The Seismicity of Indonesia and Tectonic Implications

Indonesian seismicity provides important insights into the tectonics and hazards of a region that is, characterized by a remarkable diversity in faulting, including subduction, extension, thrusting, and strike‐slip faulting. We present a synthesis of Indonesian seismotectonics by documenting the distributions of hypocenters (≥M 4.6) and focal mechanisms (≥M 5.0) over ∼20 years, quantifying seismicity rates, and comparing observed seismicity trends with recent tectonic and tomography models. Of the 20,622 events ≥M 4.6 observed in the study region, ∼77% are shallow (≤70 km depth) and of magnitudes


Sources of Uncertainty
The following sources of uncertainty in our analysis have been identified.
1. USGS hypocenters. Our completeness estimate of the USGS catalog in Indonesia provides confidence that hypocenters as low as M 4.6 are well located, though it is important to note discrepancies between earthquake catalogs. As hypocenter locations are a key component of this study, we estimate location uncertainty by comparing a subset of 450 USGS locations of varying magnitudes ≥M 4.6 and depths with equivalent locations in the International Seismological Centre (ISC) On-line Bulletin (International Seismological Centre, 2021). Figure S1 shows an example of compared locations in map view. From this comparison, latitude and longitude variances between the catalogs are generally small, varying on average between ±0.09°-0.14° (∼5-13 km) and ±0.05°-0.07° (∼5-8 km) respectively (Figure S1). Hypocentral depths for shallow, intermediate and deep events vary on average by ±12, ±16, and ±9 km respectively ( Figure S1). Deep events exhibit the lowest depth variances. Conversely, intermediate events exhibit the largest depth variances (∼22% of tested events exhibit depth differences >20 km, with a maximum variance of 165 km). These variances reflect event relocations conducted by ISC (Storchak et al., 2020). Therefore, some events in the USGS catalog may have poorly resolved depths. Shallow events that lack local data and/or depth phases may be assigned default depths of 10 and 33 km by USGS analysts. We acknowledge this will introduce artifacts in cross-sections and seismicity rate-depth histograms at these depths, though Figure S2 shows that for cross-sections the two catalogs produce comparable results. 2. Cross-sections. In addition to hypocenter location uncertainty, the chosen 100 km-wide swath of hypocenters projected on section lines may introduce some minor bias to the cross-sections, especially if there is enough curvature or along-strike changes in slab orientation within the 100 km area. To avoid this, efforts were made to draw cross-sections orthogonal to the subduction zone strike. 3. GCMT focal mechanisms. GCMT source mechanism solutions have been shown recently to be consistent and robust for events ≥M 5.0, with significant improvements added to the CMT method since 2004 that include increased station coverage, consideration of intermediate-period surface waves and addressing heterogeneity in earth models (Ekström, 2015;Ekström et al., 2012). Though the majority of events used in our analysis occurred during or after implementation of these improvements (post 2004), we do use prior events (2000)(2001)(2002)(2003)(2004) that may therefore be less accurate (Ekström et al., 2012). In addition, fewer events are available between 2000 and 2004, as the completeness of the GCMT catalog prior to 2004 is estimated at ∼M 5.3-5.4 (Ekström et al., 2012). Despite this, we are confident in the accuracy of GCMT source mechanisms for use in our analysis. Additionally, there will be location discrepancies noticed between USGS hypocenters and GCMT solutions. GCMT solutions plot the location of the centroid, or location of max stress release during an earthquake (Dziewonski et al., 1981;Ekström et al., 2012), and often do not correspond exactly with the location of the hypocenter (point of earthquake initiation). In comparing the locations of 300 equivalent events ≥M 5.0 between the USGS and GCMT catalogs, we find that latitude and longitude vary on average by ±0.08°-0.16° (9-18 km) and ±0.1°-0.13° (11-14 km) respectively, and depth locations for shallow, intermediate and deep events vary on average by ±9, ±12, and ±9 km respectively. These variances may make it difficult to associate some events with particular faults. 4. Fault and plate boundary locations. The locations of some faults and plate boundaries in Indonesia are greatly debated, and so differing interpretations are common. The faults and plate boundaries plotted on the maps used in this study are obtained from several published studies, and these provide the tectonic context of the seismicity. Here, we assume the most up-to-date published placement of the faults are correct and we do not attempt to reassign fault locations.

General Trends
Indonesia's dense seismic activity is primarily concentrated at or near the plate boundaries, with events increasing in depth inboard of major subduction zones (Figures 1 and 2a). We note that earthquakes may also occur above the megathrust on splay faults and back-thrust faults, as documented in the forearc of Sumatra and Java (Kopp, 2011;Singh et al., 2010). Figure 2a shows the locations of the 20,622 events ≥M 4.6 that occurred in the region over the 20.6-year time period. Hypocenter depths range from ≤10 to 678 km. The majority of events occur at depths ≤70 km (Table 1) and are of magnitudes < M 5.0 (Table 2), comprising  77% and 68% of the total events respectively. Events ≥M 7.0 are not uncommon ( Figure 2b), with 61 events occurring since January 2000, the largest being the 2004 M w 9.1 Great Sumatra-Andaman Earthquake (Tables 2 and 3; Hayes et al., 2017;Lay et al., 2005;U. S. Geological Survey, 2020). Table 3 lists notable events ≥M 7.0 shown in Figure 2b. The Gutenberg-Richter distribution of the data ( Figure 3a) exhibits a b-value of 1.2 when M c = 4.6, implying a greater frequency of low-magnitude events than expected regionally (perfect distribution exhibits a b-value of 1). Events >M 6.0 deviate from this b-value, likely indicating the separation in distributions between larger main shocks and smaller aftershocks (Godano et al., 2014). The b-value assigned here represents the entire study region, and may not represent individual subregions accurately, which would require further analysis beyond the scope of this study.

Seismicity Rates
Seismic activity varies greatly with location and depth. For example, earthquakes approaching the lower mantle (∼660 km) are abundant in eastern Indonesia but are not present in western Indonesia ( Figure 2a). These trends in seismic activity can be more closely observed by quantifying seismicity rates with depth. Seismicity rates provide insight into the most seismically active regions and how activity changes with depth and are compared between the entire study region ( Figure 3b) and seven individual subregions ( Figure 4). Subregions are divided into west and east Indonesian subregions (Figures 4b and 4c). All calculated seismicity rates are presented in Table 4.

Regional
The average yearly rate of events ≥M 5.0 in the study region is approximately 318 EQ/y. Seismicity is most frequent at depths ≤40 km (Figure 3b), with a rate of ∼54 EQ/y between 0 and 10 km, ∼26 EQ/y between 10 and 20 km, ∼54 EQ/y between 20 and 30 km, and ∼72 EQ/y between 30 and 40 km. Seismicity rates between 50 and 300 km decay approximately exponentially from ∼22 to <1 EQ/y following the equation log 10 (y) = log 10 (42) − h/125 (coefficient of determination from regression, r 2 = 0.96), where y is the rate in EQ/y and h is the depth in km (Figure 3b). This result is consistent with global seismicity rate-depth distributions presented by Frohlich (2006) and Houston (2015), who show that seismicity worldwide decays exponentially between 50 and 300 km proportional to log 10 (y) ∼ −h/120. Seismic gaps are present between 330 and 340 km, 410-420 km, 450-460 km and 500-510 km ( Figure 3b). These gaps are equal to the depth resolution (10-km bins) and are likely artifacts of the time period used in this study. Expanding the USGS catalog back 50 years to 1970 eliminates these gaps ( Figure S3). A small peak in seismicity is present near 410 km (∼5 events per 20.6 years). Seismicity increases within the Mantle Transition Zone (MTZ), with the most prominent peaks ∼1 EQ/y located between 510-560 km, 570-580 km, and 590-620 km. Seismic activity decreases near 660 km, and ceases between 670 and 680 km (Figure 3b). These trends remain consistent with global seismicity rates distributed with depth (Billen, 2020;Frohlich, 1989Frohlich, , 2006Houston, 2015;Zhan, 2020).

