A review on slow earthquakes in the Japan Trench

Abstract

Slow earthquakes are episodic slow fault slips. They form a fundamental component of interplate deformation processes, along with fast, regular earthquakes. Recent seismological and geodetic observations have revealed detailed slow earthquake activity along the Japan Trench—the subduction zone where the March 11, 2011, moment magnitude (M_w) 9.0 Tohoku-Oki earthquake occurred. In this paper, we review observational, experimental, and simulation studies on slow earthquakes along the Japan Trench and their research history. By compiling the observations of slow earthquakes (e.g., tectonic tremors, very-low-frequency earthquakes, and slow slip events) and related fault slip phenomena (e.g., small repeating earthquakes, earthquake swarms, and foreshocks of large interplate earthquakes), we present an integrated slow earthquake distribution along the Japan Trench. Slow and megathrust earthquakes are spatially complementary in distribution, and slow earthquakes sometimes trigger fast earthquakes in their vicinities. An approximately 200-km-long along-strike gap of seismic slow earthquakes (i.e., tectonic tremors and very-low-frequency earthquakes) corresponds with the huge interplate locked zone of the central Japan Trench. The M_w 9.0 Tohoku-Oki earthquake ruptured this locked zone, but the rupture terminated without propagating deep into the slow-earthquake-genic regions in the northern and southern Japan Trench. Slow earthquakes are involved in both the rupture initiation and termination processes of megathrust earthquakes in the Japan Trench. We then compared the integrated slow earthquake distribution with the crustal structure of the Japan Trench (e.g., interplate sedimentary units, subducting seamounts, petit-spot volcanoes, horst and graben structures, residual gravity, seismic velocity structure, and plate boundary reflection intensity) and described the geological environment of the slow-earthquake-genic regions (e.g., water sources, pressure–temperature conditions, and metamorphism). The integrated slow earthquake distribution enabled us to comprehensively discuss the role of slow earthquakes in the occurrence process of the Tohoku-Oki earthquake. The correspondences of the slow earthquake distribution with the crustal structure and geological environment provide insights into the slow-earthquake-genesis in the Japan Trench and imply that highly overpressured fluids are key to understanding the complex slow earthquake distribution. Furthermore, we propose that detailed monitoring of slow earthquake activity can improve the forecasts of interplate seismicity along the Japan Trench.

Introduction

The Japan Trench is a subduction zone located off the Pacific coast of eastern Japan (Fig. 1). In this trench, the Pacific Plate, whose age is approximately 130–140 Ma (Müller et al. 2008), is subducting beneath the Okhotsk Plate at a convergence rate of approximately 9 cm/yr (Bird 2003) (Fig. 1). Many large interplate earthquakes have been observed and actively studied in this subduction zone. In the central (37–39° N) and northern (39–41° N) Japan Trench, moment magnitude (M_w) 7–8 megathrust earthquakes have repeatedly occurred at a depth of 20–50 km (Nagai et al. 2001; Murotani et al. 2003; Yamanaka and Kikuchi 2004; Fig. 1). On March 11, 2011, the M_w 9.0 Tohoku-Oki earthquake ruptured the central and southern parts of the Japan Trench, and it was the largest earthquake ever recorded in Japan (the thick red contours in the Japan Trench in Fig. 1). In the central Japan Trench (37–39° N), the Tohoku-Oki rupture reached the trench axis, and the Japan Trench megathrust slipped more than 50 m (e.g., Lay 2018; Kodaira et al. 2020, 2021), as indicated by the seismic waveforms (Ide et al. 2011), crustal deformation (Iinuma et al. 2012), and bathymetric changes (Fujiwara et al. 2011). In addition to megathrust earthquakes, small interplate earthquakes have been vigorously studied in the Japan Trench. Small repeating earthquakes, in which the same area on the plate interface repeatedly undergoes similar ruptures (e.g., Nadeau and Johnson 1998), have been observed in the Japan Trench (Matsuzawa et al. 2002; Igarashi et al. 2003; Uchida and Matsuzawa 2013; Fig. 1). The regularity and diversity of their rupture patterns have been investigated in detail (e.g., Okada et al. 2003; Uchida et al. 2007; Okuda and Ide 2018; Ide 2019a; Chang and Ide 2021).

Fig. 1

Distribution of slow and fast earthquakes in the Japan Trench and Nankai Trough. Green and blue symbols semi-schematically show the distribution of slow earthquakes (tectonic tremors, short-term slow slip events, and long-term slow slip events). Red symbols indicate the distribution of fast, regular earthquakes. For an explanation of the other symbols, refer to the legend in the figure. The distribution of slow and fast earthquakes in the Nankai Trough is adapted from Obara and Kato (2016) and Obara (2020). Depth contours of the upper surface of the Pacific and Philippine Sea plates are based on Baba et al. (2002), Nakajima and Hasegawa (2006), Nakajima and Hasegawa (2007), Hirose et al. (2008), Nakajima et al. (2009), and Kita et al. (2010). The slip distributions of the March 11, 2011, Tohoku-Oki earthquake (10 m or larger) and other megathrust earthquakes are from Yagi et al. (1998), Yagi et al. (1999), Nagai et al. (2001), Tanioka and Seno (2001), Murotani et al. (2003), Yamanaka and Kikuchi (2004), Earthquake Research Center (2005), Mochizuki et al. (2008), Research Center for Seismology, Volcanology, and Disaster Mitigation (2008), Iinuma et al. (2012), Ohta et al. (2012), Kubo et al. (2013), and Kubo and Nishikawa (2020). The distribution of small repeating earthquakes is from Uchida and Matsuzawa (2013) and Igarashi (2020). The distribution of tectonic tremors is based on Idehara et al. (2014), Yamashita et al. (2015), Araki et al. (2017), Nishikawa et al. (2019), Yabe et al. (2019), and Yamashita et al. (2021). The distribution of short-term slow slip events is based on Ito et al. (2013), Nishimura et al. (2013), Nishimura (2014), and Nishimura (2021). The source area of the 1992 Sanriku-Oki ultraslow earthquake is from Kawasaki et al. (2001). The locations of subducting ridges and seamounts are based on Kodaira et al. (2000), Park et al. (2004), Mochizuki et al. (2008), and Yamamoto et al. (2013). The northeastern limit of the Philippine Sea Plate is from Uchida et al. (2009a). Seafloor topography is from Smith and Sandwell (1997). The plate motion is based on Bird (2003)

While much has been revealed about fast, regular earthquakes in the Japan Trench, activity of slow earthquakes, another major category of earthquake phenomenon, had remained poorly resolved until recently. Slow earthquakes are a general term for a wide variety of fault slip phenomena of much longer duration than fast, regular earthquakes of comparable seismic moment (Ide et al. 2007a). Slow earthquakes have been observed in many circum-Pacific subduction zones and are actively studied in subduction zones such as the Nankai Trough, a subduction zone located off the Pacific coast of southwest Japan (Fig. 1). In contrast to the Japan Trench, the complex distribution and occurrence patterns of various slow earthquakes were revealed along the Nankai Trough (e.g., Obara 2002; Ito et al. 2007; Obara and Kato 2016). In addition, it is noteworthy that in the Nankai Trough, the young and warm Philippine Sea Plate (approximately 15–25 Ma) is subducting (Müller et al. 2008), in contrast to the Japan Trench, where the old and cold Pacific Plate is subducting. This has been thought to cause striking differences in subduction zone processes (e.g., slow and fast earthquakes, arc volcanism, and metamorphism) between these two subduction zones (e.g., Peacock and Wang 1999; Katayama et al. 2012).

Slow earthquake research in the Japan Trench has accelerated since the 2011 Tohoku-Oki earthquake. A M_w 7.0 short-term slow slip event (SSE) accompanied by an earthquake swarm was observed approximately a month before the Tohoku-Oki earthquake on the shallow plate interface of the central Japan Trench (Kato et al. 2012; Ito et al. 2013; a green rectangle within the Tohoku-Oki earthquake rupture in Fig. 1). Furthermore, detection based on a visual inspection of broadband seismograms (Matsuzawa et al. 2015) and an analysis of repeating earthquakes (Uchida et al. 2016) also provided important insights into activity of slow earthquakes (very-low-frequency earthquakes and SSEs, respectively) in the Japan Trench, although systematic and comprehensive detection of slow earthquakes along the entire Japan Trench remained unachieved.

In 2019, pop-up-type ocean-bottom seismometers installed in the southern part of the Japan Trench (Ohta et al. 2019) and the Seafloor Observation Network for Earthquakes and Tsunamis along the Japan Trench (S-net) (NIED 2019a) finally led to revealing the detailed distribution of tectonic tremors, a type of slow earthquake, along the entire Japan Trench (Ohta et al. 2019; Tanaka et al. 2019; Nishikawa et al. 2019; Fig. 1). The tremor detection was followed by the systematic and comprehensive detection of other types of slow earthquakes (very-low-frequency earthquakes and short-term SSEs) using onshore seismic and geodetic observation networks (Baba et al. 2020; Nishimura 2021). These accumulating research findings are beginning to provide a full picture of the slow earthquake distribution along the Japan Trench (Fig. 1).

Based on the above research advances, we review the studies related to slow earthquakes along the Japan Trench and describe their research history from the first report of a large transient aseismic slip in the Japan Trench (Kawasaki et al. 1995) to the present. By compiling the observations of slow earthquakes (e.g., tectonic tremors, very-low-frequency earthquakes, and slow slip events) and related fault slip phenomena (e.g., small repeating earthquakes, earthquake swarms, and foreshocks of large interplate earthquakes), we present an integrated slow earthquake distribution along the Japan Trench. Based on the slow earthquake distribution, we discuss the occurrence processes of megathrust earthquakes along the Japan Trench, focusing on the role of slow earthquakes. Furthermore, we explore the conditions causing slow earthquakes by comparing the slow earthquake distribution with the crustal structure and geological environment of the Japan Trench.

In addition to the Japan Trench (34–41° N), we also mention slow earthquake activity in the southern end of the Kuril Trench (41–43° N). However, we limit our main discussion to the Japan Trench. Hereafter, we refer to the regions from 39 to 41° N (Iwate-Aomori-Oki), 37–39° N (Fukushima-Miyagi-Oki), 34–37° N (Boso-Ibaraki-Oki) as the northern, central, and southern Japan Trench, respectively. Because this review paper focuses on slow earthquakes along the Japan Trench, readers who are interested in the 2011 Tohoku-Oki earthquake and associated phenomena rather than slow earthquakes are referred to other review papers (Lay 2018; Kodaira et al. 2020, 2021; Uchida and Bürgmann 2021).

An overview of the subsequent sections is described below. Section 2 briefly summarizes basic knowledge about slow earthquakes for readers unfamiliar with the concept to facilitate a better understanding of the subsequent sections (Sects. 3, 4, 5, and 6); readers familiar with slow earthquakes can skip this section. Section 3 reviews observational, experimental, and simulation studies on slow earthquakes along the Japan Trench. Furthermore, we synthesized the results of observational studies of slow earthquakes and summarized what has been revealed so far about the spatiotemporal distribution of slow earthquakes before and after the 2011 Tohoku-Oki earthquake. Section 4 compares the spatial distribution of slow earthquakes along the Japan Trench with that of other fault slip phenomena. Specifically, we compared the slow earthquake distribution with the distributions of megathrust earthquake ruptures, including the 2011 Tohoku-Oki rupture, interplate coupling, small interplate earthquakes, tsunami earthquakes, and seismicity parameters (the Gutenberg–Richter relationship’s b value, tidal response of seismicity, and p value of the Omori–Utsu aftershock law). Section 5 compares the slow earthquake distribution with the crustal structure of the Japan Trench (e.g., interplate sedimentary units, subducting seamounts, petit-spot volcanoes, horst and graben structures, residual gravity, seismic velocity structure, and plate boundary reflection intensity) and describes the geological environments of the slow-earthquake-genic regions (e.g., fluid sources, pressure–temperature conditions, and metamorphism). Section 6 provides the comprehensive discussion. Specifically, we discuss the occurrence process of the Tohoku-Oki earthquake, focusing on the roles that slow earthquakes played. We also discuss the structural and geological origins of the slow earthquake distribution in the Japan Trench. Finally, we discuss how slow earthquake observations could be utilized to improve forecasts of interplate seismicity along the Japan Trench. Section 7 presents the conclusions.

Brief introduction to slow earthquakes

This section provides a brief introduction to slow earthquakes at subduction zone plate boundaries, focusing on topics essential to the subsequent sections (Sects. 3, 4, 5, and 6). Several remarkable review papers on slow earthquakes have been published (e.g., Peng and Gomberg 2010; Beroza and Ide 2011; Ide 2014; Saffer and Wallace 2015; Obara and Kato 2016; Bürgmann 2018; Obara 2020; Behr and Bürgmann 2021; Kirkpatrick et al. 2021). We refer readers to these papers for more information. Obara (2020) extensively reviewed the detailed occurrence patterns of various slow earthquakes, which are not included in this section. Obara and Kato (2016) provided a detailed discussion on the relationship between slow earthquakes and megathrust earthquakes. Geological environment and possible physical mechanisms of slow earthquakes have been described in detail by Bürgmann (2018), Behr and Bürgmann (2021), and Kirkpatrick et al. (2021). For details on slow earthquakes on shallow subduction plate interfaces, readers can refer to Saffer and Wallace (2015).

Types of slow earthquakes

There are five major types of slow earthquakes: low-frequency earthquakes (LFEs), tectonic tremors, very-low-frequency earthquakes (VLFEs), short-term SSEs, and long-term SSEs (Fig. 2).

Fig. 2

Observed signals of the slow earthquake family. The slow earthquake family includes a low-frequency earthquakes (LFEs) (Shelly et al. 2007), b tectonic tremors (Ide et al. 2008), c very-low-frequency earthquakes (VLFEs) (Ide et al. 2008), d short-term slow slip events (SSEs) (Hirose and Obara 2010), and e long-term SSEs (Takagi et al. 2016). In d the daily count of concurrent tectonic tremors is presented (Hirose and Obara 2010). The dashed lines in d indicate calculated tilt changes due to the short-term SSE (Hirose and Obara 2010). All the above signals of the slow earthquake family were observed on the deeper plate interface (25–40 km depth) of the Nankai Trough. GNSS, global navigation satellite system

Three types of slow earthquakes can be observed by seismometers (Fig. 2a–c), hereafter referred to as “seismic slow earthquakes.” The first type of seismic slow earthquake is the LFE (Fig. 2a). LFEs were first discovered by Nishide et al. (2000) at a deeper part (approximately 30 km depth) of the Nankai Trough (see also Katsumata and Kamaya 2003). LFEs are slow earthquakes with the smallest moment magnitudes (approximately M_w 1.5) and characterized by the shortest characteristic time scale (source duration of ~ 0.3 s) (Ide et al. 2007a, 2007b). They are also characterized by low dominant frequencies (1–10 Hz) compared to fast microearthquakes of comparable seismic moment (e.g., Shelly et al. 2007; Obara 2020).

The second type of seismic slow earthquake is the tectonic tremor (Fig. 2b), which is a seismic event that generates a weak seismic signal lasting more than several tens of seconds without clear P- and S-wave arrivals. The dominant frequency is 1–10 Hz (e.g., Shelly et al. 2007; Obara 2020). Similar to the LFEs, tectonic tremors were first discovered at the deeper part of the Nankai Trough in 2002 (Obara 2002). These tremors were aligned along the 30–40 km deep contours of the plate interface and located downdip of the megathrust seismogenic zones (10–20 km depth) of the Nankai Trough (Obara 2002; Obara and Kato 2016; Fig. 1). Previous reports revealed LFEs to be fault slip phenomena with a low-angle reverse fault mechanism on the plate interface (Shelly et al. 2006; Ide et al. 2007b). Furthermore, Shelly et al. (2007) demonstrated that tectonic tremor is a swarm of LFEs, by detecting many LFE signals embedded in tremor waveforms using a matched filter technique.

The third type of seismic slow earthquake is the VLFE (Fig. 2c), which is a fault slip phenomenon (M_w 3–4) comprised predominantly of seismic waves with periods of tens to hundreds of seconds, with shorter-period waves lacking (e.g., Ito et al. 2007; Obara 2020). VLFEs were first discovered near the trench axis of the Nankai Trough (Ishihara 2003; Obara and Ito 2005). Subsequently, they were found to occur simultaneously with tectonic tremors on the deeper plate interface of the Nankai Trough (approximately 30–40 km depth) (Ito et al. 2007).

There are two types of slow earthquakes observed by geodetic instruments (Fig. 2d and e), hereafter referred to as “geodetic slow earthquakes.” The first type of geodetic slow earthquake is the short-term SSE (Fig. 2d), which is an interplate fault slip event that usually lasts from a few days to a month. Short-term SSEs on subduction plate interfaces were discovered in the late 1990s and early 2000s (Sagiya 1997; Hirose et al. 2000; Dragert et al. 2001). Global navigation satellite system (GNSS) observations (Dragert et al. 2001) in the Cascadia subduction zone, a subduction zone located off the west coast of North America, captured a slow transient slip of ~ 2 cm that occurred on the deeper plate interface (30–40 km depth) in the summer of 1999. Sagiya (1997) and Hirose et al. (2000) also reported similar aseismic transients off the Boso Peninsula and beneath Tokyo Bay, respectively, in the Sagami Trough, eastern Japan. The duration of the short-term SSE in the Cascadia subduction zone was approximately a month (Dragert et al. 2001), and its moment magnitude was M_w 6.7, which is far larger than the moment magnitude of the aforementioned seismic slow earthquakes (Fig. 2a–c). This SSE occurred downdip of the megathrust seismogenic zone of the Cascadia subduction zone, similar to the deep tectonic tremors in the Nankai Trough (e.g., Obara 2002; Shelly et al. 2006; Obara and Kato 2016). In 2004, short-term SSEs were also discovered on the deeper plate interface (approximately 30–40 km depth) of the Nankai Trough by tiltmeter observations (Obara et al. 2004; Figs. 1 and 2d).

The second type of geodetic slow earthquake is the long-term SSE (Fig. 2e), which is an interplate fault slip event that lasts from several months to years. It was first discovered in 1999 on the deeper plate interface of the Nankai Trough by GNSS observations (Hirose et al. 1999). This event lasted approximately 300 days from March to December 1997 and caused a fault slip of ~ 18 cm, with an estimated magnitude of M_w 6.6. Long-term SSEs on the deeper plate interface of the Nankai Trough are located slightly shallower (approximately 25 km depth) than short-term SSEs (Obara et al. 2004; Takagi et al. 2016; Obara and Kato 2016; Fig. 1). Long-term SSEs are characterized by the largest moment magnitudes (approximately M_w 6.5–7.5) and the longest characteristic time scales (several months to years) in the slow earthquake family.

