Segmentation of XCO2 images with deep learning: application to synthetic plumes from cities and power plants
- 1CEREA, École des Ponts and EDF R&D, Île-de-France, France
- 2Laboratoire des Sciences du Climat et de l’Environnement, LSCE/IPSL, CEA-CNRS-UVSQ, Université Paris-Saclay, 91198 Gif-sur-Yvette, France
- 3Origins.S.A.S, Suez Group, Île-de-France, France
- 4Swiss Federal Laboratories for Materials Science and Technology (Empa), Dübendorf, Switerzland
- 1CEREA, École des Ponts and EDF R&D, Île-de-France, France
- 2Laboratoire des Sciences du Climat et de l’Environnement, LSCE/IPSL, CEA-CNRS-UVSQ, Université Paris-Saclay, 91198 Gif-sur-Yvette, France
- 3Origins.S.A.S, Suez Group, Île-de-France, France
- 4Swiss Federal Laboratories for Materials Science and Technology (Empa), Dübendorf, Switerzland
Abstract. Under the Copernicus programme, an operational CO2 monitoring system (CO2MVS) is being developed and will exploit data from future satellites monitoring the amount of CO2 within the atmosphere. Methods for estimating CO2 emissions from significant local emitters (hotspots, i.e. cities or power plants) can greatly benefit from the availability of such satellite images, displaying atmospheric plumes of CO2. Indeed, local emissions are strongly correlated to the size, shape and concentrations distribution of the corresponding plume, the visible consequence of the emission. The estimation of emissions from a given source can therefore directly benefit from the detection of its associated plumes in the satellite image.
In this study, we address the problem of plume segmentation, i.e. the problem of finding all pixels in an image that constitute a city or power plant plume. This represents a significant challenge, as the signal from CO2 plumes induced by emissions from cities or power plants is inherently difficult to detect since it rarely exceeds values of a few ppm and is perturbed by variable regional CO2 background signals and observation errors. To address this key issue, we investigate the potential of deep learning methods and in particular convolutional neural networks to learn to distinguish plume-specific spatial features from background or instrument features. Specifically, a U-net algorithm, an image-to-image convolutional neural network, with a state-of-the-art encoder, is used to transform an XCO2 field into an image representing the positions of the targeted plume. Our models are trained on hourly 1 km simulated XCO2 fields in the regions of Paris, Berlin and several German power plants. Each field represents the plume of the hotspot, the background consisting of the signal of anthropogenic and biogenic CO2 surface fluxes near or far from the targeted source and the simulated satellite observation errors.
The performance of the deep learning method is thereafter evaluated and compared with a plume segmentation technique based on thresholding in two contexts: the first where the model is trained and tested on data from the same region, and the second where the model is trained and tested in two different regions. In both contexts, our method outperforms the usual segmentation technique based on thresholding and demonstrates its ability to generalise in various cases: city plumes, power plant plumes, and areas with multiple plumes. Although less accurate than in the first context, the ability of the algorithm to extrapolate on new geographical data is conclusive, paving the way to a promising universal segmentation model, trained on a well-chosen sample of power plants and cities, and able to detect the majority of the plumes from all of them. Finally, the highly accurate results for segmentation suggest a significant potential of convolutional neural networks for estimating local emissions from spaceborne imagery.
Joffrey Dumont Le Brazidec et al.
Status: open (until 06 Feb 2023)
-
RC1: 'Comment on gmd-2022-288', Anonymous Referee #1, 09 Jan 2023
reply
This manuscript presents a work of developing a deep-learning-based model for plume segmentation of XCO2 over cities or power plants in European countries. They evaluated the model for model generalization on new data from the same region and model extrapolation on unseen data from another region. The results indicate the proposed segmentation model outperforms the usual segmentation technique based on thresholding.
In general, the presentation of the paper is clear, and the potential of this technique is well-suggested. However, further explanation is needed on how this technique can be applied to estimate emissions from satellite imagery.
Detailed comments:1)In the introduction section,
- The additional reference is needed for that NO2 can be a proxy to CO2 and with NO2, the plume detection capabilities are significantly improving.- Since CO2M is a satellite mission, the author is considering applying this technique; a more detailed explanation of CO2M is needed, such as the spatial resolution, channel information, etc.
2) In the 2.2 section, page 5.
- The data for Paris are selected for Jan., Mar., and Aug. Is there any specific reason to use these three month?- How much has the results performance improved using data augmentation techniques?
