Detection of ground shaking forms the basis of many lahar-warning systems. Seismic records of two lahar types at Ruapehu, New Zealand, in 2007 are used to examine their nature and internal dynamics. Upstream detection of a flow depends upon flow type and coupling with the ground. 3-D characteristics of seismic signals can be used to distinguish the dominant rheology and gross physical composition. Water-rich hyperconcentrated flows are turbulent; common inter-particle and particle-substrate collisions engender higher energy in cross-channel vibrations relative to channel-parallel. Plug-like snow-slurry lahars show greater energy in channel-parallel signals, due to lateral deposition insulating channel margins, and low turbulence. Direct comparison of flow size must account for flow rheology; a water-rich lahar will generate signals of greater amplitude than a similar-sized snow-slurry flow.
 Seismic or acoustic monitoring in catchments surrounding active volcanoes is used as the basis of lahar warning systems (e.g., Pinatubo [Marcial et al., 1996]; Indonesia [Lavigne et al., 2000]; Ruapehu [Leonard et al., 2008]) and for mass flow monitoring in other alpine areas [Hurlimann et al., 2003]. Geophones or seismometers are particularly useful because they can be placed outside the path of highly destructive/erosive flows, and can operate during multiple events [e.g., Itakura et al., 2005]. Acoustic Flow Monitor (AFM) systems are cheap, produce low data rates and have low power needs. Broad-band seismometers are more expensive and have higher data rates, but enable better resolution of flow magnitude and understanding of internal flow properties.
 Analyzing a wide range of seismic frequencies over several sites enables estimation of mass-flow frontal velocity, variations in internal flow properties and interactions with the channel. Variations in signal intensity across 3-D should reflect fundamental differences in the water content as well as particle-particle and particle-substrate interactions between flows. In order to use this information effectively in warning systems, these features need to be interpreted.
 Presented here is an analysis of seismic signals from snow-dominated flows that descended the slopes of Ruapehu Volcano, New Zealand, on 25 September 2007. These were generated by a small phreatic eruption that ejected water from Crater Lake [Lube et al., 2009]. Instrumentation installed along the Whangaehu River captured the eruption and passage of two snow-dominated lahars and a later water-rich hyperconcentrated flow. These data provide valuable insights into the physical and dynamic properties of snow slurry lahars, in relation to snow avalanches and other larger-scale lahars at this site [Manville and Cronin, 2007].
2. Geological Setting and Instrumentation Details
 Mt. Ruapehu (2797 m) is an andesitic stratovolcano in the central North Island of New Zealand (Figure 1). The acidic Crater Lake (9–10 × 106 m3) typically covers an active vent at c. 2540 m a.s.l. [e.g., Cronin et al., 1996]. Crater Lake is surrounded by several glaciers, which have fed snow and ice into lahars generated by: eruptions through the lake in 1969, 1975 and 1995 [Cronin et al., 1996], or by collapse of ice/tephra dams such as in 1953 and March 2007 [Manville and Cronin, 2007]. The majority of these lahars occurred in the Whangaehu River catchment (Figure 1).
 We installed a monitoring station alongside the Whangaehu River in 2006, c. 7.4 km downstream from the Crater Lake at the Round-the-Mountain-Track (RTMT; Figure 1). It consists of a 3-component broadband seismometer (Guralp-6TD), an AFM and a radar water-level gauge (VEGAPULS), installed along the south bank of the channel on a lava bluff. The seismometer is buried 1 m into soil directly overlying coherent lava that extends beneath most of the main channel. This sensor is located c. 4 m from the channel edge and c. 20 m above the river bed. The broadband recorded data with a sample rate of 100 Hz. Other instruments recorded observations at 5 s intervals. The AFM and broadband seismograph successfully recorded the lahars, but the radar stage gauge reacted too slowly to the sudden change in flow level caused by the onset of the lahar fronts, and/or could not process reliable reflections from the highly irregular and “solid” surfaces of the snow-slurry lahars.
