New seismic monitoring center in South America to assess the liquefaction risk posed by subduction earthquakes

The east of Cali is composed of loose sand deposits with high water table levels. This condition and the high seismic hazard of the city make cyclic liquefaction one of the main hazards in the city, which may affect more than 600,000 citizens and important infrastructures such as the city’s main drinking water treatment plants. Therefore, it was decided to design and implement a seismic monitoring center to study the behavior of liquefiable soils under local seismogenic conditions, in which subduction earthquakes predominate. First, more than 130 earthquakes from two seismic monitoring centers with liquefiable layers in the USA were studied to determine the requirements for the adequate design of the monitoring center. Then, a robust geotechnical and seismic characterization of the study area including SPT, CPTu, and seismic and ambient noise tests were carried out. From this information, the specifications and location of the instruments and, in general, the characteristics of the monitoring center were defined. The monitoring center has been planned to be established in two stages, and the first one has already been built and commissioned. The implementation of the first stage allowed to adequately record 35 earthquakes from different seismogenic sources, most of them from subduction earthquakes, and to verify that the potentially liquefiable layer remains saturated throughout the year. Subsequent ground motion sensors will allow to deeply study and understand large shear strains and excess pore pressures generation in the soil deposit, as well as their relationships with different intensity measures. The experience shared herein can benefit the design, construction, and operation of other seismic monitoring centers across the world.

liquefiable soils under local seismogenic conditions, in which subduction earthquakes predominate. First, more than 130 earthquakes from two seismic monitoring centers with liquefiable layers in the USA were studied to determine the requirements for the adequate design of the monitoring center. Then, a robust geotechnical and seismic characterization of the study area including SPT, CPTu, and seismic and ambient noise tests were carried out. From this information, the specifications and location of the instruments and, in general, the characteristics of the monitoring center were defined. The monitoring center has been planned to be established in two stages, and the first one has already been built and commissioned. The implementation of the first stage allowed to adequately record 35 earthquakes from different seismogenic sources, most of them from subduction earthquakes, and to verify that the potentially liquefiable layer remains saturated throughout the year. Subsequent ground motion sensors will allow to deeply study and understand large shear strains and excess pore pressures generation in the soil deposit, as well as their relationships with different intensity measures. The experience shared herein can benefit the design, construction, and operation of other seismic monitoring centers across the world.

Introduction
Cyclic liquefaction is a geotechnical phenomenon in which granular, poorly consolidated, and saturated soils stop behaving as a solid to behave as a viscous fluid. This phenomenon, produced by dynamic loads such as earthquakes, causes important damage to the infrastructure supported on the liquefied soil. The damages reported during the 1964 Niigata earthquake, and more recently during the earthquakes of Maule, Chile, 2010 (Verdugo and González 2015); Darfield, New Zealand, 2010 (Quigley et al. 2013);Christchurch, New Zealand, 2011(Cubrinovski et al. 2011; Tohoku, Japan, 2011 (Bhattacharya et al. 2011);Emilia-Romagna, Italy, 2012(Lai et al. 2015; and Sulawesi, Indonesia, 2018 (Sassa and Takagawa 2019), are examples of the effect of liquefaction. Given the hazard that this phenomenon implies, great efforts have been made in the last decades to understand and predict the triggering and effects of liquefaction through field and laboratory tests, numerical modeling, and analysis of case histories (Ramirez et al. 2018).
In 1982, the US Geological Survey in collaboration with several researchers implemented the Wildlife Liquefaction Array (WLA), a seismic monitoring center for potentially liquefiable soils located on the banks of the Alamo River in Imperial Valley, north of Brawley (California) (Bennett et al. 1984). This monitoring center was able to record one of the most relevant historical cases of liquefaction for the scientific community: the one caused by the 1987 Superstition Hills earthquake with a moment magnitude (M w ) of 6.6 and peak ground acceleration (PGA) of 0.21 g (Youd and Holzer 1994;Youd et al. 2004b;NEES@ UCSB 2020). Given the importance of this earthquake to understanding liquefaction, other monitoring centers have been established, such as the Garner Valley Downhole Array and the Seattle Liquefaction Array (NEES@UCSB 2020).
Liquefaction has been usually studied in the framework of shallow crustal earthquakes, which are the ones that historically have generated more cases of liquefaction around the world. Indeed, conventional methods to evaluate liquefaction potential are based on liquefaction data produced by shallow crustal earthquakes (e.g., Boulanger and Idriss 2014). However, in Colombia (e.g., 1979 Tumaco earthquake), and in other South American countries (e.g., Peru and Chile), liquefaction has been generated by earthquakes resulting from subduction between the Nazca and South American Plates (Herd et al. 1981;Audemard et al. 2005;Verdugo and González 2015;Candia et al. 2017). Even though Colombia has multiple cities with deposits of potentially liquefiable soils, until now, the country has lacked a seismic monitoring center that allows the study of liquefaction under the local seismogenic conditions.
Cali is located in the southwestern region of Colombia and is the third-largest city of the country with a population of about 2.3 million. The city is located in a high seismic hazard zone and built, to a large extent, on the sand deposits of the alluvial plain of the Cauca River, the second largest river in Colombia (Ingeominas and DAGMA 2005;Arcila et al. 2020). As a result, liquefaction has the potential to affect almost 50% of the city's urban area and about 70% of the critical public infrastructure (Alcaldía de Santiago de Cali 2016). Among the critical infrastructure at risk are two drinking water treatment plants, located in the Aguablanca District which houses more than 600,000 people, most of them belonging to low socio-economic classes. The two treatment plants in the Aguablanca District supply about 75% of the city's drinking water (Gutiérrez et al. 2016).
For the above, the School of Civil Engineering and Geomatics at Universidad del Valle, with the support of the city's public water utility company (EMCALI), decided to implement a seismic monitoring center to evaluate the behavior of soil deposits with potentially liquefiable layers in the Aguablanca District (Cali). Data collected at the center may also be useful tinform hazard, vulnerability, and risk studies for the water treatment plants and, in general, for the areas of Cali that may be affected by liquefaction. In addition, this monitoring center, which will be the first of its kind in Latin America (according to information gathered), will provide field data of scientific and engineering interest on the response of potentially liquefiable layers under subduction earthquakes.
This paper presents the studies, design, and implementation of the seismic monitoring center for potentially liquefiable soils in the Aguablanca District in Cali. First, the location and geotechnical characteristics of the study area, including previous studies on liquefaction evaluation, are discussed. Later, the behavior of two soil deposits with potentially liquefiable layers monitored by the University of California at Santa Barbara (Wildlife Liquefaction Array, WLA, and Garner Valley Downhole Array, GVDA) is analyzed, to establish the requirements for the adequate design of the monitoring center in the Aguablanca District. For this purpose, acceleration records at different depths and piezometric records of potentially liquefiable layers from more than 130 earthquakes are analyzed. Finally, the design and implementation (of the first stage) of the monitoring center are presented, based on the geotechnical and dynamic characterization of the study area and the analysis of the reference monitoring centers. To this end, the technical considerations for the location of the ground motion sensors and piezometers, the data acquisition and storage software validation, and the analysis of the earthquakes recorded during the validation stage are discussed.

