在氣候多變的今日，如何將觀測數據降尺度(Downscaling)是一個重要的議題，原因是大氣環流模式(General circulation model) 所能輸出的解析度大多為150公里以上，導致難以觀測區域的極端氣候或降雨情況，而動力降尺度需要耗費極大的計算資源，同時需要很長的執行時間。本論文的目的及動機，則是希望透過深度學習方法進行觀測雨量數據降尺度，此方法只需要在模型訓練上花費一段時間，之後使用模型便能即時的反應，而獲得降尺度後的結果，用以初步得到降雨的分布狀況。
現有深度學習進行降尺度的研究，大多是使用大氣環流模式或是區域氣候模型(Regional climate model)所預測出的氣候資料。我們則是以實際觀測到的日降雨量資料做分析及研究，論文中將會提及我們是如何將資料前處理及歸一化，並且開發了一個卷積神經網路(Convolutional neural network)，它結合了多尺度共享資訊，以及通道注意力機制，能夠找出關鍵的特徵圖(Feature map)。然而，降尺度的倍數通常很大，單一個模型無法有效處理，因此，我們提出了一個級聯式架構(Cascaded architecture)做訓練，第一個部分學習大致降雨的分布，而第二個部分則是學習細部的降雨資訊。
實驗結果方面，為了能讓讀者看出差異，我們將日累積雨量根據降雨數值畫成了假色圖，並且與雙線性內插法，以及數個經典且出色的超解析度模型做比較。在我們的實驗環境中，4倍降尺度的情形下，我們的方法獲得最低的Mean absolute error (MAE) 及最高的Correlation coefficient，分別為2.074及0.964。在20倍降尺度的情境下，我們的方法依舊在這兩個評估指標中獲得最佳的成果，分別為2.538及0.948；接著，我們分別對中雨、大雨、豪大雨這三種情況，探討模型的生成，預測出的降雨分布相似於實際觀測資料。最後，我們使用MAE 作為評估指標，探討如何得到誤差較低，同時能夠得到精細區域雨量分布的結果。

In today's changing climate, how to downscale observational data is an important issue. The reason is that the resolution of observational data obtained from General Circulation Models is mostly above 150 kilometers, which makes it difficult to observe the extreme climate or rainfall in specific regions. Besides this, dynamic downscaling requires enormous computational resources and a long execution time. The purpose and motivation of this thesis is to downscale observed rainfall data through a deep learning method. This method only requires to spend a period of time on the model training, then the model can respond immediately, and obtain downscaled results. It is useful to get the initial distribution of rainfall.
Most of the existing deep learning downscaling studies use the climate data predicted by General Circulation Models or Regional Climate Models. In this thesis, however, we do analysis and study the actual observed daily rainfall data, which will mention how we pre-process and normalize the data, and develop a convolutional neural network that combines multi-scale shared information and channel attention mechanism to find key feature maps. Unfortunately, the downscaling factor is often too large for a single model to handle effectively, so we propose a cascaded architecture for training, where the first part learns the approximate rainfall distribution, and the second part learns detailed rainfall information.
In terms of experimental results, in order to see the difference, we plot the daily cumulative rainfall as a pseudo-color map according to the rainfall value, and compare it with the bilinear interpolation method and several classic and excellent super-resolution models. In our experimental environment, with 4× downscaling, our method achieves the lowest Mean Absolute Error (MAE) and the highest correlation coefficient of 2.074 and 0.964, respectively. And in the 20× downscaling scenario, our method still attains the best in the above evaluation metrics, with 2.538 and 0.948, respectively. Next, we discuss the generation of the model for the three cases of moderate rain, heavy rain, and extremely heavy rain, and the predicted rainfall distribution is similar to the actual observation data. Finally, we use MAE as an evaluation metric to explore how to obtain results with low error and fine-grained regional rainfall distribution.

[1] C. Dong, C. C. Loy, K. He, and X. Tang,“Learning a deep convolutional network for image super-resolution,”In Proceedings of the European Conference on Computer Vision, Zurich, Switzerland, 2014, pp. 184-199.
[2] K. E. Taylor, R. J. Stouffer, and G. A. Meehl,“An overview of CMIP5 and the experiment design,”Bulletin of the American Meteorological Society, vol. 93, no. 4, pp. 485-498, 2012.
[3] R. Knutti and J. Sedláček,“Robustness and uncertainties in the new CMIP5 climate model projections,”Nature Climate Change, vol. 3, no. 4, pp. 369-373, 2013.
[4] C. Fu et al.,“Regional climate model intercomparison project for Asia,” Bulletin of the American Meteorological Society, vol. 86, no. 2, pp. 257-266, 2005.
[5] C. Chotamonsak, E. P. Salathé Jr, J. Kreasuwan, S. Chantara, and K. Siriwitayakorn,“Projected climate change over Southeast Asia simulated using a WRF regional climate model,” Atmospheric Science Letters, vol. 12, no. 2, pp. 213-219, 2011.
[6] T. Vandal, E. Kodra, S. Ganguly, A. Michaelis, R. Nemani, and A. R. Ganguly,“Deepsd: Generating high resolution climate change projections through single image super-resolution,”in Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Nova Scotia, Canada, 2017, pp. 1663-1672.