Note, however, that the distinction between short-term and long-term SSEs is ambiguous. In fact, in the Ryukyu Trench, the southwestern extension of the Nankai Trough, SSEs of medium duration (i.e., a few months) have been observed (Kano et al. 2018b). In the Nankai Trough, different characteristic time scales are often assumed a priori when systematically detecting short-term and long-term SSEs (several days and approximately one year, respectively) (Takagi et al. 2019; Okada et al. 2022). Therefore, the detailed activity of SSEs with time scales intermediate between these two, such as a few months, is less clear. Furthermore, in the Mexico subduction zone, Frank et al. (2018) suggested that a long-term SSE can be decomposed into a cluster of short-term SSEs. However, it is unclear whether such a decomposition is possible for the long-term SSEs in the Nankai Trough.

Although these five major types of slow earthquakes were first discovered on the deeper plate interfaces of the Nankai Trough and Cascadia subduction zone, they have now been observed in many circum-Pacific subduction zones (e.g., Beroza and Ide 2011; Obara and Kato 2016; Obara 2020). Slow earthquakes may universally occur in subduction zones, regardless of the tectonic characteristics of subduction zones, such as subducting plate age. Furthermore, slow earthquakes can occur under various temperature–pressure conditions (e.g., Peacock 2009; Saffer and Wallace 2015; Behr and Bürgmann 2021; Kirkpatrick et al. 2021). Slow earthquakes are predominantly observed on the plate interface downdip and/or updip of megathrust seismogenic zones (Fig. 1), although the pressure–temperature conditions of the deeper and shallower parts of the plate interface differ considerably. In the Nankai Trough (Fig. 1), for example, slow earthquakes have been observed on both the shallow plate interface less than 10 km deep (100 °C or less to approximately 200 °C) and the deeper plate interface at a depth of 25–40 km (approximately 350–450 °C) (Peacock and Wang 1999; Peacock 2009). They avoid the megathrust seismogenic zones at a depth of 10–25 km (approximately 200–350 °C) (Fig. 1).

Relationship between slow earthquakes

Simultaneous occurrence of different types of slow and fast earthquakes

Different types of slow earthquakes often occur simultaneously in close proximity. The coincidence of tectonic tremor bursts and short-term SSEs, termed episodic tremor and slip (ETS), was first discovered on the deeper plate interface of the Cascadia subduction zone (Rogers and Dragert 2003). In 2004, ETSs were also discovered in the Nankai Trough (Obara et al. 2004; Fig. 2d). Subsequently, VLFEs were also found to occur during ETS (Ito et al. 2007; Ghosh et al. 2015). Furthermore, long-term SSEs on the deeper plate interface (approximately 25 km depth) of the Nankai Trough are known to activate ETSs on their downdip side (Hirose and Obara 2005). Because of the remarkable correspondence between the short-term SSEs and tectonic tremors at the deeper parts of the Cascadia subduction zone and Nankai Trough, several attempts have been made to estimate the source properties (e.g., source duration, rupture area, and moment) of short-term SSEs too small to be geodetically detected (M_w 5.5 or less) from characteristics of tremor bursts (e.g., Aguiar et al. 2009; Obara 2010; Wech et al. 2010; Gomberg et al. 2016).

The occurrence of ETSs is not limited to the deeper subduction plate interface. On the shallow plate interface (shallower than 10 km depth) of the Nankai Trough, ocean-bottom seismometers and subseafloor borehole pore-fluid pressure measurements observed recurring ETSs (Araki et al. 2017), although shallow ETSs have not been observed in the Cascadia subduction zone (McGuire et al. 2018).

In addition to the Cascadia subduction zone and Nankai Trough, clear ETSs have been observed in the subduction zones of Costa Rica (Walter et al. 2013), Mexico (Rousset et al. 2017), and Alaska (Rousset et al. 2019). However, note that short-term SSEs and tremor bursts do not always occur at exactly the same time. In the Hikurangi Trench, the subduction zone located off the east coast of the North Island of New Zealand, a time lag of 10 days or longer between the occurrence of a short-term SSE and a tremor burst has been observed (Todd et al. 2018; Shaddox and Schwartz 2019; Nishikawa et al. 2021). Furthermore, some short-term SSEs in the Hikurangi Trench lacked detectable tectonic tremors (Todd and Schwartz 2016). However, we cannot rule out the possibility that tectonic tremors of magnitudes below the observational limit occurred. This also holds true for subduction zones where clear ETS has not been reported.

Slow and fast earthquakes can also occur simultaneously. The coincidence of short-term SSEs and swarms of interplate earthquakes (M_w $\le$ 5) has been repeatedly observed off the Boso Peninsula in the Sagami Trough, eastern Japan (Ozawa et al. 2003; Sagiya 2004). These earthquake swarms included repeating earthquakes (Kato et al. 2014; Gardonio et al. 2018). The SSE-induced stress loading is considered to be the triggering mechanism of these earthquake swarms (Fukuda 2018). Similar coincidences of SSEs and fast earthquakes have been observed in the Japan Trench (Ito et al. 2013) and the subduction zones of Mexico (Liu et al. 2007), Hikurangi (Delahaye et al. 2009), Ecuador (Vallée et al. 2013), Peru (Villegas-Lanza et al. 2016), and northern Chile (Socquet et al. 2017). Fast earthquakes that occur concurrently with SSEs are usually interplate earthquakes; however, in the Ecuador and Hikurangi trenches, swarms of intraslab or upper-plate earthquakes accompany short-term SSEs on the plate interface (Collot et al. 2017; Shaddox and Schwartz 2019; Nishikawa et al. 2021). In the Japan, northern Chile, and Mexico trenches, short-term or long-term SSEs were followed by megathrust earthquakes: the 2011 M_w 9.0 Tohoku-Oki (Kato et al. 2012; Ito et al. 2013), 2014 M_w 8.1 Iquique (Ruiz et al. 2014; Socquet et al. 2017), and 2014 M_w 7.3 Papanoa (Radiguet et al. 2016) earthquakes, respectively. The preseismic SSEs of the Tohoku-Oki and Iquique earthquakes were accompanied by swarms of foreshocks. These preseismic SSEs are considered to have triggered the megathrust earthquakes or been involved in their nucleation processes.

SSEs are not the only slow earthquakes that can occur simultaneously with fast earthquakes. In the shallow part of the southern Japan Trench, coincident tectonic tremors and swarms of interplate microearthquakes have been observed (Obana et al. 2021). In the central Hikurangi Trench, tectonic tremors, short-term SSEs, and swarms of intraslab earthquakes occur simultaneously (Romanet and Ide 2019; Nishikawa et al. 2021). Furthermore, a recent study (Yamamoto et al. 2022) reported the simultaneous occurrence of an SSE, VLFEs, and a swarm of interplate microearthquakes on the shallow plate interface of the Nankai Trough off the southeast coast of the Kii Peninsula (the Tonankai region in Fig. 1).

Unified understanding of slow earthquakes

Although slow earthquakes are diverse, efforts have been made to understand them in a unified manner. As described in Sect. 2.1, tectonic tremors are swarms of LFEs (Shelly et al. 2007). Furthermore, Ide et al. (2007a) found that the seismic moment M₀ of slow earthquakes (LFEs, VLFEs, short-term SSEs, and long-term SSEs) is approximately proportional to their source duration T (i.e., M₀ $\propto$ T) (blue symbols in Fig. 3a). In contrast, the seismic moment of fast earthquakes is proportional to the cubed source duration (i.e., M₀ $\propto$ T³) (the thick red line in Fig. 3a), and there is a significant gap between the scaling relations of slow and fast earthquakes. Aseismic transients that fall into this gap have been reported in the northern Japan Trench (Kawasaki et al. 1995) and Izu–Bonin subduction zone (Fukao et al. 2021), but are very rare (magenta circles in Fig. 3a).

Fig. 3

Slow earthquakes as broadband fault slip phenomena. a Moment–duration scaling relations of slow and fast earthquakes. Blue and magenta symbols represent slow earthquakes. LFE, VLFE, SVLFE, and S-SSE indicate LFEs (Ide et al. 2007b), deep VLFEs (Ito et al. 2007; Ide et al. 2008), shallow VLFEs (Ito and Obara 2006), and short-term SSEs (Hirose and Obara 2005) in the Nankai Trough, respectively. Other symbols are as follows: IA, a small short-term SSE near the Kii Peninsula, southwest Japan (Itaba and Ando 2011); BO, a short-term SSE near the Boso Peninsula, eastern Japan (Sagiya 2004); NZ, short- and long-term SSEs in Hikurangi, New Zealand (Wallace et al. 2012); CA, short-term SSEs in Cascadia (Dragert et al. 2004); BU, a long-term SSE in the Bungo Channel, southwest Japan (Hirose and Obara 2005); MX, a long-term SSE in Guerrero, Mexico (Kostoglodov et al. 2003; Yoshioka et al. 2004); TK, a long-term SSE in the Tokai region, central Japan (Miyazaki et al. 2006); AL, a long-term SSE in Alaska (Ohta et al. 2006); IB, aseismic transients in the Izu–Bonin subduction zone (Fukao et al. 2021); and SR, the 1992 Sanriku-Oki ultraslow earthquake in the northern Japan Trench (Kawasaki et al. 1995). The thick red line represents the scaling relationship of fast earthquakes. The red ellipse indicates tsunami earthquakes (Kanamori 1972; Ide et al. 1992). The gray curve represents the approximate observational limit of seismological instruments suggested by the US Geological Survey low-noise model (Peterson 1993). This panel is adapted from Ide (2014). b Average maximum amplitude of stacked LFE waveforms for each frequency band. The maximum amplitudes of seismic signals in each frequency band are averaged for each component of each station. Blue inverted triangles represent stacked LFE waveforms in the Nankai Trough (Masuda et al. 2020). Green symbols indicate stacked LFE waveforms in the Cascadia subduction zone (Bostock et al. 2015; Ide 2019b). For comparison, the average maximum amplitude of synthetic waveforms that obey the omega-square model (Aki 1967) with a 2 Hz corner frequency is shown. This panel is adapted from Masuda et al. (2020)

Based on the moment–duration scaling of slow earthquakes (M₀ $\propto$ T), Ide et al. (2007a) interpreted diverse slow earthquakes as the same physical phenomena with different seismic moments. They considered diffusion as a candidate for the process behind slow earthquakes, although the specific physical mechanism (e.g., slip phenomena associated with fluid diffusion, stress diffusion, or others) remained unclear. The diffusional aspect of slow earthquakes is evidenced by the parabolic migration patterns of deep tectonic tremors and LFEs in the Nankai Trough on a scale of several tens of kilometers (e.g., Ide 2010; Kato and Nakagawa 2020), although deep tectonic tremors in the Nankai Trough and Cascadia subduction zone show almost constant-velocity (approximately 10 km/day) migration patterns on a larger spatial scale (100 km or larger) (e.g., Ito et al. 2007; Houston et al. 2011; Nakata et al. 2011). Here, parabolic migration means migration with the distance traveled L proportional to the square root of elapsed time t (i.e., L $\propto$ $\sqrt{t}$), which is typical of diffusional phenomena (Ide 2010).

In support of the unified understanding proposed by Ide et al. (2007a), Kaneko et al. (2018) found a continuous seismic signal from the LFE frequency band (1–10 Hz) through the microseismic frequency band (0.1–1 Hz) to the VLFE frequency band (0.01–0.1 Hz) in a seismogram recorded during a tremor burst episode at the shallow part of the Nankai Trough. Furthermore, Masuda et al. (2020) stacked many seismograms recorded in the Nankai Trough relative to the timing of high-frequency LFE signals and found that stacked LFE waveforms have a continuous seismic signal from the LFE frequency band (1–10 Hz) to the VLFE frequency band (0.01–0.1 Hz) (Fig. 2b). These results indicate that tectonic tremors (i.e., swarms of LFEs) and very-low-frequency earthquakes are the high- and low-frequency components, respectively, of a broadband seismic wave emanating from the identical slow fault slip phenomenon. Kaneko et al. (2018) and Masuda et al. (2020) called this slow fault slip phenomenon a broadband slow earthquake.

However, with respect to geodetic slow earthquakes, the unified understanding proposed by Ide et al. (2007a) still remains under debate (e.g., Gomberg et al. 2016; Frank and Brodsky 2019; Michel et al. 2019). For example, Michel et al. (2019) reported that short-term SSEs in the Cascadia subduction zone are likely to have seismic moments proportional to their cubed durations (i.e., M₀ $\propto$ T³). They argue that the moment–duration scaling of SSEs is similar to the fast earthquake scaling (the thick red line in Fig. 3a). Their argument implies that seismic and geodetic slow earthquakes, or even short-term and long-term SSEs, might have different physical origins, and therefore requires different moment–duration scaling relations to describe different types of slow earthquakes. Of note, a single cubic scaling relation cannot explain observations of deep short-term and long-term SSEs in the Nankai Trough, while they are approximately consistent with a single linear scaling relation (Ide et al. 2007a; Okada et al. 2022).

Fast earthquakes closely related to slow earthquakes

In this section, we briefly introduce repeating earthquakes and earthquake swarms, which are fast earthquakes closely related to slow earthquakes. Repeating earthquakes are a phenomenon in which the same area on the plate interface repeatedly undergoes similar ruptures (e.g., Okada et al. 2003; Uchida et al. 2007; Kim et al. 2016), and are characterized by extremely high waveform similarity (e.g., Nadeau and Johnson 1998; Nadeau and McEvilly 1999; Igarashi et al. 2003) (Fig. 4a). The co-location, high waveform cross-correlations, or high waveform coherence of earthquakes are usually used to identify repeating earthquakes (Uchida 2019; Uchida and Bürgmann 2019, and references therein).

Fig. 4

Fast earthquakes potentially indicative of slow earthquake occurrence. a Observed waveforms of repeating earthquakes. These repeating earthquakes were observed off Kamaishi in the northern Japan Trench (Igarashi et al. 2003). b Magnitude–time diagram for an earthquake swarm accompanying an SSE. This earthquake swarm was observed during the period of the 2007 SSE off the Boso Peninsula in the Sagami Trough, eastern Japan (Fukuda 2018)

Repeating earthquakes are considered to occur due to stress loading by aseismic slip surrounding their source regions (Nadeau and Johnson 1998; Nadeau and McEvilly 1999; Matsuzawa et al. 2002; Igarashi et al. 2003; Chen and Lapusta 2009). The change in the aseismic slip rate of the surrounding region consequently fluctuates the recurrence interval of repeating earthquakes. Therefore, by examining the activity of repeating earthquakes, we can infer the spatiotemporal evolution of aseismic slip on the plate interface, such as steady slip (e.g., Nadeau and McEvilly 1999; Igarashi et al. 2003), afterslip (i.e., transient aseismic slip following a large earthquake) (e.g., Uchida and Matsuzawa 2013; Uchida et al. 2015), and SSEs (e.g., Nadeau and McEvilly 2004; Uchida et al. 2016). Once we know the occurrence time and seismic moment of a given repeating earthquake sequence and assume that a scaling relation between seismic moment and slip amount is applicable, we can infer the temporal evolution of aseismic slip surrounding the source region of the repeating earthquakes. Several scaling relationships between seismic moment and slip amount of a repeating earthquake have been proposed (e.g., Nadeau and Johnson 1998; Beeler et al. 2001; Khoshmanesh et al. 2015), as summarized and compared in Uchida (2019).

Repeating earthquakes can be used to infer the occurrence of SSEs on the subduction interface. Kato et al. (2012) detected small repeating earthquakes in the central Japan Trench prior to the March 11, 2011, Tohoku-Oki earthquake and calculated the amount of preseismic interplate fault slip. They inferred that an SSE occurred during the month prior to the Tohoku-Oki earthquake at the shallow part (approximately 10–20 km depth) of its source region. Their inference was subsequently supported by seafloor pressure gauge observations (Ito et al. 2013; a green rectangle within the Tohoku-Oki earthquake rupture in Fig. 1). There have been many similar attempts to infer the occurrence of SSEs using repeating earthquakes (e.g., Nadeau and McEvilly 2004; Bouchon et al. 2011; Kato et al. 2014; Uchida et al. 2016; Rolandone et al. 2018; Nishikawa and Ide 2018). Furthermore, several studies compared geodetically detected SSEs and repeating earthquake activity in detail and found a good correspondence between them (e.g., Shaddox and Schwartz 2019; Uchida et al. 2020; Hughes et al. 2021; Okada et al. 2022). Readers interested in repeating earthquakes can refer to review papers by Uchida (2019) and Uchida and Bürgmann (2019) for more information.

Earthquake swarms are seismic sequences without a distinguishable mainshock (Fig. 4b). They are distinct from regular mainshock–aftershock sequences and do not obey the Omori–Utsu’s aftershock law (e.g., Llenos et al. 2009; Holtkamp and Brudzinski 2011; Nishikawa and Ide 2017), which states that aftershock seismicity rates exhibit a power-law decay (e.g., Utsu 1957; Utsu et al. 1995). Earthquake swarms are thought to be triggered by transient aseismic phenomena, such as migration of crustal fluid, magma intrusion, and SSE-induced stress loading (e.g., Toda et al. 2002; Fukuda 2018; Ross et al. 2020). As already described in Sect. 2.2.1, earthquake swarms accompanying SSEs have been observed in many circum-Pacific subduction zones (e.g., Ozawa et al. 2003; Sagiya 2004; Liu et al. 2007; Delahaye et al. 2009; Ito et al. 2013; Vallée et al. 2013; Villegas-Lanza et al. 2016; Socquet et al. 2017; Yamamoto et al. 2022; Fig. 4b). Some of these earthquake swarms accompanying SSEs were followed by megathrust earthquakes (Ito et al. 2013; Socquet et al. 2017), and the earthquake swarms were considered as a swarm of foreshocks retrospectively.

Similar to repeating earthquakes, there have been many attempts to infer the occurrence of SSEs from earthquake swarm activity (e.g., Llenos and McGuire 2011; Okutani and Ide 2011; Marsan et al. 2013; Reverso et al. 2015; Nishikawa and Ide 2018; Rolandone et al. 2018; Nishikawa et al. 2019, 2021). They regarded earthquake swarms in subduction zones as potential indicators of SSEs. For example, Nishikawa et al. (2021) examined earthquake swarm activity and GNSS data along the Hikurangi Trench, New Zealand, and succeeded in detecting small SSEs (M_w $\le$ 6) that had previously been overlooked.

