On the influence of initial guess DEM for gcp-less digital elevation model extraction using photogrammetry on SPOT-6/7 tristereo imagery - A case study of South Sulawesi
Zylshal Zylshal *1, Parwati Sofan 1, Indri Pratiwi J 2, Athar Abdurrahman Bayanuddin 34

1 Research Center for Geoinformatics, National Research and Innovation Agency (BRIN), Bandung, Indonesia

2 Bureau for Public Communication, General Affairs, and Secretariat, National Research and Innovation Agency (BRIN), Jakarta, Indonesia

3 Directorate of Laboratory Management, Research Facilities, and Science and Technology Park, National Research and Innovation Agency (BRIN), Jakarta, Indonesia

4 Remote Sensing Graduate Study Program, Universitas Gadjah Mada, Indonesia

Abstract

Digital Elevation Model (DEM) generation from satellite photogrammetry is a well-established method for extracting topographic information from optical stereo imagery. With the increasing agility of modern satellites, tri-stereo acquisitions (SPOT-6/7, Worldview-3) are now commonly available. Although the use of Ground Control Points (GCPs) is standard for ensuring absolute accuracy, their collection is costly, time-consuming, and labor-intensive. To address these limitations, GCP-less DEM extraction approaches have been developed, relying on Rational Polynomial Coefficients (RPCs) and an initial seed DEM to constrain the object-space geometry. This study evaluates the influence of three different seed DEMs-ALOS World 3D (AW3D30), FABDEM, and the Indonesia-specific DEMNAS-on GCP-less DEM extraction using an iterative bundle adjustment approach on SPOT-6/7 panchromatic tri-stereo imagery. Three test sites in South Sulawesi were selected to represent varied landform and land cover characteristics. All DEMs were converted to an ellipsoidal vertical datum for consistency. A total of 27 DEMs were generated at a 6-meter ground sampling distance (GSD) and evaluated based on visual quality, horizontal alignment, and vertical accuracy. Vertical validation was conducted using 86 independent control points (ICPs) obtained through GNSS field surveys. Most outputs showed good visual fidelity, though some contained voids from cloud cover. Horizontal accuracy remained within one pixel across all outputs. Vertical Root Mean Square Error (RMSE) ranged from 1.18 to 4.79 meters. DEMNAS performed best as an initial seed DEM for GCP-less processing, with mean shift magnitude of 0.42 meters (x-direction) and 0.42 meters (y-direction). It also produced the lowest average vertical error, with RMSE, Mean Absolute Error (MAE), and Normalized Median Absolute Deviation (NMAD) of 4 m, 2.57 m, and 2.14 m, respectively. Given the limited availability of DEMNAS, AW3D30 presents a viable globally available alternative. Further testing with additional seed DEMs is recommended to enhance the generalizability of this GCP-less approach.

Keywords: DEM, GCP-less, SPOT6, Photogrammetry, FABDEM

Topic: Topic A: General Remote Sensing