Modeling the Impact of Forest Disturbance on InSAR Time Series: Synthetic Models to Real Data Analysis

Frequent and extensive tree loss in regions such as the Pacific Northwest (PNW) can introduce time-variable signals into InSAR time series, which could be misinterpreted as ground displacements. InSAR observations provide valuable constraints on a wide variety of processes and hazards that occur in the PNW, such as landslides, seismicity, volcanism, and anthropogenic activity. However, InSAR could allow for the examination of even smaller rates if the impact of forest disturbance can be mitigated. To address this issue, we first use synthetic models simulating 1) Sentinel-1/NiSAR and 2) ALOS-1 orbital geometries to constrain the impact of forest disturbance on InSAR time series. We have reformulated our time series approach to incorporate correlations between data acquisition dates, which has decreased the predicted error of the Sentinel-1-style synthetic tests, and increased the predicted error in the ALOS-1-style tests. The Sentinel-1 synthetic tests (which use relatively small, randomly distributed satellite orbits) suggest that typical errors due to forest disturbance will be 0.2 cm/yr and 10 cm/yr for 1-year secular and time-variable time series, respectively. For the ALOS-1 synthetic tests (which use actual sets of orbital geometries), we find errors on the order of 10 cm/yr with a pronounced bias associated with the ALOS-1 orbits.

The effects of time-variable forest disturbances can be mitigated if their timing is known. We illustrate this using real ALOS-1 data near Eugene, OR, where we invert for average velocity using three different approaches with a varying degree of satellite orbital geometry (baseline-dependent) effects: 1. Baseline-independent time series, 2. Constant baseline-dependent time series 3. Variable baseline-dependent time series informed by a priori forest disturbance data from Landsat optical observations, in which tree loss can be clearly identified. The mean difference in the secular velocities between a bare and deforested region for each of the 3 time series inversions above are V avg = 1.8 cm/yr, 11.9 cm/yr, and 0.01 cm/yr, respectively. This demonstrates that the inclusion of a priori information reduces the potential for misinterpretation of InSAR signals associated with forest disturbance and other types of land surface change as ground deformation.