A Novel Adaptive Spectral Drift Correction Method for Recalibrating the MarSCoDe LIBS Data in China's Tianwen-1 Mars Mission

The Tianwen-1 mission, China's first interplanetary endeavor and Mars mission, touched the surface of the Red Planet on May 15, 2021. With the successful landing of the Zhurong rover on the southern Utopian Planitia of Mars, the Mars Surface Composition Detector (MarSCoDe) on board the rover has started to analyze the material composition of the Martian surface by using the laser-induced breakdown spectroscopy (LIBS). However, changes in instrument temperature and external environment during operation will cause spectral drift in LIBS data, leading to inaccurate material composition inversion results due to shifts in elemental characteristic peak positions. To address this problem, an adaptive spectral drift correction (ASDC) approach is proposed. By considering the distribution of characteristic peaks and drift differences between different elements, the proposed ASDC method corrects the spectral drift in LIBS science data according to an adaptive spectral segmentation strategy. Experimental results obtained on 36 real LIBS science data acquired over 19 exploration activities confirmed the effectiveness of the proposed approach in terms of lower average drift values, by comparing it with two reference methods. Taking the onboard calibration data as a reference, the average spectral drift values of the LIBS science data before and after applying ASDC decreased from 0.1012 nm to 0.0179 nm, 0.0926 nm to 0.0158 nm, and 0.1697 nm to 0.0103 nm in three LIBS channels, respectively. Furthermore, the resulting decrease in the standard deviation and rms error of the average drift values also proved the robustness of the proposed method.