Subregions
We divide the study area into seven approximately equal-area subregions (Figure 4a) based on the regional distribution of hypocenters (Figure 2a), subduction, and faulting ( Figure 1) in order to investigate the diversity of seismicity-depth patterns. Figure 4b presents seismicity rates for western Indonesian subregions including NW Sumatra, SE Sumatra, and Java. Figure 4c presents seismicity rates for eastern Indonesian subregions including the Banda Sea-Timor, Sulawesi-Seram, Celebes Sea-Philippines, and West Papua. Seismicity rates for subregions with USGS locations extending back to 1970 are presented in Figure S4 for reference. Seismicity rates for all subregions are similar to the regional average ( Figure 3b) at depths ≤40 km, with peaks between 0 and 10 km, abrupt decreases between 10 and 20 km, and peaks between 20 and 40 km (Figures 4b and 4c). However, there is strong deviation between subregions at depths exceeding 40 km (Figures 4b and 4c).   NW Sumatra exhibits the highest seismicity rate of ∼17 EQ/y between 20 and 30 km depth. Many of these earthquakes may be aftershocks of the 2004 Sumatra-Andaman event. SE Sumatra exhibits the most frequent seismicity between 30 and 40 km, while Java exhibits the most frequent seismicity between 0 and 10 km ( Figure 4b). All three western subregions in Figure 4b exhibit a similar decrease in seismicity between ∼40 and 120 km, deviating below 120 km. Seismic activity in NW Sumatra is sparse deeper than 120 km, with no activity beyond 230 km depth (Figure 4b). In contrast, seismicity in SE Sumatra extends to greater depths but is sparse, with large seismic gaps between 180-230 km and 250-580 km (Figure 4b).
Deeper events become increasingly more frequent in Java, with seismicity extending to 640 km. Intermittent peaks within the MTZ in Java do not appear to coincide with the 410, 520, and 660 km discontinuities ( Figure 4b). All eastern Indonesian subregions ( Figure 4c) show the most frequent seismicity between 30 and 40 km, the greatest rate of 14 EQ/y being exhibited by the Celebes Sea-Philippines. As in western Indonesia, eastern subregions deviate with increasing depth. West Papua deviates strongest from the other subregions, rapidly decreasing in seismicity deeper than 40 km with no seismicity beyond 150 km, similar to NW Sumatra (Figures 4b and 4c). The Celebes Sea-Philippines and Sulawesi-Seram regions exhibit a similar decrease between 40 and 190 km, deviating deeper than 200 km ( Figure 4c). In contrast, the Banda Sea-Timor region decreases in seismicity rapidly between 40 and 60 km and exhibits a broad seismicity peak of ∼1-2 EQ/y between 100 and 190 km ( Figure 4c). This broad peak is not present in any other subregion. Deep seismicity, particularly within the MTZ is most frequent in the Celebes Sea-Philippines, Banda Sea-Timor, and Sulawesi-Seram regions. The Banda Sea-Timor region exhibits a peak at 410 km identical to the regional average ( Figure 4c). Seismicity in the Banda Sea-Timor and Celebes Sea-Philippines regions comprises the majority of seismicity between 500 and 680 km, extending to between 660-670 km and 670-680 km respectively.
The most significant observations regarding seismicity rates versus depth can be summarized as follows: (a) Seismicity is most frequent at depths ≤40 km, which is consistent with global seismicity rate-depth distributions obtained by Frohlich (1989Frohlich ( , 2006. However, peaks in seismicity at 0-10 km and 30-40 km may reflect poorly resolved locations, with default depths of 10 and 33 km. Despite this, it is generally known that the majority of seismicity worldwide resides within the brittle crust (Zhan, 2020), which in Indonesia can be as thick as ∼42 km (Ali & Suardi, 2018), and seismicity rates presented here are consistent with this notion; (b) The regional seismicity rate decreases exponentially with depth between 50 and 300 km at a rate consistent with the global distribution (Frohlich, 2006;Houston, 2015). A probable mechanism that governs this decrease is metamorphic dehydration within subducting slabs due to increasing temperature and pressure with depth prior to entering the MTZ (Zhan, 2020); (c) Seismic gaps are observed for all subregions at depths ≥70 km, implying regional variability in the conditions necessary for  (Table 4). Events are most frequent between 30 and 40 km depth, with frequency decreasing exponentially between 50 and 300 km. Seismicity increases again within the Mantle Transition Zone (MTZ). Depth bins (x-axis) are 10 km wide, and the y-axis is log-scaled. Locations of the 410, 520, and 660 km discontinuities in the MTZ are indicated.
deep earthquakes to occur, such as ambient mantle temperature (Houston, 2015); (d) The deepest events in Indonesia, ≥670 km depth, only occur in the Banda Sea-Timor and Celebes Sea-Philippines regions; (e) Seismicity rates deeper than 300 km decrease westward until no seismicity is evident beneath NW Sumatra, with the exception of West Papua; (f) Elevated seismicity between 100 and 200 km is only present in the Banda Sea-Timor region, implying a distinct process or boundary not evident in other subregions; And (g) the most frequent MTZ seismicity is located in the Banda Sea-Timor and Celebes Sea-Philippines regions, where seismicity peaks at 410 km and variably between 520 and 660 km imply that the 410, 520, and 660 km discontinuities may play a significant role in increasing upper mantle seismicity. While there are several factors that contribute to intraslab seismicity, phase transitions at these boundaries increase viscous resistance of the mantle to the motion of subducting slabs (Billen, 2020;Zhan, 2020). Increased mantle resistance contributes to increased strain rates and therefore elevated intraslab seismicity as slabs pass through the MTZ (Billen, 2020;Zhan, 2020). The distribution of intermediate and deep earthquakes in Indonesia will be explored in greater detail in the following section.

Cross-Sections of Intermediate and Deep Seismicity
Intermediate seismicity is defined as events occurring between 70 and 300 km depth, with deep events ≥300 km depth, together forming inclined Wadati-Benioff zones that reflect deformation within subducting slabs (Frohlich, 2006;Houston, 2015;Zhan, 2020). Wadati-Benioff zones can extend to the upper boundary of the lower mantle, typically terminating near 660 km at the deepest, with some special cases extending tens of km deeper into the lower mantle (e.g., Tonga-Kermadec subduction; Billen, 2020;Frohlich, 1989). Intermediate and deep earthquakes account for approximately 20% and 3% of recent Indonesian seismicity respectively (Table 1). Figure 5a displays the distribution of intermediate and deep earthquakes throughout Indonesia at depths ≥70 km. As observed with seismicity rates previously, the majority of intermediate and deep seismicity occurs in eastern Indonesia (Figures 4b and 4c), decreasing in abundance west of 120°E, with deep events ceasing north of ∼1°S in Sumatra (∼100°E; Figure 5a). To greater analyze these trends and their associations with Indonesian tectonics, we plot 2D cross-sections and 3D profiles for (a) the Molucca Sea, Celebes Sea, and southern Philippines (Cross-sections AA' and BB'); (b) Sulawesi island (Cross-section CC'); (c) West Papua (Cross-sections DD' and EE'); (d) the Banda Sea (Cross-section FF'); (e) Java (Cross-sections GG', HH', and II'); and (6) Sumatra (Cross-sections JJ', KK', and LL'). Cross-section locations are presented in Figure 5b. Versions of these cross-sections with hypocenters extending back to 1970 from the USGS catalog are presented for reference in the Supporting Information S1 to this study ( Figures S5 and S6).  (Figure 6a; Cardwell et al., 1980;Fan & Zhao, 2018;Hall, 1987;Hall & Spakman, 2015;Moore & Silver, 1983;Silver & Moore, 1978;Widiwijayanti et al., 2004).

Molucca Sea-Celebes Sea-Philippines
The Molucca Sea plate is comprised of two opposite-dipping slabs, the Sangihe slab dipping beneath the Sunda plate in the west, and the Halmahera slab dipping beneath the Philippine Sea plate in the east (Figures 6a and 6b;Bird, 2003;Hall & Spakman, 2015). The Sangihe slab exhibits a greater along-strike length than the Halmahera slab, trending S-N from Sulawesi to Mindanao while the Halmahera slab is constrained primarily beneath northern Halmahera island (Figures 6a and 6b). Figure 6c shows the Wadati-Benioff zones of the Sangihe, Philippine Sea and Cotabato slabs beneath Mindanao, where arc-arc collision is essentially complete (∼7°N; Hayes, 2018;Widiwijayanti et al., 2004). The Sangihe slab at this location is largely aseismic, exhibiting a seismic gap between ∼200 and 540 km, with dense seismicity between ∼540 and 680 km (Figures 6b and 6c; Fan & Zhao, 2018;Hall & Spakman, 2015). The seismic gap remains present with data extending back to 1970 but shrinks to between ∼250 and 500 km ( Figure S5a). The cause of this seismic gap is not clear from available literature. Tomography models (e.g., Chen et al., 2020;Fan & Zhao, 2018) tend to show a continuous slab near 6°N, and so one possibility is that the slab is aseismic due to ambient mantle temperatures or neutral internal stress in the slab. Another possibility is slab tearing and detachment, which typically follows a cease in subduction (Wortel & Spakman, 2000). Subduction and collision are essentially complete north of 7°N (Widiwijayanti et al., 2004), so this is a logical assumption, but further study is needed to determine the exact cause(s).
Farther south (∼3°N), where arc-arc collision remains active (Widiwijayanti et al., 2004), Figure 6d shows the Wadati-Benioff zone of the Halmahera slab dipping steeply eastward, with the Sangihe slab exhibiting a moderate dip compared to Figure 6c. These orientations are consistent with recent tomography modeling and the orientation of the Molucca Sea plate (Fan & Zhao, 2018;Hall & Spakman, 2015). The Halmahera slab is estimated to have initiated subduction ∼10 Ma (Hall & Spakman, 2015). The southern extent of the Philippine Sea slab is present east of the Halmahera slab at this location to ∼100 km depth (Figure 6d; Hayes, 2018). The tomography models of Fan and Zhao (2018) indicate the Philippine Sea slab may extend to ∼300 km depth at this location, exhibiting an overturned-to-the-east dip and possibly interacting with the Halmahera slab, though this remains unclear from seismicity.

