Aseismic fault creep rate measurement on strike-slip faults
Some fault segments slip continuously without earthquakes, accumulating or relieving seismic moment deficit. InSAR time-series over ERS, Envisat and Sentinel-1 archives resolves surface creep rates to sub-millimetre per year precision, constraining where the next locked patch begins.
Sensors
- Sentinel-1 A/B (C-band SAR, ESA): 5.6 cm wavelength, 6-day repeat at mid-latitudes with both satellites active, IW mode ground range resolution approximately 5 m x 20 m (range x azimuth), global systematic acquisition since 2014. The primary workhorse for current creep-rate time-series.
- ERS-1/2 and Envisat ASAR (C-band SAR, ESA archive): ERS archive from 1992, Envisat from 2002 to 2012. 35-day repeat, approximately 25 m resolution. Indispensable for multi-decade velocity fields and for separating secular creep from transient events. Archive accessible via ESA EOLI and Copernicus Data Space.
- ALOS-2 PALSAR-2 (L-band SAR, JAXA): 23.6 cm wavelength. L-band coherence is maintained over vegetated terrain where C-band decorrelates entirely, making PALSAR-2 the preferred sensor on forested fault segments such as portions of the Chaman system. Stripmap mode: approximately 3 m x 3 m resolution, 14-day repeat.
- ALOS-1 PALSAR archive (L-band SAR, JAXA): 2006 to 2011 archive at 46-day repeat. Provides the L-band baseline epoch needed to extend velocity time-series back before ALOS-2 and to detect any acceleration in creep rate over the past two decades.
What a creeping fault looks like to a radar
A strike-slip fault that creeps aseismically displaces the ground surface laterally, continuously, at rates typically between 2 mm/yr and 25 mm/yr depending on the fault and depth extent of creep. In a line-of-sight InSAR interferogram, this lateral motion projects onto the radar range direction and appears as a sharp, linear phase discontinuity crossing the fault trace. The discontinuity is not gradual; it steps by a fraction of a fringe over a distance of tens to hundreds of metres. That sharpness is the diagnostic signature distinguishing shallow surface creep from the broad, smooth gradient produced by deep interseismic elastic loading.
The published creep rate on the Hayward Fault in California ranges from roughly 3 mm/yr near its southern end to around 9 mm/yr near Fremont, measured consistently across ERS, Envisat and Sentinel-1 datasets. The North Anatolian Fault's Ismetpasa segment has been documented creeping at approximately 20 to 25 mm/yr in multiple InSAR studies, close to the full plate rate, implying very little moment deficit on that segment. These numbers matter operationally: a segment creeping at the full plate rate is releasing strain continuously and is unlikely to produce a large earthquake; a locked segment adjacent to it is accumulating the deficit.
The screw-dislocation model and what it tells you about depth
The standard analytical model for a vertical strike-slip fault is Savage and Burford's (1973) screw-dislocation formulation. Surface velocity parallel to the fault varies as the arctangent of distance from the fault trace, scaled by the far-field plate velocity and modulated by the locking depth. Fitting this model to an InSAR velocity profile perpendicular to the fault yields two parameters: the creep rate at the surface and the depth to which creep extends. A fault creeping to 5 km depth produces a narrower, more localised surface velocity gradient than one locked from the surface to 20 km.
The limit of this approach is that the screw-dislocation model assumes a planar, vertical fault in a homogeneous elastic half-space. Real faults are neither. Dipping fault geometry, along-strike variation in locking, and lateral contrasts in crustal rigidity all introduce systematic bias into the inferred locking depth. Published studies address this with 3-D boundary-element or finite-element models, but those require independent constraints on fault geometry and crustal structure. The InSAR velocity field alone cannot uniquely resolve the depth distribution of slip without additional assumptions or complementary geodetic data such as campaign GPS.
Atmospheric delay: the noise that mimics a fault signal
The single largest source of error in creep-rate measurement is tropospheric delay. Water vapour gradients across a fault zone, particularly where topography forces orographic moisture differences, can produce a linear phase ramp in an interferogram that is spatially correlated with the fault trace and has an amplitude of several millimetres. In a single interferogram this is essentially indistinguishable from a real creep signal.
The solution is time-series analysis over many acquisitions. Methods such as SBAS (Small Baseline Subset) and persistent-scatterer interferometry (PSI) average down atmospheric noise because tropospheric delay is largely uncorrelated in time. With Sentinel-1's 6-day repeat, a two-year stack of 120 or more interferograms reduces atmospheric noise to below 0.5 mm/yr in favourable conditions. Empirical corrections using ERA5 reanalysis weather data or GACOS (Generic Atmospheric Correction Online Service) further reduce residuals, though neither eliminates the problem entirely over complex terrain. Honest practice is to report the formal uncertainty from the time-series inversion alongside the creep rate estimate, and to flag segments where the fault-parallel topography is steep enough to make atmospheric correction unreliable.
Separating shallow creep from deep interseismic loading
This is the central interpretive challenge. Both processes produce surface velocity gradients across a strike-slip fault, but at different spatial scales. Shallow creep (from the surface to a few kilometres depth) produces a sharp, narrow signal concentrated within roughly one locking-depth-width of the fault. Deep interseismic loading from a locked fault produces a broad, smooth gradient extending tens of kilometres on each side. In practice both signals are present simultaneously on a partially creeping fault, and separating them requires either a dense velocity profile with good near-fault coverage or a two-layer model that fits both the narrow and broad components independently.
