Seismic Ground Response Prediction Based on Multilayer Perceptron
Round 1
Reviewer 1 Report
The paper is well structured and interesting, but not outstanding. It presents a machine learning approach to estimate the seismic ground response to an earthquake in Japan. Nevertheless, some issues should be solved before its publication.
General comments:
The word "predict" appears seven times in the Abstract. Over the text, "predict", "predicted", "prediction", or so forth, are found more than 60 times. I strongly recommend using a synonym.
Some figures present incorrect locations or alignments.
Please, type the equations, not copy-paste as an image.
Recent bibliography is missing: 0-2020, 5-2019, 0-2018, 1-2017, even for
MLP. Some suggestions are as follows:
https://doi.org/10.3390/app10020689
https://doi.org/10.1016/j.cageo.2020.104548
Specific comment:
Some minor spelling issues have been found:
Unnecessary use of "the" ground motion, or the missing in "frequency".
L214: I do not understand the meaning of "w" in "...80% of them w used as training... ".
L358: "is shown" instead of "are shown".
Additionally:
Figure 2: the scale bar, the legend, and the coordinate system are missing. It would be interesting depict the epicentre information in the same figure, as it can help the readers to correlate the epicentre-stations distance with the acceleration data.
L167: 9.0 in Richter scale should be replaced by a moment magnitude (Mw) datum. Mw is the most frequent (and universally accepted) magnitude in literature.
Author Response
I attached the file to provide a point-by-point response to the reviewer's comments.
Authors deeply appreciate for the excellent comments of reviewers. Vast majority of the text were rewritten and new plots were added following reviewers’ comment. The details are following.
In advance, we would like to let you know that we revised many parts of the paper we submitted, including revising your advice. Thank you.
Reviewer 1
- General comment:
- The word "predict" appears seven times in the Abstract. Over the text, "predict", "predicted", "prediction", or so forth, are found more than 60 times. I strongly recommend using a synonym.
We tried to get rid of unnecessarily duplicated sentences throughout the paper, and used sysnonym to replace “predict”
L172 “Examples of the ground motion time histories on the surfaces measured at stations IBRH11, IBRH13, and IBRH20 and estimated by the MLP model are plotted in Figure 7. The ground motion time history measured and predicted seem to have similar characteristics.”
L221 “Consequently, the MLP model estimates the seismic motions on the surface based on the motions at bedrock (or 100 m).
2. Some figures present incorrect locations or alignments.
We arranged all the figures that should be corrected such as "Figure 1", "Figure 3" to make alignments.
3. Please, type the equations, not copy-paste as an image.
We retyped the equations throughout the paper with additional necessary equations.
4. Recent bibliography is missing: 0-2020, 5-2019, 0-2018, 1-2017, even for MLP. Some suggestions are as follows:
https://doi.org/10.3390/app10020689
https://doi.org/10.1016/j.cageo.2020.104548
We referred recent study that you suggested [6,7]. Thank you.
L37 : “Developments in deep learning have opened up new possibilities in the domain of civil engineering [5-7].”
5. Some minor spelling issues have been found:
-Unnecessary use of "the" ground motion, or the missing in "frequency".
I appreciated and I tried to fix the unnecessary use of “the” and fill the missing one.
L79: 50 Earthquake events measured in two directions, east-west and north-south providing 100 sets of ground motion accelerations in each site.
-L214: I do not understand the meaning of "w" in "...80% of them w used as training...
We fixed the mistype “w” to “were”.
L86: “Out of 100 sets of seismic motion data, 80 datasets from 40 events were used for the train (Train dataset), and 20 datasets from 10 events for the test (Test dataset).”
Additionally:
6. Figure 2: the scale bar, the legend, and the coordinate system are missing. It would be interesting depict the epicentre information in the same figure, as it can help the readers to correlate the epicentre-stations distance with the acceleration data.
It could not be better to depict the epicentre information in the paper. In the future research I will try to help readers with depicting. Thank you so much. Unfortunately, it seems it is not affordable to make it in this paper.
7. L167: 9.0 in Richter scale should be replaced by a moment magnitude (Mw) datum. Mw is the most frequent (and universally accepted) magnitude in literature.
In the middle of revision, we got rid of many duplicated and unnecessary information and changed the structure of the paper. We found that the information that the sentence above provides is not necessary to this paper to make reader understand the study we conducted and deleted the sentence in the context.
Please also find the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper deals with the important valuation of the seismic ground motion responce based on machine learning approach. Method utilized are modern and appropriate. I think the topic is important for Applied Sciences and should be of interest to geophysicist and engineering working in the effect of earthquake on the surface of the Earth. In my opinion this paper should be revised before being published. The authors can find my suggestions in the attached file.
