Exploitation of ARIMA and Annual Variations Model for SAR Interferometry Over Permafrost Scenarios

Abstract

The temperature change is expected to induce nonlinear characteristics in the annual fluctuation of permafrost. The interferometric synthetic aperture radar (InSAR) technique has demonstrated its efficacy in capturing such variations by monitoring surface deformation over time. However, turbulent atmospheric phase noise often requires spatiotemporal filtering, resulting in a loss of temporal resolution for nonlinear signals. Furthermore, the influence of interannual temperature variations on annual freeze–thaw cycles has not been fully integrated into InSAR modeling thus far. In this study, we propose a methodology to enhance the effectiveness of InSAR time-series analysis in permafrost environments. Diverging from conventional filtering methods where the temporal resolution loss depends on the size of the convolution kernel, we introduce the autoregressive integrated moving average model to extract the nonlinear deformation signal component. Additionally, we derive parameters associated with annual variations from the time-series deformation data during InSAR permafrost modeling. Through synthetic data experiments incorporating various noise delays, we observe a considerable improvement in accuracy, ranging from 27.8% to 55.3% in nonlinear time-series deformation analysis. Leveraging Sentinel-1 datasets from 2017 to 2021 alongside ground truth data from northern Alaska, we ascertain an enhancement of over 22% in the accuracy of time-series deformation estimation. Furthermore, incorporating annual variations enhances the accuracy of active layer thickness estimation. Our methodology reveals a strong correlation between residual deformations and soil moisture content, shedding light on the pivotal role of soil moisture in permafrost thawing processes.