Performance comparison of short-term radar rainfall prediction models based on U-Net and U-Net++ structures

Min-Do Jeong; Kwang-Seok Yoon; Hyun-Han Kwon

doi:10.3741/JKWRA.2024.57.11.955

All Issue

2024 Vol.57, Issue 11 Preview Page Next Page

Research Article

Performance comparison of short-term radar rainfall prediction models based on U-Net and U-Net++ structures U-Net 및 U-Net++ 구조 기반의 초단기 레이더 강우예측 모델 성능 비교

30 November 2024. pp. 955-968

PDF XML

Abstract

This study developed a rainfall radar nowcasting model using the deep learning based U-Net++ model and quantitatively compared its performance with the U-Net model. Performance evaluation of rainfall prediction was performed by the lead time according to the rainfall threshold using CSI(Critical success index), RMSE(Root Mean Square Error). As a result of the study, it was confirmed that the overall RMSE of the U-Net++ model was lower than that of the U-Net model, and thed quantitative prediction performance of rainfall was improved. On the other hand, in the case of CSI, the existing U-Net model showed a higher value, showing excellent performance in predicting the occurrence of rainfall. However, both models showed a tendency to decrease CSI and increase RMSE as the threshold increased and the lead time increased. In addition, the areal rainfall in the Soyang river basin was compared using ground rainfall, radar data, U-Net model, and U-Net++ model. As a result of calculating cumulative areal rainfall every 10 minutes for a total of 8 rainfall events, the U-Net++ model tended to be slightly overestimated as the lead time increased, but it was shown to be superior to the U-Net model in the cumulative areal rainfall prediction performance up to 60 minutes. In particular, the U-Net++ model showed high similarity to ground rainfall and rainfall and radar data in events with relatively high rainfall, and the RMSE analysis also showed lower median and less variability, confirming excellent prediction performance.

Keywords

U-Net++

Deep learning

Radar

Rainfall forecast

본 연구에서는 딥러닝 기반의 U-Net++ 모형을 활용하여 강우 레이더 예측 모델을 개발하고, 기존의 U-Net 모형과의 성능을 정량적으로 비교하였다. 강우 예측의 성능 평가는 CSI(Critical Success Index)와 RMSE(Root Mean Square Error)를 이용하여 강우 임계치에 따라 선행시간별로 수행되었다. 연구 결과, U-Net++ 모형은 U-Net 모형에 비해 RMSE가 전반적으로 낮아, 양적 예측 성능이 향상되었음을 확인하였다. 반면, CSI의 경우 기존 U-Net 모형이 더 높은 값을 보여 강우 발생 여부 예측에 있어 우수한 성능을 나타냈다. 그러나 임계치가 커지고 선행시간이 증가함에 따라 두 모델 모두 CSI가 감소하고 RMSE가 증가하는 경향을 보였다. 또한, 소양강 유역의 유역 면적 강수량을 지상 강우, 강우 레이더, U-Net++ 모형, 그리고 U-Net 모형을 활용하여 비교하였다. 총 8개의 강우사상을 대상으로 10분 간격으로 누적 유역 면적 강수량을 산정한 결과, U-Net++ 모형은 선행시간이 길어질수록 다소 과대 추정되는 경향이 있었지만, 60분까지의 누적 유역 면적 강수량 예측 성능에서는 U-Net 모형보다 우수함을 보였다. 특히, 강우량이 상대적으로 많은 사상에서 U-Net++ 모형은 지상 강우 및 강우 레이더와의 높은 유사도를 나타내었으며, RMSE 분석에서도 더 낮은 중앙값과 적은 변동성을 보여 예측 성능이 우수함을 확인하였다.

키워드

U-Net++

딥러닝

레이더

강우 예측

References

Ayzel, G., Scheffer, T., and Heistermann, M. (2020). "RainNet v1.0: A convolutional neural network for radar-based precipitation nowcasting." Geoscientific Model Development, Vol. 13, No. 6, pp. 2631-2644.

10.5194/gmd-13-2631-2020

Bellon, A. , and Austin, G.L. (1978). "The evaluation of two years of real-time operation of a short-term precipitation forecasting procedure (SHARP)." Journal of Applied Meteorology and Climatology, Vol. 17, No. 12, pp. 1778-1787.

10.1175/1520-0450(1978)017<1778:TEOTYO>2.0.CO;2

Collier, C.G. (2007). "Flash flood forecasting: What are the limits of predictability?" Quarterly Journal of the Royal Meteorological Society, Vol. 133, No. 622, pp. 3-23.

10.1002/qj.29

Han, D., Choo, M., Im, J., Shin, Y., Lee, J., and Jung, S. (2023). "Precipitation nowcasting using ground radar data and simpler yet better video prediction deep learning." GIScience & Remote Sensing, Vol. 60, No. 1, 2203363.