3. Eruption Seismicity
 The seismometer at the RTMT station recorded initial tremors associated with the eruption starting at 08:16 (UT). A single explosion which expelled jets of water, lake sediment and rock debris, was recorded at 08:26 (UT) by two microbarograph stations on the western slopes of the volcano (GeoNet, New Zealand). At the RTMT, the eruption explosion and jetting event is represented as a brief spike across all frequencies (Figure 2). At least four other low-frequency spikes were recorded, but none generated an associated air wave. It is unlikely that additional material was ejected from Crater Lake, as evidenced in the deposits [Lube et al., 2009].
4. Lahar Seismicity
 Three lahars were generated from the single explosion at 08:26, two in the eastern Whangaehu catchment, and one in the north-western Whakapapa ski field area (Figure 1) [Lube et al., 2009]. We consider only those lahars that traveled down the Whangaehu Channel past the RTMT. The first snow lahar (E1) was caused by run-off of explosively ejected Crater Lake water and c. 200,000–400,000 m3 of entrained snow and ice from the Central Crater and Whangaehu Glacier, while the second (E2) was due to displacement of water over the lake outlet [Lube et al., 2009]. A third watery flood (E3) was recorded c. 1 hour after the eruption with no associated anomalous seismicity and thought to be related to further collapse and drainage from the upper-catchment ice-slurry deposits.
5. Flow Velocity and Range of Detection
 Analyses of seismic signal from the three lahars along the Whangaehu valley were made using total spectra and spectrogram profiles on all components (Figure 3). Time-series data clearly shows the approach and passage of the flows, but spectrogram profiles allow for easier detection of flow arrival at the site. Due to the unique frequency content of the signal, increases in frequencies and amplitudes can be seen earlier in the spectrogram than in the time series trace.
 The spectral plots of the September lahars show a distinct triangular shape resulting from an increase in high frequency signal with time, as is also observed in seismic signals from snow avalanches [Surinach et al., 2005]. This shape contrasts with earthquake signals [e.g., Surinach et al., 2005] and volcanic eruptions (e.g., Figure 2). We attribute the shape to either: a) anelastic attenuation of the propagation waves with distance, due to the more rapid attenuation of high frequencies compared to low [Aki, 1980], or b) increase in energy, due to the entrainment of sediment causing a rise in the signal amplitude [Surinach et al., 2005]. Since the E1 and E2 lahars had primarily entrained material in the upper reaches of their travel path, and deposit analyses indicate that they were not erosive in or near the instrumented reaches [Lube et al., 2009], we conclude that the latter explanation has, at best, a minimal effect on the amplitude of the signal. The triangular shape most likely reflects the decreasing attenuation of the signal as the flow approaches the sensor. Conversely, an inverted triangular shape seen in the spectrogram after a period of full frequency signal is attributed to the attenuation of signal as the flow-tail passed beyond the sensor. We therefore conclude that amplitude profiles are at their greatest when the flow directly passes the sensor [e.g., Caplan-Auerbach et al., 2004]. Accordingly, the arrival times of E1 and E2 at RTMT are estimated at 08:35:35 and 08:42:20 UT, respectively. Their true travel paths to this point were 6870 m for E1 and 7470 m for E2 (Figure 1). Using onset and cessation of the eruption as boundaries for flow start time, average velocities to the RTMT are 12.1–13.6 m/s for E1 and 7.7–8.2 m/s for E2. For comparison, local velocity estimates of c. 5 m/s were calculated for both flows from super-elevation measurements [Lube et al., 2009].
 The seismic spectrogram of E1 (Figure 3b) shows an unusually brief and sharply terminating tail, which either indicates movement of the bulk of the flow a critical distance away from the instrument where detection suddenly ceases, or a sudden waning of flow size and energy. As it contrasts with the long tails more typically recorded in water-rich lahars at this site (such as E3), the sharp cessation is interpreted to represent the sudden halt of the E1 ice slurry. Mapped deposits constrain the final runout of E1 to 830 m downstream of the site, with a sharp front to the deposit implying a sudden halt [Lube et al., 2009]. This suggests that the front traveled from the RTMT to the final runout point at an average velocity of 4.5 m/s, corresponding well with the super-elevation estimates. Projecting estimates of local velocity, the first vibrations relating to the detection of E1 were seen when it was c. 600–670 m upstream of the sensor.
