Next Article in Journal
Deep Learning-Based Approximated Observation Sparse SAR Imaging via Complex-Valued Convolutional Neural Network
Next Article in Special Issue
The Application of Fast Fourier Transform Filtering to High Spatial Resolution Digital Terrain Models Derived from LiDAR Sensors for the Objective Mapping of Surface Features and Digital Terrain Model Evaluations
Previous Article in Journal
An Improved Adaptive Grid-Based Progressive Triangulated Irregular Network Densification Algorithm for Filtering Airborne LiDAR Data
Previous Article in Special Issue
Neural Network-Based Fusion of InSAR and Optical Digital Elevation Models with Consideration of Local Terrain Features
 
 
Article
Peer-Review Record

Kernel Density Estimation for the Interpretation of Seismic Big Data in Tectonics Using QGIS: The Türkiye–Syria Earthquakes (2023)

Remote Sens. 2024, 16(20), 3849; https://doi.org/10.3390/rs16203849
by David Amador Luna 1, Francisco M. Alonso-Chaves 1,* and Carlos Fernández 2
Reviewer 1: Anonymous
Reviewer 3: Anonymous
Remote Sens. 2024, 16(20), 3849; https://doi.org/10.3390/rs16203849
Submission received: 3 September 2024 / Revised: 2 October 2024 / Accepted: 11 October 2024 / Published: 16 October 2024

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study employs Kernel Density Estimation (KDE) to compare two distinct  point-cloud populations from the seismic event along the Türkiye-Syria border on February 6, 2023, providing insights into the main active orientations supporting the Global Tectonics framework. The research is interesting and suitable for publication, but there are several issues that need further clarification:

(1)Why do you select this two datasets including crude and relocated-filtered? Which kind of dataset is better to be used in the analysis?  Need the two datasets be validated each other? Neither in the conclusion nor in the discussion section (section 5.2) did I see the pros and cons of the two sets of data results? Could we use only one dataset? Could you please state them clearly in the manuscript?

(2) Figure 6 and Figure 7 are the results of Kernel density maps and Point clouds, they look very similar. Could you please explain the difference between them clearly?

(3) In section 5.2, how do you evaluate the method is good or not without comparing it with other methods’ results?

Comments on the Quality of English Language

Minor editing of English language required.

Author Response

Comments 1: Why do you select these two datasets including crude and relocated-filtered? Which kind of dataset is better to be used in the analysis?  Need the two datasets be validated each other? Neither in the conclusion nor in the discussion section (section 5.2) did I see the pros and cons of the two sets of data results? Could we use only one dataset? Could you please state them clearly in the manuscript?

Response 1: First of all, we appreciate your comments, and of course, all your observations have been included in the revised version of the manuscript.

The main objective of this study is to demonstrate that a comprehensive seismic catalog with significantly higher uncertainty (unfiltered, first case) compared to a relocated catalog (second case), which offers much higher reliability and accuracy, would yield very similar interpretations when analyzed using KDE (see line 61-65). The latter, filtered database, was only used to validate the interpretations derived from the unfiltered catalog (see line 641-644). The advantage of working with raw, unfiltered data is that it allows for a quicker response and interpretation in the event of an earthquake compared to the relocated data. In the case of the unfiltered, non-relocated data, a certain degree of filtering is inherently achieved using KDE. In contrast, this additional processing step would not be necessary for relocated data. It would be ideal to work with relocated data of very high precision and reliability whenever available, but this is not always possible (see line 631-633). While relocated catalogs are ideal for more precise and well-defined studies, the unfiltered data, as demonstrated in this work, can lead to similar interpretations with a significantly shorter response time. While the first type of catalog can be processed almost instantly or within a few hours, the second type requires a more extensive effort and longer processing times (days or even months), see line 635-640.

In terms of pros and cons, unfiltered data allows for quick access and processing (typically available within few hours after the earthquake occurs). It enables the study of a larger data volume, providing a more comprehensive and broader view of seismic activity, including smaller earthquakes that could be lost after filtering. It also requires lower processing costs and offers greater ease in automation, making it possible to apply statistical analyses and filtering (as demonstrated in this study).

On the other hand, the cons include high uncertainty in both the location and magnitude of events, a higher presence of noise and anomalies due to erroneous data that require additional cleaning and filtering, and greater difficulty in analyzing patterns, critical zones, or trends due to the higher dispersion of the data. However, all of these issues can be minimized with a good filtering process.

