Preprints
https://doi.org/10.5194/gmd-2024-60
https://doi.org/10.5194/gmd-2024-60
Submitted as: methods for assessment of models
 | 
14 May 2024
Submitted as: methods for assessment of models |  | 14 May 2024
Status: this preprint is currently under review for the journal GMD.

Explaining neural networks for detection of tropical cyclones and atmospheric rivers in gridded atmospheric simulation data

Tim Radke, Susanne Fuchs, Christian Wilms, Iuliia Polkova, and Marc Rautenhaus

Abstract. Detection of atmospheric features in gridded datasets from numerical simulation models is typically done by means of rule-based algorithms. Recently, also the feasibility of learning feature detection tasks using supervised learning with convolutional neural networks (CNNs) has been demonstrated. This approach corresponds to semantic segmentation tasks widely investigated in computer vision. However, while in recent studies the performance of CNNs was shown to be comparable to human experts, CNNs are largely treated as a “black box”, and it remains unclear whether they learn the features for the correct reasons. Here we build on the recently published “ClimateNet” dataset that contains features of tropical cyclones and atmospheric rivers as detected by human experts. We adapt the explainable artificial intelligence technique “Layer-wise Relevance Propagation” (LRP) to the feature detection task and investigate which input information CNNs with the Context-Guided Network (CG-Net) and U-Net architectures use for feature detection. We find that both CNNs indeed consider plausible patterns in the input fields of atmospheric variables, which helps to build trust in the approach. We also demonstrate application of the approach for finding the most relevant input variables and evaluating detection robustness when changing the input domain. However, LRP in its current form cannot explain shape information used by the CNNs, and care needs to be taken regarding the normalization of input values, as LRP cannot explain the contribution of bias neurons, accounting for inputs close to zero. These shortcomings need to be addressed by future work to obtain a more complete explanation of CNNs for geoscientific feature detection.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.
Tim Radke, Susanne Fuchs, Christian Wilms, Iuliia Polkova, and Marc Rautenhaus

Status: final response (author comments only)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • CEC1: 'Comment on gmd-2024-60', Juan Antonio Añel, 14 Jun 2024
    • AC1: 'Reply on CEC1', Tim Radke, 16 Jul 2024
    • AC2: 'Reply on CEC1', Tim Radke, 08 Nov 2024
  • RC1: 'Comment on gmd-2024-60', Anonymous Referee #1, 27 Aug 2024
  • RC2: 'Comment on gmd-2024-60', Anonymous Referee #2, 30 Aug 2024
  • AC3: 'Response to the comments from the two anonymous referees', Tim Radke, 08 Nov 2024
Tim Radke, Susanne Fuchs, Christian Wilms, Iuliia Polkova, and Marc Rautenhaus
Tim Radke, Susanne Fuchs, Christian Wilms, Iuliia Polkova, and Marc Rautenhaus

Viewed

Total article views: 868 (including HTML, PDF, and XML)
HTML PDF XML Total Supplement BibTeX EndNote
628 210 30 868 43 19 23
  • HTML: 628
  • PDF: 210
  • XML: 30
  • Total: 868
  • Supplement: 43
  • BibTeX: 19
  • EndNote: 23
Views and downloads (calculated since 14 May 2024)
Cumulative views and downloads (calculated since 14 May 2024)

Viewed (geographical distribution)

Total article views: 820 (including HTML, PDF, and XML) Thereof 820 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 09 Dec 2024
Download
Short summary
In our study, we built upon previous work to investigate the patterns artificial intelligence (AI) learns to detect atmospheric features like tropical cyclones (TCs) and atmospheric rivers (ARs). As primary objective, we adopt a method to explain the used AI and investigate the plausibility of learned patterns. We find that plausible patterns are learned for both TCs and ARs. Hence, the chosen method is very useful for gaining confidence in the AI-based detection of atmospheric features.