OPTIMUM NUMBER AND POSITIONS OF TERRESTRIAL LASER SCANNER TO DERIVE DTM AT FOREST PLOT LEVEL

Remote sensing of forest landscape is sometimes quite challenging, considering the rugged terrain, remote areas and presence of dense canopy. LiDAR is capable of generating dense three-dimensional (3D) structure of the forest. But when research is focused on a small area that needs dense information Terrestrial Laser Scanner (TLS) proves to be an efficient way of acquiring the information. TLS is an instrument used to acquire data from the ground surface level. Ground information and under canopy structure are acquired with more accuracy and precession with TLS when compared to aerial remote sensing techniques. When a TLS survey is done, scan positions and the number of scans are important parameters to consider to acquire datasets of forest plots. In this research, an analysis is carried out to determine the optimum number of TLS scans and positions that are needed to sufficiently generate the number of ground points to generate a Digital Terrain Model (DTM). The research considers two forest plots of 25m x 25m with different tree densities. Different TLS scanning combinations are used to determine the best scanning pattern for DTM generation at the forest plot level.


INTRODUCTION
Generating Digital Terrain Model (DTM) has always been an essential part of the survey for many years. As technology advances, many techniques have been emerged and developing rapidly. An advanced remote sensing technique is Light Detection and Ranging (LiDAR). Terrestrial Laser Scanner (TLS) is one instrument based on LiDAR technique that follows the time-of-flight principle (Kushwaha, 2018). TLS is used to generate an accurate dense three-dimensional (3D) model from the ground perspective. Its dataset is also used to produce high-resolution DTM along with other applications. It is a very effective tool in the forest under canopy data acquisition compared with Un 1 manned Aerial Vehicles (UAV) based photogrammetry or any other airborne remote sensing techniques which can help in forest inventory, forest management and planning.
In a forest, plot size is initially determined, and considering the size of the plot, a number of TLS scans and their positions are planned. (Ritter et al., 2020) tried to predict different plot sizes and scanner position layouts in multi scan settings for tree detection probability. Dense 3D point clouds are generated from each TLS scan position which are further * Corresponding Author combined with all other TLS scans to produce a complete 3D point cloud dataset of the forest plot. Considering the high density of the points, the DTM generated from terrestrial LiDAR is considered to be very accurate.
Many factors can affect the quality of DTM produced in forest plots, such as the number of scans, positions of scans, tree densities, co-registration accuracy of individual scans, sensor coverage, etc. Similar research was carried out to study the parameters affecting the quality of DTM from Airborne Laser Scanning (ALS) (Hyyppä et al., 2005). The high-quality DTM is required in the forestry even to use as a reference when integrating with other datasets. In forestry accurate DTM is also used to produce Canopy Height Model (CHM) which helps in providing forest metric parameters (Singh & Kushwaha, 2021).
In this research, we have tried to focus on the number and positions of TLS scans that can produce suitable DTM in two forest plots with different tree densities. Likewise, other research has been carried out stating that close range remote sensing techniques are capable of achieving higher accuracy when compared to Airborne Laser Scanning (ALS) and Digital Aerial Photogrammetry (DAP) (Jurjevic et al., 2021).
LiDAR is capable of getting points in highly vegetated areas, One such research highlights the process of DTM generation in detecting archaeological earthwork anomalies (Limongiello et al., 2020). Different algorithms and techniques were developed to improvise the accuracy of the DTM. The evaluation of the quality of ground surfaces generated from TLS is also important for the accuracy of DTM produced (Sun, 2019). The accuracy of DTM produced from different techniques like TLS, ALS, UAV-DAP were also tested using a different algorithm and the analysis is extensively explained (Crespo-Peremarch et al., 2020). The research was carried out to evaluate the DTM quality obtained from Airborne Laser Scanners (ALS) for two seasons (spring/summer) in a year and was validated with 95 checkpoints acquired from surveying methods (Stereńczak & Kozak, 2011). The form of terrain also plays an important role in DTM generation, So, high-resolution topography survey was carried out over an 8 km 2 with 81 different TLS scan positions to analyze the terrain variables, Aspect, Slope and hillshade in Keller Peninsula region (Schünemann et al., 2018). Not only in the vegetation areas LiDAR data derived DTM can also be used to improve the geometric data of hydraulic modeling, which helps in recognition of fluvial landforms and estimation of flood inundation extents (Hohenthal et al., 2011).
TLS is also used to generate quantitative structure models (QSMs). QSMs are very accurate in order to calculate forest above-ground biomass. To do so, the tree volume and wood density are measured (Demol et al., 2021). Later the Quality of DTM produced by TLS data is analyzed (Costantino & Angelini, 2013). TLS can also be effective in mapping tree stem in dense forest canopy cover (Tao et al., 2021).
Not only in forestry, TLS improves Digital Elevation Models (DEM) and can also help in topsoil pH modelling with a complex topography and dense vegetation (Baltensweiler et al., 2017). Hence, the most reliable way of generating DTM is needed. Defining the number of scans can minimize the workload in the field also save energy. The fixed scans with the different sets of plots can help to get an idea of the scan to cover the whole plot and all the trees. This also helps to minimize the occlusion effect in the plot.

