Key factors for quantitative precipitation nowcasting using ground weather radar data based on deep learning

Han, Daehyeon; Im, Jungho; Shin, Yeji; Lee, Juhyun

doi:https://doi.org/10.5194/gmd-16-5895-2023

Articles | Volume 16, issue 20

https://doi.org/10.5194/gmd-16-5895-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/gmd-16-5895-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 16, issue 20

Model evaluation paper

|

20 Oct 2023

Model evaluation paper |

| 20 Oct 2023

Key factors for quantitative precipitation nowcasting using ground weather radar data based on deep learning

Daehyeon Han, Jungho Im, Yeji Shin, and Juhyun Lee

Data sets

Supplementary code and data: Key factors for quantitative precipitation nowcasting using ground weather radar data based on deep learning Daehyeon Han https://doi.org/10.5281/zenodo.8353423

RainNet: a convolutional neural network for radar-based precipitation nowcasting G. Ayzel https://github.com/hydrogo/rainnet

Video-Prediction-using-PyTorch A. H. Nielsen https://github.com/holmdk/Video-Prediction-using-PyTorch/tree/master

Model code and software

Supplementary code and data: Key factors for quantitative precipitation nowcasting using ground weather radar data based on deep learning Daehyeon Han https://doi.org/10.5281/zenodo.8353423

Short summary

To identify the key factors affecting quantitative precipitation nowcasting (QPN) using deep learning (DL), we carried out a comprehensive evaluation and analysis. We compared four key factors: DL model, length of the input sequence, loss function, and ensemble approach. Generally, U-Net outperformed ConvLSTM. Loss function and ensemble showed potential for improving performance when they synergized well. The length of the input sequence did not significantly affect the results.