We present evidence for significant aseismic fault slip at shallow depth above a pair of mainshock-aftershock sequences in the Zagros Mountains of Iran. The two Mw 5.9 earthquakes are each spanned by high-quality geodetic imagery and have well-recorded sequences of aftershocks that occurred beneath a salt décollement. Earlier studies of the geodetic data inferred that the mainshocks were located above the décollement, requiring a ~10 km spatial separation between aftershock cluster and earthquake centroid. We find that the geodetic data simultaneously allow two slip sources of similar magnitude: one within the basement, collocated with aftershocks, and one shallow source (also equivalent to Mw 6) responsible for the primary signal apparent in the geodetic imagery. Should this phenomenon be widespread in the Zagros, it would partially explain a previously noted discrepancy between observed seismic moment release in the Zagros and current convergence rates between the Arabian and Eurasian plates.
 A key question in active tectonics is how currently observed deformation, aseismic or seismic, is accommodated across plate boundaries and contributes to seismic hazards and long-term formation of geologic structures [Jackson and McKenzie, 1988; King et al., 1988; Masson et al., 2005]. Geodetic observations have illuminated a spectrum of episodic aseismic fault slip behavior at tectonically active boundaries [Dragert et al., 2001; Fielding et al., 2004; Linde et al., 1996; Lohman and McGuire, 2007] that may contribute significantly to the strain budget in zones of continental deformation. In the Zagros Mountains of Iran, satellite-based Interferometric Synthetic Aperture Radar (InSAR) observations spanning three moderate-sized earthquakes (Figure 1), combined with detailed aftershock locations, provide new insight into how strain accommodation varies vertically within an actively deforming mountain belt. Here we examine three thrust earthquakes—the 2005.11.27 Qeshm Island (Mw 5.9) and 2006.03.25 Fin (Mw 5.9) events in the Simply Folded Belt (SFB) and the 2006.02.28 Tiab (Mw 6.0) event in the adjoining High Zagros (Figure 1)—where surface displacements are constrained by InSAR observations and aftershocks were recorded by temporary, densely spaced local seismic networks [Gholamzadeh et al., 2009; Nissen et al., 2010; Roustaei et al., 2010]. The geodetic data for the SFB earthquakes indicate fault slip at a depth range that is significantly shallower than the depth range spanned by the aftershock cloud, although the equivalent seismic moment is similar to that detected teleseismically. We hypothesize that there are two separate deformation sources of similar magnitude for each event: one that was shallow and aseismic and that dominated the observed deformation signal, and one seismic source associated with the aftershock sequence that was deep enough to not be readily apparent in the InSAR data. We show that these results are consistent with predicted static Coulomb stress changes and regional geology.
2 The Zagros Mountains and Disagreement Between Data Types
 The Zagros SFB is an active fold-and-thrust belt characterized by an 8–10 km thick sedimentary section that is detached from Precambrian Arabian basement by the 1–2 km thick Hormuz Salt [Alavi, 1980; Falcon, 1975]. Teleseismic data and micro-earthquakes suggest that most seismicity occurs at depths of 10–25 km [Maggi et al., 2000; Tatar et al., 2004; Engdahl et al., 2006] within the uppermost portion of Arabian plate basement and deeper than the folded sedimentary section. Seismic moment tensors summed over the past century cannot account for the full convergence measured geodetically [Jackson and McKenzie, 1988; Masson et al., 2005], implying that significant shortening in the SFB may be accommodated aseismically. The high-quality InSAR data coverage for earthquakes in the Zagros (Figure 1) allows reexamination of the plate motion budget and an assessment of how strain accommodation varies between the stratigraphic section and basement.