In contrast, relocated data ensures greater precision and reliability, a significant reduction in noise due to rigorous cleaning, and allows for more reliable identification of patterns, even migration, fault activity, and risk zones with high confidence. 

However, it requires days or even months of processing time to obtain a catalog with fewer earthquakes, greater difficulty in automation, as human input is almost always needed for calculation, inability to obtain data in real time, and the potential exclusion of relevant data (whether it be precursors or aftershocks) due to filtering or because they fall below the magnitude threshold required for precise calculation (see line 535-552).

Comments 2: Figure 6 and Figure 7 are the results of Kernel density maps and Point clouds, they look very similar. Could you please explain the difference between them clearly?

Response 2: Figure 6 presents a scatter plot of seismic points, which displays the raw locations of seismic events without any statistical treatment. This representation allows for a straightforward visualization of the seismic activity and its distribution in space. In contrast, Figure 7 illustrates the same seismic data using a Kernel Density Estimation (KDE) method, which applies a statistical approach to analyze the density of seismic events over a defined area. This method smooths the data, highlighting regions with higher concentrations of seismic activity and providing a more informative representation of the underlying patterns.

Thus, while Figure 6 (now figure 7) shows the raw data distribution, Figure 7 (now figure 8) offers insights into the spatial distribution and trends of seismic events, allowing for a better understanding of interest areas (see figure captions, and lines 418-421).

We have clarified this better in the new version of the manuscript and we have modified the figures for better understanding.

Thanks to your comments, we have also detected, and fixed, an error in the figure caption.

Comments 3: In section 5.2, how do you evaluate the method is good or not without comparing it with other methods’ results?

Response 3: We would like to emphasize that our comparison serves as a validation method specifically for this case, in accordance with the publication intent of the journal. However, it is important to note that this method has two forms of validation: Internal validation: This involves comparing two populations, which is the primary objective of this study, where one of these populations presents data that is of high reliability and certainty. And an External validation: This achieved through congruence with regional geological knowledge. 

Based on these validations, we conclude that the method is valid when comparing these two populations, and it is further corroborated by aligning with another source as is the geological data. Of course, other validation methods would include geomorphology or geophysics, which converge on the same observation. Therefore, we have added a new sentence to the manuscript to clarify this point (see line 641-650).

Reviewer 2 Report

Comments and Suggestions for Authors

This study uses Kernel Density Estimation (KDE) to compare two distinct point-cloud populations from the seismic event along the Türkiye-Syria border on February 6, 2023; providing insights into the main active orientations supporting the Global Tectonics framework. 

 

The work is original, very interesting and deserves to be published. But first, the authors must further explain the methodology, improve the figures 6, 7, 8 and rearrange the conclusion.

Author Response

Comments 1: The work is original, very interesting and deserves to be published. But first, the authors must further explain the methodology, improve the figures 6, 7, 8 and rearrange the conclusion.

Response 1: We would like to thank you for your interest in our publication and for the suggestions you provided. Figures 6, 7 and 8 (now figures 7, 8 and 9) have been modified to improve comprehension and readability. Additionally, following the recommendations of other reviewers, we have included more detailed explanations of the parameter selection process in the methodology and revised the conclusion (see conclusions on section 6) to highlight the pros and cons of applying this method in seismic studies, as well as the future lines of research that may arise from the application of this method (see line 259-310). To clarify the choice of the bandwidth and Kernel Method we have included a new figure (see figure 4) .

Reviewer 3 Report

Comments and Suggestions for Authors

The paper explores the application of Kernel Density Estimation (KDE) for analyzing seismic big data, focusing on the 2023 Türkiye-Syria earthquakes. The topic is timely and important, especially given the recent seismic events in Türkiye and Syria. Using seismic big data to understand tectonic processes is a valuable contribution to both seismology and hazard assessment. However, I encourage the authors to carefully revise the manuscript according to the following comments.

 

1. While KDE is introduced as the method, the paper lacks a robust discussion of current seismic data interpretation techniques. It’s essential to provide a deeper contrast between KDE and traditional approaches, highlighting why KDE offers significant advantages in terms of accuracy, computational efficiency, or insight generation. I suggest the authors expand the review of existing methodologies, such as traditional geostatistical approaches or machine learning techniques in seismology, and clearly position KDE’s novelty within this context.

