Spatial Identification and Interactive Analysis of Urban Production—Living—Ecological Spaces Using Point of Interest Data and a Two-Level Scoring Evaluation Model

Abstract

Identifying urban production–living–ecological spaces and their interactive relationships is conducive to better understanding and optimizing urban space development. This paper took the main urban area of Hangzhou city as an example, and a two-level scoring evaluation model was constructed to accurately identify urban production–living–ecological spaces using point of interest (POI) data. Then, kernel density analysis, a spatial transfer matrix, and a bivariate spatial autocorrelation model were used to reveal the spatial patterns of urban production–living–ecological spaces and their interactive relationships during 2010 and 2019. The results showed that the proposed two-level scoring evaluation model combining both the physical area and density of POIs was effective in accurately identifying urban production–living–ecological spaces using POI data, with an identification accuracy of 88.9%. Urban production space was concentrated on the south bank of the Qiantang River and around the north of Hangzhou. Urban living space had the highest proportion, mainly distributed within the ring highway of Hangzhou in a contiguous distribution pattern, and urban ecological space was concentrated around West Lake and Xiang Lake. During 2010 and 2019, the expansion of urban production–living–ecological spaces had obvious spatial differences. Additionally, the mutual transformation between production and living spaces was more frequent during the study period and was mainly distributed within the ring highway of Hangzhou. There were significant positive spatial correlations between production and living and between living and ecological spaces, while a significant negative spatial correlation occurred between production and ecological spaces. The spatial correlations of urban production–living–ecological spaces revealed obvious spatial heterogeneity. This study proposed a two-level scoring evaluation model to accurately identify the spatial patterns of urban production–living–ecological spaces and their interactive relationships using POI data, which can provide detailed information and scientific references for urban spatial planning and management in rapidly urbanizing cities.

1. Introduction

Urban production–living–ecological spaces refer to functional spaces divided according to land use functions within a city [1,2], and they are closely related to urban economic, social, and cultural development [3,4]. Urban production–living–ecological space identification can significantly contribute to urban research and planning [5,6]. Over the past four decades, China’s urbanization has accelerated the population concentration in urban areas, resulting in substantial changes in the form and function of urban spaces [7]. However, due to limited urban space resources, some development constraints, such as imbalanced urban production–living–ecological space structures, low land use efficiency, and unreasonable urban spatial layouts, have raised a series of challenges for sustainable urban development [8,9]. In 2012, China first proposed optimizing the structure and layout of geographical space based on the development of production–living–ecological spaces and achieving the goals of ensuring that the space for production is used intensively and efficiently, that the living space is livable and proper in size, and that the ecological space is unspoiled and beautiful [10,11]. Therefore, accurately identifying urban production–living–ecological spaces and their interactive relationships can better understand the urban spatial structure and its evolution, which can provide insights for optimizing the spatial pattern of urban geographical space.

The concept of production–living–ecological spaces was proposed based on land use multifunctionality, from which production–living–ecological functions are derived [12,13]. As China emphasized production–living–ecological space optimization to support sustainable socioeconomic development, numerous studies began to focus on the concept and identification [1,14], evaluation methods and spatial pattern [15,16], and spatiotemporal evolution and management [13,17]. In these studies, two main methods were widely used to identify production–living–ecological spaces, including land use type merging and multi-criteria evaluation methods [18]. The former connects land use types and land use functions to identify the production–living–ecological spaces, but it fails to accurately evaluate the levels of production–living–ecological functions within the complex main urban area of the city [19]. Multi-criteria evaluation is used to construct a comprehensive evaluation system of geographical space functions to identify the dominant functional space based on a series of socioeconomic statistical data [20,21]. However, it cannot identify the functional spaces in highly spatially heterogeneous areas due to insufficient data precision [22]. Although emerging remote sensing technology can effectively identify surface physical conditions in urban areas, it is challenging to distinguish socioeconomic features and human activity patterns due to the diversity and complexity of surface activities [23].

