The great Mw = 9.0 2011 Tohoku earthquake appears to have complex rupture characteristics, with slower rupture velocity during the early portion of the rupture and spatial variations in the radiation frequency content. These spatial and temporal variations suggest that the subduction zone fault has spatially varying friction conditions that led to differences in the 2011 rupture characteristics, conditions that might also affect other earthquakes within the rupture zone. We find spatial variations for source parameters of 90 relocated earthquakes between 1992 and 2011 along northern Japan, with longer durations observed in shallow near trench events relative to shorter duration deeper events. A majority of these events do not lie within the high slip zone of 2011, however, and occur instead in the region of the 1896 tsunami earthquake to the north. We also find correlation between the longest duration event locations and low seismic velocities based on recent tomography models.
 The 11 March 2011 Mw = 9.0 megathrust earthquake that occurred along the subduction boundary offshore northeast Japan (Figure 1) produced strong local shaking and a devastating tsunami that led to over 15,000 fatalities and staggering economic losses. Studies of the earthquake rupture suggest that the entire seismogenic width slipped during the event, with massive amounts of slip in the near trench region and peak slip reaching 50–60 m [e.g., Simons et al., 2011; Ide et al., 2011; Koketsu et al., 2011; Yamazaki et al., 2011; YueandLay, 2011; Earle et al., 2011]. Near-trench slip can be very effective at generating large tsunami, but conventional understanding of fault processes suggests that large slip along the shallow portion of the fault is fairly unusual given the commonly assumed velocity strengthening frictional conditions near the trench [e.g.,Oleskevich et al., 1999]. Therefore, this event highlights the critical need for better understanding of how the near-trench region can slip by such large amounts.
 Observations of the 2011 mainshock also suggest depth variations in frequency content of the rupture, similar to other recent large events [e.g., Wang and Mori, 2011; Koper et al., 2012]. For the 2011 event, high frequency radiation appears to have originated primarily along downdip portions of the megathrust, with very little high frequency radiation originating from the shallow, near trench high slip zone [e.g., Koper et al., 2011; Ishii, 2011]. These rupture variations again suggest spatial variations in fault conditions.
 Here we explore whether source characteristics of other earthquakes located in the 2011 rupture zone exhibit spatial variability similar to the 2011 mainshock that may be tied to physical or mechanical features of the megathrust zone. Our earthquake catalog spans over 2 decades, including aftershocks of the 2011 mainshock and covers much of the 2011 rupture zone (Figure 1). Our dataset also includes events within the region of the 1896 tsunami earthquake, north of the 2011 event. Tsunami earthquakes, events that commonly have very shallow slip and produce large tsunami relative to their size [e.g., Polet and Kanamori, 2000], provide other examples of shallow slip events, thus comparisons of rupture characteristics in this zone can also be useful. All of the earthquakes in our dataset are relocated so that the improved hypocenters allow us to address whether other Japanese earthquakes show depth dependent rupture characteristics, whether certain areas consistently rupture with slower slip, and whether there are any differences in rupture characteristics for events before and after the mainshock.
2. Data Set and Methods
 We examine 90 events located along northern Japan with 5.6 < Mw < 7.4 from 1992 to 2011 (Figure 1) using teleseismic recordings from stations primarily in the Global Seismic Network (GSN) and Federated Digital Seismic Network (FDSN). The majority of the events have thrust focal mechanism consistent with slip on the plate boundary based on geometry from the Global Centroid Moment Tensor catalog (GCMT) [Ekström et al., 2005]. For aftershocks of the 2011 event, we examine all possible events regardless of mechanism, but for events prior to the 2011 mainshock, we analyze only those with thrust mechanisms.
 We use scaled rupture duration to serve as an indicator of slower or faster rupture, as we assume the variations in rupture duration are linked to variations in rupture velocity. For each event, we use a multi-station deconvolution procedure using bothP and SH waves to determine a source time function at a particular depth (Figure 2) [e.g., Ruff and Miller, 1994; Bilek and Lay, 2002; Bilek et al., 2011]. The procedure is repeated for a range of depths (3–65 km in 1 km increments), consistent with expected depth range of seismogenic behavior on a subduction megathrust. The optimal event depth results in the smallest misfit between the data and synthetic seismograms. We measure duration from the source time function at the lowest misfit depth and normalize it using cubed root of seismic moment scaling to account for the effect of increasing duration with increasing size [e.g., Bilek and Lay, 2002].