[7] D. Sharma, A. Das Gupta, and M. Babel,“Spatial disaggregation of bias-corrected GCM precipitation for improved hydrologic simulation: Ping River Basin, Thailand,” Hydrology and Earth System Sciences, vol. 11, no. 4, pp. 1373-1390, 2007.
[8] S. Albawi, T. A. Mohammed, and S. Al-Zawi,“Understanding of a convolutional neural network,”in Proceedings of the 2017 international conference on engineering and technology, Antalya, Turkey, 2017, pp. 1-6.
[9] G. Hu et al.,“When face recognition meets with deep learning: an evaluation of convolutional neural networks for face recognition,”in Proceedings of the IEEE International Conference on Computer Vision Workshops, Santiago, Chile, 2015 ,pp. 142-150.
[10] K. Yan, S. Huang, Y. Song, W. Liu, and N. Fan,“Face recognition based on convolution neural network,” in Proceedings of the 2017 36th Chinese Control Conference, Dalian, China, 2017, pp. 4077-4081.
[11] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi,“You only look once: Unified, real-time object detection,”in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, 2016, pp. 779-788.
[12] H. Hu, J. Gu, Z. Zhang, J. Dai, and Y. Wei,“Relation networks for object detection,”in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, Utah, 2018, pp. 3588-3597.
[13] Z. Q. Zhao et al.,“Object detection with deep learning: A review,”IEEE Transactions on Neural Networks and Learning Systems, vol. 30, no. 11, pp. 3212-3232, 2019.
[14] J. Johnson, A. Alahi, and F. F. Li,“Perceptual losses for real-time style transfer and super-resolution,”in Proceedings of European Conference on Computer Vision, Amsterdam, The Netherlands, 2016, pp. 694-711.
[15] L. A. Gatys, A. S. Ecker, and M. Bethge,“Image style transfer using convolutional neural networks,”in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, 2016, pp. 2414-2423.
[16] X. Huang and S. Belongie,“Arbitrary style transfer in real-time with adaptive instance normalization,”in Proceedings of the IEEE Conference on Computer Vision, Venice, Italy, 2017, pp. 1501-1510.
[17] S. Ren, J. Sun, K. He, and X. Zhang,“Deep residual learning for image recognition,”in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, 2016, pp. 770-778.
[18] J. Kim, J. K. Lee, and K. M. Lee,“Accurate image super-resolution using very deep convolutional networks,”in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, 2016, pp. 1646-1654.
[19] W. S. Lai et al.,“Deep laplacian pyramid networks for fast and accurate super-resolution,”in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Hawaii, 2017, pp. 624-632.
[20] I. Goodfellow et al.,“Generative adversarial nets,”Advances in Neural Information Processing Systems, vol. 27, 2014.
[21] C. Ledig et al.,“Photo-realistic single image super-resolution using a generative adversarial network,”in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Hawaii, 2017, pp. 4681-4690.
[22] B. Lim, S. Son, H. Kim, S. Nah, and K. Mu Lee,“Enhanced deep residual networks for single image super-resolution,”in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, Honolulu, Hawaii, 2017, pp. 136-144.
[23] A. Vaswani et al.,“Attention is all you need,”Advances in Neural Information Processing Systems, vol. 30, 2017.
[24] Y. Zhang et al.,“Image super-resolution using very deep residual channel attention networks,”in Proceedings of the European Conference on Computer Vision, Bavaria, German, 2018, pp. 286-301.
[25] T. Dai, J. Cai, Y. Zhang, S.-T. Xia, and L. Zhang,“Second-order attention network for single image super-resolution,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, ‎California‎, 2019, pp. 11065-11074.
[26] J. Li, F. Fang, K. Mei, and G. Zhang,“Multi-scale residual network for image super-resolution,”in Proceedings of the European Conference on Computer Vision, Bavaria, 2018, German, pp. 517-532.
[27] V. Nair and G. E. Hinton,“Rectified linear units improve restricted boltzmann machines,”in Proceedings of the 27th International Conference on International Conference on Machine Learning, Haifa, Israel, 2010, pp. 807–814.
[28] J. Hu, L. Shen, and G. Sun,“Squeeze-and-excitation networks,”in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, Utah, 2018, pp. 7132-7141.
[29] W. Shi et al.,“Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network,”in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, 2016, pp. 1874-1883.
[30] H. Noh, S. Hong, and B. Han,“Learning deconvolution network for semantic segmentation,”in Proceedings of the IEEE International Conference on Computer Vision, Santiago, Chile, 2015, pp. 1520-1528.
[31] Z. Wang, J. Chen and S. Hoi,“Deep Learning for Image Super-Resolution: A Survey”, IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 43, no. 10, pp. 3365-3387, Oct, 2021.
[32] K. Stengel, A. Glaws, D. Hettinger, and R. N. King,“Adversarial super-resolution of climatological wind and solar data,”in Proceedings of the National Academy of Sciences, vol. 117, no. 29, 2020, pp. 16805-16815.
[33] D. P. Kingma and J. Ba,“Adam: A method for stochastic optimization,”Jan 2017. [Online]. Available: https://arxiv.org/abs/1412.6980.
[34] S. Trzaska and E Schnarr, A Review of Downscaling Methods for Climate Change Projections, African and Latin American Resilience to Climate Change: Burlington, Vermont, Sep 2014, pp 7-12.