Location and geotechnical characteristics of the study area
The seismic monitoring center is set in Cali, in southwestern Colombia. This is a region of high seismic hazard, due to its proximity to the subduction zone of the Nazca Plate with the South American Plate (Ingeominas 2005) . The monitoring center is inside the Puerto Mallarino Water Treatment Plant, which is located on the banks of the Cauca River in the Aguablanca District (Fig. 1). According to Cali's seismic microzonation study, the flood plain of the Cauca River is dominated by potentially liquefiable sand deposits (Ingeominas 2005). These findings have been verified by previous studies at Universidad del Valle (e.g., Sandoval-Vallejo et al. 2013;Arango-Serna et al. 2021).
The study area was explored through nine standard penetration tests (SPT) (ASTM International 2018), six piezocone penetration tests (CPTu) (ASTM International 2020), one deep borehole (40-m wash drilling), one trial pit, and one up-hole test. In addition, ambient noise records were taken at seven different locations of the treatment plant, and the seismic records of the seismic station RAC10 of the Cali Accelerometer Network were analyzed. The assessment of liquefaction through consolidated undrained cyclic triaxial tests conducted in our previous studies on sands extracted from the treatment plant was also considered (Sandoval-Vallejo et al. 2013). Figure 1 shows the locations of the tested sites, of the RAC10 station, and of the sample extraction site of the cyclic triaxial study (site P09).

Fig. 1
Location of the monitoring center and sites where SPT, CPTu, and ambient noise records were made for the geotechnical and dynamic characterization of the study area. RAC10 corresponds to the accelerograph station of the treatment plant Figure 2b shows the typical stratigraphy of the study area. The shallow layer corresponds to a silt layer of high plasticity (MH) and rigid consistency, with an average overconsolidation ratio (OCR) equal to 3.7, undrained shear strength (s u ) greater than 40 kPa, liquid limit (LL) equal to 53% ± 10%, and thickness of 5.4 m ± 1.4 m. This is followed by a layer of potentially liquefiable silty sand (SM), with an average relative density of 42%, a peak effective friction angle (φ′ peak ) equal to 35°, a fines content of 19.5% ± 10.3%, and a thickness of 5.4 m ± 2.0 m. Below the liquefiable layer, there is intercalation of medium and coarse sands of different gradation (SP and SM), with φ′ peak > 33.5°. According to the seismic microzonation study of Cali (Ingeominas 2005), the bedrock is located at a depth of approximately 1000 m. Figure 2a shows a typical shear wave velocity profile (V s ) obtained by the up-hole test and the site fundamental frequency (F g ) as a function of weighted average shear velocity and soil deposit depth. The V s profile shows different stiffness changes (impedance) that coincide with the stratigraphic changes of the soil deposit. The most important changes presented are at a depth of 14 m (F g ≈2.6 Hz) and 22 m (F g ≈2.2 Hz). It is also observed that the soil deposit is composed of soft layers up to 14 m, stiff soils between 14 and 22 m, and very stiff soils, with V s > 400 m/s below 22 m. According to the seismic microzonation study (Ingeominas 2005) and the analysis of ambient noise and earthquake records of the study area, the fundamental frequency of the soil is between 0.6 and 0.7 Hz.
The liquefaction potential evaluation in the water treatment plant was performed using CPTu from the methodology proposed by Boulanger and Idriss (Boulanger and Idriss 2014). For this purpose, the critical surface seismic hazard condition for the study area was assumed, according to the microzonation study of the city (Ingeominas 2005). From this study, in the floodplain of the Cauca River, the expected peak ground acceleration (PGA) is equal to 265 cm/s 2 (≈0.27 g) for an earthquake of magnitude (M w ) equal to 7.5. Table 1 shows the summary of the evaluation of the liquefaction potential, where h Liq is the thickness of the liquefiable layer, q c1Ncs is the average at the liquefiable layer of the equivalent penetration resistance of CPT in clean sand, Dr ave is the average relative density of the liquefiable layer, FS min is the minimum factor of safety at the liquefiable layer, FS max is the maximum factor of safety at

Fig. 3
Evaluation of liquefaction potential in site P06 using CPTu, for an earthquake with PGA = 265 cm/s 2 (≈0.27 g) and M w = 7.5 undrained cyclic triaxial tests, for a sample consolidated under an effective confining stress (σ' c ) equal to 100 kPa and Dr equal to 51% ± 1.3%. Also, the equivalent cyclic resistance curve for the soil in the field was estimated as proposed by Finn et al. (Finn et al. 1971). According to Seed et al. (Seed et al. 1975), for an earthquake of magnitude 7.5, the number of equivalent cycles for liquefaction (N Liq ) is equal to 14. From the equivalent field cyclic resistance curve and considering the typical stratigraphy of the site shown in Fig. 2, the factor of safety against liquefaction would be FS = 0.35, which is consistent with the liquefaction potential evaluation from CPTu (Table 1).