In recent years, with the development of information technology and location acquisition technology, a series of geospatial data, such as Point of Interest (POI) data, microblogging data, and mobile phone signaling data, have been employed to describe the spatial and temporal characteristics of human activities [24,25], providing a new approach to reveal the finer spatial structure and human activity pattern within a city [26,27]. In particular, POI data have the advantages of a large data volume, detailed information coverage, high precision, and frequent updates, making it one of the most widely used data in urban research [28]. Current studies have directly identified production–living–ecological spaces by using POI semantic information, such as the density and location, which can reflect the spatial distributions and levels of urban functions respectively [29,30]. However, POIs can only be represented as points in space, they cannot reflect the scale of physical entities closely related to urban functions [23,26]. In previous studies, the physical area of POIs was not considered, which may result in a strong bias identification [31,32]. In addition, urban production–living–ecological spaces exhibit complex interactions under rapid urbanization [33]. However, the current focus of interactive relationships of production–living–ecological spaces is on the land production–living–ecological functional spaces of the entire area [34,35], and few studies have explored the interactive relationships of production–living–ecological spaces within urban areas [29,36]. Therefore, constructing an evaluation model considering both the area and density of POIs will further improve the accuracy of urban production–living–ecological space identification, and exploring the interactive relationships of urban production–living–ecological spaces will contribute to a better understanding of the evolution characteristics of urban space.

As one of the fast-expanding cities in China, Hangzhou has experienced rapid population growth and economic development. As a consequence, rapid urban expansion has resulted in dramatic changes in spatial patterns and the incoordination of urban functions [37]. Currently, Hangzhou is accelerating the construction of the national central city, aiming at optimizing the layout of urban functions and gradually forming a new optimized pattern of production–living–ecological spaces. Therefore, Hangzhou is an ideal case area for identifying urban production–living–ecological spaces and their interactive relationships, which can provide references and bases for other rapidly urbanizing cities. In this context, this study attempted to take the main urban area of Hangzhou as an example and construct a new evaluation model combining the area and density of POI data to accurately identify urban production–living–ecological spaces. A spatial transfer matrix and a bivariate spatial autocorrelation model were then used to reveal their interactive relationships. The rest of this study is organized as follows. Section 2 describes the materials and methods. Section 3 subsequently displays the results in the spatial patterns of urban production–living–ecological spaces and their interactive relationships. Section 4 reports the discussion, and the main conclusion is summarized in Section 5.

2. Materials and Methods

2.1. Study Area

Hangzhou is located in the south wing of the Yangtze River Delta and the west of Hangzhou Bay (Figure 1). Hangzhou is the political, economic, cultural, and financial center of Zhejiang Province, which plays a unique and important role in the integrated development of the Yangtze River Delta urban agglomeration and the Belt and Road. This study took the main urban area of Hangzhou as the study area, including eight districts: Gongshu, Xihu, Shangcheng, Xiacheng, Binjiang, Jianggan, Xiaoshan, and Yuhang. The main urban area is the core and engine of the high-quality development in Hangzhou. The main urban area of Hangzhou has a land area of 3370.79 km² and a permanent population of 7.78 million in 2019. The GDP of the main urban area of Hangzhou was USD 1295.6 billion in 2019, accounting for 84.27% of Hangzhou. The population and GDP in the study area grew at annual average rates of 5.32% and 10.18%, respectively, over the past ten years. Due to its superior location advantages, abundant natural resources, and rapid socioeconomic development, Hangzhou has become a representative city of the rapid urbanization phenomenon in China. The main urban area is the core growth pole of the urban space of Hangzhou and the key area of concern for spatial planning. Therefore, it is ideal to take the main urban area of Hangzhou as the research area to identify the different types of urban space.

2.2. Data Sources and Preprocessing

The POI data were obtained from the Amap API interface in 2010 and 2019 by using Web crawler technology. The amount of POI data in 2010 and 2019 was 99.76 thousand and 478.92 thousand, respectively. The attributes of POI data include longitude, latitude, name, type and management region, which can reflect the location, type and intensity of human activities [38]. First, deweighting and data cleaning methods were used to preprocess the original POI data. Then, the POI data types were classified according to the definition of the production–living–ecological spaces (

Table 1. POIs classification results based on production–living–ecological spaces.

