Quantitative Analysis Method of the Organizational Characteristics and Typical Types of Landscape Spatial Sequences Applied with a 3D Point Cloud Model

Abstract

(1) Background: Sequential landscape changes give people experiences of dynamic beauty, and the key to creating spatial sequences lies in the organization of spatial changes. The purpose of this study is to use a 3D point cloud model to achieve a refined description of spatial sequences’ organizational characteristics from information acquisition to description and explore the quantitative interpretation methods of typical sequence organizational characteristics. (2) Methods: The proposed model contains three main steps: data acquisition and extraction, characteristic index system construction and data processing, and quantitative characterization analysis. A typology research method that combines quantitative induction with qualitative statistical verification is proposed. (3) Results: Seventy-two spatial sequence point cloud models of study cases are obtained; 4 indicators are established; 3 typical organization types are summarized, namely fluctuating, reversal and moderate type; and the characterization factors and threshold intervals for each sequence organization type are analyzed to validate the type classification result. (4) Conclusions: This research improves the accuracy of spatial data, the comprehensiveness of sequence organization characterization factors, and the reliability of classification results. It supplements the existing spatial sequence theoretical knowledge system and provides parameters that can be referred to in practical design.

1. Introduction

Sequence, as an expression of spatial change, is essential to the dynamic qualities of space composition [1]. A landscape spatial sequence involves organizing a number of adjacent spatial units according to a certain order and flow [2,3]. A specific sequence relationship is established based on the interaction and interdependence of landscape units. Previously, studies on landscapes spaces were based mainly on qualitative descriptions, and the accuracy reported in some quantitative studies was insufficient. In the context of constructing high-quality human settlements, traditional spatial sequence description and analysis methods can no longer meet the needs of refined spatial analysis. The development of spatial information acquisition, modeling, and analysis technologies such as three-dimensional (3D) laser scanning has provided strong technical support for refined landscape analysis [4,5,6]. Therefore, in line with the development trend and demands of high-quality space creation in the digital age [4], the purpose of this paper is to explore a new method for quantitatively interpreting the landscape sequence organizational characteristics based on 3D point cloud technology and expand the application prospects of 3D point clouds in the study of landscape spaces. By exploring new methods of landscape spatial analysis with point cloud data, we hope to meet the demand for refined quantitative analysis in the field of high-quality space design [7].

1.1. Theoretical Studies on Spatial Sequence

Existing theoretical studies on landscape spatial sequences are mostly focused on multi-space organizational structures, spatial organizational characteristics, and visitor experiences. From the structural perspective, a sequence is an entity defined by the “path” created by the viewing route and the “nodes” encountered along the route [2,8,9]. Nodes refer to the overall external space enclosed by vegetation, terrain, water, and other landscape elements. Through the connection of routes, several spatial units are connected in a specific order, realizing the transformation from disordered units to an ordered spatial structure [10,11]. From the organizational characteristics perspective, a spatial sequence has dynamic organizational characteristics such as multiple spaces, multiple viewpoints, and continuous changes [1,8,9]. A spatial sequence is not a number of purely static 3D spaces but rather a multidimensional dynamic space that unfolds with the movement of people [12,13]. Dynamic organizational characteristics differentiate sequences from static space organization. For example, scholars often use the term “different scenes with space changes” to summarize the space characteristics of sequences. “Space changes” refer to spatial transformations in which the space unfolds along the path, and “different scenes” refer to the dynamic changes in multiple spaces. In addition, dynamic changes have composite characteristics; that is, changes in the interface, enclosure, scale, density, shape, etc., between spaces all determine the organization and presentation of the sequence as a whole [1,14]. From the experience perspective, people’s experience is determined by the overall characteristics of multi-space changes, and the feeling of overall organization change is greater than the sum of the experience of some units [3,14]. That is, compared with the experience of a single space, the impression and feeling of the sequence experience of tourists are determined by the overall characteristics of the arrangement of several spaces.

Classification and induction are common methods for studying the sequence organization theory of landscapes. Type induction works by identifying common attributes of things and building sets. It is usually based on a series of existing things or materials to find the common laws they obey. In existing studies, attempts have been made to discuss the typical sequence types of multi-space from various perspectives. Based on the path layout pattern, it can be classified as closed circular, tandem, and central sequences [9]. Robinson likened it to rhythm in music and divided sequences into types with regular accent, with patterns of repetition, or deliberately chaotic according to the rhythm features [1]. Similarly, Sun also compares a landscape space sequence to a drama script and divides it into two types, a three-stage model and a two-stage model, according to the layout development model [11].

Existing theoretical research can clarify the following two points: ① The dynamic change determined by several spatial units and paths is the core feature of spatial sequence organization; ② Type induction is a widely used theoretical research method. However, there is a lack of research that classifies the common characteristics of sequence organization from the perspective of spatial dynamic changes.

