1. Introduction

In recent years, the accelerating mass loss of the Greenland ice sheet has raised global sea level by 0.67 mm a⁻¹(Reference Spada, Ruggieri, Sørensen, Nielsen, Melini and ColleoniSpada and others, 2012). It is equally divided between negative surface mass balance and dynamic discharge to the ocean, also known as ‘dynamic mass loss’ (e.g. Reference Rignot, Box, Burgess and HannsaRignot and others, 2008, Reference Rignot, Velicogna, Van den Broeke, Monaghan and Lenaerts2011; Reference Van den BroekeVan den Broeke and others, 2009; Reference Khan, Wahr, Bevis, Velicogna and KendrickKhan and others, 2010). Discharge occurs mainly through outlet glaciers, whose flow velocities often exceed 3 km a⁻¹(e.g. Reference Joughin, Smith, Howat, Scambos and MoonJoughin and others, 2010). The dynamic mass loss component has recently received much attention, as major outlet glaciers throughout Greenland started accelerating in the early 2000s. Although this phenomenon initiated at the more southerly outlet glaciers, it now affects outlet glaciers at all latitudes (e.g. Reference Rignot and KanagaratnamRignot and Kanagaratnam, 2006; Reference Pritchard, Arthern, Vaughan and EdwardsPritchard and others, 2009).

At the termini of Greenland’s outlet glaciers, mass discharge occurs mainly through iceberg calving, although at least in some regions submarine melting may be responsible for similar amounts of mass loss (Reference Rignot, Koppes and VelicognaRignot and others, 2010). In an effort to understand and quantify Greenland’s recent mass loss, iceberg calving has become the focus of vigorous glaciological investigation. GPS measurements (Reference Amundson, Truffer, Lüthi, Fahnestock, West and MotykaAmundson and others, 2008; Reference NettlesNettles and others, 2008) and theoretical studies (Reference ThomasThomas, 2004; Reference LüthiLüthi, 2009; Reference Nick, Vieli, Howat and JoughinNick and others, 2009) show that outlet glaciers can accelerate in response to large calving events. Consequently, calving processes may also have a direct effect on ice-sheet stability. Indeed, there exists strong evidence that despite significant regional variability (Reference Moon, Joughin, Smith and HowatMoon and others, 2012), much of the recent episodic discharge increase in Greenland was triggered at calving fronts when fjord waters underwent transient warming (Reference Holland, Thomas, de Young, Ribergaard and LyberthHolland and others, 2008; Reference MurrayMurray and others, 2010; Reference Rignot, Fenty, Menemenlis and XuRignot and others, 2012).

Despite its important role in glacier and ice-sheet mass balance, iceberg calving is still poorly understood. It remains a considerable challenge to incorporate the relevant boundary conditions (e.g. water and air temperatures, proglacial water depth, strain rates near the calving front, and fracture state of the terminus) into a ‘universal calving law’ (Reference Benn, Hulton and MottramBenn and others, 2007; Reference AlleyAlley and others, 2008; Reference Amundson and TrufferAmundson and Truffer, 2010; Reference BassisBassis, 2011). One obstacle facing iceberg calving studies is the lack of supporting data. Satellite images detect and quantify calving events at limited temporal and spatial resolution (e.g. Reference JoughinJoughin and others, 2008a). Direct visual observations using time-lapse cameras can provide impressively complete catalogues of calving events (e.g. Reference O’Neel, Marshall, McNamara and PfefferO’Neel and others, 2007; Reference Köhler, Chapuis, Nuth, Kohler and WeidleKöhler and others, 2011). However, these observations often come at high costs, are prone to subjective judgment of the observer, are limited to the seasons with daylight and provide only rough estimates of calving volumes.

Seismic monitoring of iceberg calving activity is an attractive alternative to direct observations, as large calving events can generate energy across a broad frequency range (e.g. Reference O’Neel, Marshall, McNamara and PfefferO’Neel and others, 2007, Reference O’Neel, Larsen, Rupert and Hansen2010; Reference Amundson, Truffer, Lüthi, Fahnestock, West and MotykaAmundson and others, 2008, Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and Motyka2012a). Use of seismic signals to monitor calving activity at various scales has previously been documented. Initial studies using sensors with limited frequency bands focused on high-frequency components (>1 Hz) of regional calving seismograms (e.g. Reference QamarQamar, 1988). More recently, such high-frequency signals with narrowband character (1–5 Hz) have been associated with interactions between detaching icebergs and the sea surface (Reference Bartholomaus, Larsen, O’Neel and WestBartholomaus and others, 2012) as well as various englacial fracture mechanisms that are active during calving events (e.g. Reference O’Neel and PfefferO’Neel and Pfeffer, 2007; Reference Amundson, Truffer, Lüthi, Fahnestock, West and MotykaAmundson and others, 2008; Reference Richardson, Waite, FitzGerald and PenningtonRichardson and others, 2010).

The deployment of broadband seismometers has facilitated the observation of the broadband nature of signals produced by calving events. ‘Glacial earthquakes’ (Reference Ekström, Nettles and AbersEkström and others, 2003; Reference Tsai and EkströmTsai and Ekström, 2007; Reference JoughinJoughin and others, 2008a; Reference Larmat, Tromp, Liu and MontagnerLarmat and others, 2008; Reference NettlesNettles and others, 2008; Reference Nettles and EkströmNettles and Ekström, 2010) are generated by major calving events. They are identifiable on global seismic networks via detection of low-frequency (35–150 s) surface waves generated during iceberg detachment. Recent work on records at both teleseismic and local distances indicates the source mechanism of these events is most likely collision forces between detaching icebergs and the glacier terminus or fjord bottom (Reference Amundson, Truffer, Lüthi, Fahnestock, West and MotykaAmundson and others, 2008; Reference Tsai, Rice and FahnestockTsai and others, 2008; Reference Nettles and EkströmNettles and Ekström, 2010; Reference Walter, Amundson, O’Neel, Truffer and FahnestockWalter and others, 2012). The glacial earthquakes detected at teleseismic distances have a surface wave amplitude equivalent to ∼M5 tectonic earthquakes. Most glacial earthquakes are generated during calving events in Greenland; however, Antarctic events have also been confirmed (Reference Nettles and EkströmNettles and Ekström, 2010; Reference Chen, Shearer, Walter and FrickerChen and others, 2011). Seasonal and secular changes in glacial earthquake detections have been associated with changes in dynamics of Greenland outlet glaciers (Reference Ekström, Nettles and TsaiEkström and others, 2006; Reference JoughinJoughin, 2006).

