Discrimination of quarry blasts from earthquakes in Northern and Central Egypt using linear and quadratic discriminant functions

In this study, seismic events in Northern and Central Egypt are inspected to discriminate quarry blasts from earthquakes. We examine a collection of 639 events in both time and frequency domains with local magnitudes of 1.5 ≤ ML ≤ 3.3 from the Egyptian Seismological Network’s seismic event catalogue between 2009 and 2015. The maximum S-wave to the maximum P-wave amplitude ratio, complexity (C), spectral ratio (Sr), and power of events (Pe) classifiers as well as two statistical approaches, linear discriminant function (LDF) and quadratic discriminant function (QDF), are used to distinguish between earthquakes and quarry blasts. The usage of the LDF and QDF forms did not result in any major differences in the discrimination. The results obtained by the LDF and QDF from (Pe-C) are the best of all approaches. The findings of all approaches were compared to get a final categorization for each event, and a decision was achieved when at least three of the four methods provided the same event category. In Northern Egypt, 243 earthquakes and 308 quarry blasts could be identified as final decisions, with two misclassified events, resulting in an overall success rate of 99.6%. In Central Egypt, 48 earthquakes and 36 quarry blasts were classified as a final decision, with two misclassified events, for an overall success percentage of 97.6%.


Introduction
Seismic monitoring, which uses a national network of seismic stations, is a powerful tool for detecting both natural and man-made events. In zones where both events occur, it is crucial to distinguish between quarry-mining blasts and earthquakes. Chemical explosions in quarries and mine regions frequently contaminate the earthquake catalogue database. For a reliable seismic hazard assessment and recognition of any area's tectonics, the catalogue must be decontaminated. Seismologists face a challenge in determining the nature of a seismic event. The challenge of identifying quarry-mining blasts has become a major concern for enhancing seismological network performance. Taking into consideration the geographical distribution of the quarry-mining sources, Egypt can be divided into two main man-made seismic zones: Northern and Central zones. The Northern Egypt Zone is the most active blasting zone in Egypt due to a large number of quarry-mining sources. Its relatively moderate seismicity is linked to the Cairo-Suez district and Dahshour seismic sources. The Central Egypt zone is characterized by relatively low blasting activity and low seismicity. Therefore, this zone is covered by a few seismic stations. To build a reliable and classified database, quarry blasts included in the earthquake catalogue need to be removed by application of suitable classification approaches. Various parameters and methodologies have been tested to discriminate between earthquakes and man-made explosions, where the signal is reduced to several attributes. These attributes can be derived from the signal in both time and frequency domains. Different approaches have been used to distinguish between earthquakes and manmade explosions, such as the Pn/Sn, Pn/Lg, Lg/Pg, and Lg/Rg amplitude peak ratio (e.g., Baumgardt and Young 1990;Wüster 1993;Walter et al. 1995;Horasan et al. 2009), the spectral ratio of seismic phases P/S or spectrogram (Hedlin et al. 1989(Hedlin et al. , 1990Chernobay and Gabsatarova 1999;Koch and Fah 2002;Allmann et al. 2008;Korrat et al. 2008;Dahy and Hassib 2010;Hammer et al. 2013), P/S corner frequencies ratio of the spectrum (Ataeva et al. 2017), maximum S to P amplitude ratio versus logarithm of the peak amplitude (AS/AP-log AS) (Baumgardt and Young 1990;Wüster 1993;Horasan et al. 2009), complexity versus spectral ratio (C-S r ) (Gitterman and Shapira 1993;Budakoglu and Horsan,2018), power of event versus complexity (P e -C) (Kekovali et al. 2012;Badawy et al. 2019), and power of events versus spectral ratio (P e -S r ) methods (Kekovali et al. 2012;Badawy et al. 2019). The retrieved classification features from the visual inspection of the various approaches are translated into an automatic classifier of seismic events to replicate the specialist's thinking process. Automatic classification of natural and man-made seismic events has been studied in the literature using several approaches, including the artificial neural network (ANN) technique (Dowla et al., 1990), the linear discriminant function (LDF) (Horasan et al. 2006(Horasan et al. , 2009Deniz 2010;Kartal 2010;Kekovali et al., 2012;Badawy et al. 2019;Ceydilek and Horasan, 2019), quadratic discriminate function (QDF), diquadratic discriminate function (DQDF), Mahalabonis discriminate function (MDF) technique (Küyük et al. 2011), declustering-dequarry discriminate methods (Kalafat 2010), and the Fisher-Shannon discrimination technique (Telesca et al. 2011).