1.2. Quantitative Studies of Spatial Sequences

With the development of quantitative analysis methods, an increasing number of scholars have attempted to quantitatively interpret the organizational characteristics of spatial sequences. In terms of analysis methods, some studies have used the horizon analysis method of space syntax to investigate visual perception characteristics in spatial sequences through changes in the visual range and depth of space [15,16,17]. Some studies have quantitatively described the organizational characteristics of multiple spaces from the perspective of the morphological composition of pure materials; that is, the standard deviation of a single factor, such as the area of several spatial units in a sequence and the ratio of the depth or width of a public space (D) to the height of its surroundings (H) was used to express the changes in multiple spaces [18,19]. However, the information source of the above studies was mainly two-dimensional (2D) planar information, and the acquisition of 3D information mostly relied on field surveys and measurements or modeling software simulations. In addition, the spatial information used for quantitative analysis includes information only on artificial elements such as roads, squares, and buildings. Therefore, these methods fail to consider the 3D spatial morphological characteristics under vegetation cover.

In this study, we intended to refer to the research method of quantitatively describing the change in spatial morphology. However, existing quantitative research has the following two limitations: ① Lack of indicator factors system that can describe the composite spatial variation characteristics of spatial sequences; ② Insufficient spatial data accuracy and low acquisition efficiency; ③ Limited number of cases (mostly single cases), and fail to attempt to validate the method in multiple cases. The variation of a single indicator cannot comprehensively describe the organizational characteristics of a sequence space. The change relationships of sequence spatial units are reflected at multiple levels, such as interface, enclosure, scale, density, and shape [20]. Additionally, it is difficult for traditional spatial information acquisition methods to precisely and accurately record the vegetation and external 3D spatial morphology shaped by vegetation. The height, crown width, and morphological characteristics of individual vegetation types as well as the overall canopy closure and canopy line of the vegetation community all impact the external form of a landscape space [21,22,23]. The lack of 3D morphological information and the limitations of data sources have led to low spatial data acquisition efficiency and insufficient quantitative analysis accuracy, which directly affect the comprehensiveness, accuracy, and credibility of quantitative studies on spatial forms.

1.3. Application of 3D Point Cloud Data in Landscape Spatial Research

3D laser scanning technology has a prominent advantage in the acquisition of spatial information about complex substances. The working principle is to record the spatial position information of scanned surface objects in the form of a 3D point cloud through the reflection of a continuously rotating laser target [24,25]. This technology can quickly acquire massive 3D coordinate data (X, Y, Z), has the advantages of high efficiency, high precision, measurable and editable data, and high compatibility, and has been widely used in forestry surveys, architectural heritage protection, digital urban areas, and other research areas [26].

At present, the application of point cloud models in the landscape field has focused mainly on the visualization and presentation of 3D space and recording and identifying morphological characteristics [27,28,29,30] that correspond to the site survey and information acquisition stage in the early stages of design. Several researchers have also relied on 3D point cloud models to carry out spatial quantitative analysis, such as 3D green quantity analysis [31,32], visibility analysis [33,34], and climate simulation analysis [35,36,37,38]. There are also studies trying to use 3D point clouds to achieve 3D visual intelligent detection and analysis [39]. The efficiency, accuracy, comprehensive results, and feasibility of spatial quantitative analysis based on 3D point cloud data have been verified. In addition, the 3D point cloud format has the advantages of being classifiable, editable, and segmentable and has good compatibility with various spatial analysis and processing platforms. A 3D point cloud model can provide accurate spatial morphological information and meet the demands for refined spatial information acquisition, processing, and analysis.

1.4. Summary

Based on the consensus of existing theoretical research and the shortcomings of quantitative research, it can be raised that how to use 3D point cloud information to quantify the dynamic change characteristics of spatial sequences and investigate the quantitative characterization of typical types is the main content of this study. This article aims to improve the accuracy of spatial data, the comprehensiveness of sequence organization characterization factors, and the reliability of typical type induction methods. To achieve this goal, the following three key questions remain to be solved:

• How to obtain the overall spatial point cloud information of landscape space and extract spatial sequence models composed of units and paths?
• What factors can describe the organizational characteristics of spatial sequences?
• Can different organization types be described by factors and their threshold intervals?

In this study, based on the acquisition of overall spatial 3D point cloud models, a definition method for spatial sequences is proposed, an indicator system of spatial sequence organizational characteristics is established, and the conversion of spatial 3D point cloud model information into digital index information is realized. Then, by combining qualitative classification with a data quantification method, and through a comparative analysis of the quantitative characterization of a large amount of case data of different types, the spatial sequence organization types are classified from the perspective of spatial dynamic changes and described in the form of factor threshold intervals. The research intends to explore the quantitative research methods on the landscape spatial sequences in the digital technology era and to provide novel insight into related spatial research.

2. Materials and Methods

2.1. Study Cases

The landscape space studied in this article is located in the built environment and is created artificially, which is different from the natural landscape spaces shaped by purely natural forces. The built environment is shaped by both natural and artificial elements and is created based on human activity needs. It reflects the interaction between human design intervention and the natural environment. In addition, the special topography shaped by natural forces will affect the presentation of external spatial morphology. The landscape spaces with special landforms such as mountains, wetlands, farmland, terraces, etc., are not included in this study. Thus, the research methods and conclusions are only applicable to the abovementioned landscape space types.

In this paper, we selected 14 urban parks located in the built environment as study cases (

Table 1. Study cases.

