Optimization of Scanning Scheme for Low-height Vegetation Survey Based on Terrestrial Laser Scanning - A Case Study on the Restored Sand Land, Southern Qinghai-Tibetan Plateau

Background: In recent decades, vegetation surveys based on terrestrial laser scanning (TLS) have developed rapidly, especially on the forest inventory, but few studies have been conducted to the low-height vegetation. Because of the high investigation cost and subjectivity, it is impending to provide a scientic scanning scheme based on the TLS for the low-height vegetation survey (e.g. shrub, grassland, and meadow) in eco-fragile region (e.g. Qinghai-Tibetan Plateau). Method: In this study, we extracted the vegetation parameter i.e., number, height (H), and crown width (CW) of the two sample plots to evaluate the integrity of the data collected by TLS, on the restored sand land in southern Qinghai-Tibetan Plateau. We assessed the difference between the scanning mode of single-scan (SS) and multiple-scan (MS), and evaluated the inuence of terrain uctuation (windward slope, leeward slope, and the peak of slope) on the determination of scanning spots. Results: The results showed that: (1) the accuracy of vegetation parameter extracted by the mode of SS was mainly affected by the occlusion and the distance from central scanning spot, the RMSE of vegetation parameters is the smallest (RMSE H = 0.186 m; RMSE CW = 0.208 m) within 20 m from the central scanning spot. (2) For the MS mode, in addition to the central scanning spot, the scanning spot located at the peak of the slope is the most important, which was the connection of combining the data of windward slope and leeward slope. Conclusion: To sum up, the scientic layout of scanning spot is the key to collecting data by TLS eciently, and topography is the main factor affecting the layout of scanning spot. Since occlusion effect cannot be avoided, it can only be compensated by setting up more scanning points. Secondly, the accuracy of different sensors will has inuence on the distance between adjacent scanning spots.


Introduction
Vegetation types such as shrub, grassland, and meadow have the characteristics of low-height and distributed widely , and play an important role in the global carbon cycle (Scurlock and Hall. 1998), prevention of land degradation , and biodiversity conservation (Dong et al., 2020). It was reported that the grassland has been seriously degraded since the 1980s, and accompanied with shrub encroachment (Eldridge et al., 2011), causing losses in grassland biodiversity, water retention capability, and soil nutrients (Andrade et al., 2015;Chen et al., 2017). Therefore, vegetation restoration of degraded grassland is essential to maintain the functions and services of grassland ecosystems.
From 2008 to 2017, we established a set of experimental plots for vegetation restoration of degraded grasslands, and selected suitable vegetation restoration species by comparing vegetation parameters . Traditionally, the vegetation parameters for the low-height vegetation were estimated in combination with eld surveys and optical remote sensing data (Psomas et al., 2011), the Light Detecting and Ranging (LiDAR) technology, which can provide three-dimensional (3D) information, has gradually replaced traditional investigation method for vegetation survey. Terrestrial laser scanning (TLS) can provide the vertical structure information of vegetation with millimeter-level accuracy (Liang et al., 2018), which has unique advantages for collecting vegetation information on sample scale (Liang et al., 2012; Liang and Hyyppä. 2013), such as diameter at breast height (Kankare et al., 2015), height , and biomass estimation (Calders et al., 2015). However, LiDAR as a tool for vegetation investigation were, so far, primarily used for forest inventories (Liang et al., 2016), the application in lowheight eco-system was rarely investigated (Wachendorf et al., 2017;).
TLS scanning are divided into the mode of single-scan (SS) and multiple-scan (MS) (Liang et al., 2016), most of researches on scanning method are to compare the difference between these two modes in the forest inventories (Pueschel et al., 2013;Saarinen et al., 2017). For the low-height vegetation, Li et al (2019) utilized the TLS date obtained by registering 18 scanning spots to assess the effects of topography on the revegetation. Xu et al (2020) estimated the grassland aboveground biomass from the TLS data obtained by 5 scanning spots. Both studies adopted TLS data registered by the mode of MS to obtain more complete information of the sample plot. Damian et al (2019) indicated that in the process of estimating grass biomass by TLS, two scans from opposite directions slightly increased the spatial information compared to one scan only. Therefore, it is important to arrange the location of the scanning spots reasonably for the date acquisition.
As a matter of fact, we should reduce the risk and cost of the vegetation investigation in some areas with harsh environmental conditions (e.g. the plateau region like Qinghai-Tibet Plateau), and improve the data acquisition in terms of e ciency, robustness, and precision. Trochta et al. (2013) mentioned that occlusion is one of the key factors that limited the potential of TLS data acquisition. Occlusion is caused by objects shadowing each other, so that some part of the objects of interest are not visible to the scanning spots with only one side. The effect of occlusion is usually mitigated by combining TLS scans from different locations (Hilker et al., 2012). But few studies provided a standard for the location determination of scanning spots, and explored the factors affecting the completeness of data collection. Therefore, we mainly discussed the difference of the scan mode in the data acquisition of low-height vegetation, and explored the in uence of terrain features (i.e. the slope and the uctuation of terrain) for setting scanning spots. We selected some vegetation parameters for comparison, including the number (N), height (H), and crown width (CW), because these parameters can directly re ect the growth status of the vegetation (Li et al., 2016). We provided two hypotheses in this study: (1) for the low-height shrub, the mode of SS could be competent for information collection, but the occlusions from the vegetation and the distance from the central scanning spot were the main limiting factors (2) the uctuation of terrain has a great in uence on setting scanning spots. In addition, we will provide a reasonable suggestion for setting the scanning spots and optimizing the scanning scheme according to the different data requirements.

