On the remote island of Tristan da Cunha, high-resolution satellite and drone imagery will be integrated to create an accurate digital elevation model.
Medium-resolution global digital elevation models (DEMs), such as SRTM and ASTER GDEM, have supported geoscience research for more than 20 years. However, many environmental and hazard-related applications, such as landslide risk, flooding, and surface process modeling, require more detailed and accurate topographic data. Airborne LiDAR or photogrammetry flights can achieve sub-decimeter accuracy, but are expensive and logistically difficult or impossible to use in remote environments. As a result, high-resolution DEMs derived from very high-resolution (VHR) satellite imagery and low-altitude drone imagery are becoming increasingly important as an alternative.
Sensor characteristics and terminology
Although the spatial resolution terminology of ground sampling distance (GSD) varies by domain, the following definition is adopted in this study:
Low resolution: >30m Medium resolution: 30 to 5m High resolution: 5 to 1m Very high resolution (VHR): <1m
Drone imagery with a GSD approximately one order of magnitude finer than the VHR satellite is treated as a separate ultra-high resolution category.
Despite medium-resolution spectral imagery from Landsat or Sentinel missions, global DEMs still rely heavily on SRTM and ASTER, both of which contain artifacts, voids, and alignment issues. This motivated the creation of improved products, such as the void-free 90 million DEM and commercial datasets like Airbus WorldDEM. Commercial DEMs offer higher accuracy but are still expensive.
Evolution of satellite-derived DEM
Stereo DEM extraction from spaceborne images began with cross-track SPOT acquisition. Modern satellites (such as WorldView 2/3) now offer in-track stereo, more favorable acquisition geometries, and panchromatic GSDs as fine as 0.3 m. It also includes high-precision orbit/attitude information, providing absolute geolocation around 5 meters AD. Combined with powerful algorithms such as semi-global matching (SGM), it can generate dense and detailed surface models.
Advances and limitations of drone photogrammetry
Low-cost drones equipped with civilian cameras and GNSS/IMU systems have become commonplace for mapping small areas. Although these systems produce very dense point clouds (hundreds of points per square meter), absolute positioning remains weak due to uncorrected GNSS and unstable camera calibration. Strong image block geometry and ground control points (GCPs) are essential for reliable altitude accuracy.
Research goals and background
This study investigates how VHR satellite imagery and drone data can be integrated to create an accurate high-resolution DEM of Tristan da Cunha, one of the world’s most isolated inhabited islands. The island is so far away from major continents that it cannot be mapped using traditional aerial photogrammetry. Additionally, harsh weather, steep terrain, and frequent clouds complicate acquisition for both satellites and drones. A newly established geodetic control network on the island within the only settlement provided the precise ground reference necessary for accurate DEM generation.
Data and research sites
Tristan da Cunha is a roughly circular island, approximately 12 km in diameter and approximately 96 km² in area, rising steeply from sea level to Queen Mary’s Peak, 2062 m above sea level. It is located in the middle of the Atlantic Ocean and is not accessible by plane. The available datasets include:
SRTM and ASTER DEM (both contain noise and voids) DigitalGlobe/Maxar VHR archives (QuickBird, WV2, WV3) Drone imagery from two DJI Phantom 3 missions High precision GCP from GNSS/Total Station survey
SRTM and ASTER were fused using BAE’s terrain merging method to generate a medium resolution DEM (mDEM), which was later used as initial seed data for stereo matching. Of the 94 satellite scenes, 3 were selected as core “image triplets” and a further 10 were selected for residential areas based on cloud cover, radiation quality, viewing geometry, and base height ratio (~0.6).
The drone flights produced 373 images with strong geometric diversity. Nine precisely measured GCPs ensure accurate georeferencing.
methodology
satellite processing
The photogrammetry workflow was performed separately with three commercial packages:
• BAE SocetGXP
• Hexagon Eldas Imagine
• PCI Geomatica
Common processes include tie point generation, triangulation, and terrain extraction with NCC, NGATE, ATE, or SGM algorithms. Matching was performed hierarchically, starting from a 30m seed DEM and iteratively adjusting to 10m, 5m, 2m, and 1m GSDs.
drone processing
Pix4D Mapper handled self-calibration, triangulation, and point cloud generation. Multiple GCP configurations were tested.
data fusion
Spaceborne point clouds and drone point clouds were integrated by direct geodetic alignment, and ICP-based comparisons were used for validation only.
result
Satellite triangulation and DEM quality
The WV2 and WV3 scenes showed excellent initial geolocation, with residuals less than 0.5 m, better than published specifications. Only one GCP and checkpoint was required to coordinate the blocks.
DEM performance varies as follows:
• SocetGXP ATE: Cleanest 2m DEM
• SocetGXP NGATE: Dense but noisy 1m DEM
• Erdas eATE: lower noise, comparable quality
• SGM (Erdas): Very dense (~1m spacing) but requires strong noise filtering
• PCI Geomatica: High quality 2m DEM from the best stereo pair. Adding more images increases noise due to temporal differences.
Drone triangulation and accuracy
Without GCP, drone GNSS provided strong area measurements but poor height accuracy. Nine GCPs had centimeter-level residue reductions for all components. The dense point cloud reached >500 points/m² in the port area, but slightly less within the settlement.
Integration and comparison
Geodetic control enables near-perfect alignment between satellite and drone-derived data without rotation or scale correction. The ICP showed that:
• Average offset: 0.26m → 0.14m after minor movement
• Height difference: average 0.01 meters, typical 0.22 meters. developer.
The integrated product will cover the island with a resolution of approximately 1-2 m, with port details displayed at the centimeter level.
conclusion
Research has demonstrated that:
1. VHR satellite imagery, if carefully selected and processed, can reliably replace aerial mapping in remote areas.
2. Image selection (shape, radiometry, cloud cover, temporal proximity) is more important than matching algorithm selection.
3. Drone photogrammetry provides extremely fine local details when supported by high-quality GCPs and robust block geometry.
4. Combining satellite and drone data produces a complete and accurate multiscale DEM, even under harsh environmental conditions.
Future work should leverage multi-view stereo across a larger satellite subset for increased completeness in steep terrain and use multi-temporal satellite archives for hazard monitoring and change detection. The drone continues to provide ultra-high resolution details when needed.
understand
This project was funded by the University of Luxembourg and carried out by Dr. Dietmar Bax and Professor Felicia Teferle. The full text of this Space contribution can be accessed at https://doi.org/10.3389/feart.2020.00319 and forms part of Dr. Backes’ doctoral thesis.
Please note: This is a commercial profile
This article will be published in an upcoming issue of Special Focus Publication.
Source link
