Using Support Vector Machine (SVM) with GPS Ionospheric TEC Estimations to Potentially Predict Earthquake Events

Round 1

Reviewer 1 Report

Numerous studies show that the links between earthquakes and ionospheric variations make predicting earthquakes possible. Overall the manuscript is worthy for publication if the following major comments are addressed well.  

 Major comments:

 (1) the criteria of selecting the dataset for testing is not comprehensive. The ionosphere is highly variable with altitude, geographic and geomagnetic latitude and longitude, universal time, season, solar activity, and so on. The authors only examined the earthquake events where the number of sunspots was at most 50, with no solar flares on the day. As we know, solar flares could dramatically enhance the x-ray, EUV, UV radiation, which can dramatically increase the ionization of the upper atmosphere. The EM radiations travel with the speed of light, so the criteria is ok. However, the solar winds generated by the coronal mass ejection (CME) could take several days to reach the earth, and drive geomagnetic storms, leading to large ionospheric variations. Therefore, the criteria cannot exclude all the large variations of ionosphere driven by solar activity. Furthermore, the thunderstorms in the troposphere also could drive ionospheric perturbations.

 (2) the selection of the ionospheric quiet days is questionable. Obviously, Figure 6 shows a clear solar cycle in TECU data. Around 2001 is solar maximum, and around 2007 is solar minimum. At solar maximum, the ionosphere is much more active. The authors did not show the date of the selected quiet days. The comparative of the variations of TECU between the random quiet day and the earthquake day in Figure 7 does not mean much. Therefore the standard deviation TEC timeseries estimation is not reasonable, because this value could be biased by solar cycle activity.

 (3) section 3.2.3 is not necessary. The SVM is a standard method. The introduction of SVM and illustration of SVM is redunant. Instead of introducing the SVM, the author should provide more details of the SVM for this application and the experiment setup.

 (4) Finally, the extrem high accuacy of the earthquake precursor preditor is questionable, since the quiet days are not representative. The test data set should contain earthquake and high ionospheric variations periods. 

Author Response

Answer to Reviewer #1

We would like to thank the reviewer for the time and effort which spent reviewing our paper. All comments, suggestions, and questions were carefully considered, and all necessary corrections were made in the revised manuscript.

 

General comments:

Numerous studies show that the links between earthquakes and ionospheric variations make predicting earthquakes possible. Overall, the manuscript is worthy for publication if the following major comments are addressed well.

 

Specific comments:

(1) the criteria of selecting the dataset for testing is not comprehensive. The ionosphere is highly variable with altitude, geographic and geomagnetic latitude and longitude, universal time, season, solar activity, and so on. The authors only examined the earthquake events where the number of sunspots was at most 50, with no solar flares on the day. As we know, solar flares could dramatically enhance the x-ray, EUV, UV radiation, which can dramatically increase the ionization of the upper atmosphere. The EM radiations travel with the speed of light, so the criteria is ok. However, the solar winds generated by the coronal mass ejection (CME) could take several days to reach the earth, and drive geomagnetic storms, leading to large ionospheric variations. Therefore, the criteria cannot exclude all the large variations of ionosphere driven by solar activity. Furthermore, the thunderstorms in the troposphere also could drive ionospheric perturbations.

We agree that a more prefund criteria should include also solar wind particle influence originated from CME events. Therefore, we went over our data base again and double checked that there were no large-scale CMS events took place during 48 hours prior to all earthquake events used in our study. We also added this in the revised text.

Regarding thunderstorm activity in the troposphere, all possible effects of lightning activity mainly influence the lower part of the ionosphere, i.e. the D-layer, as most of the EM energy is reflected back as it bounced between the lower ionosphere and the conducting ground. The small percentage which can actually penetrate the ionosphere via VLF Whistler propagation modes is very sparse, unique and localized. Most of the studies which deals with ionospheric disturbances during lightning activity concentrate in the sporadic E layer changes. A statistical study which was conducted in 2005, found a relationship between lightning activity and the sporadic E layer intensification [Davis and Johnson, 2005]. However, Barta et al. [2013] showed that the electron density of the sporadic E layer has decreased significantly associated with lightning based on infrared maps and lightning observations over the ionospheric station of Rome.

Davis, C. J., and C. G. Johnson (2005). Lightning-induced intensification of the ionospheric sporadic E layer,Nature, 435, 799–801.

 

Barta, V., C. Scotto, M. Pietrella, V. Sgrigna, L. Conti and G. S ́atori (2013). A statistical analysis on therelationship between thunderstorms and the sporadic E Layer over Rome, Astronomical Notes, 334(9),968–971.

 

 (2) the selection of the ionospheric quiet days is questionable. Obviously, Figure 6 shows a clear solar cycle in TECU data. Around 2001 is solar maximum, and around 2007 is solar minimum. At solar maximum, the ionosphere is much more active. The authors did not show the date of the selected quiet days. The comparative of the variations of TECU between the random quiet day and the earthquake day in Figure 7 does not mean much. Therefore the standard deviation TEC timeseries estimation is not reasonable, because this value could be biased by solar cycle activity.