The purpose of this work is to use the LDF and QDF to distinguish quarry blasts from earthquakes. We used both time and frequency variant parameters (S/P maximum amplitude ratio, complexity, spectral ratio, and power of events) as a classifier estimated from the waveform data recorded by ZAF, HAG, GLL, MYD, and NBNS stations in Northern Egypt and SUT and EDF stations in Central Egypt. To improve the credibility of the results, a combination and comparison of multiple methodologies were used. This work represents an extension of Korrat et al. (2022) investigation of digital seismograms in Egypt for discriminating between earthquakes and quarry blasts.

Dataset
This study examines a dataset of 639 seismic events from 2009 to 2015 with local magnitudes ranging from 1.5 to 3.3, including 294 earthquakes (EQ) and 345 quarry explosions (QB), all of which occurred in Northern and Central Egyptian zones (NE and CE). For our study, we chose a maximum depth of 25 km for earthquakes and up to 15 km radius from the mining site for quarry blasts as the maximum error in the epicenter location. Figure 1 shows the epicentral distribution of these events. The majority of quarry blasts' origin times are reported monthly to the Egyptian National Seismological Network (ENSN) main center by quarry managers. The ATLAS software for seismic data analysis, offered by Nanometrics Inc., Canada, was used to calculate the location parameters of seismic events using a 1-D velocity-depth model of El-Hadidy (1995). The vertical component of the digital velocity waveform data extracted from the ENSN was used for discrimination between earthquakes and explosions in both time and frequency domains. Blasts produced more energy for P-wave than S-wave when compared to earthquakes of the same magnitude, which is why the vertical component of velocity seismograms was used (Baumgardt and Young 1990). Another reason for adopting the vertical component is that during the study period, the majority of ENSN stations had only one vertical component. Furthermore, the vertical component eliminates any spectrum complexity or site effects. This study analyzed recordings from the ZAF, HAG, GLL, MYD, SUT, and EDF short-period stations, as well as the NBNS broadband station since those stations recorded events more clearly and had no technical issues compared to the others (Fig. 1). Data is sampled at 100 Hz. Table 1 shows the locations of the stations used in this study. The dataset for the Northern zone of Egypt (NE) was separated into five areas (I, II, III, IV, and V; Fig. 1), with 553 seismic events (245 earthquakes and 308 possible quarry blasts) recorded by ZAF (area I), HAG (area II), GLL (area III), MYD (area IV), and NBNS (area V) stations. The Central zone of Egypt dataset was separated into two areas (VI and VII; Fig. 1), containing 86 seismic events (49 earthquakes and 37 suspected quarry blasts) recorded by the SUT (area VI) and EDF (area VII) stations. The main reasons for this division were to use only the  direct P-and Sv-waves extracted from the events of nearly the same locations which are recorded at a neighboring station in each area of each zone instead of the whole network. In this situation, the effects of propagation paths were almost identical for all events. This option has also the advantage of maintaining signal quality, achieving a close distance balance in each zone and providing a good signal to noise ratio. In this regard, seismic events (EQ and QB) are chosen from areas in Northern and Central Egypt, respectively, within epicentral distances ranging from 55 to 80 km. Table 2 shows the selected events in each area of each zone (NE and CE) as well as the stations that were used to discriminate between earthquakes and explosions in each area.

Methodologies
In this study, different approaches in the time and frequency domains have been applied to distinguish between earthquakes and quarry blasts in the Northern and Central zones of Egypt. These approaches included maximum amplitude peak ratio (A S /A P ), source complexity (C), spectral amplitude ratio (S r ), and multiparameter discrimination, as well as power spectrum density (P e ). We used a statistical technique to classify earthquakes and quarry blasts, defining quantitative variables for both linear (LDF) and quadratic (QDF) discrimination functions. The waveform data of the investigated events were initially inspected visually taking into consideration the following features: first motion polarity of P-wave, absence of S-wave recording, high amplitude of P-wave, Rg phase manifestation, shallow depth, time of occurrence, and distance to the sites of quarries. The constructed relations between the different parameters are explained as follows: 3.1 Shear-compressional wave maximum amplitude ratio (AS/AP-log AS) The amplitude ratio of the maximum S-to P-wave on the vertical seismogram was correlated to the logarithm of the peak amplitude of the S-wave (in counts) in the time domain (Wüster 1993;Baumgardt and Young 1990). The seismic analysis code (SAC) of Goldstein and Snoke (2005) was used to measure the amplitudes of the maximum S-wave (A S ) and the maximum P-wave (A P ) from the vertical component velocity seismograms of earthquakes and quarry explosions. The maximum peak amplitude ratios (A S / A P ) were then plotted against the logarithm of the maximum S-wave amplitudes (log A S ). For each station in each zone, the amplitude ratio of A S /A P versus the maximum S-wave amplitude (log A S ) plot was constructed individually.

