The transient stress change caused by the passage of surface waves resulting from the 26 December 2004 Sumatra–Andaman earthquake triggered seismic tremor in two regions in Hokkaido, in the northernmost part of the Japanese islands, where tectonic tremor associated with the subducting plate has not previously been detected. The amplitude pattern of the tremor envelope in both regions is characterized by a periodic enhancement at an interval of about 20 s, which correlates with the surface wave. One tremor that was triggered in central Hokkaido at a depth of around 10–20 km coincides with active seismicity linked to previously known, deep low-frequency microearthquakes related to volcanic activity. Another tremor occurred in northernmost Hokkaido, where there are no known active faults, volcanoes, or microearthquake seismicity. If the source of the northern tremor is located near the ground surface, it would be possible that the tremor is related to fluid pressure change, because the periodic enhancement of tremor amplitude is in phase with the largest compressional strain caused by surface waves.
 Triggered seismic phenomena reflect the Earth's crustal response by transient stress changes, providing an opportunity to better understand earthquake mechanisms, even though the triggering mechanism remains unclear [Hill and Prejean, 2007]. Microearthquake seismicity that is remotely triggered by the passage of surface waves from large teleseismic events is well known in regions of volcanic and geothermal activities [e.g., Hill et al., 1993]. Generally, remote triggering phenomena are recognized as changes in the rate of seismicity that occur after surface wave arrival [e.g., Hill et al., 1993]. On the other hand, West et al.  detected periodically triggered seismicity of microearthquakes at Mount Wrangell, Alaska, during surface wave propagation from the 2004 Sumatra–Andaman earthquake. Microearthquakes occurred at intervals of 20–30 s in phase with the largest positive vertical ground displacement during Rayleigh wave motion. Wu et al. reported microearthquakes in a non-volcanic region in China, coincident with the first cycles of Love waves from large teleseismic events.
 Another significant triggered phenomenon is non-volcanic tremor [Obara, 2002]. In southwest Japan and Cascadia, within the deeper portion beneath the locked zone along the plate interface, tremor episodes associated with short-term slow slip events (SSE) occur periodically with a specific recurrence interval for each segment, which is 6 months in Shikoku, Japan [Obara, 2011] and 14 months in northern Cascadia [Rogers and Dragert, 2003]. These SSE associated with tremor episodes usually last for several days to weeks. In contrast, short-duration tremor activity is sometimes linked to the propagation of the long-period Rayleigh wave [e.g.,Miyazawa and Mori, 2006] or Love wave [e.g., Rubinstein et al., 2007] resulting from large teleseismic events. Triggered tremor is very well correlated with each surface wave phase, indicating that tremor is very sensitive to the change in the stress field. To date, triggered tremor has been detected not only in young subduction zones, but also along the San Andreas Fault [Gomberg et al., 2008], in central Taiwan [Peng and Chao, 2008], in the North Island of New Zealand [Fry et al., 2011], and in other regions. Given that other studies have shown that triggered tremor share a common source with ambient tremor [Miyazawa et al., 2008], which is usually coincident with slow slip [Obara, 2011], it is inferred that triggered tremor is likely a response to slip driven by the stress associated with seismic waves. A benefit is that the duration of wave passage is well defined, which simplifies the detection of triggered tremor. Otherwise, dynamic triggering phenomena reflect the principle that stress level is critical to failure. In light of the above discussion, in the present study a systematic survey was conducted throughout the Japanese islands of remotely triggered, tremor-like seismic events that resulted from significant large earthquakes. The analysis revealed that tremor triggered by the 2004 Mw9.1 Sumatra–Andaman earthquake were not limited to southwest Japan; similar tremor phenomena were discovered in inland Hokkaido, in northernmost Japan, where tectonic tremor associated with subducting plate has not previously been detected.