It is generally difficult to detect far offshore SSEs only by onshore geodetic observations. Therefore, repeating earthquakes and earthquake swarms in subduction zones can be useful tools to examine far offshore regions possibly hosting SSEs (Nishikawa et al. 2019).

Possible physical mechanisms of slow earthquakes

Although in-depth investigations on slow earthquakes, with respect to their distribution and occurrence patterns, have been conducted (Obara and Kato 2016; Obara 2020, and references cited therein), their physical mechanisms remain unclear. Many computational models of slow earthquakes, especially for SSEs, have been proposed. However, no conclusion has been reached as to which model is the most plausible because verification by observations is generally difficult. The proposed models are mostly based on the rate-and-state-dependent friction law (Dieterich 1979). This friction law was derived from rock friction experiments and expresses the shear stress on a frictional surface as a linear combination of the logarithm of the slip rate and the logarithm of a variable characterizing the state of the frictional surface (state variable). Readers can refer to a review paper by Scholz (1998) for the details of the rate-and-state-dependent friction law. We divide computational models of slow earthquakes into two major groups and briefly introduce them below.

The first group is characterized by a fault patch of homogeneous friction with a mechanism to suppress the slip rate. The mechanisms include a large critical nucleation length (i.e., the minimum slipping patch size required for initiating unstable seismic slip) (e.g., Tse and Rice 1986; Kato 2004; Liu and Rice 2005; Barbot 2019), a friction that exhibits rate-weakening behavior (a decrease in fault strength in response to an increased slip rate) at low slip rates but switches to rate-strengthening behavior (an increase in fault strength in response to an increased slip rate) at high slip rates (e.g., Shibazaki and Iio 2003; Shibazaki and Shimamoto 2007; Wu et al. 2019; Im et al. 2020), and dilatant strengthening (i.e., strengthening of frictional surfaces due to reduction in pore-fluid pressure associated with pore creation induced by fault slip) (e.g., Suzuki and Yamashita 2009; Segall et al. 2010). These slip rate braking mechanisms prevent the acceleration of fault slip to the slip rate of fast earthquakes (approximately 1 m/s) and produce SSEs. A large critical nucleation length (e.g., Tse and Rice 1986; Kato 2004; Liu and Rice 2005; Barbot 2019) can be achieved by a large characteristic slip distance (i.e., a parameter of the rate-and-state-dependent friction describing a slip distance necessary for adjustment of friction to a new steady slip rate), rate-weakening but almost neutral friction, or low effective normal stress (i.e., applied fault-normal stress minus pore-fluid pressure) (e.g., Scholz 1998).

The other group regards slow earthquakes as slow fault slips resulting from heterogeneous friction or rheology (e.g., Nakata et al. 2011; Ando et al. 2012; Skarbek et al. 2012). For example, Skarbek et al. (2012) showed that slow slip naturally occurs on a fault with a heterogeneous distribution of rate-weakening and rate-strengthening frictions without requiring finely tuned frictional parameters. Ando et al. (2012) considered small brittle patches within a viscous flowing matrix and reproduced the coincidence of slow shear deformation and fast shear ruptures similar to ETS (e.g., Rogers and Dragert 2003; Obara et al. 2004; Sect. 2.2.1). This model also reproduced diffusional migration of ETS (e.g., Ide 2010; Kato and Nakagawa 2020) as stress diffusion in the viscous matrix and associated small brittle ruptures. The above models, which consider the heterogeneity of fault materials, are consistent with geological observations of exhumed subduction shear zones that suggest a mixture of brittle fractures and viscous flow in regions hosting ETS (Behr et al. 2018; Ujiie et al. 2018). Furthermore, abundant fine-scale heterogeneity on the slow-earthquake-genic plate interfaces probably requires stochastic modeling rather than continuous deterministic models, as demonstrated by Brownian slow earthquake models (Ide 2008; Ide and Yabe 2019) and a model including stochastic stress fluctuations (Aso et al. 2019). The Brownian slow earthquake models are inspired by observations of tectonic tremors, whose waveforms look like random noise, and they express slow earthquakes as rupture processes that stochastically evolve.

A vast array of previous geophysical and geological observations suggests that areas prone to slow earthquakes are characterized by abundant water and near-lithostatic pore-fluid pressure. Evidence of near-lithostatic pore-fluid pressure includes high Vp/Vs ratios at plate boundaries hosting slow earthquakes (e.g., Kodaira et al. 2004; Shelly et al. 2006), dynamic triggering of tectonic tremors by the passage of surface waves of large distant earthquakes (e.g., Miyazawa and Mori 2005; Rubinstein et al. 2007), significant tidal responses of tectonic tremors (e.g., Nakata et al. 2008; Rubinstein et al. 2008), and abundant veins in an ancient subduction shear zone exhumed from a frictional–viscous transition depth (e.g., Ujiie et al. 2018). For a detailed summary of the evidence of near-lithostatic pore-fluid pressure, refer to Bürgmann (2018), Behr and Bürgmann (2021), and Kirkpatrick et al. (2021). Computational models of slow earthquakes often consider the influence of highly overpressured fluids and assume low effective normal stress on the fault plane. Low effective normal stress stabilizes fault slip and facilitates aseismic slip (e.g., Scholz 1998). However, the process by which the highly overpressured fluids enable the occurrence of various types of slow earthquakes (Sect. 2.1) remains unclear.

Studies on slow earthquakes along the Japan Trench

Brief history of slow earthquake research in the Japan Trench

In this section, we introduce important studies related to slow earthquakes along the Japan Trench in a chronological order, with an aim to help readers understand the history of slow earthquake research in the Japan Trench. The detailed spatiotemporal distribution of slip phenomena discovered by the studies introduced in this section is presented in Sect. 3.2.

Brief history prior to the 2011 Tohoku-Oki earthquake

A large transient aseismic slip was first observed in 1992 at the shallow plate interface (approximately 10–20 km depth) of the northern Japan Trench (39–40° N) by extensometers (Kawasaki et al. 1995; Kawasaki et al. 2001; the blue dashed rectangle in Fig. 1). This transient aseismic slip was also reported by Miura et al. (1993) and Miura et al. (1994), although these reports were not peer-reviewed journal papers. Kawasaki et al. (1995) named this aseismic slip event the 1992 Sanriku-Oki ultraslow earthquake. This event can be regarded as the first-ever reported SSE at a subduction zone plate boundary. However, the duration of this event was 1 day, and the moment magnitude was approximately M_w 7.3–7.7, making it a strange event—in that it deviates significantly from the linear moment–duration scaling relation of slow earthquakes (M₀ $\propto$ T) (Ide et al. 2007a; the magenta circle labeled SR in Fig. 3a). Furthermore, the event occurred immediately after a M_w 6.9 interplate earthquake (the 1992 M_w 6.9 Sanriku-Oki earthquake) and can be interpreted as a huge afterslip. However, this event cannot be considered a typical afterslip (Alwahedi and Hawthorne 2019) because the seismic moment of this event is extremely large, ranging from four to 16 times the seismic moment of the preceding M_w 6.9 earthquake. Similar to the 1992 Sanriku-Oki ultraslow earthquake in the northern Japan Trench, an aseismic slip event with characteristics intermediate between an SSE and an afterslip has been reported in the Peru Trench (Villegas-Lanza et al. 2016). This event also had an extremely large moment, and thus, it cannot be a typical afterslip (10 times or more the seismic moment of the preceding M_w 5.8 earthquake). Villegas-Lanza et al. (2016) interpreted this event as a slow slip “helped” (or triggered) by an earthquake rather than a classical afterslip. Following the classification of Villegas-Lanza et al. (2016), we also regard the 1992 Sanriku-Oki ultraslow earthquake in the northern Japan Trench as a slow slip triggered by an earthquake.

Subsequently, in the late 1990s and 2000s, afterslip following major earthquakes, rather than SSEs, became the focus of much research. A continuous GNSS observation system (GEONET) installed by the Geospatial Information Authority of Japan (GSI) recorded the afterslip following the 1994 M_w 7.7 Sanriku-Oki earthquake (Heki et al. 1997), which ruptured the plate interface (approximately 20–30 km depth) of the northern Japan Trench (40–40.5° N). The afterslip took about a year after the mainshock to release a seismic moment greater than the mainshock. Yagi et al. (2003) found that the coseismic slip of the 1994 M_w 7.7 Sanriku-Oki earthquake and its afterslip were spatially complementary in distribution. In addition, Miura et al. (2006) reported a large afterslip following the 2005 M_w 7.1 Miyagi-Ken-Oki earthquake, which ruptured the deeper part of the central Japan Trench (37.8–38.5° N). This afterslip released a seismic moment comparable to that of the mainshock over a period of approximately one year after the mainshock.

In the 2000s, the GEONET also revealed the interplate slip deficit distribution along the Japan Trench (Ito et al. 2000; Nishimura et al. 2000, 2004; Suwa et al. 2006; Hashimoto et al. 2009; Loveless and Meade 2010). These studies consistently suggested the existence of a huge interplate locked zone in the central Japan Trench (37–39° N). This locked zone was subsequently ruptured by the March 11, 2011, M_w 9.0 Tohoku-Oki earthquake (thick red contours in the Japan Trench in Fig. 1).

In 2002, small repeating earthquakes (e.g., Nadeau and Johnson 1998; Nadeau and McEvilly 1999; Sect. 2.3) were first reported along the Japan Trench (Matsuzawa et al. 2002; small red points in Fig. 1). They were used to estimate the amount of steady (e.g., Igarashi et al. 2003; Igarashi 2010) and transient aseismic slip (e.g., Uchida et al. 2004; Matsuzawa et al. 2004). Igarashi et al. (2003) detected repeating earthquakes by a method based on waveform similarity and estimated the rate of interplate steady slip along the Japan Trench from the recurrence intervals of the repeating earthquakes. Furthermore, they found that few continuous-type small repeating earthquakes (i.e., repeating earthquakes that repeat at almost regular intervals and are not clustered in time) were distributed within the huge locked zone of the central Japan Trench (37–39° N), discovered by the GNSS observations (Ito et al. 2000; Nishimura et al. 2000). This is probably due to the absence of aseismic loading inside the locked region (see Sect. 2.3). Uchida et al. (2004) analyzed repeating earthquake activity prior to interplate large earthquakes in the northern Japan Trench and investigated preseismic changes in the interplate slip rate. They suggested the acceleration of interplate aseismic slip 6 days, 2 days, and 8 months before the 1989 M_w 7.4, 1992 M_w 6.9, and 1994 M_w 7.7 Sanriku-Oki earthquakes, respectively. These preseismic slow slips might have concentrated stress at the source regions of the large earthquakes.

In 2004, Yamanaka and Kikuchi (2004) investigated the slip distribution of M_w 7–8 class megathrust earthquakes from 1931 to 1994 along the northern and central Japan Trench by means of waveform inversion (see thin red contours in the Japan Trench in Fig. 1). They termed large slip areas of megathrust earthquakes asperities. Specifically, they regarded the area where the amount of slip is half the maximum slip or larger as an asperity. They found that in the northern Japan Trench, the asperities of several megathrust earthquakes significantly overlap. That is, the same area on the plate interface has been repeatedly ruptured by the megathrust earthquakes. Furthermore, in the central Japan Trench (38–39° N), the asperity of the 1981 M_w 7.1 earthquake, which is located at approximately 20–30 km depth (thin red contours in the Tohoku-Oki earthquake rupture in Fig. 1), slipped again during the largest foreshock (M_w 7.3) 2 days before the March 11, 2011, M_w 9.0 Tohoku-Oki earthquake (Ohta et al. 2012). These observations indicate the persistent nature of asperities. Note that there are various definitions of asperity, as summarized in Ide (2014). To avoid ambiguity, we follow Yamanaka and Kikuchi’s definition in the subsequent sections.

Brief history after the 2011 Tohoku-Oki earthquake

The M_w 9.0 Tohoku-Oki earthquake occurred on March 11, 2011 (the thick red contours in the Japan Trench in Fig. 1). After the Tohoku-Oki earthquake, aseismic slips that preceded the 2011 Tohoku earthquake were investigated in detail. Seismicity analyses (Ando and Imanishi 2011; Kato et al. 2012) and ocean-bottom pressure gauge observations (Ohta et al. 2012) revealed that afterslip had occurred on the plate interface after the largest foreshock (M_w 7.3) 2 days before the Tohoku-Oki earthquake. Ando and Imanishi (2011) suggested that the afterslip following the largest foreshock had migrated toward the rupture initiation point of the Tohoku-Oki earthquake. Furthermore, a repeating earthquake analysis (Kato et al. 2012), ocean-bottom pressure gauge observations (Ito et al. 2013), and ocean-bottom seismometer observations (Ito et al. 2015; Katakami et al. 2018) suggested that a M_w 7.0 SSE accompanied by a Japan Meteorological Agency magnitude (M_j) 5 class earthquake swarm and tectonic tremors had been occurring on the shallow plate interface (approximately 10–20 km depth) of the central Japan Trench during the month prior to the Tohoku-Oki earthquake (a green rectangle within the Tohoku-Oki earthquake rupture in Fig. 1). The source region of the preseismic slow earthquakes was ruptured by the Tohoku-Oki earthquake and slipped tens of meters (e.g., Ide et al. 2011; Iinuma et al. 2012), indicating that the identical area of the plate interface hosted both aseismic and huge seismic slips. However, we note that the signals of the preseismic slow earthquakes were observed at a small number of ocean-bottom stations (Ito et al. 2013, 2015; Katakami et al. 2018), and there should be large uncertainty in their source locations.

In 2014, GNSS data from the GEONET revealed that aseismic slip had been accelerating on the deeper plate interface (approximately 30–60 km depth) of the central and southern parts (36–39° N) of the Japan Trench in the decade prior to the Tohoku-Oki earthquake (Mavrommatis et al. 2014). A subsequent analysis by Yokota and Koketsu (2015) also supported the acceleration of the aseismic slip from 2002 to the 2011 Tohoku-Oki earthquake and regarded this transient slip as a fault slip phenomenon similar to long-term SSEs on the deeper plate interface of the Nankai Trough (Hirose et al. 1999; Ozawa et al. 2002; Fig. 1). This aseismic transient may have stressed locked parts of the Japan Trench megathrust (Yokota and Koketsu 2015). Furthermore, the decade-long aseismic transient may also have induced slips on the shallow plate interface (10–30 km depth) in the central and southern Japan, partially overlapping with the rupture area of the Tohoku-Oki earthquake, as suggested by repeating earthquake data (Mavrommatis et al. 2015).

In the southern end of the Kuril Trench (41–43° N), which is located just north of the Japan Trench, VLFEs were first observed in 2008 (Asano et al. 2008). In 2015, the F-net broadband seismograph network, which is an onshore seismograph network operated by the National Research Institute for Earth Science and Disaster Resilience (NIED) (NIED 2019b), also observed VLFEs in the Japan Trench (Matsuzawa et al. 2015). VLFEs were detected using a method based on a matched filter technique (Shelly et al. 2007). However, this detection was not sufficient to provide a full picture of the VLFE distribution along the Japan Trench because their template VLFEs were detected by visual inspection and were not spatially complete (Matsuzawa et al. 2015).

In 2016, temporal changes in interplate slip rate estimated from repeating earthquakes (Sect. 2.3) suggested that 1–6 year periodic SSEs are widespread in the Japan Trench (Uchida et al. 2016). Seismicity of M_j 5 or greater was active during the high slip rate periods of the SSEs. Uchida et al. (2016) suggested that the periodic SSEs induce periodic stress perturbations and modulate the activity of medium-to-large earthquakes in the Japan Trench. Furthermore, Nomura et al. (2016) developed a Bayesian statistical method to estimate the detailed space–time distribution of interplate slip rates from the recurrence intervals of repeating earthquakes and confirmed the quasi-periodic slip rate acceleration in the northern Japan Trench.

In 2019, pop-up-type ocean-bottom seismometers (Ohta et al. 2019) and the Seafloor Observation Network for Earthquakes and Tsunamis along the Japan Trench (S-net) (Tanaka et al. 2019; Nishikawa et al. 2019) observed clear signals of tectonic tremors on the shallow plate interface (approximately 10–20 km depth) of the Japan Trench (small green squares in the Japan Trench in Fig. 1). The S-net is an ocean-bottom seismic and pressure observation network that NIED has been operating since 2016 (NIED 2019a). These studies revealed detailed tectonic tremor activity from 2016. The tremors were widely distributed and active on the shallow plate interface of the northern and southern Japan Trench, roughly consistent with the VLFE distribution presented by Matsuzawa et al. (2015). Furthermore, Nishikawa et al. (2019) discovered an approximately 200-km-long tremor gap in the central Japan Trench (37–39° N) (Fig. 1), which corresponds well with the large slip area (10 m or larger) of the Tohoku-Oki earthquake (Iinuma et al. 2012; the thick red contours in the Japan Trench in Fig. 1).

In 2020, Baba et al. (2020) conducted a comprehensive detection of VLFEs using data recorded by the F-net broadband seismograph network. They used synthetic waveforms of low-angle reverse faulting on the plate interface as templates of the matched filter technique (Shelly et al. 2007) and revealed the spatiotemporal distribution of VLFEs along the entire Japan Trench from 2003 to 2018. The distribution of VLFEs corresponded well with the distribution of tectonic tremors revealed by the S-net (Tanaka et al. 2019; Nishikawa et al. 2019), as expected from the fact that tectonic tremors and VLFEs are seismic waves emanating from the identical slow interplate slip phenomenon (Ide et al. 2007a; Kaneko et al. 2018; Masuda et al. 2020; Sect. 2.2.2). As with the tectonic tremors (Nishikawa et al. 2019), Baba et al. (2020) discovered an approximately 200-km-long VLFE gap in the central Japan Trench, which corresponded well with the large slip area of the Tohoku-Oki earthquake. Furthermore, this VLFE gap is a feature observed both before and after the 2011 Tohoku-Oki earthquake (Baba et al. 2020). This observation indicates that the Tohoku-Oki earthquake ruptured the gap of seismic slow earthquakes.