and environment and exploitation of the compositions and soil characteristics of the Martian surface [1], [2]. To implement a detailed analysis of the Martian surface composition, the Zhurong rover carries a scientific payload named Mars Surface Composition Detector (MarSCoDe), which provides the laserinduced breakdown spectroscopy (LIBS) data, the short-wave infrared data, and the microimage camera, according to the in situ measurement [3]. In particular, when collecting the LIBS data, a high-energy pulsed laser is used to hit the surface of a sample, and a small amount of the sample is stripped and eventually excited to create plasma. The spectrum emitted by the plasma during its evolution is collected and analyzed qualitatively and quantitatively to determine the elemental composition and content of the detected material based on the location and intensity of the characteristic peaks [4]. LIBS can be used to analyze a wide range of element-related scientific issues due to its distinct advantage in identifying elements [5]. For example, the existence of hydrated minerals can be verified by identifying common elements such as silicon, sulfur, and hydrogen. Meanwhile, LIBS can be used to ablate the surface of a target with a laser and obtain the spectral information of the target rock. LIBS has become the most sophisticated and advanced technique for element-level detection in Mars exploration. Both NASA's Curiosity and Perseverance rovers are equipped with two science payloads that contain the LIBS system, namely the Chemical Camera (Chem-Cam) [6], [7] and Super Camera (SuperCam) [8], [9]. In the current Martian surface study, LIBS provides the element basis to find the evidence related to organic matter and water-bearing geological processes at the elemental level, as well as to better understand the geological background and evolutionary history of Mars. Accordingly, it has become a cutting-edge technique and a hot research topic in Mars exploration in recent years. Existing literature work can be divided into two main categories. One is dedicated to the processing of LIBS data, especially the accurate inversion of the target material composition. The other focuses on the investigation of scientific issues of Mars, such as the evolution of Mars to its geological background and the material inversion findings. For the former, a combination of nonnegative matrix decomposition and repeated k-means clustering was applied to the ChemCam LIBS data to analyze the major detection components, where six categories were then clustered based on these components [10]. By using the MarSCoDe LIBS data, quantitative analysis was carried out by comparing the content of oxide obtained by a convolutional neural network (CNN), backpropagation neural network, and two types of partial least squares regression analysis, where CNN achieved the best inversion accuracy among the compared methods [11]. Due to the difficulty of selecting local least squares parameters for quantitative analysis of ChemCam data, a particle swarm optimization-based approach was developed. The inversion accuracy was improved by iteratively searching for the optimal parameters with the minimum rms error as the fitness level [12]. For the latter, based on the ChemCam data, the soil composition near the Gale impact crater was found to be similar to the one near the Courage and Opportunity landing sites in terms of major elements, such as silicon and calcium, indicating that the distribution of Martian soils is relatively homogeneous [13]. Based on the variation of the hydrogen content of different rocks near the Gale impact crater, it was discovered that the finegrained soils of the Gale impact crater have a magnesium-iron composition with a clear hydrogen signature and low silicon content. Soils around the Gale impact crater have undergone limited aqueous alteration processes according to the results obtained by the Chemistry and Mineralogy instrument (Chemin) [14], [15]. Based on the MarSCoDe data, hydrated sulfate/silica material was successfully identified near the landing site of the Zhurong rover, suggesting that the formation with substantial liquid water was formed by rising groundwater or melting subsurface ice [16]. Fine-grained soils near the Zhurong rover's landing site are very similar to the ubiquitous surface dust found at other Mars missions' landing sites but are mixed with less chemically altered materials rich in calcium and lacking in magnesium. The formation of these mostly igneous minerals lends support to the hypothesis of a frozen ocean and sublimation [17]. Trends in the content of the major oxides based on the inversion of LIBS data from the SuperCam, together with the analysis of the remaining payloads, demonstrate a gradual increase in the mafic compositions and a gradual decrease in the level of the aqueous alteration as the stratigraphic position is lowered [18]. All these studies proved that the effective utilization of LIBS data is one of the most powerful in situ detection ways in Mars exploration, which can advance our understanding of Mars.
In addition, some works have studied the LIBS data calibration to produce more accurate LIBS spectra for further scientific analysis, such as normalization, noise removal, baseline removal, and spectral calibration [19], [20], [21], [22], [23], [24], [25], [26]. Moreover, a high-precision inversion result only can be obtained when the LIBS science data are consistent with the preflight data in terms of stable elemental peak locations. Before sending to Mars, the wavelengths corresponding to each pixel in the MarSCoDe's charge-coupled device (CCD) were calibrated in a specific laboratory environment by using a variety of elemental lamps [3]. However, the Martian environment is usually unstable due to the highly variable temperatures and external pressures, thus making it different from the one in the laboratory. Theoretically, when collecting data, differences in the external environment and instrument operating durations can cause a change in CCD temperature and pressure, and stresses within the spectrometer will change correspondingly. When stresses change, the distribution of light diffracted by the grating across the CCD surface shifts, resulting in a shift of pixel wavelength in the CCD and eventually the spectral drift and unsatisfactory material inversion accuracy [27]. In the real exploration activities on Mars, however, the spectrometer's stress change is very small, the main reason that causes the drift is mainly due to the optical system changes caused by environmental influences, especially the MarSCoDe optical system placed outside the rover which lacks temperature control. To ensure the consistency between laboratory and in situ LIBS data, initial calibration is required, with the goal of recalibrating the wavelength-pixel relationship. Within this context, particle swarm optimization was used to recalibrate the wavelength-pixel relationship of the MarSCoDe LIBS data [23], [26]. In [24], the spectral drift of MarSCoDe LIBS data was corrected by considering the relationship between the CCD temperature of spectrometers and drift. Apart from the above methods, the ChemCam team proposed an averaging segmentation correction method (denoted here as AvgSC) [28], [29]. By dividing each channel into n segments (i.e., usually eight segments), the drift distance and the pixel position of the strongest characteristic peak of the calibration target (i.e., usually the LIBS spectrum of Ti alloy) in each segment were counted. The spectral drift was then corrected on each pixel after implementing a third-order polynomial fitting model. A polynomial correction method (denoted here as PolyC) was proposed to address the same problem. A new pixel-to-wavelength conversion equation was obtained by manually selecting a few distinct characteristic peaks in each channel and fitting a second-order polynomial to their pixel position and theoretical wavelength, which also correct the spectral drift [30].
Despite the usefulness of the aforementioned methods, some open issues still required to be further addressed.
1) Existing methods (i.e., AvgSC and PolyC) mainly focus on determining the pixel-wavelength relationship, rather than correcting the LIBS spectral drift. Although the recalibration results help to mitigate the drift phenomenon presented at individual pixel positions to some extent, drifts still exist. This is due to the limited resolution of the instrument and the large wavelength interval between two adjacent pixels; therefore the positions of elemental characteristic peaks are located between two pixels. 2) The distribution of characteristic peaks differs between channels. This issue has not been properly considered in the existing methods. Manually selected characteristic peaks within each channel will significantly impact the performance of some methods (e.g., PolyC). Moreover, the uneven distribution of characteristic peaks within LIBS data will decrease the calibration performance of AvgSC. For example, in MarSCoDe LIBS data, characteristic peaks are sparsely distributed in channels 2 and 3, and densely distributed in channel 1. 3) Differences between the calibration data and laboratory data can be eliminated after the initial correction. However, as the change of data acquisition times, differences between the LIBS science data and the calibration data will gradually accrue, resulting in increased spectral drift and unreliable inversion results.
To consider the above issues, an adaptive spectral drift correction (ASDC) approach is proposed in this article. The main contributions and novelties are summarized as follows.
1) According to the actual distribution of characteristic peak positions, the proposed ASDC approach can adaptively correct the spectral drift presented at various resolution channels of the MarSCoDe LIBS data. Meanwhile, the proposed method allows the use of any spectrum as a reference, which does not rely on the availability of specific calibration targets. 2) A unified baseline for MarSCoDe LIBS science data is built, which corrects the science data obtained on different Martian days and provides a consistent baseline for their spectra. This eliminates the differences between science data, and creates a basis for further multitemporal and multisite analysis.
3) The proposed ASDC has the potential to be used in various application scenarios, such as the multitemporal spectral alignment in the same instrument, the cross-instrument (i.e., MarSCoDe, ChemCam, and the SuperCam) calibration, and the ground-flight data calibration. The rest of this article is organized as follows. Section II introduces the MarSCoDe LIBS data. Section III describes the spectral drift problem in detail. Section IV provides a detailed description of the proposed ASDC approach. Section V presents the experimental results and analysis. Finally, Section VI concludes this article.

