Lunar Seismology: A Data and Instrumentation Review

Several seismic experiments were deployed on the Moon by the astronauts during the Apollo missions. The experiments began in 1969 with Apollo 11, and continued with Apollo 12, 14, 15, 16 and 17. Instruments at Apollo 12, 14, 15, 16 and 17 remained operational until the final transmission in 1977. These remarkable experiments provide a valuable resource. Now is a good time to review this resource, since the InSight mission is returning seismic data from Mars, and seismic missions to the Moon and Europa are in development from different space agencies. We present an overview of the seismic data available from four sets of experiments on the Moon: the Passive Seismic Experiments, the Active Seismic Experiments, the Lunar Seismic Profiling Experiment and the Lunar Surface Gravimeter. For each of these, we outline the instrumentation and the data availability. We show examples of the different types of moonquakes, which are: artificial impacts, meteoroid strikes, shallow quakes (less than 200 km depth) and deep quakes (around 900 km depth). Deep quakes often occur in tight spatial clusters, and their seismic signals can therefore be stacked to improve the signal-to-noise ratio. We provide stacked deep moonquake signals from three independent sources in miniSEED format. We provide an arrival-time catalog compiled from six independent sources, as well as estimates of event time and location where available. We show statistics on the consistency between arrival-time picks from different operators. Moonquakes have a characteristic shape, where the energy rises slowly to a maximum, followed by an even longer decay time. We include a table of the times of arrival of the maximum energy tmax\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{\max}$\end{document} and the coda quality factor Qc\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$Q_{c}$\end{document}. Finally, we outline minimum requirements for future lunar missions to the Moon. These requirements are particularly relevant to future missions which intend to share data with other agencies, and set out a path for an International Lunar Network, which can provide simultaneous multi-station observations on the Moon.

and seismic missions to the Moon and Europa are in development from different space agencies. We present an overview of the seismic data available from four sets of experiments on the Moon: the Passive Seismic Experiments, the Active Seismic Experiments, the Lunar Seismic Profiling Experiment and the Lunar Surface Gravimeter. For each of these, we outline the instrumentation and the data availability.
We show examples of the different types of moonquakes, which are: artificial impacts, meteoroid strikes, shallow quakes (less than 200 km depth) and deep quakes (around 900 km depth). Deep quakes often occur in tight spatial clusters, and their seismic signals can therefore be stacked to improve the signal-to-noise ratio. We provide stacked deep moonquake signals from three independent sources in miniSEED format. We provide an arrival-time catalog compiled from six independent sources, as well as estimates of event time and location where available.
We show statistics on the consistency between arrival-time picks from different operators. Moonquakes have a characteristic shape, where the energy rises slowly to a maximum, followed by an even longer decay time. We include a table of the times of arrival of the maximum energy t max and the coda quality factor Q c .
Finally, we outline minimum requirements for future lunar missions to the Moon. These requirements are particularly relevant to future missions which intend to share data with other agencies, and set out a path for an International Lunar Network, which can provide simultaneous multi-station observations on the Moon.

Introduction
Many seismic experiments were deployed on the Moon by the astronauts during the Apollo missions. These experiments were part of the Apollo Lunar Seismic Experiments Package (ALSEP). The experiments began in 1969 with Apollo 11, and continued with Apollo 12, 14, 15, 16 and 17 ( Fig. 1; Table 1). The seismic instruments included passive seismometers, a gravimeter, and geophones which were deployed in active source experiments, and then later in passive listening mode. Fig. 2 shows the operating periods for each experiment. The passive seismic stations from Apollo 12, 14, 15 and 16 remained operational until the final transmission in 1977.  (Williams et al., 2008). From Table 5 in Wagner et al. (2017).  These remarkable experiments provide a valuable resource. Now is a good time to review this resource, since there is renewed scientific interest in planetary seismology. The Mars InSight mission carries a broadband seismometer and a short-period seismometer, which are detecting marsquakes on the surface of Mars (Lognonné et al., 2019;Banerdt et al., 2020;Giardini et al., 2020;Lognonné et al., 2020  is currently being tested in sites which are analogs for the icy moon Europa (e.g Marusiak et al., 2018;DellaGiustina et al., 2019;Marusiak et al., 2020). Efforts in many countries indicate that an International Lunar Network of seismic stations could be deployed on the Moon by the mid-2020s. In China, CNSA's Chinese Lunar Exploration Program deployed a lunar rover with the Chang'e 3 and Chang'e 4 missions. China is planning Chang'e 5 and 6 as sample return missions (Goh, 2018).
In the USA, a Lunar Geophysical Network is one of the possible candidates for the NASA New Frontiers 5 mission (Committee on the Planetary Science Decadal Survey and Council, 2011;Shearer and Tahu, 2011). The network would deploy at least three stations containing geophysical instruments, and potentially cover the farside of the Moon (Yamada et al., 2011;Mimoun et al., 2012). In Japan, JAXA's SLIM (Smart Lander for Investigating the Moon) is currently under development (JAXA, 2018). Dragonfly is a Titan mission which uses a rotorcraft-lander. It has been selected as NASA's next New Frontiers mission (APL, 2019). There is considerable interest in using seismology to explore the icy moons within our solar system (Vance et al., 2018). Lognonné and Johnson (2015) contains a review of past and future planetary seismology.
It is important that the data from the Apollo experiments can continue to be used in the future. Recent efforts have been made to preserve and document as much of the data as possible, since some of the data remain on digital tapes which are deteriorating in quality. Some tapes may have been permanently lost. The original data from the Apollo experiments were sent to the Principal Investigator (PI) for each experiment. The PIs were responsible for checking the data, and then archiving them. In some cases, especially where problems were discovered with the data, the data were not archived. Some of these data have recently been recovered (Nagihara et al., 2017). Dimech et al. (2017) analyzed thermal moonquakes with recently rediscovered data from Apollo 17. Similarly, Nagihara et al. (2018) recovered 10% of the data missing from a heat flow experiment which ran from 1974 to 1977.
The authors of this paper are members of an international team sponsored by the International Space Science Institute in Bern and in Beijing. The team formed to gather a set of reference data sets and internal structural models of the Moon. This paper reviews the available data, and the companion paper (Garcia et al., 2019) reviews lunar structural models. Within this paper, we also outline minimum requirements for a future International Lunar Network (ILN). If funded, NASA would provide two or more nodes, and other nations would provide additional nodes (Committee on the Planetary Science Decadal Survey and Council, 2011).
These requirements are particularly relevant to future missions which intend to share data with other agencies, and set out a path for simultaneous multi-station observations on the Moon.

