the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Sentinel-1 based analysis of the Pakistan Flood in 2022
Florian Roth
Bernhard Bauer-Marschallinger
Mark Edwin Tupas
Christoph Reimer
Peter Salamon
Wolfgang Wagner
Abstract. In August and September 2022, Pakistan was hit by a severe flood and millions of people were impacted. The Sentinel-1 based flood mapping algorithm developed by Technische Universität Wien (TU Wien) for the Copernicus Emergency Management Service (CEMS) global flood monitoring (GFM) component was used to document the propagation of the flood from August 10 to September 23, 2022. The results were evaluated using the flood maps from the CEMS rapid mapping component. Overall, the algorithm performs reasonably with a critical success index of up to 80 %, while the detected differences are traced back to different sensors used for the flood mapping. Over the 6 weeks timespan an area of 30,492 km2 was observed to be flooded at least once, and the maximum extent was found to be present on August 30.
Florian Roth et al.
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2022-1061', Anonymous Referee #1, 25 Jan 2023
I read the manuscript from Roth et al., with great interest. The authors evaluate the potential of the TU Wien flood mapping algorithm live on the GFM portal for the Pakistan Floods of 2022. The paper is topical, well-written and an interesting read. I particularly enjoyed the section on flood evolution and the movement of the inundation with time. I found, however, a few scientific issues that must be clarified prior to publication. As you will see below, these are fairly minor and just require correcting some sentences. Some minor comments are also included in the attached PDF file.
- The authors assert that optical data can look at flooding under vegetation canopies and SAR data cannot. While the latter is partially true, the former is not. Optical sensors are passive and can only see the first feature the sensor encounters, in case of cloud cover, this is the cloud and in case of dense canopies, it is the canopy that dominates the signal and thus the water underneath cannot normally be seen by the sensor. While it is true that the combination of SAR and optical in a classifier enables better decision boundary selection in the spectral domain, as the SAR backscatter typically increases for flooded vegetation, and indeed the addition of optical features helps in differentiating this in the feature space. I have provided some references which can provide a clearer conceptual overview of the topic in the supplementary file. Further, the authors can find details in Section 5.4 (the last paragraph addresses exactly this point) of the following reference:
Huang, Chang, Yun Chen, Shiqiang Zhang, and Jianping Wu. "Detecting, extracting, and monitoring surface water from space using optical sensors: A review." Reviews of Geophysics 56, no. 2 (2018): 333-360. - I found some algorithms used but their authors not cited, e.g. the exclusion layer in GFM. Please add these references in, they are mentioned in the supplementary file.
- The scale was missing from some maps, please add them in.Â
Thanks in advance for considering these corrections and for taking on this impotant topic. I wish the authors all the best for the publication and look forward to seeing the manuscript online!
Â
-
AC1: 'Reply on RC1', Florian Roth, 15 Mar 2023
Dear Reviewer,
Thank you very much for your helpful feedback! Please see our response below. We will submit a revised version of the manuscript once the feedback from all reviewers is available.
Thank you and best regards,
Florian Roth (on behalf of all authors)
Â
Line 28: Thank you for this advice. The risk and recovery product looks very promising for flood mapping validation. Unfortunately, there is currently no data available for the Pakistan flood, which is why we selected the rapid mapping data.
Line 33: We will clarify in the revised version that we are showing the extent of the study sites provided by the CEMS rapid mapping data.
Line 45: The internal low sensitivity masking used by our algorithm is described in detail by the cited publication (Bauer-Marschallinger et al., 2022), and we will add details on the exclusion methods to the manuscript.
Line 48: Thank you for pointing us towards this publication. The GFM exclusion layers fundamentally differ from the described methods/layers of this publication. Following a similar approach of Zhao et al., 2021, we implemented a time-series-based masking per pixel, but the layers are produced individually based on statistical parameters or external datasets. The layers will be part of a future scientific publication, and are already publicly accessible via the product definition document of the Global Flood Monitoring service (cited in the paper under Global Flood Monitoring).
Line 73: The 4th area of interest (Shanghar) is 2 days apart from a Sentinel-1 acquisition, but is currently missing in Table 1. For completion, we will add the area to Table 1, but keep the analysis to be focused on the other three areas to rely on temporally close reference data.
Line 87: Thank you for bringing this up. Since Sentinel-1 observed the situation before Spot, and the flood surface was still growing at this point in time, the underestimation is more likely related to the time difference of the acquisitions. To double-check this and to analyse the relationship with vegetation, we will add land cover data to provide more context.
