Exploring the Spatial Heterogeneity and Driving Factors of UAV Logistics Network: Case Study of Hangzhou, China

Abstract

Urban logistics is an important research topic in human and economic geography; unmanned aerial vehicles (UAVs) are an emerging technology that has huge potential in the field of logistics with the release of control restrictions on low-altitude airspace. The scientific identification of the spatial pattern and impact factors of UAV logistics networks is greatly significant in regards to UAV logistics planning and scheduling. This study considered the urban logistics network of Hangzhou in 2020 as the research topic and used kernel density estimation, a geodetector, and geographic information system (GIS) spatial analysis technology to systematically analyze the spatial patterns and influencing factors at the city and district scales. The study found that a significant spatial pattern was revealed in the UAV logistics network in Hangzhou, and the logistics nodes showed an obvious “core-edge” structure. The urban population, market scale and logistics infrastructure jointly shaped the structure and function of the UAV logistics network, and logistics nodes had a strong coupling relationship with the urban spatial structure. Through interaction detectors, the technical route of urban UAV logistics network construction was analyzed and summarized, and results can provide a scientific basis and case reference for other cities to build and plan UAV logistics networks.

1. Introduction

With the advent of the Fourth Industrial Revolution and the continuous improvement of communication network infrastructure represented by 5G, Internet of Things, industrial internet, and satellite internet, driverless and unmanned aerial vehicles (UAVs) have been integrated into people’s lives [1,2]. UAVs are an emerging technology and are expected to become the core of future intelligent logistics, owing to their advantages over traditional vehicles such as fast operation, high efficiency [3,4], environmental protection [5,6] and the ability of almost being unaffected by ground traffic conditions. The urban logistics network is the nucleus of modern logistics geography research [7]; it is the basis of the urban logistics system and is vital for organizing logistics activities and providing convenient services.

However, cities and urban spaces are characterized as centres with significant human population densities, various types of consumer products, high-value infrastructure and with a variety of associated socio-economic activities [8]. Tremendous market demand and increasingly mature remote sensing information technology have given rise to the development of urban UAV low-altitude flight path planning and construction, and relevant research teams have conducted in-depth studies on the establishment and planning of UAV low-altitude flight path networks in urban areas [9,10,11]. The above exploratory studies of UAV operations in urban environments have all demonstrated the key elements of urban social, economic and various geographic elements that influence UAV flight path planning. Therefore, studying the spatial pattern of urban UAV logistics networks and their driving mechanisms is crucial for improving the transportation efficiency and management of modern urban logistics.

Current research on the spatial pattern of urban logistics networks is mainly focused on ground transportation logistics [12,13], and changes in urban geospatial structure and urban logistics networks are generally collaborative [14,15]. Economic, social, land use, and transport data are still considered important aspects of evaluating geographical patterns of public facilities [16,17,18]. Therefore, most previous research on logistics network space discusses and analyzes the driving mechanism of logistics spatial patterns through qualitative analysis. However, urban logistics transport faces many challenges (e.g., congestion, land use conflicts, community acceptance, environmental pollution) that may arise in metropolitan areas [18]. In addition, due to the influence of urban regional economic differences and traffic accessibility, the logistics network space has a certain spatial heterogeneity. Therefore, qualitative analysis cannot fully analyze the driving mechanism of logistics networks in different regions and at different scales. Because of this, scholars have recently used a combination of qualitative and quantitative analysis to detect impact factors, which improves the objectivity and accuracy of the research results and helps further inspect the spatial and temporal patterns of logistics space.

