較佳成果；在波型相似度分析亦顯示，多數模型可掌握整體水位變化趨勢，但於多峰降雨、連續降雨、退水歷程及局部強降雨事件中，仍可能出現時間提前、延遲或波型偏差。
綜合上述而言，本研究證實深度學習模型應用於東部山區河川水位預測具有可行性，其中MLP在本研究資料條件下具較佳穩定性。未來若能整合多雨量站、前期累積雨量及土壤含水量等資料，將有助於提升洪峰時機與高水位事件之預測能力。

This study evaluates the applicability of deep learning models for river water-level prediction in mountainous catchments, using Tuban No. 1 Bridge and Yexi Xinxing Bridge in the Dazhu River Basin, Taitung County, as study sites. Water-level and rainfall data from March 2023 to March 2026 were compiled into hourly supervised-learning datasets after time alignment, missing-value treatment, and hourly-scale conversion. A baseline LSTM, a two-stage RS-LSTM-Transformer, and a multilayer perceptron (MLP) were developed and compared under different input sequence lengths. The MLP showed stable performance at both stations, achieving an NSE of 0.7304 with a 720-hour sequence at Tuban No. 1 Bridge and 0.8007 with a 504-hour sequence at Yexi Xinxing Bridge, whereas the RS-LSTM-Transformer achieved a high goodness of fit during training. The optimal sequence length was associated with catchment characteristics: the larger and more tributary-developed Tuban catchment, which is influenced by antecedent rainfall and recession processes, benefited from longer sequences, whereas the smaller and steeper Yexi sub-catchment showed a more direct rainfall-water-level response. Waveform-similarity analysis indicated that most models captured the overall trend, although timing advances, delays, and waveform deviations remained during multi-peak, continuous, recession, and localized heavy-rainfall events. Overall, this study confirms the feasibility of deep learning for water-level prediction in mountainous rivers of eastern Taiwan and suggests that incorporating multi-station rainfall, antecedent rainfall, and soil-moisture data may further improve flood-peak prediction.

[1] Bai, P., Liu, X., Liang, K., & Liu, C. (2021). Simulating runoff under changing climatic conditions: A comparison of the robustness of hydrologic models and LSTM networks. Journal of Hydrology, 601, 126777.
[2] Bates, P. D., & De Roo, A. P. J. (2000). A simple raster-based model for flood inundation simulation. Journal of Hydrology, 236(1–2), 54–77.
[3] Bahdanau, D., Cho, K., & Bengio, Y. (2014). Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473.
[4] Bengio, Y., Simard, P., & Frasconi, P. (1994). Learning long-term dependencies with gradient descent is difficult. IEEE Transactions on Neural Networks, 5(2), 157–166.
[5] Berndt, D. J., & Clifford, J. (1994). Using dynamic time warping to find patterns in time series. In Proceedings of the AAAI Workshop on Knowledge Discovery in Databases (pp. 359–370).
[6] Beven, K. J. (2012). Rainfall-runoff modelling: The primer (2nd ed.). Wiley.
[7] Box, G. E. P., Jenkins, G. M., Reinsel, G. C., & Ljung, G. M. (2015). Time series analysis: Forecasting and control (5th ed.). Wiley.
[8] Bustami, R., Bessaih, N., Bong, C. H. J., & Suhaili, S. (2007). Artificial neural network for precipitation and water level predictions of Bedup River. IAENG International Journal of Computer Science, 34(2), 228–233.
[9] Castangia, M., Grimaldi, S., Camporeale, C., & Ridolfi, L. (2023). Transformer neural networks for interpretable flood forecasting. Environmental Modelling & Software, 165, 105702.
[10] Cheng, M., Fang, F., Kinouchi, T., Navon, I. M., & Pain, C. C. (2020). Long lead-time daily and monthly streamflow forecasting using machine learning methods. Journal of Hydrology, 590, 125376.
[11] Choi, C., Kim, J., Han, H., Han, D., & Kim, H. S. (2020). Development of water level prediction models using machine learning in wetlands: A case study of Upo Wetland in South Korea. Water, 12(1), 93.
[12] Dawson, C. W., & Wilby, R. L. (2001). Hydrological modelling using artificial neural networks. Progress in Physical Geography, 25(1), 80–108.
[13] DHI. (n.d.). MIKE 11 reference manual.
[14] Di Baldassarre, G., & Montanari, A. (2009). Uncertainty in river discharge observations: A quantitative analysis. Hydrology and Earth System Sciences, 13, 913–921.
[15] Dibike, Y. B., & Solomatine, D. P. (2001). River flow forecasting using artificial neural networks. Physics and Chemistry of the Earth, Part B: Hydrology, Oceans and Atmosphere, 26(1), 1–7.
[16] Domeneghetti, A., Castellarin, A., & Brath, A. (2012). Assessing rating-curve uncertainty and its effects on hydraulic model calibration. Hydrology and Earth System Sciences, 16, 1191–1202.
[17] Dottori, F., Martina, M. L. V., & Todini, E. (2009). A dynamic rating curve approach to indirect discharge measurement. Hydrology and Earth System Sciences, 13, 847–863.
[18] Gao, S., Huang, Y., Zhang, S., Han, J., Wang, G., Zhang, M., & Lin, Q. (2020). Short-term runoff prediction with GRU and LSTM networks without requiring time step optimization during sample generation. Journal of Hydrology, 589, 125188.
[19] Goodarzi, M. R., Poorattar, M. J., Vazirian, M., & Talebi, A. (2024). Evaluation of a weather forecasting model and HEC-HMS for flood forecasting: Case study of Talesh catchment. Applied Water Science, 14, Article 34.