Sulawesi
Figure 7a displays the seismicity of Sulawesi, and Figure 7b displays a N-S cross-section through the Northern Sulawesi trench. At the trench, the subduction of Celebes Sea oceanic lithosphere beneath the island is clearly defined (Figures 7a and 7b; Hall & Spakman, 2015;Silver, McCaffrey, & Smith, 1983). Hypocenters gradually increase in depth south of the trench beneath the Gulf of Tomini, defining the Wadati-Benioff zone of the Celebes slab, as referred to by Hall and Spakman (2015) (Figures 7a and 7b). Hypocenters deviate from Slab 2.0 (Hayes, 2018) near ∼150 km depth and extend to ∼300 km depth (Figure 7b). Deeper hypocenters are likely representative of the nearby Sangihe slab (Figures 6a, 6d and 7a). Seismicity indicates subduction of the Celebes slab is deepest in the east and shallows westward, forming a wedge-shape that has been imaged with tomography ( Figure 7a; Hall, 2018;Hall & Spakman, 2015).
South of the Celebes slab is a steeply north-dipping, poorly defined Wadati-Benioff zone seismically active to near 200 km depth, tentatively termed the Sula slab by Hall and Spakman (2015) (Figure 7b). Hall and Spakman (2015) suggest that the Sula slab is a remnant of older subduction at the East Arm of the island that ceased ∼20 Ma, and that the majority of the slab has since detached near 200 km depth and descended aseismically into the mantle. The current steep dip is a likely result of slab interactions with the Celebes and Sangihe slabs, with lithospheric delamination a likely contributor to current seismicity (Hall & Spakman, 2015).
In addition to active subduction at the Northern Sulawesi trench, Bird (2003) suggested that the east-dipping Makassar thrust (sometimes labeled as a trench; Valkaniotis et al., 2018) on Sulawesi's western coast may represent early subduction development due to the presence of hypocenters ∼70 km depth beneath the interior of the island (Figure 7c). Figure 7c exhibits a cross-section through the Makassar thrust, where hypocenters are located beneath the estimated depth of the continental crust in this region (∼35 km ;Hall et al., 2009). However, there is no Wadati-Benioff zone present, and no recent tomography or gravity models confirm or deny the presence of a developing slab (Hall et al., 2009).

West Papua
Figure 8a displays the seismicity of West Papua in map view, and Figures 8b and 8c illustrate two cross sections through the island. The Caroline plate subducts at the New Guinea trench (Tregoning & Gorbatov, 2004) and a smaller developing subduction zone at the Manokwari trough north of the Bird's Head (Figure 8a; Baldwin et al., 2012;Hall, 2014Hall, , 2018Milsom et al., 1992). Hypocenters gradually increase in depth southward beneath the interior of the island, with activity deeper than 100 km not evident west of ∼136°E toward the Bird's Head (Figure 8a). Hypocenters in Figure 8b extend to ∼100 km depth at the Manokwari trough, with a separate Wadati-Benioff zone extending to ∼300 km depth beneath Seram related to subduction beneath the Banda Sea (Hall & Spakman, 2015). Figure 8c shows a poorly defined,  (2018), Valkaniotis et al. (2018), and Irsyam et al. (2020). Relative motion of the Celebes Sea at the Northern Sulawesi trench is obtained from GPS estimates (Gómez et al., 2000;Bock et al., 2003). (b) Cross-section CC' looking west beneath northern Sulawesi. The Celebes slab dips to the south and is clearly defined to ∼300 km. The Sula slab is a likely remnant of older subduction prior to subduction of the Celebes slab. Slab 2.0 predictions (Hayes, 2018) are plotted as thick black lines. Slab 2.0 geometry for the Sula slab is not available (Hayes, 2018), and is therefore estimated using the dashed black line based on Hall and Spakman (2015). Scaling is 1:1. (c) Cross-section XX' through the Makassar thrust, looking north. Some hypocenters clearly lie below the estimated thickness of the continental crust but no Wadati-Benioff zone is visible. Note that this cross-section is supplementary and is not presented in Figure 5b, so as to avoid crowding. The approximate thickness of the continental crust is indicated by the dashed red line (Hall et al., 2009). Scaling is 1:1.
southward-dipping Wadati-Benioff zone to 200 km depth beneath the interior of West Papua. A flat slab orientation is observed in Figure 8c (Hayes, 2018), consistent with Tregoning and Gorbatov (2004) who imaged the subducting Caroline plate dipping southward between 10°-30° and ∼300 km depth. Shallow seismicity observed above the slab in Figure 8c may represent activity on several thrust faults that cross the interior of the island (Baldwin et al., 2012;Watkinson & Hall, 2017).
Slab 2.0 predicts the slab to extend the entire length of the New Guinea trench (Hayes, 2018), but the nature of current subduction remains debated (Baldwin et al., 2012). The westward decrease in seismic activity observed in Figure 8a may be related to buoyant lithosphere from active seafloor spreading at the Ayu trough, and oblique convergence between the Pacific (Caroline and Philippine Sea) and Indo-Australian plates (Figure 8a; Baldwin et al., 2012;Bock et al., 2003;Fujiwara et al., 1995). The age of the Caroline plate decreases westward from 25 Ma near 136°E to 0 Ma at the Ayu trough ( Figure 8a; Seton et al., 2020). Additionally, the Pacific and Indo-Australian plates (Figure 8a) exhibit an approximate convergence angle of ∼70°NE (Hinschberger et al., 2005), producing a significant sinistral lateral Figure 8. (a) Seismicity of West Papua for all events ≥M 4.6 between January 1, 2000 and July 28, 2020. Indonesia is highlighted in green. Intermediate seismicity decreases westward toward the Bird's Head as the age of the lithosphere of the Caroline plate decreases toward the Ayu trough. Convergence between the Indo-Australian and Caroline plates is oblique with a ∼70°NE convergence angle (Hinschberger et al., 2005). Seafloor age contours (Seton et al., 2020) Figure 8a; Watkinson & Hall, 2017). These factors likely impede subduction at the western New Guinea trench and similarly at the Manokwari trough, which in addition to being a young feature resides nearest to active spreading at the Ayu trough (Figure 8a; Hall, 2014;Milsom et al., 1992). The nature of the flat slab subduction observed in Figure 8c however is not clear and poorly documented in available literature, requiring further investigation beyond the scope of this study.

Banda Sea
The seismicity of the Banda Sea is presented in map view in Figure 9a, with cross sections and 3D plots presented in Figures 9b-9d. The Banda Sea exhibits a unique distribution of hypocenters inboard of the Timor and Seram troughs that form the 180° curved collisional boundary between the Indo-Australian and Sunda plates (Figure 9a; Charlton, 2000;Hall & Spakman, 2015;Hinschberger et al., 2005;Spakman & Hall, 2010). Hypocenters in Figures 9b-9d define a highly deformed, curved Wadati-Benioff zone dipping northward at the Timor trough and southward at the Seram trough. Hypocenters highlight a slab with a gently westward-plunging synform that is flat between 600 and 660 km depth (Hayes, 2018), consistent with tomography models (Hall & Spakman, 2015;Spakman & Hall, 2010;Widiyantoro et al., 2011a). A prevailing theory suggests this lithospheric folding is a result of rollback of a single slab termed the Banda slab into a curved Jurassic embayment on the Australian continental margin beginning ∼10-15 Ma (Audley-Charles, 2011; Charlton, 2000;Hall & Spakman, 2015;Spakman & Hall, 2010).
The current nature of subduction within the Banda Sea remains controversial. Many studies indicate that subduction ceased following the arrival of Australian continental crust at the Timor and Seram troughs ∼2 Ma, switching tectonics from active subduction to collision and accretion (Audley-Charles, 2011;Bird, 2003;Hall, 2018;Hinschberger et al., 2005;Patria & Hall, 2017 Hinschberger et al., 2005) indicate that as much as 400 km of continental subduction has occurred based on tomography and geochemical evidence. Spakman and Hall (2010) proposed that lithospheric delamination may be an important mechanism that accommodates the continuing rollback and folding of the Banda slab, which may make it possible for subduction to continue beneath the active collision zone where continental subduction is becoming increasingly difficult (Harris, 2011). In addition, slab tearing beneath Seram and Timor has been proposed from tomography and seismic gaps beneath the islands (Figures 9b-9d; Widiyantoro et al., 2011a), consistent with the final stages of subduction and suggesting that subduction of the Banda slab is in the later stage prior to break-off and descent into the mantle (Hall & Spakman, 2015;Harris, 2011;Widiyantoro et al., 2011a;Wortel & Spakman, 2000). Regardless, the Banda slab remains a prevalent source of deep seismicity beneath the Banda Sea.

