Improved reconstruction of surface displacements through time-series inversion and seasonal bias correction
This repository contains scripts and examples for reconstructing a displacement signal from a network of pairwise displacement measurements through time-series inversion. Inversion is part of the processing chain for deriving a continuous time series from satellite-based measurements of displacements over landslides, glaciers, dunes or other Earth-surface processes. Prerequisite for the inversion are several temporally overlapping displacement fields obtained from image cross-correlation. The inversion process returns a multi-band raster with the cumulative displacement estimated at each time step. In addition, this repository provides functionalities to correct for seasonal bias which is common in cross-season image pairs of mountainous terrain and presents a challenge for the identification of seasonally driven displacent.
Please refer to demo.ipynb for a full walk-through of the time-series inversion process. All core functionality is implemented in timeseries_inversion.py.
In addition, there are several notebooks that explore the effect of errors, network structure and sampling interval on the inversion results based on synthetic data. These can be found in the artificial_examples folder.
To install all necessary Python packages, create a new environment using conda and the provided environment.yml file:
conda env create -f environment.yml
conda activate ts_inversion
This repository is associated with:
Mueting, A., Charrier, L., and Bookhagen, B.: Challenges in reconstructing seasonally driven landslide motion from optical satellite data: insights from the Del Medio catchment, NW Argentina (in prep.)