II. DATA DESCRIPTION
As one of the key scientific instruments carried on by the Zhurong rover, the LIBS system in MarSCoDe is crucial. When it collects data, a laser beam with a wavelength of 1064.4 nm and a frequency of 1-3 Hz is emitted to shot a sample at a distance range of 1.6-7 m. After the shot, the spectral signal generated as the plasma cools is recorded by three CCDs each with 1800 pixels. The wavelength ranges covered by three channels (denoted as CH1, CH2, and CH3) are 240-340 nm, 340-540 nm, and 540-850 nm, with a corresponding spectral resolution of 0.19 nm, 0.31 nm, and 0.45 nm, respectively. Twelve calibration sample targets were carried out. Except for the No. 8 sample, which is utilized for wavelength calibration, the remaining samples are used for onboard scientific detection (i.e., Mars igneous composition simulation, cross-validation with other LIBS systems, such as ChemCam and SuperCam) [3].
By the 276th solar day (February 21st, 2022), MarSCoDe had collected 89 LIBS spectra (including 38 science data and 51 calibration data), which provide important information for the analysis of Martian surface material composition. Fig. 1 illustrates the location of LIBS observations along the traverse path of the Zhurong rover, where three LIBS spectra of science data are shown as examples. Fig. 2 and Table I show the LIBS calibration targets assembly and their material information. Details of the acquired calibration data and science data are provided in Tables II and III, respectively.   Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply. To reduce the influence of environmental and instrument noise during the data acquisition phase, baseline correction and noise reduction were first performed on all acquired LIBS data by using the BEADS (baseline estimation and denoising with sparsity) method [22]. Meanwhile, data interpolation was carried out in all LIBS spectra to find characteristic peaks and their positions more accurately. By applying a cubic spline function interpolation, each spectrum was interpolated uniformly to 20 000 pixels in the range of 240-850 nm. Finally, 36 out of 38 science data remain for experiment due to the poor quality of the 7th and 27th science data.

