Kalman Filter Application to GBSAR Interferometry for Slope Monitoring

Abstract

The use of Kalman filtering techniques for landslide monitoring has proved effective as a tool for estimating and predicting land displacements. Ground-Based Synthetic Aperture Radars (GBSAR) are popular remote sensing instruments able to provide displacement maps of the investigated area, with submillimeter precision. These instruments outperform other sensors in several respects, such as all-weather and all-day monitoring. However, in some cases, for instance in vegetated scenarios, the displacement is affected by a significant uncertainty due to the decorrelation of the radar signal. In such a case, to retrieve any reliable information, noise must be filtered out using appropriate methods. Given the success of kinematic modeling of landslide movements through Kalman filtering, this technique seems to be the optimal candidate for processing the displacement measured by interferometric GBSAR data. This paper investigates this idea, by applying Kalman Filter to GBSAR measurements acquired in two different campaigns: a landslide monitoring in north Spain, and a sliding glacier monitoring in the Alpes, Italy. A proper initialization of the algorithm parameters is fundamental for a correct application of the Kalman filter. In this work, we present a strategy that exploits information from coherent pixels for tuning the filtering parameters and optimizing the filter performance on areas with low coherence.

A Reliable Observation Point Selection Method for GB-SAR in Low-Coherence Areas

“…GB-SAR demands greater stability from observation points, as phase instability can lead to unwrapped phase errors, thereby substantially impacting the accuracy of measurement outcomes. Filtering techniques utilized to mitigate the issue of high noise in low-coherence scenarios fall into two main categories: spatial filtering [35][36][37] and temporal filtering [38,39]. In spatial filtering scenarios characterized by relatively high noise levels, the high coherence point phase is susceptible to the influence of the surrounding random phase.…”

“…However, effectively filtering Gaussian noise and mutation by frequency domain filters is challenging. The Kalman filtering [38] in time-domain filtering methods proves effective, but it is difficult to adaptively optimize observation points with different stability in the detection area. Therefore, a methodology is needed to solve the problems encountered in the low-coherence areas.…”