2. The paper reads more like a technical report than a research article, focusing on the results of KDE application without offering substantial theoretical or methodological innovations. Authors should elaborate on the underlying assumptions, limitations, and potential enhancements of KDE for seismic analysis. They could introduce theoretical frameworks that justify the use of KDE in complex tectonic environments and discuss its limitations in different geological contexts.

3. The discussion section largely reiterates the results without deeply exploring the broader implications for tectonic research, earthquake preparedness, or future research directions. The authors should expand the discussion on how their findings could inform future studies, particularly in seismotectonic zoning or risk assessment. Additionally, they could propose future improvements for KDE or its integration with other methods like GNSS or satellite imaging.

4. The methodological steps related to KDE application, especially bandwidth selection and kernel function choice, need more explanation. The paper lacks a detailed justification for these parameter choices, which are crucial for KDE analysis. Provide clearer methodological explanations and possibly run sensitivity analyses to show how different bandwidths or kernel functions might impact the seismic density results.

Comments on the Quality of English Language

Minor editing of English language required.

Author Response

Comments 1: While KDE is introduced as the method, the paper lacks a robust discussion of current seismic data interpretation techniques. It is essential to provide a deeper contrast between KDE and traditional approaches, highlighting why KDE offers significant advantages in terms of accuracy, computational efficiency, or insight generation. I suggest the authors expand the review of existing methodologies, such as traditional geostatistical approaches or machine learning techniques in seismology, and clearly position KDE’s novelty within this context.

Response 1: First of all, we would like to thank you for your detailed comments. It was very helpful in revising the sections you mentioned to improve the understanding of the article and, as you pointed out, its significance and usefulness.

We understand the importance of situating our work within the broader context of current seismic data interpretation techniques. We have conducted thorough review of current methods and include them in the introductory part of the article. To address this, we have expanded the manuscript by incorporating a more comprehensive review of traditional geostatistical approaches (e.g., kriging, spatial interpolation) and machine learning methologies (e.g., clustering and classification algorithms) commonly used in seismological studies.

We acknowledge that traditional geostatistical approaches provide precise local estimates of seismic attributes but often require extensive prior information and are sensitive to the spatial configuration of the data. Similarly, while machine learning techniques have shown promise in handling large-scale seismic datasets, they tend to need so many data and may suffer from interpretability issues, making them less ideal for capturing the inherent complexity of seismicity distributions.

By contrast, KDE offers a non-parametric, data-driven approach that provides a smooth representation of seismic event distributions without assuming a specific functional form (non-parametric method). This flexibility allows KDE to reveal hidden patterns in seismicity with minimal assumptions, making it particularly suitable for analyzing the vertical distribution of seismic clusters and identifying seismogenic levels. Furthermore, KDE’s efficiency in big data handling and its adaptability to different contexts position it as a powerful tool for big data applications in seismology.

We have revised the manuscript accordingly to highlight these contrasts and strengthen the positioning of KDE as a robust alternative in the context of seismic data interpretation. However, none of these methods are mutually exclusive and can be applied together for a better geological understanding of these complex contexts. (See line 141-199)

Comments 2: The paper reads more like a technical report than a research article, focusing on the results of KDE application without offering substantial theoretical or methodological innovations. Authors should elaborate on the underlying assumptions, limitations, and potential enhancements of KDE for seismic analysis. They could introduce theoretical frameworks that justify the use of KDE in complex tectonic environments and discuss its limitations in different geological contexts.

Response 2: The novelty is not just the use of KDE in seismology (which is also notable), but rather the differentiation in depth of seismogenic levels on which the kernel is applied, allowing for the recognition of three-dimensional structures at depth. Thanks to your comments, we have included in the manuscript the underlying assumptions and limitations of the KDE method as applied to seismic analysis. Specifically, we have addressed the following key points:

Fundamentals and limitations of kernel: One of the primary limitations of the KDE method is that it produces an image representation of seismic data, which requires expert interpretation. We have clarified this point in the revised manuscript.

Scale, Data Volume, type of kernel method and Bandwidth: We have discussed how the scale of the analysis, the volume of data, and the bandwidth of the kernel method impact the results and interpretations. Understanding these factors is crucial for applying KDE effectively in different geological contexts. A new figure showing the distinct effects of bandwidth and kernel methods on the resulting map has been included in the manuscript.

Theoretical Frameworks: We have introduced theoretical frameworks that justify the use of KDE in complex tectonic environments, providing contexts for its application and enhancing the manuscript’s scientific rigor.