Source complexity(C) versus spectral ratio (S r )
The complexity (C) is defined as the ratio of integrated powers of the vertical component of the velocity record S 2 (t) in two selected time frames (t 0 -t 1 ) and (t 1 -t 2 ), respectively, where t 0 is the P-wave onset time which is dependent on the epicentral distance. The spectral ratio S r is a function of the frequency content of an event and is defined as the ratio of the integrated spectral amplitudes of the seismograms a(f) in each of the selected frequency bands (high-frequency band, h 1 -h 2 , and low-frequency band, l 1 -l 2 ). The frequency content of the events needed to determine S r was derived from the amplitude spectra computed using the Fast Fourier Transform in the previous work of Korrat et al. (2022). C and S r can be written as follows in Eq. 1 and Eq. 2, respectively (Arai and Yosida 2004;Gitterman and Shapira 1993). where a(f) denotes the signal spectral amplitude as a function of frequency (f). A trial-and-error approach was used to estimate the limits of the integrals (t 0 , t 1 , and t 2 ) of C given in Eq. (1), which identified the best indicative C values for both earthquakes and quarry blasts of the same magnitude (Horasan et al. 2009;Yilmaz et al. 2013). The integrals for the events recorded by ZAF, HAG, GLL, and MYD, NBNS stations in Northern Egypt, as well as SUT, and EDF stations in Central Egypt were computed separately. For both Northern and Central Egypt stations, the selected ideal time windows for determining C were 2-7 s and 2-10 s for t 1 -t 2 respectively while t 0 -t 1 was 2 s. The optimal duration of the time window was set to be capable of distinguishing between explosions and earthquakes based on the estimated C value. The limits of the integrals (h 1 , h 2 , l 1 , l 2 ) used in estimating spectral amplitudes, S r in Eq. 2 were identified based on the comparison between the spectra of earthquakes and explosions (Horasan et al. 2009). Different frequency bands were tested to identify the frequency band where S r discriminant had the most performance (Kekovali et al. 2012). The ideal high-frequency limits (h 1 -h 2 ) of 5-10 Hz and low-frequency limits (l 1 -l 2 ) of 1-5 Hz were estimated from the spectra of all seismograms in Northern Egypt. In Central Egypt, the chosen highfrequency band (h 1 -h 2 ) was 7-14 Hz, while the lowfrequency band (l 1 -l 2 ) was 1-7 Hz. Finally, a graph of C versus S r was constructed.

Multi-technique discrimination analysis P e (power of event)
This technique offered a good discrimination picture using both spectral features and amplitudes techniques. The three metrics of the multi-discrimination technique include the maximum peak amplitude ratio (A S /A P ), source complexity (C), and spectral ratio (S r ). We used this method to improve the discrimination procedure and increase the discrimination accuracy of our research work. According to Kekovali et al. (2012), a multi-parameter discriminant was defined as: In this technique, the multiplications of the squares of the amplitude ratio and spectral ratio of the two recognized populations reinforced the power of the event P e . A logarithmic scale of P e (log P e ) versus C and S r for mining blasts and earthquakes was used consistently.

