Author(s): Zhongcheng Wei; Tong Li; Luming Song

Linked Author(s):

Keywords: Convolutional neural network (CNN); Gate recurrent units (GRU); Microwave link (ML); Rainfall retrieval

Abstract: With the development of 5G and Integrated Sensing and Communication (ISAC) technology, Microwave Links (MLs) have been proven useful for providing accurate rainfall information close to the ground surface. Traditional empirical models are less adequate to address the non-linear relationship between the ML attenuation and rainfall intensity data, thus leading to low accuracy of rainfall retrieval. Therefore, this paper proposes a cascaded hybrid model combining convolutional neural network (CNN) and gate recurrent units (GRU), while incorporating an attention mechanism, named as CNN-GRU-Attention, to learn the correlation between ML attenuation and rainfall intensity data, and then to predict the rainfall intensity by inputting ML attenuation data into the CNN-GRU-Attention model. We compared the experimental results of our model with traditional ITU power-law model. The results demonstrate that the CNN-GRU-Attention model outperforms the power-law model in the consistency between predicted and actual rainfall intensities, our model achieved the correlation coefficient values of 0.97 in April and 0.92 in July, representing improvements of 7.76% and 2.22% respectively compared to the best-performing existing model.

DOI: https://doi.org/10.3850/iahr-hic2483430201-295

Year: 2024