We agree that during solar maximum, the ionosphere is much more active. Our procedure is taking this into account. We estimate the Ionospheric TEC quiet days, where a quiet day corresponding to an earthquake event is defined as the same day and time at a different year where there were no earthquake or solar disturbance (solar storm, geomagnetic storm or CMEs) events, with respect to the monthly mean SSN number during the event. For example, the 2011 Tohoku earthquake occurred in March, where the monthly mean SSN number was 78, i.e. higher than 50 (which is our fixed threshold for determining low or high solar cycle activity), thus a corresponding quite day for this event has be chosen from the same day and time at a different year where there were no earthquake or solar disturbance, and the monthly mean SSN number exceeded 50 SSN.

“Figure 5 shows an example for the TEC time series data around March 11 from 1999-2020. Using our quiet days definition described above, we get 15 days corresponding to high solar cycle activity (> 50 SSN) from which we randomly pick one quiet day for each training set. Our modified Figure 6 shows an example of a randomly picked quiet day (March 11, 2013) from all the quiet days where the monthly mean SSN number exceeded 50 SSN, compared with the March 11, 2011 Tohoku earthquake day.”

This point was clarified in the revised manuscript. Figure 6 was also modified to clarify this point.

 

 (3) section 3.2.3 is not necessary. The SVM is a standard method. The introduction of SVM and illustration of SVM is redundant. Instead of introducing the SVM, the author should provide more details of the SVM for this application and the experiment setup.

We revised this section and concentrated on a more detailed use for the SVM and our experimental setup.

 

 (4) Finally, the extreme high accuracy of the earthquake precursor predictor is questionable, since the quiet days are not representative. The test data set should contain earthquake and high ionospheric variations periods. 

As explained earlier, we have clarified this point as our quite days definition capture both low and high solar cycle activity.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper investigates possibilities of predict earthquake events with ionosphere TEC anomalies. The research background and method are clearly presented. Personally, I would like to see more specific analysis and figures of the relationship between TEC anomaly and earthquake events in Section 5 instead of purely statistics.

1. Please check if Figure 6 and 7 are incorrectly referenced in the text. In Figure 6, it is seen that several TEC anomalies happen in year 2000, 2002, but not 2011, is figure 6 for Tohoku? I may miss something but I think more discussions should be provided to show us that the TEC and earthquake are related.
2. Line 170: I think the IGS rapid ionosphere products are not available in real-time, why not use the IGS final GIM products for post-mission analysis as shown in this paper? Any comments on the applicability of this method on real-time earthquake event prediction?

Some typos should be checked, such as:

Line 221: March 11from…

Line 314: anomaly1?

Author Response

Answer to Reviewer #2

We would like to thank the reviewer for the time and effort which spent reviewing our paper. All comments, suggestions, and questions were carefully considered, and all necessary corrections were made in the revised manuscript.

 

General comments:

This paper investigates possibilities of predict earthquake events with ionosphere TEC anomalies. The research background and method are clearly presented. Personally, I would like to see more specific analysis and figures of the relationship between TEC anomaly and earthquake events in Section 5 instead of purely statistics.

Specific comments:

Please check if Figure 6 and 7 are incorrectly referenced in the text. In Figure 6, it is seen that several TEC anomalies happen in year 2000, 2002, but not 2011, is figure 6 for Tohoku? I may miss something but I think more discussions should be provided to show us that the TEC and earthquake are related.

The figures are now correctly referenced in the revised text.

We estimate the Ionospheric TEC quiet days, where a quiet day corresponding to an earthquake event is defined as the same day and time at a different year where there were no earthquake or solar disturbance (solar storm, geomagnetic storm or CMEs) events, with respect to the monthly mean SSN number during the event. For example, the 2011 Tohoku earthquake occurred in March, where the monthly mean SSN number was 78, i.e. higher than 50 (which is our fixed threshold for determining low or high solar cycle activity), thus a corresponding quite day for this event has be chosen from the same day and time at a different year where there were no earthquake or solar disturbance, and the monthly mean SSN number exceeded 50 SSN.

“Figure 5 shows an example for the TEC time series data around March 11 from 1999-2020. Using our quiet days definition described above, we get 15 days corresponding to high solar cycle activity (> 50 SSN) from which we randomly pick one quiet day for each training set. Our modified Figure 6 shows an example of a randomly picked quiet day (March 11, 2013) from all the quiet days where the monthly mean SSN number exceeded 50 SSN, compared with the March 11, 2011 Tohoku earthquake day.”

 

Additional case studies, for other large earthquakes events compared with their corresponding randomly picked quiet days are shown in the appendix.

 

Line 170: I think the IGS rapid ionosphere products are not available in real-time, why not use the IGS final GIM products for post-mission analysis as shown in this paper? Any comments on the applicability of this method on real-time earthquake event prediction?

