Geodetically Anchored 0.30m Digital Elevation Model of the Chandrayaan-3 Vikram Landing Site from Chandrayaan-2 Orbital High Resolution Camera (OHRC) Stereo Imagery

TL;DR

This work demonstrates that a sub-meter, geodetically anchored digital elevation model of the Chandrayaan-3 Vikram landing site can be produced from Chandrayaan-2 OHRC stereo imagery using an entirely open photogrammetry workflow. By applying a multi-stage pipeline (ISIS ingestion, ALE-based CSM camera modeling, and Ames Stereo Pipeline bundle adjustment and MGM stereo), the authors achieve a $0.303$ m grid DEM over $2.18\times2.24$ km with a median triangulation error of $8.1$ cm and $91.2\%$ valid pixels. Geodetic alignment to a NAC reference DEM yields a negligible vertical bias ($+0.28$ m) and NMAD of $2.88$ m over nearly $490{,}000$ pixel comparisons, validating the approach for independent hazard mapping and site characterization. A key finding is that continuous-sensor-model (CSM) camera representations are essential for stable OHRC stereo, while the legacy ISIS camera model fails to converge, underscoring the need for open, state-vector-aware photogrammetry tools in high-resolution lunar mapping. The resulting open DEM complements NAC data, enabling sub-meter hazard detection and informing future missions (Chandrayaan-4, LUPEX, Artemis) with a reproducible, openly available processing framework and data products.

Abstract

ISRO's terrain characterization and hazard mapping from Chandrayaan-2 Orbiter High Resolution Camera (OHRC) stereo imagery were central to the safe landing of Chandrayaan-3 - the first successful landing in the lunar south polar region. However, these elevation products were generated with a proprietary pipeline and have not been publicly released. We present a 0.30 m/pixel digital elevation model (DEM) of the Chandrayaan-3 Vikram landing site using a fully open workflow based on ISIS, the Ames Stereo Pipeline, and ALE, achieving sub-meter resolution comparable to mission-reported products. The reconstruction covers 2.18 x 2.24 km with 91.2% valid pixel coverage, 8.1 cm median triangulation error, and 40-50 cm relative vertical precision. The Vikram lander and Pragyan rover are individually resolved. Geodetic alignment to an LROC NAC stereo DEM achieves approximately 30 m horizontal accuracy; pixel-wise validation at 3 m resolution confirms negligible vertical bias (median dz = +0.28 m) and robust dispersion (NMAD = 2.88 m). Stable OHRC stereo convergence requires Community Sensor Model (CSM) camera models; the legacy ISIS camera model failed across two independent sites. At 0.30 m, these DEMs complement LROC NAC DTMs (approximately 1 m), resolving sub-meter hazards below the NAC detection threshold. Applied to the extensive OHRC south polar archive, this methodology provides independent capability for hazard mapping and landing site analysis for upcoming missions including Chandrayaan-4, LUPEX, and Artemis.

Figures (10)

• Figure 1: Soma lunar data portal. Left: Overview of the OHRC South Pole catalogue showing 203 data products spanning February 2020 to November 2025, with the Chandrayaan-3 landing site (Shiv Shakti Point) marked. Right: Location-based query at Shiv Shakti Point returning 37 data products at the selected coordinates, with image footprints displayed on the 3D terrain and individual product metadata (observation times, corner coordinates) available for stereo pair evaluation.
• Figure 2: Schematic of the OHRC-to-DEM processing pipeline. Stereo pairs are identified using the Soma portal (Section \ref{['sec:data']}), and Level-1 calibrated images are ingested into ISIS with SPICE kernels. CSM camera models are generated via a custom ALE integration script, and the Ames Stereo Pipeline performs bundle adjustment, stereo correlation (MGM), and DEM generation. Outputs include a 30 cm DEM, orthoimage, and triangulation error map.
• Figure 3: OHRC orthoimage of the Shiv Shakti Point region from image ch2_ohr_ncp_20240425T1406019344 (orbit 20813). Left: The full 2 $\times$ 2 km region of interest used for DEM generation. The red box marks the SPICE-projected location of the Chandrayaan-3 Vikram lander (69.5082$^\circ$S, 32.3331$^\circ$E; Table \ref{['tab:images']}), prior to geodetic alignment. Right: Zoomed view showing the Vikram lander (red arrow) and the Pragyan rover (blue arrow) approximately 8.3 m to the upper left (69.5079$^\circ$S, 32.3331$^\circ$E), consistent with the rover traverse documented by Iyer2025. Yellow arrows indicate the shadows cast by both objects, which are consistent in direction and length, confirming the solar incidence angle at the time of acquisition ($\sim$78$^\circ$). Both the lander and rover are resolved as distinct features at the 0.26 m pixel$^{-1}$ GSD. Image credit: ISRO/C. Tungathurthi.
• Figure 4: Color-coded elevation map of the Shiv Shakti Point DEM with 50 cm contour intervals, shown at three zoom levels. Left: Full 2.18 $\times$ 2.24 km DEM extent; the color scale spans approximately 57 m of total relief. Red inset: Intermediate zoom of the Vikram landing area, showing widely spaced contours that confirm the exceptional flatness of the selected landing site --- elevation varies by less than 3 m across several hundred meters. Blue inset: Fine-scale zoom showing sub-meter contour detail around individual craters and the lander location. Elevations are datum-shifted to positive values (referenced to the 1737.4 km lunar sphere). DEM shown in oblique stereographic projection centered on the scene. Image credit: ISRO/C. Tungathurthi.
• Figure 5: Triangulation error map overlaid on the orthoimage. Blue/dark tones indicate low intersection error ($<$5 cm); green/yellow indicate moderate error (5--15 cm); warm tones (orange/red) indicate elevated error ($>$20 cm). The majority of the scene is dominated by low error (median 8.1 cm), with elevated values concentrated along crater walls and rims (where stereo occlusion and shadow reduce matching quality), at strip margins (weaker stereo geometry), and in along-track bands (a known ASP MGM tile-blending artifact). Red inset: Zoomed view of the Vikram lander area (dashed circle), which sits in a low-error zone consistent with the flat, well-illuminated terrain at the landing site. Image credit: ISRO/C. Tungathurthi.