A voxel-based fine-scale 3D landscape pattern analysis using laser scanner point clouds

SUN Hongzhan and WU Qiong

College of Geo-Exploration Science and Technology, Jilin University, Changchun 130026, China

Abstract: The landscape pattern metrics can quantitatively describe the characteristics of landscape pattern and are widely used in various fields of landscape ecology. Due to the lack of vertical information, 2D landscape metrics cannot delineate the vertical characteristics of landscape pattern. Based on the point clouds, a high-resolution voxel model and several voxel-based 3D landscape metrics were constructed in this study and 3D metrics calculation results were compared with that of 2D metrics. The results showed that certain quantifying difference exists between 2D and 3D landscape metrics. For landscapes with different components and spatial configurations, significant difference was disclosed between 2D and 3D landscape metrics. 3D metrics can better reflect the real spatial structure characteristics of the landscape than 2D metrics.

Keywords: 3D landscape metrics; 3D laser scanner; voxel; point clouds

0 Introduction

A voxel-based landscape model based on point clouds was constructed in this study, and a series of voxel-based 3D landscape metrics were constructed to analyze the characteristics of 3D landscape pattern, and the landscape pattern analysis results in 3D space were compared to that in 2D space to verify the feasibility of 3D voxel landscape pattern analysis.

1 Methods

The study area is located in the Yuhuayuan of Changchun City (Fig. 1a). According to the landscape composition and spatial configuration characteristics, the study area is divided into 3 zones (Fig. 1). Point clouds were collected by laser scanner and were divi-ded into three classes including low vegetation (0.2--3.5 m), medium height vegetation (3.5--8.5 m), and high vegetation (above 8.5 m) (Fig. 1b). The raster map was obtained by vectorizing the classified point clouds (Fig. 1c).

(a) Aerial image; (b) point clouds; (c) raster map; Ⅰ-zone 1;Ⅱ-zone 2; Ⅲ-zone3.Fig.1 Schematic diagram of geographical location of the study area and landscapes

1.1 Landscape voxel model construction

The construction of landscape voxel model is the process of voxelization of point clouds in 3D space. In this paper, the voxel model is established based on the spatial grid division method. The space was divi-ded into equal-spaced spatial grids according to a certain resolution, and the voxels of landscape was defined by the cube with point clouds. Considering the original data point cloud density and the detail of the 3D landscape, the voxel model of this research is constructed with the resolution of 0.1 m (Fig. 2).

1.2 Metrics construction and calculation

In this study, a patch was defined as a stereoscopic patch that is not voxel-connected with other patch of the same type in 3D space. 3D landscape metrics includeCA(CArepresents the total surface area (m2) of the focal class),PLAND(PLANDrepresents the proportion of the focal class in the landscape),NP(NPcan be used as a simple measure of the fragmentation of the focal class),DIVISION(DIVISIONrepresents the probability that two randomly chosen voxels in the landscape are not situated in the same patch of the focal class, and is affected by the proportion, fragmentation and patch size of the focal class.DIVISION=0 when the focal class consists of a single patch and approaches 1 as the proportion and patch size of the focal class decrease) andAI(AIis a mea-sure of the aggregation and compaction of patches in the focal class.AI=100 when the focal class consists of a single patch, and decrease as the aggregation and compaction of the patches in the focal class decrease) were constructed based on 3D surface area, volume, and adjacency (Table 1). The calculation results of 3D metrics were compared with that of 2D metrics to analyze the landscape pattern.

(a) High vegetation; (b) medium height vegetation; (c) low vegetation.Fig.2 Voxel model

Table 1 3D landscape metrics based on voxel model

2 Results and discussion

The difference in the calculation results of 2D and 3D landscape metrics is mainly attributed to the difference in the relative position, spatial extent, compaction level, and space proportion of the patches in 2D space and 3D space, which leads to different change trend between 2D and 3D landscape metrics among the 3 zones with different landscape components and spatial configuration characteristics (Table 2), which are represented by:

(1)CA

3DCAcalculates 3D surface area of the patch, while 2DCAcalculates plane area of the patch, which is the projected area of 3D patch on 2D plane, result-ing in greater 3DCAcalculation result of each class than 2DCAcalculation result.