Figure 4: Thank you for this comment. The validation is limited to areas showing a valid result from both the Sentinel-1 flood mapping and the optical reference data. Consequently, cloud-covered areas are excluded. We will add the information about the coverage of the validation!
Line 89: Thank you for putting this into the correct context and pointing us towards two useful publications (Schumann et al., 2022 and Dasgupta et al., 2022). Optical observations are not able to detect flood under dense vegetation. In the case of Shikarpur, we assumed advantages of optical observations for detecting flood close to sparse vegetation, but we have not fully communicated this in the paper. However, as you mentioned, the differences are more likely to be related to the time difference between Sentinel-1 and Spot (please see comment on line 87).
Line 91: See comment on line 89.
Line 99: Thank you for your advice. We will include a zoomed subset to ease the interpretation of the confusion maps.
Line 100: The areas of overestimation mostly correspond to cropland, which occasionally shows fast backscatter changes due to different agricultural activities. Low-backscatter over vegetation appearing with water-like signature can occur due to ploughing in case of agricultural fields or attenuation of the signal within the vegetation (Vreugdenhil et al., 2018, Harfenmeister et al., 2019). A more speculative explanation is that the flood surface changed between the two satellite observations due to manipulation of local dams in reaction to the flood threat.
Line 101: Thank you for pointing this out. You are correct, the time difference can generally only correspond to an increased or decreased flood surface. Our interpretation of the overestimation can be seen in the comment for line 100.
Figure 7: We will add a scale bar to the figure.
Line 114: We agree!
Line 140: As this large flood is not covered by a single Sentinel-1 overpass, each orbit covers different parts of the event, at different times. To allow an estimate of the progress of the flooded area over time, we decided to base the analysis on a single orbit, which covers about the same area at each overpass.
Citation: https://doi.org/10.5194/egusphere-2022-1061-AC1
- The authors assert that optical data can look at flooding under vegetation canopies and SAR data cannot. While the latter is partially true, the former is not. Optical sensors are passive and can only see the first feature the sensor encounters, in case of cloud cover, this is the cloud and in case of dense canopies, it is the canopy that dominates the signal and thus the water underneath cannot normally be seen by the sensor. While it is true that the combination of SAR and optical in a classifier enables better decision boundary selection in the spectral domain, as the SAR backscatter typically increases for flooded vegetation, and indeed the addition of optical features helps in differentiating this in the feature space. I have provided some references which can provide a clearer conceptual overview of the topic in the supplementary file. Further, the authors can find details in Section 5.4 (the last paragraph addresses exactly this point) of the following reference:
-
CC1: 'Comment on egusphere-2022-1061', Wang Jin, 09 Mar 2023
This paper presents a new sentinel-1 based flood mapping algorithm to record flood propagation in Pakistan from August 10 to September 23, 2022. The algorithm is compared to the results of the CEMS rapid mapping component, and the results suggest that this algorithm is able to provide information about large-scale flood events in NRT. Although this paper presents an interesting and potentially useful idea, there are some issues that need to be addressed for it to be considered for publication.
Firstly, the research background may needs to be supplemented in the Section of Introduction and the shortcomings of previous works may need be analyzed with one or two sentences to extend the method of this paper, explaining the importance for the proposed algorithm, and give an insight into how it differs from existing methods.
Secondly, the comparison between the TU Wien flood mapping algorithm and CEMS fast mapping component should be shown in more detail and clearly. This comparison should demonstrate the advantages of the proposed algorithm, and explain why it is more effective than existing methods.
Thirdly, the results should be explained in more detail, and why they were obtained should be discussed. The paper should provide a clear analysis of the results and explain the implications of the findings.
Fourthly, a specific description of the TU Wien flood mapping algorithm should be added to Section of Methodology. This section should provide a comprehensive overview of the algorithm, including the data sources and tools used.
Finally, the discussion may need to be improved by providing a comprehensive overview of the implications of the results. The paper should also provide recommendations for future research and practical applications.
I believes that if these issues are addressed in the paper, the essential contribution of this paper would be important for flood mapping.
Citation: https://doi.org/10.5194/egusphere-2022-1061-CC1 -
AC2: 'Reply on CC1', Florian Roth, 17 Mar 2023
Dear Dr. Jin,
Thank you for your interest and your comment on our manuscript.