Study Area
There are two sample plots in this study, the plot #1 was located on the north of the highway from Zedang to Sangye in Shannan city (91.324°E, 29.181°N), and the altitude is about 3560 m above sea level. The plot #2 were located on the third uvial terrace of the Yarlung Zangbo River in Gongga County, Tibetan Autonomous Region of China (90.889 °E and 29.337 °N near the Lhasa Airport), with altitudes ranging from 3560 m to 3730 m above sea level. These areas belong to the semi-arid plateau temperate monsoon climate zone with the annual precipitation of approximately 300-450 mm and the annual daily mean temperature between 6.3 and 8.7℃.

Data Collection
We used the RIEGL VZ-400i scanning system to conduct TLS scanning of sample plot #1 on June 30th, 2017, and the scanning of sample plot #2 was carried out on July 1st, 2017. The VZ-400i TLS has a eld view of 360° (horizontal) × 100° (vertical), an accuracy of ± 5 mm per 100 m. This TLS system acquires 3D point cloud at a speed up to 500,000 points per second, and every scanning spot spent about 20 seconds with a high-speed model. When collecting TLS data for the sample plot (square), one central scanning spot was determined rstly in the center of the plot, and four peripheral scanning spots were located near the four corners of the sample plot (Fig. 1c). In addition, we recorded the terrain features (including the slope and the uctuation of terrain) and the vegetation coverage of the sample plot (Table 1).

TLS pre-processing
The point cloud data collected at each scanning spot was freely combined and registered according to the experimental requirements, and we uniformly named the data set to facilitate understanding and writing. The DS 1−(r) and DS 2−(r) represent the reference data set collected by 5 scanning spots of the sample plot #1 and #2, the DS 1−(n) and DS 2−(n) represent the experimental data set collected by the n'th scanning spot, and the DS 1−(n−) and DS 2−(n−) represent the experimental data set collected by all the scanning spots except the n'th scanning spot (the speci c number of the scanning spots could be found in Fig. 1c). In previous research (Tian et al., 2020), we have evaluated the extraction accuracy of the H and CW between the reference data set (5 scanning spots) and the eld-measured data (R 2 = 0.944 and 0.970, respectively), which veri ed the integrity of reference data set based on TLS.
The registration was based on the close-range point clouds (Besel and McKay., 1992) and registration poles using the RiSCAN Pro software (http://www.riegl.com). All the registered data were preprocessed in the LiDAR360 software (https://www.lidar360.com), and the procedure including the denoising, ltering, and normalization. Firstly, the statistical outlier removal lter was utilized in the step of denoising. Secondly, the ground point was separated based on the progressive triangulated irregular network densi cation lter (Zhao et al., 2016), and the digital elevation model (DEM) at a resolution of 0.02 m was obtained by the Kriging interpolation (Guo et al., 2010). Finally, the normalized point cloud data was obtained by subtracting the corresponding DEM value from the height value of point.