Analysis of reference sites
To understand the behavior of potentially liquefiable layers and determine the minimum instrumentation necessary to effectively study their response to earthquakes, the databases of the Wildlife Liquefaction Array (WLA) and Garner Valley Downhole Array (GVDA), both located in California (USA), were used (http:// nees. ucsb. edu). Both monitoring sites aim to study the behavior of the soil deposits with potentially liquefiable layers (NEES@UCSB 2020).
The selection of these sites as a reference for the implementation of the monitoring center in Cali is largely due to the geotechnical similarities, such as the superficial non-liquefiable layer (crust), the potentially liquefiable layer composed of silty sand, the presence of granular layers below the potentially liquefiable layer (a commonality with GVDA), and the presence of bedrock at great depth (a commonality with WLA).

Site description and data processing
The WLA soil deposit consists of a shallow layer of silty clay down to a depth of 2.7 m, underlain by a liquefiable silty sand layer down to a depth of 6.5 m, followed by silt and clay intercalations (Youd et al. 2004a, c). The bedrock is at a depth of around 1000 m (Youd et al. 2004b). The weighted arithmetic mean shear wave velocity down to a depth of 100 m (which is where the deepest accelerometer is located) is about 250 m/s. The GVDA soil deposit consists of a shallow layer of sandy silt and silty sand down to a depth of 6 m, underlain by a layer of liquefiable silty sand down to a depth of 12 m, followed by a liquefiable sand layer down to a depth of 17 m. Below this layer, there is a coarse granular material (weathered granite) that extends down to a depth of 90 m, where the bedrock is located (Youd et al. 2004b;NEES@UCSB 2020). The weighted arithmetic mean shear wave velocity from the surface to the bedrock is 505 m/s, and the mean shear wave velocity at the bedrock is 1650 m/s. For each monitoring site, seismic records with local magnitude (M L ) greater than 3 were selected from January 2005 to January 2019. For the WLA site, 1699 earthquakes were recorded, where 85% were earthquakes with M L between 3 and 4 and 15% with M L greater than 4. For the GVDA site, the number of recorded earthquakes was 762, where 80% had M L between 3 and 4, and 20% of them had M L greater than 4.
For the analysis, stratified sampling from the earthquake magnitude was performed, as shown in Arango-  Cyclic resistance curve for silty sand of site P09 (FC = 8%) obtained from consolidated and undrained cyclic triaxial tests (σ' c = 100 kPa and D r = 50%). The black line with squares is the cyclic resistance obtained from the triaxial tests, and the dotted black line with circles is the field cyclic resistance obtained from laboratory results and the correction factor proposed by Finn et al. (Finn et al. 1971). Adapted from Sandoval et al. (Sandoval-Vallejo et al. 2013) pore pressure and initial effective confining stressr u ) greater than 2% and acceleration records at all sensors of interest were also added to the sample. Regarding GVDA, earthquakes with excess pore pressure were neglected because there were only five records with r u > 2% (poor sample). The WLA sample, shown in Fig. 5a, had 42 earthquakes with M L between 3 and 4, 37 earthquakes with M L greater than 4, and 11 earthquakes with r u > 2%. The GVDA sample, shown in Fig. 5b, had 24 earthquakes with M L between 3 and 4 and 21 earthquakes with M L greater than 4.
To study the seismic behavior of potentially liquefiable soils, the maximum shear strain at the top and bottom of the liquefiable layer for each earthquake at WLA was estimated according to Eq. 1, where d(t) is the displacement, z the sensor's depth, s the nearest sensor to the surface, and dh the deepest sensor. For this, the records of accelerometers located at depths of 2.5 m and 5.5 m were used to estimate shear strain at the top of the liquefiable layer, and records at depths of 5.5 m and 7.7 m were used to estimate shear strain at the bottom. The accelerograms were preprocessed following the recommendations proposed by USGS (Converse and Brady 1992) and Boore (2005), which were used in similar studies (Chandra et al. 2015(Chandra et al. , 2016Bonilla et al. 2017). For each acceleration record, the mean and trend of the data were removed, a 5% Tukey tapering function was applied, and zero padding was used as proposed by Boore (2005). Finally, a third-order Butterworth filter was applied between 0.2 and 25 Hz. The displacements were then obtained by the double integration of the accelerograms by the trapezoidal method.
To evaluate the amplification of earthquakes from the bedrock to the surface, energy reduction profiles (r b ) were calculated as proposed by Kayen and Mitchell (1999), where r b is the ratio of the Arias intensity (AI) (Arias 1970) at a given depth and the Arias intensity at surface according to Eq. 2 and Eq. 3. The energy reduction profiles obtained were then compared with the theoretical r b proposed by Kayen and Mitchell (1998), as shown in Eq. 4.

Behavior of the liquefiable layer
Analyses of the information obtained for the potentially liquefiable layer at WLA were conducted. More specifically, as shown below, the relationship between excess pore pressure and shear strain, the shear strain as a function of depth, and the excess pore pressure generation as a function of time and depth were investigated.  Figure 6 shows the relationship between the peak excess pore pressure ratio (ru) and the maximum shear strain at the liquefiable layer (γ max ) in the EW direction of WLA (in the NS direction, a similar behavior was observed). According to the results, excess pore pressure is generated when the shear strain of the liquefiable layer is greater than 0.008% (dotted line in the figure). Beyond this threshold, r u and γ max correlate according to the black line (trendline) with the equation shown in the figure, which has a coefficient of determination (R 2 ) equal to 0.82. From the trendline, liquefaction triggering (r u ≈100%) in WLA would occur when the shear strain of the potentially liquefiable layer exceeds 0.2%. Figure 7 shows the maximum shear strain (γ max ) at the top and bottom of the liquefiable layer in the EW direction of WLA. As shown in the figure, the WLA deposit is more susceptible to large shear strains at the bottom of the liquefiable layer during an earthquake, as most of the points (84%) had a maximum shear strain at the bottom part of the liquefiable layer, i.e., they were located above the 45° line in the figure. It must also be stated that the earthquakes that showed the greatest shear strain in the upper part of the liquefiable layer were those with r u > 10%. Figure 8a shows the excess pore pressure profile (isochrones) recorded at WLA during one of the Brawley earthquake swarm events (2012 August 27). As shown in the figure, the peak r u is presented at the top of the liquefiable layer, and it was equal to 45% (peak r u is assumed to be the average measured by piezometers at the top of the liquefiable layer). It is also seen that at the depths of 3.3 m and 4.4 m, the excess pore pressure buildup occurs at a slower rate than at other depths in the potentially liquefiable layer. This behavior was also evidenced in other earthquakes with excess pore pressure generation and during the 1987 Superstition Hills earthquake (r u = 100%) recorded at the old WLA monitoring center (80-m upstream from the study site), as shown in Fig. 8b.

