Sub-metre Lunar DEM Generation and Validation from Chandrayaan-2 OHRC Multi-View Imagery Using an Open-Source Pipeline

Abstract: High-resolution digital elevation models (DEMs) of the lunar surface are essential for surface mobility planning, landing site characterization, and planetary science. The Orbiter High Resolution Camera (OHRC) on board Chandrayaan-2 has the best ground sampling capabilities of any lunar orbital imaging currently in use by acquiring panchromatic imagery at a resolution of roughly 20-30 cm per pixel. This work presents, for the first time, the generation of sub-metre DEMs from OHRC multi-view imagery using an exclusively open-source pipeline. Candidate stereo pairs are identified from non-paired OHRC archives through geometric analysis of image metadata, employing baseline-to-height (B/H) ratio computation and convergence angle estimation. Dense stereo correspondence and ray triangulation are then applied to generate point clouds, which are gridded into DEMs at effective spatial resolutions between approximately 24 and 54 cm across five geographically distributed lunar sites. Absolute elevation consistency is established through Iterative Closest Point (ICP) alignment against Lunar Reconnaissance Orbiter Narrow Angle Camera (NAC) Digital Terrain Models, followed by constant-bias offset correction. Validation against NAC reference terrain yields a vertical RMSE of 5.85 m (at native OHRC resolution), and a horizontal accuracy of less than 30 cm assessed by planimetric feature matching.

• The paper introduces a robust open-source pipeline that generates sub-metre lunar DEMs from multi-view Chandrayaan-2 OHRC imagery.
• The methodology integrates custom PDS4 data ingestion, SPICE-based camera models, and stereo geometric analysis achieving horizontal accuracy <30 cm and vertical RMSE ~6 m.
• Implications include enhanced lunar mapping for mission planning and hazard analysis, paving the way for advanced machine learning-based matching techniques.

Sub-Metre Lunar DEM Generation from Chandrayaan-2 OHRC Multi-View Imagery with an Open-Source Pipeline

Introduction

This work introduces a robust open-source photogrammetric pipeline for generating sub-metre lunar digital elevation models (DEMs) from multi-view Chandrayaan-2 Orbiter High Resolution Camera (OHRC) imagery. The OHRC instrument provides unprecedented panchromatic coverage of the lunar surface at 20–30 cm GSD, representing a substantial improvement over widely used LRO NAC stereo-derived DEMs constrained to 1–5 m posting. As the OHRC data are not distributed as explicit stereo pairs and their ingestion is unsupported in default planetary stereo pipelines, two primary technical innovations were required: (a) development of an open-source import and sensor model configuration for OHRC’s PDS4 data products, and (b) algorithmic identification and ranking of viable multi-temporal stereo combinations based on geometric quality—specifically, baseline-to-height (B/H) ratio and convergence angle.

Pipeline Overview

The pipeline follows a rigorous photogrammetric workflow (Figure 1). OHRC PDS4 data ingestion is enabled through a custom-developed import template and Community Sensor Model (CSM) configuration, compatible with ISIS and the NASA Ames Stereo Pipeline (ASP). SPICE kernel acquisition provides geometric initialisation for spacecraft ephemeris and attitude, enabling accurate camera models via the ALE (Abstraction Layer for Ephemerides) framework.

Figure 1: OHRC DEM generation pipeline, from raw multi-view imagery and metadata ingestion to final DEM mosaic via camera geometry, bundle adjustment, stereo correspondence, surface reconstruction, and void filling.

Candidate stereo pairs are selected through geometric analysis, computing B/H ratios and convergence angles from SPICE and image metadata. Bundle adjustment (BA) performs joint refinement of camera parameters and tie-point clouds to reduce residual geometric inconsistency, robustified via a Cauchy loss. Dense stereo matching employs normalised cross-correlation-based block matching, followed by 3D point cloud triangulation. DEM interpolation and gridding produce regular raster DEMs for each site, which are aligned via ICP to NAC DTM references and further corrected for mean vertical offset. Where matching fails (in shadows or textureless regions), NAC elevation is used to fill holes, and overlapping products are blended using priority-based mosaicing.

Data and Experimental Regions

Five OHRC stereo pairs were identified, spanning diverse lunar locations and acquisition geometries. Respective B/H ratios range from 0.396 to 1.161, with corresponding convergence angles from approximately $22^\circ$ to $61^\circ$, probing both the theoretically optimal and limit regimes of lunar photogrammetric stereo geometry.

Generated DEMs achieve grid resolutions between 24 and 54 cm, with sub-metre morphological detail clearly preserved (Figure 2). NAC coverage limitations restrict validation for certain regions; Region 5 demonstrates the geometric/photometric limits of high B/H, while Regions 1–4 provide comprehensive accuracy and completeness validation.