Space Type	Sub-Space	Class I	Class II
Production	Commercial production	Company	Advertisement, internet, commercial trade
		Financial and insurance	Insurance institute, bank, securities company
	Industrial production	Industrial park	Factory, industrial park
		Warehousing logistics	Storage, logistics
Living	Commercial services	Food and beverages	Restaurant, cafeteria, dessert shop
		Shopping	Convenience shop, supermarket
		Life service	Post office, petrol station
		Sports leisure	Entertainment venues, sports buildings
		Accommodation services	Hotel, guesthouse
	Public services	Government agencies	Government agency, social organization
		Traffic facilities	Subway, bus, train
		Medical security	Hospital, clinic
		Science and education culture	Museum, scientific research institution, school
	Residential guaranteeing	Commercial housing	Commercial office building, residential district
Ecological	Green ecological	Urban parks	Botanical garden, park plaza
		Scenic spot	Scenic area

). Additionally, the land use planning data came from Hangzhou Bureau of Planning and Natural Resources, which was generated by high-resolution remote sensing images and numerous manual visual interpretation work.

Specifically, urban production space refers to place carriers that provide human beings with production and management activities, such as material products, transportation, and commerce [29]. It can be classified into two types: commercial production and industrial production, which correspond to four types of POI data including company, financial and insurance, factory and industrial park, warehousing and logistics. Urban living space refers to the places that provide living, consumption, leisure and entertainment for human beings [39], including three categories: residential guaranteeing, commercial services, and public services. These spaces correspond to ten types of POI data, including food and beverages, shopping, life service, sports leisure, accommodation services, government agencies, traffic facilities, medical security, science and education, and commercial housing. Urban ecological space refers to the green space that provides ecological products and ecosystem services for residents, which corresponds to two types of P0I data, including urban parks and scenic spots [39].

The main steps to identify urban production–living–ecological spaces and their interactive relationships can be divided into four parts, as shown in Figure 2.

2.3. Methods

2.3.1. Kernel Density Analysis

Kernel density analysis is an important statistical method for extracting the distribution features of geospatial facilities, which can analyze the correlation and density expansion between elements [40,41]. In this study, the density of each POI data in its surrounding fields was calculated by a function to determine the agglomeration area of the POI distribution. The formula is as follows:

$$f(s)={\displaystyle \sum _{i=1}^n\frac{1}{h^2}}\times k\times (\frac{s-c_i}{h})$$

(1)

where f(s) is the calculated value of kernel density, h is the distance decay threshold, n is the total number of samples, and $k\left(\frac{s-c_i}{h}\right)$ is the spatial weight function.

2.3.2. Two-Level Scoring Evaluation Model

The functions of production–living–ecological spaces are significantly influenced by the density of POIs and their physical areas. Therefore, this study constructed a two-level scoring evaluation model considering the physical area and density of POIs. The physical area of POIs refers to the area occupied by the POI building entity. First, according to the “Code for classification of urban land use and planning standards of development land”(GB50137-2011), different categories of POI data were geo-registered with the land use planned construction land data of Hangzhou’s main urban areas; thus, the average area of land patches corresponding to each type of POI data was obtained [30]. Then, based on the average area of each POI entity, the ratio of land patch of POI data to the total area was calculated to measure the area-weights of various POIs as the first-level weight value (

Table 2. The area-based weight of each POI category.

Space Type	Average Physical Area (hm2)	Weight	Space Type	Average Physical Area (hm2)	Weight
Companies	1.06	0.046	Accommodation services	1.03	0.045
Financial and insurance	0.57	0.024	Government agencies	1.01	0.044
Industrial park	1.97	0.085	Traffic facilities	0.99	0.043
Warehousing logistics	2.19	0.094	Medical security	1.48	0.064
Food and beverages	0.57	0.025	Science and education culture	3.28	0.141
Shopping	1.27	0.055	Commercial housing	2.26	0.097
Life service	0.93	0.040	Green lands of parks	0.74	0.032
Sports leisure	3.13	0.135	Scenic spot	0.72	0.031

). Second, the kernel density analysis method was used to obtain the density of POIs, which was defined as the second-level weight value. Finally, the comprehensive influence of each POI was calculated by using the following formula:

$$W_i=W_1\times W_2$$

(2)

where W_i is the comprehensive weight of the i-th POI, W₁ is the area-based weight of the i-th POI, and W₂ is the density-based weight of the i-th POI.