Vegetation parameter extraction between SS and MS
Due to the large number of individual vegetation in sample plot #1, we set up 12 small quadrats (10 m × 10 m) according to different distances from the central scanning spot (Fig. 2), and the distance was 20 m (quadrat #9-12), 35 m (quadrat #5-8), and 50 m (quadrat #1-4), respectively. We extracted the N, H, and CW of the vegetation in the 12 quadrats from the DS 1−(r) and DS 1−(5) . Then, the difference between MS (5 scanning spots) and SS (central scanning spot) could be compared from the vegetation parameters of 12 quadrats, and the in uence of distance between the quadrat and the central scanning spot for data collection.
The difference of the slope between the sample plot #1 and #2 is obvious, 3.8% and 56.2%, respectively.
The terrain has a uctuation in the sample plot 2, while the sample plot #1 is at. We extracted the same vegetation parameters of all vegetation from the DS 2−(r) and DS 2−(5) , for testing the applicability of SS (central scanning spot) in steep slope area.

The in uence of terrain features
In this part, we obtained the terrain features at the location of the n'th scanning spot through DEM data of the sample plot 2#. The uctuation of the dunes is mostly caused by wind erosion , and the highest location between the windward slope and the leeward slope is known as the slope peak ( Fig. 3). In order to analyze the importance of each scanning spot for data collection by TLS, we take any scanning spot in sample plot #2 as an impact factor, and compared the vegetation parameter extracted from the DS 2−(r) and DS 2−(n−) .

The error analysis
The error analysis of the vegetation parameters extracted by the experimental Data Set and the reference Data Set (DS 1−(r) and DS 2−(r) ) was determined based on the root mean squared error (RMSE), the equations as follow: where VP represent the vegetation parameter (including the H and CW), VP ref is the vegetation parameter extracted by the DS 1−(r) and DS 2−(r) , VP e is the vegetation parameter extracted by the experimental Data Set, and the experimental Data Set include the data combination of different scanning spots.

Vegetation parameter extraction between SS and MS
As shown from Table 2  We compared the results of N, H-mean, and CW-mean extracted from the DS 1−(5) , DS 2−(5) , DS 1−(r) and DS 2−(r) ( Table 3). In the sample plot #1, the proportion of extracted vegetation number (Ns/Nr) followed the order of Quadrat #9-12 (97.9%) > #5-8 (94.8%) > #1-4 (40.6%), and the RMSE of the H-mean and the CW-mean followed the same order, the minimum value all appears in the Quadrat #9-12, is 0.186 m and 0.208 m, respectively. There are total 40 shrubs in the sample plot #2, and 95 percent of the vegetation was extracted, the RMSE of the H-mean and CW-mean are 0.4 m and 0.854 m, respectively.

Terrain feature of the scanning spots
We extracted the terrain type at the location of 5 scanning spots of the sample plot #2 according to the DEM data (Fig. 4a), the terrain feature of scanning spot #1, #2, and #5 is windward slope, scanning spot #3 is leeward slope, and scanning spot #4 is slope peak. The number of vegetation on the windward slope and slope peak was signi cantly more than that on the leeward slope (Fig. 4b). Table 4 the error analysis of vegetation parameters extracted between DS 2-(n-) and DS 2- could also provide about 80% accuracy, and the occlusion also was the major factor for the error. In addition, we found that the occlusions come not only from other vegetation (mainly affect the extraction of H, Fig. 5b), but also from some vegetation itself (mainly affect the extraction of CW, Fig. 5d), because only one side of the vegetation information was collected by the mode of SS (Damian et al., 2019).