Despite these promising results, there remain open questions about the completeness of calving catalogues generated by ‘glacial earthquake’, or teleseismic surface wave detection, particularly since only some classes of large-scale calving events appear to generate significant low-frequency surface wave energy. Specifically, Reference Nettles and EkströmNettles and Ekström (2010) suggested that only capsizing icebergs appear to generate observable low-frequency surface wave energy, whereas tabular, non-capsizing icebergs can calve ‘quietly’. Similarly, they note that relatively few glacial earthquakes can be attributed to floating ice tongues, which suggests that coupling of the terminus to the solid Earth is essential for surface wave generation. Consequently, the glacial earthquake catalogue may not reflect the full extent of seasonal variations in calving activity, because large, tabular and smaller capsizing icebergs typically can calve at different times of the year (Reference Amundson, Fahnestock, Truffer, Brown, Lüthi and MotykaAmundson and others, 2010). Furthermore, the single-force source conventionally used to model glacial earthquakes (e.g. Reference Ekström, Nettles and AbersEkström and others, 2003; Reference Tsai and EkströmTsai and Ekström, 2007) provides a force amplitude, whose relationship with iceberg volume is not fully understood (Reference Amundson, Burton and Correa-LegisosAmundson and others, 2012b; Reference Walter, Amundson, O’Neel, Truffer and FahnestockWalter and others, 2012). Correspondingly, the upper limit for glacial earthquake size indicated by surface wave magnitudes (Reference Nettles and EkströmNettles and Ekström, 2010) may not only be the result of a limit of glacier calving volume.

Normal modes of proglacial fjord waters, which are excited as icebergs detach from glacier termini, are another type of calving-generated signal. The low-frequency (<0.01 Hz) signals from such ‘seiches’ can be recorded by seismometers in response to major calving events at Jakobshavn Isbræ, one of Greenland’s largest ice streams (Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others, 2012a). In other closed and semi-closed water bodies, such as lakes (e.g. Reference ForelForel, 1904) and harbours (e.g. Reference MilesMiles, 1974), seiches are commonly induced by winds. Seiches triggered through earthquakes (Reference KvaleKvale, 1955; Reference Ichinose, Anderson, Satake, Schweikert and LahrenIchinose and others, 2000), landslides (Reference Bondevik, Løvholt, Harbitz, Mangerud, Dawson and SvendsenBondevik and others, 2005) and ship traffic (Reference McNamara, Ringler, Hutt and GeeMcNamara and others, 2011) have also been reported.

In the present paper we show that, using data from three broadband stations in the vicinity of fjords into which calving occurs in Greenland, the long-period component of the seismic recordings can effectively be used to detect seiches. We analyse the performance of a purely automatic detection algorithm and discuss result improvement when some user interaction is added. We corroborate our detections using local tide gauge records, time-lapse photographs, satellite images and the teleseismic glacial earthquake catalogue (Reference Veitch and NettlesVeitch and Nettles, 2012). Furthermore, we analyse our derived catalogue to discuss calving activity of several Greenlandic glaciers. Due to their remote locations, the dynamics of these glaciers are difficult to study with in situ field techniques.

2. Seismic and Fjord Pressure Data

The Greenland Ice Sheet Monitoring Network (GLISN; http://glisn.info/; Reference Dahl-Jensen, Larsen and VossDahl-Jensen and others, 2010; Reference Husen, Clinton, Olivieri and GiardiniHusen and others, 2010) is a recent broadband seismic infrastructure in and around Greenland (Fig. 1). GLISN is a joint collaboration between the USA, Denmark, Switzerland, Germany, Canada, Italy, Japan, Norway, Poland and France. The purpose of this network is to enhance the capability of the pre-existing Greenland seismic infrastructure for detecting, locating and characterizing both tectonic and glacial earthquakes, together with other cryo-seismic phenomena. As glacial earthquake activity has been shown to exhibit seasonal and secular fluctuations likely caused by changes in full-thickness iceberg calving rate (Reference Ekström, Nettles and TsaiEkström and others, 2006), by improving the detection of these events the GLISN data can provide powerful monitoring of glaciological processes. All data from GLISN are freely and openly available through various institutes, including the ORFEUS Data Center (ODC) and the Incorporated Research Institutions for Seismology (IRIS). GLISN stations are almost uniformly equipped with Streckeisen STS-2 seismometers (flat response between 120 s and 50 Hz) and Quanterra Q330 digitizers. Except for on-ice and extremely remote stations, data are acquired and distributed in real-time.

Fig. 1. (a) Overview of GLISN seismic network installed across Greenland. (b–d) Detailed images of the regions near the seismic stations discussed (Landsat images and Landsat mosaic for NUUG region). A water pressure sensor was installed near the shore within 100 m of NUUG. At ILULI, at different periods, water pressure sensors were installed in Ilulissat harbor (ILULH, ∼3 km from seismic sensor) and at the mouth of the fjord (ILULS, ∼2 km distant). The location of another pressure sensor, ILULF, near the calving front of Jakobshavn Isbræ, is indicated by yellow dot.

The present study uses regional seismic broadband recordings of calving events in northwestern Greenland. We use data from three stations (KULLO, NUUG and ILULI; Fig. 1), which were installed in the summers of 2009 and 2010 near the shoreline in the vicinity of major calving glaciers. Since their installation the instruments have been operating continuously with only minor interruptions. For each of the three stations, this has resulted in a data return above 99% with only ∼20 data gaps per year.

Associated with this project, offline pressure sensors were temporarily installed in the fjords close to stations ILULI and NUUG. The sensor specifications and operational periods are shown in Table 1. The pressure data effectively measure water level above the sensor. In the event of seiching in the fjord, the water pressure sensors directly measure the amplitude of the seiche at the site, whereas the seismometers respond to ground tilt induced by the changing fjord water heights in the vicinity of the station. Seismometer response to tilt induced by water waves has been documented in a variety of previous cases, including seiche signals induced by ship traffic in the Panama Canal (Reference McNamara, Ringler, Hutt and GeeMcNamara and others, 2011) and tsunamis (Reference OkalOkal, 2007). A tilt signal on a seismometer is characterized by significantly larger signals on the horizontal components compared to the vertical (Reference Wieldant and ForbrigerWielandt and Forbriger, 1999).