Numerous studies have examined the driving mechanisms of urban logistics networks using various quantitative methods, including factor analysis [19], multiple linear regression [20], average nearest neighbor analysis [21], weighted model [22], analytic hierarchy process [23], hot spot area analysis [24], and geographically weighted regression [25,26]. Based on the above studies, we found that most of the academic research has not sufficiently considered the interaction effect of the impact factors of the spatial gathering and the study scale is mostly limited to provinces and cities, leading a lack of micro-level perspectives on the logistics networks. The Geodetector() is a new statistical model based on spatial analysis and geostatistical theory to quantify the factors of impact on the spatial stratified heterogeneity of geographic phenomena [27,28]. This model does not require too many assumptions compared to traditional regression models and is immune to the covariance of multiple independent variables. Therefore, it has obvious advantages in the analysis of complex factor interactions and the study of spatial and temporal scale effects of various geographic phenomena [29,30].

However, UAV logistics is in the initial test stage and its logistics network space needs to be planned and arranged according to the existing urban space and economic development level [31]. Unlike ground vehicles that can refuel during transportation, UAVs are limited by fuel and power, and their flight distance is relatively short; therefore, they essentially have a terminal logistics distribution mode as UAV loading goods are transported from the seller to the consumer through the logistics network [32]. Unlike ground logistics, aerial logistics do not need to consider traffic congestion and distribution route planning and scheduling and are generally a straight-line delivery from business to consumer. Because of this, the research scale of urban UAV logistics networks is mostly at city and district levels. With a decrease in spatial scale, the spatial pattern and functions of urban space become more complex [33] and interactions between impact factors will also differ from ground transportation. Therefore, it is improbable to accurately detect the driving mechanism of UAV logistics spatial pattern at different scales based only on the research experience of ground logistics.

Thus, in the present study, we took the Autonomous Delivery Network (ADNET) logistics network of the Hangzhou as the research object and used the kernel density estimation (KDE) method to depict the take-off and landing points of the logistics network and the aggregation of navigation routes in the study area. Subsequently, using spatial analysis and Geodetector, dominant factors of spatial pattern of city-scale UAV logistics distribution network and impact factors interactions were analyzed. The main contribution of this study is to reveal the differentiated characteristics of UAV logistics network and its driving mechanism at city and regional scales. The results in this study provide a basis for the spatial layout, structural system optimization and related infrastructure spatial planning of UAV logistics nodes and urban UAV logistics service facilities in order to realize the advantages of UAV in urban smart logistics. This study also provided a case reference for other cities to build a standardized urban UAV logistics network and optimize the logistics resource allocation and strategic planning. The remainder of this paper is organized as follows. Section 2 introduces the study area, the research data preprocessing, and the methodology. Section 3 presents the results, which are subsequently discussed in Section 4. The paper concludes with a summary of the results and their implications for future research.

The characteristics of UAV logistics network differentiation and its driving mechanism were revealed at the city and district scales to provide a basis for the spatial layout, structural system optimization, and related infrastructure support of UAV logistics nodes and urban UAV logistics service facilities to realize advantages of UAVs in urban intelligent logistics. This study also provided a case reference for other cities to build a standardized urban UAV logistics network and optimize the logistics resource allocation and strategic planning.

2. Materials and Methods

2.1. Study Area

Hangzhou is a metropolis of East China in the southern Yangtze River Delta and is an important e-commerce center in China. Hangzhou, including its six core districts, has experienced rapid economic growth and urbanization, with its urbanization rate changed from 29.40% in 1990 to 79.5% in 2020, which has driven a rapid increase in the quality and level of supply of goods and services. In recent years, Hangzhou has focused on the digital economy, using 5G technology to provide development impetus for manufacturing, traditional trade, and urban governance with the aim of building a digital city with first-class digital economic concepts and technologies. These developments provide a strong base for the construction of city-level UAV smart logistics.

This study considered the coverage area of the ADNET smart logistics network of Hangzhou as the research area, which is centered on the central city of Hangzhou and spreads approximately 30 km outward to form a city-level drone distribution logistics circle around the boundary of the middle ring road section of Hangzhou. The logistics network consists of 144 unmanned vehicle terminal stations (a comprehensive terminal station integrating the functions of UAV landing pads, unmanned vehicle garages, cargo transit and staging, and human-machine interaction) and 265 routes, with a total cumulative route length of 927 km and a logistics circle covering the Shangcheng, Xiacheng, Gongshu, Binjiang, and Jianggan districts of Hangzhou, as well as parts of Xiaoshan, Yuhang, and Xihu districts. In total. The network has a network coverage area of 325 km² with a serviced population of more than 5 million, as shown in Figure 1.