In 2021, Nishimura (2021) carried out a systematic and comprehensive short-term SSE detection using GNSS data from the GEONET in the southern Japan Trench. Nishimura (2021) discovered a bimodal depth distribution of SSEs (M_w 5.8–6.9) on the shallow (10–30 km depth) and deeper (40–60 km depth) parts of the plate interface (Fig. 1). SSEs avoid a depth of 30–40 km, where past megathrust earthquakes have repeatedly occurred (e.g., Mochizuki et al. 2008; Kubo et al. 2013). The bimodal depth distribution of SSEs is similar to that of shallow and deep slow earthquakes (shallower than 10 km depth and 25–40 km depth, respectively) in the Nankai Trough (Fig. 1).

Research on the relationship between slow and fast earthquakes along the Japan Trench has also made progress in recent years. In 2020, Kubo and Nishikawa (2020) revealed the along-dip complementary distribution of M_w 7 class megathrust earthquake ruptures (30–40 km depth) (e.g., Yamanaga and Kikuchi, 2004) and tectonic tremors (10–20 km depth) (e.g., Nishikawa et al. 2019) in the northern and southern Japan Trench. They suggested that the shallow slow-earthquake-genic regions in the northern and southern Japan Trench impede rupture propagation of megathrust earthquakes. In 2021, synchronous phenomena of tectonic tremors and swarms of small interplate earthquakes were observed in the shallow part of the southern Japan Trench (Obana et al. 2021). This observation implies that SSEs too small for onshore geodetic detection often trigger both tectonic tremors and earthquake swarms on the shallow plate interface of the Japan Trench.

Observational studies on slow earthquakes in the Japan Trench

In this section, we present the detailed slow earthquake distribution revealed by observational studies in the Japan Trench. We first present the spatiotemporal distribution of slow earthquakes from August 2016 (Sect. 3.2.1) because it is much more clearly resolved than before August 2016 (Sect. 3.2.2); the details of the spatiotemporal distribution of slow earthquakes before August 2016 are still enigmatic. We believe that a clear picture presented in Sect. 3.2.1 facilitates understanding of the subsequent Sects. (3.2.2 and 3.2.3). In the last part of this section, we review studies on repeating earthquakes, earthquake swarms, and foreshocks because the studies are closely related to slow earthquake observations in the Japan Trench (Sects. 3.2.1 and 3.2.2) and provide insights into slow earthquake activity before March 2011.

Spatiotemporal distribution of slow earthquakes from August 2016

The detailed tectonic tremor activity has been revealed by the S-net since August 2016. Figure 5 shows the spatiotemporal distribution of tectonic tremors (Nishikawa et al. 2019), VLFEs (Baba et al. 2020), and short-term SSEs (Nishimura 2021) from August 2016 to December 2021 along the Japan Trench. Nishikawa et al. (2019) detected the tectonic tremors until August 2018 by applying the envelope correlation method (Obara 2002; Ide 2010) to the S-net seismograms. We updated their tremor catalog until December 2021 using the same method as Nishikawa et al. (2019). Baba et al. (2020) detected VLFEs from January 2003 until July 2018 by a matched filter technique using synthetic waveforms of low-angle reverse faulting on the plate interface as templates. The templates were spaced approximately 40 km apart in the latitudinal and longitudinal directions. The SSE distribution was based on Nishimura (2021), who detected SSEs using the GNSS data of the GEONET until December 2019 in the southern Japan Trench. We updated the catalog until September 2021 using the same method as Nishimura (2021).

Fig. 5

Tectonic tremor activity in the Japan Trench from August 2016 to December 2021. a Epicenters of the tectonic tremors. Colored points indicate the epicenters of the tremors. Yellow and red stars represent VLFEs (Baba et al. 2020) and M_j ≥ 4 fast earthquakes, respectively. The fast earthquakes are from the Japan Meteorological Agency (JMA) catalog. Open and solid dashed green rectangles denote deep and shallow short-term SSEs, respectively (Nishimura 2021). The upper edges of the SSE fault models are indicated by the green solid lines. Black triangles represent S-net stations. The orange shaded polygons are subregions used for tremor detection by Nishikawa et al. (2019). The upper surface of the subducting Pacific Plate is indicated by the black contours at 10-km-depth intervals. b Spatiotemporal distribution of the tectonic tremors. We used Japanese Standard Time (UTC + 9 h). c–h Enlarged views of (b). The spatiotemporal windows of c–h are shown in (b)

We observed good correspondence between tectonic tremors and VLFEs in both their epicenters and origin times (Fig. 5b, d, and e). This is not surprising because tectonic tremors and VLFEs originate from the identical slow interplate slip phenomenon (Ide et al. 2007a; Kaneko et al. 2018; Masuda et al. 2020; Sect. 2.2.2; Fig. 3b). Based on this fact, hereafter, we assume that, as with the VLFEs, the detected tectonic tremors are located along the plate interface, although the source depths of the tectonic tremors are not well constrained (Nishikawa et al. 2019).

In the Japan Trench, tectonic tremors and VLFEs are distributed along the depth contours of 10–20 km of the plate interface (Ohta et al. 2019; Tanaka et al. 2019; Nishikawa et al. 2019; Baba et al. 2020). However, the tremors and VLFEs avoid the central Japan Trench (37–39° N), making the region an approximately 200-km-long gap of seismic slow earthquakes (Fig. 5a). This gap corresponds well with the large slip area of the 2011 Tohoku-Oki earthquake (e.g., Iinuma et al. 2012; the thick red contours in the Japan Trench in Fig. 1). This correspondence is the most prominent feature of the slow earthquake distribution in the Japan Trench.

The tectonic tremor activity along the Japan Trench and the southern end of the Kuril Trench has some common features with tectonic tremor activities in other subduction zones, implying that the slow-earthquake-genic regions in the Japan and Kuril trenches and other subduction zones share basically similar frictional and rheological properties. There are three prominent common features. The first one is tectonic tremor migration. In the southern end of the Kuril Trench, which is located to the north of 41° N, along-strike tremor migration episodes with a speed of ~ 10 km/day are observed approximately once a year (Tanaka et al. 2019; Nishikawa et al. 2019; Fig. 5b and c). Similar along-strike migration episodes with comparable speeds have been observed at the deeper (approximately 10 km/day; Obara 2002) and shallower (approximately 10–60 km/day; Yamashita et al. 2015; Annoura et al. 2017) parts of the Nankai Trough and at the deeper part of the Cascadia subduction zone (approximately 10 km/day; Rogers and Dragert 2003; Houston et al. 2011).

The second common feature is the simultaneous occurrence of tectonic tremors and short-term SSEs (Nishikawa et al. 2019; Fig. 5f and h). This composite fault slip phenomenon, called ETS (Sect. 2.2.1), has also been observed at the deeper part of the Cascadia subduction zone (Rogers and Dragert 2003) and at the deeper and shallower parts (Obara et al. 2004; Araki et al. 2017) of the Nankai Trough. However, in the Japan Trench, many tremor bursts occur without detectable SSEs (Nishimura 2021; Fig. 5b). This is probably because the SSE catalog (Nishimura 2021) is not complete near the trench axis. It is generally difficult to detect small (M_w $\le$ 6) SSEs near the trench axis using only onshore GNSS observations. Furthermore, it is also worth noting that deep ETS has not been reported in the Japan Trench. Although SSEs exhibit a bimodal depth distribution (10–30 km and 40–60 km) (Nishimura 2021; Fig. 5a), tectonic tremors have been observed only on the shallower plate interface.

The third common feature is the correlation between the recurrence intervals of tremor bursts and their seismic energy rates. Yabe et al. (2021) compared the recurrence intervals of tremor bursts with the energy rates of their 2–8 Hz bandpass-filtered seismic waves to find a correlation between them. In the southern end of the Kuril Trench (north of 41° N), tectonic tremor bursts repeated approximately every 0.5–1 year (Fig. 5b), and their median seismic energy rate was high (1700 J/s). In the northern Japan Trench, tremor episodes repeated approximately every 1–2 months (Fig. 5b), and their median seismic energy rate was low (830 J/s). The southern Japan Trench was characterized by medium recurrence intervals of approximately 3 months and medium median seismic energy rates (approximately 1400 J/s). The correlation between recurrence intervals of tremor bursts and their seismic energy rates has also been observed for deep tremors in the Cascadia subduction zone in both the along-strike and along-dip directions (Wech and Creager 2011; Idehara et al. 2014; Yabe and Ide 2014). Yabe et al. (2021) suggested that differences in the frictional strength of faults producing tectonic tremors cause the correlation. A stronger fault may endure stress loading for a longer period and produce more energetic tremors.

The tectonic tremor activity along the Japan Trench also has a feature distinct from that in the other subduction zones: We observed a coincidence of tectonic tremors, short-term SSEs, and a swarm of moderate interplate earthquakes in the Japan Trench. In July and August 2021, a M_w 6.6 short-term SSE and a burst of tectonic tremors occurred in the southern Japan Trench (35.5–36.5° N) (Fig. 5h). In early August, a swarm of moderate interplate earthquakes, with a maximum magnitude of M_w 5.8, occurred in an adjacent region. The fault model of the SSE was located to the north (around 36.4° N) of the earthquake swarm (around 36.2° N), and the tremor burst occurred to the south (around 35.8° N). We also observed a similar coincidence of tectonic tremors, short-term SSEs, and three M_j $\ge$ 4 interplate earthquakes in November and December 2018 in the same region (Fig. 5g). Composite fault slip phenomena of tectonic tremors, short-term SSEs, and a swarm of moderate interplate earthquakes have rarely been observed in subduction zones other than the Japan Trench. In the Nankai Trough and Cascadia subduction zone, ETSs on the deeper plate interface are not accompanied by swarms of interplate earthquakes (Rogers and Dragert 2003; Obara et al. 2004). Furthermore, moderate interplate seismicity (M $\ge$ 4) is very rare in the Nankai Trough and Cascadia subduction zone (Ide 2013). However, we note that a recent study (Yamamoto et al. 2022) has reported the simultaneous occurrence of a short-term SSE, VLFEs, and a swarm of interplate microearthquakes (M $\le$ 3) on the shallow plate interface (10 km depth or shallower) of the Nankai Trough. At the Boso Peninsula in the Sagami Trough, eastern Japan, swarms of moderate interplate earthquakes accompany short-term SSEs (Ozawa et al. 2003; Sagiya 2004), but concurrent tectonic tremors have not been observed. These differences in composite fault slip phenomena between subduction zones may reflect differences in interplate frictional properties of slow-earthquake-genic regions between subduction zones. Note, however, that each subduction zone has different observation conditions, such as station density and noise level, and that such differences might lead to apparent differences in composite fault slip phenomena.

Spatiotemporal distribution of slow earthquakes before August 2016

With respect to slow earthquake activity prior to the start of the S-net observations in August 2016, the VLFE activity along the entire Japan Trench (Baba et al. 2020) and SSE activity in the southern Japan Trench (Nishimura 2021) were revealed by the F-net and GEONET, respectively. Systematic and comprehensive detection of SSEs in the central and northern Japan Trench has not yet been conducted. However, in the central Japan Trench, ocean-bottom pressure gauge observations detected short-term SSEs in November 2008 and from the end of January 2011 to just before the largest foreshock (M_w 7.3) of the March 11, 2011, Tohoku-Oki earthquake (Ito et al. 2013). We note that one should be careful of the treatment of the SSE preceding the Tohoku-Oki earthquake, especially when discussing the slow earthquake distribution prior to the Tohoku-Oki earthquake. It is natural that the period just prior to and the region close to the epicenter of the Tohoku-Oki earthquake be analyzed in greater detail than the other periods and regions, given their scientific importance. This causes a spatiotemporal bias in the detectability of SSEs. Moreover, several SSEs similar to the SSE preceding the Tohoku-Oki earthquake (i.e., SSEs accompanied by earthquake swarms and/or tectonic tremors) have recently been reported in the southern Japan Trench (Nishimura 2021; Fig. 5h), implying that such SSEs are more common in the Japan Trench than previously thought.

Figure 6 shows the slow earthquake activity before and after the March 2011 Tohoku-Oki earthquake. VLFEs frequently occurred in the northern and southern Japan Trench both before and after the 2011 Tohoku-Oki earthquake (Fig. 6b and c). In the gap of tectonic tremors and VLFEs that we identified in Fig. 5a (37–39° N), VLFE activity was low even before the Tohoku-Oki earthquake (Fig. 6a and b). The VLFEs before the Tohoku-Oki earthquake appear to have avoided the huge interplate locked zone in the central Japan Trench (Suwa et al. 2006; the red shaded area in Fig. 6a), although ocean-bottom pressure gauge and seismometer observations suggested that a short-term SSE and tectonic tremors preceding the March 2011 Tohoku-Oki earthquake had occurred in the VLFE gap (Ito et al. 2013; Ito et al. 2015; Katakami et al. 2018; a green rectangle within the Tohoku-Oki earthquake rupture in Fig. 6a and b).

Fig. 6

Slow earthquake activity in the Japan Trench before and after the 2011 Tohoku-Oki earthquake. a Epicenters of the slow earthquakes. b Spatiotemporal distribution of the slow earthquakes before the March 11, 2011, Tohoku-Oki earthquake. Potential ETSs before the Tohoku-Oki earthquake are numbered 1–6. c Spatiotemporal distribution of the slow earthquakes from January 2015 to December 2021. For an explanation of the symbols, refer to the legend in the figure. The epicenter of the Tohoku-Oki earthquake and its largest foreshock are from the JMA catalog. The coseismic slip (10 m or larger) and afterslip (April 23, 2011 to December 10, 2011) of the Tohoku-Oki earthquake are based on Iinuma et al. (2012) and Iinuma et al. (2016), respectively. The afterslip distribution is shown at 0.4 m intervals. The interplate locked region prior to the Tohoku-Oki earthquake is based on the slip deficit distribution estimated by Suwa et al. (2006) (larger than 6 cm/year, i.e., approximately 70% plate coupling). Tectonic tremors and VLFEs are from Nishikawa et al. (2019) and Baba et al. (2020), respectively. In the figure, we showed tectonic tremors with a duration of more than 80 s. Short-term SSEs are from Ito et al. (2013) and Nishimura (2021). The decade-long acceleration of aseismic slip prior to the Tohoku-Oki earthquake (10 cm or larger) is from Yokota and Koketsu (2015). The transient aseismic slip in early 2015 in the northern Japan Trench is based on Honsho et al. (2019)

As shown in Fig. 6b, short-term SSEs and VLFEs sometimes occur simultaneously in close proximity. These coincidences are potential ETSs (Sect. 2.2.1). Here, if the distance between a VLFE epicenter and an SSE fault model was within 40 km and the VLFE occurred within 2 weeks before or after the short-term SSE period, we considered the event as a potential ETS. As a result, we identified six potential ETSs before the March 2011 Tohoku-Oki earthquake (Fig. 6b). However, Fig. 6b shows that most of the VLFEs were not accompanied by detectable short-term SSEs (Nishimura 2021). This may be due to the incompleteness of the SSE catalog near the trench axis, as mentioned in Sect. 3.2.1 Furthermore, the uncertainty in location of the VLFEs and SSEs may also make the correspondence between the two slow earthquakes worse.

Figure 6 also shows two slow fault slip phenomena similar to SSEs. One is the decade-long acceleration of aseismic slip on the deeper plate interface (approximately 30–60 km depth) of the central Japan Trench, which was discovered by the GEONET (Ozawa et al. 2012; Mavrommatis et al. 2014; Yokota and Koketsu 2015; Sect. 3.1.2; the yellow shaded area in Fig. 6a and b). This transient lasted from around 2002 to just before the 2011 Tohoku-Oki earthquake. Note that this decade-long transient may also have induced slips on the shallower plate interface (approximately 10–30 km depth), as suggested by repeating earthquake data (Mavrommatis et al. 2015). This transient slip resembles long-term SSEs (Yokota and Koketsu 2015). However, the duration of this transient slip (9 years) is longer than that of typical long-term SSEs (several months to several years), and the magnitude of this transient slip (M_w 7.7) is also larger than that of typical long-term SSEs (M_w 6.5–7.5) (Fig. 3a). Furthermore, it is unclear whether this transient slip recurs like the deep long-term SSEs in the Nankai Trough (e.g., Kobayashi 2014; Takagi et al. 2019), as it was observed only once in the decade prior to the 2011 Tohoku-Oki earthquake. Mavrommatis et al. (2014) suggested that this transient slip might be an aseismic slip intruding from the deep plate interface into the inside of the locked region prior to the coseismic rupture. A definite conclusion has not yet been reached as to the identity of this aseismic transient slip.

The other slow fault slip phenomenon similar to SSEs is a transient aseismic slip in the northern part (39.2–40.2° N) of the Japan Trench in early 2015 (Honsho et al. 2019; the blue dashed square in Fig. 6a). Honsho et al. (2019) detected this transient slip (approximately M_w 7.3) based on GNSS-acoustic observations and repeating earthquake activity. Fujiwara et al. (2022) also detected the probably identical transient slip as Honsho et al. (2019) using onshore GNSS observations. Although Honsho et al. (2019) considered this transient slip as a shallow SSE, it may not be a typical SSE. This is because the transient slip followed a large interplate earthquake of M_w 6.7 (Fujiwara et al. 2022). Therefore, one might consider the transient slip as a huge afterslip of the M_w 6.7 interplate earthquake, although it is too large to be a typical afterslip (Alwahedi and Hawthorne 2019). We found that the region where the transient slip was observed significantly overlaps with the region where Kawasaki et al. (1995) observed the 1992 Sanriku-Oki ultraslow earthquake, a transient aseismic slip of M_w 7.3–7.7 that followed a M_w 6.9 interplate earthquake (Sect. 3.1.1 and the blue dashed rectangle in Fig. 1). Due to the similarity and proximity between the 1992 and 2015 transient slip events, we interpret the 2015 event as a slow slip triggered by an earthquake as well as the 1992 event (Sect. 3.1.1). A transient aseismic slip with characteristics intermediate between SSEs and afterslip (Villegas-Lanza et al. 2016) may be a typical slip behavior on the shallow plate interface of the northern part (39–40° N) of the Japan Trench.