Limitations in Geological Contexts: We are working on applying the method to other tectonic contexts, and the preliminary results are encouraging. The final results will be presented in future publications. (See line 553-583)

Comments 3: The discussion section largely reiterates the results without deeply exploring the broader implications for tectonic research, earthquake preparedness, or future research directions. The authors should expand the discussion on how their findings could inform future studies, particularly in seismotectonic zoning or risk assessment. Additionally, they could propose future improvements for KDE or its integration with other methods like GNSS or satellite imaging.

Response 3: Due to the focus of this journal, we presented only the methodology in this manuscript. However, as we have mentioned before, we are investigating the applicability of the method in other geological contexts (Granada basin, La Palma volcanic zone, other transform faults and subductive contexts). Those studies, we are modifying the bandwidth according to the scale of the studied regions.

Accordingly, in the revised discussion section, we have emphasized that despite its limitations, this technique provides more information compared to other methods. Regarding earthquake preparedness, these techniques can help better define seismic hazard, identify critical zones, segment faults, and delineate segments that have been active but are currently inactive.

We propose that this technique should be a routine method used in seismic centers due to its versatility, agility, and ease of use. It serves as an initial approximation, almost in real time, while awaiting data refinement, with subsequent interpretations not differing significantly from previous results. These new data can provide more detailed precision and can be processed automatically by computers for immediate calculation.

Furthermore, this technique, combined with a tectonic and rheological study of the studied portion of the lithosphere, is key for the definition of the different seismotectonic levels at that zone. This could lead to the generation of a catalog of slices or layers for each section of the Earth, similar to how we currently perform seismic zoning for crustal resistance in plan view, but now vertically within the lithosphere.

We believe these enhancements to the discussion will better address your concerns and provide a more comprehensive understanding of the implications of our article.

Comments 4: The methodological steps related to KDE application, especially bandwidth selection and kernel function choice, need more explanation. The paper lacks a detailed justification for these parameter choices, which are crucial for KDE analysis. Provide clearer methodological explanations and possibly run sensitivity analyses to show how different bandwidths or kernel functions might impact the seismic density results. 

Response 4: The choice of a 0.05º bandwidth was determined considering the study area, which covers approximately 250,000 km2, and the spatial distribution of more than 5,500 seismic events recorded. This value provides an adequate balance between spatial detail and the smoothing necessary to capture coherent seismic patterns in such an extensive region. A smaller bandwidth would increase local resolution but could introduce noise in areas with lower event density. Conversely, a larger bandwidth would excessively smooth the seismic distribution patterns, decreasing the ability to identify key geological structures. Therefore, the 0.05º value allows for capturing both global trends and regional details of seismicity in the area of interest.

In QGIS, several kernel functions are available by default, including quartic, triweight, uniform, Epanechnikov, and triangular. For this study, the quartic function was selected due to is ability to provide an adequate balance between smoothing and spatial accuracy. Unlike the triweight function, which decreases more rapidly towards the edges, the quartic function offers a smoother transition, allowing for better capture of heterogeneous distributed patterns. The uniform function, by assigning equal weight to all points within the bandwidth, is not suitable for capturing the gradual variation of the data. On the other hand, the Epanechnikov function provides similar smoothing to the quartic but with less capability to detect subtle variations. Finally, the triangular function tends to excessively prioritize nearby points, which could exclude the influence of distant but relevant seismic events. Therefore, the choice of the quartic function ensures an optimal balance in the seismic density analysis. However, it is important to highlight that the difference in the kernel method employed are minimal compared to the impact that the bandwidth has on the results.

To evaluate the impact on the distribution of seismic density, a sensitivity analysis was conducted in which different bandwidth values (0.01º, 0.03º, 0.05º, and 0.07º) and kernel functions (including Epanechnikov, triweight and quartic) were tested. As shown in the figure that is included in the new version of the manuscript, lower bandwidth values generate more detailed results but with greater noise, while higher values highlight regional patterns at the expense of local patterns. (See line 259-310, we have included a new figure -num.4- to present the effect of bandwidth and the Kernel Method on the resulting maps and why we have selected these parameters )

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thanks for the author's careful consideration and modification of the  comments. After the modification, the logic is smoother ,  the content is richer, and the illustration of the figures are also clearer.  The manuscript meets the requirements for publication.

Back to TopTop