Park Name	Abbr.	Location	Area (ha)
QingLv Garden	QLY	Nanjing	26.97
Hexi Ecological Park	HX	Nanjing	22.12
Expo Park	LBY	Nanjing	67.22
Vondel Park	VP	Amsterdam	44.92
Rembrandt Park	RP	Amsterdam	48.27
Beatrix Park	BP	Amsterdam	28.40
Ooster Park	OP	Amsterdam	11.65
Sarphati Park	SP	Amsterdam	5.02
Morningside Park	MP	New York	20.63
Constitution Gardens	CG	Washington D.C.	19.93
West Potomac Park	WPP	Washington D.C.	68.34
Gold Star Family Park	GSFP	Chicago	5.15
Miller River Park	MRP	Stamford	7.76
Buffalo Bayou Park	BBP	Houston	72.32

). They are artificial external spaces created for the purpose of public leisure activities and scenic sightseeing, and with no special natural landforms. These cases are also highly recognized by the public or have been awarded. Moreover, the spatial data of study cases are obtainable, that is, the site is outside the drone no-fly zone, or the area is where public spatial data are available.

2.2. Workflow

To address the key questions summarized above, the methodological framework of the present study (Figure 1) includes three main steps: point cloud data acquisition and extraction, characteristic index and data processing, and typical sequence type characteristic quantitation. From information acquisition to quantitative expression, a refined interpretation of the spatial sequence organizational characteristics and typical landscape patterns is achieved.

2.3. Point Cloud Data Acquisition and Extraction

Obtaining point cloud information that records spatial sequences is the first step in research. However, what is acquired by UAV or laser scanner is the overall model of the external space. Therefore, it is necessary to propose a method to define and extract the components of the spatial sequence to obtain a point cloud model that meets the analysis requirements.

2.3.1. Overall Space Information Acquisition

Both airborne light detection and ranging (LiDAR) scanning and unmanned aerial vehicle (UAV) oblique photography can achieve the rapid acquisition of medium- and large-scale landscape space information, such as information about urban parks and public green spaces [25,26,36]. Airborne LiDAR scanning equipment can directly obtain the space coordinate set of sampling points on the surface of a measured object through the execution of a flight task. The data acquired by the oblique camera of a UAV include image data and positioning and orientation system (POS) data. In the context capture modeling platform, the original data acquired by oblique photography can be generated through 3D aerial encryption, triangulation model construction, texture mapping, and other modeling steps to generate a 3D spatial reality model, which is ultimately output in the LASer (LAS) 3D point cloud format. The point clouds that record different types of space elements are classified in ENVI LiDAR. The point cloud set can be classified into a point cloud set that records information about the land surface, vegetation, and buildings (Figure 2).

2.3.2. Spatial Unit Extraction

A landscape spatial unit is the smallest spatial node in sequence organization, construction, and design. A spatial unit is a spatial node where tourists can carry out activities, and it is composed of an inner domain and an enclosure interface that can accommodate activities [40]. The definition of a spatial unit should satisfy the following principles [41]:

• The internal domain of the unit can accommodate people’s activities and has an independent enclosure structure;
• It is connected by garden roads and is accessible;
• It meets the minimum external spatial modulus defined by Yoshinobu [42].

According to the above definition principles, the extraction of spatial units includes the following steps. First, in ArcMAP, through rasterization processing and an overlay analysis of the 3D point cloud model, the inner domain of the “virtual volume” space that can accommodate the activities is extracted. Next, the boundary range of the domain within the enclosed spatial unit is extracted through aggregation analysis. Finally, based on the unit area (400 m² was used as the minimum unit area modulus in this paper) and path accessibility, spatial units that play an important role in sequence organization and construction are further screened (Figure 3).

2.3.3. Spatial Sequence Acquisition

A spatial sequence is organized by a number of adjacent spatial units in a certain order. However, the number of sequences in green space is neither finite nor specific. In this paper, the extraction principles are further agreed upon as follows from four levels: the starting point, end point, path, and number of units.

• Starting point: The space at the main and secondary park entrances is the starting point of the sequence.
• Ending point: In addition to the spaces located at the main and secondary entrances and exits of the park, nodes in the park where tourists stay for a long time or with outstanding tourism value are included.
• Path: A path developed based on the first-level and second-level park roads.
• Number of units: In this paper, the combination of 4–7 spatial units is uniformly used as the modulus for sequence extraction.

According to the above principles, the point cloud model of the spatial sequence can be obtained by determining the starting and ending points and the number and order of the units connected by the path (Figure 4).

2.4. Characteristic Index and Data Processing

The method that converts the model information of a spatial sequence into index information based on 3D point cloud data is proposed in this section.

2.4.1. Characteristic Index for Sequence Organizational Characteristics

The organization of a sequence is not the simple superposition of several units. It is the creation of changes, which brings the transformation of adjacent units. Changes in spaces are dynamic, and the overall sequence is the superposition and integration of several changes. Therefore, in this paper, we first introduce the relative variance (RV) to describe the spatial variation characteristics of adjacent units; subsequently, the average RV (${AVE}_{RV}$), standard deviation of the RV (${SD}_{RV}$) and number of direction conversions ($N_c$) were used to describe the spatial organizational characteristics of the sequences as a whole.