(2)PLAND

In zones 1 and 3, the patch compaction level of the upper-layer vegetation patch is high, and the height is higher than the lower-layer vegetation patch. After the vertical height information is introduced, the proportion of 3D volume of the upper-layer vegetation classes (the high vegetation of zone 1, medium height vegetation of zone 3) is greater than the proportion of 2D area, and the proportion of 3D volume of other classes in the corresponding zone is less than the proportion of 2D area. The patch compaction level in the upper classes of zone 2 is low, and the patch compaction level in the lower classes is high, resulting in smaller proportion of 3D volume of upper-layer classes in zone 2 than the proportion of 2D area, and the proportion of 3D volume of other classes is greater than that of 2D area. As a result, 3DPLAND< 2DPLANDin medium height vegetation and low vegetation of zone 1, medium height vegetation of zone 2, low vegetation of zone 3; 3DPLAND>2DPLANDin high vegetation of zone 1, medium height vegetation and low vegetation of zone 2, medium height vegetation of zone 3.

(3)NP

The patches of upper classes have a high fragmentation, which divides the complete patches of lower classes into multiple small patches, resulting in 3DNP< 2DNPin low vegetation of zone 3.A patch in the projection direction (2D plane) may be composed of multiple patches interlaced in the vertical direction in 3D space, and the covering effect of the upper classes on the lower classes lead to 3DNP> 2DNPin medium height vegetation and low vegetation of zone 1, all classes of zone 2, medium height vegetation of zone 3.

(4)DIVISION

In high vegetation of zone 1, the calculation result of 3DPLANDis greater than 2DPLAND, which leads to 3DDIVISION< 2DDIVISIONin high vegetation of zone 1. From 2D plane to 3D space, the proportion of focal class decreases (3DPLANDis smaller than 2DPLAND), the uniformity of patch size increases, and the fragmentation level increases (3DNPis greater than 2DNP), resulting in the 3DDIVISION> 2DDIVISIONin medium height vegetation and low vegetation of zone 1, all classes of zone 2 and zone 3.

(5)AI

The adjacency number of similar pixels in 2D plane calculates the adjacency of the pixels in four directions in the plane. The pixels inside the patch are all "full adjacency", while in 3D space, the voxels calculate the adjacency of the voxels in six directions. At the same time, the spatial shape of 3D patch is complicated, and the voxels inside the patch are not all "full adjacent" voxels, resulting in smaller 3DAIthan 2DAIin all classes of all zones.

Table 2 Calculation results of 2D and 3D metrics

Compared with 2D landscape metrics, 3D landscape metrics can quantitatively evaluate the characteristics of the landscape pattern in 3D space, whereas 2D metrics may misjudge the spatial heterogeneity of the landscape in some cases, such as:

(1) Misjudgment of the proportion of landscape components: in 2D calculation result ofPLANDin zone 1, medium height vegetation accounted for the highest proportion (PLAND=38.0%), which is the dominant land cover type in the zone, but 3D calculation results show that high vegetation occupies a larger proportion of the space (PLAND=44.9%).

(2) Misjudgment of the level of aggregation and compaction of patches: due to the large number of gaps in the understory structure, the adjacent conditions between various parts of the ground features in 3D space are complicated, 2D metrics will misjudge the level of aggregation and compaction of various landscape patches. In zone 2, 2DAIof high vegetation is the highest (AI=99.42), but 3DAIof high vegetation is smaller than that of the other classes in zone 2 (AI=37.91).

3 Conclusions

Based on the point clouds, this paper constructed a high-resolution voxel model, compared the calculation results of 2D and 3D metrics of each zone in the study area, and verified the reasonableness of above evaluation method. According to the experimental results, the following conclusions are drawn:

(1) The voxel model is convenient for metric calculation, which can accurately describe the hierarchical structure of the landscape, and has a strong ability to express the characteristics of the space under the landscape. The high-resolution voxel model can better engrave the details of landscape entities and can be used for fine-scale 3D landscape pattern feature analysis.

(2) Compared with 2D landscape metrics, 3D landscape metrics can accurately reflect the relative position, spatial extent, compactness level, space proportion and adjacent status of the patch, without misinterpretation of landscape features, and can reflect the characteristics of the true 3D landscape pattern.