Table 1. Operational periods and sampling rates for pressure sensors located in fjords near the seismic sensors. The channel names VDH and LDH conform to the Standard for the Exchange of Earthquake Data (SEED). Note that the numerical labels for ILULS channels refer to two different epochs

3. Study Sites on Greenland’s Northwest Coast

Installed in the town of Ilulissat in July 2009, the broadband station ILULI is located ∼60 km from the calving front of Jakobshavn Isbræ (Fig. 1), one of Greenland’s largest and fastest-flowing outlet glaciers draining ∼7% of the entire ice sheet (Reference BindschadlerBindschadler, 1984). Following ∼50 years of stability (Reference Sohn, Jezek and Van der VeenSohn and others, 1998), the glacier began a rapid retreat in 1998, losing its 15 km floating tongue (Reference Luckman and MurrayLuckman and Murray, 2005). The retreat was accompanied by thinning rates as high as 15 m a⁻¹ (Reference Thomas, Abdalati, Frederick, Krabill, Manizade and SteffenThomas and others, 2003; Reference KrabillKrabill and others, 2004), and the flow rate near the terminus accelerated from ∼6000 m a⁻¹ in 1997 to 12 000 m a⁻¹in 2003 (Reference Joughin, Abdalati and FahnestockJoughin and others, 2004). The glacier has maintained its high velocities, with calving front positions fluctuating seasonally on the order of 6 km (Reference JoughinJoughin and others, 2008b). There exists evidence that the glacier is currently undergoing dynamic changes allowing for iceberg discharge in winter (Reference Cassotto, Fahnestock and AmundsonCassotto and others, 2010; Reference FahnestockFahnestock and others, 2010; Reference Truffer, Amundson, Fahnestock, Motyka and JoughinTruffer and others, 2010). However, until recently, the glacier formed a small floating tongue in winter as calving typically ceased. This temporary tongue disintegrates in early summer via calving of large tabular icebergs (Reference Amundson, Fahnestock, Truffer, Brown, Lüthi and MotykaAmundson and others, 2010), the typical calving style for the previously floating terminus. During the subsequent summer, iceberg discharge occurs mainly via calving of smaller, full-thickness icebergs, which capsize upon detachment (Reference Amundson, Fahnestock, Truffer, Brown, Lüthi and MotykaAmundson and others, 2010). Jakobshavn Isbræ is the only major calving front in the vicinity of ILULI and thus the only candidate front for the events we observe at this station. The ice debris cover may not be typical of Greenland’s glacierized fjords: in recent years it has been present year-round, although its thickness and integrity change seasonally (Reference JoughinJoughin and others, 2008b; Reference Amundson, Fahnestock, Truffer, Brown, Lüthi and MotykaAmundson and others, 2010). In contrast, Reference Howat, Box, Ahn, Herrington and McFaddenHowat and others (2010) document the clearing of fjord debris cover for glaciers within 300 km north of Jakobshavn Isbræ. Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012a) demonstrate that calving events at Jakobshavn Isbræ produce seiche signals that are visible on station ILULI (Fig. 1). Up to six glacial earthquakes in a single year have been located at this calving front (Reference Veitch and NettlesVeitch and Nettles, 2012).

The broadband station NUUG is located just outside the small fishing community of Nuugaatsiaq in Greenland’s Uummannaq district. It was installed in July 2010 near glacier termini located within a system of fjords likely suitable for sustained seiches (Fig. 1). The calving fronts of Ingia and Umiamako Isbræ underwent rapid retreats in 2003, increasing their flow speeds by 20% and 300%, respectively (Reference Howat, Box, Ahn, Herrington and McFaddenHowat and others, 2010). Umiamako Isbræ retreated by ∼4 km, constituting the largest retreat in this area (Reference Howat, Box, Ahn, Herrington and McFaddenHowat and others, 2010; Reference McFadden, Howat, Joughin, Smith and AhnMcFadden and others, 2011). Between 2004 and 2008 it thinned by ∼66 m. Concurrently, it underwent acceleration, which may continue into the future (Reference McFadden, Howat, Joughin, Smith and AhnMcFadden and others, 2011). Kangerdlugssup Sermerssua has exhibited a stable calving front position although it doubled its speed between 2000 and 2005 and thinned by nearly 60 m in recent years (Reference Howat, Box, Ahn, Herrington and McFaddenHowat and others, 2010; Reference McFadden, Howat, Joughin, Smith and AhnMcFadden and others, 2011). During the same period, Rink Isbræ showed little change in terminus position and flow speed (Reference Howat, Box, Ahn, Herrington and McFaddenHowat and others, 2010; Reference Joughin, Smith, Howat, Scambos and MoonJoughin and others, 2010). Around the turn of the century, the discharge of Rink Isbræ was five times as large as Kangerdlugssup Sermerssua and about half as large as Jakobshavn Isbræ (Reference Rignot and KanagaratnamRignot and Kanagaratnam, 2006). It is therefore likely the dominant producer of icebergs in the Uummannaq district. This is also in agreement with glacial earthquake studies, which ascribe all detected events in this region (0–2 a⁻¹) to Rink Isbræ (Reference Tsai and EkströmTsai and Ekström, 2007; Reference Nettles and EkströmNettles and Ekström, 2010; Reference Veitch and NettlesVeitch and Nettles, 2012).

Near the broadband station KULLO, located at the village of Kullorsuaq since July 2009, the Greenland ice sheet drains through large ice streams and wide calving margins (Fig. 1). The ice streams calve directly into the open ocean, with buttressing rock outcrops forming only few water-filled fjords. Compared to the region near NUUG this suggests fewer water basins suited for seiches. Ice-sheet changes have been moderate over the past decade (Reference Rignot and KanagaratnamRignot and Kanagaratnam, 2006), except for Alison Gletscher (∼30 km from KULLO), which from around 2002 began an 8.7 km retreat and doubled the peak speeds (Reference Moon and JoughinMoon and Joughin, 2008; Reference Joughin, Smith, Howat, Scambos and MoonJoughin and others, 2010; Reference Howat and EddyHowat and Eddy, 2011; Reference McFadden, Howat, Joughin, Smith and AhnMcFadden and others, 2011). The glacier’s terminus and speed stabilized relatively quickly around 2007, arguably because of steep slopes reaching far inland (Reference McFadden, Howat, Joughin, Smith and AhnMcFadden and others, 2011). As expected for a rapid retreat, Alison Gletscher has produced up to four glacial earthquakes a year between 2003 and 2008 (Reference Veitch and NettlesVeitch and Nettles, 2012). In contrast, Hayes Glacier (∼40 km from KULLO) experienced a minor slowdown between the early and mid-2000s with approximately constant mass balance (Reference Rignot and KanagaratnamRignot and Kanagaratnam, 2006). Since 2006, it has only produced one glacial earthquake (Reference Veitch and NettlesVeitch and Nettles, 2012).

4. Event Characterization

Broadband calving seismograms recorded nearshore within 100 km of the source are characterized by a rich and distinctive wave train. Figure 2 demonstrates the different character of typically observed signals recorded on the north–south component at station NUUG. The records include a calving event from 23 August 2010. For this event, as well as for many calving events we have analysed, the north–south component most clearly exhibits the typical features of a calving seismogram as discussed below. Satellite images from the nearby glaciers taken before and after this date confirm the contemporaneous occurrence of iceberg calving and are discussed later. The general character of the calving seismogram in Figure 2 is also representative of other calving events at the seismic broadband stations ILULI, NUUG and KULLO. Figure 2 also includes signals from a large teleseism (M7.6 Kermadec Islands, 6 July 2011), a regional earthquake (M3.2, epicentre ∼120 km northwest of station ILULI, ∼200 km away from NUUG) and a sample of seismic noise.