2.3.3. Identification Method of Production–living–Ecological Spaces

Based on the above processing results of the comprehensive influence of POIs, the POI data and grid unit were spatially connected so that the POIs could fall within the corresponding grid unit. Then, the comprehensive weights of all POI data in each grid were summed to obtain the function values of production, living and ecology. Third, the quadrat proportion method was used to calculate the proportion of the function values of each type to the total function values in the grid unit. The calculation formula was as follows:

$$S_j={\displaystyle \sum _{i=1}^n{c}_i\times W_i}$$

(3)

$$R_j=\frac{S_j}{{\displaystyle \sum _{j=1}^3{S}_j}}$$

(4)

where R_j is the proportion of the j-th functional value within the spatial unit, S_j is the functional value of the j-th space, c_i is the number of the i-th POI within the spatial unit, and W_i is the comprehensive weight of the i-th POI. When the value of R_j within the spatial unit is higher than 50%, this unit is determined as the corresponding j-th space. When the values of all types of R_j within the spatial unit do not reach 50%, it is determined to be a mixed functional space [30].

The scale of the spatial unit is the key to the identification of production–living–ecological spaces. According to the related literature on Chinese cities and the measurement of Hangzhou map, the length of the streets in southern China is between 300 and 1000 m [42,43]. Therefore, the “Create Fishnet” tool in ArcGIS 10.6 was used to divide the whole study area into 300 × 300 and 500 × 500 m grid units, respectively. Then, according to the land use type of the planning map in Hangzhou city, combined with POI data, 400 and 330 spatial units were manually identified as the sample data of PLE spaces within the 300 × 300 and 500 × 500 m grid units, respectively. Third, a random forest model was used to determine which scale of the spatial unit was more suitable for this study [29]. Specifically, the type of space within the spatial unit was used as the dependent variable, and the number of various POIs in the sample data was used as the independent variable. In total, 70% of the sample data was randomly selected as training samples and imported into the random forest model, and 30% of the sample data was used as validation data. The predicted function in this model was used to predict the type of space in spatial units.

The forecast results showed that compared with the 500 × 500 m grid unit, the prediction accuracy of production–living–ecological spaces significantly increased within the 300 × 300 m grid unit, and the overall prediction rate was 86.02%, indicating that the 300 × 300 m grid unit was more suitable for identifying the production–living–ecological spaces in Hangzhou city (

Table 3. The forecast accuracy of the random forest model at different scales.

Space Type	Production Space		Living Space		Ecological Space
Scale of spatial unit	300 m	500 m	300 m	500 m	300 m	500 m
Number of samples	140	120	180	160	80	50
Number of forecasts	119	91	165	134	64	32
Accuracy	84.29	75.83	91.67	83.75	80.00	64.00

).

2.3.4. Spatial Transition Matrix

To analyze the interactive relationships between different functional spaces, a spatial transition matrix was used to reveal the dynamic process of mutual transformation among different functional spaces at the beginning and end of a certain period [44]. The specific formula was as follows:

$$S_{ij}=\left[\begin{array}{cccc}{S}_{11}& S_{12}& \dots & S_{1n}\\ S_{21}& S_{22}& \dots & S_{2n}\\ \dots & \dots & \dots & \dots \\ S_{n1}& S_{n2}& \dots & S_{nn}\end{array}\right]$$

(5)

where, S_ij is the area of the i-th space converted to the j-th space, n is the number of space types before and after the transfer, and i and j are the space types before and after transfer, respectively. The row label of each element in the matrix (i) represents the original space type, and the column label of each element in matrix (j) represents the current space type.

