High Time Resolution Displacement Monitoring using a Multiple Stack Adaptive Multilooking InSAR Technique
Rabus, Bernhard; Eppler, Jayson; Sharma, Jayanti
MDA, CANADA

Advanced third-generation InSAR stack analysis techniques have been useful in providing time series of accurate high spatial resolution displacement risk maps for semi-urban as well as industrial (mining, oil and gas, pipeline) infrastructure in a variety of environments. Homogenous Distributed Scatterer (HDS)-InSAR recently developed by MDA is an example of these techniques, which operate on a single InSAR stack. HDS-InSAR exploits both persistent point scatterers (PS) and coherent distributed scatterers (DS) through adaptive phase filtering over statistically homogenous pixel neighborhoods, evaluated from the amplitude statistics of the InSAR stack. The technique further involves the use of temporal matched filters in the comprehensive iterative phase component modeling to separate long/short range, long-term/seasonal displacements, topographic error, static and dynamic atmospheres, and noise in the signal. The density of HDS exceeds that of PS by more than an order of magnitude, often providing essentially continuous high phase quality coverage over infrastructure and surroundings. The temporal resolution of HDS-InSAR is limited by the sensor repeat interval inducing a direct constraint when observing faster, variable displacements (e.g. monitoring pit/tailing stability for mining applications and rapidly deforming oil fields). An even more important indirect constraint derives from the fact that minimum stack thicknesses of about 15 images are generally needed for construction of the adaptive filter kernel to achieve accuracies of mm/yr. For RADARSAT-2 (repeat pass interval 24 days) this translates into >1 year of data acquisition before a new site can be monitored. The only way to shorten the lead in time to less than 15 x 24 days is through the simultaneous use of two or more, shorter interleaving InSAR data stacks in the HDS analysis. In this paper we describe and demonstrate a new multi incidence angle (MIA) version of HDS-InSAR that exploits the significant multi-acquisition diversity present for sensors with longer repeat pass interval such as RADARSAT. MIA HDS-InSAR allows optimum synergistic joint processing of two or more interleaving interferometric stacks of different incidence angle but same look direction. We show that the MIA concept boosts temporal resolution and reduces acquisition lead in time by a factor proportional to the number of stacks interleaved. A key feature of MIA HDS-InSAR is the joint determination of the HDS neighborhoods for each pixel after all SLC images have been projected accurately to a chosen master stack geometry. HDS neighborhoods represent adjacent pixels with an intensity distribution statistically similar above a certain threshold to that of the investigated pixel. To determine the HDS neighborhoods jointly the statistical probabilities in a maximum neighborhood square are averaged across stacks weighted by the number of acquisitions in each stack before the probability threshold is applied to define the HDS neighborhood. Interferograms for master and slave stacks are then adaptively filtered using the jointly estimated MIA-HDS neighborhoods. Projection of the neighbourhoods to the common RDC geometry of the master uses a novel rational function based high precision geocoding module. The alignment of the maximum neighborhood squares from different stacks depends on the accuracy of the initial topography; initial alignment is improved through iterative refinement of the projection after a first solution for the topographic error. The adaptively filtered (wrapped) interferograms are jointly modeled in MIA HDS-InSAR achieving a robust phase model of the topographic error and displacement components. Due to the different incidence angles for the multiple stacks only a 1D model of the topographic error would be strictly independent of knowing the displacement direction. Other than the joint estimation of the HDS neighborhoods (which depends on the topographic error) joint phase modeling in MIA is dependent on displacement direction, which we a-priori specify as either vertical or along the topographic fall line.
After phase unwrapping and SVD network inversion (done separately for each stack), and the joint estimation of topographic error and displacement rates, MIA HDS-InSAR combines results into a single time series which is temporally filtered for final MIA-HDS deformation traces. This step is equivalent to the Lowes filter of conventional HDS-InSAR, which already is optimally adapted to data points irregularly spaced in time (originally to deal with missing data points in a single interferometric repeat pass stack). Depending on the interleaving of the stacks the Lowess filter width can be reduced to achieve the higher temporal resolution of the output traces (for two stacks down to a minimum of half the satellite repeat interval, or 12 days for RADARSAT).
The prototype was tested against a pair of interleaving RADARSAT-2 Ultra-Fine interferometric data stacks (U4 and U24, Descending, with 15 degree incidence angle difference) over the Lost Hills Oilfield near Belridge, California. Oilfield subsidence in this area is characterized by high maximum surface displacement rates of 10-100 cm/yr with superimposed significant non-linear time series behavior due to steam injection for enhanced oil recovery. Both conventional HDS-InSAR and MIA HDS-InSAR are evaluated for acquisitions over the same short period. In the absence of ground truth we validate against a corresponding temporal subset of a "gold standard" time series that was created from a large number of scenes of one stack. U24 with greater incidence angle and correspondingly better ground resolution was used to create the "gold standard" solution and also provides the master track geometry for the MIA processing.
Increased phase quality of MIA HDS-InSAR from joint HDS neighbourhood formation and topographic error modelling is shown to translate into significantly higher density of reliable HDS points compared to the conventional HDS-InSAR result. We show that MIA HDS-InSAR obtains robust and accurate results with RADARSAT-2 for time periods as short as 7 repeat cycles that would be considered far insufficient for conventional HDS-InSAR or any other single-stack based technique.