Amplification of seismic waves
The energy reduction profiles (r b ) as a function of the earthquake magnitude obtained for GVDA and WLA are shown in Fig. 9 and Fig. 10, respectively. In the figures, the gray lines correspond to the r b profile of each earthquake, the black line corresponds to the average r b profile, the dotted black lines to the standard deviation, and the red lines to the theoretical r b profile proposed by Kayen and Mitchell (1998).

Fig. 6
Correlation between peak excess pore pressure ratio (r u ) and maximum shear strain (γ max ) at the potentially liquefiable layer for WLA in EW direction At both sites, the greatest amplification of seismic wave energy occurs in the shallow layers (change of gradient in the r b curve). This surface amplification occurs when r b exceeds a threshold close to 0.25. In GVDA, this occurs at a depth of approximately 20 m regardless of the magnitude of the earthquake (Fig. 9). That is, the greatest energy amplification of seismic waves always occurs in the shallower 20 m. In contrast, in WLA, the greater the magnitude of the earthquake, the greater the depth at which r b exceeds the 0.25 threshold (Fig. 10). This coincides with the formulation proposed by Kayen and Mitchell (Eq. 4) where r b is a function of depth and magnitude, e.g., the greatest energy amplification in earthquakes with In GVDA at a depth of 22 m, there is 12% ± 2% of the Arias intensity measured at the surface and 150 m (bedrock) 5% ± 1%. This means that in GVDA, the amplification of seismic wave energy between bedrock and 22 m is small compared to the amplification from 22 m to surface. The seismic wave energy was amplified about 20 times between bedrock and surface while between bedrock and 22 m only 2.5 times. This behavior confirms that the greatest amplification of seismic waves occurs in the shallow layers. This analysis was not performed for WLA because there was no bedrock sensor.  To evaluate the influence of the depth of the reference sensor in the study of wave amplification, the average of the surface-to-bedrock spectral ratio of the horizontal components of the earthquake (commonly known as standard spectral ratio -SSR) was calculated for different depths of the reference sensor. Figure 11 shows the results obtained for the NS direction of GVDA (left) and WLA (right) (in the EW direction, a similar behavior was observed). In the figure, each line corresponds to the average SSR obtained for each reference sensor depth. For example, the gray line in Fig. 11b corresponds to the average SSR obtained between the surface and 30-m-deep records at WLA. According to previous studies, the fundamental frequency of the sites is equal to 1.7 Hz in GVDA and 0.9 Hz in WLA (Arango-Serna et al. 2021). As seen in the figures, the peak corresponding to the fundamental frequency appears in GVDA when the reference sensor is deeper than 22 m while in WLA when the reference sensor is deeper than 30 m. Also, the amplitude of these peaks is very low compared to the amplitude recorded when the reference sensor is in the bedrock (GVDA case in Fig. 11a), which makes difficult their visual identification.