2.2. Materials

2.2.1. Logistics Network Data

The data of the ADNET urban logistics distribution network in the study area were provided by Hangzhou Xunyi Company. The network was verified through continuous field tests and gradually shaped into a stable UAV logistics network in urban scenarios under social and economic effects. The data included geographic location data of unmanned hub stations and waypoint GPS information of air routes (including takeoff and landing points, as well as waypoint data of flight operations).

To reveal the spatial distribution characteristics and spatial stratified pattern [27] mechanism of unmanned vehicle terminal stations and routes in the UAV logistics network, and to explore impact factors of the geographic pattern of the urban UAV logistics network, the following processing steps were carried out on the logistics network data: for the route data, because the research perspective of this study was two-dimensional geographic space, z-values in the GPS information of waypoints were unified and transformed into two-dimensional plane data. Then, an urban UAV logistics network map was constructed. Finally, the centrality of logistics nodes calculated by network analysis was used as the weight for KDE, which was in turn used as the explanatory variable for the spatial layout of the UAV logistics network.

2.2.2. Driving Factor Data

Based on previous studies and experiences of ground logistics networks and UAV delivery [13,14,32,34], this study selected 10 representative indicators to analyze the specificity of the driving mechanism for detecting the spatial pattern of UAV logistics networks (

Table 1. An index system for detection results of spatial pattern of unmanned aerial vehicles (UAV) logistics network in Hangzhou.

Factor	Description
X1: population density	Population density values (people/km2)
X2: residential land density	Residential land use POI kernel density values (POI/km2)
X3: building density	Percentage of building area within the grid unit (%)
X4: road density	Total length of roads in grid unit (km/km2)
X5: 5G base station density	5G base station kernel density values (POI/km2)
X6: night light intensity	$\mathrm{Annual} \mathrm{night} \mathrm{light} \mathrm{values} (\left(\mathrm{W}\right)/\left({\mathrm{m}}^2\mathsf{\mu }\mathrm{m} \mathrm{s}\mathrm{r}\right)$)
X7: business service capability	Business service POI kernel density values (POI/km2)
X8: consumption vitality	Index calculated based on the weighted density of the five POI core categories: dining, shopping, living, sports and leisure, and accommodation services, and finally normalized
X9: housing price	Average value of house price POI data by township-level administrative block (RMB)
X10: medical facility density	Medical facility POI (pharmacies, clinics, hospitals at all levels) kernel density values (POI/km2)
Y: kernel density of UAV logistics network	Indicators of explanatory variables calculated using the weighted KDE method based on UAV logistics network data

). These indicators were chosen to accurately portray impact factors that influence the spatial distribution characteristics and spatial stratified heterogeneity of logistics network using geographic and economic factors, while also considering the spatiotemporal granularity and accessibility of data.

Administrative division data from 2020 were obtained from the National Catalogue Service for Geographic Information, and the township-level division vector data were obtained from the Geographical Information Monitoring Platform Cloud Platform (http://www.dsac.cn/ (accessed on 20 June 2020). Because of the fine granularity of this study, population density data were selected from high-resolution population data products generated by the WorldPop project based on nighttime lights, land use types, and elevations through the random forest model and the built-settlement growth model [35], which has a spatial resolution of 100 m.

The UAV logistics in this study system works for 24 h, so annual nighttime light data were selected to characterize the economic activities of the city [36]; the nighttime light data used was from the 2012–2020 global annual nighttime light dataset produced by Earth Observation Group based on VIIRS monthly data [37], which has a spatial resolution of the Hangzhou area of approximately 500 m.