III. ANALYSIS OF THE SPECTRAL DRIFT IN MARSCODE LIBS DATA
As mentioned in Section I, although the initial spectral correction was carried out, spectral drifts still exist due to the Martian environment and instrument temperature. To quantitatively analyze the drift problem, characteristic peaks within each channel of the 2B-level MarSCoDe LIBS data were selected for calculating the statistics. Based on the adaptive segmentation result obtained by using the proposed ASDC on calibration sample No. 11 (i.e., basalt), at least one high-intensity characteristic peak within each segment was selected. Finally, 34 characteristic peaks were selected (as shown in Table IV). The peak-seeking and element-matching algorithms will be described in detail in Section IV-B.
Table V provides the spectral drift statistics calculated on 36 considered LIBS science data. We can observe the following points.
1) It is obvious that the spectral drifts are still presented in the 2B-level science data even though the initial correction has been implemented, with the minimum average drift value equal to 0.0926 nm (in CH2), which is approximately 0.3 pixel.
2) The spectral drift varies by channel, where CH3 has the highest average drift value (i.e., 0.1697 nm), and then CH1 (i.e., 0.1012 nm) and CH2 (i.e., 0.0926 nm);  III  DETAILED INFORMATION OF THE ACQUIRED LIBS SCIENCE DATA   TABLE IV  CHARACTERISTIC PEAKS CHOSEN FOR CALCULATING THE SPECTRAL DRIFT  STATISTICS 3) The lower the channel resolution (i.e., CH1: 0.19 nm, CH2: 0.31 nm, and CH3: 0.45 nm), the higher the standard deviation of the average drift values (i.e., CH1: 0.0484, CH2: 0.0536, and CH3: 0.0463). This indicates that as the resolution decreases, drift becomes more erratic. The obtained quantitative analysis results reveal that the spectral drifts remained in the 2B-level LIBS science  data is nonnegligible, which require a further removal and amendment.
As shown in Table III, two science data were collected for each single solar day, with the longest time interval of 15 minutes. During this limited period, assume that the external environment is relatively stable, thus its impact on the spectrum can be ignored. Therefore, the increase of spectrometer's temperature is most likely the primary cause of the spectral drift. Fig. 3 categorizes the science data by solar day, where blue and orange represent the first and second science data sets acquired on the same solar day, respectively. One can see that in all observations of three channels, the temperature of instrument rises from the first to second observation, while the drift value also increases following the same trend. This also confirms that spectral drifts in the MarSCoDe LIBS science data are highly correlated with the increasing temperatures of instrument.

IV. METHODOLOGY
The proposed ASDC consists of three steps: 1) reference spectrum selection; 2) adaptive segmentation of the reference spectrum; 3) ASDC. The block diagram of the proposed ASDC approach is illustrated in Fig. 4. Details of each step are described as follows.

A. Reference Spectrum Selection
As mentioned before, spectral drift differences between the onboard calibration data and the laboratory data have been corrected in the obtained 2B-level MarSCoDe LIBS data. Therefore, the remaining spectral drifts between the calibration data and science data that are caused by variations of the measurement environment are still required to be corrected. To this end, a reference spectrum in the calibration data should be selected. To verify the stability of the same calibration sample data, two most intense characteristic peaks within each channel were selected (as shown in Table VI). The maximum difference values between multiple observations and their average value on the same characteristic peak of a given calibration target (except the No. 2 calibration target, which only has one observation) was calculated. Numerical results are provided in Table VII, and the average spectrum for each of 12 calibration targets is finally used. Detailed analysis can be found in Section V-A. The intensity of the spectrum varies when the same LIBS system acquires multiple LIBS spectra for the same substance. Therefore, the intensity of the spectrum cannot be used to determine the similarity between the science data and the calibration data. However, while the intensity varies, the distribution of characteristic peaks is very similar, indicating that the shape of multiple spectra for the same substance is almost the same. Therefore, the spectral angle mapper (SAM) is applied, which measures the similarity by the shape of spectral signatures. Its formula is provided in as follows: where x and y represent the spectrum of each LIBS science data and calibration data, respectively. Finally, the spectrum of calibration data that matches each science data with the lowest θ value is selected as the reference spectrum (S r ) for the subsequent segmentation and drift correction.