The following paper introduced the data augmentation technique for weather applications considering major wind direction. Like this, have you considered the domain characteristics in data augmentation methods?
"Seo, Minseok, et al. "Domain Generalization Strategy to Train Classifiers Robust to Spatial-Temporal Shift." arXiv preprint arXiv:2212.02968 (2022)."3) In the 3.4 section,
- The results showed when the concentration is low or signal-to-noise is small, the performance is significantly degraded. The author mentioned NO2 is helpful for that in the introduction section. Then, why is NO2 data not used as an additional input to solve this problem?- In the deep-learning approaches, the data split is important. Generally, the training and validation dataset are randomly split, while the test is separated from the training and validation. It would be best if you used separate datasets, not days in the middle of the same month used in the training dataset. And please indicate how many datasets are in each training, validation, and test dataset.
4) In the results,
- Most plume smoke shapes are long-tailed, and when the smoke does not spread and gathers in the middle, the segmentation results are not as good as those from long-tailed shapes. There has been a bias towards the plume shape. It seems necessary to analyze whether the result of having a higher wbce score was influenced by the shape of the plume.- How you get the emission amount in the Figure 13.
-
CEC1: 'Comment on gmd-2022-288', Juan Antonio Añel, 15 Jan 2023
reply
Dear authors,Unfortunately, after checking your manuscript, it has come to our attention that it does not comply with our "Code and Data Policy".https://www.geoscientific-model-development.net/policies/code_and_data_policy.htmlYou have archived your code on GitHub. However, GitHub is not a suitable repository. GitHub itself instructs authors to use other long-term archival and publishing alternatives, such as Zenodo. Therefore, please, publish your code in one of the appropriate repositories, and reply to this comment with the relevant information (link and DOI) as soon as possible, as it should be available for the Discussions stage.In this way, you must include the modified 'Code and Data Availability' section in a potentially reviewed version of your manuscript, the DOI of the code (and another DOI for the dataset if necessary).Please, be aware that failing to comply promptly with this request could result in rejecting your manuscript for publication.Juan A. AñelGeosci. Model Dev. Exec. Editor
-
AC1: 'Reply on CEC1', Joffrey Dumont Le Brazidec, 16 Jan 2023
reply
Dear Editor,
The github code repository provided in the article has been archived with zenodo: this corresponding zenodo doi has been published in the README.md (https://github.com/cerea-daml/co2-images-seg/blob/main/README.md) of the github repository.
More precisely, here are the doi and zenodo urls to access the archived code:
- doi: 10.5281/zenodo.7371413
- zenodo url: https://zenodo.org/record/7371413
Both can be found on the github page: https://github.com/cerea-daml/co2-images-seg
We should have been clearer from the start and provided the doi directly in the "code and availability" section of the article: sorry for the lack of clarity. We will add this in a corrected version of the paper.
Kind regards,
Joffrey Dumont Le Brazidec-
CEC2: 'Reply on AC1', Juan Antonio Añel, 16 Jan 2023
reply
Dear authors,
Thanks for your reply.
Providing the link to the Zenodo repository in GitHub did not make any sense. The whole point of not accepting repositories such as GitHub is that they are unreliable. GitHub could be closed immediately by the sole decision of Microsoft, and all the material would be lost, and as a consequence, it would be impossible to reach the assets of your manuscript from its text.
Juan A. Añel
Geoesci. Model Dev. Executive Editor
-
AC2: 'Reply on CEC2', Joffrey Dumont Le Brazidec, 16 Jan 2023
reply
Dear editor,
You are right: the Zenodo link will therefore be directly and primarily given in the revised paper.
Kind regards,
Joffrey Dumont Le Brazidec
-
AC2: 'Reply on CEC2', Joffrey Dumont Le Brazidec, 16 Jan 2023
reply
-
CEC2: 'Reply on AC1', Juan Antonio Añel, 16 Jan 2023
reply
-
AC1: 'Reply on CEC1', Joffrey Dumont Le Brazidec, 16 Jan 2023
reply
Joffrey Dumont Le Brazidec et al.
Joffrey Dumont Le Brazidec et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
274 | 87 | 16 | 377 | 3 | 4 |
- HTML: 274
- PDF: 87
- XML: 16
- Total: 377
- BibTeX: 3
- EndNote: 4
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1