• RV

Adjacent units can be considered units generated by spatial changes, and they are the basis for the generation of complex sequence changes. The morphological changes in adjacent units are complex and are reflected by five aspects: interface, enclosure, capacity, density, and shape. The RV was used as an indicator of the degree of composite change in adjacent units for each of the five morphological change aspects. The indicators permeation degree (PD), D to H (D/H), spatial volume (V), spatial density (SD), and extension degree (ED) were used to represent the spatial interface, enclosure, capacity, density, and morphological characteristics [20]. The selection of the above five factors is determined by the coefficient of variation of all case data. The higher the coefficient of variation indicates that the difference in this indicator between spatial units is the most significant, that is, this factor can better represent the spatial variation characteristics of the sequence. The change in the unit Um to Un was chosen as an example. The calculation formula of the RV is as follows:

$${RV}_{m-n}={PD}_{m-n}+{D/H}_{m-n}+V_{m-n}+{SD}_{m-n}+{ED}_{m-n}$$

(1)

And

$$X_{m-n}=\frac{2(X_n-X_m)}{X_m+X_n}\times \lambda$$

(2)

Xm = X factor attribute of unit Um

Xn = X factor attribute of unit Un

λ = Factor influence direction modification coefficient (λ = 1 or −1)

X ∈ {PD, D/H, V, SD, ED}

The sign of the RV represents the change direction of the adjacent unit space under the joint influence of multiple change levels. A larger absolute value of the RV represents a more obvious and intense change. A smaller absolute value indicates that the spaces are more similar or that the change trend is not obvious.

• ${AVE}_{RV}$

The ${AVE}_{RV}$ is an indicator that describes the average degree of change between adjacent units that constitute a series and is determined by the average of the absolute RV values of several adjacent units. These values are used in the calculation to avoid the influence of the positive and negative change directions on the calculation results. The calculation formula for the ${AVE}_{RV}$ is as follows:

$${AVE}_{RV}=\frac{\sum _{i=1}^n\left|{RV}_i\right|}{n}$$

(3)

${RV}_i$ = RVs of adjacent units

$n$ = number of adjacent units

• ${SD}_{RV}$

The ${SD}_{RV}$ is an indicator that reflects the difference in morphological changes among sequences and reflects the degree of dispersion of the RVs of several adjacent units. The larger the ${SD}_{RV}$, the more fluctuation in the morphological changes of the sequences. With reference to the calculation formula of the standard deviation, the calculation of the ${SD}_{RV}$ is as follows:

$${SD}_{RV}=\sqrt{\frac{{\sum }_{i=1}^n{\left(\left|{RV}_i\right|-{AVE}_{RV}\right)}^2}{n}}$$

(4)

${RV}_i$ = RVs of adjacent units

$n$ = number of adjacent units

• $N_c$

The $N_c$ is an indicator that reflects the frequency of morphological changes. The number of positive changes and the number of negative changes in the morphology of several adjacent spatial units that constitute the sequence were calculated. The more frequent side is the dominant directional trend of the sequence, while the less frequent side is the $N_c$. When the morphological change directions of all adjacent units are the same, $N_c$ is 0. The calculation formula for the $N_c$ is as follows:

$$\begin{gathered} if N_p\ge N_n N_c=N_n \\ if N_p\le N_n N_c=N_p \end{gathered}$$

(5)

Nc = Number of direction conversions

Np = Number of positive changes

Nn = Number of negative changes

2.4.2. Point Cloud Data Processing and Calculation

After obtaining spatial information about the landscape, this information was converted into indicator data according to the definitions of the indicators. The software platforms used in this operation process were ArcGIS 10.5 and CloudCompare V2.10. The 3D analysis and space analysis tool modules in ArcGIS can analyze and calculate the planar geometric form information and 3D spatial information. The built-in tool analysis module of CloudCompare can directly analyze and calculate the 3D spatial morphological information. The above data analysis platform can calculate the PD, D/H, V, SD, and ED of the spatial units. The quantitative representation of the characteristic index factors of each sequence can be obtained based on the algorithm formula. On this basis, a spatial variation chart can be established to assist in expressing the dynamic change characteristics of the sequence space.

2.5. Sequence Type Classification and Quantification

Classification and induction are effective ways to study the intrinsic patterns of complex entities. The classification of landscape spatial sequences involves the clustering and extraction of common organizational orders. The general classification method is mainly qualitative and consists of two steps: one is to abstract typical types from specific objects, and the other is to combine the types with scenes and restore them to specific forms. This article proposes an analysis method that combines qualitative classification with quantitative statistical analysis, which can not only summarize typical sequence types that are consistent with cognitive understanding but also verify the rationality of the classification results through statistical analysis results. The data analysis methods used in this research stage included the multiple independent sample Kruskal–Wallis (K–W) nonparametric test, multiple comparison analysis method, and threshold interval statistical analysis method (Figure 5).

The method consists of three steps. First, based on the dynamic change trend shown in the spatial variation charts, we summarize the typical sequence space organization types and their common characteristics and select the representative cases that best represent this type from the case library. Second, use the multiple independent sample Kruskal–Wallis (K–W) nonparametric test and multiple comparison analysis method to verify the reliability of the classification results. The K–W nonparametric test was used to test whether there were significant differences among the factors of different types. Multiple comparison analysis (homogeneous subset analysis) can divide a sequence type into several subsets according to the value characteristics of the factors in different classifications. The sequence types with no significant differences in factor values were classified into the same subset. The types with statistically significant differences were divided into different subsets. Finally, the percentiles of the quantitative characterization factors in different types were analyzed, and the common spatial composition patterns were expressed through threshold intervals. Through the above analysis, typical types of the sequence organization of a large number of similar cases were interpreted as the threshold interval of the variable factors.