Discrimination functions
The procedures mentioned above are quick and simple to use, but they are not completely reliable. Therefore, the LDF (Fisher 1936) and QDF are used in this study to discriminate efficiently between earthquakes and quarry blasts (Krzanowski 1998;Seber 1984). The LDF analysis produces a single covariance matrix for all classes. QDF analysis, on the other hand, calculates one covariance matrix for each class (Küyük et al. 2014;Yavuz et al. 2019). Discriminant functions are a type of classification approach that uses distinct classes to generate data from various Gaussian distributions. The fitting function determines the parameters of a Gaussian distribution for each class to produce a classifier (Yavuz et al. 2019). The generated classifier selects the class with the minimum misclassification to forecast the classes of new data. In this paper, a Python code for linear and quadratic discriminant analysis, which used algorithms to classify various events specified by the user, was created. LDF and QDF algorithms were used to construct the discriminating line and curve between the scattering variables following the method described by Küyük et al. (2014). In general, the functional form of LDF and QDF can be demonstrated simply as follow: K and K1 are the boundary equation's constant terms, L(1), L(2), and L1 are the boundary equation's linear coefficients, and x and y are the discriminatory variables. Q is the boundary equation's quadratic coefficient matrix. Figure 2 displays the derived values of amplitude peak ratios of S to P waves against the logarithms of the peak amplitudes of S waves for investigating seismograms which were recorded by ZAF, HAG, GLL, MYD, and NBNS stations in Northern Egypt (NE), and SUT and EDF stations in Central Egypt (CE). LDF and QDF algorithms were used to generate the discrimination lines and curves that separate earthquakes from quarry explosions for each station. For Northern Egypt, the investigated seismic records (EQ and QB) in ZAF station (area I) manifested clear discrimination between 64 earthquakes and 25 quarry blasts using the LDF algorithm (Fig. 2a1). Only ten events were misclassified. The QDF method effectively distinguished 63 earthquakes and 27 quarry blasts while 9 events were misclassified ( Fig. 2a2. Table S1 in the supplementary material shows the classification results of the events recorded by the ZAF station. Investigation of the events recorded by HAG stations in region II using the LDF algorithm indicated that 47 earthquakes and 101 quarry explosions were accurately discriminated, whereas 12 events are misclassified (Fig. 2b1). The QDF technique correctly recognized 52 earthquakes and 98 quarry blasts (Fig. 2b2), but 10 events were miscategorized. Table S2 in the supplementary material shows the classification results of the events recorded by the HAG station. For events recorded by the GLL station (Fig. 2c1), the LDF algorithm identified 48 earthquakes and 120 quarry blasts in area III (NE), whereas four events could not be discriminated. The QDF approach identifies 51 earthquakes and 118 quarry blasts (Fig. 2c2); however, three events are misclassified. Table S3 in the supplementary material shows the classification results of the events recorded by GLL. The LDF algorithm discriminates 21 earthquakes and 15 quarry explosions for events recorded by MYD station in area VI (NE), while 6 events are misclassified (Fig. 2d1). 22 earthquakes and 14 quarry blasts were recognized by the QDF method (Fig. 2d2), while 6 events were not classified. Table S4 in the supplementary material shows the classification of the events recorded by MYD. For events recorded by the NBNS station in area V (NE), the LDF algorithm (Fig. 2e1) discriminated 32 earthquakes against 40 quarry explosions, while 8 events were not classified. On the other hand, the QDF algorithm (Fig. 2e2) identified 35 earthquakes and 39 quarry blasts, with six events remaining unclassified. Table S5 in the supplementary material shows the classification of the events recorded by NBNS. Table 3 summarizes the classification results of both LDF and QDF algorithms for each recording station in NE. The classification accuracies of the LDF algorithm for ZAF, HAG, GLL, MYD, and NBNS stations in Northern Egypt were 89.9%, 92.5%, 97.7%, 85.7%, and 90.0%, respectively, according to this method. The classification accuracies of the QDF algorithm, on the other hand, were 90.9%, 93.8%, 98.3%, 85.7%, and 92.5% (Table 3). The analysis also revealed that the QDF algorithm performs slightly better than the LDF approach in terms of classification except for the MYD station which showed the same percentage of separation for both algorithms.

S/P amplitude ratio (AS/AP-log AS) method outputs
For events recorded by the SUT station in area VI (CE) of Central Egypt (EQ and QB), 24 earthquakes and 15 quarry explosions were distinguished with the LDF algorithm (Fig. 2f1) and 4 events were misclassified. Applying the QDF approach, 24 earthquakes and 14 quarry blasts were discriminated (Fig. 2f2), whereas 5 events are misclassified. Table S6 in the supplementary material shows the classification results of the events recorded by the SUT station. For events recorded by the EDF station in area VII of Central Egypt, the LDF algorithm identified 18 earthquakes and 20 quarry blasts while 5 events were not classified (Fig. 2j1). Applying the QDF algorithm, 21 earthquakes and 20 quarry blasts were recognized (Fig. 2j2), whereas two events were not classified. Table S7 in the supplementary material shows the classification results of the events recorded by EDF. Table 3 summarizes the classification results of both LDF and QDF algorithms for each recording station in CE. For stations SUT and EDF, the classification accuracies, using the LDF method were 90.7% and 88.4%, respectively. The percentages of proper identification with the QDF algorithm, on the other hand, were 88.4% and 95.3%, respectively (Table 3). The analysis also explained that the QDF algorithm performs better than the LDF approach for the EDF station while the SUT station gives the opposite behavior in terms of classification. Evaluation of the amplitude ratio method in both Northern and Central Egypt zones revealed a tendency for the S/P amplitude ratio values for earthquakes to be larger than the events generated by quarry blasts because the S-wave amplitude of an earthquake is often larger than the P-wave amplitude. The isotropic character of the explosion sources, which produces predominantly compressional P-waves with extremely weak S-waves, accounts for the smallness of the amplitude ratio of the quarry blasts. Calculated parameters for As/Ap-log (As) methods for ZAF and EDF stations with LDF and QDF forms are in the following equations: The equations for all the stations are listed in the supplementary material Equations (S1 to S14).