Urban road and building vector data were obtained from OpenStreetMap (https://www.openstreetmap.org (accessed on 20 June 2020)), which provides free vector map data downloads. Point of interest (POI) data related to business services, medical facilities, housing prices, 5G base stations, and consumption vitality related to this study were obtained from the application programming interface (API) provided by Baidu Map, and the final ensemble obtained POI data in 17 categories with 321,443 pieces of valid POI data for the study area. The consumption vitality index, a comprehensive index that integrates various types of consumption [38], was calculated by weighting the nuclear density of five types of POI data: food service, shopping service, living service, sports and leisure, and accommodation service.

2.3. Methods

2.3.1. Social Network Analysis

Logistics refers to the process of the physical flow of goods from the place of supply to the place of demand, and urban logistics network planning and layout is the process of selecting a number of addresses as commodity logistics nodes within the existing urban spatial area [39,40]. Social network analysis is a common method for urban network structuring that uses network centrality to portray the overall shape, structure, and characteristics of urban UAV logistics spatial networks. The degree centrality of node $i$, $C_D\left(i\right)$ is defined as the ratio of the number of neighboring nodes that are connected to node $i$ to the maximum number of possible connections [41]. Centrality in the network space indicates the number of points directly connected with other points. The larger the ‘centrality’ value of a point in the urban logistics network, the higher the logistics activity and thus the stronger the corresponding centrality of its logistics hub. The formula for determining the degree centrality is as follows:

$$C_D\left(i\right)=g_i/(N-1)$$

(1)

where $C_D\left(i\right)$ represents the degree centrality of the hub station $i$ of the city logistics network; $g_i$ represents the number of connecting edges for a logistics hub in cities; $N$ is the sum number of the network node and $N-1$ represents the maximum number of possible connections.

2.3.2. Kernel Density Estimation

KDE is based on the first law of geography and is a theoretical basis for exploring the spatial distribution pattern of point sets [42]; its estimated value is a field simulation of the spatial phenomenon where each location determines the local aggregation intensity according to its spatial relationship with neighboring sample points. Most traditional point pattern analysis is based on isotropic Euclidean space; however, for the network space of facility point distribution, the actual point density is influenced by the spatial accessibility of urban space and the service capacity of the facility points [43]. Therefore, to more accurately express the distribution density of UAV logistics terminal stations, this paper used the network centrality of each terminal station as the weight for KDE and set 500 m as the bandwidth parameter for KDE based on the design service range of the terminal stations and considering the scope of the study area. The calculation formula is as follows:

$$f(x,y)=\frac{1}{n\ast h}{\displaystyle \sum _{i=1}^n{C}_{D}K\left(\frac{dis{t}_i}{h}\right)}$$

(2)

where $f(x,y)$ is the estimated kernel density at $(x,y)$; $h$ is the bandwidth (distance decay threshold); $n$ is the number of element points with distance less than or equal to $h$ from location $(x,y)$; $dis{t}_i$ is the distance between the sample points; $K$ is the function of spatial weights; and $C_D$ is the spatial weight function calculated by Eq.(1). The weight function in this study was network centrality; the stronger the centrality, the greater the weight, indicating that the denser the UAV logistics network near the region, the stronger the logistics distribution force.

2.3.3. Geodetector

In the study, the estimated kernel density were used as the logistics network capacity explanatory variable y. y is collected in a system of sampling points consisting of sampling cells $i$($i=1,2,\dots ,n$, where $n$ is the total number of sampling cells) in the study area. For some factors that may affect the spatial variance of the explanatory variable $A=\left\{A_h\right\}$ ($h=1,2,\dots ,L$ is the number of classifications based on expert knowledge or natural interruption points, representing the different categories of the factor, representing one or more sub-regions in the geographic map image, i.e., the stratum Strata in statistics). To explore the extent to which the factor $A$ explains the spatially stratified pattern of logistics aggregation $y$, the explanatory variable layer is overlaid with the factor A layer, and t the metric is measured using the $q$ value, which is calculated as