Comments for author File: Comments.pdf
Author Response
I attached the file to provide a point-by-point response to the reviewer's comments.
Authors deeply appreciate for the excellent comments of reviewers. Vast majority of the text were rewritten and new plots were added following reviewers’ comment. The details are following.
In advance, we would like to let you know that we revised many parts of the paper we submitted, including revising your advice. Thank you.
Reviewer 2
- “In addition, the seismic wave disseminated to the earth and sand ground through the bedrock area has ground amplification, depending on the properties of the bedrock while passing the soil stratum, and then puts weight on the upper structure” - This sentence is not clear. Please explain better.
We not only made sentence clear but also made introduction clearer than before.
L26“Earthquake disasters can cause enormous social and economic damage [1, 2], and therefore the sustainability of infrastructure requires the mitigation of earthquake consequences. The disaster mitigation should consider the seismic performance of structures due to earthquakes. The bedrock motion may be remarkably amplified on the ground surface due to the soil layers [3]. The estimation of the amplification in ground response is challenging for the designers. Geotechnical engineers have been developing quantitative methods for the seismic ground response. Seismic ground response analysis refers to the evaluation of acceleration response on the surface based on the characteristics of the soil deposits and the bedrock motions.”
2. “and the soil and sand” - Why the authors use the litotype sand together with the litotyper soil? Typically, is used soils, or soft soils, to identify the shallower layers where seismic amplification occurs.
“sand” was not proper word to be typed in there. We fixed it.
L26“Earthquake disasters can cause enormous social and economic damage [1, 2], and therefore the sustainability of infrastructure requires the mitigation of earthquake consequences. The disaster mitigation should consider the seismic performance of structures due to earthquakes. The bedrock motion may be remarkably amplified on the ground surface due to the soil layers [3]. The estimation of the amplification in ground response is challenging for the designers. Geotechnical engineers have been developing quantitative methods for the seismic ground response. Seismic ground response analysis refers to the evaluation of acceleration response on the surface based on the characteristics of the soil deposits and the bedrock motions.”
Equation (2) - There are many methods that can provide the natural period (or resonance frequency) of the soil. For istance, I suggest the authors to apply the method proposed by Nakamura (1989) to the seismic ambient noise recorded by the seismic network stations. Nakamura have been shown that the fundamental resonance frequency, which corresponds to the lowest amplification frequency, to be successfully determined by the horizontal-to-vertical spectral-ratio (HVSR) method, especially when a sharp velocity contrast is present at depth (BONNEFOYCLAUDET et al., 2009).
- BONNEFOY-CLAUDET, S., BAIZE, S., BONILLA, L.F., BERGE-THIERRY, C., PASTEN, C., CAMPOS, J., VOLANT, P., and VERUGO, R. 2009), Site effect evaluation in the basin of Santiago de Chile using ambient noise measurements, Geophys. J. Int., 176, 925-937.
- Nakamura, 1989. A method for dynamic characteristics estimation of subsurface using microtremor on the ground surface. Railway Technical Research Institute, Quarterly Reports]
Getting natural period itself is not the scope of the study. But since you let us know the research that is worth, we refer the paper you mentioned as in the following.
L108: “It is noted that the natural period can also be evaluated from the field measurements. For example, Nakamura [18] showed that the fundamental resonance frequency, which corresponds to the lowest amplification frequency, to be successfully determined by the horizontal-to-vertical spectral-ratio (HVSR) method, especially when a sharp velocity contrast is present at depth.”
4. Figure 3 - I think that the figure is overlapped to the text and it's not clear to me what is written under it. Please check.
We arranged the figure to make it clear to read.
5. “Out of 100 seismic wave datasets, 80% of them w used as training datasets” - were
We fixed the mistype “w” to “were”.
L86: “Out of 100 sets of seismic motion data, 80 datasets from 40 events were used for the train (Train dataset), and 20 datasets from 10 events for the test (Test dataset).”
6. ”The seismic ground response analysis of the physical model used for seismic ground response analysis is a method to predict the free-field movement on the ground surface when an earthquake by modeling the strata on the bedrock,” / when an - "during an" or "generated by an”
We replace “when an” to “during an” as your advise.
L140: “The seismic ground response analysis is to predict the free-field movement on the ground surface during an earthquake [27].”
7. The program also introduces the concept of equivalent line analysis to produce relatively reliable results for the seismic ground response analysis results. / line – linear
We made some changes in explanation of the conventional physical model.
L153: The nonlinearity of the shear modulus and damping is taken into consideration for using equivalent linear soil properties with an iterative procedure to obtain values for modulus and damping compatible with the effective strains in each layer [28].