Key aspects for the design of the monitoring center
The results obtained by correlating the peak excess pore pressure ratio with the maximum shear strain of the liquefiable layer (Fig. 6) show the high codependence of these two parameters during the occurrence of an earthquake. As shown in Fig. 6, the excess pore pressure buildup starts after a shear strain of 0.008%, which is consistent with the shear strain threshold proposed by Dobry and Abdoun of 0.01% (Dobry et al. 1982, and previous forced vibration tests performed in WLA (Cox 2006). Also, the shear strain for liquefaction triggering (r u ≈100%) of 0.2% obtained for WLA is in agreement with that obtained in other recent studies (e.g., El-Sekelly et al. 2017;Dobry and Abdoun 2017). These results show that the seismic response of the potentially liquefiable layer can be studied as a function of excess pore pressure and/or shear strain (both parameters can be used as seismic demand parameters in performance-based earthquake engineering studies) (Kramer and Mitchell 2006;Dashti and Karimi 2017). Thus, it is necessary for the design of seismic monitoring centers in soil deposits with liquefiable layers to combine ground motion sensors and piezometers to study the behavior of the potentially liquefiable layer. The results obtained for the shear strain of the liquefiable layer as a function of depth show that the maximum strain occurs in the lower part of the liquefiable layer when the earthquake does not generate excess pore pressure, or they are very low (for WLA when r u ≤ 10%). This behavior was unexpected because, in the bottom part of the liquefiable layer, the confining stress is lower. In contrast, when r u > 10%, the maximum shear strain occurs at the top of the liquefiable layer. This is because as excess pore pressure increases, the soil loses stiffness resulting in an increase in shear strain (e.g., Zeghal and Elgamal 1994;Holzer and Youd 2007;Abdoun et al. 2013), as evidenced in Fig. 6. In summary, the maximum shear strain of the liquefiable layer may occur at the top when there is considerable excess pore pressure or at the bottom when they are very low or are not generated. Therefore, to study the maximum shear strain of the potentially liquefiable layer in any scenario, it is necessary to have ground motion sensors at the top, middle, and bottom of the layer.
The r u isochrones (Fig. 8) show that at the beginning of excess pore pressure buildup, the maximum r u may be developed during the first seconds at the bottom of the liquefiable layer and then migrate to the top. They also show that when the excess pore pressure during the earthquake is large (r u > 10%), the maximum r u values of the earthquake are seen in the upper part of the liquefiable layer. These findings are in agreement with the behavior seen during the 1987 Superstition Hills earthquake, as well as in several laboratory tests, where liquefaction (r u ≈100%) occurred first in the upper part of the liquefiable layer (where there are lower confining stresses) and then propagated to the middle and lower part (Zeghal and Elgamal 1994;Youd et al. 2004d;Dobry et al. 2011). Thus, like the ground motion sensors, it is also necessary to have piezometers at the top and bottom of the potentially liquefiable layer, to consider different scenarios of excess pore pressure generation.
Another important aspect observed in Fig. 8a is that at a depth of 4.5 m (not at the base of the potentially liquefiable layer at 6.5 m, where the confining stress is higher), r u values are the smallest and increase at a slower rate with time compared to other depths. This same effect has been evidenced in large-scale shake tests using a laminar box and in centrifuge tests (Gonzalez 2008;Dobry et al. 2011;Abdoun et al. 2013). Previous researchers attributed this effect to density changes in the sand, which resulted in areas less susceptible to liquefaction. In WLA, according to the geotechnical characterization (e.g., borehole P2, P5, and P6), the sand layer is denser at this depth (higher number of SPT blows) so the r u increases at a slower rate. Accordingly, it is important to detect the densest zones of the liquefiable layer and allocate them a lower priority for the installation of piezometers, since they are zones less susceptible to liquefaction and do not represent the critical conditions of the site.
The r b profiles ( Fig. 9 and Fig. 10) show that the greatest energy amplification of seismic waves occurs in the shallow layers, starting at the depth where r b exceeds a threshold close to 0.25. In GVDA, this takes place at a depth of 20 m. In WLA, the depth at which this occurs varies according to the magnitude of the earthquake, with a depth of 35 m being the most critical condition for earthquakes with a magnitude greater than 5. Due to the above, locating the reference sensor at a depth where r b ≤ 0.25 will allow studying the greatest energy amplification of the seismic waves. As a starting point for the design of monitoring centers, a minimum depth of 35 m can be used, which was the most critical in the sites analyzed.
The SSR obtained in GVDA and WLA by varying the depth of the reference sensor (Fig. 11) shows that below 22 m in GVDA, and 30 m in WLA, it is possible to identify the frequency corresponding to the first mode of ground vibration. For GVDA, it is observed that, although the fundamental frequency is identified with a sensor at a depth of 22 m, the amplification factor is lower than that obtained by SSR between the surface and bedrock records. This implies that there is a loss of information in the first vibration mode when the reference sensor is not at the bedrock. Consequently, the reference sensor below 35 m, as recommended in the previous paragraph, may be a valid alternative in sites where the bedrock is at great depth, but it does not completely replace positioning the sensor at the bedrock or outcrop rock.

Design considerations
The site P06 at Fig. 1 was selected for the location of the seismic monitoring center. This area is potentially liquefiable, protected by the flood embankment, and not affected by subsurface structures or expansion projects. The stratigraphy is composed of a shallow layer of silts and clays of high plasticity of approximately 5.5 m thick, followed by a layer of potentially liquefiable silty sand of approximately 5.5 m thick (Fig. 3). Below this layer, there are intercalations of medium and coarse sands with gravels of high stiffness.
For the design of the monitoring center, it was necessary to consider different technical aspects based on geotechnical exploration, scientific literature, and data obtained from the analysis of seismic records at the reference monitoring centers (i.e., WLA and GVDA). The main considerations for the design of the monitoring center are presented below.

Piezometers
To investigate the excess pore pressure during and after an earthquake, at least one piezometer located at the top of the liquefiable layer and sufficiently separated from the silt-sand interface is necessary. For this research, depths between 6.0 and 7.0 m were considered adequate. To study the instant at which the excess pore pressure starts during an earthquake, the installation of a piezometer at the lower of the liquefiable layer is also important. To that end, depths between 9.5 and 10.5 m were considered adequate. As discussed in the previous section, piezometers should be avoided in dense zones within the liquefiable layer, as they are less propense to exert high excess pore pressure values. For the case study, and as shown in Fig. 3 (evaluation of liquefaction potential in site P06 using CPTu), between depths of 8.0 and 9.5 m, the potentially liquefiable layer has a higher density (D r ≈60%); thus, piezometers in this depth should be avoided.
The triggering of excess pore pressure and its propagation through the potentially liquefiable layer could happen in fractions of a second (Youd and Holzer 1994). Therefore, for the recording of pressure buildup during an earthquake, the use of strain-gauge semiconductor piezometers, which have a minimum data acquisition rate of 50 readings per second, is considered adequate. For periodic calibration of the piezometers, it is also necessary to make periodic readings of the water table level at the area. For this purpose, it is possible to use a direct measurement method in inspection wells near the monitoring center with electronic reading units.

Ground motion sensors
Depending on the magnitude of the earthquake and site drainage conditions, the maximum shear strain of the potentially liquefiable layer may occur at different depths in the layer. To analyze any possible scenario, ground motion sensors are required in the middle of the liquefiable layer, and immediately above and below of the liquefiable layer, as done in the reference projects (WLA and GVDA) (NEES@UCSB 2020). For the case study, it was decided to install the sensors between depths of 4 m and 5 m (above the liquefiable layer), 7 m and 8 m (middle part of the liquefiable layer), and 12 m and 13 m (below the liquefiable layer).
To investigate local site effects and for complementary analyses using numerical modeling, it is necessary to have surface and bedrock earthquake records that allow analyzing the amplification of waves and vibration modes of the soil deposit (Yoshida 2015). However, the depth of the alluvial deposit of the Cauca River at the study area (approximately 1 km) makes the installation of a reference sensor at the bedrock impractical. Because of this, and as discussed in the previous section, locating a reference sensor below a depth of 35 m (where r b ≤ 0.25) and another at the surface is considered adequate. It must be highlighted however that, given the loss of information that may occur for the fundamental vibration mode of the site, it is important to complement this setup with a sensor on a nearby rock outcrop in the future.

