the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Enhanced Land Subsidence Interpolation through a Hybrid Deep Convolutional Neural Network and InSAR Time Series
Abstract. Land subsidence, the gradual or sudden sinking of the land, poses a global threat to infrastructure and the environment. This paper introduced a hybrid method based on deep convolutional neural networks (CNN) and persistent scattered interferometric synthetic aperture radar (PSInSAR) to estimate land subsidence in areas where PSInSAR cannot provide reliable measurements. This approach involves training a deep CNN with subsidence driving forces and PSInSAR data to learn patterns and estimate subsidence values. Our evaluation of the model shows its efficiency in overcoming the discontinuities observed in the PSInSAR results, producing a continuous subsidence surface. The deep CNN was evaluated on training, validation, and testing data, resulting in mean squared errors of 5 mm, 9 mm, and 11 mm, respectively. In contrast, the kriging interpolation method showed a mean square error of 37.19 mm in the experimental data set. subsidence prediction using the deep CNN method showed a 70 % improvement compared to the Kriging interpolation method.
- Preprint
(2358 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 30 Jul 2024)
-
RC1: 'Comment on gmd-2024-15', Anonymous Referee #1, 21 Jun 2024
reply
General Comments:
The paper explores the innovative use of deep convolutional neural networks (CNN) combined with PSInSAR time series for land subsidence interpolation. The methodology is scientifically sound, and the results show significant practical value. However, the paper needs improvements in data description, methodological details, result presentation, and depth of discussion. Addressing these aspects will enhance the paper's scientific rigour, transparency, and applicability.
Detailed Comments:
- Clarity of Abstract Evaluation
The abstract's description of the model's evaluation, stating that "Our evaluation of the model demonstrates its proficiency in addressing the discontinuities evident in PSInSAR results, resulting in a continuous subsidence surface," is insufficiently precise. It is advisable to incorporate specific evaluation metrics and outcomes to better understand the model's performance.
- Insufficient Comprehensive Review in the Introduction
The introduction section ought to encompass a broader literature review focusing on recent advancements in the field of land subsidence, particularly studies that incorporate PSInSAR technology and deep learning techniques. This comprehensive review would better contextualize the research background and underscore its significance.
- Insufficient Description of the Dataset
The paper should furnish comprehensive details regarding the Sentinel-1A images employed in the study. This includes specifying the acquisition intervals, spatial resolution, and other pertinent parameters. Additionally, a thorough description of the GNSS (Global Navigation Satellite System) and groundwater monitoring data, including their source, frequency of collection, and processing methods, is essential to enhance the transparency and reproducibility of the research.
- Methodology Section Lacks Detail
The methodology section requires further elaboration, particularly about the training process of the CNN model. Providing details such as the rationale behind selecting specific loss functions, the hyperparameter tuning process, and the specific data preprocessing steps employed will greatly aid readers in understanding the implementation process. Clarifying these aspects will strengthen the credibility and reproducibility of the methodology.
- Lack of Performance Comparison Analysis
While the paper mentions that the CNN model outperforms the Kriging interpolation method, a more in-depth performance comparison analysis is warranted. Incorporating additional performance metrics, such as the coefficient of determination (R2), mean absolute error (MAE), and root mean square error (RMSE), would provide a more comprehensive evaluation of the models. Furthermore, conducting statistical significance tests, such as t-tests or ANOVA, would strengthen the conclusions regarding the superiority of the CNN model. This enhanced analysis will lend greater credence to the findings and facilitate their interpretation.
- Insufficient Explanation of the Selection of Driving Factors
The paper should provide a more thorough explanation of why the selected driving factors (e.g., land use/cover changes, geology, aquifer characteristics, water management, etc.) are significant in predicting subsidence. This explanation should be supported by relevant literature to demonstrate that these factors have been identified as important predictors in previous studies. Clarifying the rationale behind the choice of these factors will strengthen the credibility of the model.
- Lack of Visualization in the Results Section
Visualizations are crucial in presenting the results in an intuitive and comprehensible manner. The results section should include subsidence prediction maps generated by the CNN model and the Kriging interpolation method, as well as error distribution maps that show the spatial distribution of prediction errors. These visualizations will help readers better understand the performance of the different methods and identify areas of high or low prediction errors.
- Discussion Section Not In-Depth
The discussion section should expand on the advantages and limitations of the proposed CNN model. It should discuss how the model performs in different scenarios, such as urban and rural areas, and identify potential factors that may affect its accuracy. Additionally, the section should explore potential improvements to the model, such as incorporating additional data sources or employing more advanced deep learning architectures. Furthermore, it should discuss the broader implications of the findings and suggest future research directions.
- Literature References Are Not Comprehensive
To ensure the paper is situated within the current research landscape, it should cite more recent studies published in the field of subsidence prediction using remote sensing and deep learning. These references should cover both methodological advances and case studies relevant to the current research. Incorporating a more comprehensive literature review will strengthen the paper's theoretical foundation and position it within the broader research context.
- Lack of In-Depth Discussion on Traditional Methods Comparison
The paper mentions the Kriging interpolation method but lacks a thorough comparative analysis with other traditional interpolation methods. Conducting a comparison with methods such as inverse distance weighting (IDW), radial basis functions (RBF), and others would provide a more comprehensive evaluation of the CNN model's performance. This comparison would help readers understand the advantages and disadvantages of different approaches and facilitate the selection of the most appropriate method for their specific application.
- Conclusion Needs Strengthening
The conclusion should summarize the key findings and contributions of the study. It should highlight the advantages of the proposed CNN model in predicting subsidence and discuss its potential applications and limitations. Additionally, the conclusion should provide specific suggestions for improving the model and identify potential future research directions. By strengthening the conclusion, the paper will leave a lasting impression on readers and motivate further research in this important area.
Citation: https://doi.org/10.5194/gmd-2024-15-RC1
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
108 | 31 | 9 | 148 | 5 | 5 |
- HTML: 108
- PDF: 31
- XML: 9
- Total: 148
- BibTeX: 5
- EndNote: 5
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1