3. Application and Results

3.1. Case Data Processing

The 3D model data of the cases located in Nanjing were acquired via UAV oblique photography, and the data were collected between May and July. The spatial 3D point cloud model of the foreign cases was obtained through the Agriculture, Nutrition and Health (ANH) Academy public database and OpenTopography database. The data showed that the acquisition time was distributed between May and September. To unify the accuracy difference between different data acquisition methods, we used the point cloud thinning processing of CloudCompare to unify the accuracy of the point cloud model in the distribution of 15–200 points/m². According to the spatial sequence extraction principles mentioned in Section 2.3.3, a total of 72 cases were extracted (

Table A1. 72 Sequences extracted from 14 study cases.

Park Name	Number of Sequences Extracted	Sequence Names a
QingLv Garden	6	QLY-SE (U1–U8), QLY-SE (U20–U15), QLY-SE (U28–U15), QLY-SE (U20–U12), QLY-SE (U18–U12), QLY-SE (U18–U15)
Hexi Ecological Park	4	HX-SE (U2–U6), HX-SE (U10–U6), HX-SE (U14–U4), HX-SE (U12–U6)
Expo Park	4	LBY-SE (U6–U12), LBY-SE (U15–U19), LBY-SE (U22–U25), LBY-SE (U41–U38)
Vondel Park	8	VP-SE (U32–U12), VP-SE (U32–U14) a, VP-SE (U32–U14) b, VP-SE (U33–U14), VP-SE (U33–U17), VP-SE (U4–U14), VP-SE (U4–U27), VP-SE (U32–U26)
Rembrandt Park	10	RP-SE (U1–U4), RP-SE (U1–U6), RP-SE (U10–U4), RP-SE (U15–U6), RP-SE (U16–U20) a, RP-SE (U16–U20) b, RP-SE (U16–U20) c, RP-SE (U16–U32), RP-SE (U32–U20) a, RP-SE (U32–U20) b
Beatrix Park	6	BP-SE (U1–U16), BP-SE (U1–U2), BP-SE (U12–U23), BP-SE (U15–U17), BP-SE (U21–U16), BP-SE (U21–U15)
Ooster Park	5	OP-SE (U2–U10), OP-SE (U3–U13), OP-SE (U10–U2), OP-SE (U13–U2), OP-SE (U11–U2)
Sarphati Park	3	SP-SE (U1–U3), SP-SE (U4–U1) a, SP-SE (U4–U1) b
Morningside Park	5	MP-SE (U14–U2), MP-SE (U1–U2), MP-SE (U19–U14), MP-SE (U30–28), MP-SE (U29–23)
Constitution Gardens	6	CG-SE (U1–U5), CG-SE (U12–U1), CG-SE (U12–U18), CG-SE (U20–U9), CG-SE (U9–U17), CG-SE (U1–U15)
West Potomac Park	5	WPP-SE (U1–U18), WPP-SE (U1–U15), WPP-SE (U8–U18), WPP-SE (U8–U15), WPP-SE (U25–U21)
Gold Star Family Park	2	GSFP-SE (U1–U7), GSFP-SE (U13–U7)
Miller River Park	2	MRP-SE (U8–U4), MRP-SE (U5–U11)
Buffalo Bayou Park	6	BBP-SE (U1–U10), BBP-SE (U11–U17), BBP-SE (U37–U43) a, BBP-SE (U37–U43) b, BBP-SE (U47–U53), BBP-SE (U56–U51)

^a If the first and last units of the sequence are the same, they are distinguished by tail labels such as a, b, etc.

). These sequences were named as follows: “park name abbreviation-SE (starting unit number—ending unit number)”. In platforms such as CloudCompare and ArcGIS, the value of each indicator can be calculated by processing the point cloud data. In addition, we used the composite chart of the spatial changes to express the organizational characteristics of the spatial sequence more intuitively. The size and sign of the interface, enclosure, capacity, density, and shape convergence of each adjacent unit are shown in a superimposed histogram, and the RV under the joint influence of the five indicators is shown in a line graph. The results showed that the ${AVE}_{RV}$ of 72 cases was in the range of 0.54–4.07, with a median of 2.30. The coefficient of variation was 0.36. The ${SD}_{RV}$ ranged from 0.64 to 4.34, with a median of 2.59 and a coefficient of variation of 0.35. The $N_c$ ranged from 0 to 3, with a median of 2 and a coefficient of variation of 0.44.

3.2. Typical Types of Landscape Spatial Sequences

Based on the sequence change trends and development characteristics presented in the composite chart of the spatial changes of the 72 cases, we further classified the typical spatial sequence organization types. The change trend refers to the trends of spatial change and development, and the development characteristics reflect the organizational order of the changes in a number of adjacent units. The typical organization types used for spatial sequences can be classified into three types: fluctuating, reversal type, and moderate type.