$$q=1-\frac{1}{n{\sigma }^2}{\displaystyle \sum _{h=1}^L{n}_h{\sigma }_h^2}$$

(3)

where $h=1,2,\dots ,L$ is the strata of factor $A$, that is, the classification or partition; $n_h$ and $n$ are the sample size in strata $h$ and the whole area, respectively, where $n={\displaystyle \sum _{h=1}^L{n}_h}$; ${\sigma }_h^2$ and ${\sigma }^2$ are the variance of logistics network capacity in strata $h$ and the whole region, respectively. $q$ statistic range from 0 to 1, with larger values indicating a stronger explanatory power of the factor $A$ for the spatial variance of the explanatory variables.

Geodetector also provides interaction detector through spatial interaction of spatial stratification of the dependent variable based on q-value calculation methods in Eq. (3). The function of interaction detector is to determine whether the interaction between different driving factors affect the explanatory variable, or whether the effect of each driving factor on explanatory variable is independent [28]. The method is to calculate the value of two different driving factors on explanatory variables: $X_1$ and $X_2$, then calculate the interaction result of the value of two different driving factors: $(X_1\cap X_2)$, and finally compare the calculated results. The interaction type between the two factors can be divided into the following categories (

Table 2. Types of interaction between two independent variables and dependent variables.

Description	Interaction Type
$q(X_1\cap X_2)<Min(q(X_1),q(X_2))$	Non-linear reduction
$Min(q(X_1),q(X_2))<q(X_1\cap X_2)<Max(q(X_1),q(X_2))$	Single-factor nonlinearity reduction
$q(X_1\cap X_2)>Max(q(X_1),q(X_2))$	Two-factor enhancement
$q(X_1\cap X_2)=q(X_1)+q(X_2)$	Independent
$q(X_1\cap X_2)>q(X_1)+q(X_2)$	Nonlinear enhancement

.):

3. Results

3.1. Spatial Distribution Pattern of UAV Logistics Network

The results of the weighted kernel density analysis of UAV logistics data in Hangzhou (Figure 2) showed that the spatial distribution of the logistics network in Hangzhou had obvious aggregation characteristics. To further explore the spatial layout of the logistics network, KDE with the network centrality of UAV terminal stations was used as the weight to identify core areas where the spatial distribution of UAV logistics was concentrated.

In the urban area of Hangzhou, the spatial distribution of the UAV logistics network was centered in the central city, however there was pattern within the central city, concentrated in residential land, business gathering places, IT parks, and other areas. According to the nuclear density estimation results, the UAV logistics network was divided into one primary core area (business and population concentration zones of Xihu District, Gongshu District, and Xiaocheng District), three secondary core areas (business and population concentration zone of Jianggan District and the IT park and business district concentration zone of Binjiang District), and five tertiary core areas (relatively scattered) (Figure 2).

The above analysis results showed significant spatially divergent characteristics in the spatial layout of UAV logistics in Hangzhou, and most logistics networks were concentrated in business districts, medical care areas, and densely populated areas. Because explanatory variables and their respective variables were all area data and had different spatial granularity, this study finally chose 20,456 sampling points with a 200 m interval for multi-factor spatial correlation as the input data of the Geodetector model. The model detected dominant factors affecting the spatial pattern of the UAV logistics network and their interaction mechanisms from the city and district scales, respectively, and analyzed corresponding locational characteristics of the gathering areas of the UAV logistics network.

3.2. The Detection of Factors Influencing UAV Logistics Network

3.2.1. City Scale

The factor detector can quantitatively detect the driving force of impact factors on explanatory variables and identify dominant factors affecting the spatial pattern of the explanatory variables by comparing the magnitude of the factor driving force. In this study, spatial indicators of urban UAV logistics calculated by KDE were taken as explanatory variables (Y) and the 10 impact factors were taken as independent variables (Table 1).

