Assessing Stereo DEM Quality under Large Resolution Mismatch: Results from Galileo Europa Imagery

INTRODUCTION

Stereo photogrammetry has been extensively applied to reconstruct the topography of various planetary bodies. Its application yields a digital representation of surface topography, typically referred to as a Digital Elevation Model (DEM), whose quality determines how reliably it can be employed for subsequent geophysical and geological investigations. The successful application of stereogrammetry primarily depends on appropriate viewing conditions from the cameras and, secondarily, on the geometric and illumination characteristics of the images [1]. Aside from the perspective-related differences required to generate sufficient parallax, images should ideally be as similar as possible, since most stereo algorithms rely on feature matching between two images. Such ideal conditions, however, are not always met, especially for outer Solar System bodies, which often present heterogeneous datasets [2].

In particular, large resolution mismatches between stereo images, common in “serendipitous” observations from multi-flyby missions, constitute one of the main complications in the generation of reliable stereo DEMs, because features resolved in one image may appear blurred or absent in the other. To reduce mismatch and enhance similarity between the two stereo images, a commonly adopted approach is to leave the lower-resolution image as it is and downsample the higher-resolution image. In addition to improved similarity, such an approach also minimises computational effort, which was particularly valuable in the past given the technological limitations. However, downsampling inevitably suppresses high-frequency spatial information, potentially reducing the effective resolution of the resulting stereo DEMs. Conversely, upsampling the lower-resolution image preserves the native detail of the finer image but may increase matching ambiguity and noise-related artefacts. Computational cost is generally higher in this case, though with current technology this is not an important concern for small images such as the Galileo Solid-State Imaging (SSI) data [3] employed here.

The relative performance of these opposite strategies for stereopairs with substantial resolution mismatch remains poorly constrained. Here we present results from an ongoing investigation of how image resampling strategies and stereo-processing parameters affect DEM quality using Galileo SSI data of Conamara Chaos on Europa. The study is motivated by the need to identify optimal settings for the future generation of reliable and high-quality DEMs, maximise our topographic knowledge of the Jovian moons using existing image datasets, and provide useful constraints for stereo-planning activities relevant to the JUICE (ESA) and Europa Clipper (NASA) missions.

METHODS

Topographic reconstruction was performed using the Ames Stereo Pipeline (ASP) [4] on Galileo SSI image pairs covering Conamara Chaos. Two categories of stereopairs were analysed: (1) high-resolution pairs, used to generate reference DEMs, and (2) mismatched pairs, combining one high-resolution image with a lower-resolution frame.

Two opposite resampling approaches were systematically explored, one involving upsampling of the lower-resolution image and the other involving downsampling of the higher-resolution image. Additionally, stereo-processing parameters were systematically explored through hierarchical testing of different parameter combinations, consistent with methodologies adopted in previous studies [5-6].

DEM quality was evaluated quantitatively through a smoothing-based approach adapted from previous studies on Martian DEMs [6], using RMS deviation analyses to estimate the horizontal resolution and vertical precision of the target DEMs. Additional qualitative inspection of hillshaded DEMs, error maps, and topographic profiles was also performed.

RESULTS

The analysis indicates that significantly different stereo DEMs can be generated from the same set of images depending on how the images are pre-processed and how the stereo software is configured. Some parameters appear to exert a stronger control than others on the balance between spatial detail, DEM smoothness, and relative vertical precision.

Systematic differences in DEM quality are primarily associated with the adopted resampling strategy and the image ordering within the stereo pair (because the matching process does not treat the two images symmetrically) and, secondarily, with the stereo-processing parameters, particularly the size of the matching kernels. Under appropriate tuning, fine-sampling approaches appear capable of preserving geomorphological detail more effectively without excessively penalising vertical precision, particularly in rough-textured terrains and across narrow morphologic features. In contrast, coarse-sampling configurations generally produce smoother but less detailed topography.

These results suggest that preserving the native detail of higher-resolution images through fine-sampling approaches may be advantageous even in the presence of substantial resolution mismatches, provided that appropriate stereo configurations are adopted. Future work will further quantify these relationships, establish practical guidelines for optimising stereo photogrammetry under limited and heterogeneous imaging conditions, and provide additional constraints for planning the acquisition of image pairs suitable for stereogrammetry during the JUICE and Europa Clipper missions.

ACKNOWLEDGMENTS

G.C. and G.M. acknowledge support from the Italian Space Agency (2023-6-HH.0).

REFERENCES

[1] Becker, K. J. et al. (2015) 46^th LPSC, 2703.

[2] Schenk, P. M. (2008) ISPRS Archives, XXXVII.

[3] Belton, M. J. S. (1992) Space Sci. Rev., 60, 413-455.

[4] Beyer, R. A. et al. (2018) Earth Space Sci., 5, 537-548.

[5] Bland, M. T. et al. (2021) Remote Sens., 13, 5097.

[6] Kirk, R. L. et al. (2021) Remote Sens., 13, 3511.