 Each event is also relocated in order to provide an improved spatial comparison than could be possible with PDE catalog locations alone. We use the method of Pesicek et al.  to identify depth phases (pP, pwP, sP) and improve phase onset times. The additional depth phases are incorporated with existing phase catalogs and relocated using the Engdahl, van der Hilst, and Buland (EHB) teleseismic location approach [Engdahl et al., 1998]. For each event, we first calculate the power spectral density function at 1 Hz for vertical component velocity and displacement waveforms. The resulting time series reflects the magnitude of the discrete Fourier transform at that frequency. We next find the gradient of the time series by calculating the derivative, apply minimum smoothing, and normalize to 1.0. To identify changes in the gradient time series, points less than the mean are set to zero. Abrupt changes in the gradient occurring on both time series are considered triggers. Triggers that occur within a few samples on both time series are associated with P, pP, or sP. We report onset times from the velocity time series. Observed scatter in the fit to depth phases using the EHB approach can be largely attributed to the effects of complexities in structure across the region on the arrival times of depth phases that have been modeled using the ak135 model [Kennett et al., 1995] overlain by water depth estimates at their bounce points. Relocated event hypocenters are provided in Table S1 in the auxiliary material.
 Depth estimates based on the deconvolution and the relocation are evaluated for similarity. If these estimates are more than 10 km different (in this case ∼17% of events), we do a manual review of waveforms and results for both techniques. In most of these cases, there is a 2nd minimum in the deconvolution misfit curve and we choose this depth and corresponding duration if this new depth is within 10 km of the relocation depth. In other cases, relocations are computed again following manual review of waveform picks. Comparisons between the different methods suggest that the EHB relocated depths are deeper than both the deconvolution based depth and NEIC catalog values, likely due to velocity model differences used in the calculations. Synthetics generated in the deconvolution method are based on a simple water layer over a half-space (Vp = 6.7 km/s) velocity model, and the EHB method uses ak135. Based on depth phase reading errors and effects of 3D structure, we estimate depth errors on the order of 10 km with respect to the ak135 global velocity model. Our relocated depths are also similar to the JMA produced depths.
3. Spatial Variations of Source Characteristics
 Normalized source duration estimates for events both before and after the mainshock are similar, and a general trend of decreasing duration with increasing depth exists, similar to results for other subduction zones [e.g., Bilek and Lay, 2002; Bilek et al., 2011] (Figure 3). Overall the mean normalized duration is 4.2 s with a 2 s standard deviation, with a mean of 4.1 s for the pre-mainshock event population and a mean of 4.5 s for the aftershock population.
 Whereas all events in our catalog prior to the mainshock have thrust mechanisms, a range of focal mechanisms exists in the aftershock population. Several of the aftershocks are very shallow, as we might expect for upper plate events, and these have very short durations. However, there are a significant number with depths >15 km that have longer than average durations.
 We also find along-strike variations in normalized duration along this entire segment of the Japan Trench (Figure 1). Along the northernmost section (A-A′), all the analyzed events (pre and post mainshock) are thrust mechanism, lie very close to the estimated slab position ofHayes et al. , and have shorter than average durations throughout the entire depth range (15–35 km). Longer than average duration events occur both before and following the Tohoku earthquake within section B-B′, which is also the region of the 1896 tsunami earthquake. These events are all thrust mechanism, and, along with their location relative to the slab position, are consistent with slip on the plate interface. This region shows overlap of long duration events that have occurred in nearly the same location but over tens of years, and continued after the 2011 mainshock. Four of these long duration events extend deeper along the plate interface, down to ∼40 km.
 Within region C-C′, all events are thrust mechanism with locations consistent with the projected slab position. The longest duration event of the entire catalog occurred at shallow depth (16.5 km relocated depth, 12 km deconvolution depth) in this section, 1 day prior to the 2011 mainshock (2011/03/10 08:08:20 UTC). To the south along section D-D′, nearest the mainshock epicenter, the majority of pre-2011 earthquakes had short durations. The majority of long duration events within this section are aftershocks with non-thrust mechanism; their normal or oblique slip mechanisms and hypocenter locations suggest that some of these events were likely within the upper plate or within the subducting slab. Within sections E-E′ and F-F′, 2 long duration thrust events occur within 50 km of the trench in the very shallow portion of the seismogenic zone. There are also 2 long duration non-thrust mechanism events within 100 km landward of the trench that lie at or near the top of the projected slab.