Apollo Seismic Instruments
More than 40 years after the termination of the experiments, the Apollo data continue to provide important insights for lunar seismology. The Apollo Lunar Surface Experiment Packages (ALSEPs) were a unique series of in-situ geophysical experiments, which included seismic experiments. No seismic observations have been performed on the Moon since Apollo. The experiments included the Passive Seismic Experiment (PSE), the Active Seismic Experiment (ASE), and the Lunar Surface Profiling Experiment (LSPE). For decades, these data have been used to investigate the internal structure of the Moon (e.g. Nakamura, 1983;Lognonné et al., 2003;Weber et al., 2011;. In addition to these experiments, the Lunar Surface Gravimeter (LSG) also provides some seismic information . In this section, we review the instrumentation.

Passive Seismic Experiments (PSE)
The Passive Seismic Experiments (PSE) were performed at Apollo 11, 12, 14, 15, and 16. Fig. 2 shows the observation period of each station. Apollo 11 functioned for only about 3 weeks. Stations 12, 14, 15 and 16 operated continuously since their deployment and functioned as a seismic network until September 1977, when all the remaining experiments were shut down. More than 13000 seismic events were cataloged using data from the mid-period instruments during the operation of the network (Nakamura et al., 1981). The four stations formed an almost equilateral triangle, with stations 12 and 14 at one corner (Fig. 1). The network covered only a portion of the lunar nearside. This is likely one of the reasons that most of the detected seismic events are from the lunar nearside. Each PSE station was equipped with a 3-component (two horizontal and one vertical) mid-period displacement sensors and a vertical-component short-period (SP). Earlier papers referred to the mid-period seismometer as long-period. We use the designation midperiod to be consistent with the IRIS naming conventions, and to better describe the capabilities of the seismometer.
The mid-period (MP) sensors were feedback displacement transducers (Sutton and Latham, 1964), with a single-pole high-pass output level stabilizer, and an 8pole low-pass output anti-aliasing filter for each. The SP sensor was a standard coilmagnet velocity transducer, also with a single-pole high-pass output level stabilizer and an 8-pole low-pass output anti-aliasing filter. The feedback signals from the MP sensors were recorded as tidal (TD) signal outputs.
The MP sensor had two modes for seismic observation. These were the peaked mode and the flat mode. The peaked mode was the natural response of the seismometer, and the seismometer did not include a feedback filter. The flat mode was designed to be sensitive to a broader range of frequencies, and used a feedback filter in the circuit. Unfortunately, the flat mode was not very stable. Therefore, the seismometers were mainly operated in peaked mode. All of these outputs went through pre-and post-amplifiers before they were fed to the input of the analogto-digital converter for digitization. Table 2 summarizes the periods when the MP seismometer was functioning in flat mode.
The peaked mode of the MP sensor has a peak at about 0.45 Hz while the flat mode has flat response (for displacement) from about 0.1 to 1 Hz.
Although the two horizontal components for the MP sensor were intended to point north and east, they were misaligned for stations S12 and S16. Section S1 in the electronic supplement contains the correct orientations. We provide only the nominal sampling rates for all the seismometers. Small variations in the actual sampling rates were observed at all sites (Nunn et al. (2017) and Knapmeyer-Endrun and Hammer (2015, Supplement)

Flat-Response Mode of the Mid-Period Seismometer
In flat-response mode, the seismometer response A M P F (ω) for acceleration is represented by: where ω is the angular frequency, and K 3 is the amplifier gain of the feedback output.
F a (ω) is the transfer function of the single-pole high-pass filter in the output amplifier, where ω a is the output high pass cut-off angular frequency, and j 2 = −1.
F l (ω) is the transfer function of the 8-pole output low-pass anti-aliasing filter, where ω l is the output low-pass cut-off angular frequency and ω l = 2πf l .
F sf (ω) is the transfer function of the feedback component of the seismometer, K 1 is the gain of the displacement transducer in V/m, and K 2 is the coil-magnet is the transfer function of the demodulator low-pass filter, where ω d is the demodulator low-pass cut-off angular frequency.
S(ω) is the transfer function of the seismometer for acceleration: where f 0 is the resonant frequency of the pendulum and h is the damping constant. F f (ω) is the transfer function of the feedback low-pass filter, where ω f is the feedback low-pass cut-off angular frequency. The parameters for the mid-period seismometer have the following values (Yamada, 2012): To convert the seismometer response to velocity in V/(m/s), we multiply A M P F (ω) by the function s(ω). To convert it to displacement in V/m, we multiply it by the square of s(ω), as follows: The instrument output voltages between -2.5 V and +2.5 V and the digitizer recorded digital units between 0 and 1023. Therefore, we can convert the transfer function from V/m to DU/m by multiplying by 1024 DU/5 V , which is the reciprocal value of the 1-LSB (least significant bit) of the analog-to-digital converter: The transfer function in flat mode is shown in Fig. 3.

