Subpixel offset tracking for landslide deformation monitoring: optimization of input conditions and assessment of vegetation noise

Abstract

Surface deformation analysis is a crucial task in landslide research, as it plays an important role in landslide monitoring and serves as a prerequisite for landslide mechanism analyses and risk assessments. High-accuracy surface displacement fields can be derived rapidly by the subpixel offset tracking (sPOT) algorithm based on multitemporal very-high-resolution (VHR) remote sensing data without tedious manual interpretation work. However, some technical characteristics of the availability of sPOT, such as the input conditions and vegetation noise, have not been fully discussed thus far, and these factors may compromise the accuracy of the derived surface displacement fields and thus limit the applicability of sPOT as a high-accuracy image interpretation algorithm. In this study, several quantitative indices from information theory and digital image processing methods were introduced to support the quantitative analysis of the sPOT inputs and outputs. We used two sets of tests to show how different input conditions (including different image band and window parameters) affect the accuracy of the displacement field. Areas with different vegetation statuses in the region of interest (RoI) were then segmented using pattern recognition methods, and the noise level in each area was successfully quantified. Finally, some practical guidelines on the selection of input conditions and filtering of vegetation noise were proposed. The results of this study help to improve the accuracy of the sPOT algorithm from a user’s perspective and are expected to contribute to the in-depth application of sPOT in various fields of geomorpho-dynamics, including landslide deformation monitoring.