Java
The seismicity of Java and the surrounding islands is displayed in map view in Figure 10a, with cross-sections presented in Figures 10b-10d. Subduction of the Indo-Australian plate beneath the Sunda plate at the Sunda-Java trench is clearly visible (Figure 10a). Hypocenters extend to ∼660 km beneath Java, highlighting a steeply north-dipping slab consistent with tomography models (Figures 10a-10d; Hall & Spakman, 2015;Widiyantoro et al., 2011b). At this location, relatively old lithosphere between 100 and 155 Ma (Jacob et al., 2014;Seton et al., 2020) subducts with a ∼7°-15°E convergence angle (Figures 10b and 10c; Hall & Spakman, 2015). Nearly direct convergence and lithospheric age may contribute to the observed steep dip (Hall & Spakman, 2015).
Notable changes in seismicity are observed in cross-sections across Java (Figures 10b-10d). East of Java (Figure 10b) hypocenters form a relatively continuous Wadati-Benioff zone to ∼660 km depth, with shallower seismicity in the upper plate likely related to back-arc thrusting along the Flores thrust (Figure 10a). In contrast, hypocenters beneath eastern Java (Figure 10c) exhibit a prominent seismic gap from approximately 150-600 km depth, with seismicity resuming below 600 km. This gap has been proposed to correspond with a hole in the slab that extends for ∼400 km laterally beneath east Java as imaged by tomography (Dokht et al., 2018;Hall & Spakman, 2015;Widiyantoro et al., 2011b). Beneath west Java (Figure 10d) this gap is no longer visible, with the Wadati-Benioff zone relatively continuous to ∼200 km, and intermittent between 200 and 660 km. In all cross-sections, the majority of seismicity appears to lie below the Slab 2.0 estimate for the slab top (Hayes, 2018), suggesting seismicity is primarily within the slab rather than the overlying plate or at the plate interface (Figures 10b-10d).

Sumatra
The seismicity of Sumatra is displayed in Figure 11a, with cross-sections presented in Figures 11b-11d. North of ∼1°S, hypocenters occur no deeper than ∼250 km (Figures 4b, 11a and 11c; Hayes, 2018). Slab material however has been imaged with seismic tomography to depths of at least 660 km throughout the majority of Sumatra (Hall & Spakman, 2015). Figure 11b shows that in southeastern Sumatra, seismicity traces a relatively continuous and steeply dipping Wadati-Benioff zone active to ∼250 km, with Slab 2.0 predicting the slab extends aseismically to near 600 km (Hayes, 2018). In contrast, Figure 11c in central Sumatra near 1°N exhibits a more shallow-dipping Wadati-Benioff zone to ∼200 km, with Slab 2.0 predictions extending no further than 250 km depth (Hayes, 2018). A similar dip is observed in NW Sumatra near 5°N (Figure 11d), but the Wadati-Benioff zone is poorly defined to ∼200 km. The majority of seismicity in   (Hayes, 2018) in Figures 11b-11d to illustrate slab dip changes. Hall and Spakman (2015) suggest a N-S trending tear in the slab is located between 1° and 5°N (between cross-sections KK' and LL'), which corresponds to slight dip changes between these crosssections. Subducting lithosphere at section LL' is older than at section KK' (Seton et al., 2020). Scaling is 1:1. Figure 11d appears clustered around the predicted shallow plate interface, suggestive of increased megathrust activity following the 2004 M w 9.1 Sumatran-Andaman earthquake (Lay et al., 2005). Additionally, Figure 11e illustrates slight dip changes between Figures 11c and 11d predicted by Slab 2.0 (Hayes, 2018), which are noted and proposed to coincide with a N-S trending tear in the slab located between ∼1° and 5° N by Hall and Spakman (2015).
The observed changes in subduction geometry and lack of deep seismicity moving NW from Java to Sumatra may be due to increasing oblique convergence, decreasing plate velocity, decreasing lithospheric age and increasing mantle temperatures (DeMets et al., 2010;Hall & Spakman, 2015;McCaffrey, 2009;Saita et al., 2002;Widiyantoro & van der Hilst, 1996). The western Sunda-Java trench is a partitioned subduction zone, where oblique convergence is divided between the trench and trench-parallel strike-slip faults in the upper plate such as the Sumatran fault (Figure 11a; McCaffrey, 2009). Convergence angles between the Indo-Australian and Sunda plates are estimated to change from below 20° in southern Sumatra to near 30° in northern Sumatra, producing a significant dextral lateral component (Baroux et al., 2002;Bock et al., 2003;McCaffrey, 2009  In addition, NW Sumatra has been determined to exhibit an abnormally thin MTZ and depressed 410 km phase boundary, indicating the mantle is warmer than average, a likely result of mantle upwelling through slab tears (Kong et al., 2020;Saita et al., 2002). Deep earthquake generation is highly dependent on temperature, and where temperatures are too great slabs will resort to ductile deformation (Houston, 2015). Mantle temperatures increase between Java and Sumatra as deep seismicity decreases, suggesting temperature is a primary control on deep seismicity beneath Sumatra (Figure 11a; Kong et al., 2020;Saita et al., 2002).

Intraslab Mechanisms
Intraslab earthquakes are similar in many respects to shallow crustal events but occur at much higher temperatures and pressures and by different mechanisms including slab flexure, metamorphic dehydration, thermal shear instability, and mineral phase transitions (Billen, 2020;Houston, 2015;Romeo & Álvarez-Gómez, 2018;Zhan, 2020). Figure 12 shows that intraslab focal mechanisms in Indonesia are quite heterogeneous. Thermomechanical models expect events ≥400 km to be down-dip compressional (reverse faulting) and intermediate events to be down-dip tensional (normal faulting; Houston, 2015). Indonesian slabs appear to deviate from this prediction however, with normal faulting mechanisms commonly observed for events ≥300 km and reverse and oblique mechanisms common at intermediate depths (Figure 12). Though unclear, this deviation may be related to a number of factors, including slab age, subduction rates, strain heterogeneity, and inherited faults within the slab (Billen, 2020;Houston, 2015;Wiseman et al., 2012). Intraslab events can reach magnitudes ≥M 7.0, and particularly those that occur at shallower depths can pose serious seismic hazards. One such example is the 2009 M w 7.6 Padang event, which occurred ∼80 km depth (Wiseman et al., 2012).

Shallow Seismicity and Focal Mechanisms
In contrast to intermediate and deep seismicity, shallow seismicity ≤70 km depth is more representative of plate motions and associated deformation nearest to the surface, accounting for the majority of seismicity in Indonesia (∼77%; Table 1; Figure 2a). Shallow events pose a greater seismic hazard than deeper events due to hypocenter locations close to the surface (Table 3). Sources of shallow seismicity in Indonesia are diverse, but for ease of organization, the primary sources are defined as (a) megathrust, (b) crustal, and (c) shallow intraslab.