B. Adaptive Segmentation of the Reference Spectrum
To overcome the limitation in AvgSC and PolyC methods that ignore the distribution of individual characteristic peaks, an adaptive segmentation strategy was proposed in this work. There are two parameters in the proposed algorithm, i.e., the number of characteristic peaks N cp and the intensity threshold T. In particular, N cp determines the initial number of characteristic peaks used for segmentation, to keep as many characteristic peaks in each channel. T is set to ensure that the spectrum remains continuous after correction, and should be smaller than the intensity of most characteristic peaks in each channel.
The main steps of the proposed adaptive segmentation are provided in Algorithm 1. First, a peak-seeking algorithm is used to find the top N cp characteristic peaks having the highest intensity in reference spectrum S r . By applying the second-order derivative to the LIBS data and calculating the local minimum value, the spectral characteristic peaks are identified. Meanwhile, to get a more accurate position of characteristic peaks by applying a second-order derivative, the cubic spline function interpolation is applied around the characteristic peaks. Then, the element type of the characteristic peak is identified according to the nearest distance principle, by comparing the position of the identified characteristic peaks with the ones in the ChemCam LIBS library [31]. Then, a merging interval range U i = [P L i , P R i ] is defined for the characteristic peak CP i (i = 1,2, …,N cp ), where P L i and P R i is the lower and the upper bound of the interval, respectively, and Algorithm 1: Adaptive Segmentation of Reference Spectrum.
is less than T. Close characteristic peaks are merged according to the defined merging interval. Starting from the U i with the greatest intensity, positions of the remaining characteristic peaks P j are analyzed in turn to determine whether they fall in the U i . The above process iterates until all N cp characteristic peaks have been analyzed and marked sequentially. Finally, the number of segmentations (N s ) is obtained, as well as the range of each segment with their characteristic peaks. The proposed adaptive segmentation algorithm allows for an automatic separation of characteristic peaks but also ensures an independent correction of strong peaks that are divided into different segments. The maximum number of these segments in each channel is defined depending on N cp in the proposed approach. Meanwhile, it utilizes the most intense peak in those complex overlapping peaks as the calibration reference, which avoids the inaccurate peak position estimation issue.

C. Adaptive Spectral Drift Correction (ASDC)
The segmentation results obtained on the reference spectrum (Seg r ) are applied to the science spectrum (S c ) to generate the corresponding segment result (Seg s ) and the number of segmentations (N s ). Then, the peak-seeking algorithm is also applied to each segment of the science data to identify the characteristic peaks. Accordingly, the position of the strongest characteristic peak in each segment of the reference spectrum (P str ) is extracted. A wavelength location P pos ([P str -M, P str + M]) on the science spectrum is defined, where M is a small constant. In a given segment, if the strongest characteristic peak of the science spectrum (S str ) falls within P pos , characteristic peaks of the science spectrum and the reference spectrum are successfully matched for further drift correction. Otherwise, the drift in the segment will not be corrected if it failed in the match process.
For the segment which is to be corrected, the pixel distance D between the matched peaks in the target and reference spectra is calculated using If D < 0, the science spectrum will be shifted left, and d pixels at the beginning of this segment will be moved to the end of this segment. On the contrary, if D >0, d pixels will be moved from the end to the beginning of this segment. Note that such an operation enables the correction of characteristic peaks within each segment, but also keeps the continuity of the spectrum after correction. The whole process of the ASDC is provided as follows.

