Chandrayaan 3 data: Unveiling the challenges of performing LIBS on the Lunar Surface

I. INTRODUCTION

The Chandrayaan 3 (C3) mission was developed and launched by the Indian Space Research Organization (ISRO) in July 2023. After the failure of Chandrayaan 2 lander in 2019, the goal of Chandrayaan 3 was to achieve the objectives of its predecessor, i.e., demonstrate full capability in safe landing, roving, and conducting in situ experiments on the lunar surface. This mission contained a lander including a 26-kg rover named Pragyan that carried the Low-energy Eye-safe Laser-Induced Breakdown Spectroscopy (LE-LIBS) instrument. The laser operates at 1540 nm, delivering 20 pulses at 5 Hz, with an energy between 3 and 4 mJ per pulse on target.

During its 100-meter traverse (Fig. 1), the rover conducted 66 LIBS investigations. At each of those locations, there are between 1 and 10 LIBS spectra. The results have been studied in this work and compared with the preliminary results of Sridhar et al., (2025) [1].

Fig. 1: Path of Pragyan rover (from ISRO).

 

II. DATA PROCESSING

The C3 archive [5] contains both raw data (level-0) and wavelength-calibrated data (level-1). Starting from level-0 data, we built our own calibration. Data were first de-noised with the Undecimated Wavelet Transform method (UWT) [3], followed by baseline removal with Asymmetrically Iterative Reweighted Penalized Least Squares (AIRPLS) function [4]. We found that our results do not match level-1 data provided by the C3 team [5] with unexplained discrepancies up to 8 nm. Thus, for the rest of the study, we used our own calibration, starting from level-0 data.

Next, emission peaks were fitted with a Lorentzian function that led to peak identification and resolution from wavelength and full width at half-maximum (FWHM), respectively. The NIST database was used for peak element identification.

 

III. RESULTS

We have processed all 66 investigations, and when significant, classified the results (see Fig. 2 below) on the basis of Signal to Noise Ratios (SNR) and noise levels. We then focused on Target 6 to compare our element identification results with those reported in [1], and subsequently with Target 19.

 

     A. Spectra classification

We have developed a Python tool to process 34 of the 66 operations, which are considered by the C3 team as significant for science, resulting in 170 LIBS spectra Those spectra were compared and classified based on their main characteristics, i.e., noise and SNR. SNR is defined as the ratio of the maximum peak amplitude versus the standard deviation of the signal between 680-760 nm, a zone considered as pure noise. With this information, we found that 13 of the 66 investigations have the best signal with the most peaks (tens to hundreds of peaks as expected in LIBS spectra [2]) and are therefore scientifically usable.

Fig 2: Operations classification regarding the noise and the SNR of the strongest peak. Data from operations 14, 21, 22, 23, 24, and 47 are damaged and not shown. The 13 optimal investigations are circled in red.

 

     B. Comparison with Sridhar et al., (2025) [1]  

Target 6 reported in Sridhar et al., (2025) [1], was acquired on August 27, 2023 at 18:12.23 UT. Figure 3 shows our processing of the LIBS spectrum, which we compared with the one processed by the C3 team.

Both spectra appear similar, but differences arise after removing the baseline and fitting the peaks. The same number of peaks could be fitted in both of the studies, however, our element identification based on NIST database do not fully agree with the results in [1].

We have identified 28 peaks. 11 match in wavelength and element identification results of [1]. We found that 17 do not match in wavelength (i.e., shifted by more than 1 nm). Specifically, 6 do not match in element identification because another element is more probable while the other 11 do not have an unequivocal element candidate. This means those peaks do not match regarding the element in the NIST database, or the presence of the element is in doubt because the major emission peaks of this element are missing (e.g., manganese, chromium, titanium). Finally, some elements like silicon, iron and aluminum are clearly present in this sample, but among the few peaks attributed to those elements some do not include the most intense peaks like the one attributed to Si I at 792 nm. Our wavelength recalibration did not change these observations.

Fig. 3: Spectrum of Target 6 derived in this work. Some peaks could not be associated with specific elements, as no unequivocal correspondence using NIST could be established.

 

     C. Comparison with Target 19

The measurements of target 19 took place 2 days after Target 6, and their spectra are both clean and strong (Fig. 2). Thus, the comparison can remove doubt and confirm peak wavelengths. It appears that the spectra of these targets are very similar, which confirms that the same regolith was sampled in both cases. We will continue to investigate other targets to confirm the presence of additional elements beyond Fe, Al, Mg, Ca, O, and Si.

 

IV.   CONCLUSION

Chandrayaan-3 has demonstrated that LIBS can work on the Moon, despite challenges with the current setup (no active refocus) to systematically acquire a meaningful signal. One of the next steps is to investigate the origin of the observed nonlinear wavelength shifts. A possible explanation is an issue with the beam focus, which we are currently investigating in our laboratory.

 

REFERENCES

[1] Sridhar, R. V. L. N., et al. « Chandrayaan-3 LIBS Sensor: Preflight Characterization, Inflight Operations, and Preliminary Observations ». IEEE Sensors Journal 25, n^o 2 (2025): 2554‑66. https://doi.org/10.1109/JSEN.2024.3506037.

[2] Wiens, Roger C., Sylvestre Maurice, et al. « The SuperCam Instrument Suite on the NASA Mars 2020 Rover: Body Unit and Combined System Tests ». Space Science Reviews 217, n^o 1 (2021): 4. https://doi.org/10.1007/s11214-020-00777-5.

[3] Starck, Jean-Luc, et al. « The Undecimated Wavelet Decomposition and its Reconstruction ». IEEE Transactions on Image Processing 16, n^o 2 (2007): 297‑309. https://doi.org/10.1109/TIP.2006.887733.

[4] Baek, Sung-June, et al. « Baseline Correction Using Asymmetrically Reweighted Penalized Least Squares Smoothing ». The Analyst 140, n^o 1 (2015): 250‑57. https://doi.org/10.1039/C4AN01061B.

[5] pradan.issdc.gov.in/ch3/