Repeating earthquakes, earthquake swarms, and foreshocks before March 2011

Slow earthquake activity along the Japan Trench prior to the 2011 Tohoku-Oki earthquake has been inferred from fast earthquake activity potentially indicative of slow earthquake occurrence (i.e., repeating earthquakes, earthquake swarms, and swarms of foreshocks) (Sect. 2.3). In the Japan Trench, repeating earthquakes have been vigorously studied since the first report by Matsuzawa et al. (2002). The spatiotemporal distribution of repeating earthquakes has been revealed in detail (Igarashi et al. 2003; Igarashi 2010; Uchida and Matsuzawa 2013; Igarashi 2020; red points in Fig. 7). As already described in Sect. 3.1.1, Igarashi et al. (2003) found that few continuous-type small repeating earthquakes (i.e., repeating earthquakes that repeat at almost regular intervals and are not clustered in time) were distributed within the huge locked zone in the central Japan Trench (Ito et al. 2000; Nishimura et al. 2000). Consistent with Igarashi et al. (2003), repeating earthquakes detected by Uchida and Matsuzawa (2013) are sparse within the large slip area of the Tohoku-Oki earthquake in the central Japan Trench (37–39° N) in comparison with the adjacent regions in the northern and southern Japan Trench (39–40° N and 36–37° N) (Fig. 7).

Fig. 7

Activity of repeating earthquakes and earthquake swarms along the Japan Trench before the 2011 Tohoku-Oki earthquake. a Epicenters of the repeating earthquakes and earthquake swarm events. b Spatiotemporal distribution of the repeating earthquakes and earthquake swarm events before the March 2011 Tohoku-Oki earthquake. Coincident events of SSEs and earthquake swarms are indicated by numbers 1–6 encircled in green. Coincident events of VLFEs and earthquake swarms are indicated by numbers 1–9 encircled in yellow. For an explanation of the other symbols, refer to the legend in the figure. The epicenter of the Tohoku-Oki earthquake and its largest foreshock are from the JMA catalog. The coseismic slip of the Tohoku-Oki earthquake (10 m or larger) is based on Iinuma et al. (2012). The VLFEs are from Baba et al. (2020). The short-term SSEs are from Ito et al. (2013) and Nishimura (2021). The repeating earthquakes are from Uchida and Matsuzawa (2013). The earthquake swarms (M_j ≥ 3) are from Nishikawa and Ide (2018) and Nishikawa et al. (2019). The foreshock-prone areas are based on Maeda and Hirose (2016) and Hirose et al. (2021). The 3 years periodic slow slip in the northern Japan Trench is from Uchida et al. (2016)

Repeating earthquakes have been used to detect transient aseismic slip on the plate interface in the Japan Trench (Matsuzawa et al. 2004; Uchida et al. 2004, 2016; Kato et al. 2012; Uchida and Matsuzawa 2013; Khoshmanesh et al. 2020). A M_j 5 class earthquake swarm including repeating earthquakes occurred on the shallow plate interface in the central Japan Trench a month before the March 2011 Tohoku earthquake, with their epicenters migrating toward the rupture initiation point of the Tohoku-Oki earthquake (Kato et al. 2012). This observation is a representative example of inferring the occurrence of a propagating SSE from repeating earthquake activity (Sect. 2.3).

Uchida et al. (2016) analyzed repeating earthquake activity from 1984 to 2011 along the entire Japan Trench and found that the interplate slip rates estimated from the repeating earthquake change with periods of 1–6 years. The periodic slip rate changes inferred from repeating earthquakes were widespread in the Japan Trench. They were also suggested by temporal changes in spatial gradients of the surface displacement rate field (Iinuma 2018). Furthermore, Uchida et al. (2016) showed that the occurrence rate of M_j 5 or greater earthquakes was positively correlated with the interplate slip rate. Based on these observations, Uchida et al. (2016) suggested that periodic SSEs widespread in the Japan Trench had stressed the surrounding regions and triggered M_j 5 or greater earthquakes.

Uchida et al. (2016) identified the northern part (39–40.4° N) of the Japan Trench (the black squares in Fig. 7a) as an area of pronounced 3-year periodicity. In and near this region, a repeating earthquake analysis by Uchida et al. (2004) also suggested SSEs preceding the 1989 M_w 7.4, 1992 M_w 6.9, and 1994 M_w 7.7 Sanriku-Oki earthquakes (Sect. 3.1.1). Furthermore, the far offshore region of the pronounced 3-year periodicity significantly overlaps with an area where SSEs triggered by earthquakes were geodetically observed in 1992 (Kawasaki et al. 1995; Kawasaki et al. 2001; Sect. 3.1.1; the blue dashed square in Fig. 1) and 2015 (Honsho et al. 2019; Fujiwara et al. 2022; Sect. 3.2.2; the blue dashed square in Fig. 6a). Uchida et al. (2016) probably captured the recurrence of these SSEs. In fact, the 1992 event (i.e., the 1992 Sanriku-Oki ultraslow earthquake) (Kawasaki et al. 1995, 2001) corresponds with a peak of the periodic interplate slip rate in Uchida et al. (2016).

In the Japan Trench, there have been a number of attempts to infer the occurrence of SSEs from earthquake swarm activity (Marsan et al. 2013; Nishikawa and Ide 2017, 2018; Nishikawa et al. 2019) and activity of swarms of foreshocks (i.e., earthquake swarms followed by megathrust earthquakes) (Matsumura 2010; Kato et al. 2012; Maeda and Hirose 2016; Nishikawa and Ide 2018; Kubo and Nishikawa 2020; Hirose et al. 2021).

Nishikawa and Ide (2018) considered increases in the seismicity rate that do not obey the Omori–Utsu’s aftershock law (Utsu 1957; Utsu et al. 1995; Sect. 2.3) as earthquake swarms and detected such seismicity rate increases in the central and southern Japan Trench using a statistical seismicity model called the epidemic-type aftershock-sequence (ETAS) model (e.g., Ogata 1988; Zhuang et al. 2002). The ETAS model usually expresses the seismicity rate as the summation of a constant background seismicity rate and aftershock rates derived from Omori–Utsu’s aftershock law. Therefore, increases in the seismicity rate that do not obey the Omori–Utsu’s aftershock law appear as anomalous increases in the seismicity rate in the ETAS model analysis (e.g., Llenos et al. 2009; Okutani and Ide 2011). As shown in Fig. 7, M_j $\ge$ 3 earthquake swarms frequently occur in the southern Japan Trench. Furthermore, Nishikawa and Ide (2018) found that earthquake swarms including repeating earthquakes had repeatedly occurred in a region around 36.2° N, 142° E (Fig. 7a). The earthquake swarm activity was the most active during the weeks prior to the 1982 and 2008 M_j 7 Ibaraki-Oki earthquakes (Matsumura 2010; Nishikawa and Ide 2018), which ruptured the plate interface downdip of the recurrent earthquake swarms. Nishikawa and Ide (2018) inferred from these observations that SSEs recur in the region around 36.2° N, 142° E and that the SSEs preceding the M_j 7 Ibaraki-Oki earthquakes were possibly aseismic slip acceleration in the nucleation phase of the Ibaraki-Oki earthquakes (e.g., Dieteich, 1992; Ohnaka 1992; McLaskey 2019). Recently, seismic slow earthquakes (i.e., tectonic tremors and VLFEs) were observed in the region around 36.2° N, 142° E (Nishikawa et al. 2019; Baba et al. 2020; Fig. 6a). The coincidence of a short-term SSE, burst of tectonic tremors, and earthquake swarms was also observed in July and August of 2021 in the same region (Fig. 5h). These recent observations are consistent with the inference made by Nishikawa and Ide (2018).

Maeda and Hirose (2016) and Hirose et al. (2021) systematically surveyed foreshock activity (M_j 5 or greater) of M_j $\ge$ 6 earthquakes in the Japan Trench from 1961 to 2010. They identified areas of pronounced foreshock activity far offshore in the Japan Trench (the orange shaded areas in Fig. 7a). In these areas, 38% of the M_j $\ge$ 6 mainshock earthquakes were preceded by a swarm of foreshocks (three or more M_j $\ge$ 5 earthquakes within 10 days before the mainshock) (Maeda and Hirose 2016). These foreshock-prone areas are close to areas where tectonic tremors and VLFEs occur (Figs. 6a and 7a). Furthermore, Hirose et al. (2021) showed that the occurrence rate of foreshock activity increases as it approaches the time of mainshock occurrence and that this foreshock rate acceleration cannot be fully reproduced by the ETAS model (Ogata 1988; Zhuang et al. 2002), which describes the earthquake-to-earthquake triggering. These results suggest that the foreshock activity in the Japan Trench cannot be explained by earthquake-to-earthquake triggering and that transient aseismic phenomena such as slow earthquakes may be involved in the triggering of the foreshocks. Moreover, Hirose et al. (2021) found that foreshock activity in the northern Japan Trench had been synchronized with the 3-year periodic SSEs reported by Uchida et al. (2016) (the far offshore black square in Fig. 7a) and suggested that the periodic SSEs had excited the foreshock activity.

Although repeating earthquakes and earthquake swarms have been used to infer the occurrence of SSEs in the Japan Trench (e.g., Kato et al. 2012; Marsan et al. 2013; Uchida et al. 2016; Nishikawa and Ide 2018), they have hardly ever been compared with independently observed SSEs or VLFEs in the Japan Trench. Here, we compared the catalogs of small repeating earthquakes (Uchida and Matsuzawa 2013; Nishikawa et al. 2019), M_j $\ge$ 3 earthquake swarms (Nishikawa and Ide 2018; Nishikawa et al. 2019), SSEs (Nishimura 2021), and VLFEs (Baba et al. 2020). In Fig. 8a and b, we counted repeating earthquakes and earthquake swarm events around SSE fault models (inside the rectangular faults or within 20 km of the fault models) within 80 days before or after the central dates of the SSE occurrence periods. Here, we excluded SSEs associated with M_w $\ge$ 5.8 fast earthquakes (Nishimura 2021) because not SSEs but afterslip following the M_w $\ge$ 5.8 earthquakes may have triggered repeating earthquakes and earthquake swarms.

Fig. 8

Coinciding events of repeating earthquakes, earthquake swarms, and slow earthquakes in the southern Japan Trench. a and b Histograms of the occurrence times of repeating earthquakes and earthquake swarm events, respectively, relative to an SSE central date. These repeating earthquakes and earthquake swarm events occurred inside the SSE fault models or within 20 km of the fault models. c Epicenters of the coincident events. The coincident events of earthquake swarms and SSEs are indicated by numbers 1–6 in green, as in Fig. 7b. The coinciding events of earthquake swarms and VLFEs are indicated by numbers 1–9 in yellow, as in Fig. 7b. The VLFEs are from Baba et al. (2020). Note that the VLFEs sometimes have identical epicenters because they were detected using a matched filter technique (Baba et al. 2020). The short-term SSEs are from Nishimura (2021). Tectonic tremors are from Nishikawa et al. (2019) and this study. Repeating earthquakes are based on Uchida and Matsuzawa (2013). The earthquake swarms are from Nishikawa and Ide (2018) and Nishikawa et al. (2019). The VLFEs, SSEs, earthquake swarm events, and repeating earthquakes are colored according to their occurrence time

Figure 8a and b shows that the counts of repeating earthquakes and earthquake swarm events have a maximum within 10 days before or after the central dates of the SSE occurrence periods. This is consistent with the idea that SSEs trigger repeating and earthquake swarms in their vicinities (e.g., Kato et al. 2012; Marsan et al. 2013; Uchida et al. 2016; Nishikawa and Ide 2018). However, with respect to earthquake swarms, the sample size (18 earthquake swarm events) was too small to make a statistical argument. As for repeating earthquakes, the tendency that repeating earthquakes are more likely to occur within 10 days before or after the SSE central dates than in the other periods is statistically significant (p = 1%). Here, we conducted a statistical test with a null hypothesis that repeating earthquakes are equally likely to occur on the dates within 80 days before or after the SSE central dates and used a significance level of 5%. However, in Fig. 8a, the repeating earthquakes also occurred on days not close to the central dates of the SSEs. This is probably because repeating earthquakes are triggered by not only SSEs but also interplate steady slip (e.g., Nadeau and Johnson 1998; Nadeau and McEvilly 1999; Matsuzawa et al. 2002; Igarashi et al. 2003). Furthermore, as mentioned in Sect. 3.2.2, the incompleteness of the SSE catalog near the trench axis may have affected the results in Fig. 8a and b.

Figure 8c shows the spatial distribution of coincident events of earthquake swarms and SSEs or earthquake swarms and VLFEs. From January 1995 to March 2011, we observed six coincident events of earthquake swarms and SSEs and nine coincident events of earthquake swarms and VLFEs (Fig. 7b). Here, if an earthquake swarm event occurs within 2 weeks before or after an SSE period inside the SSE rectangular fault or within 20 km of the fault, we regard it as a coincident event. Similarly, if an earthquake swarm event occurs within 2 weeks before or after a VLFE within 40 km of the VLFE epicenter, we regard it as a coincident event. Small repeating earthquakes included in earthquake swarm activity are indicated by colored stars in Fig. 8c. We found that the coincident events had repeatedly occurred at the same locations (Fig. 8c). Specifically, they recurred in regions around 36.3° N, 142.5° E, 36.2° N, 142.0° E, and 35.1° N, 141.4° E. The region around 36.3° N, 142.5° E corresponds with the area where Obana et al. (2021) observed coinciding events of tectonic tremors and swarms of interplate microearthquakes in March and June of 2017 (Sect. 3.1.2). Furthermore, in the region around 36.2° N, 142.0° E, the coincidence of a short-term SSE, burst of tectonic tremors, and an interplate earthquake swarm was also observed in July and August of 2021 (Fig. 5h). These observations imply that coincident events of slow and fast earthquakes are common on the shallow plate interface of the southern Japan Trench.

Experimental studies on slow earthquakes in the Japan Trench

Experimental studies on slow earthquakes in the Japan Trench have been focused on the slip behavior of the shallow plate interface (Ikari et al. 2015; Ito and Ikari 2015; Ito et al. 2017; Ikari and Kopf 2017; Sawai et al. 2017). These studies were motivated by the observational findings that an SSE preceding the Tohoku-Oki earthquake occurred on the shallow plate interface of the central Japan Trench and that the shallow plate interface subsequently underwent a huge coseismic slip breaching the trench axis (Kato et al. 2012; Ito et al. 2013). Before the 2011 Tohoku-Oki earthquake, the shallow subduction plate interface was thought to exclusively host aseismic creep (e.g., Nishizawa et al. 1992; Ikari et al. 2015). This is consistent with the standard model of the megathrust slip behavior at that time (Scholz 1998), in which the shallowest part of the plate interface is characterized by rate-strengthening behavior and stable sliding. However, this view was challenged by the aforementioned observations implying the unstable nature of the shallow plate interface. With this background, the above experimental studies reconsidered the frictional behavior of the shallow plate interface using core samples from the shallow plate boundary of the Japan Trench.

Ikari et al. (2015) conducted an experiment in which a core sample retrieved from the shallow plate boundary fault zone (~ 7 km landward of the trench axis and 822 m below sea floor) by the Japan Trench Fast Drilling Project (JFAST) (Chester et al. 2013) was sheared at a rate comparable to the plate convergence rate (8.5 cm/year). They found that the clay-rich sample exhibited rate-weakening behavior, which is suitable for coseismic rupture propagation, and observed spontaneous slow strength perturbations similar to SSEs. Ikari et al. (2015) proposed that these experimental results explain both the preseismic SSE and coseismic rupture of the Tohoku-Oki earthquake. Furthermore, Ikari and Kopf (2017) showed that weak clay-rich fault samples from shallow parts of several subduction zones also exhibit rate-weakening behavior and produce slow strength perturbations when sheared at a rate comparable to plate velocities. They suggested that the observed unstable nature of the shallow fault zone materials may facilitate the propagation of coseismic rupture and slip at shallow depths in subduction zones.

Ito et al. (2017) also used a core sample retrieved by JFAST and showed that an increase in the slip rate can induce a change from slip-strengthening friction to slip-weakening friction. They found that velocity steps with initial slip rates consistent with afterslip following the largest (M_w 7.3) foreshock of the Tohoku-Oki earthquake exhibit significant slip weakening. Based on this experiment, they suggested that the accelerated interplate aseismic slip due to the afterslip following the largest foreshock (Ando and Imanishi 2011; Kato et al. 2012; Ohta et al. 2012; Sect. 3.1.2) might have been a favorable initial condition for the huge coseismic slip of the Tohoku-Oki earthquake.

Numerical simulations on slow earthquakes in the Japan Trench

There are several simulation studies that consider slow earthquakes (especially SSEs) in the Japan Trench (Mitsui et al. 2012; Ohtani et al. 2014; Shibazaki et al. 2019; Barbot 2020; Nakata et al. 2021). These are earthquake cycle simulations based on rate-and-state-dependent friction (Dieterich 1979; Sect. 2.4). Some studies reproduced SSEs in the Japan Trench by assuming large characteristic slip distances (Ohtani et al. 2014; Nakata et al. 2021; Sect. 2.4), and others reproduced the SSEs by assuming friction that exhibits rate-weakening behavior at low slip rates but switches to rate-strengthening behavior at high slip rates (Shibazaki et al. 2019; Sect. 2.4).

Ohtani et al. (2014) assumed that regions outside asperities show rate-weakening behavior but have a large characteristic slip distance. Asperities here mean large slip areas of past megathrust earthquakes (Yamanaka and Kikuchi 2004), as described in Sect. 3.1.1. Ohtani et al. (2014) found that recurrent SSEs occur on the shallow plate interface in the southern Japan Trench and pointed out that these SSEs might correspond to a decrease in the interplate coupling in the southern part (~ 37° N) of the Japan Trench prior to the 2011 Tohoku-Oki earthquake detected by GNSS observations (Geospatial Information Authority of Japan GSI 2011). Shibazaki et al. (2019) assumed that the shallow plate interface near the trench axis has a frictional property that exhibits rate-weakening behavior at low slip rates but switches to rate-strengthening behavior at high slip rates. Their model reproduced the preseismic SSE, the largest foreshock, and coseismic rupture of the Tohoku-Oki earthquake, although the location of the simulated SSE (the shallow part of the northern Japan Trench) was different from that of the observed SSE (the shallow part of the central Japan Trench) (Ito et al. 2013).

