GAN–argcPredNet v1.0: a generative adversarial model for radar echo extrapolation based on convolutional recurrent units

Zheng, Kun; Liu, Yan; Zhang, Jinbiao; Luo, Cong; Tang, Siyu; Ruan, Huihua; Tan, Qiya; Yi, Yunlei; Ran, Xiutao

doi:https://doi.org/10.5194/gmd-15-1467-2022

Articles | Volume 15, issue 4

https://doi.org/10.5194/gmd-15-1467-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/gmd-15-1467-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 15, issue 4

Model description paper

|

18 Feb 2022

Model description paper |

| 18 Feb 2022

GAN–argcPredNet v1.0: a generative adversarial model for radar echo extrapolation based on convolutional recurrent units

Kun Zheng, Yan Liu, Jinbiao Zhang, Cong Luo, Siyu Tang, Huihua Ruan, Qiya Tan, Yunlei Yi, and Xiutao Ran

Download

Final revised paper (published on 18 Feb 2022)
Preprint (discussion started on 06 Jul 2021)

Interactive discussion

Status: closed

CEC1:
'Comment on gmd-2021-165', Astrid Kerkweg, 14 Jul 2021
Dear authors,
in my role as Executive editor of GMD, I would like to bring to your attention our Editorial version 1.2: https://www.geosci-model-dev.net/12/2215/2019/
This highlights some requirements of papers published in GMD, which is also available on the GMD website in the ‘Manuscript Types’ section: http://www.geoscientific-model-development.net/submission/manuscript_types.html
In particular, please note that for your paper, the following requirement has not been met in the Discussions paper:
"The main paper must give the model name and version number (or other unique identifier) in the title."

Please add the software acronym (GAN-arcPredNet) and the version number v1.0 as mentioned in the code availability section in the title upon your revised submission to GMD.
Yours, Astrid Kerkweg
Citation: https://doi.org/10.5194/gmd-2021-165-CEC1
- AC1: 'Reply on CEC1', Yan Liu, 15 Jul 2021
  
  Dear Astrid Kerkweg ,
  We will modify the title of the paper and add the model name and version number later.
  The title of the paper will be changed to GAN-argcPredNet v1.0: A Generative Adversarial Model for Radar Echo Extrapolation Based on Convolutional Recurrent Units.
  The modification of the description of the code availability section has been completed at the time of the last submission.
  Yours, Yan Liu
  
  Citation: https://doi.org/10.5194/gmd-2021-165-AC1
RC1:
'Comment on gmd-2021-165', Anonymous Referee #1, 06 Aug 2021

In this paper, a new GAN based sequence prediction model is developed for precipitation nowcasting. Experimental results are reported to demonstrate the effectiveness of the proposed method. I have the following suggestions:

(1) There are many deep learning extrapolation methods developed for precipitation nowcasting, for example, ConvGRU, TrajGRU, PredRNN, etc. The authors are suggested to compare with some of these methods.

(2) Some important references are missing, for example

[a] Chuyao Luo, Xutao Li, Yongliang Wen, Yunming Ye, Xiaofeng Zhang:

A Novel LSTM Model with Interaction Dual Attention for Radar Echo Extrapolation. Remote. Sens. 13(2): 164 (2021)

[b] Xie P , X Li, X Ji, et al. An Energy-Based Generative Adversarial Forecaster for Radar Echo Map Extrapolation[J]. IEEE Geoscience and Remote Sensing Letters, 2020, PP(99):1-5.

[c] Chuyao Luo, Xutao Li, Yunming Ye:

PFST-LSTM: A SpatioTemporal LSTM Model With Pseudoflow Prediction for Precipitation Nowcasting. IEEE J. Sel. Top. Appl. Earth Obs. Remote. Sens. 14: 843-857 (2021)

Citation: https://doi.org/10.5194/gmd-2021-165-RC1
- AC2: 'Reply on RC1', Yan Liu, 12 Aug 2021
  
  1. There are many deep learning extrapolation methods developed for precipitation nowcasting,for example, ConvGRU, TrajGRU, PredRNN, etc. The authors are suggested to compare with some of these methods.
  Thank you for your suggestions. One of the key points of this paper is to solve the problem of predictive image blur. Previous studies have shown the advantages of generative countermeasure network (GAN) in this problem. Your suggestion is great, we will add relevant experiments in the revision later.
  2. Some important references are missing.
  Thank you for your reminder. The missing references you mentioned will be added when we submit the revision later.
  