 Previous studies of InSAR data spanning the 2005 Qeshm and 2006 Fin earthquakes [Lohman and Barnhart, 2010; Nissen et al., 2007; Roustaei et al., 2010] (Figure 1) inferred that coseismic slip was restricted to the sedimentary section between 3 and 10 km (Figure 2). The available SAR imagery (Table A1) brackets a short (weeks) time interval for the Qeshm earthquake and a longer (months) range for the Fin event. In these two examples, inferred fault slip does not overlap with the depth range of aftershocks (10–30 km) recorded by local arrays deployed days after each event [Nissen et al., 2007; Roustaei et al., 2010] or micro-earthquakes recorded in the region [Tatar et al., 2004]. Detailed teleseismic body wave modeling of each mainshock is consistent with centroid depths within either the basement or sedimentary section [Nissen et al., 2007; Roustaei et al., 2010; Engdahl et al., 2006], as has been found elsewhere for earthquakes of similar size [Devlin et al., 2012]. Although aftershocks generally fill a region several times larger than the area that ruptured coseismically, the highest density of aftershocks is usually closely associated with the ruptured region itself—a relationship that is violated if the Qeshm and Fin coseismic ruptures are located in the sedimentary section. For this reason, we examine the possibility that the mainshock did occur in the basement for which the resulting deformation signals are too broad and low magnitude to be apparent in the InSAR data.
3 Fault Slip Resolution
 We explore the significance of the apparent separation of mainshock and aftershocks for the Fin and Qeshm earthquakes by reproducing the inversions of the available InSAR data while also examining the sensitivity of the InSAR to fault slip at the depths spanned by the aftershocks. For each event, we generate interferograms using Envisat ASAR images acquired by ESA (Figure 1 and Table A1). We use the JPL/Caltech ROI_PAC software package [Rosen et al., 2004] and the 90 m resolution Shuttle Radar Topography Mission digital elevation model [Farr et al., 2007]. We use a spatial resolution of ~31 × 55 m, then estimate interferogram noise structure, and downsample the resulting interferograms from ~106 pixels to a computationally manageable ~102 pixels using a model resolution–based quadtree method [Lohman and Simons, 2005] (Figures A1–A3 and Tables A3–A5).
 To determine a best fit fault geometry for each observed deformation signal, we first invert the interferograms spanning each event for the geometry of a single rectangular fault plane with uniform slip [Okada, 1992] varying strike, dip, slip direction, hypocentral location, fault length, and fault width using the neighborhood algorithm [Sambridge, 1999]. We cannot discriminate within error between the two potential nodal planes, so we consider two planes for each event in our conclusions. For the Fin and Qeshm events, which have relatively complicated surface deformation signatures, we fix the fault geometry to that of the best fit fault patch with uniform slip and extend the fault both along strike and down dip so that a distributed slip inversion does not produce artifacts from interactions with the edges of the model. We then discretize the fault model with triangular dislocations [Meade, 2007], whose size varies with model resolution [Barnhart and Lohman, 2010], and invert for the best fit slip distribution (Figures A5 and A6). We impose nonnegative slip constraints and fix the rake to that from the uniform slip inversion. We find that fixing the rake does not produce a noticeably different slip solution than when we allow rake to vary freely.
 In general, the resolution of geodetic inversions decreases with distance from the data (in this case, depth below the surface). We perform Monte Carlo sensitivity tests that constrain the appropriate error bounds on our inferred slip models due to noise in the data (Figure 3). For both the Qeshm and Fin earthquakes, we find that slip with the observed Global CMT magnitudes of Mw 5.9 at 10–22 km—the depths of aftershocks—are permissible given the level of noise in the InSAR data [Lohman and Simons, 2005] in addition to the shallow slip located in the stratigraphic section (Figure 3). Our tests do not account for the contribution from errors in crustal elastic parameters, nonplanar fault geometries, etc. Accounting for these errors would tend to increase the range of possible slip values, making it even more likely that the earthquakes could “hide” at depth.
 The 2006 Tiab earthquake (Figure 1), located outside the salt-dominated SFB, provides a counterexample and does not share the apparent separation of aftershocks and mainshock when we compare geodetic and seismic data. Inversions of the InSAR observations produce fault slip within the depth ranges of locally recorded aftershocks (Figure 2c) [Gholamzadeh et al., 2009], consistent with the behavior of typical mainshock-aftershock sequences. The collocation of InSAR-derived slip models and aftershock locations for the Tiab earthquake suggests that the separation observed in the SFB is not simply an artifact of our approach and is likely due to differences in behavior between the two regions.
 From these observations, we find that the InSAR and aftershock data for the Fin and Qeshm events are each consistent with two slip sources: one shallow source within the sedimentary section, and one deeper source within the basement. Because the deep sources are collocated with aftershocks in the basement, we infer that the deep sources are the coseismic ruptures recorded teleseismically. This removes the necessity to explain an extraordinary lack of aftershocks near the coseismic rupture. Furthermore, coseismic rupture in the basement is more consistent with perceived ground motions, which were initially overpredicted by a seismic source within the sedimentary section during the Qeshm Island earthquakes [Jaiswal et al., 2009].