Peaked-Response Mode of the Mid-Period Seismometer
The seismometer response during peaked-response mode A LP P (ω) is represented by eliminating the transfer function of the feedback low-pass filter F f (ω) from the equation of A LP P (ω): The transfer function in peaked mode is shown in Fig. 3, and a block diagram which covers both the peaked and flat modes is included in the Electronic Supplement.

Tidal-Response of the Mid-Period Seismometer
The tidal output is the un-amplified feedback signal proportional to the mid-period boom motion (the feedback component of the seismometer F sf (ω), followed by an additional low-pass feedback F f (ω)). This signal potentially gives changes to the gravity field and tidal acceleration, since it has higher sensitivity than the midperiod output at longer periods. It records only once every eight samples of the mid-period instrument, giving a nominal sampling rate of 0.828125 Hz. The flat tidal-mode response in acceleration is: or alternatively: We noticed problems with earlier formulations of the tidal mode. Fig. 4.2 in Teledyne (1968) (reproduced in the Electronic Supplement) does not include a second wire between the filter switch and the feedback resistor (R f b in their diagram). We found a different problem in Fig. 3 in Yamada (2012), which was based on Fig. 2 in Horvath (1979). The tidal output should be connected to the peakedmode output of the switch, and thus to the input of K 2 . Instead it is connected to the input of the mode switch.
There is also a peaked mode of this signal, which is as follows: For both the flat and peaked tidal modes, we multiply by the square of the function s(ω) to convert the response to displacement. Finally, the conversion K between volts and digital units (DU/V) is applied. Fig. 3 shows the transfer function for the tidal mode.

Response of the Short-Period Seismometer
The transfer function of the short-period sensor A SP (ω) in acceleration is expressed by where G 1 is the generator constant of the magnet-coil system and G 2 is the preamplifier gain. G is the resistance ratio of the damping circuit, which is expressed where R s is the damping resistance and R g is the coil resistance in ohms. S p (ω) is the transfer function of the short-period sensor in acceleration.
where ω 0 is the resonant frequency in rad/s. The values are from Yamada (2012), (except K, which was derived in section 2.1.1). The short-period transfer function is shown in Fig. 3, and a block diagram is included in the Electronic Supplement.

Active Seismic Experiment (ASE)
Active seismic experiments were performed at stations 14 and 16 with a small array of geophones. In contrast to the passive experiments, which were primarily designed to detect natural seismic events, the active experiments were designed to evaluate the subsurface structure around the landing site using controlled seismic sources. For both stations, three geophones were deployed to form a linear array and station 16 (right). Reproduced from Kovach et al. (1971) and Kovach et al. (1972).
( Fig. 4). The nominal distance between the geophones was 45.7 m (Kovach et al., 1971). The geophones were labeled as geophone 1, 2 and 3, with geophone 1 closest to the Central Station. Two types of seismic sources were used for the exploration.
The second seismic source used rocket-launched grenades which impacted at a location distant from the geophone array. The grenades were designed to probe different depths at the landing site. Unfortunately, the grenade experiment was not performed at station 14 due to the fear that the back-blast might damage the other instruments. Table 3 shows the launch details for station 16. The grenades reached approximate distances of 914 m, 305 m and 152 m from the array. Kovach et al. (1971) and Kovach et al. (1972) monitored several additional signals, including the thrust of the Apollo 14 and Apollo 16 Lunar Module ascent. They estimated the structure of the local subsurface using a combination of active and passive sources. Kovach et al. (1971), Kovach et al. (1972) and Brzostowski and Brzostowski (2009) describe more details of the experiment. The active seismic experiments (ASE) used geophones, which covered higher frequencies compared to the passive experiments. The transfer function A ASE (ω) for acceleration is represented by: where A is the amplifier gain, G is the generator constant and S p is a transfer function for acceleration (Eq. 18).
In addition, the experiment used an 8th-order low-pass filter (McAllister et al., 1969). The filter type is not specified. However, we find a reasonable fit to Fig. 7-5 of Kovach et al. (1971) with a Butterworth filter: where n is the order of the filter, and ω l is the cutoff angular frequency.   Kovach et al. (1971). The low-pass filter order and cutoff are from McAllister et al. (1969). We estimated the damping constant by fitting it to Fig. 7-5 of Kovach et al. (1971). We calculated the values for the conversion coefficient Kg and the conversion constant D using Table 5-VI in Lauderdale and Eichelman (1974). Yamada (2012) estimated the logarithmic compression parameters (Mneg, Mpos, bneg, bpos and M 1 ) using calibration data provided by Y. Nakamura. The nominal sampling rate is from Table A1 in MSC (1971). We noticed that the sampling rate is sometimes incorrectly quoted as 500 Hz.   Kovach et al. (1972). Other parameters from the same sources as The output voltage of the ASE signal was 5 V and the digital output was recorded in 5-bit integers. The following expression recovers the seismometer output voltage V out from the digital output D out : We can recover the pre-compressed input voltage V in using the following expression from Yamada (2012): Table 4 and Table 5 include the parameters for station 14 and 16, respectively.
One of the transfer functions for station 14 is shown in Fig. 5.