We are using the IGS final GIM products and not the IGS rapid ones. This point has been clarified in the revised manuscript.

Since for a proof-of-concept demonstration it is applicable to use the IGS final GIM products for post-mission analysis, however any future real-time earthquake event prediction platform will require the ability for producing TEC maps at shorter latency time scales. For a specific designated area, it is possible to use a local GNSS receivers real time rinex data to estimate the TEC values in near real time latency, thus producing every hour the previous 23- or 47-hours TEC values and test them via a post-mission training set to statistically determine whether an event would occur or not. We added a comment regarding this matter in the discussion section in the revised manuscript.

 

Minor comments:

Some typos should be checked, such as:

 

Line 221: March 11from

Changes were made in the revised manuscript.

 

Line 314: anomaly1? Corrected

Changes were made in the revised manuscript.

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Major comments:

Obviously, the results in the Appendix from Figures A11 to A14 show clear diurnal variations. I believe this diurnal variation is residual diurnal variations after the detrend. Probably it won’t affect the results, but it suggests the detrending method doesn’t work well.

 Section 3.1.3, I would like to suggest that the authors examine the KP index found from NOAA SWPC data. KP index is a good indicator showing the geomagnetic activity.  

 Minor comments:

Line 113, the phase speed of the GNSS signal increase, but the group speed is reduced. Please add “phase speed” in the context.

 Line 141, there is no direct evidence that can show the surface charges are transported into the ionosphere. Actually, this hypothesis is controversial. If the authors have, please add a reference. If not, it’s better to modify the sentence. It’s possible that the charges are transported, or the charges build up an electric field that can affect the ionosphere.

 Line 212, “Where” -> “where”

 Line 215 “we s generate” -> “we generate”? Please verify.

 Line 220, there are numerous ways of doing moving window detrend. It’s better to show the mathematical expression. For example, xi = xi – sum(xj) j from i-5 to i+5.

 Line 246, “As such, We seek to” -> “As such, we seek to”

 Line 249, “such as such as” -> “such as”

 Line 278 “Following are the different skill scores…” what are the “different skill scores”?

 Line 297, “quiets years. Following” -> “quiet years. Following”

Author Response

Answer to Reviewer #1

We would like to thank the reviewer for the time and effort which spent reviewing our paper. All comments, suggestions, and questions were carefully considered, and all necessary corrections were made in the revised manuscript.

 

Specific comments:

Obviously, the results in the Appendix from Figures A11 to A14 show clear diurnal variations. I believe this diurnal variation is residual diurnal variations after the detrend. Probably it won’t affect the results, but it suggests the detrending method doesn’t work well.

We thank the reviewer for pointing out this issue. The figures which were presented in the appendix (including Figure 7 in chapter 4), are before the detrending step, as we only normalized the TEC time series (TEC – minimum(TEC))/maximum(TEC) to point out that within a two day’s time window before the earthquake event, the ionospheric TEC values exceeds 4 times the standard deviation of the randomly chosen quiet day. we have clarified this point in the revised manuscript.

Section 3.1.3, I would like to suggest that the authors examine the KP index found from NOAA SWPC data. KP index is a good indicator showing the geomagnetic activity.  

We did examine the Kp index for excluding any geomagnetic activity. As such we filtered out all the relevant maps during days with high Kp index values (kp ≥ 6). we have clarified this point in the revised manuscript.

 

Minor comments:

Line 113, the phase speed of the GNSS signal increase, but the group speed is reduced. Please add “phase speed” in the context.

Changes were made in the revised manuscript.

Line 141, there is no direct evidence that can show the surface charges are transported into the ionosphere. Actually, this hypothesis is controversial. If the authors have, please add a reference. If not, it’s better to modify the sentence. It’s possible that the charges are transported, or the charges build up an electric field that can affect the ionosphere.

We have modified the sentence accordingly. “The accumulated surface charge over land or ocean drives the current outward. After the charge neutralization time, it is possible that some surface charges are transported into the ionosphere.”

 

 Line 212, “Where” -> “where”

Changes were made in the revised manuscript.

 

Line 215 “we s generate” -> “we generate”? Please verify.

Changes were made in the revised manuscript.

 

Line 220, there are numerous ways of doing moving window detrend. It’s better to show the mathematical expression. For example, xi = xi – sum(xj) j from i-5 to i+5.

Changes were made in the revised manuscript.

 

 Line 246, “As such, We seek to” -> “As such, we seek to”

Changes were made in the revised manuscript.

 

 Line 249, “such as such as” -> “such as”

Changes were made in the revised manuscript.

 

 Line 278 “Following are the different skill scores…” what are the “different skill scores”?

We have changed the sentence to: “Following are the results of the different skill score metrics we used for our best model.”

 

 Line 297, “quiets years. Following” -> “quiet years. Following”

Changes were made in the revised manuscript.

Author Response File: Author Response.pdf