The Impact of the Type and Abundance of Urban Blue Space on House Prices: A Case Study of Eight Megacities in China

Abstract

Urban blue spaces (UBS) have been shown to provide a multitude of cultural ecosystem services to urban residents, while also having a considerable impact on the surrounding community’s house prices. However, the impact of different types of UBS and the effect of their abundance on house prices have been largely understudied. This study aims to address this gap by examining the impact of different types of UBS on house prices using eight megacities in China as a case study. Spatial hedonic price models are developed to assess the impact of different types of UBS on house prices, and differences in their impact across cities are identified. Variance partitioning analysis is also used to decompose the relative contributions of UBS variables and explore the relationship between UBS-attributable premiums and the abundance of UBS. The results indicate that lakes and the main river have a significant positive impact on house prices in most cities, while the impact of small rivers on house prices varies across cities. The influence of UBS variables differs significantly across cities, but these differences are not solely driven by the abundance of UBS. This study provides valuable information for UBS planning and management and contributes to the equitable distribution of urban public services.

1. Introduction

1.1. Gap Exists in the Abundance of Urban Blue Spaces between Cities

Urban blue–green spaces, a crucial component of urban ecosystems, offer numerous benefits to urban residents. These spaces have been shown to ameliorate air pollution and improve the thermal environment [1,2], foster connections between human beings and wildlife [3], enhance the physical and mental health of residents [4,5], and mitigate the risk of natural disasters [6,7].

While blue spaces and green spaces may offer similar ecosystem services, they differ in certain aspects. Numerous studies have established a strong positive correlation between urban green spaces coverage and wealth, as measured by GDP or GDP per capita [8,9,10]. The growth of cities and the rise in wealth often lead to an increased demand for urban green spaces, and well-developed cities have the resources to plan, construct, and manage such spaces [11,12]. However, the pressure to allocate land for development can result in the shrinkage of urban blue spaces (UBS) through the filling of lakes [13]. While some cities have created constructed wetlands, the abundance of UBS is largely determined by geographical factors such as landforms, water sources, and soil permeability [14]. Cities located along the coast or near large rivers typically have higher levels of UBS abundance compared to cities in arid or semi-arid regions [8,15,16], creating significant disparities in UBS abundance across different regions.

1.2. Supply, Demand, and Value of Urban Blue Spaces

The ecosystem services provided by UBS are reflected in the surrounding house prices in the form of an implied value [17]. Numerous studies have established a link between the supply and demand of ecosystem services and their value [18,19], leading us to hypothesize that the abundance of UBS in different cities affects the premiums of house prices attributed to UBS. From an economic and social psychological perspective, scarcity can increase the perceived value of an object; when demand is high and supply is low, people are likely to place a higher value on the object [20,21]. Hence, houses near UBS may command higher premiums in cities with scarce UBS resources. However, market behavior also exhibits herd behavior, where residents in cities with abundant UBS might prefer to purchase houses close to UBS due to its ample availability [22,23]. For UBS, which is a unique component of the urban ecosystem that has an implied value but is non-commodity and non-essential, understanding the relationship between its supply and demand and how it affects people’s evaluation of its worth can support equitable distribution of urban public services and inform UBS planning and management in different cities [24,25].

1.3. Capitalized Value of Urban Blue Spaces

Real estate prices are known to have close ties to the surrounding landscapes, as demonstrated by various studies [26,27,28]. Attractive landscapes, such as those with a desirable open view, proximity to the ocean [29], moderate grassland areas, an appropriate distance from wetlands [28], and a higher density of vegetation [30], are likely to result in a significant positive premium in house prices. Conversely, unpleasant views or the presence of undesirable facilities, such as industrial installations and cemeteries [31,32], polluted urban rivers [33], roads and railways [34], and agricultural land [35], may negatively impact house prices.

UBS are sometimes viewed as sources of mosquito breeding and potential risks of pollution and flooding [33,36,37]. However, modern cities have implemented measures to minimize these limitations in their ecological governance systems [38]. Despite not being considered commodities, the recreational benefits and landscape features generated by UBS are reflected in the price of housing through accessibility and visibility, which are rare in high-density cities and highly valued by urban residents [39]. Most studies that utilized the hedonic price method found that UBS have a positive impact on house prices, as indicated by positive premiums [29,40,41,42].