Megathrusts
Megathrusts define the shallow plate interface (≤50 km depth) between subducting and overriding plates, behaving as large-scale thrust faults that accommodate plate convergence at subduction zones (Bilek & Lay, 2018;Kanamori, 1986). Able to generate the largest earthquakes recorded by modern instrumentation (∼M 9.0), megathrust events are often associated with destructive ground shaking and tsunamis, creating serious seismic hazards (Bilek & Lay, 2018;Polet & Kanamori, 2000). Several subduction zones in Indonesia host active megathrusts, the most prominent being the Sunda megathrust within the Sunda-Java trench (Figures 13 and 14).
The Sunda megathrust is most active offshore of Sumatra (Figure 13 Table 3; Hughes et al., 2010;Hayes et al., 2017). In Java however, megathrust activity appears more subdued, with fewer events than Sumatra ≥M 5.0 supporting a north-dipping thrust interface (Figure 14a), very few events ≥M 6.0 (Figure 14b), and no great earthquakes known to have occurred in at least a century (Newcomb & McCann, 1987;Okal, 2012). The largest recent event was a M w 7.7 SW of Java on July 17, 2006 ( Figure 14b; Table 3; Hayes et al., 2017). Irsyam et al. (2020) indicate that this lack of notable megathrust events implies interplate coupling on the Sunda megathrust is much lower in Java than in Sumatra. However   Table 3). Focal mechanisms are diverse, with normal faulting common for deep earthquakes and oblique mechanisms common for intermediate earthquakes. Megathrusts in eastern Indonesia are not as large or active as the Sunda megathrust but are nonetheless notable sources of seismicity. The Northern Sulawesi trench for example ( Figure 15) produced a destructive M w 7.4 event in November 2008 ( Figure 15; Table 3), the latest of several ≥M 7.0 events since 1990 including a M w 7.8 in 1990, a M w 7.5 and M w 7.0 in 1991, a M w 7.9 and M w 7.0 in 1996, and a M w 7.0 in 1997 (Gómez et al., 2000;Hayes et al., 2017;U.S. Geological Survey, 2020). Though the 2008 M 7.4 earthquake is the only event ≥M 6.0 that has occurred in the trench since January 2000, based on past events a future ≥M 7.0 rupture is likely, with the trench designated by others as a serious seismic hazard (Cipta et al., 2016;Irsyam et al., 2020). Figure 14. Shallow GCMT focal mechanisms ≤70 km for Java between January 1, 2000 and July 28, 2020 for (a) Events ≥M 5.0 and (b) Events ≥M 6.0. Events are colored by mechanism and scaled by magnitude. Thrusting mechanisms are less abundant along the Sunda-Java trench in Java than in Sumatra, likely indicating low coupling on the megathrust (Irsyam et al., 2020). Other thrust mechanisms are located along the Flores back-arc thrust. Normal faulting in the outer-rise of the trench indicates down-dip extension and low megathrust coupling (Christensen & Ruff, 1988). Notable events in panel (b) are presented in Table 3. Faults and boundaries are obtained from Sliver, Reed, et al. (1983), Hall (2012), Koulali et al. (2016), Supendi et al. (2018). White arrows are plate velocity vectors obtained with respect to the Sunda plate from the MORVEL velocity model (DeMets et al., 2010). SU=Sunda plate and IA=Indo-Australian plate.
The western New Guinea trench exhibits less-frequent seismicity than other megathrusts in Indonesia (Figure 16), and the nature of subduction remains debated (Baldwin et al., 2012). No thrust events ≥M 6.0 since January 2000 are evident within the western New Guinea trench (Figure 16b), though some events <M 6.0 do support a shallow southward-dipping thrust interface primarily near 138°E (Figure 16a). Despite exhibiting low seismicity, the New Guinea trench is capable of generating great magnitude events, the most recent being the February 1996 M w 8.2 Biak earthquake (Henry & Das, 2002;Okal, 1999). Another similar Figure 15. Shallow GCMT focal mechanisms ≤70 km depth for Sulawesi between January 1, 2000 and July 28, 2020 for events ≥M 5.0. Events are colored by mechanism and scaled by magnitude. Subduction is active in the North Sulawesi trench, with strike-slip mechanisms supporting E-W sinistral slip along the Palu-Koro and Montano faults. Strike-slip motion is present in the Banggai islands. Other faults do not exhibit notable seismic activity but are likely tectonically active (Socquet et al., 2006;Watkinson and Hall, 2017). Normal faulting is present throughout the Gulf of Tomini, associated with rollback of the Celebes slab (Hall & Spakman, 2015;Watkinson and Hall, 2017). Faults and boundaries are obtained from Silver, McCaffrey, and Smith (1983), Hall (2012), Cipta et al. (2016), Watkinson and Hall (2017), Hall (2018), Nugraha and Hall (2018), Valkaniotis et al. (2018), and Irsyam et al. (2020). White arrows are relative motion velocity vectors of the Celebes Sea obtained from GPS (Gómez et al., 2000;Bock et al., 2003). Notable events are labeled and presented in Table 3. Figure 16. Shallow GCMT focal mechanisms ≤70 km depth for West Papua and Seram between January 1, 2000 and July 28, 2020 for (a) Events ≥M 5.0 and (b) Events ≥M 6.0. Events are colored by mechanism and scaled by magnitude. The western New Guinea trench exhibits little to no seismic activity. Lithospheric age and oblique convergence may contribute to this. Strike-slip mechanisms supporting E-W sinistral shear are evident along the Yapen, Lowland, Terera-Aiduna, and western Sorong faults. Broad sinistral shear is transferred from the Terera-Aiduna fault to the western Sorong fault across Seram (Patria & Hall, 2017;Watkinson & Hall, 2017). Normal faulting is active in the Aru trough, associated with rollback of the Banda slab (Adhitama et al., 2017;Spakman and Hall, 2010). Notable events labeled in panel (b) are presented in Table 3. Faults and boundaries are obtained from Saputra et al. (2014), Adhitama et al. (2017), Patria and Hall (2017), Watkinson and Hall (2017), and Hall (2018). White and dark blue arrows represent plate velocities with respect to the Sunda plate obtained from the MORVEL plate velocity model (DeMets et al., 2010). SU=Sunda plate, CA=Caroline plate, and IA=Indo-Australian plate. event in 1914 is speculated to have occurred near the 1996 rupture (Okal, 1999), suggesting the trench is often seismically locked and prone to infrequent great magnitude ruptures. A possible explanation is that though the trench is active, it may not absorb all of the estimated ∼70 mm/y of convergence (Tregoning & Gorbatov, 2004) between the Indo-Australian and Caroline plates, with fold and thrust belts on the interior of the island providing additional accommodation (Bock et al., 2003;Hinschberger et al., 2005;Tregoning & Gorbatov, 2004). The majority of recent active thrusting activity appears clustered to the west in the Manokwari trough, which exhibits abundant thrust mechanisms ≥M 5.0 supportive of a SSW-dipping thrust interface and southward subduction (Figure 16a; Hall, 2014). Two ≥M 7.0 thrust events occurred near the trough in 2009, a M w 7.7, and a M w 7.4, and seven events M 6.0-7.0 occurred between 2000 and 2009, suggesting the Manokwari trough is an active megathrust and a notable seismic hazard, despite being a developing subduction zone (Figure 16b; Hall, 2014).
Megathrusting within the southern Philippine trench (Figure 17) appears most active near the Philippine island of Mindanao with abundant focal mechanisms <M 6.0 supportive of a west-dipping thrust interface (Figure 17a). Activity is not as abundant farther south near the Indonesian island of Morotai where the trench terminates (Figure 17a; Hall, 2018). Very few events ≥M 6.0 are evident within the trench (Figure 17b), and a great magnitude event has not occurred within or near the trench east of Mindanao since ∼1972 (U.S. Geological Survey, 2020). Few events ≥M 7.0 have been characterized along the trench since ∼1600, with such activity occurring primarily south of ∼9°N (Ye et al., 2012). In 2012, a M w 7.6 thrust event occurred in the outer-rise east of the Philippine trench in a region of low megathrust activity (Figure 17b; Table 3), suggesting strain accumulation in the oceanic crust due to strong coupling on the megathrust Figure 17. Shallow GCMT focal mechanisms ≤70 km depth for the southern Philippines and Celebes Sea region between January 1, 2000 and July 28, 2020 for events (a) ≥M 5.0 and (b) ≥M 6.0. Events are colored by mechanism and scaled by magnitude. Thrusting is active within the Philippine trench but events are <M 6.0. Active megathrusting is prevalent at the Cotabato trench. Strike-slip motion is located primarily along the Philippine fault. Outer-rise normal faulting is similar to Java, with a single reverse event in 2012 suggesting a locked megathrust (Christensen & Ruff, 1988;Ye et al., 2012). Notable events labeled in panel (b) are presented in Table 3. Faults are obtained from Moore and Silver (1983), Widiwijayanti et al. (2004), Besana and Ando (2005), Tsutsumi and Perez (2013), and Hall (2018). White arrows represent plate velocities with respect to the Sunda plate from the MORVEL velocity model (DeMets et al., 2010). SU=Sunda plate, and PS=Philippine Sea plate. (Christensen & Ruff, 1988;Ye et al., 2012). It is not currently understood however if a large megathrust event can be expected in the near future as a result.
Conversely, the Cotabato trench SW of Mindanao has produced several notable megathrust events in the past century. Several shallow focal mechanisms inboard of the trench ≥M 5.0 support an ENE-dipping thrust interface with three events exceeding M 6.0, the largest being a M w 7.5 in 2002 (Figure 17; Hayes et al., 2017). The trench also produced the destructive ∼M s 7.8 Moro Gulf earthquake and tsunami of 1976, with another likely similar event ∼M s 8.0 occurring in 1918, making the Cotabato trench a notable source of seismic hazard in the Celebes Sea region (Stewart & Cohn, 1979).
Megathrust activity may still occur in regions where collision and accretion have recently overprinted subduction (Figures 18 and 19). Bird (2003) indicated that the Molucca Sea Collision Zone is likely capable of megathrusting if an interface with the subducted Molucca Sea plate is present beneath the colliding Sangihe and Halmahera arcs ( Figure 18). Some focal mechanisms ≥M 5.0 suggest thrust faults that dip to the west near Sulawesi and to the east near Halmahera, consistent with the orientations of the Molucca Sea plate (Figures 6 and 18a). However, it is not clear whether these events may represent interplate slip. Similarly, megathrusting may be possible within the Timor and Seram troughs ( Figure 19; Liu & Harris, 2014). Despite often being interpreted as foreland basins that formed during collision rather than as active subduction zones (Audley-Charles, 2011;Patria & Hall, 2017), some authors have argued that the Seram trough in particular is still an active subduction zone (Bird, 2003;Hinschberger et al., 2005). Focal mechanisms ≥M 5.0 are supportive of north-dipping thrusts inboard of the Timor trough and south-dipping thrusts inboard of the Seram trough, consistent with the orientation of subduction at these locations (Figure 19a;Hall & Spakman, 2015;Spakman & Hall, 2010;Widiyantoro et al., 2011a). However, it is not clear whether these events are representative of ruptures at a plate interface, deformation within the overlying continental crust, or within underlying continental crust due to continental subduction (Hinschberger et al., 2005;Tate et al., 2015). Using historical records, Liu and Harris (2014) modeled a megathrust event that likely occurred within the Seram trough in 1629 that resulted in a destructive tsunami. This result indicates that megathrust events are possible beneath Seram but are rare.