[20] Gupta, H. V., Kling, H., Yilmaz, K. K., & Martinez, G. F. (2009). Decomposition of the mean squared error and NSE performance criteria: Implications for improving hydrological modelling. Journal of Hydrology, 377(1–2), 80–91. doi:10.1016/j.jhydrol.2009.08.003
[21] Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735–1780.
[22] Horritt, M. S., & Bates, P. D. (2002). Evaluation of 1-D and 2-D numerical models for predicting river flood inundation. Journal of Hydrology, 268(1–4), 87–99.
[23] Kim, S., Tachikawa, Y., & Sayama, T. (2022). Effects of input variable selection in artificial neural network models for hydrological prediction. Journal of Japan Society of Civil Engineers, Ser. B1 (Hydraulic Engineering), 78(2), I_145–I_150.
[24] Kratzert, F., Klotz, D., Brenner, C., Schulz, K., & Herrnegger, M. (2018). Rainfall–runoff modelling using Long Short-Term Memory (LSTM) networks. Hydrology and Earth System Sciences, 22, 6005–6022.
[25] Li, W., Liu, C., Xu, Y., Niu, C., Li, R., Li, M., Hu, C., & Tian, L. (2024). An interpretable hybrid deep learning model for flood forecasting based on Transformer and LSTM. Journal of Hydrology: Regional Studies, 54, 101873. doi: 10.1016/j.ejrh.2024.101873
[26] Li, X., Sun, Q.-L., Zhang, Y., Sha, J., & Zhang, M. (2024). Enhancing hydrological extremes prediction accuracy: Integrating diverse loss functions in Transformer models. Environmental Modelling & Software, 177, 106042. doi: 10.1016/j.envsoft.2024.106042
[27] Milly, P. C. D., Betancourt, J., Falkenmark, M., Hirsch, R. M., Kundzewicz, Z. W., Lettenmaier, D. P., & Stouffer, R. J. (2008). Stationarity is dead: Whither water management? Science, 319(5863), 573–574.
[28] Pearson, K. (1895). Notes on regression and inheritance in the case of two parents. Proceedings of the Royal Society of London, 58, 240–242.
[29] Qian, X., Wang, B., Chen, J., Fan, Y., Mo, R., Xu, C., Liu, W., Liu, J., & Zhong, P.-A. (2025). An explainable ensemble deep learning model for long-term streamflow forecasting under multiple uncertainties. Journal of Hydrology, 662(Part B), 133968. doi: 10.1016/j.jhydrol.2025.133968
[30] Rajurkar, M. P., Kothyari, U. C., & Chaube, U. C. (2002). Artificial neural networks for daily rainfall–runoff modelling. Hydrological Sciences Journal, 47(6), 865–877.
[31] Rinderer, M., McGlynn, B. L., & van Meerveld, H. J. (2017). Groundwater similarity across a watershed derived from time‐warped and flow‐corrected time series. Water Resources Research, 53, 3921–3940. doi:10.1002/2016WR019856
[32] Sakoe, H., & Chiba, S. (1978). Dynamic programming algorithm optimization for spoken word recognition. IEEE Transactions on Acoustics, Speech, and Signal Processing, 26(1), 43–49. doi:10.1109/TASSP.1978.1163055
[33] Salas, J. D. (1980). Applied modeling of hydrologic time series. Water Resources Publications.
[34] Salas, J. D. (1982). ARMA model identification of hydrologic time series. Water Resources Research, 18(4), 1011–1021.
[35] Shen, C. (2018). A transdisciplinary review of deep learning research and its relevance for water resources scientists. Water Resources Research, 54(11), 8558–8593.
[36] Snieder, E., Shakir, R., & Khan, U. T. (2020). A comprehensive comparison of four input variable selection methods for artificial neural network flow forecasting models. Journal of Hydrology, 583, 124299.
[37] Solomatine, D. P., & Ostfeld, A. (2008). Data-driven modelling: Some past experiences and new approaches. Journal of Hydroinformatics, 10(1), 3–22. doi: 10.2166/hydro.2008.015
[38] Talei, A., & Chua, L. H. C. (2012). Influence of lag time on event-based rainfall–runoff modeling using the data driven approach. Journal of Hydrology, 438–439, 223–233. doi:10.1016/j.jhydrol.2012.03.027
[39] U.S. Army Corps of Engineers. (n.d.). HEC-RAS hydraulic reference manual.
[40] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30.
[41] Wang, Q., Guo, Y., Yu, L., & Li, P. (2020). Earthquake prediction based on spatio-temporal data mining: An LSTM network approach. IEEE Transactions on Emerging Topics in Computing, 8(1), 148–158.
[42] Wei, X., Wang, G., Schmalz, B., Hagan, D. F. T., & Duan, Z. (2023). Evaluation of Transformer model and Self-Attention mechanism in the Yangtze River basin runoff prediction. Journal of Hydrology: Regional Studies, 47, 101438. doi: 10.1016/j.ejrh.2023.101438
[43] You, X., Jiang, P., & Wang, J. (2023). A study on loss function against data imbalance in deep learning-based correction of heavy precipitation forecasts. Atmospheric Research, 284, 106597.
[44] 交通部中央氣象局。（2021）。因應聯合國最新氣候變遷報告之科學說明與回應。交通部中央氣象局。
[45] 交通部中央氣象署。（無日期）。全球暖化與氣候變遷。交通部中央氣象署。
[46] 交通部中央氣象署。（2024）。氣候變遷下，臺灣的極端降水將如何變化？氣候服務入口網。
[47] 臺東縣政府。（2025）。臺東縣管河川界點圖冊。臺東縣政府。