• Fluctuating spatial sequence

A fluctuating spatial sequence represents an organization type in which the number of adjacent units that constitute the whole sequence changes significantly, and the change direction is volatile. In this type, the relationship between adjacent spatial units sometimes exhibits a negative change and sometimes exhibits a positive change. These changes are not clear or consistent, but the uncertainty of these changes makes this sequence type more intense.

The spatial sequence RP-SE (U1–U4) in Rembrandt Park was chosen as an example (Figure 6). This case contains 6 spatial units, and the RVs of the adjacent units were 2.91→ −2.26 → 3.85 → −2.92 → 3.81. The $N_c$ of the RP-SE (U1–U4) was 2, the ${AVE}_{RV}$ was 3.15, and the ${SD}_{RV}$ was 3.02. The sequence started with a positive change. Then, the direction of change was continuously reversed, and it ultimately ended with positive change. In addition, the space transformation resulted in multiple fluctuations. In the space experience of the Rembrandt Park sequence RP-SE (U1–U4), people’s emotions rise and fall with the reversal of the spatial changes among the paths.

• Reversal spatial sequence

A reversal type refers to a spatial sequence with a consistent and significant change trend; however, when a heterogeneous mutation occurs in one or several consecutive adjacent units, the consistent trend is suddenly reversed.

The study case BP-SE (U15–U16) located in Beatrix Park in Amsterdam was used as an example (Figure 7). The BP-SE (U15–U16) is composed of four spatial units, and the RVs of the adjacent units are −2.10→−2.48→4.99. The $N_c$ of the BP-SE (U15–U16) was only 1, the ${AVE}_{RV}$ was 2.94, and the ${SD}_{RV}$ was 3.43. The front and middle segments of BP-SE (U15–U16) exhibited a consistent negative change trend, which reversed to a positive change trend at the end of the sequence, and the degree of positive change was extremely high. In terms of the experience, people’s emotions are continuously suppressed as the space shrinks, and then, with the appearance of the last space, they suddenly feel enlightened and uplifted.

• Moderate spatial sequence

A moderate spatial sequence refers to an organization type in which the development of the ordinal space exhibits no obvious highs or lows, and the change trend is not prominent. The direction of change between adjacent units in a moderate spatial sequence may not be completely consistent, but the RV is not significant.

The study case LBY-SE (U6–U12) in Nanjing Expo Park was selected as an example (Figure 8). The spatial units that constitute the LBY-SE (U6–U12) are all hard paved squares. The sequence consists of five spatial units, and the RVs of the adjacent units were 0.09→1.04→0.42→0.63. The $N_c$ in the LBY-SE (U6–U12) was 0, the ${AVE}_{RV}$ was 0.54, and the ${SD}_{RV}$ was 0.64. Although the RVs in the four change groups were all positive, the RVs were all low; that is, the overall development of the sequence did not significantly increase. Therefore, a moderate organization type can often bring the visitors a continuous and soothing experience.

3.3. Quantitative Characterization of Typical Types

Sequence cases that were most consistent with the typical types’ organization characteristics were selected as representative cases of each type (

Table 2. Cases of three typical landscape spatial sequence types.

Spatial Sequence Types	Cases a
Fluctuating spatial sequence	HX-SE (U6–U10), QLY-SE (U1–U8), RP-SE (U1–U4), BP-SE (U21–U15), SP-SE (U1–U3), VP-SE (U32–U12), VP-SE (U32–U14) b, BBP-SE (U1–U10), CG-SE (U12–U1), MP-SE (U1–U2)
Reversal spatial sequence	LBY-SE (U41–U38), RP-SE (U10–U4), RP-SE (U16–U20) a, RP-SE (U32–U20) b, BP-SE (U1–U16), BP-SE (U12–U23), BP-SE (U15–U16), OP-SE (U13–U2), GSFP-SE (U13–U7), WPP-SE (U8–U18)
Moderate spatial sequence	LBY-SE (U15–U19), LBY-SE (U6–U12), BP-SE (U1–U2), MP-SE (U19–U14), MP-SE (U30–U28), MP-SE (U29–U23)

^a If the first and last units of the sequence are the same, they are distinguished by tail labels such as a, b, etc.

). The results of the K–W nonparametric test showed that the values of the three factors, i.e., ${AVE}_{RV}$, ${SD}_{RV}$, and $N_c$, were significantly different among the three types (

Table 3. Test Statistics ^a,b.

	${\mathit{N}}_{\mathit{c}}$	${\mathit{A}\mathit{V}\mathit{E}}_{\mathit{R}\mathit{V}}$	${\mathit{S}\mathit{D}}_{\mathit{R}\mathit{V}}$
Kruskal–Wallis H(K)	13.03	14.57	14.44
Asymp. Sig.	0.001	0.001	0.001

^a Kruskal–Wallis Test. ^b Grouping Variable typical types of spatial sequences.

).

Multiple comparison analysis (homogeneous subset) showed (Figure 9) that the value of a single indicator could only explain part of the difference. The homogeneous subset created based on the ${AVE}_{RV}$ and ${SD}_{RV}$ consisted of two subsets. The first subset consisted of the moderate type, and the second subset consisted of the fluctuating and reversal types. The homogeneous subset created based on the $N_c$ consisted of two subsets, where the first subset consisted of the reversal and moderate types, and the second subset consisted of the fluctuating type. The ${AVE}_{RV}$ and ${SD}_{RV}$ values can explain the differences between the moderate type and the other two types, and the $N_c$ value can explain the differences between the fluctuating type and the other two types. Therefore, the three types of sequences need to be characterized by the values of the three indicators together.