Thrust Faulting
Thrust faulting is commonly observed inboard of subduction zones and within regions of active collision. In the Molucca Sea Collision Zone (Figure 18), prominent thrust faults on the margins of the colliding Sangihe and Halmahera volcanic arcs accommodate the majority of E-W convergence (∼80 mm/y) between the Sunda and Philippine Sea plates (Figures 17 and 18; Bock et al., 2003;Rangin et al., 1999). The primary faults are the West Halmahera thrust on the western margin of the Halmahera arc, and the East Sangihe thrust on the eastern margin of the Sangihe arc, which dip toward each other toward the center of the collision zone ( Figure 18; Moore & Silver, 1983). Abundant reverse and thrust faulting mechanisms ≥M 5.0 with NE-SW strikes located between the thrusts reflect deformation between the colliding arcs ( Figure 18a). Five events ≥M 7.0 have occurred in the region since January 2000, the largest being a M w 7.5 rupture in 2007 (Figure 18b; Table 3). Some events have caused damage to Sulawesi and Talaud islands, and particular events such as the 2019 M w 7.1 Molucca Sea earthquake generated small tsunamis, indicating that frequent thrust activity in the Molucca Sea region creates notable seismic hazard for nearby islands including Sulawesi, Talaud, Halmahera, and Sangihe (Table 3; U.S. Geological Survey, 2020).
In the Banda and Flores seas, the Flores and Wetar back-arc thrusts form an approximately 800 km-long south-dipping thrust system that traces offshore north of the islands of Lombok, Sumbawa, Flores, and Wetar (Figures 14 and 19). Bock et al. (2003) show from GPS measurements that the majority of convergence between the Indo-Australian and Sunda plates at this location is accommodated by the thrusts (up to ∼67 mm/y), representing the transfer of convergence into the back-arc region and early subduction polarity reversal (Koulali et al., 2016;Silver, Reed, et al., 1983). Focal mechanisms ≥M 5.0 support south-dipping thrusts (Figures 14a and 19a). Back-arc thrusting activity decreases west of the Flores thrust toward the Kendeng thrust in Java, where two events ≥M 5.0 support a south-dipping thrust (Figure 14a). The majority of the Kendeng thrust appears to be seismically inactive due to a lack of events near the fault (Figures 14a  and 14b), however, Koulali et al. (2016) indicate from GPS data that the fault is active and a likely extension of the Flores thrust. Both the Flores and Wetar thrusts have generated several notable and damaging events recently, such as the 2018 Lombok Sequence which produced two M w 6.9 events north of Lombok (Figure 14b; Table 3; Yang et al., 2020), and a M w 7.5 event near Wetar in 2004 (Figure 19b; Table 3; Hayes et al., 2017;U.S. Geological Survey, 2020). As much of the active thrusting occurs offshore, there is great risk for tsunamis. For example, the 1992 M w 7.9 Flores Earthquake generated a tsunami with wave heights up to 25 m (Yang et al., 2020;Yeh et al., 1993).  The back-arc thrusts may represent subduction polarity reversal (Sliver, Reed, et al., 1983). Strike-slip and reverse events throughout the Banda Sea indicate complex transpressive motion. Notable events labeled in panel (b) are presented in Table 3. Faults are obtained from Silver, Reed, et al. (1983), Roosmawati and Harris (2009) Survey, 2020). Other thrusts such as Sulawesi's Batui, Tolo, and Makassar thrusts exhibit little to no seismicity ≥M 5.0 ( Figure 15; Watkinson & Hall, 2017). The Makassar thrust however has been designated a notable seismic hazard based on earthquake hazard modeling (Cipta et al., 2016;Irsyam et al., 2020).

Strike-Slip Faulting
Oblique convergence in Indonesia produces a considerable amount of lateral motion accommodated by strike-slip fault zones. In western Indonesia, the ∼1,900 km-long, segmented dextral Sumatran fault traces the length of Sumatra and partitions oblique subduction at the Sunda-Java trench (Figure 13a; McCaffrey, 2009;Sieh & Natawidjaja, 2000). Partitioning of dextral slip along this major plate boundary fault has been proposed to form a "sliver plate" in the forearc between the fault and the megathrust that behaves separately from the rest of the Sunda plate (Bradley et al., 2017;McCaffrey, 2009). The nature of this sliver plate is debated, with some authors proposing that increasing slip rates along the Sumatran fault moving north inferred by GPS imply the forearc is stretching (Bock et al., 2003;McCaffrey, 2009). More recent modeling by Bradley et al. (2017) however suggests that the forearc is rigid, and the Sumatran fault accommodates a constant slip rate ∼15-16 mm/y. The Sumatran fault exhibits notable seismic activity, with five events between M 6.0 and 7.0 confirmed to have occurred on the fault in the past two decades, the largest being a M w 6.6 event in 2009 (Figure 13b; Salman et al., 2020;U.S. Geological Survey, 2020). Some of these events were damaging (Salman et al., 2020). The fault zone is estimated to be capable of ∼M 7.5-7.7 ruptures (McCloskey et al., 2005), with a 1995 M 7.0 rupture in southern Sumatra being the most recent major earthquake (McCaffrey, 2009;Omang et al., 2016;Sieh & Natawidjaja, 2000). Seismicity may be induced on the Sumatran fault directly by great megathrust events generated at the Sunda megathrust (McCloskey et al., 2005), and related faults that lie outside of the Sumatran fault zone pose additional hazards, such as the fault responsible for the damaging 2016 M w 6.6 Aceh earthquake (Figure 13b; Salman et al., 2020) Offshore of Sumatra, a series of N-S trending strike-slip faults within the oceanic crust of the Indo-Australian plate further add to seismicity (Figure 13a). Abundant strike-slip mechanisms ≥M 5.0 west of Sumatra between 0° and 5°N support activity on these faults (Figure 13a), which likely represent reactivated fracture systems formed during now extinct seafloor spreading in the Wharton Basin (Indian Ocean; Satriano et al., 2012). Activity on these faults may be indicative of diffuse deformation between the Indian and Australian plates, referred to as a single plate (Indo-Australian plate) throughout this study (Bradley et al., 2017). While reasons for this deformation are debated, Bradley et al. (2017) indicate that for their rigid Sumatran forearc model to be plausible, the majority of deformation must be accommodated within the oceanic crust of the Indo-Australian plate. These remarkable faults generated the largest strike-slip events ever recorded in 2012, a M w 8.6 and a M w 8.2 ( Figure 13b; Table 3), which were likely induced by Coulomb stress changes following the December 2004 M w 9.1 Sumatra-Andaman and March 2005 M w 8.6 Nias events (Delescluse et al., 2012).
Eastern Indonesia is decidedly more complex, with several strike-slip fault zones accommodating E-W sinistral shear between the obliquely converging Pacific (Philippine Sea and Caroline plates) and Indo-Australian plates ( Figure 16). The Sorong fault for example trends E-W from the Bird's Head to near eastern Sulawesi (Figures 1 and 16), splitting into several "horsetail" splays west of the Bird's Head (Figures 1 and 16; Saputra et al., 2014;Watkinson & Hall, 2017;Watkinson et al., 2011). A lack of strike-slip focal mechanisms ≥M 5.0 near the Sorong fault at the Bird's Head suggests the fault is not very seismically active at this location ( Figure 16a; Watkinson & Hall, 2017). Increased strike-slip activity ≥M 5.0 is evident to the west near Halmahera island (Figure 16a), where the fault on average is estimated to accommodate ∼19 mm/y of slip (Bock et al., 2003;Watkinson & Hall, 2017). The western splays of the Sorong fault have generated notable events, such as 2019 M w 7.2 earthquake south of Halmahera (Figure 16b; Table 3; U.S. Geological Survey, 2020).
Strike-slip mechanisms ≥M 5.0 supportive of E-W sinistral shear are also evident east of the Bird's Head, where the Sorong fault connects to the Yapen fault ( Figure 16; Hinschberger et al., 2005;Patria & Hall, 2017;Watkinson & Hall, 2017). The Yapen fault is estimated to accommodate ∼46 mm/y of slip, with the majority of seismicity ≥M 6.0 clustered near its connection with the Sorong fault where a destructive M w 7.6 earthquake occurred in 2002 (Figure 16b; Table 3; Hayes et al., 2017). Strike-slip events ≥M 5.0 also support E-W sinistral motion south of the Yapen fault to the E-W trending Terera-Aiduna and NE-SW trending Lowland faults (Figure 16; Watkinson & Hall, 2017). This supports the hypothesis that sinistral shear on a broad scale is transferred south from the Yapen fault to the Terera-Aiduna fault, effectively bypassing the Sorong fault at the Bird's Head (Bock et al., 2003;Hinschberger et al., 2005;Watkinson & Hall, 2017). The Terera-Aiduna and Lowland faults have produced several events ≥M 6.0 recently, the largest being the 2004 M w 7.3 Nabire earthquake (Figure 16b; Table 3; Hayes et al., 2017). Strike-slip seismicity extends westward beyond the Terera-Aiduna fault and into the Banda Sea, where sinistral shear is likely transferred to the western Sorong fault via transpressive faults across Seram (Patria & Hall, 2017;Watkinson & Hall, 2017). This proposed transfer of sinistral shear from the Terera-Aiduna fault to faults on Seram is not distinguishable from seismicity alone, and Watkinson and Hall (2017) suggest any connection may be aseismic or there is no structural connection at all.
Notable strike-slip faults on Seram include the sinistral Kawa and Bobol faults (Figure 16; Patria & Hall, 2017;Watkinson & Hall, 2017). Few strike-slip focal mechanisms ≥M 5.0 are evident along these faults, though geomorphic expressions interpreted by Watkinson and Hall (2017) indicate the Kawa fault is capable of ground-rupturing events, as are other faults on the island, such as the recently reactivated fault responsible for the destructive 2019 M w 6.5 Ambon earthquake (Figure 16b; Table 3; Sahara et al., 2021). Mixed strike-slip and reverse mechanisms ≥M 5.0 between Seram island and the western Sorong fault suggest the transfer of sinistral shear between them is very complex. Additionally, strike-slip mechanisms ≥M 5.0 in the Banda Sea between Seram and Wetar to the south suggest that lateral motion may be partitioned by faults throughout the interior of the Banda Sea, possibly along remnants of older seafloor spreading, though this is speculative (Figures 16 and 19; Bock et al., 2003;Hinschberger et al., 2005;Seton et al., 2020).
Farther west on the island of Sulawesi, sinistral-shear is accommodated primarily by the Montano and Palu-Koro faults (Figure 15; Silver, McCaffrey, & Smith, 1983;Socquet et al., 2006;Watkinson & Hall, 2017). The Montano fault is a highly segmented SE-NW trending fault system that extends offshore to the east toward the Tolo thrust, and inland to connect with the southern terminus of the Palu-Koro fault though the nature of this connection is not well understood (Figure 15; Silver, McCaffrey, & Smith, 1983;Watkinson & Hall, 2017). Strike-slip mechanisms ≥M 5.0 supportive of E-W sinistral shear are clustered along the eastern Montano fault with the largest event being a M w 6.1 in 2011 ( Figure 15; Watkinson & Hall, 2017). The western segments of the Montano fault do not exhibit any recent seismic activity ≥M 5.0 ( Figure 15). The ∼220 km long Palu-Koro fault trends NW from near the Bone Bay and extends offshore to the western extent of the North Sulawesi trench (Figure 15; Hall, 2018;Patria & Putra, 2020;Silver, McCaffrey, & Smith, 1983;Watkinson & Hall, 2017). Strike-slip mechanisms near the fault support NW-SE sinistral shear with slip rates estimated from GPS at ∼38 mm/y ( Figure 15; Bock et al., 2003;Watkinson & Hall, 2017). The Palu-Koro and Montano faults have been recognized as serious seismic hazards (Cipta et al., 2016;Irsyam et al., 2020;Watkinson & Hall, 2017). The Palu-Koro fault in particular has produced several events ≥M 6.0 in the past two decades, the largest being the September 2018 M w 7.5 Palu earthquake, which caused an uncharacteristically large tsunami and widespread liquefaction that devastated Palu ( Figure 15; Table 3; Socquet et al., 2019).
Other strike-slip faults on Sulawesi have exhibited little to no seismic activity recently, though may be tectonically active and sources of future seismicity based on earthquake hazard modeling and GPS studies ( Figure 15; Cipta et al., 2016;Irsyam et al., 2020;Socquet et al., 2006;Watkinson & Hall, 2017). The onshore segment of the Lawanopo fault ( Figure 15) for example, which runs roughly parallel and south of the Montano fault has been previously classified as a serious earthquake hazard (Cipta et al., 2016;Irsyam et al., 2020;Watkinson & Hall, 2017). The offshore segment of the fault or a related fault likely produced the 2001 M w 7.5 Banda Sea earthquake ( Figure 15; Table 3; Hayes et al., 2017;Watkinson & Hall, 2017). The Walanae and Gorontalo faults ( Figure 15) have also been identified as potential sources of seismic hazard (Cipta et al., 2016;Irsyam et al., 2020). Though they exhibit little to no seismicity, there is indication that the faults are currently locked, based on low slip rates derived primarily from GPS data (Cipta et al., 2016;Socquet et al., 2006;Watkinson & Hall, 2017). In eastern Sulawesi, the E-W trending dextral Balantak fault exhibits little to no apparent seismicity supporting dextral shear from available focal mechanisms but has been suggested to be tectonically active (Figure 15; Watkinson & Hall, 2017). The majority of strike-slip activity in eastern Sulawesi lies on a series of NE-SW trending faults that cross the Banggai islands ( Figure 15; Watkinson & Hall, 2017;Watkinson et al., 2011). These complex dextral-transpressive faults pose serious seismic hazards and produced the destructive M w 7.6 Banggai islands earthquake and tsunami in May 2000 (Hayes et al., 2017;Watkinson & Hall, 2017).