 The present analysis used seismogram data recorded by velocity seismometers of the High Sensitivity Seismograph Network (Hi-net) administered by the National Research Institute for Earth Science and Disaster Prevention (NIED) of Japan [Obara et al., 2005]. In order to detect remotely triggered tremor, a search was made for periodic tremor wave trains coincident with the long-period surface wave phase. The seismogram envelopes of bandpass-filtered traces with a passband frequency of 4–16 Hz were compared with the long-period displacement waveform after deconvolving the instrument response of the STS-2 from Hi-net velocity seismometers [Maeda et al., 2011].
 Based on manual inspections of envelope data from 800 stations for 25 large earthquakes with magnitudes of 7–9 that occurred after 2001, it was found that 10 teleseismic events triggered non-volcanic tremor in southwest Japan, as previously studied byMiyazawa and Mori . I also detected triggered tremor excited by the 2004 Sumatra–Andaman earthquake, in northernmost and central Hokkaido (Figure 1), which has not been documented previously. In the central region, detected tremor also occurred during passing of surface waves of the M8.6 Sumatra–Nias earthquake on 28 March 2005.
Figures 2a and 2bshow seismograms of triggered tremor from northernmost Hokkaido. The higher-frequency seismogram components are characterized by a periodic amplitude enhancement at around 20 s intervals with a total duration of approximately 200 s when the surface wave amplitude is greatest. The dispersion on the original displacement seismograms makes it difficult to discern any correlation with the tremor envelope pattern; however, the pattern is similar to the bandpass-filtered radial and vertical displacement components with a narrow passband of 16–20 s (Figure 2a). The amplitudes and periodicity of the triggered tremor wave train correlate well with the amplitude variations of the traces Rayleigh wave. However, the triggered tremor amplitude pattern in the central region is not coincident with that of the surface wave for any component, although the periodicity is almost the same (Figure 2c). Each wave train exhibits a spindle-shaped amplitude pattern where the onset of P or S is unclear (Figure 2d). The peak amplitude arrival time differs between neighboring stations, reflecting the distance traveled from the tremor source. The dominant frequency components of the tremor wave trains are 2–15 Hz and 2–5 Hz for the northernmost and central tremor activities, respectively (Figure 2d). The waveform property of the latter tremor activity is somewhat similar to that of non-volcanic tremor in southwest Japan and of other inland low-frequency microearthquakes [Obara and Hirose, 2006].
3.1. Location of Tremor
 The tremor source was located using the envelope shape calculated from the filtered seismogram data obtained with the previously described dominant frequency range and by applying the envelope correlation method [Obara, 2002]. The time lag between station pairs was assumed to be the S wave travel time difference. Figure 3shows triggered tremor hypocentral distributions with depth errors within 1 and 2 km for the central and northern regions, respectively. In the central region, well-defined tremor were concentrated at depths of 10–20 km by using 9 stations of NIED Hi-net, the Japan Meteorological Agency (JMA) and Hokkaido University. In the northernmost region, the sources of well-defined tremor are distributed from the near-surface to a depth of 20 km, as located by using 6 stations of Hi-net. Because the measurement error of the time lag between station pairs is approximately 2 s, the estimated depth range reflects the uncertainty in the source depths; however, a comparison of the locations of tremor activity in the two regions arises the possibility that tremor was located at a relatively shallow depth in the northern region. The difference in the dominant frequencies of tremor wave trains in the two regions (Figure 2d) may indicate a difference in depth and causative mechanism.
3.2. Tectonic Environment of Tremors
 In Hokkaido, there are many volcanoes and clusters of low-frequency earthquakes whose hypocenters are located at depths of 10–40 km along an east–west trend (Figure 1). In central Hokkaido, there exist at least four clusters of low-frequency earthquakes. The central tremor activity coincides with one of the volcanic-related, deep low-frequency earthquake clusters both in epicenter location and in depth (Figure 3). Therefore, the tremor activity is inferred to reflect the continuous occurrence of small, volcanic-related low-frequency earthquakes similar to non-volcanic tremor in the Nankai subduction zone [Shelly et al., 2007]. However, in northernmost Hokkaido there are no volcanoes or known active fault systems. Seismicity rates are very low in the tremor source area, but relatively high in the western part of this region (Figure 3). One particularly relevant geological aspect is the existence of a limestone cave system, called the Nakatonbetsu caves [Tajika, 1989], located close to the estimated tremor epicenter.