Complexity-spectral ratio (C-S r ) method output
The estimated values of complexity (C) versus spectral ratio (S r ) for the vertical component of the investigated seismograms recorded by ZAF, HAG, GLL, MYD, and NBNS stations in Northern Egypt and SUT and EDF stations in Central Egypt are shown in Fig. 3. The discriminating lines and curves that distinguish earthquakes from quarry explosions for each station were created using LDF and QDF algorithms.
With the LDF algorithm, 67 earthquakes and 29 quarry explosions were distinguished from the events recorded by the ZAF station in region I of Northern Egypt (EQ and QB), and four events were misclassified (Fig. 3a1). For the same station, the QDF algorithm correctly classified 68 earthquakes and 28 quarry explosions (Fig. 3a2). Three events, on the   Table S1 in the supplementary material shows the classification of the events recorded by ZAF station. The LDF algorithm recognized 53 earthquakes and 101 quarry explosions among events (EQ and QB) recorded by the HAG station in area II (NE), while 6 events were misclassified (Fig. 3b1). The QDF approach identified 58 earthquakes and 99 quarry explosions (Fig. 3b2), but three events were misclassified. Table S2 in the supplementary material shows the classification of the events recorded by HAG. The LDF algorithm distinguished 51 earthquakes and 120 quarry explosions from seismic events (EQ and QB) recorded by the GLL station in area III (NE), (Fig. 3c1). One event, however, was misclassified. The QDF approach (Fig. 3c2) identified 52 earthquakes and 119 quarry explosions, with one remaining unclassified event. Table S3 in the supplementary material shows the classification of the events recorded by the GLL station. The LDF algorithm distinguished 24 earthquakes and 16 quarry explosions for seismic events recorded by MYD station (EQ and QB) in area IV (NE) (Fig. 3d1). The QDF approach identified 25 earthquakes and 14 quarry explosions (Fig. 3d2), while three events were incorrectly classified. Table S4 in the supplementary material shows the classification of the events recorded by MYD. For seismic events (EQ and QB) in area IV (NE) recorded by NBNS station (Fig. 3e1), the LDF algorithm distinguished 36 earthquakes and 41 quarry explosions, while three were misclassified. The QDF approach identified 38 earthquakes and 41 quarry explosions (Fig. 3e2), whereas only one event was not categorized. Table S5 in the supplementary material shows the classification of the events recorded by the NBNS station.  In Central Egypt, the LDF algorithm succeeds to distinguish 24 earthquakes and 16 quarry blasts between the events (EQ and QB) recorded by the SUT station (area VI) (Fig. 3f1). Only three events were misclassified. On the other hand, the QDF approach determined 26 earthquakes and 16 quarry explosions (Fig. 3f2), whereas one event was not correctly identified. Table S6 in the supplementary material shows the classification of the events recorded by the SUT station. For events recorded by the EDF station (area VII), 22 earthquakes and 21 quarry blasts were discriminated, and three events were misclassified using both the LDF algorithm (Fig. 3j1) and the QDF algorithm (Fig. 3j2). Table S7 in the supplementary material shows the classification of the events recorded by the EDF station. Table 3 summarizes the classification results of both LDF and QDF algorithms for each recording station in CE. In Central Egypt, the classification accuracies with the LDF algorithm were 93.0% and 100% for stations SUT and EDF, respectively, while the QDF algorithm exhibited 97.7% and 100% classification accuracy for the same stations (Table 3). These results showed that QDF provided better discrimination than LDF for the SUT and EDF stations. Furthermore, due to the greater frequency content of earthquake seismograms, S r is larger for earthquakes than for quarry blasts. Calculated parameters of C versus S r analyses with LDF and QDF forms are given in the supplementary material Equations (S15 to S28).