The Geodetector model used was the Geodetector package in R language; the relevant variables were input into the computational model and the factor detection results are shown in Figure 3. Among them, business services (X7), residential land density (X2), medical facility density (X10), consumption vitality (X8), and building density (X4) were the leading impact factors leading to the spatial pattern of the UAV logistics network in Hangzhou, and their corresponding values all exceeded 0.2 (0.5173, 0.3374, 0.2685, 0.2636 and 0.2090, respectively) and all of their p-values < 0.05. The q value of the business service driving factor was greater than 0.5, indicating that its explanatory power for the spatial divergence of the logistics network was more than 50%. From a geospatial perspective, as can be observed from Figure 4, there is a certain coupling between the spatial density of the logistics network and the spatial layout of the five leading impact factors. For example, the business service capability that drives the strongest forces has a strong spatial similarity, and this feature is detected by the q-value of the Geodetector.

3.2.2. District Scale

To investigate the influence of scale on the spatial pattern of the UAV logistics network, the data of sampling points were partitioned into district-level units and input into the factor detector model. Differences were observed in the differentiation mechanism of UAV logistics network density within each district-level administrative unit (Figure 5 heat map, all statistics passed the significance test). As the number of UAV terminal stations within the non-central city blocks of Yuhang District and Xiaoshan District was small, the detection value of the partition factor was small. Although business service capacity was the dominant influencing factor in each district-level unit, dominant factors within different units demonstrated different degrees of influence. The influence of business services in Xiacheng District (78.6%) was the largest, followed by Shangcheng District (73.8%), Gongshu District (64.2%), and Xihu District (60.5%). The sub-dominant factors of the above four district-level units were residential land density, so the spatial pattern of UAV logistics in the above four units was mainly influenced by business services and residential land use. Although the explanatory power of the dominant influencing factors was not high in Binjiang District, there were more types of dominant factors, including business services, building density, night lighting intensity, medical facility density, and consumption vitality. In addition, the influence of residential land and housing prices was not significant, which was consistent with the more developed IT industry in Binjiang District. In Shangcheng District, the impact of the level of medical facilities was significantly higher than that in other district-level units, indicating that Shangcheng District is rich in medical resources and is the main output unit of UAV medical logistics in Hangzhou.

3.3. The Interaction Influence between Factors

The above analysis analyzed the degree of influence of single factors on the spatial pattern of UAV logistics networks, however in practice, complex interactions between multiple factors jointly influence the spatial pattern of explanatory variables. Therefore, this study used the interaction detector to demonstrate that at the city scale. Figure 6 shows the results of interaction detector, different colors represent different interaction strengths, where the solid square symbol indicates the interaction type is two-factor enhancement and the hollow triangle indicates the interaction type is Nonlinear enhancement. As a further exploration, factor interactions had a higher influence on the spatial variation of UAV logistics networks compared with individual factors, where the interaction type was mainly two-factor enhancement and mostly related to economic and demographic factors. Using the interaction detector tool, the value of the interaction between factors increased to different degrees than that of the single factors, among which the explanatory power of the interaction between business service capability and night lighting intensity was the highest (0.5567), followed by business service capability and population density (0.5555). As shown, interactions between business service capability and other factors had high explanatory power, which indicated that improvements in business service capacity in Hangzhou urban area significantly increased the influence of each factor and thus the explanation of the spatial variation of the UAV logistics network by each factor as independent variables.

4. Discussion

4.1. Implication for UAV Logistics Planning

Analyzing the weighted kernel density analysis of the UAV logistics network revealed a highly uneven spatial pattern of the UAV logistics network in Hangzhou; the logistics terminal hubs showed an obvious axial radiation structure in space, the core of which was mainly concentrated in the central city that had strong overall economic strength while the rest was scattered in urban spaces dominated by residential areas, IT parks, and medical facilities. Differences in the density of logistics networks corresponded to geographic distributions of partitions of different levels of the impact factors, which illustrated the strong spatial stratified heterogeneity of urban UAV logistics networks.