2.3.5. Bivariate Spatial Autocorrelation Model

To explore the spatial interactive relationships between the three different functional spaces, the bivariate spatial autocorrelation model was employed for the spatial correlation between different functional spaces [45]. The spatial autocorrelation model was constructed based on the spatial proximity of geographical locations. Moran’s I and Local Moran’s I (LISA) index were used to measure global and local spatial correlations between them, respectively [46]. The specific formulas are as follows:

$$\begin{array}{l}Moran’s \mathrm{I}=\frac{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j\ne 1}^n{W}_{ij}·(X_i-\overline{X})·(X_j-\overline{X})}}}{S^2{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j\ne 1}^n{W}_{ij}}}} \\ Local Moran’s \mathrm{I}=\frac{X_i-\overline{X}}{S^2}{\displaystyle \sum _{i=1,j\ne 1}^n{W}_{ij}·(X_j-\overline{X})} \end{array}$$

(6)

where X_i and X_j are the values of unit i and unit j, n is the number of spatial units, W_ij is the spatial weights matrix, $\overline{X}$ is the average value of the unit, and S² is the variance of the observations.

3. Results

3.1. Identification and Validation of Urban Production–Living–Ecological Spaces

As shown in Figure 3, the distribution pattern of production–living–ecological spaces displayed significant regional differences. The production space was mainly distributed in the Xiaoshan and Yuhang districts, and the production space close to the city center tended to be clustered. The living space accounted for the majority of the study area and was mainly concentrated within the Hangzhou ring highway. The scale of ecological space was small, mainly surrounding the West Lake and Xixi Wetland. During 2010 and 2019, the amount of production–living–ecological spaces in Hangzhou expanded rapidly (

Table 4. The number of PLE spaces in Hangzhou in 2010 and 2019.

Space Type	2010	2019	Annual Growth Rate (2010–2019)
Production space	3251	5571	7.14
Living space	5933	10485	7.67
Ecological space	256	551	11.52
Mixed functional space	26	89	24.23

). The number of production spaces increased by 2320, with an annual growth rate of 7.14%. The living space expanded rapidly, with an increase of 4552 during the study period and an annual growth rate of 7.67%. The ecological space and mixed functional space increased less, with an increase of 295 and 63, respectively, during the study period.

Validation of identification results was necessary. In this study, a total of 20 validated plots, including 180 grids, were randomly selected to superimpose the land use situation of the study area in 2019. Based on the above validation method, the accuracy of using POI data to identify the production–living–ecological spaces was obtained (

Table 5. The validation results of PLE spaces.

Number	Space Type	Land Use Planning Types	The Correct Grid Proportion
1	Production	Industrial land	8/9
2	Living	Highway land	8/9
3	Living	Urban residential land	9/9
4	Living	Highway land, forest land	7/9
5	Production–living	Mining, highway land	8/9
6	Production	Logistics storage, commercial service facilities land	7/9
7	Production	Industrial land	9/9
8	Living–ecological	Highway land, forest land	7/9
9	Production–living	Railway, urban residential land	8/9
10	Production	Industrial land	9/9
11	Ecological	Green, forest land	6/9
12	Production–living	Science and education, commercial service facilities land	8/9
13	Production	Industrial, rural residential land	8/9
14	Living	Commercial service facilities, urban residential land	9/9
15	Ecological	Commercial service facilities, forest land	7/9
16	Production	Industrial, urban residential land	9/9
17	Production-living	Industrial, road land	8/9
18	Production	Mining, road land	7/9
19	Production–living	Industrial, rural residential land	9/9
20	Production–living	Industrial, urban residential land	9/9
Total			160/180

). Generally, the number of grids with correct identification of production–living–ecological spaces in 180 grids in the validation area was 160, and the identification accuracy was 88.9%, indicating that the identification method used in this study was highly reliable, and that the precision of the identification results was relatively high. Specifically, the identification results of production and living spaces were relatively accurate, while the matching degree of ecological space identification was relatively low. The main reason was that the POI data of scenic spots and parks existed in the form of points, resulting in a much smaller coverage than the actual areas, which affected the identification accuracy of some ecological spaces.

3.2. Spatial Pattern of Production–Living–Ecological Spaces

To analyze the spatial evolution characteristics of urban production–living–ecological spaces, the kernel density analysis was used to reveal the agglomeration pattern of different functional spaces (Figure 4).