No previous simulation studies have reproduced or considered the complex slow earthquake distribution along the Japan Trench (Figs. 5 and 6). However, Nakata et al. (2021) attempted to reproduce along-strike changes in the slip behavior of the Japan Trench. They assumed rate-weakening behavior with a large characteristic slip distance in the southern and northern Japan Trench and rate-weakening behavior with a small characteristic slip distance in the central Japan Trench, considering the along-strike distribution of interplate sedimentary units revealed by multichannel seismic reflection surveys (Tsuru et al. 2002; Sect. 5.1.1). They reproduced the coseismic rupture of the Tohoku-Oki earthquake in the central Japan Trench (e.g., Ide et al. 2011; Iinuma et al. 2012; Lay 2018) and the large afterslip in the shallow part of the southern Japan Trench (Uchida and Matsuzawa 2013; Sun and Wang 2015; Iinuma et al. 2016; Tomita et al. 2017; Honsho et al. 2019; Tomita et al. 2020; Watanabe et al. 2021; the blue contours in Fig. 6a). The simulation results of Nakata et al. (2021) also show that transient slow fault slips repeatedly occur in the southern Japan Trench during interseismic periods. These transient aseismic slips might correspond to the slow earthquakes observed in the southern Japan Trench (Nishikawa et al. 2019; Baba et al. 2020; Nishimura 2021; Figs. 5 and 6), although the characteristic time scale of the transient slips (tens of years) and that of the observed SSEs (tens of days) are different.

Comparison of the slow earthquake distribution along the Japan Trench with the distribution of other fault slip phenomena

Coseismic slip, afterslip, and interseismic interplate coupling of the M_w 9.0 Tohoku-Oki earthquake

As already described in Sect. 3.2.1, the approximately 200-km-long along-strike gap of tectonic tremors in the central Japan Trench corresponds to the region ruptured by the Tohoku-Oki earthquake (Figs. 5a and 6). In other words, the tremor-genic regions in the southern and northern Japan Trench did not slip significantly during the Tohoku-Oki earthquake (Figs. 5a and 6). The huge coseismic slip in the central Japan Trench and the rupture termination in the southern and northern Japan Trench are features common to most coseismic slip models of the Tohoku-Oki earthquake (e.g., Ide et al. 2011; Simons et al. 2011; Iinuma et al. 2012). For more details on the coseismic rupture characteristics of the Tohoku-Oki earthquake, refer to a review paper by Lay (2018).

The northern and southern Japan Trench experienced afterslip following the Tohoku-Oki earthquake. The afterslip on the shallow plate interface has been suggested by GNSS-acoustic observations and repeating earthquake analyses in both the regions (Uchida and Matsuzawa 2013; Sun and Wang 2015; Iinuma et al. 2016; Tomita et al. 2017; Honsho et al. 2019; Tomita et al. 2020; Watanabe et al. 2021; blue contours in Fig. 6a). The VLFE occurrence rates in the shallow parts of these two regions also sharply increased after the Tohoku-Oki earthquake (Matsuzawa et al. 2015; Baba et al. 2020). A similar correspondence between VLFEs and afterslip following the 2003 M_w 8.0 Tokachi-Oki earthquake has been reported in the southern end of the Kurile Trench (Miyazaki et al. 2004; Asano et al. 2008; Uchida et al. 2009b; Itoh et al. 2019; Baba et al. 2020). Simons et al. (2011) point out that the area near the trench axis of the southern Japan Trench (around 36–37° N, 142.5° E) is a gap of megathrust earthquakes and a region of possible high seismic hazard. However, the large afterslip of the Tohoku-Oki earthquake (Uchida and Matsuzawa 2013; Sun and Wang 2015; Iinuma et al. 2016; Tomita et al. 2017, 2020; Honsho et al. 2019; Watanabe et al. 2021) and the numerous slow earthquakes (i.e., tectonic tremors, VLFEs, and short-term SSEs) (Matsuzawa et al. 2015; Nishikawa et al. 2019; Baba et al. 2020; Nishimura 2021) imply that the accumulated elastic strain in this area has so far been released aseismically.

The spatial distribution of VLFEs before the 2011 Tohoku-Oki earthquake (Matsuzawa et al. 2015; Baba et al. 2020) corresponds well with the spatial distribution of tectonic tremors after the Tohoku-Oki earthquake (Fig. 6). Given the related physical origin of tectonic tremors and VLFEs (Kaneko et al. 2018; Masuda et al. 2020; Sect. 2.2.2), this observation suggests that there was no significant change in the distribution of seismic slow earthquakes (tectonic tremors and VLFEs) before and after the Tohoku-Oki earthquake. Even before the Tohoku-Oki earthquake, a 200-km-long gap of seismic slow earthquakes existed in the central Japan Trench. As already mentioned in Sect. 3.2.2, this gap corresponds well with the huge interplate locked zone revealed by the GNSS observations before the Tohoku-Oki earthquake (Ito et al. 2000; Nishimura et al. 2000; Nishimura et al. 2004; Suwa et al. 2006; Hashimoto et al. 2009; Loveless and Meade 2010; the red shaded area in Fig. 6a). However, we note that these studies could not resolve the slip deficit near the trench axis due to the limited offshore resolution of onshore GNSS stations. The slip deficit near the trench axis is substantially affected by model assumptions. Some assumed complete interplate decoupling at the trench axis (e.g., Suwa et al. 2006), and others assumed complete coupling (Wallace et al. 2009). Lindsey et al. (2021) recently proposed a new method for estimating interplate coupling that assumes nonnegative shear stress accumulation rates on the plate interface during an interseismic period, to improve the offshore resolution. The application of this new method to the Japan Trench (Lindsey et al. 2021) suggested that the region near the trench axis in the central Japan Trench had been almost completely coupled before the 2011 Tohoku-Oki earthquake, while the shallow parts of the northern and southern Japan Trench had been partially coupled. This coupling distribution corresponds well with the spatial distribution of VLFEs before the 2011 Tohoku-Oki earthquake (Matsuzawa et al. 2015; Baba et al. 2020; Fig. 6a).

It is still not clear whether the spatial complementarity between M_w 9 class megathrust coseismic rupture and seismic slow earthquakes (Fig. 6a) is true only for the 2011 Tohoku-Oki earthquake or also for past and future M_w 9 class earthquakes in the Japan Trench. However, Ikehara et al. (2016) examined thick turbidite units in sediment cores collected at the trench axis and suggested that the shallow part of the central Japan Trench had repeatedly experienced huge earthquakes similar to the 2011 Tohoku-Oki earthquake (i.e., the 1454 Kyotoku and 869 Jogan earthquakes). Furthermore, the thick turbidite units related to the 2011 Tohoku-Oki, 1454 Kyotoku, and 869 Jogan earthquakes (Ikehara et al. 2016) were observed only in the central Japan Trench (Ikehara et al. 2018), corresponding well with the observed distribution of slip to the trench during the 2011 Tohoku-Oki earthquake (Kodaira et al. 2020, 2021). In other words, in the northern and southern Japan Trench, they found no evidence implying a slip to the trench during the 1454 Kyotoku and 869 Jogan earthquakes. Therefore, slip-to-the-trench events (i.e., Mw 9 class earthquakes breaching the trench axis) might be characteristic of the central Japan Trench (Kodaira et al. 2020), and the spatial complementarity between M_w 9 class megathrust coseismic rupture and seismic slow earthquakes might be true for past M_w 9 class earthquakes in the Japan Trench.

Large and small interplate earthquakes

Figure 9 shows asperities (i.e., areas where the amount of slip is half the maximum slip or larger) of M_w 7–8 class interplate earthquakes in the Japan Trench from 1936 (Nagai et al. 2001; Murotani et al. 2003; Yamanaka and Kikuchi 2004; Earthquake Research Center 2005; Mochizuki et al. 2008; Research Center for Seismology, Volcanology, and Disaster Mitigation 2008; Ohta et al. 2012, Kubo et al. 2013; Kubo and Nishikawa 2020; Sect. 3.1.1).

Fig. 9

Comparison between megathrust asperities and the slow earthquake distribution in the Japan Trench. a Megathrust asperities and epicenters of the slow earthquakes. b and c Enlarged views of the northern and southern Japan Trench, respectively. For an explanation of the symbols, refer to the legend in the figure. The megathrust asperities are from Nagai et al. (2001), Murotani et al. (2003), Yamanaka and Kikuchi (2004), Earthquake Research Center (2005), Mochizuki et al. (2008), Research Center for Seismology, Volcanology, and Disaster Mitigation (2008), Ohta et al. (2012), Kubo et al. (2013), and Kubo and Nishikawa (2020). The slip distribution of the March 11, 2011, Tohoku-Oki earthquake and the 1896 Meiji Sanriku tsunami earthquake is from Iinuma et al. (2012) and Tanioka and Seno (2001), respectively. The rupture initiation points of the 1968 M_w 8.2 Tokachi-Oki and 1994 M_w 7.7 Sanriku-Oki earthquakes are from the JMA catalog. The possible rupture area of the 1677 Enpo Boso-Oki tsunami earthquake is based on the Japan Society of Civil Engineers (2002). Small repeating earthquakes are from Uchida and Matsuzawa (2013). Repeating earthquakes included in earthquake swarms are based on Nishikawa and Ide (2018) and Nishikawa et al. (2019). Tectonic tremors are from Nishikawa et al. (2019) and this study. The VLFEs are from Baba et al. (2020). Short-term SSEs are from Ito et al. (2013), Nishimura (2021), and this study. The location of the subducting seamount in the southern Japan Trench is based on Mochizuki et al. (2008). The northeastern limit of the Philippine Sea Plate is from Uchida et al. (2009a). The intersection of slab and upper-plate Moho is based on Bassett et al. (2016)

The asperities of M_w 7–8 class interplate earthquakes (red contours in Fig. 9) are complementary to the slow earthquake distribution in the entire Japan Trench (green symbols in Fig. 9). The asperities are mainly distributed at 20–40 km depth, whereas slow earthquakes avoid this depth range. The Tohoku-Oki earthquake is the only instrumentally recorded megathrust earthquake that significantly ruptured the shallow slow-earthquake-genic region (10–20 km depth) in the central Japan Trench (Fig. 9a). Nishimura (2021) pointed out that short-term SSEs in the southern Japan Trench predominantly occur at depths of 10–30 km and greater than 40 km, avoiding the source depths of large interplate earthquakes (30–40 km) (Fig. 9a and c). The complementarity between large coseismic slip areas and shallow slow earthquakes has also been observed at the southern end of the Kuril Trench (north of 41° N) (Asano et al. 2008; Baba et al. 2020). No instrumentally recorded megathrust earthquakes are known to have significantly ruptured the shallow slow-earthquake-genic region (10–20 km depth) at the southern end of the Kuril Trench (Fig. 5a), although the M_w 9 class megathrust earthquake in the seventeenth century, which was identified by tsunami deposits (Nanayama et al. 2003), might have ruptured the shallow slow-earthquake-genic region (Satake et al. 2008; Ioki and Tanioka 2016).

In contrast to large interplate earthquakes, we observed no clear complementarity between small interplate earthquakes and slow earthquakes in the dip direction. The red dots in Fig. 9 show the distribution of small repeating earthquakes on the plate interface (mostly from M_j 2.5–4.5) (Uchida and Matsuzawa 2013). Repeating earthquakes are located at 10–60 km depth in the northern and central Japan Trench. In the southern Japan Trench, they occur at depths ranging from 10–80 km. Note that the upper plate of the deeper part of the southern Japan Trench is the subducting Philippine Sea Plate (Uchida et al. 2009a, b) and not a typical mantle wedge. The subducting Philippine Sea Plate is thought to lower the temperature of the upper surface of the Pacific Plate (Hasegawa et al. 2007) and allow small interplate earthquakes to occur at greater depths in the southern Japan Trench (Nakajima et al. 2009). As shown in Fig. 9, the source depths of the repeating earthquakes significantly overlap those of the slow earthquakes (i.e., 10–30 km and 40 km depth or deeper). These observations indicate that in contrast to large interplate earthquakes of M_w 7–8, small interplate earthquakes and slow earthquakes (i.e., tectonic tremors, VLFEs, and short-term SSEs) can occur under similar pressure–temperature conditions. Furthermore, these observations are slightly incompatible with the widely accepted view that slow earthquakes are intermediate-type fault slips that are transitional between fast earthquakes and stable sliding, with the transition proceeding predominantly in the dip direction (Obara and Kato 2016).

In addition, we note that some of the repeating earthquakes in Fig. 9 (red dots) occurred during periods of earthquake swarm activity (yellow stars). As described in Sect. 2.3, they were potentially triggered by slow earthquakes (Nishikawa and Ide 2018; Nishikawa et al. 2019). Consistent with this idea, they are mainly concentrated in the shallow part of the Japan Trench and located near the source regions of shallow slow earthquakes (i.e., tectonic tremors, VLFEs, and short-term SSEs) (Fig. 9b and c), while the whole of the repeating earthquakes is more widely distributed along the plate interface.

Tsunami earthquakes

Interplate earthquakes that generate huge tsunamis despite weak ground motions are called tsunami earthquakes. Tsunami earthquakes are considered to be fault slip events with a longer duration than regular earthquakes of comparable seismic moment (Kanamori 1972). They occur on the plate interface near the trench axis (e.g., Kanamori 1972; Ide et al. 1993; Kanamori and Kikuchi 1993). Soft subducted sediments on the shallow plate interface are thought to cause their rupture processes to be slower than those of regular earthquakes (Kanamori and Kikuchi 1993; Tanioka and Sataka 1996).

In 1896, the M_w 8.0 Meiji Sanriku tsunami earthquake occurred in the northern Japan Trench (Kanamori 1972; Tanioka and Sataka 1996; Tanioka and Seno 2001; the open purple rectangle in Fig. 9a and b). The earthquake is considered to have ruptured the very shallow plate interface of the northern Japan Trench. The source region is close to and slightly shallower than the shallow tremor-genic region (approximately 10–20 km depth) in the northern Japan Trench (Fig. 9b). According to a fault slip model inverted from tsunami waveform data of the 2011 Tohoku-Oki earthquake (Satake et al. 2013), this shallow plate interface of the northern Japan Trench slipped again during the 2011 Tohoku-Oki earthquake, and the slip amount reached up to 36 m. However, bathymetry observations (Fujiwara et al. 2017) have not found evidence of such a large slip breaching the trench axis in the northern Japan Trench.

Besides the 1896 Meiji Sanriku tsunami earthquake, many large interplate earthquakes have repeatedly occurred in the northern Japan Trench (Yamanaka and Kikuchi 2004; red contours in Fig. 9b). Among them, the 1994 M_w 7.7 Sanriku-Oki earthquake is known to have started its coseismic rupture like a tsunami earthquake (Nakayama and Takeo 1997). The epicenter of this earthquake was located at 40.43° N and 143.75° E, close to the shallow tremor-genic region (an open purple diamond in Fig. 9a and b). After the rupture initiation, the coseismic rupture propagated downdip toward the west (landward) at a slow speed of approximately 1.8 km/s. To the west of 143°E, the rupture propagated at a speed of 3.0 km/s. A similar slow rupture (1.1–1.5 km/s) was also observed during the 1992 M_w 7.6 Nicaragua tsunami earthquake (Ide et al. 1993; Kanamori and Kikuchi 1993).

Tsunami earthquakes may also have occurred close to the shallow slow-earthquake-genic region of the southern Japan Trench (Fig. 9c). The 1677 Enpo Boso-Oki earthquake is considered a tsunami earthquake because of its huge tsunami and weak ground shaking (the dashed purple rectangle in Fig. 9a and c). From historical documents and tsunami deposits, Yanagisawa et al. (2016) estimated a slip model for this earthquake. They suggested that the earthquake may have ruptured the area near the trench axis of the southern Japan Trench (mainly south of 36° N) and caused a fault slip of 11–16 m.

In the central Japan Trench, a typical tsunami earthquake has not been observed. However, Ide et al. (2011) pointed out that the shallow rupture of the Tohoku-Oki earthquake in the central Japan Trench had only weakly radiated high-frequency seismic waves despite the prodigious slip amount (tens of meters) and resembled tsunami earthquakes.

In summary, the tsunami earthquakes in the Japan Trench ruptured the plate interface close to the shallow slow-earthquake-genic regions (Fig. 9). They are characterized by a duration longer than and a rupture propagation speed slower than regular earthquakes (e.g., Kanamori 1972; Nakayama and Takeo 1997; Ide et al. 2011), which is probably caused by soft subducted sediments on the shallow plate interface (Kanamori and Kikuchi 1993; Tanioka and Satake, 1996). These rupture characteristics of tsunami earthquakes are qualitatively similar to those of slow earthquakes. However, the rupture speed of tsunami earthquakes, which is roughly 50% slower than fast regular earthquakes (Ide et al. 1993; Kanamori and Kikuchi 1993; Nakayama and Takeo 1997), is approximately four orders of magnitude faster than the typical rupture propagation speed (10 km/day) of short-term SSEs (e.g., Dragert et al. 2001). This indicates that tsunami earthquakes and slow earthquakes are quantitatively very different phenomena, although they are both slow compared to fast, regular earthquakes. The difference is also evident in their moment–duration scaling relations (Fig. 3a).

Seismicity parameters

The Japan Trench has a high interplate seismicity rate compared to the Nankai Trough and Cascadia subduction zone (Ide 2013), and the characteristics of the interplate seismicity have been studied vigorously. In this section, we compare the slow earthquake distribution with the spatial distributions of parameters characterizing seismicity. Here, we selected the Gutenberg–Richter relationship’s b value (Nanjo et al. 2012; Tormann et al. 2015), tidal response of seismicity (Tanaka 2012), and p value of the Omori–Utsu’s aftershock law (Ogata 2011; Ueda 2021). These parameters showed a good correspondence with the slow earthquake distribution.

Gutenberg–Richter relationship’s b value and tidal response of seismicity

Spatiotemporal variations of the Gutenberg–Richter relationship’s b value (Nanjo et al. 2012; Tormann et al. 2015) and tidal response of seismicity (Tanaka 2012) in the Japan Trench have been investigated in detail (Fig. 10a and b). The b value is the power exponent of the magnitude–frequency distribution of earthquakes (Guternberg and Richter 1944). It reflects the relative number of small earthquakes compared to large ones. A low b value indicates relatively frequent large earthquakes. According to Nanjo et al. (2012) and Tormann et al. (2015), the b value inside the source region of the March 2011 Tohoku-Oki earthquake gradually decreased from around 2005 until the occurrence of the earthquake. The decrease in the b value in the central Japan Trench may reflect an increase in the differential stress in the locked region prior to the Tohoku-Oki earthquake because rock experiments (e.g., Scholz 1968; Goebel et al. 2013) and natural earthquake observations (e.g., Schorlemmer et al. 2005; Spada et al. 2013; Nishikawa and Ide 2014; Scholz 2015; Petruccelli et al. 2019) suggest that the b value is negatively correlated with differential stress in the crust.