Lunar Seismic Profiling Experiment (LSPE)
Another active experiment was performed at station 17. The aim of Lunar Seismic Profiling Experiment (LSPE) was to explore the subsurface down to a few kilo- meters, which was much deeper than the previous active seismic experiments. A larger geophone array was established with four geophones (Fig. 6, top panel).
Eight explosive packages, equipped with different amounts of high explosives, were used as the seismic source. The four geophones formed a triangular array with an additional geophone at the center of the triangle. The outer sensors were approximately 100 m apart. The geophones were miniature moving coil-magnet seismometers. All eight explosives were successfully deployed during the extravehicular activity (EVA), and detonated after the astronauts left the Moon (Fig. 6, lower panel). Table 6 shows the amount of explosives and the detonation time for each explosive package. The LSPE was also turned on to observe the impulse produced by the thrust of lunar module ascent engine. Geophone 1 was approximately 148 m west-northwest of the lunar module (Kovach et al., 1973). The LSPE also detected the impact of the lunar module, which impacted approximately 8.7 km away. Finally, the LSPE was also turned on from August 15, 1976to April 25, 1977 for passive observation. Haase et al. (2013) improved on the original approximate estimates of the coordinates for the dimensions of the geophone array and the locations of the explosives using images from Lunar Reconnaissance Orbiter. Heffels et al. (2017) used these coordinates to re-estimate the subsurface velocity structure. Kovach et al. (1973) and Brzostowski and Brzostowski (2009) contain further details about the experiment.   (Kovach et al., 1973).
We get better results using this modified equation, which adjusts the zero displacement on the seismometer to zero voltage. Calibration data are included in section S7 of the Electronic Supplement.
The Lunar Surface Profiling Experiment has the same transfer function as the active experiments (Eq. 19), with different parameters (Table 7). As with the Active Seismic Experiment, we suspect that there was a pre-amplifier for the lower frequencies, but we have been unable to find the equation for it. Thus, the actual sampling rate was higher than the nominal rate shown in Table 7.
A transfer function for one of the geophones is shown in Fig. 5.  Kovach et al. (1973). The resonant frequencies and generator constants are from Table 10-I in Kovach et al. (1973). We obtained the conversion coefficient Kg, the conversion constant D and the logarithmic compression parameter values Mneg, Mpos, bneg, bpos and M 1 using calibration data originally provided by R. Kovach (via Y. Nakamura). We estimated the nominal values of the cutoff to the low-pass anti-aliasing filter f l and the damping constants, since these were not available in the original documentation. The sampling rate is from

Lunar Surface Gravimeter (LSG)
The Lunar Surface Gravimeter (LSG) was originally designed to detect gravitational waves on the Moon, as predicted from general relativity, and taking advantage of the very low noise conditions. The instrument was a high-sensitivity vertical accelerometer that sensed a local change in gravity. Unfortunately, the engineers miscalculated the compensating mass to deal with the reduced gravity on the Moon. Consequently, the instrument did not provide satisfactory data for its primary objectives. However, in addition to the primary objective, the LSG also functioned as a seismometer to detect ground motion. Recently,  verified that the data quality were sufficient for seismic analysis. Kawamura et al. (2015)  Due to the malfunction, the LSG went through a series of operations to recover the functionality (see Giganti et al. (1977) and  for more details). Initially, the sensor beam could not be centered to the equilibrium position. Additional force was applied to center the beam. This enabled the sensor beam to oscillate and the LSG was able to function as a seismometer. However, this also changed the sensitivity of the gravimeter from its original design. The gravimeter was originally designed to have a flat response between 0.1 and 16 Hz in seismic mode. Instead, the gravimeter had a peaked response at around 1.9 Hz, and sensitivity at low frequencies was degraded significantly after the recovery operation. The data were sampled at the same sampling rate as the short-period seismometers (∼ .02 s).
The Lunar Surface Gravimeter (LSG) was changed to open loop mode with maximum seismic output on December 7, 1973 (Giganti et al., 1977). Consequently, all of the available data were recorded in this mode. The transfer function for open seismic mode is as follows: where S(ω) is a transfer function of the seismometer for acceleration (Eq. 7).
We defined the transfer function using the block diagram in Fig. 2 of Weber and Larson (n.d.) The diagram is reproduced in the electronic supplement. G dc is a DC coupled gain, which is missing from the block diagram but described in p2, Weber and Larson (n.d.). K s is the sensitivity of the displacement transducer, G a is the adjustable gain which varied from 1 to 86.4 in 16 discrete steps (Fig. 2, Weber and Larson, n.d.). G s is the seismic-mode amplifier gain. F l (ω) is a low-pass filter, and F h (ω) is a high-gain high-pass filter.
The experiment used an 8th-order low-pass Butterworth filter (described in Eq. 20). It also used a high-gain 4th-order high-pass Butterworth filter as follows: where G 1 is the gain, m is the order of the filter, and ω h is the cutoff angular frequency.
After the corrections were made, the quality factor was estimated to be about We estimated the order for the high and low-pass filters, and the cutoff frequency for the high-pass filter using the transfer function produced by the original team (Fig. 5, Weber and Larson, n.d.). Finally, the conversion K between volts and digital units (DU/V) is applied. Although we reproduce the peak at 1.9 Hz, we were unable to reproduce the sharp peak in the original. Since the instrument had to be adjusted after deployment, we stress that many of the parameters described here are only estimates. The estimated transfer function for the Lunar Surface Gravimeter is shown in Fig. 3. After the malfunction and reconfiguration, the average noise level of the Lunar Surface Gravimeter (LSG) was higher than the other Apollo seismometers (Lauderdale and Eichelman, 1974).

Seismic Sources
Seismologists have observed and categorized several types of moonquakes. These include deep moonquakes, meteoroid impacts, shallow moonquakes, thermal moonquakes and also artificial impacts ( Fig. 7; Table 8; Schematic in  Type of Moonquake No.