Normal Faulting
Crustal normal faulting is abundant in eastern Indonesia, commonly intermixed with reverse and strike-slip faulting in transpressive regions such as the Banda Sea, Sulawesi, and West Papua (Figures 15, 16 and 19). Prominent zones of normal faulting include the Aru trough ( Figure 16) and the Gulf of Tomini (Figure 15), with extension attributed to slab-rollback of the Banda and Celebes slabs respectfully (Adhitama et al., 2017;Hall & Spakman, 2015;Spakman & Hall, 2010). Normal faulting mechanisms ≥M 5.0 in the Aru trough support N-S trending faults and E-W extension (Figure 16a; Bird, 2003;Adhitama et al., 2017), where the Banda Sea moves approximately ∼15-18 mm/y westward away from the Indo-Australian plate (Bock et al., 2003;Hayes et al., 2017). The largest recent event in the Aru trough was a M w 7.0 in 2010 (Table 3; Hayes et al., 2017). In addition to the Aru trough, slab rollback in this region likely formed the Weber Deep, where a low angle normal fault termed the Banda detachment has never generated any recorded seismicity but may have caused the 1852 Banda Tsunami . Similar to the Aru trough, normal faulting events ≥M 5.0 in Sulawesi's Gulf of Tomini ( Figure 15) are supportive of NW-SE-striking normal faults and NE-SW extension that extends into the Togean islands consistent with rollback of the Celebes slab (Hall, 2018;Hall & Spakman, 2015;Watkinson & Hall, 2017).
Normal faulting is also common within the outer-rise regions of subducting plates, forming narrow trench-parallel seismic zones within the oceanic crust. Several extensional events ≥M 5.0 south of the Sunda-Java trench near Java (Figure 14a), and a similar zone east of the Philippine trench ( Figure 17a) support down-dip extension (slab-pull) associated with outer-rise normal faulting (Christensen & Ruff, 1988). Outer-rise faulting can provide insight into the state of stress on the megathrust. Abundant outer-rise normal faulting combined with few megathrust events suggests weak coupling on the megathrust, whereas reverse faulting in the outer-rise suggests strong coupling (Christensen & Ruff, 1988). The lack of large megathrust events and abundant outer-rise normal faulting from Java to Sumba suggests weak coupling on the eastern Sunda megathrust, supporting prior interpretations (Irsyam et al., 2020). The Philippine trench in contrast may exhibit strong coupling east of Mindanao due to infrequent thrust faulting in the outer-rise ( Figure 17; Ye et al., 2012). Though rare, outer-rise normal faults can generate events ≥M 7.0. The largest normal faulting event ever recorded occurred within the outer-rise south of Sumba in 1977, producing a massive M w 8.3 earthquake and an eight-meter tsunami that devastated Sumba and Sumbawa islands (Gusman et al., 2009;Lynnes & Lay, 1988). The possibilities of damaging singular events in the outer rise, or events preceding or following large megathrust events further compounds elevated seismic hazards at these subduction zones.

Shallow Intraslab Seismicity
Shallow intraslab events can occur by the same diverse mechanisms discussed in Section 4.3.7 but are constrained to between ∼20 and 60 km depth and often occur directly below the active megathrust interface (Seno & Yoshida, 2003;Wiseman et al., 2012). Individual events are capable of reaching magnitudes ≥M 7.0 and pose notable seismic hazard. The June 2000 M w 7.9 Enganno event for example occurred as an oblique mechanism ∼33-50 km depth beneath the southern Sumatran forearc and caused significant damage and fatalities to local communities ( Figure 13b; Table 3; Abercrombie et al., 2003;Wiseman et al., 2012;U.S. Geological Survey, 2020). Events can also compound seismic hazards already inherent with active megathrusting, capable of both inducing and being induced by megathrust ruptures (Wiseman et al., 2012).