Fig. 10

Comparison between spatial distributions of seismicity parameters and the slow earthquake distribution in the Japan Trench. a Gutenberg–Richter relationship’s b value (Tormann et al. 2015). The b values were estimated from January 1, 2004, to March 11, 2011. In gray areas, Tormann et al. (2015) did not estimate b values because the magnitude–frequency distributions of the areas were not well described by the Gutenberg–Richter relationship. b Tidal response of seismicity (Tanaka 2012). p values of Schuster’s test during the 3000 days prior to the Tohoku-Oki earthquake are shown. A low p value indicates a high tidal response. The p values were calculated for seismicity in spatial windows of 200 km × 200 km, whose centers correspond with the centers of the squares of 50 km × 50 km in the figure (the black squares). c p value of the Omori–Utsu’s aftershock law (Ogata 2011). The p values were estimated using M_j 5 or greater earthquakes from 1926 to 2008. Note that the slow and fast earthquakes are colored gray in (c). For an explanation of the symbols, refer to the legend in the figure. The slow earthquakes, megathrust earthquakes, and small repeating earthquakes are from Fig. 9

Note that the b value also depends on the focal mechanisms of sampled earthquakes, with thrust faults being characterized by lower b values than normal and strike faults (e.g., Schorlemmer et al. 2005; Bürgmann et al. 2016). The decade-scale decrease in the b value might have been affected by a temporal change in the fraction of the faulting types of the earthquakes sampled by Nanjo et al. (2012) and Tormann et al. (2015) because their analyses included small earthquakes whose faulting types were unclear. One should be careful when interpreting spatiotemporal changes in the b value because the fraction of the faulting types can vary in space and time in the Japan Trench (Nakamura et al. 2016). However, the b value changes due to the faulting types are only about 0.2 even in very extreme cases (e.g., 100% thrust to 100% normal faulting) (Bürgmann et al. 2016) and realistically would be about 0.1 or less, considering the observed temporal changes in the faulting types prior to the Tohoku-Oki earthquake (Bürgmann et al. 2016). Therefore, it is unlikely that the decade-scale decrease in the b value (~ 0.4) was solely due to the temporal change in the fraction of the faulting types.

In Fig. 10a, we compared the slow earthquake distribution with the spatial distribution of the b value prior to the Tohoku-Oki earthquake estimated by Tormann et al. (2015) (January 1, 2004, to March 11, 2011). Tectonic tremors and VLFEs are predominantly distributed outside the region of a particularly low b value (b $\le$ 0.5) in the central Japan Trench (the purple area in Fig. 10a), implying that the differential stress in the tremor-genic regions of the northern and southern Japan Trench was relatively low before the Tohoku-Oki earthquake. These observations are consistent with the frequent stress release due to slow earthquakes in the shallow parts of the northern and southern Japan Trench (Fig. 6). However, the SSE that occurred during the month prior to the Tohoku-Oki earthquake ruptured the very region where the b value was particularly low (b $\le$ 0.5) (Kato et al. 2012; Ito et al. 2013; a green rectangle within the Tohoku-Oki earthquake rupture in Fig. 10a). This observation implies that the preseismic SSE of the Tohoku-Oki earthquake ruptured the region where the differential stress at the plate boundary was high.

Tanaka (2012) compared the occurrence times of M_w 5 or greater earthquakes in the Japan Trench with tidally induced increases in shear stress. The amplitude of the shear stress perturbation was less than 1 kPa. Tanaka (2012) observed a significant correlation between the tidal stress and earthquake occurrence times from December 2002 to March 2011 in the central Japan Trench (p = 0.34%). The region of high tidal response corresponds well with the region of a particularly low b value (b $\le$ 0.5) (Fig. 10a and b). Tanaka (2012) interpreted the high tidal response in the central Japan Trench to indicate that the source region of the Tohoku-Oki earthquake had been critically stressed prior to the Tohoku-Oki earthquake and was sensitive to the small stress perturbations induced by the tides. Figure 10b presents the comparison of the slow earthquake distribution and the spatial distribution of the tidal response prior to the Tohoku-Oki earthquake (December 2002–March 2011). Similar to the b value (Fig. 10a), tectonic tremors and VLFEs are mainly distributed outside the region of particularly high tidal response in the central Japan Trench. The SSE during the month prior to the Tohoku-Oki earthquake (Kato et al. 2012; Ito et al. 2013) ruptured the very region where the tidal response was particularly high (a green rectangle within the Tohoku-Oki earthquake rupture in Fig. 10b).

In summary, the above comparisons of the b values, tidal response, and slow earthquake distribution imply that the SSE during the month prior to the March 2011 Tohoku-Oki earthquake ruptured the region where the differential stress or shear stress on the plate interface was increased and that seismic slow earthquakes (i.e., tectonic tremors and VLFEs) predominantly occurred outside that region both before and after the Tohoku-Oki earthquake (Nishikawa et al. 2019; Baba et al. 2020).

p value of the Omori–Utsu’s aftershock law

The p value of the Omori–Utsu’s aftershock law (Utsu 1957; Utsu et al. 1995) is the power exponent of the power-law decay of the aftershock occurrence rate; the larger the p value, the faster the aftershock rate decays. Ogata (2011) and Ueda (2021) used the hierarchical space–time ETAS (HIST-ETAS) model, a spatiotemporal statistical model describing earthquake-to-earthquake triggering, to investigate the spatial distribution of the p value of the Omori–Utsu’s aftershock law in the Japan Trench. Figure 10c shows the comparison of the p value distribution with the slow earthquake distribution. The p value distribution was estimated using M_j 5 or greater earthquakes from 1926 to 2008 (Ogata 2011). Figure 10c shows that the p value is high (reddish areas) near the tremor-genic regions in the northern and southern Japan Trench (p $\ge$ 1.3). That is, aftershock rates decay faster near the tremor-genic regions.

It is not straightforward to understand why the p value is high near the tremor-genic regions. Utsu et al. (1995) pointed out that aftershock rates of earthquakes within earthquake swarm activity tend to decay faster (i.e., a higher p value). Therefore, the high p value in the tremor-genic regions may indicate frequent earthquake swarms in these regions. Consistent with this idea, earthquake swarms (M_j $\ge$ 5) have been repeatedly observed in these regions (Kawasaki et al. 1995, 2001; Matsumura 2010; Nishikawa and Ide 2018; Fig. 5h). Furthermore, earthquake swarms in subduction zones are considered to be potential indicators of short-term SSEs (e.g., Marsan et al. 2013; Nishikawa et al. 2019; Sect. 2.3), which are often accompanied by tectonic tremors. Based on the above reasoning, our idea may consistently explain the p value and slow earthquake distributions along the Japan Trench.

Moreover, as mentioned in Sect. 3.2.3, it is known that earthquake swarms cannot be well described within the usual framework of the ETAS model (e.g., Llenos et al. 2009; Okutani and Ide 2011). Therefore, areas that frequently experience earthquake swarms may appear as some kind of anomaly in the ETAS model analysis; the high p values (p $\ge$ 1.3) in Fig. 10c might reflect such anomalies.

Crustal structure and geological environment of the slow-earthquake-genic regions in the Japan Trench

In this section, we present comparisons of the slow earthquake distribution in the Japan Trench with the crustal structure (Sect. 5.1). We also describe the geological environment of the slow-earthquake-genic regions, focusing on water sources, pressure–temperature conditions, and metamorphism (Sect. 5.2). Through these comparisons, we aim to gain insight into the factors governing the slow earthquake distribution in the Japan Trench.

Crustal structure of the slow-earthquake-genic regions

Interplate sedimentary units

Many previous studies have investigated the along-strike changes in the crustal structure of the Japan Trench in detail. Tsuru et al. (2002) investigated the distribution of interplate sedimentary units subducted in the Japan Trench by conducting multichannel seismic reflection surveys (Fig. 11). In the northern Japan Trench, a deformed wedge-shaped sedimentary unit is located updip (from the trench axis to approximately 10 km depth) of the tremor-genic regions (Fig. 11a and b). In contrast, in the southern Japan Trench, a channel-like sedimentary unit with a thickness of approximately 2 km extends to depths deeper than 15 km (Fig. 11a and d). Both the sedimentary units are characterized by low P-wave velocity (2–4 km/s) (see also Azuma et al. 2018). In the central Japan Trench, the shape of the sedimentary unit switches from wedge shape to channel-like shape (Fig. 11a and c), and the interplate sedimentary units of the Japan Trench change their shape on a scale of hundreds of kilometers along the strike direction of the plate interface.

Fig. 11

Comparison between the distribution of interplate sedimentary units and the slow earthquake distribution in the Japan Trench. a Map view of the interplate sedimentary units and slow earthquakes. The distribution of the interplate sedimentary units is from Tsuru et al. (2002). For an explanation of the symbols, refer to the legend in the figure. The slow earthquakes, megathrust earthquakes, and small repeating earthquakes are from Fig. 9. b–d Seismic reflection images of survey line 3, 6, and 14, respectively. The slip behavior of the plate interface is indicated in the images. The seismic reflection images are adapted from Tsuru et al. (2002). Unit U in d indicates the channel-like sedimentary unit

Seismic slow earthquakes (i.e., tectonic tremors and VLFEs) are distributed in the northern and southern Japan Trench (Fig. 11a), with an approximately 200-km-long along-strike gap of seismic slow earthquakes in the central Japan Trench. This along-strike change appears to correspond to the along-strike change in the shape of the interplate sedimentary units (Fig. 11a). The low P-wave velocity sedimentary units are thought to contain large amounts of water (Tsuru et al. 2002). Therefore, the along-strike change in the shape of the fluid-rich sedimentary units may cause an along-strike change in the fluid distribution, which, in turn, might affect the distribution of seismic slow earthquakes (Nakajima and Hasegawa 2016; Delph et al. 2018; Yabe et al. 2021; Sect. 2.4). However, the correspondence between the interplate sedimentary units and seismic slow earthquakes at a smaller spatial scale (tens of kilometers) is not clear. The tremors prefer some small spots on the shallow plate interface (e.g., 38.9° N and 36.6°–36.8° N), but the correspondence between these spots and the shape of the sedimentary units is unclear.

Figure 11b–d shows the locations of slow earthquakes on seismic reflection images obtained by Tsuru et al. (2002). These images provide insight into the crustal structures generating slow earthquakes. On survey line 3 in the northern Japan Trench, tectonic tremors (Nishikawa et al. 2019) and small repeating earthquakes (Uchida and Matsuzawa 2013) have been observed on the plate interface deeper than the deformed wedge-shaped sedimentary unit (Fig. 11b). Previous ocean-bottom seismic observations (Hino et al. 1996, 2000; Obana et al. 2018) also reported the absence of small interplate seismicity where the wedge-shaped sedimentary unit is present. Soft sediments constituting the wedge-shaped unit may suppress the small interplate seismicity (Hino et al. 1996). In contrast to tectonic tremors and small interplate earthquakes, the 1896 M_w 8.0 Meiji Sanriku tsunami earthquake is thought to have ruptured the plate interface beneath the wedge-shaped sedimentary unit (Kanamori 1972; Tanioka and Satake, 1996; Tanioka and Seno 2001; Sect. 4.3; the purple rectangle in Fig. 11a). The soft sediments on the shallow plate interface probably slowed down the rupture process compared to regular earthquakes (Tanioka and Satake, 1996).

On survey line 6 in the central Japan Trench (Fig. 11c), repeating earthquakes have been observed on the plate interface deeper than the deformed wedge-shaped sedimentary unit, similar to survey line 3. The SSE that occurred during the month prior to the Tohoku-Oki earthquake is also thought to have mainly ruptured this plate interface (Kato et al. 2012; Ito et al. 2013; a green rectangle within the Tohoku-Oki earthquake rupture in Fig. 11a). The plate interface beneath the wedge-shaped sedimentary unit on survey line 6 is thought to have slipped during the 2011 Tohoku-Oki earthquake in addition to the 1896 earthquake (e.g., Tanioka and Seno 2001; Ide et al. 2011; Iinuma et al. 2012; Kodaira et al. 2012; Kodaira et al. 2021).

Figure 11d shows the location of the slow earthquakes on the seismic reflection image of survey line 14 in the southern Japan Trench. In this region, the approximately 2-km-thick channel-like sedimentary unit characterized by low P-wave velocity (3–4 km/s) extends to depths greater than 15 km. Shallow SSEs (Nishimura 2021) probably occur within this possibly fluid-rich sedimentary unit (Fig. 11a).

Subducting seamounts and petit-spot volcanoes

Subducting topographic features such as seamounts and petit-spot volcanoes may affect the slip behavior of the Japan Trench megathrust. Petit-spot volcanoes are volcanoes that are thought to be formed by asthenospheric melts ascending along lithospheric fractures in response to plate flexure during subduction (Hirano et al. 2006), distinct from the other types of volcanoes, such as those at divergent plate boundaries (mid-ocean ridges), convergent plate boundaries (e.g., island arcs), and hotspots. For more details on petit-spot volcanism, refer to a review paper by Hirano (2011).

In the southern Japan Trench, the Joban seamount chain is subducting (Fig. 12a). A seismic survey and gravity anomalies revealed a subducted seamount near 36° N, 142° E on the shallow plate interface (Mochizuki et al. 2008; Bassett et al. 2016; the brown dashed circle in Fig. 12a). Although there is debate as to whether seamounts accumulate elastic strain and host megathrust earthquakes (Scholz and Small 1997) or subduct aseismically (Wang and Bilek 2011; Sun et al. 2020), the seamount in the southern Japan Trench is considered to be subducting aseismically (no megathrust earthquakes but many small earthquakes and slow earthquakes) (Mochizuki et al. 2008; Wang and Bilek 2014; Kubo and Nishikawa 2020). Fluid-rich sediments entrained with the seamount (Mochizuki et al. 2008) and the development of a complex fracture network in the upper plate around it (Wang and Bilek 2011, 2014) are thought to suppress large earthquakes and promote aseismic creep and numerous small earthquakes. Consistent with this idea, tectonic tremors and swarms of small interplate earthquakes have been observed in the vicinity of the subducting seamount (Nishikawa and Ide 2018; Nishikawa et al. 2019; Sun et al. 2020; Yamaya et al. 2022; Figs. 9c and 12a). Rupture areas of megathrust earthquakes are located downdip of the subducting seamount (Mochizuki et al. 2008; Kubo and Nishikwa, 2020; red contours in Figs. 9c and 12a). This might be because the enhanced tectonic compression on the downdip side of the seamount causes large fault-normal stress and promotes sediment consolidation to achieve a favorable condition for fast earthquake rupture (Sun et al. 2020).

Fig. 12

Comparison between the crustal structure and the slow earthquake distribution in the Japan Trench. a Residual gravity anomalies (Bassett et al. 2016). The residual gravity anomalies are residuals from the along-strike average of satellite-derived free-air gravity anomalies. b Residual P-wave velocity (Vp) tomography along the upper surface of the subducting Pacific Plate. The residual Vp is the residual from the along-strike average of the original Vp model obtained by the tomographic inversion (Hua et al. 2020). c Plate boundary PP reflection intensity (Mochizuki et al. 2005; Mochizuki 2017). The colored line segments indicate the spatial variations of reflection intensity from the plate interface. For an explanation of the other symbols, refer to the legend in the figure. The slow earthquakes, megathrust earthquakes, and small repeating earthquakes are from Fig. 9

In the northern Japan Trench, petit-spot volcanoes are subducting (Hirano et al. 2006; Fig. 12a). According to Fujie et al. (2020), the petit-spot volcanic area near the trench axis of the northern Japan Trench is characterized by thinner sediment (≤ 160 m) than the other areas along the trench axis (240–400 m) (the blue section of the trench axis in Fig. 12a). This may be because the sediments in the petit-spot volcanic area were disturbed by the petit-spot volcanism and thermally metamorphosed (Fujie et al. 2020).

Fujie et al. (2020) also pointed out that the area where the petit-spot volcanoes are subducting corresponds to the northern limit of the coseismic rupture of the 2011 Tohoku-Oki earthquake (Iinuma et al. 2012; Fig. 12a). Furthermore, we found that tectonic tremors and small repeating earthquakes are densely distributed in this area, whereas they are sparse to the south of 39° N (Fig. 12a). These observations indicate a striking change in the slip behavior of the shallow plate interface near the petit-spot volcanic area in the northern Japan Trench. The frictional properties of the plate interface may have been altered by the disturbance and thermal metamorphism of the subducting sediments due to the petit-spot volcanism (Fujie et al. 2020). This, in turn, might have caused the rupture termination of the Tohoku-Oki earthquake (Fujie et al. 2020) and the clear contrasts in the tectonic tremor and repeating earthquake distributions. Due to the disturbance by the petit-spot volcanism, smectite in a pelagic clay layer of the subducting sediments may have transformed into illite, which has a significantly larger frictional coefficient than smectite (Saffer and Marone 2003).

In addition, horst and graben structures are particularly well developed near the petit-spot volcanic area in the northern Japan Trench (Tanioka et al. 1997). Such subducting relief may also influence the frictional properties and slip behavior of the plate interface because it makes the interplate contact rough (Ruff 1989; Tanioka et al. 1997; Heuret et al. 2012). The rough interplate contact may impede the propagation of coseismic rupture.

Residual gravity anomalies

Bassett et al. (2016) investigated residual gravity anomalies along the whole Japan Trench (Fig. 12a). The residual gravity anomalies are residuals from the along-strike average of satellite-derived free-air gravity anomalies. They found that the residual gravity has a particularly large contrast (approximately 60 mGal) between the central (a high residual gravity anomaly) and southern (a low residual gravity anomaly) Japan Trench. Bassett et al. (2016) called the boundary between the high and low residual gravity anomalies the forearc segment boundary (the thick gray dashed line in Fig. 12a). They suggested that this boundary might be the offshore continuation of the Median Tectonic Line, which is the most prominent tectonic boundary in Japan. The crust of the upper plate in the central Japan Trench might be composed of igneous rocks, whereas that of the upper plate in the southern Japan Trench might be composed of accretionary complexes.

When comparing the forearc segment boundary with the tectonic tremor distribution (Nishikawa et al. 2019), we found a striking difference in the tectonic tremor distribution (small dark green squares in Fig. 12a) across the boundary. Numerous tectonic tremors are distributed in the southern Japan Trench, whereas the central Japan Trench lacks tremors and is characterized by megathrust earthquake ruptures, including the 2011 Tohoku-Oki rupture (e.g., Yamanaka and Kikuchi 2004; Iinuma et al. 2012). The difference in the rocks constituting the upper-plate crust (i.e., igneous rocks and accretionary complexes) might contribute to the contrasting tectonic tremor distribution because upper-plate lithology can modulate the frictional behavior of the plate interface (Bassett et al. 2016). However, the residual gravity anomalies alone cannot explain the slow earthquake distribution along the entire Japan Trench. There is no large contrast in the residual gravity between the shallow parts of the northern and central Japan Trench (Bassett et al. 2016), although we observed a striking contrast in the tectonic tremor distribution (Fig. 12a).