8. “As a result, the ground motion time history predicted from the MLP presents a shape similar to the measured result” / ”measured result” – measure one
We changed the whole sentences for better explanation.
L172” Examples of the ground motion time histories on the surfaces measured at stations IBRH11, IBRH13, and IBRH20 and estimated by the MLP model are plotted in Figure 7. The ground motion time history measured and predicted seem to have similar characteristics.”
9. “Figure 8. Measured and prediction results of ground motion time history due to earthquake at each station:” / Measured -Earthquake accelerations measured (grey line) , results - Earthquake accelerations measured (grey line)
We replaced the title of the figure.
‘Figure 7. Earthquake acceleration measurements and acceleration predictions by MLP’
10. “where n is the number of (Sa.measured – Sa.predicted), Sa.measured is the spectral acceleration of 328 the measured results, Sa.measured, and the spectral acceleration of the predicted results.” – ‘and Sa.predicted is the’
We added additional equation which is necessary and correct the explanation on it.
L265: where is the i th spectral acceleration of the measurement, is the th spectral acceleration of the prediction from MLP or SHAKE2000 model, and is the number of ground motions in each station.
11. “The seismic ground response analysis results using SHAKE2000, Figure 9–14, show that the spectral acceleration in each station is higher than the measured at a particular period.”
- In my opinion this depends on the variability of the input parameters and in particular the dynamic properties of the soil. It would be useful to show a table in which the individual models for each site are indicated (Vs, thickness, shear modulus degradation and damping). In addition, also the depth of the bedrock must be reported.
We made an explanation of the input parameters and the dynamic properties of the soil for SHAKE2000.
L157: “The KiK-net provides the profile information of the ground with the types of strata, classified as top soil, sandy gravel, clayey gravel, and granite. The nonlinear dynamic properties of each soil are employed from the proposals in the literatures as shown in Table 2. For the top soil, the mean values of normalized shear modulus and damping ratio curves proposed for sands by Seed & Idriss [29] were taken. For both sandy and clayey gravels, the mean values proposed for gravel by Seed et al. [30] were implemented. The mean values of rocks by Schnabel [31] were used for Granite. The shear modulus reduction and damping curves employed are shown in Figure 6.”
‘Table 2: Type of strata and Normalized shear modulus and damping ratio curve’ and ‘Figure 6: Dynamic soil properties: (a) Normalized shear modulus curves; (b) Damping ratio curves’ also will help readers to understand it.
Additionally, we inserted the depth of the bedrock in the Table 1.
12. “The seismic ground response analysis results using SHAKE2000 showed that the spectral acceleration in each station was higher than the measured at a specific period. The seismic ground response analysis results generally amplified the most in the natural period of each station. The ground profile information applied in the seismic ground response analysis using the physical model analysis program SHAKE2000 was insufficient. Thus, the re-397 sults were obtained using dynamic properties applied by a technician, but these results differed from the actual results.”
- Please be clearer. In Fig.3 you show for each site the Vs velocity model up to 100 m depth. These values are from DH carried out inside the wells. Probaly the major differences are due to the presence of lower Vs velocity values (< 1000 m/s) in the top part of the granitc rock (see for instance IBRH11, IBRH13 and IBRH18). It would also be useful to report the depth of the bedrock in the Fig.3. This would help the reader to better understand.
We set the structure of the conclusion and made it clear to understand.
L306:
“This study proposes an MLP based model to evaluate the seismic response of the surface based on the ground motion at bedrock (or 100 m) level and compare its performance with that of conventional physical model SHAKE2000. Total 6 sites in Japan were selected and 100 ground motion at each site were made up for the models. The acceleration response spectra were calculated from the predicted and measured (baseline) acceleration histories for comparison.
The proposed MLP model predicted the magnitudes of response and the natural periods where the response amplifies closely with the measured ground motions (baseline). The proposed model did not perform well for the earthquakes whose response spectra exceeds about 2 g due to deficiency in large earthquake measurements in training datasets.
The MLP model outperformed the conventional model for seismic ground response analysis with few exceptions. It is noted that the seismic ground response analysis by SHAKE2000 was conducted based on the dynamic soil properties in literatures widely used. The results of conventional seismic ground response would rely much on the input properties.
The proposed MLP model incorporates the input and output seismic motions with no soil properties, and therefore only applicable for certain sites where the earthquake measurements are available. As subsequential development, it is aimed to implement the information on soil layers to the deep learning model so that it can also evaluate the surface ground motion for the site where only soil information is known.”
Additionally, we inserted the depth of the bedrock in the Table 1.
Please also see the attachment.
Thank you
Author Response File: Author Response.pdf
