Towards 3D InSAR time series displacement with Kalman filtering - Feasibility and theoretical accuracy

Li-Chieh J. Lin, & Gareth J. Funning

Submitted August 30, 2026, SCEC Contribution #15597, 2026 SCEC Annual Meeting Poster #TBD

InSAR velocity products have been widely used in various fields to reveal long-term deformation associated with different processes. If sufficient information from different radar lines of sight (LOS) is available, we may decompose such velocities into a 3D InSAR velocity field to capture the full details of the surface deformation, and to infer the underlying mechanisms. However, studies using GNSS measurements show that the deformation is complex in time. Also, it has been suggested that the deformation signals are often distributed in space. Thus, a 3D time series deformation that expands spatially is required to capture the complex surface deformation. Due to the density of GNSS stations, it is often difficult to discern the spatial pattern of deformation. With modern satellite SAR missions and the aid of UAVSAR which offers various LOS directions, it is possible to identify the complex 3D deformation time series using such InSAR data.

Here we aim to produce 3D InSAR displacement time series by using Kalman filtering. We first generate synthetic deformation data using Okada’s model and project the synthetics into the LOS directions of Sentinel-1, NISAR and UAVSAR. Next, we add seasonal deformation and correlated and uncorrelated noise to these time series to simulate real-world conditions. We generate seasonal deformation signals using the sine and cosine functions in the GNSS trajectory model and make the vertical signal 5 times larger than the horizontal. For the noise component, we generate both realistic spatially correlated and uncorrelated noise to add to each LOS dataset. Lastly, we apply Kalman filtering to different LOS combinations time series to decompose them into 3D displacements. Here we show the accuracy (cm/day) of the three directions of possible situations one would encounter.

2✕Sentinel-1: 0.07 cm/day (EW), 1.56 cm/day (NS), 0.22 cm/day (Vertical)
2✕NISAR: 0.05 cm/day (EW), 1.38 cm/day (NS), 0.18 cm/day (Vertical)
2✕Sentinel-1 + 2✕NISAR: 0.02 cm/day (EW), 0.10 cm/day (NS), 0.01 cm/day (Vertical)
2✕Sentinel-1 + 2✕NISAR + 3✕UAVSAR: 0.03 cm/day (EW), 0.06 cm/day (NS), 0.04 cm/day (Vertical)

The study shows the feasibility of how well the 3D time series displacement could be resolved using Kalman filtering. In the case of the central San Andreas fault area, the maximum number of LOS measurements one can obtain is 2 Sentinel-1, 2 NISAR and 3 UAVSAR, which suggests the potential for future time-dependent modeling.

Key Words
Kalman filter, 3D InSAR time series

Citation
Lin, L. J., & Funning, G. J. (2026, 08). Towards 3D InSAR time series displacement with Kalman filtering - Feasibility and theoretical accuracy. Poster Presentation at 2026 SCEC Annual Meeting.

Related Projects & Working Groups
Tectonic Geodesy