In Figure 2a, all the records are shown without any processing or filtering. The calving event is characterized by a high-frequency onset followed by a nearly monochromatic long-period energy that resonates for many hours that is absent from all other signal types. Figure 2b shows the same data, after integration and bandpass-filtering between 0.001 and 0.01 Hz. Additionally, data recorded on the nearly co-located water pressure gauge are plotted. Though large teleseisms also excite energy with similar amplitude in this frequency band, the duration is short compared to the several hours of resonance seen during the calving. As the frequency, phase, envelope amplitude and duration of the calving seismogram closely match the water pressure data, it is clear that the seismometer is responding to the seiche measured with the pressure gauge.

In Figure 2c, all the seismic data are presented using a bandpass filter that accentuates large-amplitude surface waves at low frequencies (0.02–0.05 Hz). All signals use the same scale, except the teleseism, which is scaled down by a factor of 250. Though the calving event does generate low-frequency surface waves (in this case, two distinct peaks separated by 20 min which may be indicative of the detachment of several icebergs as documented by Reference Walter, Amundson, O’Neel, Truffer and FahnestockWalter and others, 2012), the amplitudes do not exceed the background noise by more than a factor of 5. In Figure 2d, the waveforms are bandpass-filtered between 1 and 3 Hz to highlight the high-frequency components of the signal. All the signals are on the same scale. The teleseism is at such great distance that the high-frequency signal is barely perceptible, with similar amplitude to the noise. The calving event produces clear energy only during the second of the clear low-frequency surface wave energy transients. During calving, high-frequency energy is generated during iceberg detachment (Reference O’Neel, Marshall, McNamara and PfefferO’Neel and others, 2007) and water surface impact (Reference Bartholomaus, Larsen, O’Neel and WestBartholomaus and others, 2012), as well as motion of ice debris in the fjord, and typically lasts several minutes (Reference Amundson, Truffer, Lüthi, Fahnestock, West and MotykaAmundson and others, 2008; Reference Walter, Amundson, O’Neel, Truffer and FahnestockWalter and others, 2012). The high-frequency calving seismicity typically precedes the seiche signal by some tens of minutes (Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others, 2012a). It tends to be emergent, and cultural noise can often generate a stronger signal.

The coincidence of the surface wave arrival (Fig. 2c) with a high-frequency seismicity burst is in agreement with the conceptual model that glacial earthquakes are generated by contact forces between a detaching iceberg and an obstacle coupled to the solid Earth, such as the fjord bottom or glacier terminus (e.g. Reference Amundson, Truffer, Lüthi, Fahnestock, West and MotykaAmundson and others, 2008, Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and Motyka2012a; Reference Tsai, Rice and FahnestockTsai and others, 2008; Reference Walter, Amundson, O’Neel, Truffer and FahnestockWalter and others, 2012). However, there also exist examples when seismic surface wave generation and high-frequency fracture seismicity do not clearly coincide (Reference Walter, Amundson, O’Neel, Truffer and FahnestockWalter and others, 2012). For the example shown in Figure 2 this is the case for the first surface wave arrival. This may indicate that during a multiple-iceberg calving event, some icebergs detach and capsize more ‘freely’ with relatively little englacial fracturing or displacement of fjord debris cover.

Fig. 2. Comparison of different signals recorded on the north–south component at NUUG. Records are presented from seismic background noise, a local calving event (23 August 2010), a teleseismic event (M_w = 7.6; Kermadec Islands, 6 July 2011) and a regional earthquake (M_w = 3.2 on 20 August 2010 at 20:53 UTC, ∼120 km northwest of ILULI, 200 km south of NUUG; Fig. 1). (a–d) share the same scale on the x-axis (note that the scale bar in (c) is omitted for clarity). Seismograms have not been corrected for instrument response. Note that in (b–d) the time series shown in (a) (raw data) is integrated. The low-frequency corner of the flat response of the STS-2 sensor is 0.0083 Hz (120 s), hence the time series in (c) and (d) are proportional to displacement. The vertical axes of (a) and (b) are labelled ‘Velocity*’ and ‘Displacement*’ respectively, because the signals include frequencies outside the sensor’s flat response. (e) The respective low-frequency velocity spectra. For the same calving event, pressure data from the NUUG fjord hydrograph (location shown in Fig. 1) are plotted in (b) and (e). Each seismogram in (a) is normalized to its maximum. For (c), the blue number indicates that the scale of the teleseism seismogram was reduced by a factor of 250 with respect to the other seismograms.

In Figure 2e, the long-period components of the signals are displayed in the frequency domain. The calving event is characterized by a number of narrow and distinct spectral peaks. The pressure gauge data also share several of these peaks, in particular the dominant resonance near 0.006 Hz. None of the other event types excite significant energy near this frequency. The seismic spectrum contains various additional low-frequency peaks, which are only weakly present in the fjord pressure data. This may be due to the nature of the sensor installation: near NUUG, the fjord system is complex (Fig. 1). Depending on the specific glacier terminus, any single calving event may induce seiching with different resonating frequencies in nearby basin systems. Whereas the water pressure sensor is only sensitive to resonances in the local fjord, the seismic sensor can respond to seiching-induced tilts that occur in nearby basins with different resonance periods.

Figure 2 demonstrates why we focus on identification of calving events using long-period energy (<0.01 Hz) associated with seiching. Regional and teleseismic earthquakes and even seismic background noise can generate energy in all three frequency bands shown in Figure 2. However, the 0.001–0.01 Hz range is least contaminated by other signals, which include very low-frequency surface waves from relatively rare major teleseismic events, local tilting from wind or cultural noise, and instrument glitches such as mass re-centrings. Cultural noise and electronic instrument glitches, in particular, can produce false alarms for our automatic calving seiche detector at station KULLO, as discussed below. Nevertheless, these sources have a far shorter duration than seiche signals. Furthermore, in this frequency band, teleseismic events can be clearly distinguished from seiches by comparison between vertical and horizontal amplitudes: teleseisms have similar amplitudes between these components, but in the case of calving seiches the sensor responds to minor tilts. In this case the vertical component has order-of-magnitude smaller amplitudes than the horizontal components (Reference ClintonClinton, 2004; Reference PinoPino, 2012). Accordingly, the seiche seismogram in Figure 2 has a horizontal-to-vertical (H/V) ratio of 20 (using the north–south component). In contrast, the teleseismic surface waves typically have an H/V ratio of ∼1.

Although calving events generate seismic signals in each frequency band, the seiche signal consistently has the best signal-to-noise ratio. Indeed, during some seiche detections, surface wave and/or the high-frequency energy does not emerge above the noise even at distances of less than 100 km. Furthermore, these higher-frequency bands are rich in frequently occurring transient signals produced by cultural noise and, in particular, earthquake sources. The 0.001–0.01 Hz passband is therefore suitable for automatic detection of iceberg calving events using the seiche approach.