A. Results of Reference Spectrum Selection
Since the composition of calibration samples is fixed, we assume relative stability of the LIBS measurements on the same calibration target. Therefore, for a given calibration target, even if multiple calibration spectra were acquired over multiple solar days, positions of the characteristic peaks almost have no change. This is confirmed by the numeric results provided in Table VII. One can see that for each characteristic peak, differences between the MarSCoDe LIBS data obtained for the same calibration target under different solar days are very small. The mean values of the maximum differences in three channels are 0.0135 nm (CH1), 0.0119 nm (CH2), and 0.0222 nm (CH3), respectively, whereas the maximum differences are 0.0314 nm (CH1), 0.0399 nm (CH2), and 0.0813 nm (CH3), respectively. It is worth noting that the maximum differences are still much smaller than the channel resolution (i.e., 1 pixel) of MarSCoDe LIBS, which is 0.19 nm (CH1), 0.31 nm (CH1), and 0.45 nm (CH3), respectively. This allows us to use the averaging LIBS spectrum of each calibration target for the reference spectrum selection.
Then, SAM scores were calculated between each of 36 science data and the averaging LIBS spectrum of 12 calibration targets. The calibration data that best matched the science data were selected based on the minimum SAM principle. Results are

B. Segmentation Results on the Reference Spectrum
Based on numerous trials, it was discovered that when N cp = 20 and T = 0.001, a satisfactory segmentation result can be obtained by capturing the most distinctive peaks in all three channels. Even the sparse characteristic peaks distributed in CH2 and CH3 can be captured successfully. Fig. 5(a)-(c) shows the most intense 20 characteristic peaks in each of the three channels of the reference spectrum. Fig. 5(d)-(f) shows the final segmentation results in each channel. By visually analyzing the segmentation results, it can be clearly seen that the individual significant peaks are separated and followed into different segments, while characteristic peaks close to the significant peaks are all merged into the same segment. This offers a good basis for the following drift correction within each segment.

C. Adaptive Correction Results on the Science Data
To quantitively evaluate the correction results obtained by the proposed ASDC approach, drift values were calculated based on the 34 selected characteristic peaks as shown in Table IV. Accordingly, by counting the absolute value of the drift before and after correction, drift values are calculated for each of 34 characteristic peaks, as well as the average drift values in each channel of the science data. Statistics results are provided in Fig. 6, one can see that in all three channels, the spectral drift values are significantly decreased after correction (in orange) than the ones before correction (in blue). In particular, the decreasing trend is presented more clearly in CH3 [see in Fig are all smaller than 0.018 nm, which is only 9.4%, 5.1%, and 2.3% of the spectral resolution in each of three LIBS channels. The largest drift before calibration in science data No. 22 is 0.2547 nm, but after calibration, the drift was reduced to 0.0124 nm, which is only 1/25th of the original drift value. The obtained numeric results demonstrate the good performance of the proposed ASDC approach in correcting the real MarSCoDe LIBS science data.