In terms of the production space, the high-value areas of kernel density of production space were concentrated on the south bank of the Qiantang River and the north of Hangzhou in 2010. These areas are the major economic development zones and industrial parks of Hangzhou, such as the Xiaoshan economic development zone, Hangzhou high-tech industrial development zone, and Linping Industrial Park. In 2019, the production space of Hangzhou changed obviously, and the high-value areas of kernel density of production space transferred to the east and north of Hangzhou, mainly distributed in Xiaoshan and Yuhang districts. Additionally, some production space clustering areas began to appear in the south of Xiaoshan District and the west of Yuhang District. For the living space, the high-value areas of kernel density of living space were distributed within the ring highway of Hangzhou and along the Qiantang River in 2000, presenting a contiguous distribution pattern. These areas are the core urban areas of Hangzhou, with high population concentration. In 2019, the high-value areas of kernel density of living space expanded to the east and west along urban main roads, and the living space in northern Yuhang District and eastern Xiaoshan District experienced rapid expansion. In terms of the ecological space, the high-value areas of kernel density of ecological space were mainly concentrated around the West Lake and Xiang Lake in 2000. These areas are famous scenic spots in Hangzhou with rich ecological landscape resources. In 2019, the high-value areas of kernel density of ecological space expanded to the surrounding area, and some new ecological space appeared sporadically around the living space.

3.3. Spatial Interactive Analysis of Production–Living–Ecological Spaces

3.3.1. Spatial Transformations of Production–Living–Ecological Spaces

To quantify the scale and direction of mutual transfer among production, living, and ecological spaces, the spatial transition matrix was used to reveal the temporal interactive relationships between different functional spaces. As shown in Figure 5, the transfer from the production space to other space was the largest between 2010 and 2019, and the number of transferred spaces reached 1102. Among them, the proportion of transfer scale from production space to living space accounted for the largest, accounting for 95.28% of the total amount. The number of transfer scales from other spaces to production space was 584, mainly obtained from living space. This indicated that the interaction between production and living spaces was the most frequent during the study period. Additionally, the transfer number from ecological space to living space was 91, while the number of transfer scales of living space to ecological space was 55, indicating that there was a trend of transferring from ecological space to living space. Other types of spatial transformations among production, living, and ecological spaces were not evident during the study period.

In the spatial distribution, the areas where the production space was transferred to living space were mainly distributed in western Xiaoshan District, Gongshu District, and Jianggan District, which were clustered along the main roads in Hangzhou. The areas transferring from living space to production space were mainly concentrated in the north of Xihu District, Jianggan District, and the east of Xiaoshan District. These areas are agglomeration areas of enterprises and industrial parks in Hangzhou.

3.3.2. Spatial Correlations of Production–Living–Ecological Spaces

To conduct the bivariate autocorrelation analysis, a spatial proximity matrix was constructed using GeoDa software, and the global Moran’s I statistic for kernel density values of production–living–ecological spaces in 2010 and 2019 were then calculated. As shown in Figure 6, the global Moran’s I statistics for production and living as well as living and ecological spaces were greater than 0, indicating that the spatial distributions of production and living spaces, and living and ecological spaces were not random but exhibited a significant positive correlation in Hangzhou. Additionally, the global Moran’s I statistic for production and living spaces was greater than that for living and ecological spaces, indicating that the positive spatial correlation between production and living spaces was higher than that for living and ecological spaces. The global Moran’s I statistic for production and ecological spaces was less than 0, indicating that the spatial distribution for production and ecological spaces displayed a negative correlation in Hangzhou. During 2010 and 2019, the global Moran’s I statistic for production and living space and for living and ecological spaces decreased from 0.365 to 0.206 and from 0.232 to 0.202, respectively, indicating that the positive spatial correlation between production and living spaces and between living and ecological spaces weakened. The global Moran’s I statistic for production and ecological spaces for production and ecological spaces decreased from -0.036 to -0.151, indicating that the negative spatial correlation between production and ecological spaces was enhanced during the study period.