Power of event and complexity (P e -C) method outputs
The discriminant criterion for each station was constructed by plotting logarithmic P e values against complexity values C from the vertical components of the velocity seismograms of the seismic events (EQ and QB) in the study area. Figure 4 displays the power of seismic events (log P e ) versus their complexity (C). LDF and QDF algorithms were used to create the discrimination lines and curves that separate earthquakes from quarry explosions among the distribution parameters (EQ and QB) (Fig. 4). In Northern Egypt, the LDF algorithm discriminated 68 earthquakes and 29 quarry explosions for events located in the area I (NE) utilizing ZAF station records (Fig. 4a1), whereas two events were misclassified. Applying the QDF algorithm to the dataset provided by the same stations (Fig. 4a2), 68 earthquakes and 30 quarry explosions were distinguished, but one event is incorrectly classified. Table S1 shows the classification of the events recorded by ZAF station. The LDF algorithm identified 57 earthquakes and 97 quarry explosions using the events recorded by the HAG station (Fig. 4b1), while 6 events were not classified. The QDF algorithm (Fig. 4b2) correctly classified 57 earthquakes and 99 quarry explosions out of the same dataset, but four events were misclassified. Table S2 shows the classification of the events recorded by HAG station. For the GLL station (area III in the NE), 52 earthquakes and 120 quarry explosions were distinguished using both LDF (Fig. 4c1) and QDF (Fig. 4c2) algorithms. Table S3 shows the classification of the events recorded by GLL station. For MYD station (area IV in NE), 25 earthquakes and 15 quarry explosions were distinguished using both LDF (Fig. 4d1) and QDF (Fig. 4d2) algorithms, while 2 events were not correctly classified. Table S4 shows the classification of the events recorded by MYD station. For the NBNS station (area V in the NE), 38 earthquakes and 41 quarry explosions were discriminated using both LDF (Fig. 4e1) and QDF (Fig. 4e2) algorithms, while only one event was misclassified. Table S5 shows the classification of the events recorded by NBNS station. Table 3 summarizes the classification results of both LDF and QDF algorithms for each recording station in NE. According to these discriminant results in Northern Egypt, the classification precisions with the LDF algorithm were 98%, 96.3%, 100%, 95.2%, and 98.8% for stations ZAF, HAG, GLL, MYD, and NBNS, respectively, whereas the percentages of proper classification with the QDF algorithm were 99%, 97.5%, 100%, 95.2%, and 98.8% (Table 3). The results indicated that the QDF provided slightly better discrimination than LDF in the study area for ZAF and HAG. Both algorithms manifested the same percentage of separation for GLL, MYD, and NBNS stations. In Central Egypt, LDF distinguished 27 earthquakes and 15 quarry explosions for the examined events of SUT station (area VI, Fig. 4f1). There was only one unclassified event. When the same dataset was repeatedly applied to the QDF algorithm (Fig. 4f2), 26 earthquakes and 15 quarry explosions were identified, while two events were misclassified.  Table S6 shows the classification of the events recorded by SUT station. For the EDF station (area VII in the CE), 22 earthquakes and 21 quarry explosions were distinguished using both LDF (Fig. 4j1) and QDF algorithms (Fig. 4j2). Table S7 shows the classification of the events recorded by EDF station. Table 3 summarizes the classification results of both LDF and QDF algorithms for each recording station in CE. In Central Egypt, the classification accuracies with the LDF algorithm were 97.7% and 100%, respectively, for stations SUT and EDF, according to this discriminant criterion. The QDF algorithm's accuracy percentages were 95.3% and 100%, respectively ( Table 3). The results indicated that the LDF provided better discrimination than QDF for a SUT station in the study area. The EDF station showed the same percentage of separation for both LDF and QDF algorithms. It is also significant to note that P e for earthquakes is substantially higher than P e for quarry explosions. Furthermore, earthquakes have positive P e values, but quarry blasts have typically negative P e values (Fig. 4). The calculated parameters for P e versus C analyses with LDF and QDF forms are in the supplementary material Equations (S29 to S42).

Power of event and spectral ratio (P e -S r ) method outputs
Plotting the logarithmic P e values versus the spectral ratio values for each station yields the discriminant criterion. Figure 5 depicts the power of event (log P e ) versus the spectral ratio (S r ) for the vertical components of the investigated velocity seismograms (EQ and QB). The LDF and QDF algorithms resolve the discrimination lines and curves separating earthquakes from quarry blasts among the distribution variables (EQ and QB) (Fig. 5).