6. Flow-Substrate Interaction, Flow Dynamics and Rheology
6.1. Vibrational Energy and Frequency Spectra
Suwa et al.  proposed that the peak vibrational energy recorded by a seismometer is proportional to the peak discharge of a passing lahar, while volume estimations could be made by integrating the acceleration amplitude. Direct comparisons of vibrational energies between similar flows recorded at the same site should, therefore, provide a means for calculating their relative size and peak discharge (or mass flux). The peak discharge of E1 was independently calculated as 1700 m3/s, with a sharp-fronted peak wave of 6.2 m in height [Lube et al., 2009]. In comparison, the break-out lahar of 18 March 2007, at the same site, had an estimated peak discharge of 2000–2200 m3/s and a maximum stage height of 8 m. The peak wetted perimeter of the E1 snow-slurry lahar was c. 70% of the March event, but it only produced 10–12% of the peak vibrational energy of the 18 March sediment-laden flow. It is less easy to compare these flows with E2 and E3, as the relative amplitudes of the signals are likely to have been dampened by the loosely compacted deposits of E1 that would decrease the efficiency of wave propagation through the unit. The discrepancy between flow volume and seismic amplitude for the different flow types is an important observation, as it implies seismic amplitudes alone cannot be used to determine flow volume. We infer the differences are a result of the freedom movement of large particles within the flow, especially their ability to saltate/collide with the substrate and flow-substrate frictional interaction.
 Previous seismic studies of hyperconcentrated and debris flows suggest that particle collisions transferred to the channel bed or sides generate signals of higher frequency than does motion of frictional-sliding bedload. Huang et al.  measured bed-friction generated signals with frequencies in the range 10–300 Hz, though predominantly between 20–80 Hz. Collisional motion produced signals between 10–500 Hz. Other studies have concluded energy from hyperconcentrated and debris flows are concentrated in the 10–100 Hz range [Marcial et al., 1996; Lavigne et al., 2000].
 Depositional indicators of flow-parallel lateral shear structures in the September flows show that the ice-slurries moved by frictional sliding as a plug-like Bingham material [e.g., Ancey, 2007], with little free motion of particles within the flow, above a basal zone of high pore water pressures [Lube et al., 2009]. As such, we interpret the seismic signal to show little component relating to particle collisions; instead, we infer that the dominant 5–20 Hz spectral peaks of the E1 signal (Figure 3b) are representative of the frictional interaction of the moving snow-rich material over the channel substrate. In comparison, visual observations of the 18 March 2007 lahar imply it was a sediment-rich hyperconcentrated streamflow, producing high levels of bedload saltation and particle collisions. Correspondingly, higher seismic frequency ranges were recorded (Figure 4c). From this, we infer that the water content of E1 was very low (consistent with physical estimates from deposits [Lube et al., 2009]). If it were higher, saturation of air-filled porosity would be expected, along with vertical drainage, generation of a watery underflow, and a seismic profile more typical of a hyperconcentrated flow.
 Lahar E2 produced signals that were generally less energetic than those of E1 (Figure 3d), with a less-pronounced increase in frequency and amplitude as the flow front passed the sensor and a more rapid tail drop-off, despite traveling c. 2 km farther (Figure 3e). The stratigraphy of the deposit shows that E2 traveled on top of the already-frozen E1 deposits. We infer that the reduced seismic amplitudes result from insulation/attenuation of seismic energy by the E1 deposits, reducing coupling between E2 and the channel base. The tail of E2 showed an increase in amplitude in the higher frequency range as it passed the sensor (Figure 3e), which may reflect a more watery and turbulent tail.