Figure 6 displays that the bivariate spatial autocorrelation for production–living–ecological spaces revealed obvious spatial heterogeneity patterns. In 2010, the high–high correlation areas of production and living spaces were mainly distributed in the periphery of the core area of the main urban area of Hangzhou, including the northern parts of Xihu District, Gongshu District, Jiangganqu District, the Binjiang District, and western Xiaoshan District. In these areas, the population and high-tech enterprises are highly concentrated, and urban production and living spaces are intertwined. The low–high correlation areas were located around the Xihu Lake, and these areas are the core areas with high population density and complete commercial facilities and infrastructure. The high–low correlation areas were scattered on the edge of Xiaoshan and Yuhang districts, and these areas have many industrial parks, which are important industrial clusters in Hangzhou. In 2019, the high–high correlation areas reduced obviously and were transferred to the periphery. The low–high correlation areas expanded rapidly, mainly to the east and west. The high–low correlation areas increased in the southern and eastern parts of the main urban area of Hangzhou. For the spatial correlation of living and ecological spaces, the high–high correlation areas were located around the West Lake and Xiang Lake, and these areas are rich in natural landscapes and high-quality environments. There were few low–high correlation areas, which were mainly concentrated in southwestern West Lake. The high–low correlation areas were mainly concentrated in the east of Jianggan District and Yuhang District; these areas are densely populated and dominated by farmland. In 2019, the high–high correlation areas expanded to the surrounding areas, and the high–low correlation areas expanded to the core areas and the south of the Qiantang River. For the spatial correlation of production and ecological spaces, the area of high–high correlation areas was small, mainly distributed around Xiang Lake and Banshan National Forest Park. The low–high correlation areas were concentrated around West Lake, and these areas are rich in ecological resources and low-intensity human activities. The high–low correlation areas were widely distributed, mainly around the eastern and northern ring highway and in the eastern Xiaoshan District. These areas are emerging industrial parks in Hangzhou with convenient transportation. In 2019, the low–high correlation areas expanded to the west, while the high–low correlation areas became more spatially concentrated. Additionally, the low–low correlation areas of production–living spaces, living–ecological spaces, and production–ecological spaces were mainly distributed on the edge of the main urban area of Hangzhou, covered by contiguous farmland and forests.

4. Discussion

With the continuous expansion of global cities and the increasing shortage of resources, optimizing the urban spatial pattern to achieve sustainable development has become one of the biggest challenges in this century [47,48]. Urban production–living–ecological space identification and optimization in rapidly urbanized areas are receiving increasing attention [49,50]. The traditional identification and analysis of urban production–living–ecological spaces rely on remote sensing data and socioeconomic statistics. The application of remote sensing data is limited because it cannot reveal the diversity of human activities in urban areas [51]. The use of statistical data is subjective, and the statistical scale is usually based on administrative divisions, which cannot satisfy the high spatial heterogeneity in urban areas [22]. Emerging POI data can provide detailed information on human activities and meet high precision; thus, how to apply these data to accurately identify urban production–living–ecological spaces has become a popular topic.

This study proposed a new evaluation model to identify urban production–living–ecological spaces using the POI-based method. This method introduced the average physical area of POIs, together with the density of POIs, to further improve the recognition accuracy of urban production–living–ecological spaces. Compared with previous studies, the kernel density index usually represents the distribution characteristics of different land uses from POIs, but it fails to reveal the physical area of POIs [52]. For example, the physical area of enterprise POIs is larger than that of commercial residence POIs, but the density of enterprise POIs is lower than that of commercial residence POIs, which may result in a lower recognition probability of industrial production space [23]. In this study, the proposed two-level scoring evaluation model considered both the physical area and density of POIs, and the identification accuracy reached 88.9%, which was higher than that of other research on urban production–living–ecological space identification [29,30]. Therefore, the proposed evaluation model has high application prospects and can contribute to guiding urban space optimization and planning management.