Near-fault coherence is the practical bottleneck. Urban environments on the Hayward Fault maintain coherence well; sparsely vegetated desert terrain on the Chaman Fault in Pakistan is manageable in L-band but loses coherence in C-band during seasonal vegetation cycles. Agricultural land, which covers much of the North Anatolian Fault's western segments, decorrelates at C-band over summer months, creating seasonal gaps in the time-series that can alias the creep signal if not handled carefully. Selecting interferogram pairs that span full seasonal cycles, rather than arbitrary time windows, reduces this bias.
From velocity map to seismic moment deficit
The practical output for a hazard assessment is not the creep rate itself but the moment deficit rate: the rate at which elastic strain energy accumulates on the locked portions of the fault. This is computed by subtracting the geodetically observed surface creep from the long-term geological or GPS-derived slip rate. If a fault segment is slipping at 5 mm/yr aseismically on a fault with a 20 mm/yr plate rate, the locked portion accumulates moment at a rate equivalent to 15 mm/yr of slip over the locked area. Integrated over decades or centuries, this constrains the maximum credible earthquake magnitude for that segment.
Satellize processes Sentinel-1 and archive C-band and L-band stacks for clients requiring fault-segment creep inventories, delivering georeferenced velocity grids and along-fault creep-rate profiles with formal uncertainties. The workflow is the same class of time-series InSAR used in published Hayward and North Anatolian studies. Creep-rate products are typically delivered as GeoTIFF velocity grids and CSV fault-profile exports compatible with standard geodetic modelling packages.
Honest limits of the method
InSAR measures only the line-of-sight component of displacement. For a near-vertical strike-slip fault, the horizontal fault-parallel motion projects poorly onto the radar look direction for certain orbit geometries. Ascending and descending pass combinations recover both the east-west and vertical components of surface motion, but the north-south component remains poorly constrained by C-band or L-band side-looking radar. On faults oriented close to north-south, this geometric insensitivity can underestimate creep rates by a factor of two or more. Azimuth offset tracking or multiple-aperture interferometry (MAI) can partially recover the along-track displacement, but at coarser precision than range-direction measurements.
Minimum detectable creep rate depends on the length of the time-series, the atmospheric conditions, and the coherence of the scene. In ideal conditions over urban or rocky terrain with a five-year Sentinel-1 stack, rates below 1 mm/yr are resolvable. Over agricultural or forested terrain with seasonal decorrelation, the practical detection floor rises to 3 to 5 mm/yr. Faults with creep rates below that threshold require either L-band data, a longer archive, or campaign GPS to constrain the signal. These are not caveats to hide; they are the inputs a hazard analyst needs to decide whether InSAR alone is sufficient or whether a combined geodetic approach is warranted.
Typical figures
| Typical spatial resolution (Sentinel-1 IW) | ~5 m range x ~20 m azimuth; velocity grids commonly posted at 50–100 m after multilooking |
| Revisit period (Sentinel-1, both satellites) | 6 days at mid-latitudes; 12 days with single satellite |
| Archive depth | C-band from 1992 (ERS-1); L-band from 2006 (ALOS-1 PALSAR) |
| Radar frequency / wavelength | C-band: 5.405 GHz / 5.6 cm (Sentinel-1, ERS, Envisat); L-band: 1.27 GHz / 23.6 cm (ALOS-2 PALSAR-2) |
| Line-of-sight displacement precision (time-series) | 0.3–1 mm/yr in coherent, low-atmospheric-noise conditions over multi-year stacks |
| Minimum detectable creep rate (practical) | ~1 mm/yr over urban/rocky terrain; 3–5 mm/yr over vegetated or agricultural terrain |
| Swath width (Sentinel-1 IW) | 250 km; enables full fault-system coverage in a single pass |
| Delivery formats | GeoTIFF velocity grids, NetCDF time-series cubes, CSV fault-perpendicular velocity profiles, PDF technical report |
| Atmospheric correction | ERA5-based or GACOS empirical correction applied; residual uncertainty reported per pixel |
Analytics Satellize can run
| Mean line-of-sight velocity map | SBAS or PSI time-series InSAR over multi-year Sentinel-1 or archive C/L-band stack | GeoTIFF velocity grid with per-pixel formal uncertainty, georeferenced to WGS84 |
| Along-fault creep rate profile | Fault-perpendicular velocity profile extraction and near-fault phase discontinuity measurement at 1–5 km along-strike intervals | CSV table of along-strike position vs. creep rate with 1-sigma uncertainty bounds |
| Locking depth estimate per fault segment | Savage-Burford screw-dislocation model fit to fault-perpendicular velocity profiles; grid-search or Bayesian inversion | PDF report with locking depth and creep rate posterior distributions per segment |
| Seismic moment deficit rate map | Subtraction of InSAR-derived surface creep from published geological or GPS plate-rate; elastic dislocation forward modelling | GeoTIFF moment deficit rate grid and tabulated maximum credible magnitude estimates per segment |
| Atmospheric delay correction assessment | ERA5 and GACOS tropospheric delay modelling; empirical phase-elevation correlation removal | Before/after interferogram comparison report; corrected velocity grid |
| Creep rate change detection (multi-epoch comparison) | Split time-series inversion comparing velocity fields from two consecutive periods to detect acceleration or deceleration | GeoTIFF velocity-difference map; alert flag if rate change exceeds 2 mm/yr between epochs |
| Coherence and data-gap assessment | Temporal coherence mapping across interferogram stack; identification of decorrelation-prone segments by land cover | Coherence mask GeoTIFF and written recommendation on sensor selection (C-band vs. L-band) per fault segment |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.