In Northern Egypt, 68 earthquakes and 30 quarry explosions were correctly classified using both the LDF (Fig. 5a1) and QDF algorithms (Fig. 5a2), with only one misclassified event using ZAF station records. Table S1 shows the classification of the events recorded by ZAF station. Using the records of the HAG station, 59 earthquakes and 101 quarry explosions were completely distinguished using both LDF (Fig. 5b1) and QDF algorithms (Fig. 5b2) of events that occurred in area II (NE). Table S2 shows the classification of the events recorded by HAG station. For the datasets of GLL station (area III in NE), 52 earthquakes and 120 quarry explosions were separated with both LDF (Fig. 5c1) and QDF algorithms (Fig. 5c2). Table S3 shows the classification of the events recorded by GLL station. For MYD station, 25 earthquakes and 15 quarry explosions were discriminated (area IV in NE) using both LDF (Fig. 5d1) and QDF algorithms (Fig. 5d2), while 2 events were not discriminated. Table S4 shows the classification of the events recorded by MYD station. Thirty-eight earthquakes and 41 quarry explosions were distinguished with both LDF (Fig. 5e1) and QDF algorithms (Fig. 5e2) for the data of NBNS station (region V in NE), whereas only one event was misclassified. Table S5 shows the classification of the events recorded by NBNS station. Table 3 summarizes the classification results of both LDF and QDF algorithms for each recording station in NE. According to this method, the classification percentages for stations ZAF, HAG, GLL, MYD, and NBNS using the LDF algorithm were 99%, 100%, 100%, 95.2%, and 98.8%, respectively, in Northern Egypt. The QDF algorithm, on the other hand, had accuracy percentages of 99%, 100%, 100%, 95.2%, and 98.8% (Table 3). These results indicated that both LDF and QDF algorithms gave the same percentage of separation.
In Central Egypt, 27 earthquakes and 15 quarry explosions were clearly distinguished from the events recorded by the SUT station with LDF technique (Fig. 5f1), while only one event was not classified. Applying the QDF algorithms on the same dataset (Fig. 5f2), 26 earthquakes and 15 quarry explosions were separated with two misclassified events. Table S6 shows the classification of the events recorded by SUT station. For the dataset of EDF station (region VII in the CE), 22 earthquakes and 21 quarry explosions were completely differentiated using both LDF (Fig. 5j1) and QDF algorithms (Fig. 5j2). Table S7 shows the classification of the events recorded by EDF station. Table 3 summarizes the classification results of both LDF and QDF algorithms for each recording station in CE. According to this approach, the classification accuracies of the LDF algorithm for the SUT and EDF stations in Central Egypt were 97.7% and 100%, respectively. The QDF algorithm, on the other hand, had classification accuracies of 95.3% and 100% (Table 3). These results showed that the percentage of separation for both the LDF and QDF algorithms was the same for , SUT (f), EDF (j). The discrimination lines and curves that separate the earthquakes from the quarry blasts are determined by the LDF and QDF algorithms the EDF station. The LDF algorithm discriminated better than the QDF algorithm for the dataset of SUT station. The calculated parameters for P e versus S r relation with LDF and QDF forms are in the supplementary material Equations (S43 to S56).

Discussion
In this work, 639 seismic events (EQ and QB) in Northern and Central Egypt were classified using four different techniques to discriminate between earthquakes and quarry explosions between 2009 and 2015, based on linear (LDF) and quadratic (QDF) discriminant algorithms for event classification. The analysis includes 553 seismic events from Northern Egypt and 86 events from Central Egypt, which were recorded by seven stations from seven separated areas. Each station is dedicated to one of these areas. The applied techniques include maximum S to P amplitude ratio vs logarithm of the peak amplitude (AS/AP-log AS), complexity vs spectral ratio (C-S r ), power of event vs complexity (P e -C), and power vs spectral ratio (P e -S r ) methods. The usage of multi-approach techniques to KRSK station data in the vicinity of explosion sites in Armutlu Peninsula, Turkey has proven to be an outstanding success in discrimination between earthquakes and quarry blasts with a very high rate (Yavuz et al. 2019).