Artificial Impacts 9
Meteoroid Impacts 1743 Shallow Moonquakes 28 Deep Moonquakes (assigned to nests) 7083 Deep Moonquakes (not assigned to nests) 317 Other Types (including thermal quakes) 555 Unclassified 3323 Total 13058 Lunar events typically have a very long duration, and indirect scattered energy can arrive tens of minutes after the direct waves (e.g. Fig. 7). The scattered energy is known as the seismic coda. These long, reverberating trains of seismic waves were interpreted as scattering in a surface layer overlying a non-scattering elastic medium (e.g Dainty et al., 1974). Diffusion scattering is important when the mean free path (the average distance seismic energy travels before it is scattered) is short compared to the seismic wavelength. In comparison with terrestrial environments, Dainty and Toksöz (1981) showed very short mean free paths for the Moon. Dainty et al. (1974) and Aki and Chouet (1975) distinguished the diffusion model of seismic wave propagation (which applies to a strongly scattering medium) from a single scattering model (which applies to a weakly scattering medium). The much larger amplitude (relative to direct phases) and much greater duration of lunar seismograms compared to terrestrial seismograms suggests both more intense scattering and much lower attenuation on the Moon than on the Earth (Dainty and Toksöz, 1981). Sato et al. (2012) provide an extensive review of the theoretical developments in the field of scattering and attenuation of high-frequency seismic waves (particularly when applied to the Earth).

Artificial Impacts
Nine impacts occurred when the Saturn third stage boosters or the ascent stages of the lunar module were deliberately crashed into the Moon. These observations are particularly valuable, since the timing of the impact, the location, and the impact energy are known (see Section S9 in the electronic supplement). Unfortunately, the tracking was prematurely lost for Apollo 16's Saturn booster, meaning that both the location and timing were poorly known for this impact. Plescia et al.
(2016), Wagner et al. (2017) and Stooke (2017) estimated the location of many impacts using remarkable images from the camera on Lunar Reconnaissance Orbiter.
Photographs of the impact craters can be viewed online (LROC, 2017).

Meteoroid Impacts
More than 1700 events recorded during the operation of the Apollo stations were attributed to meteoroid impacts (e.g. the Nakamura et al. (1981) catalog, provided within the electronic supplement). Oberst and Nakamura (1991) found two distinct classes of meteoroids impacting the Moon, originating from either comets or asteroids, and estimated the mass for the meteoroids to range from 100 g to 100 kg.
The waveforms of meteoroid and artificial impacts differ significantly from fault-generated quakes. They do not have a double-couple source. Since the Moon has no significant atmosphere, impacts have high velocities, and the impactor tends to fragment and vaporize. Teanby and Wookey (2011) noted that this leads to the creation of radially symmetric craters, except for very low-angle impacts (with respect to the horizontal). Therefore, the most appropriate seismic source is purely isotropic (explosive) (Stein and Wysession, 2003;Teanby and Wookey, 2011;. Gudkova et al. (2011)

Shallow Moonquakes
Shallow moonquakes are rare events (with only 28 events in the catalog of Nakamura et al. (1981)), which have larger magnitudes than the other naturally occurring events. There is some variation in the estimated depth ranges for these events.
In the VPREMOON model of , they occur at depths from 0 to 168 km. In contrast, Khan et al. (2000) preferred a depth range of 50 to 220 km, and suggested that they occur in the upper mantle. Similarly, Nakamura et al. (1979) suggested that the amplitude decay function of shallow moonquakes implies that they are likely to be located shallower than 200 km depth but deeper than the crust-mantle boundary. Oberst (1987) estimated the equivalent body-wave magnitudes to be between 3.6 and 5.8. He also estimated unusually high stress drops.
Shallow moonquake spectra include high frequencies, which are clearly visible on the short-period seismographs. While the deep moonquakes have little seismic energy above 1 Hz, energy for the shallow moonquakes continues up to about 8 Hz and then rolls off. This is the reason that shallow moonquakes were initially called high-frequency teleseismic events. No correlation between shallow moonquakes and the tides has been observed (e.g. Nakamura (1977)). Nakamura (1980) showed a strong similarity between these quakes and intraplate earthquakes on Earth, particularly considering the relative abundance of large and small quakes.

Deep Moonquakes
Deep Moonquakes are the most numerous events, and are found at depths from 700 to 1200 km (Nakamura et al., 1982;Nakamura, 2005). They have highly repeatable waveforms, suggesting that they originate from source regions (or 'nests') which are tightly clustered. The quakes have been classified into numbered groups or clusters (e.g. Nakamura, 1978;Bulow et al., 2007;Lognonné et al., 2003). The exact number of nests varies between studies, but Nakamura (2005) (2006) found that the A1 group was large enough to distinguish subgroups of events with slightly different waveforms. When they processed the stacks separately, the final waveform stacks of these subgroups were somewhat different, but the delays between P and S arrival times obtained by correlation implied that the distance between sources was at most one kilometer. Nakamura (2003) correlated every pair of events using a single-link cluster analysis. Events belonging to one source region correlated to a high degree, while those belong to separate source regions correlated to a lesser degree. A surprising finding was that some events that were originally thought to be belonging to two separate source regions were found to be highly correlated.
Many studies, including Lammlein et al. (1974), Lammlein (1977) and Nakamura (2005), have noted an association between the occurrence times of deep moonquakes and the tidal phases of the Moon. Analysis of the periodicity of deep moonquake occurrence shows the strongest peak at 13.6 days, followed by a peak around 27 days (e.g. Lammlein, 1977). Additional 206-day variation and 6-year variation, due to tidal effects from the Sun, are also observed (Lammlein et al., 1974;Lammlein, 1977). However, analysis of individual clusters by Frohlich and Nakamura (2009) shows tidal periodicity for each cluster, but not necessarily the same dependence on the tidal cycle for all clusters.
Although the deep moonquakes appear to be tidally triggered, the exact cause remains unclear. Saal et al. (2008) argued that the presence of fluids (especially water) explained the mechanism. Instead, Frohlich and Nakamura (2009)  600-1200 km, which covers the range of estimated deep moonquake depths (e.g. Cheng and Toksöz, 1978).
The majority of the deep moonquakes have been located to the nearside of the Moon, with around 30 nests attributed to the farside (Nakamura, 2005)). Since none of the events have been located to within about 40 degrees from the antipode of the Moon, Nakamura (2005) suggested that this region of the farside is aseismic, or alternatively that the very deep interior of the Moon severely attenuates or deflects seismic waves.