5. Seiche Detection

Our strategy for the automated detection of calving events is to target the seiche signals. In order to identify all transient energy signals at long periods, we apply a simple triggering detector to the bandpassed continuous waveforms. In order to maximize the completeness of this first-stage catalogue we use a conservative parameter set for the triggering algorithm. The detector was first tuned using known calving events from first-hand observations at Jakobshavn Isbræ (personal communication from M. Lüthi, 2011) and the glacial earthquake catalogue (most recent update from Reference Veitch and NettlesVeitch and Nettles, 2012). Our conservative parameter choice also produces a significant number of false detections. We present an automated approach to solving this problem, and a second solution which requires user interaction. For the latter we manually review all STA/LTA (short-term average/long-term average) detections. This is the most reliable, although tedious, solution for removing the false detections from the automatic algorithm. For the automated approach, we propose a second stage that filters the initial catalogue, only including events in the final automatic catalogue that exhibit certain characteristics expected of seiches. The first check in this second stage discriminates seiching from other signals by setting a threshold for signal duration. This removes brief transients and all but the largest teleseisms. A further check that removes teleseismic signals requires the calving events to have a minimum ratio between the peak amplitudes of the horizontal and vertical components. A final check that removes spurious noise, or indeed small calving events that are difficult to verify, requires events to reach a minimum amplitude for the characteristic spectral peaks. As calving events from different sources exhibit quite different characteristics – in particular, the seiche frequencies and durations vary according to the fjord geometry – the actual threshold parameters are station-dependent (Table 2).

Table 2. Individual station parameters for automatic seiche detection. Columns 2–4 define the initial detection phase: high-pass (HP) and low-pass (LP) corners of the four-pole bandpass filter; lengths used to define the short-term (STA) and long-term (LTA) averages for the trigger; threshold of STA/LTA algorithm to initiate and terminate detection, indicating event duration. Columns 5–7 describe the parameters used in the secondary event filtering: minimum event duration; frequency bands for determining station-dependent characteristic amplitude; minimum ratio of the average horizontal to vertical components within the bandpass

5.2. Detection verification

Visual calving event confirmation with first-hand observations (e.g. Reference O’Neel, Marshall, McNamara and PfefferO’Neel and others, 2007; Reference Köhler, Chapuis, Nuth, Kohler and WeidleKöhler and others, 2011) or time-lapse photography (Reference Amundson, Truffer, Lüthi, Fahnestock, West and MotykaAmundson and others, 2008, Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and Motyka2012a; Reference Walter, Amundson, O’Neel, Truffer and FahnestockWalter and others, 2012) is unrealistic for the 250 detected calving events, which occurred over >2 years in extremely remote terrain. Instead, when possible, we verified our detections with water pressure data in nearby fjords. However, the pressure sensors were only installed near ILULI and NUUG (Table 1), and even for these data there are extended periods without measurements. Furthermore, a potential problem with pressure gauge data is that occasionally they may record seiches caused by mechanisms other than calving, such as freely rotating icebergs, landslides or weather-related events.

Satellite images are an alternative to direct observations. They can be used to verify that our automatic seiche detections correspond to iceberg calving events. However, the spatial resolution of satellite images is limited. In addition, cloud-free image pairs are rarely taken immediately before and after a calving event, so the temporal resolution of satellite-based calving event detection cannot compete with seismic methods. In this study, we use an image library provided daily by the Danish Meteorological Institute (http://ocean.dmi.dk/arctic/modis.uk.php). It contains data from the Envisat satellite and its Advanced Synthetic Aperture Radar (ASAR), © European Space Agency. The images have sufficient resolution (∼150 m) to identify major calving episodes and are generally available at intervals of tens of days or less. A systematic effort to create an independent calving catalogue for each calving front by using satellite images to monitor changes in terminus geometry could be used to estimate the completeness of our calving catalogue. Though this is beyond the scope of the present study, we demonstrate the potential of satellite images for calving event confirmation in Figures 3 and 4. The shown events at KULLO and NUUG do correspond to coeval mass wastage at a calving front.

Fig. 3. Envisat ASAR image pair from terminus of Alison Gletscher taken about 11 hours before (top) and 2.5 days after (bottom) calving detection at KULLO on 30 January 2011 at 10:53. The red line traces the pre-calving terminus and highlights the calved area in the bottom image (blue circle). In addition to the chunk missing from the terminus, there is a major change in ice melange near the terminus.

Fig. 4. Envisat ASAR image pair of the NUUG region (Uummannaq district), taken about 5 days before (left) and 1.5 days after (right) the seiche detection on 23 August 2010 at 03:20 (Fig. 2). The pre- and post-calving termini are difficult to trace in these images (due to lighting and the presence of melange cover); however, the appearance and changes of the proglacial melange (red circles), in particular at Ingia Isbræ (I.I.) and Kangerdlugssup Sermerssua (K.S.), suggest calving events likely happened in between the images. R.I. refers to Rink Isbræ (shown in Fig. 1).

Figure 3 shows ASAR images of Alison Gletscher’s terminus near KULLO taken on 30 January 2011 and 2 February 2011 (see also Fig. 1). A seiche was detected within this time window at KULLO on 30 January 2011 at 10:53. The second image presented clearly shows a missing piece in the terminus due to one or more calving events. The image pair furthermore highlights a change in the proglacial melange cover, which may be due to upwelling of fresh water from subglacial discharge (e.g. Reference Motyka, Hunter, Echelmeyer and ConnorMotyka and others, 2003; Reference Rignot, Koppes and VelicognaRignot and others, 2010) or local winds.

There exist ASAR images taken ∼5 days before and 1.5 days after the calving seiche detection at NUUG on 23 August 2010 at 03:20 (Figs 2 and 4). On these images it is more difficult to trace the terminus positions and to detect changes in terminus geometry. There are changes in proglacial melange cover at all shown calving fronts. However, as these changes may be due to changes in fjord currents or wind patterns they carry little significance. On the other hand, we note the appearance of severalkilometre-long debris covers in the fjords of Ingia Isbræ (I.I.) and Kangerdlugssup Sermerssua (K.S.), which suggests that substantial calving occurred around the detection time.

5.3. Detection statistics

Figure 5 illustrates the statistics of our automatic detection time series with events binned in 1 month intervals. The figure also shows the result of the visual review of all initial STA/LTA detections, i.e. all automatic detections as well as those rejected by the thresholds for amplitude and H/V ratio. Moreover, where possible, we visually confirmed that the seiche signal was present in the fjord pressure sensor data. This manually reviewed catalogue best represents the calving activity derived from our seiche detections. The manual calving catalogues for each station are included in Table 3 in the Appendix. Significant additional confidence in the manual catalogue was gained by visual inspection of the entire 2 year continuous seismic archive subjected to a bandpass filter between 200 and 1000 s. We note that missed events can occur due to gaps in data, which also produces a blind period for the detector, corresponding to the duration of the long-term average window. However, this occurs rarely, as the data return of all three stations exceeds 99%. Furthermore, data gap occurrence is not influenced by the seasons and thus cannot mimic or mask seasonal fluctuations in seiche detection.

