HKO pioneered development of deep learning in precipitation nowcasting under the collaboration with the Hong Kong University of Science and Technology (HKUST). Two novel AI/ML models for predicting the evolution of radar images and rainfall based on recurrent neural network, namely convolutional long-short-term-memory (ConvLSTM) and trajectory gated recurrent units (TrajGRU) were developed in 2015 and 2017 respectively. They demonstrated promising performance in forecasting the location and development of precipitation as compared to the operational variational optical flow algorithm. HKO has made available a 7-year radar image dataset called HKO-7 to promote research development of AI/ML in precipitation nowcast, training and benchmarking of deep learning precipitation nowcast algorithms.

In recent years, several new deep learning model frameworks have been developed and implemented for real-time trial in SWIRLS operational environment. They generally produce more realistic details and intensities of radar reflectivity nowcast with increasing forecast lead time. Research developments will continue to enhance AI/ML techniques and their applications in SWIRLS operation and RSMC for Nowcasting products.

2-hour reflectivity nowcasts from 6 deep learning models: ResConvLSTM-GAN (upper left), TrajGRU (upper centre), Earthformer (upper right), IAM4VP (lower left), Fourier (lower centre) and ensemble mean of diffusion nowcast ensemble (lower right). Actual radar reflectivity is shown on right panel.

Reference:

Wong, W. K. (2024): AI Technology in Nowcasting. WMO WMC Beijing Workshop on New Technology and Products, Guangzhou, 12-14 November 2024.

Wong, K. H., Wong, W. K. & Lau, H. W. (2023): A New Deep Learning Nowcast Model of Radar Imagery using Generative Adversarial Network for Operational Rainfall Nowcasting. 40th Conference on Radar Meteorology (online), American Meteorological Society, Minneapolis, USA (Online), 28 August – 1 September 2023 (presentation)

HKO-7 dataset (Information available from: https://github.com/sxjscience/HKO-7)

Shi, X., Gao, Z., Lausen, L., Wang, H., Yeung, D. Y., Wong, W. K., & Woo, W. C. (2017): Deep Learning for Precipitation Nowcasting: A Benchmark and A New Model. In Advances in Neural Information Processing Systems (pp. 5622-5632).

Shi, X., Chen, Z., Wang, H., Yeung, D. Y., Wong, W. K., & Woo, W. C. (2015): Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting. In Advances in Neural Information Processing Systems (pp. 802-810).