Recent studies have advanced our understanding of the mechanism by which UBS impact house prices. A summary of relevant empirical studies on the relationship between UBS characteristics and house prices is presented in

Table 1. Summary of recent hedonic studies involving UBS.

Type	Study	Location	Key Finding(s)
Canals and streams	[46]	Milan, Italy	Streams in the province of Milan are a negative factor in the decline of property values, while artificial canals are seen as a positive feature.
Streams	[47]	Guangzhou, China	Proximity to polluted urban streams significantly reduces house prices, and the marginal implied value of these stream views is ambiguous.
Streams, rivers, bays, bayous and water	[48]	Mobile and Daphne, Mexico	Coastal residents view proximity to the waterfront as one of the most important factors when purchasing houses and pay higher prices for houses near most waterfront types.
Wetland parks, lakes and rivers	[45]	Hangzhou, China	Urban wetlands significantly increase the value of housing within a 5 km radius and more significantly within a 1 km radius, and the housing market cycle also affects the estimated amenity value of urban wetlands.
Lakes	[43]	the United States	The water-clarity of lakes is most valued by homebuyers; a one-tenth of a meter change in water-clarity can result in a one percent change in house prices.
Canal	[41]	Hangzhou, China	Accessibility of the Grand Canal significantly boosts house prices, and the impact of the Grand Canal exhibits distance and regional heterogeneity
Wetlands	[44]	Franklin County, USA	Urban residents’ preference for wetland size and proximity to wetlands has an inverted U-shape, and people prefer broader upland buffers, green spaces, and protected wetlands.
Rivers	[48]	Chongqing, China	People will pay 0.92% more for a house that is 10% closer to two urban rivers, and a combination of river landscape and mountain views can further increase the value of the property.
Rivers, wetland parks, and lakes	[49]	Changsha, China	With the exception of small lakes, proximity to blue space can significantly increase the value of houses.
Ocean	[29]	Yokohama, Japan	“Very nice” open views and ocean views may have a positive premium, while “slightly nice” open views and ocean views may not.
Lakes and wetlands	[50]	Ramsey and Dakota County, USA	The landscape context around a blue or green spaces influences its value.

. The quality of water, presence of buffers, hydrological activity, special wetlands, surrounding land use, spatial dependence, and the housing market cycle are all factors that can influence residents’ preferences for UBS [43,44,45]. However, most of these studies have been limited to individual cities, leaving the evidence on the impact of different types of UBS on house prices in different cities, such as lakes of varying sizes, the main river (defined as the major river that crosses the urban center), and other rivers, vague and indistinct.

On the other hand, there has been a growing interest among researchers in recent years in the impact of the scarcity and abundance of urban amenities on house prices [51,52,53]. For example, Yuan et al. used the number of amenities per capita as an indicator of scarcity to examine the effect of both substitutable and irreplaceable amenities on house prices [52]. Dijk et al. defined water abundance as the percentage of water bodies within 1 km of houses [53]. Meanwhile, Brander and Koetse incorporated population density into a hedonic price meta-analysis to measure the scarcity of open spaces [51]. Despite the importance of scarcity and abundance in commodity prices, there is still a need for further discussion in the hedonic study of house prices.

To address these gaps in knowledge, a cross-city study was conducted to examine the impact of various types of UBS on house prices. We developed the following three hypotheses:

Hypothesis 1:

The effects of different types of UBS (such as large lakes, small lakes, small rivers, or the main river) on house prices vary among adjacent communities.

Hypothesis 2:

The impact of UBS on house prices varies across cities included in the study.

Hypothesis 3:

The variation in the impact of UBS on house prices across the cities is related to the abundance of UBS.

2. Materials and Methods

2.1. Methodological Flow

The study’s methodological flow is depicted in Figure 1. Our study proceeded as follows: (1) a sample of study cities was selected based on research requirements, and (2) the corresponding data were collected. (3) An ordinary least squares (OLS)-based characteristic price model was developed and evaluated. (4) Spatial regression was employed to estimate the model coefficients, which were used to examine the impact of various types of UBS on surrounding house prices. (5) The Tiao–Goldberger test was conducted to determine if the effects of each variable were statistically significant across cities. (6) Variance partitioning was utilized to determine the relative contribution of UBS variables to house prices and to investigate whether the abundance of UBS affects its perceived value by residents.