  Citation: https://doi.org/10.5194/gmd-2021-165-AC2
CC1:
'Comment on gmd-2021-165', Yuyao Ci, 03 Nov 2021

This article mainly introduces a new type of generative adversarial network, which can be used for radar echo extrapolation, and the effect is obvious. I have the following suggestions:

1. The content of the related work part is a bit lacking, and some related research work can be supplemented.

2. The summary of the future outlook is a bit vague and can be written more clearly.

Citation: https://doi.org/10.5194/gmd-2021-165-CC1
- AC3: 'Reply on CC1', Yan Liu, 04 Nov 2021
  
  Thank you for your valuable suggestions. We will supplement the relevant work after the public discussion, and clarify our future prospects.
  
  Citation: https://doi.org/10.5194/gmd-2021-165-AC3
EC1:
'Comment on gmd-2021-165', David Topping, 23 Nov 2021

Dear authors. Than you for your submission. Clearly this is a timely area of work and the proposed developments fit within the broad readership of the journal. I request you respond to the following minor comments before publication.
General comments:
The multi model image problem is really a core component. Is it possible to provide a visualisation of this, or schematic, for the readers in the introduction?
How did you arrive at the number of convolutional layers? Did you go through a period of architecture optmisation? e.g. using hypertune?
What is the role of data gumentation, if any, on the model performance? Is it possible to use synthetic data?
The formatting of the equations seems to be wrong in places. Please check this. e.g. in equation 1, the subscripts under argmax should be smaller than the main variables.
Minor comments:
Abstract - 'The model composed..' should read the 'model is composed of the argcPredNet..' . In the proceeding line 'In generator, a gate control data memory and output are designed in the rgcLSTM prediction unit of the generator, thereby reducing the loss of spatiotemporal information' should read (i guess), 'In the generator, a gate controlling the memory and out are designed in the rgcLSTM component..'??
There are multiple instance of e.g. 'In discriminator..' please change these to 'In the..' and check throughout the document.

Page 2, line 42: 'which prefers to model unimodal distribution.' should read 'which is better suited to a unimodel distribution'?
50 2 Relate work - this should be 'Related work'.
Page , line 57: please define GRU for the readers.
Page 3 '3.1 Model overview' Please change this to include the name of your model. Please also provide a few sentences telling the reader what each following section is describing.
Page 3 line 86. 'WGAN-gp' please define what this is. expand the acronym for each first instance. Do bare in mind that readers should not need to know implicitly what a model acronym means.
Page 3 line 91: 'Finally, use Adam to optimize the adversarial loss and then update the parameters of the

discriminator, optimize the loss function of the' should read, Finally, we use the Adam optimiser for training..'
I think your paper would really benefit from a table listing all model components that you bring together and a general alrogithm workflow.
Please add clarify in the figure 1 caption this is the new model [and name it]

Citation: https://doi.org/10.5194/gmd-2021-165-EC1
- AC4: 'Reply on EC1', Yan Liu, 24 Dec 2021
  
  The comment was uploaded in the form of a supplement: https://gmd.copernicus.org/preprints/gmd-2021-165/gmd-2021-165-AC4-supplement.pdf
  
  Citation: https://doi.org/10.5194/gmd-2021-165-AC4

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

AR by Yan Liu on behalf of the Authors (25 Dec 2021) Author's response Author's tracked changes Manuscript

ED: Publish subject to technical corrections (13 Jan 2022) by David Topping

AR by Yan Liu on behalf of the Authors (17 Jan 2022) Manuscript

Short summary

In extrapolation methods, there is a phenomenon that causes the extrapolated image to be blurred and unrealistic. The paper proposes the GAN–argcPredNet v1.0 network model, which aims to solve this problem through GAN's ability to strengthen the characteristics of multi-modal data modeling. GAN–argcPredNet v1.0 has achieved excellent results. Our model can reduce the prediction loss in a small-scale space so that the prediction results have more detailed features.