The percentiles of 10% and 90% of the above three factors in the three organization types were obtained respectively. When the indicators could not explain the difference between the two sequence types, the union set of the percentile intervals of the factor values was used as the threshold interval of the factors in the set. Therefore, by combining the division results and percentiles of the homogeneous subsets, we were able to obtain a quantitative characterization of each type (

Table 4. Quantitative characterization of different spatial sequence types.

	${\mathit{A}\mathit{V}\mathit{E}}_{\mathit{R}\mathit{V}}$	${\mathit{S}\mathit{D}}_{\mathit{R}\mathit{V}}$	${\mathit{N}}_{\mathit{c}}$
Fluctuating spatial sequence	[1.54, 3.68]	[1.59, 4.03]	[2, 3]
Reversal spatial sequence			[0, 2]
Moderate spatial sequence	[0.54, 1.21]	[0.35, 1.37]

).

• Quantitative characterization of fluctuating spatial sequences

The ${AVE}_{RV}$ of the fluctuating type was between the range of [1.54, 3.68], the ${SD}_{RV}$ was between the range of [1.59, 4.03], and the $N_c$ was mostly between 2 and 3. The spatial changes in the fluctuating sequences exhibited large differences, and the ${AVE}_{RV}$ between adjacent units was significant. The ${AVE}_{RV}$ and ${SD}_{RV}$ of the fluctuating and reversal types were not significantly different; both of these values were significantly greater than those of the moderate type. However, the $N_c$ of the reversal type was significantly less than that of the fluctuating type, and the change direction of the fluctuating type was constantly reversed.

• Quantitative characterization of reversal sequences

The ${AVE}_{RV}$ of the reversal type was between the range of [1.54, 3.68], the ${SD}_{RV}$ was between the range of [1.59, 4.03], and the $N_c$ was less than 2. Although the $N_c$ values between the reversal and moderate types were not significantly different, the box plot of the $N_c$ of the reversal type showed that the $N_c$ values were mostly 1, and an $N_c \mathrm{o}\mathrm{f}$ 2 was less common. These findings indicate that a single change in the direction is more conducive to creating contrast, leading to a reversal spatial sequence with sudden changes in heterogeneity.

• Quantitative characterization of moderate sequences

The ${AVE}_{RV}$ of the moderate type was between the range of [0.54, 1.21], the ${SD}_{RV}$ was between the range of [0.35, 1.37], and the $N_c$ was usually no more than 2. The thresholds of ${AVE}_{RV}$ and ${SD}_{RV}$ for the moderate sequences were significantly lower than those of the other two types. In this type, the RVs of adjacent units were generally less than 1.21, the ${SD}_{RV}$ was generally less than 1.37, and the intergroup intensity of change and the degree of difference between adjacent units were not significant.

In summary, the organization of spatial sequence is the result of the joint influence of the ${AVE}_{RV}$, ${SD}_{RV},$ and $N_c$. No single factor could completely explain and differentiate the differences among the three organization types. The difference between the fluctuating and reversal types is reflected in the threshold of the $N_c$; the difference between the reversal and moderate types is reflected in the thresholds of the ${AVE}_{RV}$ $\mathrm{a}\mathrm{n}\mathrm{d} {\mathrm{S}\mathrm{D}}_{\mathrm{R}\mathrm{V}}$. The threshold intervals of the three factors were different for the fluctuating and moderate spatial sequences.

4. Conclusions

4.1. Comparison with Previous Studies

This study follows the existing theoretical research foundation, takes dynamic change characteristics as the core of sequence space research, and uses typology as a common method for interpreting organizational characteristics. Although active attempts have been made to conduct quantitative research on existing spatial sequences, due to limitations in data acquisition accuracy and efficiency, the number of study cases is limited and a method path that comprehensively quantifies the dynamic changes cannot be formed. The spatial 3D point cloud model provides comprehensive, accurate, and reliable data for quantitative research on landscape space and also provides the possibility for refined processing and analysis of spatial information. Therefore, compared with previous research, this article improves the accuracy of data acquisition, the comprehensiveness of sequence organization characterization factors, and integrates research methods of qualitative classification and quantitative description to achieve a refined analysis of typical sequence organization types.

• Improved spatial data acquisition efficiency and accuracy

Traditionally, spatial morphological characteristics are mostly collected through field surveys and measurements. The obtained information is mostly length, width, and height scale information, and the survey tools used are mostly rangefinder and total station instruments. The abovementioned instruments and equipment are suitable for small-scale spatial data acquisition. However, the equipment needs to be manually moved to obtain comprehensive information. The labor and time costs needed by traditional methods are high, and relatively few types of data are acquired. Three-dimensional laser scanning or oblique photography equipment mounted on a UAV can quickly and efficiently acquire large-scale spatial data. A UAV flight task can be completed by one person, and the spatial data of multiple cases can be collected in one day. In addition, compared to traditional survey methods, which use single-size information, 3D laser point cloud models can record more comprehensive 3D spatial information, such as scale, volume, capacity, and form. Additionally, affected by personal experience and uncertainty, the accuracy of spatial information that depends on manual measurements is also difficult to guarantee. The 3D point cloud model uses data processing and analysis platforms such as ArcGIS and CloudCompare, and the analysis results are objective and accurate and are not influenced by personal experience.