4. Comparison With Slip Distributions
 Slip distributions of the 2011 event vary in detail depending on the data used in the inversions, but gross features of the inversions are consistent, such as the largest slip patch concentrated at the shallow portion of the subduction zone [e.g. Koketsu et al., 2011; Yokota et al., 2011; Lay et al., 2011; Simons et al., 2011; Earle et al., 2011]. The majority of events within our catalog prior to the mainshock occurred within the northern portion of the 2011 rupture zone but outside of the area of high slip (Figure 4). The longest duration event in the entire dataset occurred within the area of eventual high slip 1 day prior to the mainshock. Several of the long duration thrust events, both pre- and post-Tohoku mainshock, occurred within 50–200 km of the 1896 tsunami earthquake. Aftershocks are scattered throughout the rupture zone without much overlap in the high slip region. The one aftershock that occurred within the high slip zone had shorter than average rupture duration, although this event had a strike slip focal mechanism and hypocenter depth at 22–26 km within the subducted slab. The majority of thrust events in the downdip 2011 rupture zone and area of high frequency radiation are short duration, although a few (<5) have average to above average durations. We also compared our results to other published slip distributions and find similar results. We find 1–2 more events within moderate slip areas (20–30 m) when comparing with models with wider regions of near trench slip [Simons et al., 2011; Koketsu et al., 2011; Yamazaki et al., 2011; Yokota et al., 2011], but none in the peak slip zone.
 The lack of events in our catalog within the area of highest slip makes it difficult to draw conclusions about how conditions within that specific region might impact other earthquake ruptures. This lack of seismicity in the region of eventual high slip is not unexpected, as it was also observed in Sumatra [e.g., Engdahl et al., 2007]. One area that does appear to show consistent long duration earthquake characteristics is within the segment between 39.5°–40.5°N in the area of the 1896 tsunami earthquake. The majority of events in our catalog that are located within this region have longer than average normalized durations, and we see these events extend along most of the seismogenic width. This suggests the possibility of long-lived (over a century since the 1896 event) conditions that can lead to long duration events, and may impact future events that occur within the 2011 rupture zone. We observed similar behavior in the northern Sumatra subduction zone [Bilek et al., 2011].
 This observation of spatial source variations leads to questions about what on the megathrust might cause the variations. Zhao et al.  show that the 2011 mainshock epicentral area and area of high slip occur within a region of high velocity in tomography models through the area [Huang et al. 2011]. The authors relate the velocity variations to variations in coupling and/or friction conditions on the megathrust, with the high velocity regions indicative of strongly coupled regions. Comparison of our source parameters with the tomography results shows that of 18 events that have normalized duration 1 standard deviation longer than the mean (6.2 s), 11 of these are located either in a region of low velocity or in a transition between low to high, 5 events occur in high velocity regions, and 2 occur in region of no tomography coverage (although the southernmost low velocity region can likely be extrapolated to these 2 earthquakes). Zhao et al.  suggest that the high velocities may indicate the presence of subducting ridges or other features, and that low velocities may indicate the presence of subducted sediments and/or fluids [e.g., Mishra et al., 2003; Hyndman and Peacock, 2003]. These correlations between the longest duration events and areas of low or transitional velocity provide further support for the ideas that the frictional conditions are linked to the rupture characteristics of these earthquakes.
 Using a catalog of 90 relocated earthquakes that occurred between 1990 and late 2011 (after the Mw= 9.0 Tohoku earthquake), we find rupture duration varies both along strike and with depth for events throughout the entire time period, with longer duration events generally at the shallower updip end of the seismogenic zone. In the area of the 1896 tsunami earthquake, north of the 2011 high slip zone, long duration events occurred over the entire time period, and extend from very shallow to 40–50 km depth. One very long duration event occurred within the shallow high slip zone 1 day prior to the 2011 mainshock. The majority of thrust events likely on the plate interface that occurred within the 2011 zone of high frequency radiation had short rupture durations. Within the non-thrust mechanism aftershocks, there is a diversity of duration. We suggest that the rupture variations observed with this catalog and the 2011 mainshock are spatially controlled by conditions along the megathrust that may be fairly stable over long (decadal) timescales and that produce low seismic velocities along the megathrust.
 We thank Andrew Newman and Keith Koper for constructive reviews. We thank G. Hayes for supplying the slip distribution of the 2011 mainshock, and Shishay Bisrat for aiding with waveform review. All waveform data was obtained from the IRIS Data Management Center. This work was supported by NSF-OCE funding (0840908 to S.L.B., 0841022 to H.R.D., and 0841040 to E.R.E.).
 The Editor thanks the two anonymous reviewers for their assistance in evaluating this paper.