3.3. Strain Field and Tremor Occurrence
 It is usual to calculate the change in Coulomb and other stress components with time; however information on the target fault type and slip parameters is unavailable for the detected tremor. Therefore, the relationship between the strain field generated by large-amplitude surface waves and tremor occurrence was investigated. First, the displacements of horizontal components at each grid point (spacing of 6 minutes) were calculated at 1-second intervals from deconvolved data at all Hi-net stations in Hokkaido. Then, the strain components were obtained from the difference in the displacement field for neighboring time slices at each grid cell.Figure 4 compares snapshots of the areal strain field at the surface with tremor activity in the northern region. The bottom traces are time series of the areal strain estimated at the epicenter of northern tremor activity. Envelope seismograms observed at the HTBH station are illustrated with time corresponding to the source process with correction of travel distance from the station to the tremor source. Clearly, the seismic energy of tremor is inversely correlated with the areal strain at the tremor source, and this indicates that tremor source amplitude correlates with the contraction of areal strain.
 In geothermal areas, dynamically triggered microearthquakes are interpreted to result from geothermal and volcanic activities, such as fluid flow and bubble excitation [Hill et al., 1993]. In contrast, West et al. proposed that periodic earthquake occurrence triggered by each Rayleigh wave phase arriving from the 2004 Sumatra–Andaman earthquake are the result of extensional stress changes in the originally extensional field, with fluid effects being secondary. Similarly, non-volcanic tremor in southwest Japan is triggered by each peak in the amplitudes of dilatational and extensional normal stress due to the long-period Rayleigh wave. This is interpreted as a reduction in normal stress with a higher friction coefficient, with fluid being driven to the plate interface [Miyazawa et al., 2008]. Wu et al.  explained triggered microearthquakes in China by a simple Coulomb failure criterion applied to the transition zone at shallow depth on a fault plane.
 The mechanisms proposed in the above studies (e.g., fault slip, fluid effects) may also explain the triggered tremor detected in the present study. In northernmost Hokkaido, the detected tremor might reflect the occurrence of slip on an unknown fault plane, in the case that the slip parameters are matched to appropriate strain changes. If we assume that the northernmost tremor occurred at a shallow depth, one possible mechanism might be related to the occurrence of a nearby limestone cave. Because the tremor is modulated by the compressional strain estimated from the strain field analysis, if we speculate that the cave is filled with water, then the fluid pressure would increase as a consequence of compressional strain. The peak fluid pressure might cause hydraulic fracturing along cracks in the rocks surrounding the cave, and the radiating energy may correlate to the variations in areal strain.
 The central tremor activity might be related to volcanic activity, given its spatial similarity to the deep low-frequency earthquakes near volcanoes [Katsumata and Kamaya, 2003]. Hasegawa et al. suggested that deep low-frequency earthquakes occur adjacent to a low-velocity zone, which is a magmatic body. Therefore, the tremor might be affected by strain perturbation from this anomalous structure. Inconsistencies in amplitude patterns between the input surface wave and the resultant tremor envelope suggest that the triggering mechanism is related to shear strain components rather than isotropic components. Indeed, the strain component along the ray path produced by Rayleigh waves should be small at a depth of 10–20 km. Therefore, the tremor might be affected by shear strain along an unknown fault.
 I thank the staff of NIED's Hi-net for providing quality waveform data. For the hypocentral determination of triggered tremor, data were used from NIED Hi-net, JMA, and Hokkaido University. All waveform data were obtained from the NIED Hi-net data server. I used the earthquake catalogs constructed by JMA, which mainly use Hi-net data. The figures in this paper were created using the Generic Mapping Tools [Wessel and Smith, 1998], SAC, and R.
 The Editor thanks Zhigang Peng and an anonymous reviewer for their assistance in evaluating this paper.