Configuration of the monitoring center
The configuration of the seismic monitoring center, as determined by the stratigraphy and design conditions described above, is summarized in Fig. 12 and Table 2. The seismic monitoring center is planned to be installed in two stages. The first stage, already completed, included the following: (i) four wells for the borehole motion sensors, the slab for the surface motion sensor, the protection huts, and the data and electrical power networks; (ii) the installation of the operations center; (iii) the installation of a surface ground motion sensor (01 in Fig. 12) and a borehole ground motion sensor at 40 m (00 in Fig. 12); (iv) validation of the correct functioning of the electrical networks, data network, data acquisition, and storage system; (v) collection of water table level data and installation of one piezometer on top of the potentially liquefiable layer at a depth of 6.2 m (P2 in Fig. 12); and (vi) preliminary analysis of the seismic records obtained.

Constructions and installations
For the construction of the deep wells for the motion sensors (00, 02, 03, and 04 in Fig. 12), four 8″ diameter boreholes were conducted by rotation using bentonite slurry to stabilize the borehole. A 4″ diameter PVC pipe was then installed, and the space between the excavation and the pipe was backfilled with cement slurry. The 4 m (04), 7 m (03), and 40 m (00) deep wells were equipped with 70 × 70 × 100 cm and 10-cm-thick reinforced concrete inspection boxes (Fig. 13c), which have two 2″ PVC pipes that allow connecting the data and power supply of the equipment to the surface sensor instrumentation hut (where the monitoring center acquisition system will be installed). Finally, the inspection boxes were protected with steel huts (Fig. 13b). The instrumentation hut of the surface sensor (01) and the 13-m-deep well (02) had a box foundation of 90 × 90 × 90 cm, composed of a 10-cm-thick reinforced concrete perimeter wall, and compacted selected material inside. A 20-cm-thick reinforced concrete slab was built on top of the foundation, embedded 10 cm into the soil (Fig. 13d). The 2″ connection pipes from the other huts are connected to the slab. On top of the foundation slab, an 80-cm-tall steel hut was installed, with a gable roof and airflow slots to prevent moisture condensation inside the hut (Fig. 13b). In addition, a 110-V socket with a groundbed was installed to supply electrical power to the Triaxial sensor for reference seismic records. Monitoring well at a depth of 40 m and protection hut. Waterproof wideband borehole motion sensor 01 Triaxial sensor for seismic recording on the surface. Foundation slab and waterproof protection hut. Wideband motion sensor with acquisition system located on foundation slab. Minimum acquisition recommendations: 24 bits, sampling frequency 100 Hz 02 Triaxial sensor for seismic recording at the base of the liquefiable layer. Monitoring well at a depth of 13 m embedded in the foundation slab of the surface sensor. Waterproof wideband borehole motion sensor 03 Triaxial sensor for seismic recordings at the liquefiable layer. Monitoring well at a depth of 7-m-depth protection hut. Waterproof wideband borehole motion sensor 04 Triaxial sensor for seismic recordings at the upper part of the liquefiable layer. Monitoring well at a depth of 4 m and protection hut. Waterproof wideband borehole motion sensor P1 Sensor for measuring excess pore pressure at the base of the liquefiable layer. Strain gauge semiconductor piezometer located between depths of 9.5 m and 10.5 m. Requirements: 0-5 V; maximum pressure 250 kPa P2 Sensor for measuring excess pore pressure at the base of the liquefiable layer. Strain gauge semiconductor piezometer located between depths of 6.0 m and 7.0 m. Requirements: 0-5 V; maximum pressure 250 kPa A Operation center of the monitoring center at the CINARA foundation. Laptop computer. Requirements: RAM equal or greater than 8 GB; storage equal or greater than 1 TB; processor speed equal or greater than 2 GHz B Electrical connection for monitoring center equipment. Socket and groundbed. Requirements: AC: 120 V C Light pole. Electrical connection for the monitoring center. Installation of a single-pole breaker. Requirements: AC: 110 V; 10 A D Electrical distribution panel of the Puerto Mallarino treatment plant grit chambers system. Adjacent to this hut, there is a transition box (B in Fig. 12 and Fig. 13), where the power supply for the monitoring center arrives and the data cable (Ethernet CAT5E) transfers the information to the operation center (A in Fig. 12 and Fig. 13). A surface triaxial ground motion sensor Lennartz LE-3D (hut 01) and a triaxial borehole ground motion sensor Lennartz LE-3D/BH (40-m-deep borehole, hut 00) were installed according to the manufacturer's recommendations, as shown in Fig. 13c and Fig. 13d. The borehole sensor tolerates an inclination with respect to the vertical of up to 5°. A Geokon model 3400 piezometer was also installed at 6.25 m depth (P2), in the middle of a layer of clean fine sand approximately 40 cm thick. The piezometer borehole was backfilled, as recommended by the manufacturer. The motion sensors and piezometer were connected to the National Instruments NI-9234 acquisition modules, which in turn were connected to a National Instruments cDAQ-9174 chassis that allows continuous and simultaneous data acquisition. The acquisition system was connected to the electrical supply and the operation center through the data network.
The operation center is equipped with a computer that runs continuously and is connected to the electrical and Internet network of the plant. This allows constant cloud storage of data and remote management of the system. Data acquisition and storage are performed using the SISMOS software developed in our previous studies (Gallo et al. 2017), which resamples the information acquired in real time, stores this information in a digital.bin format that occupies little space in hard disk memory, visualizes the information in real time, removes the data from the RAM after it is stored on the hard disk, and uses the STA/LTA seismic detection algorithm in real time.

