Elastic Wave Denoising in the Case of Bender Elements Type Piezoelectric Transducers
Round 1
Reviewer 1 Report
Overall
This study aims to obtain high-quality elastic waveform data used in the early warning of rainfall-induced landslides. The authors propose to use a wavelet threshold denoising (WTD) technique together with a superimposed denoising (SD) technique. They demonstrate the denoising results with a WTD only, an SD only, and SD and WTD techniques applied for heavy sine signal, bumps signal, doppler signal, and lab data
It is needless to say that an SD technique is well-known, and a WTD also has a long history since the 1990s. Then, the originality of this study should be the combination of two techniques and application to elastic waves obtained in the tests using bender elements type piezoelectric transducers.
Considering this, there are some concerns about the contribution of this study to the early warning of rainfall-induced landslides.
1. In this study, only small noise is applied for heavy sine signal, bumps signal, and doppler signal. Original SNR is set between 12 and 20. However, the maximum value of the signal-to-noise ratio in the frequency domain is about 100 (I calculate it for heavy sine data; see Fig. ref1). It should be noted that the SNR in the time domain is lowered due to noise in a frequency range without signals. The original signals have enough high SNR to read the arrival time or timing of some phases (Figures 2-4).
Even in the case of elastic wave data, if one wants to know the arrival time of the phase around 350 samples in Figure 5, SD is enough to read the sample point of 361.
I wonder if the elastic wave data in this study were not obtained under noisy conditions such as fields of landslides but under quiet conditions such as in the laboratory.
The authors must show the effects of denoising for much worse data (signals with low SNR around 2-5 in the frequency domain).
In the first place, the elastic wave signals must be small enough not to affect the subsurface structure even if the ground is much soft (the ground to be monitored tends to be soft with a high risk of landslides).
2. In this study, only Gaussian white noise is applied for heavy sine signal, bumps signal, and doppler signal. The dominant frequency ranges of these signals and elastic wave signals (see Fig. ref2) are enough lower than the Nyquist frequency. Then, the effects of Gaussian white noise and the noise in elastic wave data are quite limited and only in higher frequencies than signal frequencies.
The real noise in fields of soft ground should be pulsatory motions, raindrops, ground motion with social activities, elect-magnetic noises, and so on. Such noises are not necessarily white nor steady and can be periodic, pulse-like, monotonic, and/or unsteady. Also, the dominant frequency ranges of such noises are wide and often overlapped with those of elastic wave signals.
The authors must show the effects of denoising the signals with much more realistic noises.
3. If the authors focus on the velocity of elastic waves, they must show not only the images, SNR, and MSE of denoised data but the improvement of the detection of phase arrival.
4. The details of elastic wave data must be provided, such as the condition, path length, the shape of the source signal, and sampling frequency.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 2 Report
The manuscript proposed a denoising method for measurements obtained from bender element piezoelectric transducers. The denoising performance looks good, and the method may be useful to experimental researchers. Still, manuscript has several flaws and requires further revisions in order to be accepted. Detailed comments are provided below:
1. Title: This manuscript is almost irrelevant to early warning of rainfall-induced landslides because bender element tests have rarely been used in practice of landslide early warning. “Early Warning of Rainfall-induced Landslides Based on Elastic Wave:” should be removed from the title.
2. Introduction: According to Comment 1, the background of rainfall-induced landslides should be removed due to comment 1. Instead, the background should be re-written and may re-focus on bender element tests that have been widely adopted in soil experimental studies. In addition, there are too many citations about bender element studies by Y.L. Chen. Such over citation should be avoided.
3. Section 3.4: It is preferred to add some additional and long actual elastic wave signals to verify the proposed method. Instead, some artificially generated signals in sections 3.1 – 3.3 could be minimized.
4. If possible, it is recommended that the authors provide their algorithm code as an appendix of the manuscript for contributing to the research community by sharing.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This study investigated the effectiveness of elastic waveform signal improvement technique, that combined the superposed denoising and wavelet threshold denoising method. While the proposed method indicated a reasonble signal denoising performance,it is hard to find a direct relationship between the proposed method and the landslide early warning system. The test setup details of the signal simulation tests should be supplemented. Also, additional explanations should be provided on how to utilize the study results on accurately predicting the rainfall-induced landslides.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The description of the scope of the study is much improved after removing the phrases recalling the field monitoring and early warning from the title and introduction. The remained concerns are as follows:
1. The authors should show the characteristics of Gaussian noise. As the authors use, Gaussian white noise is a type of Gaussian noise, and there are other Gaussian noises. Some more information such as how to calculate or the name of Gaussian noise is required so that readers can get the spectral information of the noise.
2. In Figure 8, the wave amplitude of each raw record should also be plotted to show the effectiveness of the denoising method. I feel that a lower input voltage is suitable to emphasize the effectiveness of the denoising method.
The followings are minor concerns.
Line 14. The phrase “which is caused by factors such as current and infiltration” should be changed into “which is caused by factors such as fluid current and infiltration” or “which is caused by factors such as water current and infiltration”.
Line 140. The article by Mallat (1989) should be listed in References because some notations in this manuscript are the same as those of Mallat (1989) and the article helps readers to understand the description in SubSection 2.2.1.
Lines 143-145. Some parameters and functions must be in italic (e.g., f(t) in L143, h and g in L144, j and N in L145).
Lines 135, 144, and 163. I think that indents must be removed because they are not heads of paragraphs.
L 279. What is “t” after “,”?
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 2 Report
This manuscript has improved after the revision and becomes suitable for publication.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