10.1080/15481603.2023.2203363

Jeon, B.K., Lee, C.K., and Kim, Y.S. (2012). "Evaluation of rainfall measurement capability of dual polarization radar." Journal of the Korean Society of Hazard Mitigation, Vol. 12, No. 2, pp. 215-224.

10.9798/KOSHAM.2012.12.2.215

Kang, M., Ku, J.M., and Yoo, C. (2016). "Usefulness of precipitable water for improving the accuracy of rainfall occurrence decisions using radar data." Journal of the Korean Society of Hazard Mitigation, Vol. 16, No. 1, pp. 37-50.

10.9798/KOSHAM.2016.16.1.37

Keys, R. (1981). "Cubic convolution interpolation for digital image processing." IEEE Transactions on Acoustics, Speech, and Signal Processing, Vol. 29, No. 6, pp. 1153-1160.

10.1109/TASSP.1981.1163711

Kim, H.S., Kim, H.S., Jeon, G.I., and Gang, S.U. (2016). "Assessment of 2014~2015 drought." Water for Future, Vol. 49, No. 7, pp. 61-75.

Kim, J., Yoon, S., Kim, G., and Moon, Y. (2015). "Application of urban runoff simulation using TREC technique for short-term radar rainfall prediction." Journal of Korea Water Resources Association, Vol. 48, No. 5, pp. 409-423.

Kim, T.J., Lee, D.R., and Kwon, H.H. (2015). "Development of a weather radar rainfall field tracking method for short-term rainfall prediction." Journal of Korea Water Resources Association, Vol. 48, No. 12, pp. 995-1009.

10.3741/JKWRA.2015.48.12.995

Krizhevsky, A., Sutskever, I., and Hinton, G.E. (2012). "ImageNet classification with deep convolutional neural networks." Advances in Neural Information Processing Systems, Vol. 25, pp. 1097-1105.

Laroche, S., and Zawadzki, I. (1994). "A variational analysis method for retrieval of three-dimensional wind field from single-Doppler radar data." Journal of Atmospheric Sciences, Vol. 51, No. 18, pp. 2664-2682.

10.1175/1520-0469(1994)051<2664:AVAMFR>2.0.CO;2

Ravuri, S., Lenc, K., Willson, M., Kangin, D., Lam, R., Mirowski, P., Fitzsimons, M., Athanassiadou, M., Kashem, S., and Madge, S. et al. (2021). "Skilful precipitation nowcasting using deep generative models of radar." Nature, Vol. 597, No. 7878, pp. 672-677.

10.1038/s41586-021-03854-z34588668PMC8481123

Ronneberger, O., Fischer, P., and Brox, T. (2015). "U-Net: convolutional networks for biomedical image segmentation." ArXiv, abs/1505.04597.

10.1007/978-3-319-24574-4_28

Shi, X., Chen, Z., Wang, H., Yeung, D.-Y., Wong, W.-k., and Woo, W.-c. (2015). "Convolutional LSTM Network: a machine learning approach for precipitation nowcasting." Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 1, Montreal, Canada.

Sun, J., Xue, M., Wilson, J.W., Zawadzki, I., Ballard, S.P., Onvlee Hooimeyer, J., Joe, P., Barker, D.M., Li, P.-W., Golding, B., Xu, M., and Pinto, J. (2014). "Use of NWP for nowcasting convective precipitation: Recent progress and challenges." Bulletin of the American Meteorological Society, Vol. 95, pp. 409-426.

10.1175/BAMS-D-11-00263.1

Tan, Y., Zhang, T., Li, L., and Li, J. (2024). "Radar-based precipitation nowcasting based on improved U-Net model." Remote Sensing, Vol. 16, No. 10, 1681.

10.3390/rs16101681

Yoon, S., Park, H., and Shin, H. (2020). "Very short-term rainfall prediction based on radar image learning using deep neural network." Journal of Korea Water Resources Association, Vol. 53, No. 12, pp. 1159-1172.

Yoon, S., Shin, H., and Heo, J. (2023). "Radar-based short-term rainfall prediction using generative adversarial networks." Journal of Korea Water Resources Association, Vol. 56, No. 8, pp. 471-484.

Zhou, Z., Siddiquee, M.M.R., Tajbakhsh, N., and Liang, J. (2018). "UNet++: A nested U-Net architecture for medical image segmentation." Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, Vol. 11045, pp. 3-11.

10.1007/978-3-030-00889-5_132613207PMC7329239

Information

Publisher :KOREA WATER RESOURECES ASSOCIATION
Publisher(Ko) :한국수자원학회
Journal Title :Journal of Korea Water Resources Association
Journal Title(Ko) :한국수자원학회 논문집
Volume : 57
No :11
Pages :955-968
Received Date : 2024-10-02
Revised Date : 2024-11-11
Accepted Date : 2024-11-11
DOI :https://doi.org/10.3741/JKWRA.2024.57.11.955

Journal of Korea Water Resources Association ISSN:2799-8746(Print) 2799-8754(Online) 한국수자원학회 논문집

All Issue