Laboratoire d'Informatique de Grenoble Équipe Ingénierie de l'Interaction Humain-Machine

Équipe Ingénierie de l'Interaction

InSarViz, an open source interactive visualization tool for satellite SAR interferometry

In European Spatial Agency Living Planet Symposium. 2022.

Margaux Mouchené, Renaud Blanch, Erwan Pathier, Franck Thollard

Poster présenté à :


Satellite SAR interferometry (InSAR) is a well-established technique in Earth Observation that is able to monitor ground displacement with a high precision (up to mm/year), combining high spatial resolution (up to a few m) and large coverage capabilities (up to continental scale) with a temporal resolution from a few days to a few weeks. It is used to study a wide range of phenomena (e.g. earthquakes, landslides, permafrost, volcanoes, glaciers dynamics, subsidence, building and infrastructure deformation, etc.).

For several reasons (data availability, non-intuitive radar image geometry, complexity of the processing, etc.), InSAR has long remained a niche technology and few free open-source tools have been dedicated to it compared to the widely-used multi-purposes optical imagery. Most tools are focused on data processing (e.g. ROI_PAC, DORIS, GMTSAR, StaMPS, ISCE, NSBAS, OTB, SNAP, LICSBAS), but very few are tailored to the specific visualization needs of the different InSAR products (interferograms, network of interferograms, datacube of InSAR time-series). Similarly, generic remote-sensing or GIS software like QGIS are also limited when used with InSAR data. Some visualization tools with dedicated InSAR functionality like the pioneer MDX software (provided by the Jet Propulsion Lab, were designed to visualize a single radar image or interferogram, but not large datasets. The ESA SNAP toolbox also offers nice additional features to switch from radar to ground geometry.

However, new spatial missions, like the Sentinel-1 mission of the European program COPERNICUS with a systematic background acquisition strategy and an open data policy, provide unprecedented access to massive SAR data sets. Those new datasets allow to generate a network of thousands of interferograms over a same area, from which time-serie analysis results in spatio-temporal data cube: a layer of this data cube is a 2D map that contains the displacement of each pixel of an image relative to the same pixel in the reference date image. A typical data cube size is 4000x6000x200, where 4000x6000 are the spatial dimensions (pixels) and 200 is a typical number of images taken since the beginning of the mission (2014). The aforementioned tools are not suited to manage such large and multifaceted datasets.
In particular, fluid and interactive data visualization of large, multidimensional datasets is non-trivial. If data cube visualization is a more generic problem and an active research topic in EO and beyond, some specifics of InSAR (radar geometry, wrapped phase, relative measurement in space and in time, multiple types of products useful for interpretation…) call for a new, dedicated visualization tool.
We started the InSARviz project with a survey of expert users in the French InSAR community covering different application domains (earthquake, volcano, landslides), and we identified a strong need for an application that allows to navigate interactively in spatio-temporal data cubes.

Some of the requirements for the tools are generic (e.g., handling of big dataset, flexibility with respect to the input formats, smooth and user-driven navigation along the cube dimensions) and other more specific (relative comparison between points at different location, selection of a set of pixels and the simultaneous vizualisation of their behavior in both time and space, visualization of the data in radar and ground geometries…)

To meet those needs we designed the InSARViz application with the following characteristics:
- A standalone application that takes advantage of the hardware (i.e. GPU, SSD hard drive, capability to run on cluster as a standalone application). We choose the Python language for its well-known advantages (interpreted language, readable, large community) and we use QT for the graphical user interface and OpenGL for the hardware graphical acceleration.
- Using the GDAL library to load the data. This will allow to handle all the input formats that are managed by GDAL (e.g. GeoTIFF). Moreover, we designed a plug-in strategy that allows users to easily manage their own custom data formats.
- We take advantage of Python/QT/OpenGL stack that ensures efficient user interaction with the data. For example, the temporal displacement profile of a point is drawn on the fly while the mouse is hovering over the corresponding pixel. The “on the fly” feature allows the user to identify points of interest. The user can then enter another mode in which they can select a set of points. The application will then draw the temporal profiles of the selected points, allowing a comparison of their behavior in time. This feature can be used when studying earthquakes as users can select points across a fault, allowing to have a general view of the behavior of the phenomenon at different places and times.
- Multiple windows design allows the user to visualize at the same time data in radar geometry and in standard map projection, and also to localize a zoomed-in area on the global map. A layer management system is provided to quickly access files and their metadata.
- Visualization tools commonly use aggregation methods (like e.g. smoothing, averaging, clustering) to drastically accelerate image display, but they thus induce observation and interpretation biases that are detrimental to the user. To avoid those bias, the tool focuses on keeping true to the original data and allowing the user to customize the rendering manually (colorscale, outliers selection, level-of-detail)
In our road map, we also plan to develop a new functionality to visualize interactively a network of interferograms.

We plan to demonstrate the capabilities of the InSARviz tool during the symposium.
The InSARviz project was supported by CNES, focused on SENTINEL1, and CNRS.