When looking at the first Sentinel-1 scenes covering this outstandingly large flood event happening in Pakistan, we decided on presenting the abilities of our flood mapping algorithm for the event. The main goal was to quickly share our results with the scientific and other interested communities, while providing details on the performance of the algorithm. This was the reason why we went for the pre-print option and made the data publicly available, while the event was still on going. As our algorithm is included in the Copernicus Emergency Management Service's (CEMS) Global Flood Monitoring (GFM) component, it contributed to the data used by DG Echo and other users for emergency response. Additionally, the topic was taken up by EUMETSAT in their article on flood and drought monitoring.
The used algorithm itself is presented in detail in a dedicated methodology paper (Bauer-Marschallinger et al., 2022). For this study, we focused on applying the algorithm on a large-scale event and showing how to describe the impact of this kind of event based on flood mapping results. This is why we do not present the full methodology in this topical paper.
Please let us point you to some relevant links to this topic:
Methodology paper: https://www.mdpi.com/2072-4292/14/15/3673/htm
GFM Product Definition Document: https://extwiki.eodc.eu/GFM/PDD
EUMETSAT article: https://www.eumetsat.int/features/towards-better-flood-and-drought-monitoring
DG Echo map: https://erccportal.jrc.ec.europa.eu/ercmaps/ECDM_20220902_SW_Pakistan.pdfThank you and best regards,
Florian Roth (on behalf of all authors)Citation: https://doi.org/10.5194/egusphere-2022-1061-AC2
-
AC2: 'Reply on CC1', Florian Roth, 17 Mar 2023
-
RC2: 'Comment on egusphere-2022-1061', Anonymous Referee #2, 19 Mar 2023
In the manuscript entitled "Sentinel-1 based analysis of the Pakistan Flood in 2022" authors present an analysis of the flood event in Pakistan in 2022 using the Sentinel-1 based flood mapping algorithm developed by Technische Universität Wien (TU Wien) for the Copernicus Emergency Management Service (CEMS). The results were evaluated using flood maps from the CEMS rapid mapping component, and the overall performance of the algorithm was assessed. The study area, methodology, evaluation, and results are well presented and contribute to the understanding of the 2022 Pakistan flood event.
However, there are a few concerns and recommendations that I believe should be addressed by the authors in order to improve the quality of the manuscript.
In the Introduction, the authors mention that the uneven topography and heavy rainfall during the monsoon season make Pakistan a flood-prone country. While this statement is accurate, it would be beneficial to provide more context on the hydrology, climate, and land use in the region to better understand the factors contributing to flood events in Pakistan. Providing this context would help readers to better appreciate the significance of the study. The evaluation of the algorithm's performance is based on a limited dataset, as the authors acknowledge in the Evaluation section. While the results from the selected areas of interest (AOI) are informative, it is important to emphasize that the evaluation may not be representative of the algorithm's performance on a larger scale. The authors should consider discussing the limitations of the evaluation and potential implications for the algorithm's performance in other regions or flood events.
In the Results and discussion section, the authors mention that the detected underestimation in the Shikarpur AOI could be attributed to differences in the used sensor for flood mapping. It would be helpful if the authors could provide more information on the differences between the sensors used and how these differences might impact the flood mapping results.
The manuscript would benefit from a more detailed discussion of the potential applications of the flood mapping algorithm and the implications of its performance for flood management and disaster response. For example, the authors could discuss how the algorithm could be integrated into existing flood forecasting and early warning systems or used to inform post-disaster recovery efforts.
In the Conclusions section, the authors briefly mention that the study shows the potential of providing information in near-real-time on large-scale flood events. It would be valuable if the authors could expand upon the significance of near-real-time flood mapping for flood management and disaster response, as well as potential future developments and improvements to the algorithm.
Overall, the manuscript presents a valuable analysis of the 2022 Pakistan flood event using the Sentinel-1 based flood mapping algorithm. Addressing the concerns and recommendations outlined above would help to improve the quality of the manuscript and provide a more comprehensive understanding of the algorithm's performance and potential applications.
Â
Citation: https://doi.org/10.5194/egusphere-2022-1061-RC2
Florian Roth et al.
Data sets
Sentinel-1 based analysis of the Pakistan Flood in 2022 Florian Roth, Bernhard Bauer-Marschallinger, Mark Edwin Tupas, Christoph Reimer, Peter Salamon, Wolfgang Wagner https://doi.org/10.48436/zvvmh-nan78
Florian Roth et al.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 438 | 284 | 16 | 738 | 8 | 6 |
- HTML: 438
- PDF: 284
- XML: 16
- Total: 738
- BibTeX: 8
- EndNote: 6
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1