2.2. Study Areas Selection

A comparative analysis across cities was conducted to explore the impact of the type and abundance of UBS on house prices differed in cities [54]. The main urban areas of eight Chinese megacities (Beijing, Chengdu, Chongqing, Guangzhou, Kunming, Shenyang, Suzhou, and Wuhan) were selected as the study areas (Figure 2).

Since our study did not include seascapes, all coastal cities were excluded from the selection. These eight selected megacities represent the vast geo-locational and climate diversity in China, with significant differences in the abundance of UBS. At the same time, all these main urban areas are well developed, with sufficient populations to support prosperous real estate trading markets. Data from these megacities should offer a more in-depth understanding of whether people’s value assessment of UBS differs in cities with different UBS abundances.

2.3. Data Collection and Processing

China initiated the reforms of housing commercialization and housing distribution monetization in 1998. Over the past two decades, market-oriented condominiums have gradually become the dominant type of urban housing. These condominiums are typically developed and built by property developers as large communities and consist of multiple-unit buildings with 6–34 floors, each floor usually accommodating 1–6 households. Besides new condominiums developed by real estate developers, individuals can also buy and sell pre-owned condominiums. For most megacities, the transaction volume of second-hand condominiums is higher than that of new ones; thus, the average price of second-hand condominiums in the community was selected as the dependent variable [55].

Our study obtained the necessary data between July and August 2021. The average prices of second-hand condominiums in communities were gathered from the second-hand housing trading website, Beike (https://ke.com/ (accessed on 17 February 2023)), which is a New York Stock Exchange-listed company and holds a significant share of China’s real estate trading platform. After collecting the house price data and removing any incomplete, incorrect, or villa community data, we obtained the geographic coordinates of all communities through the Amap API (https://lbs.amap.com (accessed on 17 February 2023)). In addition, point-of-interest (POI) data, such as metro stations, primary schools, middle schools, 3A hospitals, cemeteries, funeral homes, and shopping centers, were also accessed from the Amap API. All UBS data, park data, and some large-scale facility data, such as universities and airports, were collected as polygon data from the Open Street Map (OSM) to avoid the deviation that might occur from predicting these large facilities as points. We collected all OSM and POI data within the study area and its outer administrative regions to ensure that data near the boundaries were not ignored. The distributions of UBS and community data in the study cities are presented in Figure 3.

2.4. Variable Selection and Model Derivation

According to the hedonic price model, the price of a market-oriented commercial condominium is influenced by several variables, including building characteristics, neighborhood characteristics, and UBS characteristics, and these variables may vary in different countries and regions. For instance, in China, the assignment of houses to compulsory education school districts implies that high-quality primary and middle schools can have a significant impact on house prices [56]. In consideration of the characteristics of China’s real estate market and prior research, we selected the following variables to form the hedonic price model (

Table 2. Definition of main variables.

Characteristic Type	Variables	Variable Definition
Dependent variable	PRCIE	Logarithm of the average house price of the community (CNY/m2).
Building characteristics	AGE	Year of the building construction (year).
Neighborhood characteristics	D_METRO	Logarithm of the straight-line distance between the community and nearest metro station.
	D_PRIM	Logarithm of the straight-line distance between the community and nearest key primary schools.
	D_MID	Logarithm of the straight-line distance between the community and nearest middle schools.
	D_UNI	Logarithm of the straight-line distance between the community and nearest university boundary.
	D_HOS	Logarithm of the straight-line distance between the community and nearest 3A hospital.
	OBNO_FAC	Obnoxious facility: the influence of various obnoxious facilities was set as follows: airport (3000 m outside), train station (500 m outside), railway (500 m per side), highways and overpass (500 m per side), cemetery (1000 m outside), funeral home (1000 m outside). It equals 0 when there is no obnoxious facility around the community, and one point is added for each.
	SHOP_CON	Shopping convenience: shopping center POI data were used to generate kernel density then obtain the corresponding value.
	D_PAKR	Logarithm of the straight-line distance between the community and nearest park.
	PARK_AREA	Logarithm of the area of the nearest park.
UBS characteristics	D_LAKE	Logarithm of the straight-line distance between the community and nearest lake.
	LAKE_AREA	Logarithm of the area of the nearest lake.
	D_RIVER	Logarithm of the straight-line distance between the community and nearest river.
	D_MRIVER	Logarithm of the straight-line distance between the community and the main river of the city.