Fig. 5. Automatic (black) and manual (red) detections per month at the three seismic stations. Note that in contrast to KULLO and ILULI, the NUUG detection time series shows a seasonal fluctuation. During some months in 2011, the number of automatic detections at KULLO exceeds 50 and thus the limits of the vertical axis.

Figure 5 shows that the automatic catalogue can miss up to six events per month. Furthermore, it can include an even greater number of false triggers. The problem of false detections is particularly severe for KULLO, which has >50 monthly false detections in summer 2011. In total, 79% of all KULLO detections were false, while the automatic detector incorrectly rejected 18% of the manually confirmed events. Visual inspection of the continuous waveforms shows that station KULLO generates a large amount of long-duration long-period noise, which falsely triggers our detection algorithm. Similar to data gaps (also more frequent in the KULLO record), such noise can decrease the trigger sensitivity for several hours, potentially leading to missed calving detections. We believe this noise occurs because, though the sensor is located on rock, it is underneath an occupied house near one of the four foundation piers. It may thus be susceptible to local tilting induced by transient forces on the foundation due to wind or house use.

In an attempt to reduce the high rate of false detections at KULLO, we varied the amplitude and duration cut-off values (Table 2), but were not able to significantly improve performance. On the other hand, the performance of the automatic detector at ILULI and NUUG is more reliable, with 17% and 6% of the manual detections missed respectively. Conversely, 29% and 33% of the automatic detections for the respective stations were false.

6. Comparison with other Catalogues

To further evaluate the performance of our calving detector we compare our automatic and manual catalogues with existing iceberg calving catalogues based on glacial earthquake detection (Reference Veitch and NettlesVeitch and Nettles, 2012) and a calving catalogue derived from a combination of seismic and visual observations (Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others, 2012a). It should be stressed that these catalogues all use different datasets and use different frequency bands of calving seismograms. Consequently, the detection techniques may well be sensitive to different styles and sizes of calving events. Therefore, the ideal catalogue would most likely be a combination of the existing catalogues and our manual seiche detection record.

We focus on the detections of Reference Veitch and NettlesVeitch and Nettles (2012) in our three candidate regions. This catalogue spans the years 2006–10 and thus builds on earlier glacial earthquake catalogues of Reference Tsai and EkströmTsai and Ekström (2007). For the particular case of calving events from Jakobshavn Isbræ, our catalogues are compared with the updated version (personal communication from J.M. Amundson, 2012) of the Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012a) catalogue. Whereas the Reference Veitch and NettlesVeitch and Nettles (2012) catalogue also served as a guideline for the choice of our detection parameters, the updated version of Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012a) was not used in the development of our trigger algorithm. The Amundson catalogue documents the calving history of Jakobshavn Isbræ between 1996 and 2011 and is based on a manual analysis of local high-frequency seismicity, satellite imagery and, when possible, time-lapse footage of Jakobshavn Isbræ’s terminus. For Jakobshavn Isbræ, this is presently the most complete record of calving events. Requiring low-frequency surface wave energy recorded at global distances, the trigger algorithm of Reference Veitch and NettlesVeitch and Nettles (2012) targets relatively large calving events only. However, it can potentially detect calving events at all major outlet glaciers of Greenland.

Our automatic seiche detections include all but one of the ten glacial earthquake events (nine detections at Jakobshavn Isbræ and one detection at Rink Isbræ) detected by Reference Veitch and NettlesVeitch and Nettles (2012). Our automatic and manual catalogues at ILULI are compared with the updated catalogue of Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012a) in Figure 6. Figure 6a compares their catalogue with our automatic catalogue. The number of events occurring each month detected by either catalogue varies between 0 and 10. During most months, there exist events that are detected by both catalogues, as well as events that are detected only by one of the two methods. When we consider our manual catalogue (Fig. 6b), the number of events only present in our seiche catalogue decreases slightly while the number of events observed by both methods increases. This confirms (1) our manual catalogue, based on a visual inspection of all STA/LTA triggers, correctly eliminates a number of false calving events in the automatic catalogue; and (2) in a small number of cases, a few events that match the STA/LTA trigger criterion but are subsequently rejected are likely true events as they also appear in the updated catalogue of Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012a). On the other hand, ∼30 calving events (the summation of green bars in Fig. 6b) in the updated catalogue of Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012a) do not trigger the STA/LTA algorithm. The reason is most likely that these calving events do not produce a seiche strong enough to be detected by our seismic sensor at ILULI.

Fig. 6. Comparison between monthly detection in the automatic (a) and manual (b) catalogues (‘seiche’) and the updated catalogue by Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012). Note that this comparison is limited to calving events at Jakobshavn Isbræ detected at station ILULI.

The majority of events in the manual catalogue are also in the updated catalogue of Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012a). Combining both catalogues, there are on average four to five events at Jakobshavn Isbræ per month. Nevertheless, there are events (represented by red bars) which produce seiches but which are not included in the updated version of Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012a). These are less straightforward to interpret. One reason may be that events in the Amundson catalogue are required to have observable high-frequency seismic energy and they have to be visible in satellite and/or time-lapse images. Furthermore, Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012a) only consider calving events, which produce multiple icebergs as indicated by more than one burst in high-frequency seismicity. Consequently, the seiche catalogue may in fact identify previously unknown calving events that may be too small to be visually identified or emit significant high-frequency energy. At the same time we cannot exclude the possibility that at least occasionally our trigger algorithm may also detect seiches which are unrelated to calving events (e.g. freely capsizing icebergs, strong wind or landslides).

7. Discussion

7.2. Changes in signal characteristics

The recordings at NUUG and KULLO show seasonal changes of seiche amplitudes (calculated as maximum amplitude recorded on any component during the seiche). This is most clearly seen on the east–west component shown in Figure 7. At NUUG and KULLO, seiche amplitudes tend to be highest during late summer. Such an amplitude variation over time and season cannot be observed in the corresponding low-frequency surface wave frequency band on the same component (Fig. 8).

Fig. 7. Maximum amplitudes on east (E)-component seiche seismograms for all calving events in the manual catalogue (see Table 2 for bandpass used for each station). Note that the seiche signals at KULLO and NUUG exhibit a clear seasonal fluctuation, with relatively strong amplitudes in late summer and early autumn.

Fig. 8. Maximum amplitudes on east (E)-component calving seismograms in the 0.0167–0.033 Hz range (long-period surface wave energy), for all calving events in the manual catalogue. In contrast to the seiche signal amplitudes (Fig. 7), no seasonal variation is apparent in the surface wave amplitude at any station.