 We infer that the coseismic rupture in the basement induced aseismic fault slip within the sedimentary section, resulting in the primary signal apparent in the InSAR observations. Laboratory experiments on halite layers within sandstone produce both stick-slip behavior and ductile flow when an abrupt stress change is imposed [Shimamoto and Logan, 1986; Shimamoto, 1986]. These experiments indicate that salt present along fault zones throughout the SFB [Jahani et al., 2009] may permit triggered aseismic slip when exposed to stress changes of the sort considered here. Furthermore, predicted static Coulomb stress changes [King, 2009; Lin and Stein, 2004] for an earthquake occurring within the cloud of basement aftershocks and with the mechanism reported by the Global CMT would encourage slip at the location where we infer shallow aseismic creep (Figure A7).
 Additionally, the observed surface deformation during the Fin aseismic slip event is consistent with the long-term evolution of the individual folds. The line-of-sight uplift observed during the Fin sequence occurs primarily on the dipping limb of a fold identified in optical imagery and digital elevation models (Figure 4a) and is consistent with fault-bend folding inferred elsewhere in the SFB [Burberry et al., 2008]. Kinematic fault-bend fold models predict that, in a mature fold where total accumulated slip exceeds the width of the ramp, slip on the dipping ramp produces uplift of the limb above the ramp alone (Figure 4b), whereas fault-propagation and detachment fold models would predict uplift at the crest [Suppe, 1983]. The Hormuz Salt (~11–12 km) and shallower (3–4 km and ~6 km) evaporite and shale horizons [Jahani et al., 2009; Sherkati et al., 2005] bracket the inferred depths of the Fin aseismic slip event, which is consistent with shallow active detachments (Figure 4b). Earthquake-related fold shortening has been inferred for the 2005 Qeshm event [Nissen et al., 2007] and elsewhere for other events, including the 1980 Algerian earthquake [King and Vita-Finzi, 1981] and the 1983 Coalinga earthquake [e.g., Hill, 1984; Stein and Ekström, 1992].
 The data used in this study place no constraint on the geometric relationship between faults in the basement and sedimentary section or the processes by which slip is transferred through the intervening Hormuz Salt. While salt at 10–12 km depth is likely to behave ductilely at longer time scales, it behaves elastically at short time scales (seconds) and perhaps can sustain the static coseismic Coulomb stress change long enough to initiate aseismic deformation in the sedimentary layers above it. Alternatively, the convergence history between Arabia and Eurasia may have resulted in basement relief that allows direct coupling between a single fault in the basement and the upper cover rocks. These questions may be resolved when we have further knowledge of the exact geometry of the two slip sources or the character of interseismic deformation associated with the fold belt. Where sufficient data exist, InSAR time series analysis can have submillimeter per year detection thresholds [e.g., Finnegan et al., 2008], suggesting that future InSAR missions with frequent “background” acquisitions may allow us to resolve the timing between coseismic rupture and triggered aseismic slip.
 This work shows that aseismic shortening in mountain belts such as the Zagros occurs, at least in part, episodically as seismically triggered, aseismic events. The inferred aseismic deformation accompanying the Fin and Qeshm earthquakes is equivalent to, if not greater than, the magnitude of the coseismic deformation (Figure 3 and Table A2). Accordingly, the aseismic deformation may effectively double the moment release during each earthquake sequence. This indicates that a significant portion of the inferred seismic deficit [Jackson and McKenzie, 1988; Masson et al., 2005] is accommodated over short periods (days to weeks) following earthquakes rather than through steady interseismic motion.
 We thank M. Barazangi, C. Andronicos, and C. Connors for extensive critical discussions leading to this manuscript. We are grateful to R. Bilham, R. Mellors, and three other anonymous reviewers for their constructive criticism. SAR imagery was provided through a Category-1 proposal through the European Space Agency. Detailed methods, derived data products, and modeling results are available in the supplemental material. This work was funded in part by the American Chemical Society Petroleum Research Fund grant 49877-DNI8. W.D.B. was funded through NASA graduate fellowship NNX09AO30H.