Thermal Moonquakes
Duennebier and Sutton (1974) showed that the majority of the many thousands of seismic events recorded on the short-period seismometers were small local moonquakes triggered by diurnal temperature changes. More recently, Dimech et al. (2017) found and categorized 50,000 events recorded by the Lunar Seismic Profiling Experiment at Apollo 17. The events occurred periodically, with a sharp double peak at sunrise and a broad single peak at sunset.

Compilation of Reference Data
We have compiled reference data from various sources, and provide these data sets within the Electronic Supplement. This section describes these data sets.

Deep Moonquake Stacks
As described above, waveforms from each deep moonquake source region are highly repeatable. Researchers have used the repeatability of the waveforms to use stacking and cross-correlation methods to enhance the signal-to-noise ratio. It is easier to pick the arrival times on the stacked waveforms, which are considerably clearer.
The quality of the stack will depend on a number of factors including the number of stacked events, the signal-to-noise ratio of the individual events and the filtering applied. Nakamura (1978) showed that source regions also produce events with similar waveforms but with flipped polarity. He suggested that this was caused by similar events being triggered by different parts of the tidal cycle.
In section S2 of the Electronic Supplement, we provide deep moonquake stacks from three independent sources in miniSEED format (Nakamura, 2005;Lognonné et al., 2003;Bulow et al., 2007). Nakamura (2005) correlated deep moonquakes to determine clusters, and stacked the seismograms when he detected 10 or more events within a cluster. The individual traces were weighted to maximize the final signal-to-noise ratio. The stacks were made from cross-correlations between events using single-link cluster analysis. He made P and S arrival time picks and estimated hypocenters for many of the stacks. Using a slightly different process, Lognonné et al. (2003) stacked seismograms after time alignment relative to a reference event. Bulow et al. (2007) also stacked these data, which were originally included in Bulow et al. (2005).
They used a median-despiking algorithm to produce improved differential times and amplitudes, which enabled them to produce cleaner stacks. Lognonné et al. (2003) recorded which event was the reference in the header to the file. However, Nakamura (2005) did not use reference times in his process, and we do not have reference times from Bulow et al. (2005). This unfortunately makes it more difficult to compare stacks or calculate arrival times. For the stacks provided in the Electronic Supplement, we are unable to confirm exactly which individual events were used in each stack, which traces of which individual events were flipped relative to the reference event, and the filtering or pre-processing carried out by the researchers. We expect that slightly different criteria were used by different researchers to accept or reject each trace. The stacks of Nakamura (2005), Bulow et al. (2007), and Lognonné et al. (2003) are 500 s, 4200 s and 1600-3500 s long, respectively. A1 contains the largest number of events with good signal-to-noise ratio, followed by A40. Since the different studies used different reference traces, several of the stacked traces were of reverse polarity. For example, for the A1 cluster MH1 and MH2 from Bulow et al. (2007) and all three traces from Nakamura (2005) were of reverse polarity to those from Lognonné et al. (2003). The reverse traces were flipped before calculating the correlation coefficients or plotting.

Lunar Catalog of Arrival-Time Picks
We compiled arrival times from Goins (1978), Horvath (1979), Nakamura (1983), Lognonné et al. (2003), Bulow et al. (2007) and Zhao et al. (2015)). We provide the arrival times within section S3 of the Electronic Supplement. The P and/or S arrivals were picked for artificial impacts, meteoroid impacts, shallow moonquakes, and deep moonquake stacks. Nakamura (1983) summarized the published results of Horvath (1979) and Goins (1978) (Lognonné et al., 2003;Bulow et al., 2007;Nakamura, 2005). We aligned the independent stacks using cross-correlation. The correlation coefficients for each pair are shown in the boxes. The reference times are from Lognonné et al. (2003), and are shown beneath each stack. The plot shows S wave arrivals (green lines), and P wave arrivals (blue line; only available for the A1 cluster) from Lognonné et al. (2003). For A97, the Bulow et al. (2007) and Nakamura (2005)  Bulow/ Nakamura Fig. 9 Histogram of the correlation coefficients between the vertical component of the stacked deep moonquake traces for each named cluster compiled by different authors. Turquoise lines compare Lognonné et al. (2003) and Bulow et al. (2007); purple lines compare Lognonné et al. (2003) and Nakamura (2005); blue lines compare Bulow et al. (2007) and Nakamura (2005).
The number of clusters with correlation coefficients greater than 0.7 or below 0.7 is shown.
Correlation window length is 300 s, and begin 50 s before the S-arrival time pick, when it is available. When the S-arrival has not been picked, the window length is the full length of the shortest trace.
locity structure of the crust and upper mantle (Nakamura, 1983). In contrast, the source locations for the other quakes rely on previously determined models. Garcia et al. (2019), also written by our group, provides a discussion of the determination of the source locations. We provide two possible origin locations. One set of locations come from Lognonné et al. (2003). The second set comes from , and were calculated using the velocity model Very Preliminary Reference Moon (VPREMOON).
For individual events, such as shallow moonquakes or impacts, a reference timestamp is provided. However, the P and S arrival times of the deep moonquakes are picked on stacked waveforms for which a single reference time is not always available. Unfortunately, different studies used different reference events to align their stacks. We decided to present here only arrival times of deep moonquake events for which a quake location is available. When the deep moonquake stacks used the same reference time as Lognonné et al. (2003), the P and S arrival times are provided. When the deep moonquake cluster is not clearly identified as the one used in Lognonné et al. (2003), the P and S arrival times are provided with their own reference time. When the arrival time does not have a clear reference date and time, or a different one from Lognonné et al. (2003) for the same deep moonquake cluster, only S-P differential travel times are provided.
Later studies of deep moonquakes, such as Bulow et al. (2007), were able to include 503 more individual events than Nakamura et al. (1981). By using crosscorrelation to identify new events, some moonquake stacks had up to 53% more events than Nakamura et al. (1981). The compiled arrival times relative to the reference time, and the S-P times when a reference time is not available, are provided in Section S3 of the electronic supplement. For some events, there were arrival times from multiple stations and from multiple studies. In the instances where more than two studies cite P, S, and/or (S-P) values for an event, we computed the mean and standard deviations. Fig. 10 shows P and S arrival times from Lognonné et al. (2003), plotted by epicentral distance. The plot shows some scatter for both P and S arrival times. We expect some scatter in this plot, since the events are estimated to originate from different depths. In addition, it may be difficult to estimate epicentral distance.
Some variation is also expected from differences in crustal thickness or seismic velocities between different regions of the Moon. for the inversion tests presented into our companion paper (Garcia et al., 2019). Error bars are available but not presented for clarity.