At periods between 0.1 and 1 Hz, sea ice can substantially dampen ocean gravity waves in Arctic and Antarctic waters (Reference Grob, Maggi and StutzmannGrob and others, 2011; Reference Tsai and McNamaraTsai and McNamara, 2011). However, recent observations (Reference Bromirski and StephenBromirski and Stephen, 2012) of ice-shelf response to infra-gravity waves only document a minor damping effect of sea-ice cover at lower frequencies (0.004–0.02 Hz). The seiche signals observed at KULLO, NUUG and ILULI contain energy between 0.001 and 0.01 Hz, which suggests that the observed seasonality in seiche amplitudes cannot be fully attributed to damping effects of sea ice. Instead, we suggest that seasonal changes in seiche amplitudes are the effect of changes in melange and sea-ice cover on the generation rather than the transmission of calving seiches. Laboratory studies (Reference Amundson, Burton and Correa-LegisosAmundson and others, 2012b; Reference BurtonBurton and others, 2012) as well as theoretical (Reference Tsai, Rice and FahnestockTsai and others, 2008; Reference Amundson, Fahnestock, Truffer, Brown, Lüthi and MotykaAmundson and others, 2010) and observational studies (Reference Walter, Amundson, O’Neel, Truffer and FahnestockWalter and others, 2012) suggest that melange cover in the fjord affects not only the detachment but also the capsizing of icebergs by adding drag forces during the rotation. Accordingly, less rotational energy may be converted into standing fjord waves when a thick sea-ice or melange cover is present, resulting in lower seiche amplitudes.

The frequency content of seiche signals also exhibits considerable temporal variation, which is particularly pronounced on the north–south components of NUUG (Fig. 9). Here a strong spectral peak at 0.006 Hz is particularly interesting, as it is absent during the earlier months of 2011 and immediately following the station installation. There are several potential explanations for changing boundary conditions, which may give rise to the observed variation in spectral character. First, individual spectral peaks may be characteristic for a particular fjord basin (e.g. Reference LeeLee, 1971). However, this cannot explain spectral changes of records at ILULI, which we have argued is only sensitive to calving events from Jakobshavn Isbræ. Furthermore, this line of reasoning would imply that, near NUUG, different glaciers calve during different times of the year, which seems unlikely for neighbouring glaciers. An alternative explanation could be seasonal fluctuations in terminus positions. Although terminus positions currently vary <1 km in the NUUG region (Reference Howat, Box, Ahn, Herrington and McFaddenHowat and others, 2010), these fluctuations may be enough to subtly but measurably alter the resonance frequencies of fjord water bodies. Finally, our preferred explanation for changing seiche spectra is the variation in proglacial ice melange cover combined with the seasonal presence of sea ice. This is supported by recent numerical calculations (Reference MacAyeal, Freed-Brown, Zhang and AmundsonMacAyeal and others, 2012) showing that the presence of closely spaced icebergs within a fjord introduces ‘band gaps’ in normal oscillations of proglacial fjords. In addition, such closely spaced icebergs can lengthen the periods of certain seiche modes by >400%.

Fig. 9. Spectra from seiche seismograms for all calving events in the manual catalogue. Waveforms have instrument response removed. North (n) components are shown. Events are sorted by date of occurrence. Note the relative amplitudes and periods of spectral peaks vary over time. At NUUG, the peak at 0.006 Hz (black arrow) has dramatic changes in amplitudes. At ILULI, the fundamental frequency peak appears to lie closer to 0.003 in late summer and 0.004 in late winter.

7.3. Changing seismic background noise

In order to understand whether our detection threshold was being influenced by variations in seismic noise, we followed the method of Reference Vila, Macià, Kumar, Ortiz, Moreno and CorreigaVila and others (2006). We measure seismic background noise levels in the seiche (0.001–0.01 Hz) and low-frequency surface wave (0.02–0.05 Hz) frequency range for the three stations over the entire available dataset. To estimate the noise for each day, the average amplitude of the bandpassed velocity signal over a 3 hour time window is computed, and then the minimum 3 hour average amplitude value for each day is selected. Selecting a 3 hour period to estimate noise removes transient spikes from, for example, earthquakes that may occur elsewhere in the day, but also provides a fair indication of the average background noise, rather than the minimum. We can compare this noise measure with the recorded peak seiche amplitudes. Figure 10 shows that in both frequency ranges, the events detected by our method generally carry energy significantly above our measure of background noise. Exceptions are the LHN components of NUUG and KULLO.

Fig. 10. Average daily seismic noise levels in the seiche frequency (0.001–0.01 Hz; red) and surface wave frequency (0.02–0.05 Hz; blue) bands at ILULI, NUUG and KULLO. Triangles indicate the maximum amplitudes of calving seismograms in the manual catalogue in the same respective frequency band. The algorithm detects calving events using the vertical component, and though the amplitude of the signal tends to be smaller than the horizontal component because the noise is substantially lower, the signal to noise is most favourable on this component.

We can draw several specific conclusions from Figure 10: First, the noise level of the horizontal components generally exceeds the noise on the vertical components by an order of magnitude. Thus, calving seismograms on the vertical component, on which our STA/LTA detector operates, have the best signal-to-noise ratio due to the low noise levels on these components. Second, there exist noise variations on a seasonal timescale in both the seiche and surface wave bands. On the other hand, the NUUG record shows that a peak in seiche detections does not coincide with the annual minimum in background noise. The last point, in particular, indicates that the seasonality in seiche detections at NUUG (Fig. 5) is not an artefact of seasonal changes in background noise levels. Concerning the reason for seasonal noise level variations we note that they are most evident on the horizontal components of ILULI and KULLO and have a similar phase in both frequency ranges, with noise minima and maxima occurring in the first and second half of 2010, respectively. Though it is possible that the noise is associated with external factors, we believe there may be a more simple explanation. The seismometers at these two sites are placed under residential buildings, as although the ambient noise is high, the sites are secure, protected from the elements and provide power and communications. Occupation is subject to weather conditions as well as human habit.

7.4. Comparison with global surface wave detector

Our calving detection catalogue (manually confirmed detections) consists of 257 events. This includes nine of the ten events detected as glacial earthquakes by teleseismic surface wave detection (nine detections at Jakobshavn Isbræ and one detection at Rink Isbræ; Reference Veitch and NettlesVeitch and Nettles, 2012) during the same time period and within the same regions. The single missed event occurred on Jakobshavn Isbræ on 21 May 2010; the seiche detector rejected the event because the seiche duration of 300 s is well below the discrimination threshold (1200 s), and the seiche amplitude is small.