Recorded earthquakes
From October 2019 to February 2021, 35 seismic events with hypocentral distances between 75 and 572 km were recorded at the monitoring center. The SISMOS data acquisition software was able to detect and store 21 of these earthquakes completely. The remaining 14 were partially detected and stored from the arrival of the S-waves, i.e., the first few seconds  (01) of these earthquakes were not recorded by the algorithm due to the low amplitude of the waves. In these cases, the records were obtained manually from the backup data. Also, an angular correction was applied to the seismic signals obtained by the motion sensor at 40 m depth following the methodology proposed by Yamazaki et al. (1992), which consists of finding an angle at which the horizontal components of the borehole sensor present the maximum cross-correlation with respect to the horizontal components of the surface sensor. For the recorded earthquakes, the angle of correction between axes corresponded to 244.7° ± 1.4° (when more data is acquired, this angular correction will be implemented directly in the SISMOS acquisition software). Figure 14 shows the location and type of seismogenic sources in the country, according to the Colombian Geological Survey (Arcila et al. 2020).
Seismogenic sources are grouped into four categories: shallow crustal, Pacific subduction zone, Benioff zone, and Bucaramanga seismic nest (Taboada et al. 1998;Arcila et al. 2020). Shallow crustal (surface) earthquakes occur within the continental plate and reach depths of up to 70 km. The earthquakes of the Pacific subduction zone (interplate subduction) occur at the convergence limits of the Nazca and South American plates and reach depths of up to 50 km. The earthquakes of the Benioff zone (intraplate subduction) occur due to the subduction of the Nazca plate in South America and have depths above the interplate earthquakes and reach up to 200 km depth. The Bucaramanga seismic nest earthquakes are earthquakes generated in a region at the north of Colombia that presents a high volume of intense seismic activity, which is persistent in time and isolated from nearby seismic activity. These earthquakes commonly have depths between 140 and 200 km (Prieto et al. 2012).
The summary according to the seismogenic source of the 35 earthquakes recorded by the monitoring center was as follows: 18 shallow crustal earthquakes (C), 7 earthquakes at the Bucaramanga seismic nest (N), 6 earthquakes at the Pacific subduction zone (S(I)), and 4 earthquakes at the Benioff zone were recorded (S(B)), as shown in Fig. 14 and Table 3. Ten of the 18 shallow crustal earthquakes are part of a seismic swarm in which there was seismic activity for more than 60 continuous hours, between December 24 and 26, 2019, in Mesetas (earthquakes no.3 to no.12). Assuming this seismic swarm as a single event, it can be concluded that most of the recorded earthquakes correspond to subduction earthquakes, which are one of the main justifications for the construction of this monitoring center.
The shallow crustal earthquakes (C) had magnitudes (M L ) between 3.6 and 6.0 and depths between 5 and 30 km. The earthquakes at the Bucaramanga seismic nest (N) had magnitudes (M L ) between 3.7 and 4.9 and depths between 146 and 164 km. The earthquakes of the Pacific subduction zone (S(I)) had magnitudes (M L ) between 4.0 and 4.8 and depths between 8 and 33 km. The earthquakes at the Benioff zone (S(B)) had magnitudes (M L ) between 4.1 and 5.1 and depths between 37 and 158 km. The peak ground acceleration (PGA) of the recorded earthquakes was between 0.125 and 5.887 cm/s 2 , the peak ground velocity (PGV) between 0.003 and 0.317 cm/s, and the peak ground displacement (PGD) between 0.002 and 0.592 mm. Earthquakes no. 3 and no. 4 in Mesetas (shallow The reduction factor (r b -Eq. 3) at a depth of 40 m was equal to 0.13 ± 0.07, which means that the reference sensor is located below the depth where the greatest energy amplification of the seismic waves begins, as analyzed in previous sections (r b ≤ 0.25). Figure 15 shows the standard spectral ratio (SSR) between the earthquakes recorded at a depth of 0 m and 40 m. The figure shows amplitude peaks at frequencies at approximately 1.8 Hz, 2.4 Hz, 3.2 Hz, 3.9 Hz, and 5.1 Hz, without peaks below 1.5 Hz, including for the fundamental frequency of the site (between 0.6 and 0.7 Hz).

Water table level
Before the installation of the piezometer between depths 6.0 m and 7.0 m (P2 in Fig. 12), it was decided to monitor changes in the water table level at the monitoring center. For this purpose, a water level meter was used to measure the water table level periodically in a well adjacent to the grit chamber (D in Fig. 12) of the treatment plant. Figure 16a shows the water table measurement from October 2019 to November 2020. As seen in the figure, the water table level presented a strong variation during the recording time, ranging from 2.5 to 5.2 m, with an average of 4.3 m ± 0.6 m. As shown in the geotechnical characterization, the potentially liquefiable silty sand layer in this site must be at an approximate depth of 5.5 m; thus, it can be concluded that the layer will always be saturated. The results also guarantee that the piezometers planned to be installed will remain underwater. The P2 piezometer (Fig. 12) was installed in September 2021 at the top of the potentially liquefiable layer, at a depth of 6.25 m. Before installation, the piezometer was subjected to several calibrations and data acquisition tests. To validate the correct performance of the piezometer, the water table level measured directly was compared with that estimated from the hydrostatic pressure recorded by the piezometer by 2 months. As shown in Fig. 16b, both readings are consistent. The average difference between the readings was equal to 9 cm (less than 1 kPa) and does not affect the calculation of effective stresses and, therefore, of the excess pore pressure ratio. To date, excess pore pressures have not been produced by earthquakes. Fig. 15 Standard spectral ratio (SSR) between surface and 40-m-depth earthquake records. The SSR is shown on the left for the east-west (EW) direction and on the right for the north-south (NS) direction