Statistical analysis
The accuracy of measuring the arrival times of both the P-wave and S-wave for different events strongly affects the travel-time measurement, which is the key parameter for further inversions on both source location and velocity structure.
However, due to the signal characteristics of the lunar seismic records, accurate and reliable arrival-time measurements are challenging. The arrival times of artificial impacts and deep moonquakes can have large differences between different studies.
For example, for the impact of Apollo 15's Lunar Module (15LM), the P-wave arrival time from Nakamura (1983) is 5.5 s earlier than that of Lognonné et al. (2003).
We calculated the variation of arrival-time picks (Fig. 11). Since we calculate a mean for each event, we require at least two independent observations. Where available, we show P arrivals, S arrivals, and the difference between the S and P arrival times (S-P time). For the P arrivals, the small number of artificial impacts show high consistency (the standard deviation is 1.3 s). The second lowest standard deviations are for the shallow moonquakes (3.0 s), followed by the meteoroid strikes (4.0 s) and a small number of deep moonquake observations (10.1 s). In general, the S arrivals have lower consistency than the P arrivals. There are too few observations for meaningful statistics for the artificial impacts for S and also S-P. The standard deviations are lowest for the stacked deep moonquake events (3.7 s), followed by the shallow moonquakes (13.2 s) and then the meteoroid strikes

Arrival Time of the Maximum Energy and the Coda Decay Time
Lunar seismograms are characterized by strongly scattered waves with a long duration, when compared with their terrestrial counterparts. The coda, which can be thought of as the tail of the seismogram, is formed from the scattered waves which arrive after the direct waves. There is a long delay time between the onset of the signal and the arrival of the maximum energy, also known as the rise time (Latham et al., 1971;Blanchette-Guertin et al., 2012). A long rise time indicates multiple scattering in a strongly heterogeneous medium, and that the waves are strongly dispersed. The rise is followed by an even longer decay time, where energy from the scattered waves continues to arrive at the seismic station. An accurate measurement of the rise time requires an accurate pick of the S-wave arrival, which is not always possible. Instead, Gillet et al. (2017) used t max , which is the time elapsed from the energy release at the source (at time t 0 ) to the arrival of the maximum of the energy (Fig. 12). Although measurement of t max does not require a pick of the S-wave arrival, it is affected by any error in the estimation of the origin time t 0 .
The long duration of the coda on the Moon is the result of a very low noise level and significantly lower anelasticity than Earth. Using the diffusion model of scattering of Dainty et al. (1974) and Aki and Chouet (1975), we can quantify the decay of a seismogram after the arrival of the maximum energy. Aki and Chouet (1975) introduced a quality factor Q c , such that the energy varies in the coda as On Earth, the exponent is usually chosen between 1 and 2 depending on the geological context and the wavefield content. In the case of the Moon, the dissipation is so weak and the propagation time is so long that waves have the time to explore the entire volume of the planet. In such a scenario, one expects the signal to simply decay exponentially at long lapse time, thereby suggesting α = 0 (Blanchette-Guertin et al., 2012;Gillet et al., 2017). We also adopt this value. τ d is the time taken for the (smoothed) coda to be reduced to 1/e times its initial value.
Observations by Latham et al. (1971); Dainty et al. (1974)  A typical choice of filter is the 4-pole Butterworth with a bandwidth equal to 2/3 of the central frequency (Aki and Chouet, 1975). The next step is to convert amplitude to energy. The simplest procedure is to square the filtered traces which directly yields a quantity proportional to the kinetic energy of wave motion.
Finally, the squared trace is smoothed to reduce the fluctuations. A customary choice is to apply a moving-average filter with a typical duration of 8 to 16 peri-ods. Longer windows provide smoother envelopes at the expense of reducing the signal dynamics.
Once smooth energy envelopes have been obtained, it is straightforward to find the maximum of the energy and its associated time of arrival. Estimating the uncertainty is difficult due to residual random fluctuations of the envelope. Gillet et al. (2017)   given epicentral distance, followed by shallow moonquakes and then impacts. t max could not be measured for the deep moonquakes at high frequencies. Modeling (at 0.5 Hz) predicts a dependence of t max on epicentral distance, as well as a sharp increase in t max around 10°epicentral distance for the impacts (Fig. 7, Gillet et al. (2017)). Note that t max combines the travel time for the initial energy, as well as the rise time from initial energy to maximum, and that both quantities depend on epicentral distance.

