午後雷陣雨具備短延時強降雨、發生於局部空間，以及非線性發展過程的特性，需要考量運算成本與網格空間解析度的數值天氣預報模式於午後雷陣雨的預報上面臨挑戰；另一方面，由於軟硬體的成熟與資料完整性的提升，人工智慧演算法於各領域蓬勃發展，因其快速推論、易於新增潛在預報因子，與擅長捕捉資料中非線性特徵等優點，有越來越多的研究嘗試使用人工智慧演算法進行預報模型的建構。
　　仰賴於全球衛星導航系統(Global Navigation Satellite System, GNSS)，生活中由定位衍生的服務相繼推出；於氣象上更可利用衛星訊號反推大氣的水氣含量(Bevis et al., 1992)，衛星訊號由衛星發射至地面接收器時，將通過大氣層，其傳遞過程並非以直線前進的方式進行，因而造成時間上的延遲。延遲量可區分為對流層延遲(Zenith Tropospheric Delay, ZTD)與電離層延遲，其中的ZTD是與水氣含量正向相關的指標，可以提供大氣水氣資訊。本研究使用2010年至2020年之氣象與ZTD資訊，針對臺南地區暖季並且符合大環境為弱綜觀的資料，建構深度學習時間序列模型，用於午後1至3小時之短時預報，並客觀量化GNSS ZTD對於預報模型的優化效果。

This study uses Global Navigation Satellite System (GNSS) Zenith Tropospheric Delay (ZTD) and deep learning time series forecasting model to perform afternoon thunderstorm prediction under warm season and weak synoptic environment over Tainan. The main goals are (1) quantifying the performance gain when using ZTD and (2) deciding the best forms of model input for different tasks.
In terms of single-station classification problems, using ZTD as a predictor, when predicting whether it will rain after 3 hours, the overall AUCs at the mountain, the plain, and the near-coast areas increase around 10.60%, 8.13% and 9.77%, respectively, comparing with solely using CWB dataset. When predicting whether it will rain after 1 hour, the overall AUCs increase 1.96%, 4.89%, and 9.77%. Furthermore, the forecasting model has better performances when considering dataset from multi-station if the target hour is closer to the 3rd hour.
In terms of single-station regression problems, comparing with solely using CWB dataset, the overall RMSEs of predicting hourly rainfall at the mountain area decrease around 3.25%, 1.33% and 5.33% after 1 to 3 hours after using ZTD as a predictor. At the near-coast area, the overall RMSEs decrease 1.82%, 3.34% and 0.09%. Moreover, for the multi-station experiments, the rainfall predicting models for the mountain and the near-coast areas further indicate that they have better performances if excluded the dataset far from these areas.

Bevis, M., Businger, S., Herring, T. A., Rocken, C., Anthes, R. A., & Ware, R. H. (1992). GPS meteorology: Remote sensing of atmospheric water vapor using the global positioning system. Journal of Geophysical Research: Atmospheres, 97(D14), 15787-15801.
Chang, C.-P., & Chen, G. T.-J. (1995). Tropical Circulations Associated with Southwest Monsoon Onset and Westerly Surges over the South China Sea. Monthly Weather Review, 123(11), 3254-3267. Retrieved from https://journals.ametsoc.org/view/journals/mwre/123/11/1520-0493_1995_123_3254_tcawsm_2_0_co_2.xml. doi:10.1175/1520-0493(1995)123<3254:Tcawsm>2.0.Co;2
Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
Kingma, D. P., & Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. nature, 521(7553), 436-444.
Leick, A., Rapoport, L., & Tatarnikov, D. (2015). GPS satellite surveying: John Wiley & Sons.
Lin, P.-F., Chang, P.-L., Jou, B. J.-D., Wilson, J. W., & Roberts, R. D. (2011). Warm Season Afternoon Thunderstorm Characteristics under Weak Synoptic-Scale Forcing over Taiwan Island. Weather and Forecasting, 26(1), 44-60. Retrieved from https://journals.ametsoc.org/view/journals/wefo/26/1/2010waf2222386_1.xml. doi:10.1175/2010waf2222386.1
Mohammed, R., Rawashdeh, J., & Abdullah, M. (2020). Machine learning with oversampling and undersampling techniques: overview study and experimental results. Paper presented at the 2020 11th international conference on information and communication systems (ICICS).
O'Shea, K., & Nash, R. (2015). An introduction to convolutional neural networks. arXiv preprint arXiv:1511.08458.
Saastamoinen, J. (1972). Atmospheric correction for the troposphere and stratosphere in radio ranging satellites. The use of artificial satellites for geodesy, 15, 247-251.
Smola, A. J., & Schölkopf, B. (2004). A tutorial on support vector regression. Statistics and computing, 14(3), 199-222.
Tan, P.-H., Soong, W.-K., Tsao, S.-J., Chen, W.-J., & Chen, I.-H. (2022). Impact of Lidar Data Assimilation on Simulating Afternoon Thunderstorms near Pingtung Airport, Taiwan: A Case Study. Atmosphere, 13(9), 1341. Retrieved from https://www.mdpi.com/2073-4433/13/9/1341.
Tieleman, T., & Hinton, G. (2012). Lecture 6.5-rmsprop: Divide the gradient by a running average of its recent magnitude. COURSERA: Neural networks for machine learning, 4(2), 26-31.
Zadeh, L. A. (1988). Fuzzy logic. Computer, 21(4), 83-93.
Zhang, C. J., Wang, H. Y., Zeng, J., Ma, L. M., & Guan, L. (2020). Tiny‐rainnet: a deep convolutional neural network with bi‐directional long short‐term memory model for short‐term rainfall prediction. Meteorological Applications, 27(5), e1956.
Zhao, Q., Liu, Y., Yao, W., & Yao, Y. (2021). Hourly rainfall forecast model using supervised learning algorithm. IEEE Transactions on Geoscience and Remote Sensing, 60, 1-9.
朱瑞鼎、陳昭銘、張家治. (2017). 臺南地區弱綜觀天氣之降雨分析. 106年天氣分析與預報研討會.
宋偉國、朱宗良. (2014). 台灣南部地區午後對流系統個案研究. 103年天氣分析與預報研討會.
宋偉國、顏罡志. (2021). 應用剖風儀與數值模式於雷暴之研究. 110年天氣分析與預報研討會.
林品芳、張保亮、周仲島. (2012). 弱綜觀環境下臺灣午後熱對流特徵及其客觀預報. 大氣科學, 40 卷, 77-108.
張惠玲, & 林沛練. (1987). 台灣地區午後對流降水之研究. 研究, 8, E1.
陳延昇. (2022). 評估區域性電離層資料同化模式對於精密單點定位之影響. 碩士論文.
陳泰然、周鴻祺、廖佩娟、楊進賢. (2009). 暖季弱綜觀強迫下中北臺灣午後熱對流的氣候特徵. 大氣科學, 37 卷, 69-108.
陳鈞澤、洪炯宗、廖宇慶、劉千義、鍾佳儒、葉大綱. (2021). 弱綜觀午後雷暴事件即時預報-機器學習方法. 110年天氣分析與預報研討會.
黃國維. (2021). 臺灣地區天頂向對流層延遲量預測短期夏季降雨可行性評估. 碩士論文.