Evaluation of the Terrain features
For the sample plot #2 with steep slope and a uctuation, we believed that each scanning spot contributes to the data acquisition rstly (Table.4). The central scanning spot #5 have a strong ability of collecting the overall vegetation information in the sample plot #2 (Table.3), but some scanning spots (like the scanning spot #1) have a small contribution, which could be replaced by other scanning spots (like the scanning spot #2), because the location of scanning spot #1 and #2 belongs to windward slope, and there are more vegetation around the scanning spot #2 (Fig. 4b). Scanning spot #4 was located at the peak of the slope (Fig. 4a) and had an excellent scanning vision, which was the connection scanning spot of combining the data of windward slope and leeward slope, and compensated for the lack of information due to the uctuation of terrain. From this result, we can infer that if the central scanning spot #5 moves down to the position of the slope peak (Fig. 4a), it can make the data quality collected by SS mode better, because of the supplement to the vegetation information on the leeward slope. Scanning spot #3 mainly provided the vegetation information of the leeward slope, because of the far distance from scanning spot #4 and the occlusions caused by the tall vegetation on the leeward slope, the complete vegetation information of the leeward slope cannot be obtained by the single scanning spot #4, so the scanning spot #3 is very important for the data collection of sample plot #2.

Suggestions for setting the scanning spots of TLS
For the forest inventories, Abegg et al. (2017) have explained the in uence of the scanner placement on the quality of the scans in terms of completeness, and found that stand describing parameters (e.g. diameter distribution, stem number and dominant diameter), distance to the scanner, and object size all have an impact on data acquisition quality. Too many factors will increase the di culty of data collection and test the data acquisition experience of the collector. But for the low-height vegetation, we just need to consider the terrain factor and the occlusion in the scanning direction. Because most of the occlusion problems have been addressed by the height difference between the scanner and the vegetation, if the scanner is at the same height level as the vegetation, or below the height level of the vegetation, it will face the same problems encountered by forest inventories.
In addition, we summarized some key points of setting scanning spots according to the different requirements of data collection and the different situation of the sample plot. (1) The scanning mode of SS is suitable for the rapid acquisition of vegetation information in the sample plot, which has a low requirement for the extraction accuracy of vegetation parameters. Besides, the location of central scanning spot should consider the uctuation of terrain rather than the slope, and it is best to choose the slope peak near the center of sample plot, which could reduce the occlusion of the terrain. (2) The scanning mode of MS could be used when we need the vegetation parameters with high precision. Firstly, we need survey the situation of terrain and vegetation before setting up the scanning spots in the sample plot, and determine the central scanning spot based on the SS method preferentially. Secondly, the determination of peripheral scanning spots should consider the occlusion caused by the terrain and vegetation in the scanning direction. Therefore, the peripheral scanning spots need to be located in the areas with different terrain features (including windward slope, leeward slope, and slope peak). Finally, if some large vegetation was found in one scanning direction of the central scanning spot, a peripheral scanning spot need to be set in the opposite direction. (3) Taking the equipment used in this study as an example (RIEGL VZ-400i), we need pay attention to the size of the sample plot or the distance between the scanning spots. The size of the sample plot (square) should not exceed 50 m for the scanning mode of SS (Table.3), or the information on the edge of the sample plot is not reliable. In addition, the distance between the adjacent scanning spots should also not exceed 50 m for the scanning mode of MS, to ensure the repeatability and complementarity of the point cloud. Finally, it should be noted that the reasonable distance between scanning spots varies greatly depending on the speci c parameters of the scanner, it is necessary to test the reasonable distance between scanning spots before the formal data acquisition.

Conclusion
In this paper, we evaluated the applicability of the SS mode by comparing the vegetation parameters (number, height, and crown width) extracted with the scanning mode of MS, and analyzed the in uence of terrain features for setting scanning spots. We con rmed that the SS mode has strong applicability in the eld-survey of low-height vegetation. However, the accuracy will be affected by the size of sample plot and the occlusion caused by the terrain and vegetation. In addition, we have provided a relatively complete method to set up the scanning spots for SS and MS modes, and we believed that the quickly acquirement of the terrain information is a prerequisite for reasonably determining the location of the scanning spots. Therefore, the adoption of unmanned aerial vehicle for terrain data collection before the TLS scanning can greatly reduce the cost and improve the collecting e ciency.   The distribution of quadrats in the sample plot 1.

Figure 3
Page 16/17 the diagrammatic drawing of the slope peak