tmax (s) tmax (s)
τd (s) τd (s)   operating simultaneously on the Moon. However, in order to be able to analyze the data of these simultaneously operating instruments as a network, the missions/instruments/sensors need to fulfill a minimum set of requirements. Our team sets out these minimum requirements in Fig. 14. The requirements are built upon a single objective: to ensure the capability of researchers to analyze data simultaneously acquired by similar geophysical sensors on the Moon. These requirements do not only apply to seismic sensors but to any geophysical sensors deployed on the Moon. For each requirement, we justify the flow-down from science objectives to station/mission and instrument/sensor requirements. However, performance requirements are not specified in order to allow low performance sensors to still fulfill these requirements. Therefore, decisions on performance considerations can be decided by the organization funding the instrument or the mission. The science return as a function of instrument performance is not considered in these requirements because it often depends not only on sensor self noise, but also on mission design, deployment capabilities, and many other factors.

Resources within the Electronic Supplement
Section S1 contains parameters describing the location of the Apollo passive seismometers, including longitude, latitude, elevation, azimuth of the horizontal seismometer components and distance between stations. Section S2 contains stacked traces from deep moonquake clusters from three independent studies, in miniSEED format. Section S3 contains arrival-time catalogs from six independent sources, as well as estimates of event time and location where available. Section S4 contains the full lunar catalog which contains over 13,000 events (Nakamura et al. (1981) and updated in Oct. 2008). Section S5 contains attenuation parameters from Gillet et al. (2017). Section S6 contains a pdf version of the Minimum Requirements for an International Lunar Network (Fig. 14). Section S7 contains a Jupyter Notebook to plot the transfer functions and the logarithmic compression parameters.
WARNING: These requirements must be applicable to any instrument type (geophone, seismometer, VBB seismometer etc) and any mission type (hard or soft lander) from any agency (NASA, CNSA etc). Data from geophysical stations deployed on the Moon must allow an international community of researchers to locate events, perform waveform analysis and develop structural models.

Major
These requirements cover seis-mic data. Additional requirements should be stated for sensors sensing ground temperature, gravity field, magnetic field, ground rotations, low frequency planetary scale rotations and deformations, and lunar laser ranging. This section outlines the reasons for the requirements in the following sections. In defining the requirements, we make the assumption that the internal structure is perfectly known.

Source Timing
The origin time of geophysical events detected by the seismic network must be recovered with an accuracy better than the sampling rate of the recording stations.

Source Location
The source location of geophysical events detected by the network must be recovered with an accuracy better than a quarter of the wavelength of the signals used to locate the source.

Source Energy
The energy of the source of geophysical events detected by the network must be recovered with an accuracy better than 20%. Source radiation = X°T he orientation of the instruments must be provided with a precision ensuring that the errors on the estimate of the source radiation are smaller than 10°.
ILN-REQ-0.5 ILN-REQ-1.5 Data archiving and documentation Data content, documentation, storage and archiving must ensure that an international community of researchers are able to understand these data and to implement research activities. Accuracy on the dating of the samples must be better than one tenth of the average sampling rate of the data channel.

STATION REQUIREMENTS
The aim is to use the data from various stations for network analysis. The amplitude of the instrument response must be provided with an accuracy better than 10% over the bandpass of the instrument during the entire lifetime of the instrument.

Calibration Information
The phase of the instrument response must be provided with an accuracy better than 10°over the bandpass of the instrument during the entire lifetime of the instrument.

Metadata and processing
Metadata must contain a description of all the processing steps from the physical unit to the digital (count) output of all data channels. Compression/ Decompression If lossy compression is applied, it should allow signal reconstruction with an accuracy better than 10% of the signal energy.

INSTRUMENT REQUIREMENTS
ILN-REQ-1.3; ILN-REQ-1.5 ILN-REQ-2.10 Aliasing The instrument must be designed so that less than 0.1% of the signal above the Nyquist frequency is aliased in the bandpass of the instrument.
ILN-REQ-1.5 ILN-REQ-2.11 Noise Estimates Sensor and instrument noise must be estimated over the bandpass of the instrument and provided for each seismic channel in m/s 2 /√(Hz).
ILN-REQ-1.5 ILN-REQ-2.12 Archiving Data and metadata for all the instrument channels must be archived both in planetary databases and in geophysical sensor databases.
ILN-REQ-1.5 ILN-REQ-2.13 Naming A network code and a station code must be assigned to the geophysical station by the International Federation of Digital Seismograph Networks (FDSN).
ILN-REQ-1.2; ILN-REQ-1.5 ILN-REQ-2.14 Station Location The station location must be provided in a standard reference system defined by the International Astronomical Union (IAU).

ILN-REQ-1.2 ILN-REQ-2.15 Station Location
The station location coordinates must be provided with an accuracy better than 25 m.
Half a wavelength, assuming highest frequency is 10 Hz and lowest S-wave velocity is 500 m/s.

ILN-REQ-1.4 ILN-REQ-2.16 Axis Orientation
The sensing direction of the instrument data channels must be provided with an accuracy better than 10°.
ILN-REQ-1.5 ILN-REQ-2.17 Operations Mission, platform and instrument operation activities impacting the signals above instrument noise level must be time-stamped, recorded and archived in the metadata for the instrument.

Fig. 17 Continued
Section S8 reproduces block diagrams for the mid-and short-period seismometers.
Section S9 contains a table of the artificial impacts. Section S10 summarizes the current data availability. Section 11 contains the response files for the mid-and short-period seismometers.