Automated pipeline for DEM generation from SPOT-5 stereo imagery for glacier elevation change assessment

Accurate estimation of glacier elevation change is crucial for long-term geodetic mass balance and assessing glacier response to climate change. This study introduces an automated pipeline for generating Digital Elevation Models (DEMs) from satellite stereo imagery to quantify glacier elevation changes.  

We focus on SPOT-5 High-Resolution Stereoscopic (HRS) imagery, recently freely accessible through the SPOT World Heritage program by CNES. SPOT-5 is an underutilized archive with global coverage from 2002 to 2015 and stereoscopic capabilities, making it valuable for reconstructing glacier elevation changes and complementing stereo imageries from more recent satellites. However, its inherent challenges, such as limited radiometric resolution, rectangular pixel geometry, and absence of camera's Rational Polynomial Coefficient model, require specific attention. We apply our workflow to Hofsjökull, Iceland’s third-largest ice cap, because of extensive SPOT-5 temporal coverage, further complemented by ArcticDEM, SPOT-6 and Pleiades for recent years.

Our workflow addresses key steps in DEM generation such as stereo pair selection, bundle adjustment, stereo correlation, noise filtering, point cloud gridding, void filling, and co-registration. Each of these steps significantly affects DEM quality and glacier elevation change estimates. Therefore, we compare and evaluate various approaches to identify optimal solutions for automation. We benchmark open-source photogrammetry tools, including Ames Stereo Pipeline and MicMac, and geospatial libraries like xDEM, GeoUtils, and OPALS, integrating them for interoperability.

We tested different stereo-matching algorithms and found that the More Global Matching algorithm performs best for SPOT-5 data under diverse illumination and viewing conditions. For DEM gridding and void filling, we use a Robust Moving Planes fitting method in OPALS. Co-registration is performed using the globally available Copernicus DEM (GLO-30) as reference, using appropriate masks to exclude glaciers, forests, water bodies and steep areas. The least-squares template matching algorithm implemented in OPALS enhances alignment accuracy by estimating full affine transformations, while sub-pixel refinement is achieved with the Nuth and Kääb method. Finally, we derive elevation-band-based trends from spaceborne DEM time series to extrapolate elevation changes over decadal intervals. This enables us to calculate area-weighted mean elevation change estimates for each glacier and the entire ice cap over defined periods.

This study contributes to the Glacier Mass Balance Intercomparison Exercise (GlaMBIE) by advancing scalable, open-access methodologies for glacier elevation change assessments. Additionally, our systematic comparison and integration of algorithms and techniques for each stage ensures optimized performance, making the pipeline reproducible across regions, temporal scales, and satellite platforms.