Through further analysis of the Geodetector model, the level of business services, the density of residential land, medical facilities, consumption dynamics, and building density were determined to be main influencing factors of the spatial pattern of the UAV logistics network in Hangzhou. The two impact factors with the strongest explanatory power of spatial pattern (the level of business services, the density of residential land) corresponded to the business and consumer sides of the business-to-consumer (B2C) logistics model. The results of the interaction detector again confirm these two most leading factors. We can clearly see from Figure 6 that two impact factors play an extremely clear positive role in the interaction with other factors while the others do not have such a strong influence. These dominant impact factors are roughly the same as those identified in ground logistics research [44,45], but have some variations. Among factors influencing the spatial pattern of UAV logistics networks, the driving force of medical facilities in UAV logistics is higher than that of ground logistics; as medical distribution has higher requirements for speed, UAV logistics has suitable characteristics [4]. Accordingly, pilot studies of the commercial applications of UAV logistics in many countries have involved medical distribution [46]. In addition, the two logistics network spatial pattern mechanisms were varied in terms of traffic accessibility. Traffic accessibility refers to the opportunity for interaction between nodes in a traffic network [44] and a high traffic accessibility can be conducive to expanding user groups and logistics markets, facilitating transportation of logistics activities, improving logistics organization efficiency, and reducing logistics transportation costs [6]. The strength of ground road connectivity (measured in road mileage per unit area) is a key factor affecting the spatial pattern of ground logistics networks [14,44].

However, for air logistics transportation, road density has insufficient explanatory power for the spatial pattern of logistics networks. Although UAVs have more freedom to operate at low altitudes and can achieve point-to-point linear flight, the development of UAV intelligence is not as mature as that of ground-based unmanned transportation and specific flights could be impacted by weather (e.g., high-rise winds generated between buildings) and signal strength (poor local signals due to shading or problems with the deployment of related communication equipment) [47]. Hangzhou city is a digital city with mature communication facilities, and its center is at a low-altitude and has low surface undulation. These ideal geographical conditions and infrastructure construction environment play a key role in the safe operation of UAV logistics. Therefore, although the explanatory power of 5G base station density in Hangzhou in this study was insufficient for spatial pattern in Geodetector results, in terms of building a more general logistics network, the basic communication facilities, mainly 4G/5G base stations, were the key elements affecting spatial heterogeneity.

Most studies on the driving mechanisms of spatial pattern focus on the degree of influence of single impact factors on the spatial pattern of explanatory variables [20,22,26]; however, the driving mechanisms of spatial pattern of most geographic phenomena in cities are complex and analyzing interactions among factors is more meaningful for exploring driving mechanisms. In this study, the interaction of all factors was investigated using Geodetector interaction detector, and the interaction of any two factors selected in this study was shown to be greater than the influence of any single factor. In particular, interactions between business service capacity and residential land density with other factors enhanced the explanatory power of those factors for the spatial pattern of logistics networks. These two factors correlated to business and consumer sides of the B2C operation mode of UAV logistics. Interaction detector results revealed that results of spatial pattern of UAV logistics networks were not caused by a single driving factor, but rather by the combined effect of different impact factors, which were mainly influenced by the business and consumer sides of the logistics operation mode combined with various hardware facilities, population, and economic factors.

4.2. Impact of Regional Differences

Generally, the spatial pattern of the research object is dependent on the spatial scale analyzed [15,26,32,48], therefore examining and studying the same research object at multiple scales can reveal deeper patterns. In this study, the spatial pattern of the urban UAV logistics network differed when observing at the city and district scales, and the factor detection results at the city (Figure 3) and district levels (Figure 5) were combined and analyzed together. The spatial pattern was more complex at the district scale than the city scale, influenced by factors such as industrial structure, economic strength, and regional policies within each district. In addition, the dominant influencing factors differed between districts and differed from the detection results at the city scale.