Seismic velocity structure

The seismic velocity structure along the plate interface has been investigated in detail in the Japan Trench (e.g., Zhao et al. 2011; Liu and Zhao 2018; Hua et al. 2020). Hua et al. (2020) used S-net ocean-bottom stations and onshore Hi-net stations to obtain the P-wave velocity structure along the plate interface (Fig. 12b). The Hi-net is a high-sensitivity seismograph network operated by NIED (NIED 2019c). They found that the epicenter of the 2011 Tohoku-Oki earthquake (the black cross in Fig. 12b) is located between a deeper high-velocity anomaly (blue) and a shallower low-velocity anomaly (orange) in the central Japan Trench. The deeper high-velocity anomaly corresponds to the large coseismic slip areas of past large interplate earthquakes (M_w $\ge$ 7) (e.g., Yamanaka and Kikuchi 2004), and this deeper plate interface strongly radiated high-frequency seismic waves during the Tohoku-Oki earthquake (e.g., Ide et al. 2011; Koper et al. 2011). The shallower low-velocity anomaly corresponds to the large coseismic slip (tens of meters) of the Tohoku-Oki earthquake, and this shallower plate interface only weakly radiated high-frequency seismic waves during the Tohoku-Oki earthquake (e.g., Ide et al. 2011; Koper et al. 2011). Hua et al. (2020) suggested that the shallow low-velocity anomaly may reflect low-rigidity materials such as subducted sediments on the plate interface.

In contrast to the above correspondence between the seismic velocity structure and the Tohoku-Oki earthquake rupture, the correspondence between the seismic velocity structure and the slow-earthquake-genic regions is unclear (Fig. 12b). Hua et al. (2020) pointed out that the SSE during the month prior to the March 2011 Tohoku-Oki earthquake (Kato et al. 2012; Ito et al. 2013; a green rectangle within the Tohoku-Oki earthquake rupture in Fig. 12b) occurred in the vicinity of the low-velocity anomaly on the shallow plate interface of the central Japan Trench. Tectonic tremors are also distributed near low-velocity anomalies in the central and southern Japan Trench. However, in the northern Japan Trench, tremors are distributed near a high-velocity anomaly. Therefore, no simple relationship can be found between the seismic velocity structure and the slow earthquake distribution.

Plate boundary reflection intensity

Seismic refraction–reflection surveys have been used to investigate the relationship between the plate boundary PP reflection intensity and the distribution of microearthquakes in the shallow part of the north-central Japan Trench (38–40° N) (Fujie et al. 2002; Mochizuki et al. 2005; Mochizuki 2017). These studies pointed out that microearthquakes avoid the plate interface with strong reflection. Figure 12c shows the spatial variation of the plate boundary reflection intensity revealed by these studies. In Fig. 12c, small repeating earthquakes (red points) (Uchida and Matsuzawa 2013), which are repeating ruptures of the identical small areas on the plate interface, are distributed in the vicinity of regions of weak reflection (the white line segments). The regions of strong reflection (the red line segments in Fig. 12c) correspond with the regions of less active microseismicity (the indigo region) (Mochizuki et al. 2005; Mochizuki 2017). Small repeating earthquakes (red points) also avoid the regions of strong reflection.

In contrast to fast earthquakes, tectonic tremors (small dark green squares) and VLFEs (open yellow squares) occur near the regions of strong reflection (around 38.9° N, 143.4° E) (Fig. 12c). The results of the seismic surveys (Fujie et al. 2002; Mochizuki et al. 2005) suggest that a thin layer (approximately 200 m) with a low P-wave velocity of 3–4 km/s exists at the plate boundary in the area of strong reflection. Tectonic tremors and VLFEs probably occur within this thin low-velocity layer. Fujie et al. (2002) suggested that this low-velocity layer may represent a layer of aqueous fluids and/or rocks with hydrated minerals such as serpentine. Such fluids and rocks may generate tectonic tremors and VLFEs and suppress the occurrence of small repeating earthquakes. In contrast, the short-term SSE that occurred during the month prior to the Tohoku-Oki earthquake (Kato et al. 2012; Ito et al. 2013; a green rectangle in Fig. 12c) is considered to have mainly ruptured the plate interface close to a region where the plate boundary reflection is weak and small repeating earthquakes occur. This observation implies that short-term SSEs can occur outside the possibly fluid-rich layer.

Geological environment of the slow-earthquake-genic regions

In this section, we describe the geological environment of the slow-earthquake-genic regions in the Japan Trench, focusing on their water sources, temperature–pressure conditions, and metamorphism. As described in Sect. 2.4, slow-earthquake-genic regions are characterized by abundant fluids and near-lithostatic pore-fluid pressure (e.g., Kodaira et al. 2004; Shelly et al. 2006; Ujiie et al. 2018; Bürgmann 2018; Behr and Bürgmann 2021; Kirkpatrick et al. 2021). Understanding water sources in the Japan Trench may help identify the factors governing the slow earthquake distribution in the Japan Trench.

Figure 13a and b shows the flux of water released upward from subducted sediments and the oceanic crust in the Japan Trench (Hyndman and Peacock 2003) and the thermal structure of the Japan Trench (Peacock and Wang 1999; Hyndman and Peacock 2003), respectively. Tectonic tremors, VLFEs, and shallow short-term SSEs occur in the shallow part of the Japan Trench (approximately 10–20 km depth). At these depths, the temperature at the plate boundary is estimated to be between 80 °C and 130 °C (Peacock and Wang 1999; Hyndman and Peacock 2003). Under this pressure–temperature condition, subducted sediments provide substantial water. The water sources include sediment compaction, the diagenetic transformation from opal to quartz, and the phase transition from smectite to illite (Hyndman and Peacock 2003; Kimura et al. 2012; Saffer and Tobin 2011; Saffer and Wallace 2015; Katayama 2016; Fig. 13a). The water flux due to the sediment compaction is dominant near the top of the shallow slow-earthquake-genic regions (10 km depth and 80 °C). As the temperature increases, the water flux due to the dehydration reactions becomes significant. At the bottom of the shallow slow-earthquake-genic regions (20 km depth and 130 °C), the phase transition from smectite to illite is the primary water source. Along with the low permeability of marine sediments, the water supplied by these sources may cause near-lithostatic pore-fluid pressure at the plate boundary (Saffer and Tobin 2011; Saffer and Wallace 2015) and enable shallow slow earthquakes to occur.

Fig. 13

Thermal and metamorphic environment of the Japan Trench. a Amount of water released upward from subducted sediments and the oceanic crust as estimated by Hyndman and Peacock (2003). For the details of the fluid flux estimation, see Hyndman and Peacock (2003). The green areas indicate the locations of the slow-earthquake-genic regions. b Thermal structure of the Japan Trench as estimated by Peacock and Wang (1999). This panel is adapted from Hyndman and Peacock (2003). The green sections of the plate interface indicate the locations of the slow-earthquake-genic regions. The question mark indicates the possible deep slow-earthquake-genic region. c Pressure–temperature paths for the top of the subducting oceanic crust in the Japan Trench, Nankai Trough, and Cascadia subduction zone (Peacock and Wang 1999; Peacock 2009). The pressure–temperature ranges of the slow-earthquake-genic regions are shown by the green sections of the pressure–temperature paths. The question mark indicates the possible deep slow-earthquake-genic region in the Japan Trench. Metamorphic conditions of the subducting oceanic crust (Peacock 2009) are also shown. eA, epidote–amphibolite; eB, epidote–blueschist; egA, epidote–garnet–amphibolite; jeB, jadeite–epidote–blueschist; laE, lawsonite–amphibole eclogite, PA; prehnite–actinolite; PP, prehnite–pumpellyite; and Z, zeolite facies

At the deeper plate interface of the Japan Trench (approximately 40–60 km depth), deep short-term SSEs occur in the southern Japan Trench (Nishimura 2021; Fig. 9a). As mentioned in Sect. 3.2.2, systematic and comprehensive detection of short-term SSEs has not yet been conducted in the central and northern Japan Trench; thus, it is unclear whether deep short-term SSEs occur along the entire Japan Trench. However, the decade-long acceleration of aseismic slip prior to the Tohoku-Oki earthquake, which resembles long-term SSEs, occurred on the plate interface at a depth of approximately 30–60 km in the central Japan Trench, possibly accompanied by slips on some areas of the shallower plate interface (10–30 km depth) (Ozawa et al. 2012; Mavrommatis et al. 2014, 2015; Yokota and Koketsu 2015; Sect. 3.2.2). Furthermore, the 3-year periodic slow slip on the deeper plate interface of the northern Japan Trench, which has been inferred from repeating earthquake analyses (Uchida et al. 2016; Khoshmanesh et al. 2020; Sect. 3.2.3; Fig. 7a), is also thought to occur at a similar depth range (approximately 30–60 km depth). Considering these observations, it is possible that certain types of slow earthquakes occur on the deeper plate interface in the central and northern Japan Trench as well as in the southern Japan Trench.

The temperature on the deeper plate interface (40–60 km depth) of the Japan Trench is estimated to be approximately 180–240 °C (Peacock and Wang 1999; Hyndman and Peacock 2003). Pore collapse within basalt constituting the subducting oceanic crust (200–400 °C) can be a water source under this pressure–temperature condition (Hyndman and Peacock 2003; Saffer and Tobin 2011; Katayama 2016; Fig. 13a). The supplied water may be related to the occurrence of deep short-term SSEs (Nishimura 2021) and transient slip phenomena similar to long-term SSEs (e.g., Mavrommatis et al. 2014; Yokota and Koketsu 2015; Uchida et al. 2016; Khoshmanesh et al. 2020; Sects. 3.2.2 and 3.2.3). At the deep plate interface of the Japan Trench, the subducting oceanic crust experiences the blueschist → eclogite facies metamorphism and releases substantial amounts of water (Peacock and Wang 1999). However, this transition occurs on the plate interface at a depth of approximately 110 km (approximately 500 °C) (Peacock and Wang 1999), which is much deeper than the possible deep slow-earthquake-genic region (40–60 km depth).

Figure 13c shows the pressure–temperature ranges of the shallow and deep slow-earthquake-genic regions in the Japan Trench and metamorphic conditions of the subducting oceanic crust. For comparison, the pressure–temperature paths at the plate boundaries of the Nankai Trough and Cascadia subduction zone are also shown (Peacock and Wang 1999; Peacock 2009). The slow-earthquake-genic pressure–temperature ranges greatly vary among the three subduction zones. The shallow part of the Japan Trench (10–20 km depth) has the lowest temperature in Fig. 13c, ranging from 80–130 °C. In this pressure–temperature range, the subducting oceanic crust has not yet experienced a metamorphic facies transition. The possible deep slow-earthquake-earthquake-genic region of the Japan Trench (40–60 km depth and 180–240 °C) corresponds to the jadeite-lawsonite blueschist facies. The highest temperature in Fig. 13c is found in the deeper part of the Cascadia subduction zone (30–50 km depth and 500–600 °C), where the young and warm Juan de Fuca Plate (less than 10 Ma) (Müller et al. 2008) is subducting. In this pressure–temperature range, the subducting oceanic crust experiences the epidote–amphibolite → epidote–garnet–amphibolite → eclogite facies metamorphism. It is surprising that despite the substantial differences in the pressure–temperature and metamorphic conditions, the same types of slow earthquakes (i.e., tectonic tremors, VLFEs, and short-term SSEs) have been observed at the shallow plate boundaries of the Japan Trench and Nankai Trough and at the deep plate boundaries of the Nankai Trough and Cascadia subduction zone (Dragert et al. 2001; Obara 2002; Rogers and Dragert 2003; Obara et al. 2004; Obara and Ito 2005; Ito et al. 2007; Ghosh et al. 2015; Araki et al. 2017; Ohta et al. 2019; Tanaka et al. 2019; Nishikawa et al. 2019; Baba et al. 2020; Nishimura 2021). This suggests that the global slow earthquake distribution is not governed by a single specific temperature, pressure, mineral dehydration, metamorphic facies, or lithology (Peacock 2009; Saffer and Wallace 2015; Kirkpatrick et al. 2021).

Discussion

In this section, by synthesizing the results and implications of slow earthquake studies in the Japan Trench (Sects. 3, 4, and 5), we comprehensively discuss the roles of the slow earthquakes in the occurrence process of the March 11, 2011, M_w 9.0 Tohoku-Oki earthquake (Sect. 6.1), factors governing the slow earthquake distribution in the Japan Trench (Sect. 6.2), and the use of the slow earthquakes for improving the forecasts of interplate seismicity along the Japan Trench (Sect. 6.3).

Overview of the occurrence process of the 2011 Tohoku-Oki earthquake focusing on the roles of slow earthquakes

Prior to the Tohoku-Oki earthquake

In the interseismic period before the 2011 Tohoku-Oki earthquake, slow earthquake activity on the shallow plate interface of the Japan Trench (10–20 km depth) probably varied substantially in the along-strike direction. As suggested by the VLFE activity from January 2003 (Baba et al. 2020; Sect. 3.2.2; Fig. 6), the slow earthquake activity may have been particularly active in the northern and southern Japan Trench. Note that, however, along-strike variations in moment release due to the shallow slow earthquakes are unclear because the spatiotemporal distribution of SSEs (i.e., slow earthquakes with the largest seismic moments) along the entire Japan Trench is unknown (Nishimura 2021). The interplate coupling ratio before the Tohoku-Oki earthquake was high in the central Japan Trench and relatively low in the northern and southern Japan Trench (Ito et al. 2000; Nishimura et al. 2000, 2004; Suwa et al. 2006; Hashimoto et al. 2009; Loveless and Meade 2010; Uchida and Matsuzawa 2011; Lindsey et al. 2021; Fig. 6a), and this coupling distribution corresponds well to the variation of the shallow slow earthquake activity. However, these observations do not mean that slow earthquakes did not occur in the central Japan Trench before the Tohoku-Oki earthquake. VLFEs repeatedly occurred in a small spot around 38.9° N, 143.4° E (Matsuzawa et al. 2015; Baba et al. 2020; Sect. 5.1.5; Figs. 6a and 12c), which corresponds to the northern edge of the huge locked zone in the central Japan Trench (Suwa et al. 2006; Fig. 6a) and the region of strong plate boundary reflection (Fujie et al. 2002; Mochizuki et al. 2005; Mochizuki 2017; Sect. 5.1.5; Fig. 12c).

Large interplate earthquakes (M_w $\ge$ 7) before the Tohoku-Oki earthquake avoided the shallow slow-earthquake-genic regions (10–20 km depth) (Sect. 4.2; Fig. 9). Most of them ruptured the plate interface deeper than the shallow slow-earthquake-genic regions (e.g., Yamanaka and Kikuchi 2004; Mochizuki et al. 2008; Research Center for Seismology, Volcanology, and Disaster Mitigation 2008), although the 1896 Meiji Sanriku tsunami earthquake ruptured the shallow plate interface close to and slightly shallower than the slow-earthquake-genic region in the northern Japan Trench (Tanioka and Satake, 1996; Tanioka and Seno 2001; Sect. 4.3; Figs. 9 and 14). Furthermore, in the southern Japan Trench, large interplate earthquakes (Mochizuki et al. 2008; Research Center for Seismology, Volcanology, and Disaster Mitigation 2008) avoided the source depth of deep short-term SSEs (40 km depth or deeper) as well as that of shallow short-term SSEs (10–30 km depth) (Nishimura 2021; Figs. 9 and 14).

Fig. 14

Schematic of the interplate slip behavior of the Japan Trench. For an explanation of the symbols, refer to the legend in the figure

The clear complementarity between the slow-earthquake-genic regions and the rupture areas of the large interplate earthquakes indicates that the coseismic ruptures did not propagate deep into the slow-earthquake-genic regions. Based on this observation, we suggest that the slow-earthquake-genic regions impeded the coseismic rupture propagation. According to linear elastic fracture mechanics, coseismic rupture terminates when the strain energy released by unit rupture advance falls below the fracture energy consumed (e.g., Ke et al. 2018). Based on this idea, a simple interpretation is that during the coseismic rupture propagation in the slow-earthquake-genic regions, the fracture energy consumed by unit rupture propagation might have exceeded the strain energy released. The large fracture energy in the slow-earthquake-genic regions might have been caused by the mechanisms suppressing fault slip rates considered in the physical models of slow earthquakes (Sect. 2.4). Alternatively, it is also possible that the coseismic rupture propagation terminated due to the small strain energy released by unit rupture propagation in the slow-earthquake-genic regions, which may have resulted from a small stress drop. However, we are aware that linear elastic fracture mechanics may be too simplistic to describe the rupture propagation in slow-earthquake-genic regions because fault zone deformation in these regions is thought to involve not only brittle fractures but also viscous flow (e.g., Ando et al. 2012; Behr et al. 2018; Ujiie et al. 2018), which is not within the framework of elastic fracture mechanics.

Slow earthquakes were involved not only in the rupture termination process but also in the rupture initiation process of the large interplate earthquakes before the 2011 Tohoku-Oki earthquake. SSEs triggering swarms of foreshocks are thought to have occurred during the weeks prior to the 1982 and 2008 M_j 7 Ibaraki-Oki earthquakes in the southern part of the Japan Trench (Matsumura 2010; Nishikawa and Ide 2018; Kubo and Nishikawa 2020; Hirose et al. 2021; Sect. 3.2.3; Fig. 14). Moreover, the 3-year periodic slow slip in the northern Japan Trench (Uchida et al. 2016) might have triggered foreshock activity of M_j $\ge$ 6 interplate earthquakes (Hirose et al. 2021; Sect. 3.2.3).

In the decade prior to the March 2011 M_w 9.0 Tohoku-Oki earthquake, various changes occurred on the plate interface of the Japan Trench. From around 2002 until the 2011 Tohoku-Oki earthquake, an acceleration of aseismic slip similar to a long-term SSE occurred on the deeper plate interface (approximately 30–60 km depth) in the central Japan Trench (Ozawa et al. 2012; Mavrommatis et al. 2014, 2015; Yokota and Koketsu 2015; Sect. 3.2.2; Fig. 6a and b). Furthermore, some areas of the shallow plate interface (10–30 km depth) may also have slipped during this transient (Mavrommatis et al.