• Improve the comprehensiveness of the quantitative index factor system

In previous theoretical studies, the analysis of the change characteristics of sequential spaces mostly relies on the use of the semantic description method, that is, the use of relative concepts such as strength and size, to describe the space organizational characteristics. Existing studies fail to quantify the complex change characteristics between several spaces and could not further establish an indicator system for the overall organizational characteristics of the sequences. In this study, the RV was first introduced to describe the degree of complex changes in adjacent units, and the factors used to quantify the relative variance between units are selected by the coefficient of variation of all case data. Then, the ${AVE}_{RV}$, ${SD}_{RV},$ and $N_c$ were introduced to quantitatively describe the overall spatial sequence characteristics under the joint influence of several changes. The expression methods of the above indicators were also consistent with the expressions commonly used in theoretical research, which is convenient for scholars and designers to understand. The advantage is that, first, the refined measurement of the spatial organizational characteristics of a specific case can be obtained; second, it is convenient for the refined comparison between different cases.

• Improve the reliability of classification results combined with quantitative analysis methods

This study attempted to conduct qualitative type induction and quantitative characterization analysis on a large number of cases. Methods such as nonparametric testing, multiple comparisons, and threshold analysis are used to achieve quantitative interpretation of typical sequence types. The nonparametric tests and multiple comparison analysis can verify whether variable factors can explain different types of differences, and threshold interval analysis can obtain the value range of factors for similar cases. The results of type classification are based on the inductive extraction of dynamically changing features of 72 sequence cases, which facilitates cognitive understanding and imitation learning. Typical types of quantitative representation analysis further explain the rationality and reliability of the classification results through data statistical analysis methods.

4.2. Implications

• Theoretical implications

This research supplements the existing spatial sequence knowledge system and achieves the connection between theoretical research and quantitative research. Existing theoretical research regards the dynamic change as the core of sequence spatial organization, but existing quantitative research fails to achieve a quantitative description of dynamic changes. Theoretical research mostly takes qualitative classification as research methods but fails to verify the classification results through data statistics. Based on obtaining high-precision 3D spatial information, this article proposes a quantitative index factor system for spatial organization characteristics, optimizes the quantitative description method of typical organization types, and improves the deficiencies of existing research.

This method can be applied in many scenarios in the future. For example, taking the objective sequence spatial organization characteristics as independent variables and human behavioral activities and psychological perception characteristics as dependent variables, we can further explore the interaction mechanism between people’s subjective experience and the objective spatial organization.

• Practical implications

The classification and induction results of the sequences provide the spatial types that can be referred to and imitated in practical design. The threshold interval is a quantitative characterization of a large number of cases with common characteristics, and it is also a sufficient and necessary condition for the creation of a specific spatial type. The quantitative research results on the organizational characteristics and typical types of landscape spatial sequences can more scientifically and rationally reveal the internal order constituted by sequences, which is highly important for space and teaching and compensates for the deficiencies in existing studies. Additionally, the research results can also provide a database and parameters for intelligent design generation. At present, intelligent design generation lacks the support of spatial data derived from actual cases. The typical types and their threshold intervals obtained in this article can be used as parameters for intelligent design to guide, regulate, and inspect design generation.

From the perspective of technical application, the application of 3D point cloud data has gradually shifted from the stage of site survey and information acquisition to the stage of space processing, operation, and analysis based on point cloud information, which can satisfy the diverse quantitative analysis needs faced in landscape research and practice.

4.3. Limitations and Prospect

With the advancement of technical methods, the traditional topic of landscape space also urgently needs to introduce new methods and perspectives. This article is an exploration of traditional spatial sequence research methods using digital information technology. But it is undeniable that, as an attempt to update the method, the article still has many limitations. First, the classification perspective needs to be diversified. This article only attempts one classification perspective, that is, based on the dynamic changing characteristics of the sequence. In the future, we plan to apply diversified classification perspectives to explore quantitative research conclusions on spatial sequence composition from different angles. Second, the case database needs to be expanded. Although, compared with the existing quantitative research with a limited number of cases, the article has greatly expanded the number of cases. However, due to the large workload of data processing and analysis, the conclusions of this study are based on the analysis results of 72 cases in 14 parks. In the future, we can use computational programming to realize automated analysis and extraction of 3D spatial information to incorporate more case data. Third, there are limitations of the application scenarios. As mentioned in the Section 2, the case studied in this article is an artificially designed landscape located in a built environment, rather than a natural space, and has no special topography. Under the influence of the external environment, spatial sequence definition methods and classification perspectives may be different. Therefore, the application of the research conclusions is limited to the same scenarios as this article.

With the iterative updating of technology, methods such as machine learning, intelligent construction, and artificial intelligence have also emerged. In the future, more intelligent spatial analysis and generation methods need to be explored from three directions: intelligent data processing of three-dimensional point cloud data, machine learning technology-assisted spatial sequence type classification, and the use of artificial intelligence to assist in the generation of spatial design.