The coordinated development of production–living–ecological spaces is the key to optimizing urban spatial patterns [53]. Therefore, revealing the interactive relationships of production–living–ecological spaces was of significance to better understand the evolution of urban spatial patterns. Previous studies have mainly focused on the coupling coordination of production–living–ecological spaces using a coupling coordination degree model and correlation analysis [23], ignoring the impact of spatiotemporal changes in production–living–ecological spaces on their interactive relationships. This study employed the spatial transfer matrix and bivariate spatial autocorrelation model to explore the interactive relationships of production–living–ecological spaces from the perspective of temporal transformation and spatial correlation, which can better understand the spatiotemporal interactions of urban production–living–ecological spaces in rapid urbanization.

The results showed that the mutual transformation between production and living spaces was more frequent during 2010 and 2019. This phenomenon mainly occurred within the ring highway of Hangzhou. These areas are the main areas for urban renewal in Hangzhou, a large number of industrial enterprises have moved out to industrial parks, and some old residential land was renovated into land for high-tech enterprises [54]. Spatially, the interactions between production and living spaces and between living and ecological spaces had significant positive spatial correlations. This indicated that urban production and ecological space are important supports for the development of living space. While there was a negative spatial correlation between production and ecological spaces, this can be explained by the fact that the ecological space of Hangzhou was mainly concentrated around West Lake and Xiang Lake, while the production space of Hangzhou was mainly distributed in the north and east regions, resulting in a negative spatial correlation between them. In addition, the high–high correlation areas of production–living spaces were far more than those of living–ecological spaces, indicating that there was unequal access to green space for residents in different locations. This phenomenon is more pronounced in southern Chinese cities than in northern cities [55,56]. Urban managers could optimize the distribution pattern of urban ecological space resources by building more pocket parks and road green belts and by formulating green infrastructure planning to match urban living space [57].

Although the proposed evaluation model is proven to be effective in identifying urban production–living–ecological spaces, there were still some limitations in the current study. First, there is some uncertainty in POI data. To some extent, POI data are only a representation of different elements of virtual geographical space, which is still different from the actual development of urban spatial elements [58]. Therefore, other types of data, such as remote images, land use data, government statistics, and field survey data, could be used to supplement related research. Second, this study only focuses on the identification of urban production–living–ecological spaces and their interactions but lacks exploration of the driving mechanism and optimization mode of the production–living–ecological spaces. These limitations need to be further improved upon future research to provide a scientific basis for the sustainable development of urban space.

5. Conclusions

As the optimization of production–living–ecological spaces has become the directional guidance of urban development and planning, it is essential for urban planners and managers to accurately identify and evaluate the production–living–ecological spaces within the urban areas and their interactive relationships. Taking Hangzhou as an example, this study proposed a two-level scoring evaluation model to identify urban production–living–ecological spaces in Hangzhou in 2010 and 2019. A spatial transfer matrix and bivariate spatial autocorrelation model were then used to explore their interactive relationships. The main conclusions are as follows:

(1)

The proposed two-level scoring evaluation model considering both the physical area and density of POIs can be effective in accurately identifying urban production–living– ecological spaces, and the identification accuracy reached 88.9%.

(2)

Urban production space was concentrated on the south bank of the Qiantang River and the north of Hangzhou, urban living space was distributed within the ring highway of Hangzhou and along the Qiantang River in a contiguous distribution pattern, and urban ecological space was concentrated around West Lake and Xiang Lake, which are rich in natural and cultural landscapes.

(3)

During 2010 and 2019, urban production space transferred to the east and north of Hangzhou, urban living space rapidly expanded to the east and west along urban main roads, and urban ecological space expanded to the surrounding living areas.

(4)

The mutual transformation between production and living spaces was more frequent during the study period and was mainly distributed within the ring highway of Hangzhou. There were significant positive spatial correlations between production and living, and between living and ecological spaces, while significant negative spatial correlation appeared between production and ecological spaces. The spatial correlations of urban production–living–ecological spaces had obvious spatial heterogeneity.

Urban areas are complex heterogeneous spaces with land use multifunctionality. This study proposed a new two-level scoring evaluation model to accurately identify urban production–living–ecological spaces and their interactions, which can contribute to the literature and are useful in promoting the optimization and management of urban spatial patterns.