the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Automating tephra fall building damage assessment using deep learning
Abstract. In the wake of a volcanic eruption, the rapid assessment of building damage is paramount for effective response and recovery planning. Uninhabited aerial vehicles, UAVs, offer a unique opportunity for assessing damage after a volcanic eruption, with the ability to collect on demand imagery safely and rapidly from multiple perspectives at high resolutions. In this work, we established a UAV-appropriate tephra fall building damage state framework and used it to label ~50,000 building bounding boxes around ~2,000 individual buildings in 2,811 optical images optical images collected during surveys conducted after the 2021 eruption of La Soufrière volcano, St Vincent and the Grenadines. We used this labelled data to train convolutional neural networks (CNNs) for: 1) Building localisation (average precision = 0.728); 2) Damage classification into two levels of granularity: No Damage vs Damage (F1 score = 0.809); and Moderate damage vs Major damage, (F1 score = 0.838) (1 is the maximum obtainable for both metrics). The trained models were incorporated into a pipeline along with all of the necessary image processing steps to generate spatial data (a shapefile with damage state attributes) for rapid tephra fall building damage mapping. Our pipeline is expected to perform well across other volcanic islands in the Caribbean where building types are similar, though would benefit from additional testing. Through cross validation, we found that the UAV look angle had a minor effect on the performance of damage classification models, while for the building localisation model, the performance was affected by both the look angle and the size of the buildings in images. These observations were used to develop a set of recommendations for data collection during future UAV tephra fall building damage surveys. This is the first attempt to automate tephra fall building damage assessment solely using post-event data. We expect that incorporating additional training data from future eruptions will further refine our model and improve its applicability worldwide. All trained models and pipeline code can be downloaded from GitHub to facilitate collaboration and development.
- Preprint
(15233 KB) - Metadata XML
-
Supplement
(9394 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on nhess-2024-81', Sebastien Biass, 18 Jun 2024
The comment was uploaded in the form of a supplement: https://nhess.copernicus.org/preprints/nhess-2024-81/nhess-2024-81-RC1-supplement.pdf
-
AC1: 'Reply on RC1', Eleanor Tennant, 02 Aug 2024
The comment was uploaded in the form of a supplement: https://nhess.copernicus.org/preprints/nhess-2024-81/nhess-2024-81-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Eleanor Tennant, 02 Aug 2024
-
RC2: 'Comment on nhess-2024-81', Anonymous Referee #2, 20 Aug 2024
The manuscript “Automating tephra fall building damage assessment using deep learning” by Eleanor Tennant et al. presents a pipeline based on UAV imagery and deep learning to automatically localize buildings and make a classification of the damage caused by tephra fall.
The manuscript is generally well written, logically organized, and adequately illustrated. I recommend it for publication after few minor points are addressed in revision, as described in the following.
- The manuscript is too focused on methodology. I understand that it represents the heart of the work, but the text seems too unbalanced in relation to the results and the chosen volcanic application (the 2021 eruption of the La Soufrière volcano, St Vincent and the Grenadines). I suggest lightening Sections 2 and 3, moving even more details into the supplementary material and simplifying the main text. This would definitely make reading faster and more fluent.
- The location of Table 1 cannot be the Introduction. It provides a performance comparison of several models, including the one described in this manuscript, and should therefore be included in the Discussions. It is not logically correct to introduce F1, mean average precision and accuracy scores before even introducing the model. It is also not immediately clear what "P", "P & P", "C1" and "C2" mean.
- In Figure 1, please include the location of Georgetown.
- The text is full of diagrams and tables. Perhaps it would be more attractive to a wider audience if the authors included some figures on the case study (for example, some images used for the model development, currently in the supplementary material).
Citation: https://doi.org/10.5194/nhess-2024-81-RC2 -
AC2: 'Reply on RC2', Eleanor Tennant, 30 Aug 2024
The comment was uploaded in the form of a supplement: https://nhess.copernicus.org/preprints/nhess-2024-81/nhess-2024-81-AC2-supplement.pdf
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
286 | 105 | 65 | 456 | 32 | 17 | 17 |
- HTML: 286
- PDF: 105
- XML: 65
- Total: 456
- Supplement: 32
- BibTeX: 17
- EndNote: 17
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1