In Northern Egypt, the classification achieved using the last approach (P e -S r ) is the best of all four approaches. It manifested the best discrimination results with the highest classification accuracy. The LDF and QDF versions of this approach yield only one unclassified event for ZAF and NBNS stations, and two misclassified events for the MYD station while HAG and GLL stations provide a 100% classification success rate without any misclassified events. The LDF and QDF versions of the third approach (P e -C) produced the same number of misclassified events for GLL, MYD, and NBNS stations as the fourth approach. The QDF version of the same approach from ZAF station yields the same misclassified events as the fourth approach; however, the LDF yields two. The results obtained by LDF version of the third approach (P e -C) belonging to HAG station manifest 6 misclassified events while the QDF version of the same approach manifests 4 misclassified events.
In Central Egypt, the third approach (P e -C) and the fourth approach (P e -S r ) are the best of all four approaches for a SUT station with only one misclassified event for LDF technique and two misclassified events for the QDF technique while the second, third, and fourth approaches provide 100% classification success rate without any misclassified events for EDF station. These results also show that both quadratic and linear forms produce the same results in terms of misclassified events for the EDF station.
The performed discriminant analyses suggest that the power of event-complexity (P e -C) and the (P e -S r ) techniques provide an improved discriminatory result compared with any of the other preceding approaches. The results obtained by Kekovali (2012) and Badawy et al. (2019) from C-LogP e and S r -log P e analyses in the Kütahya Province, Turkey, and Egypt agree well with our results, with the high rate of discrimination success. These methods, according to Kekovali (2012), are the most reliable for distinguishing mining blasts from tectonic earthquakes. These approaches successfully classified the events jointly with both LDF and QDF. It is also evident that the optimal discriminating approach differs from one site to the other. This is due to the diverse crustal structures, site effects, and quarry blasting procedures in each of the seven different sites investigated in this study. Each site is unique and has its own set of data. Therefore, LDF and QDF functions' parameters may change when the locations of events vary (Yavuz et al. 2019).
In terms of C-S r approach, the majority of stations exhibited a high separation rate, which is similar to the findings of Horasan et al. (2009) and Badawy et al. (2019) in Istanbul and its vicinity, Turkey and Northern Egypt, respectively. Yavuz et al. (2019) revealed a lower separation percentage rate for Armutlu Peninsula, Turkey than those seen in the other studies. Badawy et al. (2019), Yavuz et al. (2019), and Budakoğlu and Horasan (2018) obtained a lower separation percentage value in Egypt, Armutlu Peninsula, Turkey and Sakarya region of the North Anatolian fault, Turkey than those manifested by Horasan et al. (2009) in Istanbul and its vicinity, Turkey, using the S-wave/P-wave amplitude peak ratio (AS/AP-log AS). Our findings revealed lower separation accuracy than Horasan et al. (2009), but still slightly higher than Badawy et al. (2019) and Yavuz et al. (2019). In our work, the implementation QDF algorithms did not significantly improve the resulting discrimination obtained from the LDF algorithm in any of the four approaches. Both forms gave the same result for the amplitude peak ratio and the C versus S r approaches in Armutlu Peninsula, Turkey (Yavuz et al. 2019).

Conclusion
The findings of all approaches extracted from LDF and QDF algorithms are integrated to get a final judgment on the event type classification. When at least three of four techniques provide the same event type, a decision was reached. Hence, from 553 events in Northern Egypt, 243 earthquakes and 308 possible quarry blasts had been discriminated while two events were misclassified, obtaining a success rate of 99.6% with respect to the manual identification. Out of the 86 seismic events, 48 earthquakes and 36 quarry blasts in Central Egypt could be discriminated while two events were misclassified with a success rate of 97.6%. The proposed method can be integrated in the catalogue screening in fully automatic process allowing the automatic peaking of P and S wave maximum amplitudes and spectrum ratio.
The (P e -S r ) method with the LDF presented in this study achieved better success where applications of algorithm are simple and faster than other methods. Generally, the LDF requires less computations effort while the QDF is based on more complex equations which require more computational power.
The majority of quarry blasts' origin times are reported monthly to the ENSN main center by authorized quarry managers. However, not all the quarries are authorized and therefore, there are some unreported explosions that need to be screened out of the seismic catalogue. The proposed approach can enhance the catalogue screening review process and can spot misclassified events during the manual identification process.
Author contribution Adel S. Othman preparation of the dataset and performed the data analysis; Hesham M. Hussein wrote the manuscript; Mohamed N. ElGabry and Ibrahim Korrat contributed to the conception of the study and revise the manuscript; Ahmed Lethy revise the manuscript; Evrim Yavuz has made some contributions to the code development and formulation of the LDF and QDF equations.
Funding Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB).

Competing interests The authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.