Discussion
To select the location of the ground motion sensors and piezometers at the potentially liquefiable layer, the result of the CPTu performed during the characterization of the study area was fundamental. Because CPTu allows acquiring soil data almost continuously, it was possible to accurately establish the limits of the potentially liquefiable layer (an aspect that was verified during the drilling of the wells). It was also possible to identify a densified zone within the potentially liquefiable layer unsuitable for the installation of the piezometers since it is less susceptible to generating excess pore pressure and liquefaction during seismic events.
The reference sensor was located based on economic and constructive conditions since the bedrock is located at a depth of approximately 1000 m. Based on the analysis of the reference monitoring centers (GVDA and WLA), it was decided to locate the reference sensor in a rigid soil at a depth of 40 m, under the assumptions that at this depth, the energy reduction factor (r b ) would be less than 0.25 and the amplification that the seismic waves have had from the bedrock is small. The results of the earthquakes recorded at the monitoring center showed that the r b at 40 m is less than 0.25.
Although earthquakes at this depth can be used as input signals for theoretical analyses, e.g., numerical models, they should be used with caution. As observed in WLA and GVDA, there may be a loss of information at several frequencies of interest. Indeed, and as observed in the analysis of seismic records from our monitoring center, no amplification of frequencies below 1.5 Hz has been observed between the surface and 40 m depth. Because of this, it is necessary to plan the installation of a sensor in a nearby rock outcrop or to drill a deeper well to study the amplification of waves in a larger frequency range.
The acquisition of seismic records shows that SIS-MOS software and the hardware of the monitoring center are working adequately. Acquisition, resampling, data storage, and uploading to the cloud worked as expected. However, the detection and partial storage of 14 of the 35 recorded earthquakes show that it is necessary to calibrate the parameters of the algorithm for detecting the onset of the earthquake (STA/ LTA algorithm) using the recorded earthquakes.
The monitoring center was able to record earthquakes from the main seismogenic sources in Colombia, which will allow the analysis of the different seismic hazard scenarios at the study area. Among the different types of earthquakes acquired, those of the subduction zone stands out, as this is the same seismogenic source that has generated liquefaction in other Latin American countries such as Peru (Audemard et al. 2005) and Chile (Verdugo and González 2015;Candia et al. 2017). This type of earthquake is of high research relevance since they represent a seismic scenario different from the shallow crustal earthquakes commonly used to study liquefaction.
The water table level measurement over 1 year showed that:the projected piezometers will remain underwater; the potentially liquefiable layer will remain saturated during the year (susceptible to liquefaction); and the water table level shows abrupt changes with no trend over time (e.g., November 19 WT = 3.3 m; November 2020 WT = 5.2 m). Validation tests of the P2 piezometer, which was installed to measure excess pore pressure during earthquakes, show that it also allows estimating the water table and, therefore, the effective soil stresses. Since these readings are continuous in time, the data acquired will allow studying the changes in the water table level of the site as a function of the Cauca River level, rainfall intensity, among others.

Conclusions
In this paper, the process for the design and implementation of a seismic monitoring center in Cali (Colombia), whose main objective is to study the seismic response of soil deposits with potentially liquefiable layers in the east of the city, and under the local seismogenic conditions, was presented. It must be noted that, according to the information gathered, this will be the first seismic monitoring center (including ground motion sensors and piezometers) designed to study the behavior of a liquefiable soil deposit in Latin America. To establish the required instrumentation for the monitoring center, the behavior of two soil deposits with potentially liquefiable layers monitored by the University of California at Santa Barbara was analyzed (Wildlife Liquefaction Array -WLA and Garner Valley Downhole Array -GVDA). Then, based on the analyses performed at WLA and GVDA, and on the geotechnical and dynamic characterization of the study area, a seismic monitoring center was designed with the capability to study the seismic response of the soil deposit and the behavior of the potentially liquefiable layer. Finally, the construction and start-up of the first stage of the monitoring center were presented. The insights and lessons shared herein will hopefully prove valuable to other researchers and scientists involved in the design, construction, and operation of monitoring centers in other parts of the world.
The main aspects that were considered in the design of the monitoring center from the analysis of the GVDA and WLA seismic records are as follows: (i) To adequately study the behavior of the potentially liquefiable layer, the monitoring of pore pressure and shear strain during the occurrence of the earthquake is required due to the codependence of these two parameters. (ii) Pore pressure should be recorded at the top and bottom of the liquefiable layer because the depth where the peak excess pore pressure ratio (r u ) changes during the earthquake. (iii) The shear strain should be recorded at the top and bottom of the potentially liquefiable layer because the maximum shear strain occurs at the top when the earthquake generates r u > 10% and at the bottom in other cases. (iv) There may be denser zones within the potentially liquefiable layer where the installation of piezometers should be avoided since these zones are less susceptible to generating excess pore pressure and therefore liquefaction. (v) The reference sensor should be below the depth where the greatest energy amplification of seismic waves starts, which occurs in the shallow layers from the depth where the energy reduction factor (r b ) is greater than 0.25. (vi) Although the reference sensor is below this depth, all the information that can be recorded in the bedrock at some frequencies of interest may be missed.
The geotechnical and dynamic characterization of the study area showed the susceptibility of the soil deposit to liquefaction during an earthquake.
According to the analyses performed, the potentially liquefiable layer starts below 5 m depth, has an average thickness of 5.4 m, remains saturated during the year, and has relative densities low enough to present liquefaction during earthquakes with PGA > 98 cm/ s 2 (> 0.1 g) and M w = 7.5. This condition makes liquefaction one of the main hazards for the treatment plant and the more than 600,000 inhabitants living in the surrounding areas. Because of this, and given the local seismogenic conditions, it is expected that the monitoring center will provide data that will contribute to scientific research on the seismic behavior of soil deposits with potentially liquefiable layers and that will allow hazard and vulnerability studies for the treatment plant and, in general, for the urbanized areas on the alluvial plain of the Cauca River.
The design and implementation of the seismic monitoring center implied evaluating engineering, economic, administrative, and social aspects. Consequently, aspects like the security situation of the area and the depth of the bedrock influenced the location and final design of the monitoring center. The implementation of the first stage of the monitoring center was able to demonstrate the good performance of the adopted software for data acquisition and storage. It also showed the necessity to continuously calibrate the parameters of the earthquake detection algorithm so that more earthquakes may be detected. Up to today, the seismic monitoring center has been able to detect earthquakes satisfactorily from the different seismogenic sources.
For the final implementation of the monitoring center, the installation of the piezometer at the base of the potentially liquefiable layer, the ground motion sensors at the potentially liquefiable layer, and the ground motion sensor in a nearby rock outcrop are planned. This stage will also include the development of a web portal to share seismic and piezometric records with the scientific community. Future research developments around the monitoring center include the development of an artificial intelligence algorithm for earthquake detection, the construction of a numerical model of seismic response that may be calibrated by artificial intelligence algorithms, and the development and construction of ground motion sensors.