 The third flow (E3; Figure 3g) occurred c. 1 hour after the eruption event, and lasted over 90 minutes. The arrival of E3 generated a similar spectrogram profile to E1 and E2, but with lower energy amplitudes (Figure 3i). An abrupt energy increase across the entire frequency range occurred only in the vertical component c. 15 minutes after E3's arrival (Figure 3h). This is interpreted to represent the moment when E3 eroded through the deposits of E1 and E2 to directly contact the channel base. A site survey showed that E3 had cut a c. 10 m-wide slot through the earlier frozen deposits. A corresponding increase in the horizontal components is absent, probably due to the attenuation of signals through the porous frozen deposits that filled the c. 80 m-wide channel. The frequency distribution after c. 09:55 (UT) shows a similar, though less energetic, profile to the March 2007 lahar (Figure 4c). This suggests that E3 was dynamically similar, being turbulent and water-dominated. E3 left only a few cm of mantling deposit at the monitoring site, consistent with low sediment concentration.
6.2. Signal Directionality
 Comparisons of the channel-perpendicular and channel-parallel vibrational energies of the March and September 2007 lahars show significant differences in the excited frequencies and amplitudes. We hypothesize that this can provide further information on flow dynamics and rheology. The water-rich, turbulent March lahar had flow-perpendicular amplitudes up to 3 times higher than those in flow-parallel directions (Figure 3c). By contrast, the ice slurry flows E1 and E2 produced more flow-parallel energy (Figures 3c and 3f). Depositional evidence of the E1 and E2 flows indicate a plug-like behavior, with little internal motion, minimizing collisions within the flow. Water-rich hyperconcentrated flows, however, allow for greater freedom of particle motion, leading to random collisions within the flow and against channel margins. We suggest that it is this behavior that produces the significant lateral, channel-dispersive and collisional forces responsible for the observed higher amplitudes in cross-channel directions in the March flow compared to the September flows. Deposition inward from the margins is implied for the ice slurry flows [Lube et al., 2009], as exhibited by lateral-shear structures in the deposits; this would likely increase damping of seismic energy within the cross-channel orientation.
 The directionality signals generated by the E3 flow should be similar to the March 2007 lahar, with higher cross-channel amplitudes. However, they were more like those of E1 and E2 with proportionally lower cross-channel energy (Figure 3i). We infer this pattern to be due to attenuation as the signal passed through the channel-lining E1 and E2 deposits, damping the likely higher frequency content signal induced by E3.
7. Discussions and Conclusions
 We find that water- and snow-dominated lahars show distinctly different seismic signatures. Strong seismic signals and broad frequency responses relate to water-rich flows and high turbulence as they cause both bedload movement and frequent randomly oriented particle-particle and particle-bed collisions. The multiple trajectories, along with collisional contact with channel margins, produce more energetic cross-channel vibrations than flow-parallel. By contrast, snow-slurry flows have very low seismic energy compared to watery lahars of similar size as they involve little-to-no collisional interaction. They also generate stronger flow-parallel signals due to insulation of channel margins by lateral deposition and low incidence of channel-side collisions.
 Distinguishing different lahar types by these features will provide a more complete assessment of the nature and degree of hazard posed. This step must be made before simple interpolations of size are concluded from direct examination of signal strength. For snow-slurry lahars, signal strength estimates using thresholds developed for more dilute flows will strongly underestimate flow size (discharge/volume). This implies that typical AFM warning systems, with minimum recording thresholds designed to avoid trigger by background volcanic seismicity, may not be triggered by the seismically quieter snow-rich lahars. This was notably the case at Ruapehu in September 2007, where despite successful recording at RTMT, the lahar-warning system using AFMs was not triggered (H. Keys, personal communication, 2007). This is of concern, as ice-slurry lahars are highly efficient at transporting large volumes of ice and snow from volcanoes to generate significant downstream lahar hazards [Pierson et al., 1990].
 SEC thanks the Commonwealth Scholarship Scheme and Massey University Graduate Research School. SJC and VM acknowledge support from the Marsden Fund (MAUX0512) and SJC from the NZ Foundation for Research, Science and Technology (MAUX0401).