STUDY AREA
In this research, two forest plots were chosen to test the hypothetic idea. The plots were considered with different tree densities plot 1 with 49 trees and plot 2 with 102 trees. Forest plots considered in this research are located in Zvolen, the central part of Slovakia. The location of the study area is represented in Figure 1.

DATASET USED
The dataset required for this research was acquired with TLS I n pre-established plots. A total of 9 scans were done in each plot. Forest plots were considered with dimensions of 25 x 25 m, plots were scanned from nine scan positions. The pattern of all the nine scan positions in the forest plots is depicted in Figure 2. Spherical reflectors were used in the forest plots for coregistration of the individual scans. Ground markers were installed in all four corners of the plots for identification in the pointcloud datasets. After pre-processing the individual scans were used to produce the 3D point cloud dataset of the overall plots, respectively. Other six markers (Checkerboards) were also placed facing the centre scan position which was used to geo-reference the point clouds.

RESEARCH HYPOTHESIS
The research hypothesis is how many TLS scans and their spatial combination are sufficient to generate detailed and accurate DTM with less data absence and redundancy.
We came up with six possible combinations that can be considered in this research. The first combination is only with the Centre Scan (CS). The second combination is with all the Four Corner Scans (FCS). The third combination is with Four Corner with Centre Scans (FCwCS). The fourth combination is with Four Sides Centre Scans (FSCS). The fifth combination is with Four Sides Centre with Centre Scans (FSCwCS). Finally, the sixth combination is with All Nine Scans (ANS) in the forest plot which is also used as the reference scan combination dataset. Figure  Case-I, When CS combination is considered, there are high chances that the areas near the corners may miss data either due to occlusion or due to less density of points as the distance increases radially from the TLS. Case-II, when the FCS combination is considered, there is a high probability of missing the plot's central area, but with more details in the corner regions. Case-III, when FCwCS is considered chances to cover the whole plot are increased. Case-IV, when FSCS is considered again there are high chances that the centre area data may get missed. Case-V, when FSCwCS is considered, then most parts of the whole plot are covered. Case-VI, when ANS is considered whole plots are covered like in case-III and case-V but with an increase in data and regions. As the number of scans is increased the region of coverage increases and density of points increases enormously and data redundancy also increases. The total number of sphere of influence of TLS scan positions are 1, 1, 2, 2, 3, 4 respectively in case -I, II, III, IV, V, VI respectively. Case-III and case-V have an equal number of scans, but TLS scans position has a good advantage in case-V, as it has one extra sphere of influence with same number of scans. Theoretically, the authors would like to predict that case-V i.e., combination FSCwCS would be more suitable to generate a sufficient number of ground points to produce DTM of the forest plot.