The large discrepancy in number of teleseismic surface wave detections and local seiche detections indicates that the second approach is significantly more sensitive. It should be noted that observed glacial earthquakes in the Reference Veitch and NettlesVeitch and Nettles (2012) catalogue are close to the glacial earthquake detection threshold. For instance, during a 50 day period in 2007, Reference NettlesNettles and others (2008) report six automatic glacial earthquake detections at Helheim Glacier, East Greenland, with magnitudes of 4.5–4.8. However, a more sensitive, interactive detection method identified five additional events. Another reason may be that the generation of glacial earthquakes during iceberg calving is more dependent on calving style and terminus geometry than is the generation of seiches. Specifically, in order to generate sufficient energy for detection as a glacial earthquake, it appears necessary that capsizing icebergs collide with the glacier terminus or fjord bottom (Reference Amundson, Truffer, Lüthi, Fahnestock, West and MotykaAmundson and others, 2008; Reference Tsai, Rice and FahnestockTsai and others, 2008; Reference Nettles and EkströmNettles and Ekström, 2010; Reference Walter, Amundson, O’Neel, Truffer and FahnestockWalter and others, 2012). Iceberg calving from floating termini likely transfers less low-frequency surface wave energy into the solid Earth (Reference Nettles and EkströmNettles and Ekström, 2010), and likely inhibits the detection of calving from floating termini using this method. In this context it is interesting to note that between 2008 and 2010 no glacial earthquakes were detected at Alison Glacier, near KULLO (Reference Veitch and NettlesVeitch and Nettles, 2012). The reason may be that from ∼2009 the calving front was afloat, and calving of tabular, non-capsizing icebergs became the predominant mechanism (Reference Veitch and NettlesVeitch and Nettles, 2012). On the other hand, we continue to observe calving seiches at KULLO during this period (Figs 5 and 6), and Alison Glacier likely produces a substantial fraction of these events.

By targeting an entirely independent physical mechanism, our seiche detector can serve as a complement to, and verification of, existing seismic detections of iceberg calving. It remains to be shown to what extent seiches can be generated by calving of non-capsizing icebergs (e.g. tabular icebergs calving from floating tongues). If sensitive to tabular iceberg calving, our detection scheme may find events which at distances of tens of km are more difficult to identify with higher-frequency detectors. At the same time, seiches could be generated by freely floating icebergs, which capsize in the middle of the fjord and may produce false detection in our calving catalogue. At least occasionally, such spontaneous capsizing can occur, and it can generate a typical high-frequency (>2 Hz) calving signature (Reference Amundson, Fahnestock, Truffer, Brown, Lüthi and MotykaAmundson and others, 2010). These cases may be responsible for at least part of the discrepancy between our catalogue at ILULI and the updated version of the catalogue by Reference Amundson, Clinton, Fahnestock, Truffer, Lüthi and MotykaAmundson and others (2012a). However, if and how often such spontaneous capsizing events can ‘falsely’ trigger our seiche detection algorithm should be systematically investigated using satellite images.

If the magnitudes of the teleseismic surface wave detections and seiche amplitudes scaled with calving volume, for example, one would expect a clear, perhaps even linear, relationship between the amplitude of the low-frequency surface waves and seiche signals at the local station. Figure 11 suggests that this is not necessarily the case: on the east–west component, only a weak trend between the two signal amplitudes exists, with some of the strongest (weakest) seiche signals associated with some of the strongest (weakest) surface wave amplitudes. The absence of a clearer relationship is not surprising considering that the relationship between low-frequency surface wave signals and volumes of detaching icebergs is likely complicated. Hydrodynamic drag forces and melange cover play an important role in the generation of contact forces between detaching icebergs and the glacier terminus, which in turn can cause significant low-frequency surface wave energy (Reference Tsai, Rice and FahnestockTsai and others, 2008; Reference Amundson, Burton and Correa-LegisosAmundson and others, 2012b; Reference Walter, Amundson, O’Neel, Truffer and FahnestockWalter and others, 2012). Accordingly, more theoretical analysis is needed in order to accurately link centroid single force amplitudes (Reference Ekström, Nettles and AbersEkström and others, 2003) to calving volumes. Similarly, one should investigate the influence of hydrodynamic drag forces on seiche amplitudes.

Fig. 11. Comparison between amplitudes of surface wave (0.0167–0.033 Hz) and seiche signals (0.0012–0.007 Hz) at each of the three stations. East (E) component is shown. There is no clear correlation between the amplitudes of seiche and surface wave signals at any of the locations.

An open question and interesting prospect for future work therefore concerns the existence of a clear physical relationship between the observed seiche characteristics and calved iceberg properties such as location, volume or style of calving. Our catalogue indicates there are considerable variations in seiche amplitude, with little correlation with low-frequency seismic surface wave amplitude. Future investigations should aim to establish a physical meaning for seiche signal characteristics, such as maximum amplitude, durations and dominant frequency. This would mean that monitoring calving-induced seiches could provide additional quantitative monitoring of calving volume.

8. Conclusion and Outlook

The present investigation demonstrates that fjord seiche detection can be used to identify and monitor calving events in northwest Greenland. Seiching events are clearly visible as tilt signals on broadband seismometers located near the shoreline in the vicinity (within 100 km) of calving fronts. At this point, the most reliable seiche-based calving catalogue can only be attained after visual review of continuous seismic records and the output of trigger algorithms. However, although the numbers of missed events and false detections given by our automatic detector can be considerable and at certain times dominant, the presented results nevertheless indicate that real-time seismic monitoring of major iceberg discharge events with little or no user interaction is, in principle, possible. In its current state, the automatic catalogue is already capable of highlighting qualitative aspects of calving activity in a specific region, such as the seasonal fluctuations near NUUG (Fig. 5). According to our background noise analysis, this pattern cannot be attributed to seasonal changes in trigger sensitivity. Consequently, the automatic catalogue could already be used as a monitoring tool to identify transient and long-term changes in calving activity.

Nevertheless, at this point, instrumental and/or site noise inhibits reliable detection records at one particular station, KULLO. Incorporating more sophisticated waveform recognition techniques will likely mitigate this problem and significantly improve the overall performance of our detection algorithm. This may also allow inclusion of additional seismic signals associated with iceberg calving such as high-frequency (>1 Hz) seismicity and long-period (35–150 s) surface wave energy. Another possibility would be to include other features of seiche seismograms, such as polarization and relative heights of spectral peaks.

The calving catalogue for NUUG identified seasonal fluctuations in calving activity at the glaciers in the Uummannaq district (Fig. 1). One explanation could be the previously reported seasonal fluctuations in terminus positions (Reference Howat, Box, Ahn, Herrington and McFaddenHowat and others, 2010). For Jakobshavn Isbræ, the occurrence of events throughout the year at ILULI is consistent with the observation that the glacier had started to calve in winter as previously suggested by other workers (e.g. Reference FahnestockFahnestock and others, 2010; Reference Podrasky, Fahnestock, Amundson, Cassotto and JoughinPodrasky and others, 2012). These observations indicate that seismic monitoring of calving activity using seiche signals can serve as a means to study changes in glacier dynamics, which have been attracting much scientific attention, especially in Greenland.

The present seiche detection technique should be extended to incorporate more GLISN stations at the periphery of the Greenland ice sheet. At the same time, it will be important to establish relationships between physical iceberg parameters and seiche duration, amplitude and frequency content. High-resolution satellite images and denser networks of seismometers and fjord pressure sensors will be a valuable tool for this task. Ideally, this will allow for identification of various calving styles and estimation of calving volumes, which are difficult to quantify with existing methods.