At present, UAV logistics is mostly limited to the supply of medical, cold chain, and other goods with high requirements for high-speed delivery. In addition, the current market scale of UAV delivery is far smaller than that of ground logistics transportation and transportation costs are higher. Therefore, the degree of supply and demand for goods with tight time requirements and high delivery costs will be detected at the district scale. For Shangcheng District, Xiacheng District, and Xihu District, the q-values of each driving factor were higher than the city scale because the overall economic strength of these regions was stronger, and the manufacturing service industry was more fully developed compared with other regions. These characteristics increase the logistics demand, enabling UAV logistics to gather in these regions.

The above analysis showed that the spatial pattern of urban drone logistics network was scale dependent. Therefore, when carrying out spatial planning and determining the layout of a logistics network, as well as considering interactions and overall construction of various factors at the macroscopic scale, we should also consider the differences of various industries and socio-economic factors in different regions at the microscopic scale and functional roles played by regions in the city. In addition, understanding the intensity factors influencing different urban areas at different scales can improve resource allocation, regulation, and guidance. This improved understanding will improve the utilization rate of logistics facilities and resources, strengthen the linkage development and coordination between various regions of the city, and cultivate and expand the scale of demand for the UAV logistics market.

In summary, the construction of a city-level UAV logistics network was more complicated than that of a ground logistics network. First, there are necessary technical conditions in terms of the hardware and policy environment, including the establishment of extensive communication infrastructure for the service coverage area, the support of government and relevant authorities, the safe and reasonable application of airspace, and the improvement of UAV operation supervision. Second, the logistics network layout and optimization need to combine the local regional economic development level, population density, and business and medical level, as well as medical and cold chain goods for pilot application. Finally, the use of geographic information technology mining and analysis of the urban UAV logistics market demand can complement ground logistics transport; the formation of an air-ground dual logistics chain is vital for the development of transport for the logistics industry.

5. Conclusions

The study of the spatial pattern of city-level UAV logistics networks and the driving of their geographical association is an essential component of UAV logistics planning and resource scheduling. This study aimed to improve our understanding of the driving mechanism of UAV logistics networks and provide suggestions for the layout and planning of urban UAV logistics networks. The main contributions of this study are as follows:

(1) The UAV logistics network of Hangzhou shows spatial pattern at the city scale. From the perspective of the functional structure of the network, under the influence of the UAV logistics mode of B2C, the logistics hubs presented an obvious ‘core-margin’ structure; in this structure, the logistics hubs in the core areas of the city largely accepted the role of the business while the logistics hubs in the non-core areas of the city, mainly IT parks and residential areas, performed the role of consumer.

(2) The spatial pattern of the urban UAV logistics network was the result of the combined effect of several impact factors and the impact level varied by scale. As the scale became finer, the correlation between each influencing factor and the spatial distribution increased. In terms of the formation mechanism, urban population, market scale, and logistics infrastructure jointly shaped the form, structure, and function of the urban UAV logistics network. At the district level, the scale of UAV logistics nodes and locations dominated by medical and business services were strongly coupled.

(3) Economic factors such as business services, consumption vitality, and market size were important impact factors for the size of city-level logistics nodes. In contrast, geographical factors such as population density, operating signal environment, and building density can limit the safe and efficient operation of UAVs. Therefore, the construction of UAV logistics networks in urban centers requires more secure and reliable technical means and a more accurate and perfect operation and supervision system.

By sorting out the relevant contents of this study, the following perspectives are given for future related research. The urban logistics network will be dynamically adjusted according to the changes of urban industry and economic structure, and this feature is especially obvious in the field of UAV logistics. Future research should further analyze the influencing factors on urban UAV logistics networks at spatial and temporal scales and validate the detection results with actual logistics order data in logistics networks. This will improve our understanding of the spatial heterogeneity and the driving mechanism of dynamic adjustment of city-level logistics networks and provide further reference for urban UAV logistics network planning and resource allocation.