D. Comparison With Other Methods
To verify the effectiveness of the proposed ASDC approach, quantitative results obtained by using other two reference methods, i.e., PolyC and AvgSC, were further compared (as shown in Fig. 7). The usefulness of the three methods is confirmed by the decrease of overall drift values obtained in three methods from the original data to the corrected ones. In particular, the proposed ASDC approach outperformed the other two methods in terms of    Fig. 7(a)-(c)]. The average drift value of ASDC (i.e., CH1: 0.0179 nm, CH2: 0.0158 nm, and CH3: 0.0103 nm) is significantly reduced when compared to the ones in PolyC (i.e., CH1: 0.0548 nm, CH2: 0.0510 nm, and CH3: 0.0673 nm) and in AvgSC (i.e., CH1: 0.0270 nm, CH2: 0.0560 nm, and CH3: 0.0412 nm). The standard deviation of the average drifts in ASDC (i.e., CH1: 0.0118 nm, CH2: 0.0056 nm and CH3: 0.0018 nm) is also lower than the ones in PolyC (i.e., CH1: 0.0264 nm, CH2: 0.0259 nm, and CH3: 0.1224 nm) and in AvgSC (i.e., CH1: 0.0160 nm, CH2: 0.0547 nm, and CH3: 0.0312 nm). Meanwhile, the rms error of prediction (RMSEP) is also used to measure the effectiveness of proposed ASDC [24] given by where A i is the average drifts of each science data in a given channel and n is the number of science data (in our case n = 36). Table IX shows the obtained numeric result of RMSEP, illustrating that ASDC is more potential for spectral drift correction in terms of the lowest RMSEP values in all three channels. Furthermore, the benefit of the proposed method becomes more apparent as the density of emitting lines in three channels decreases. The sparser the emission spectrum, the fewer significant characteristic peaks exist, and the less likely they are to be evenly distributed across the channel's positions. For example, in CH3, the average drift correction of ASDC is 0.1594 nm, whereas the one in CH1 and CH2 is 0.0833 nm and 0.0768 nm, respectively. This indicates that the proposed method overcomes the limitations of existing methods by considering the actual distribution of the characteristic peaks.

E. Analysis of the Drift Correction Effects on Different Characteristic Peaks
As the principle of the proposed ASDC approach is based on the characteristic peaks with the highest intensity (HP) within each segment, the remaining peaks with low intensity (LP) are corrected accordingly. So, the correction effects are only related to the LP peaks required to be further analyzed.
In CH1 and CH2, the spectral lines are relatively dense, having the first, second, and third intense characteristic peaks within each segment. However, in CH3, the spectral lines are relatively sparse, they may only have one characteristic peak. Meanwhile, characteristic peaks found in both reference data and science data were kept to verify the correction results. The final selected LP peaks within each channel and their positions are listed in Table X. Based on the considered 36 science data points, we calculated the average values of LP peak drift before and after correction. Results are shown in Table XI. One can observe that although the ASDC approach is designed by considering the HP peaks within each segment, it also exhibits a good correction effect on the LP peaks within the same segment. The drift of LP peaks reduced from 0.1123 nm, 0.1207 nm, and 0.3553 nm to 0.0794 nm, 0.0759 nm, and 0.2596 nm in three channels, respectively. However, there is still a distance in the correction accuracy between the LP and HP peaks. The HP drift correction values in the three channels are 0.0833 nm, 0.0768 nm, and 0.1594 nm, respectively, whereas the ones of LP in the three channels are 0.0329 nm, 0.0448 nm, and 0.0957 nm, respectively. This demonstrates the superiority of the proposed ASDC method in the correction of the HP peaks in the LIBS spectrum, but further improvements are still required to improve the accuracy of the correction of LP, especially those LP peaks that are associated with some chemical elements with high scientific value on the Martian surface.

VI. CONCLUSION
The spectral drift phenomenon presented in the flight MarSCoDe LIBS data is caused by the external environment of Mars and the duration of instrument operation times. A novel ASDC method has been proposed in this article to address this problem. Compared with the initial calibration, which only corrects the correspondence between pixels and wavelengths, the proposed ASDC method further corrects positions of individual characteristic peaks, but also constructs a unified reference for all LIBS science data by using the calibration data. This provides an important data consistence for the subsequent multitemporal analysis and an accurate quantitative inversion. Experimental results obtained on the considered 36 scientific data confirmed the effectiveness of the proposed method in terms of the significant decreasing of drift values in three channels (from 0.1012 to 0.0179 nm in CH1, 0.0926 to 0.0158 nm in CH2, and 0.1697 to 0.0103 nm in CH3), which reduce about 86.7% average drift values in the original LIBS data. Meanwhile, the proposed method overcomes the limitations of existing stateof-the-art calibration methods, thus achieves better correction performance when dealing with the practical LIBS calibration applications. It is potential for multitemporal spectral alignment in a same instrument, the cross-instrument (i.e., MarSCoDe, ChemCam, and the SuperCam) calibration and the ground-flight data calibration.