Figure 2: DEMs at 24–54 cm resolution for four representative lunar sites, illustrating spatial completeness and high morphometric fidelity; voids and NAC infill are region-dependent.

Quantitative Validation

DEM vertical accuracy is validated via terrain profile comparison and residual analysis against co-located NAC DTMs (Figures 5, 6, 7, 8). For Regions 1, 2, and 4, vertical RMSE ranges from 3.0–8.5 m with a mean RMSE of 5.85 m (native OHRC grid). These values reflect excellent internal consistency and high external accuracy given the differences in posting and the propagation of SPICE kernel errors.

Figure 3: Profile comparison for Region 4: OHRC DEM (red) matches NAC DTM reference (green) with high fidelity; minor discrepancies are attributed to SPICE-based vertical bias.

Figure 4: Region 1 elevation profile; the sub-30 cm OHRC DEM preserves all NAC-resolvable features and reveals finer surface slopes.

Figure 5: Region 2 profile shows consistent alignment and suggests robust elevation modeling even beyond the equatorial zone.

Figure 6: Detail profile for Region 2; even fine-scale features are accurately recovered by the OHRC DEM.

Horizontal accuracy is assessed via planimetric feature matching, with a maximum mean horizontal offset below 30 cm for all validated regions and a mean triangulation error of 0.21 ± 0.16 m, demonstrating coherence with OHRC’s native sampling.

Computational Performance

On high-memory, many-core workstations, ingestion, geometry initialisation, and stereo pair analysis require only minutes, while bundle adjustment and dense stereo dominate wall-clock time. DEM gridding and post-processing scale sublinearly with output resolution and area. Total pipeline time per region is 1–3 hours, establishing the computational feasibility of large-scale high-resolution DEM production.

Analysis of Stereo Geometry Limits and Algorithmic Tradeoffs

The results substantiate the critical role of stereo geometry in lunar DEM generation. Usable B/H is bounded above by photo-consistency constraints: Region 5, with B/H ≈ 1.16 and convergence angle of $61^\circ$, suffers significant void fraction despite exceptionally fine nominal gridding (Figure 7), confirming a practical upper bound at B/H ≈ 0.9 for OHRC-based lunar stereo. Photometric and shadow mismatches dominate failure modes at high B/H, and further progress will demand more advanced learned matching algorithms.

Figure 7: High-void-fraction DEM for Region 5; high B/H yields theoretical precision but practical matching failure due to radiometric disparity and shadowing.

ICP-informed bundle adjustment and iterative refinement loops further enhance spatial consistency, particularly in hole recovery and alignment with absolute reference, motivating future hybrid approaches leveraging both stereo and shape-from-shading or deep correspondence models.

Pipeline Flexibility and Targeted Region-of-Interest Processing

The open-source nature and modular design of the presented pipeline enable flexible region selection (Figure 8), facilitating focused site characterization (e.g., candidate landing zones) while reducing resource requirements for large OHRC datasets. ROI-based processing strategies accelerate throughput and densify valid DEM contributions, especially where environmental constraints (illumination, overlap) limit stereo matchability over full swaths.

Figure 8: Targeted ROI stereo processing—native pixel image chips (left, center) and cropped high-resolution DEM (right)—enables application-driven, efficient mapping.

Implications and Prospects

This work demonstrates that high-precision, sub-metre lunar topography can be reliably produced from Chandrayaan-2 OHRC data using a completely open-source pipeline. The new PDS4/CSM import pathway and pipeline configuration may serve as a model for future instrument integration in planetary photogrammetry. DEMs at this scale are critical for mission planning, boulder-scale hazard analysis, traverse design, and geomorphological science.

Primary limitations are the accuracy of Chandrayaan-2 SPICE kernels and the availability of suitable stereo baselines. Forthcoming SPICE kernel releases from ISRO, as well as machine learning-based matching methods, will directly improve vertical accuracy and completeness. Broader application of these methods is expected to accelerate high-resolution lunar and planetary mapping and facilitate geospatial interoperability in planetary data archives.

Conclusion

The presented pipeline establishes a new standard for open access, sub-metre lunar DEM generation leveraging Chandrayaan-2 OHRC multi-view imagery, achieving validated horizontal accuracy <30 cm and vertical RMSE ~6 m across diverse regions. The approach overcomes previous limitations in data format compatibility and stereo pair availability, expanding the lunar mapping toolkit for both operational and scientific objectives. Future work will integrate improved sensor models, advanced correspondence algorithms, and broader coverage for comprehensive topographic characterization at spatial scales relevant to current and prospective lunar missions.