Discussion and Conclusions
It is clear from this synthesis that the Indonesian region has a very high level of seismic activity, exhibiting an average of about 320 ≥ M 5.0 earthquakes each year with events ≥M 7.0 not uncommon, including the occurrence of five great (≥M 8.0) earthquakes since January 2000. The sources of seismicity are diverse, and include active megathrusts, crustal faults, and normal and reverse intraslab mechanisms. The majority of seismicity (77%) is shallow (≤70 km) and concentrated at or near plate boundaries and major faults. Abundant intermediate and deep seismicity (≥70 km depth) accounts for ∼23% of all seismicity and traces the Wadati-Benioff zones of subducting slabs with orientations generally consistent with recent tomography models and Indonesian tectonics.
Intermediate and deep seismicity distributions and mechanisms vary greatly by location and depth. Regionally, seismicity rates decrease exponentially with depth and increase within the MTZ, likely due to increasing pressure and temperature conditions with depth (Zhan, 2020), consistent with global seismicity-depth distributions (Billen, 2020;Frohlich, 1989Frohlich, , 2006Houston, 2015). Viscous resistance from phase changes at the 410, 520, and 660 km discontinuities likely contributes to elevated seismicity rates within the MTZ (Billen, 2020;Zhan, 2020), with the highest rates in the Celebes and Banda Sea regions of eastern Indonesia. The frequency of deep events decreases westward across Java toward Sumatra, likely due to a number of factors including increasing oblique convergence, decreasing lithospheric age, decreasing plate velocity, and increasing mantle temperatures. Similarly, lithospheric age and oblique convergence likely contribute to poorly developed Wadati-Benioff zones in the Manokwari trough and western New Guinea trench. Seismic gaps in Wadati-Benioff zones are common, with some inferred to be related to slab tearing (e.g., east Java and Timor), while others are not clearly understood (e.g., Sangihe and Philippine Sea slabs). Intraslab source mechanisms are diverse, with normal faulting commonly observed for deep events, and oblique-reverse mechanisms common at intermediate depths.
Individual events are capable of reaching magnitudes ≥M 7.0 and can pose notable seismic hazards such as the 2009 M w 7.6 Padang earthquake.
As tomography is an important verification for the locations of intermediate and deep seismicity, especially for the analysis of seismic gaps observed in this study, we compare our cross-sections with images of subducted slabs obtained by seismic tomography to evaluate their consistency ( Figure 20). We make our comparisons with teleseismic P-wave tomographic models (Fan & Zhao, 2018;Pesicek et al., 2010;Spakman & Hall, 2010;Widiyantoro et al., 2011b) because these models provide the highest resolution of the steeply dipping oceanic lithosphere subducted to the MTZ and deeper. Seismicity beneath the Molucca Sea ( Figure 20b) shows a seismicity gap at a depth of ∼250-350 km, whereas the slab appears to be continuous in the seismic tomographic image (Fan & Zhao, 2018). This result implies that seismicity gaps cannot be interpreted as slab breakoff unless it can be confirmed with seismic tomography. The gap in the Sangihe slab may be an aseismic section of a continuous slab. The seismicity beneath the Banda Sea (Figure 20c) also shows a seismicity gap at a depth of ∼300-500 km. The seismic tomography image shows the slab to be continuous, similar to the slab beneath the Molucca Sea in Figure 20b. Furthermore, both the deep seismicity and the image from seismic tomography show the slab flattens to a horizontal geometry within the MTZ with another seismic gap present. This is similar to the situation observed in Figure 9d, and indicates the slab is continuous where seismic gaps are present. In contrast, the pronounced seismicity gap from a depth of ∼200-550 km beneath central Java (Figure 20d) coincides with a slab gap observed in the seismic tomography image. This slab gap in central Java has been extensively discussed by Widiyantoro et al. (2011b) and Hall and Spakman (2015). As observed previously, the seismicity beneath Sumatra is notable for the lack of events deeper than 230 km (Figures 4b, 11b-11d and 20e). However, the image from seismic tomography shows a continuous slab to a depth of at least 300 km (Figure 20e), with some researchers indicating deeper to at least 660 km (Hall & Spakman, 2015). These comparisons indicate that (a) gaps in deep seismicity exist where the slab is continuous, (b) physical gaps in the slab coincide with seismicity gaps, and (c) slab bending inferred from seismicity patterns can be confirmed with seismic tomography images.
As observed from focal mechanism distributions, shallow seismicity is diverse, consistent with recent tectonic models for the location of major faults. The abundance and high magnitudes of shallow earthquakes also highlights the locations and sources of significant seismic hazards throughout the region. Shallow events pose greater seismic hazard due to hypocenters residing closer to the surface than intermediate and deep events.
Megathrusting is a primary source of shallow seismicity, and a prevalent source of seismic and tsunami hazard. Crustal faulting occurs primarily on major thrust, strike-slip, and normal fault zones, with individual events more than capable of reaching ≥M 7.0 and even great magnitudes. Regions with elevated seismic activity such as Sumatra, Sulawesi, the Banda Sea, and the Molucca Sea Collision Zone are clearly prone to prevalent seismic hazards, which beyond ground shaking may include liquefaction, landslides, and tsunamis. However, it is important to note that even less obvious regions with relatively subdued seismic activity, such as Java or the western New Guinea trench still pose notable seismic hazards (e.g., M w 7.7 Java 2006). Therefore, nearly all regions of Indonesia should be quantitatively assessed for the seismic and tsunami hazard. It is also useful to compare this study to previous comprehensive works on seismicity and tectonics, particularly the global tectonic plate modeling of Bird (2003) and the slab morphology modeling of Slab 2.0 (Hayes, 2018;Hayes et al., 2018). In terms of analysis, this study differs from that of Bird (2003) in several ways. Bird (2003) used the ISC catalog from 1964 to 1991, while the present study uses the USGS catalog from 2000 to 2020 (U.S. Geological Survey, 2020). The latter catalog is more accurate due to the significantly larger number of seismographic stations available in this more recent time period. Likewise, the focal mechanism catalog used by Bird (2003) is based on data from 1977 to 1998, whereas this study uses more reliable focal mechanisms from 2000 to 2020. Additionally, Bird (2003) did not consider seismicity cross sections or 3D plots to image the geometry of the subducting plates. As there is debate over the accuracy of the Bird (2003) boundaries, the boundaries would likely benefit from a reevaluation using more recent data. Bird (2003) tends to simplify the most complex faulting regions, particularly in east Indonesia, which arguably holds the most intense seismic activity in the region. Hall (2018) states that the fully connected and enclosed microplate boundaries chosen by Bird (2003) are useful for plate reconstructions but are in actuality inaccurate and discount the intense internal deformation within the plates. Plotting of fault locations rather than the enclosed plate boundaries is typically favored by researchers in this region, and for this reason we chose to use more recent fault locations to provide a more accurate picture of seismotectonics.
In contrast to Bird (2003), the Slab 2.0 model by Hayes (2018) provides an overview of the geometry of all currently subducting slabs and employs a different methodology than the present study. The model provides a useful visual tool for estimating slab geometry in comparison to the Wadati-Benioff zones presented in this study. Hayes (2018) defines the depth of the slab by using an iterative procedure with multiple constraints, including seismic catalogs, active-source seismic data interpretations, tomographic models, and receiver functions, where we simply employ the USGS seismic catalog locations and the Global CMT solutions (focal mechanisms). Hayes (2018) did not consider focal mechanisms, focusing rather on the slab geometry as a best-fitting 3D surface with depth contours. Importantly, the work of Hayes (2018) and the present study are anchored on the USGS earthquake catalog (U.S. Geological Survey, 2020), and therefore provide consistent results that provide insight into the seismotectonics of Indonesia.

Further Study
While this study provides a comprehensive overview of the current knowledge of Indonesian seismotectonics, there are several regions where further study is needed to provide a more complete analysis. For example, in West Papua, the nature of subduction at the Manokwari trough and western New Guinea trench is not well documented and would benefit from further study. Other faults on land that are not evaluated here need further study, such as the faults in West Java, and other back-thrust and splay faults associated with the Sunda arc. In addition, the origin of seismic gaps, such as the gap in the Sangihe slab beneath Mindanao or the aseismicity of the Philippine Sea slab require further study. Addressing these regions in greater detail would improve the assessment of Indonesian seismotectonics. Figure 20. Comparisons of the seismicity data used in this study with four recent P-wave seismic tomography models. (a) Map with seismicity deeper than 70 km, major plate boundaries, and locations of four tomography cross-section. IA, Indo-Australian plate; PS, Philippine Sea plate; CA, Caroline plate, and SU, Sunda plate. (b) Cross-section AA' through the southern Philippines, coincident with tomography model of Fan and Zhao (2018). The dashed black lines correspond to the +0.3% P-wave velocity perturbation contours that define the morphology of the slabs (Fan & Zhao, 2018). Seismicity is spatially consistent with the slab. A 100 km-wide gap is present in the Sangihe slab between ∼250 and 350 km depth, though the slab is interpreted to be continuous based on the tomography. The Philippine Sea slab is aseismic below a depth of ∼250 km. This section is comparable to Figure 6c. (c) Cross-section BB' through the western Banda slab, coincident with a seismic tomography model of Spakman and Hall (2010). Dashed black line follows the +1.2% P-wave velocity perturbation contour and outlines the slab. A seismic gap is evident at a depth of 280-500 km, but the slab is continuous. Both the seismicity and tomography indicate slab folding below a depth of 500 km. Compare with Figure 9d. (d) Cross-section CC' through Java coincident with the seismic tomography model of Widiyantoro et al. (2011b). The dashed black lines follow the +0.8% P-wave velocity perturbation contour used by Widiyantoro et al. (2011b) to construct 3D models of the slab. The tomography suggests a gap in the slab (Hall & Spakman, 2015;Widiyantoro et al., 2011b) and we find a coincident seismicity gap. Compare with Figure 10b. (e) Cross-section DD' through central Sumatra coincident with the tomography model of Pesicek et al. (2010). The dashed black line follows the +1.2% P-wave velocity perturbation contour used in the study. The tomography model by Pesicek et al. (2010) only extends to 300 km depth, but Hall and Spakman (2015) report that the slab extends to 660 km without showing tomograms. Seismicity appears to trace the slab and there is a seismic gap below 200 km. Compare with Figures 11b-11d. All cross-sections shown are 1000 km long and scaled 1:1. Hypocenters are colored by depth following the scale in panel (a).