RESEARCH METHODOLOGY
Initially, Forest plots were identified for the proposed research. TLS survey was carried out in both the plots identified in the forest with different tree densities.
Pre-processing of the TLS scans were done in FARO Scene software. Then the different scans were combined depending upon the possible scan pattern combinations as mentioned in Figure 3, ground points were filtered out from the point clouds obtained from all the six combinations using LAStools. Ground points were interpolated, and DTM was produced for all the possible combinations (for illustration refer Figure 5). DTM produced from the ANS combination is used as a reference and a DTM comparison is made with DTM's obtained from other scan combinations. The methodology is represented in the workflow in Figure 4.

RESULTS
Ground points are filtered out from forest plot point clouds. The ground points are interpolated with the average elevation points wherever the data is missing in each scanning combination mentioned in earlier sections. The filtered ground points obtained in all the scan combinations are also shown in Figure 6, along with the percentage of ground coverage. The ground points are further used to produce DTM's, respectively. Among the DTM obtained in each scanning combination, it is challenging to inspect which DTM produced is close to the reference DTM obtained from the ANS combination. So, the relative difference is calculated between all the scan combinations DTM's to the reference DTM. The results obtained are shown in Figure 7.
Hypothetically scan combination FCwCS and FSCwCS produce good results as both the combinations have five scan positions each. However, among both the combinations FSCwCS has produced the best results which is evident from Figure 7 and the histogram values obtained in Figure 8.

Ground Coverage
The portion of ground covered in each scanning combination pattern is shown in Figure 6, and it is evident that with the increase in the number of scans, the percentage of ground cover has increased. Ground coverage is higher in plot 2 comparatively to plot 1 despite nearly double the number of trees. This is because the trees present in plot 2 are less thick than the trees in plot 1. FCS and FSCS both have four scans, but the position of scans has improved in FSCS, similarly, FCwCS and FSCwCS both have five scans, but the ground cover is more in FSCwCS. This is evident from Figure 6 for both plots 1 and 2, respectively.

Ground Filtering
We used lasground_new tool from LAStools plugin in QGIS software to filter out ground points from the overall point cloud of forest plots. The tool classifies the point cloud into the ground and non-ground classes.

DTM Generation
The ground points obtained in each scan combination do not have full coverage due tree stems, fallen logs, big stones, occlusions beyond the tree stems, etc. Therefore, the filtered ground points are used to interpolate and produce DTM.

DTM Assessment
In this research, we have considered the ANS scan combination as the reference scan combination as it has the highest number of scan positions and percentage of ground coverage. So, the DTM's obtained from all other scan combinations are comparatively analyzed with DTM obtained from ANS combination to find out the equal number of points which are common ground points. This can be further understood from the graph in Figure 8.

CONCLUSIONS
We have presented the possibility of ground coverage in each scanning combination and DTM generated in two forest plots with different densities. From the results, obtained in both plots with different tree densities and as per the results obtained it is evident that among all the six scan combinations considered for this research Four Sides Centre with Centre Scans (FSCwCS) is the best suitable scan combination that can be used to generate DTM as close to All Nine Scans (ANS) combination without increasing more scans, data redundancy and size of the data.
The main advantages of minimizing the number of TLS scan positions are i) Reducing the data acquisition survey time. ii) Reduces the data redundancy; as we keep increasing the number of scan positions it is likely that we can cover the missing areas from other scans but it also highly increases the redundancy of points in the datasets by scanning the other same region many times. iii) As the number of scans is minimized the overall size of the dataset is also reduced which further helps in processing and handling the datasets.
The authors would like to recommend that if the forest plot is around 25 x 25 m and the aim is to produce DTM, then the FSCwCS scan pattern combination can be sufficient to fulfil the required objective effectively.

FUTURE SCOPE
Further, the authors would also like to investigate the effect of scanning pattern combinations and the number of scans on the under canopy information like stems, leaves, branches, etc., and also for individual tree detection and tree parameters retrieval.