). To offer a comprehensive understanding of the selected variables in each city, descriptive statistics are presented in

Table A1. Descriptive statistics of data in Beijing.

	Number	Mean	Min.	Max.	S.D.
PRCIE	4031	85,515.23	23,496	216,153	30,664.37
AGE	4031	22.17	3	91	8.69
D_METRO	4031	694.25	0.92	6621.60	552.77
D_PRIM	4031	1150.49	24.90	7058.71	881.33
D_MID	4031	1165.91	12.99	6237.26	888.31
D_UNI	4031	3246.91	0	12,318.07	1978.08
D_HOS	4031	1492.63	18.46	12,935.60	1290.42
OBNO_FAC	4031	0.76	0	3	0.74
SHOP_CON	4031	0.36	0	3.19	0.43
D_PAKR	4031	458.48	0	2387.06	310.65
PARK_AREA	4031	175,799.78	182.94	6,220,382.06	599,235.60
D_LAKE	4031	1977.13	10.70	10,516.20	1341.27
LAKE_AREA	4031	234,764.48	3256.07	3,511,994.66	371,126.57
D_RIVER	4031	1853.08	17.05	13,297.69	1601.75

,

Table A2. Descriptive statistics of data in Chengdu.

	Number	Mean	Min.	Max.	S.D.
PRCIE	5404	14,979.85	3414	47,606	5517.04
AGE	5404	17.24	0	74	6.96
D_METRO	5404	1038.65	32.48	13,812.13	1650.59
D_PRIM	5404	2045.74	57.71	18,047.10	2381.58
D_MID	5404	3155.31	41.18	20,995.26	4064.85
D_UNI	5404	2512.17	0	17,970.03	2255.43
D_HOS	5404	2189.38	26.33	12,865.17	2003.02
OBNO_FAC	5404	0.66	0	5	0.80
SHOP_CON	5404	0.37	0	5.15	0.50
D_PAKR	5404	671.87	0	8845.37	677.65
PARK_AREA	5404	128,047.18	596.60	9,268,506.00	595,473.11
D_LAKE	5404	2790.45	114.92	15,713.46	2291.58
LAKE_AREA	5404	144,296.74	3960.20	5,245,391.13	343,453.91
D_RIVER	5404	2556.32	14.75	17,679.39	3402.40

,

Table A3. Descriptive statistics of data in Guangzhou.

	Number	Mean	Min.	Max.	S.D.
PRCIE	3325	42,765.94	8789	156,648	20,971.48
AGE	3325	20.63	1	89	6.36
D_METRO	3325	1671.35	34.93	18,770.19	1555.01
D_PRIM	3325	2013.40	5.95	27,387.10	2647.12
D_MID	3325	652.37	34.49	6694.11	626.46
D_UNI	3325	3325.47	0	16,984.32	2379.64
D_HOS	3325	1642.69	34.56	19,585.22	1972.97
OBNO_FAC	3325	0.93	0	3	0.74
SHOP_CON	3325	0.69	0	3.31	0.62
D_PAKR	3325	552.76	0	4020.87	442.29
PARK_AREA	3325	119,984.73	116.80	6,287,184.80	469,010.00
D_LAKE	3325	1860.89	14.11	22,184.87	1535.04
LAKE_AREA	3325	121,421.78	1525.95	732,683.34	152,916.98
D_RIVER	3325	2515.72	12.36	14,586.98	2067.53
D_MRIVER	3325	2894.44	21.71	27,920.23	3072.27

,

Table A4. Descriptive statistics of data in Kunming.

	Number	Mean	Min.	Max.	S.D.
PRCIE	1371	14,195.41	5728	40,517	4026.52
AGE	1371	16.63	0	70	6.55
D_METRO	1371	786.65	29.57	6436.58	590.28
D_PRIM	1371	1672.93	43.64	9915.10	1361.88
D_MID	1371	1677.64	39.17	7455.01	1087.40
D_UNI	1371	2324.01	0	11,335.91	1956.31
D_HOS	1371	1811.33	54.63	9458.52	1479.94
OBNO_FAC	1371	0.47	0	3	0.66
SHOP_CON	1371	0.44	0	2.53	0.59
D_PAKR	1371	969.17	0	6340.09	776.86
PARK_AREA	1371	87,683.27	76.48	1,188,255.57	196,275.58
D_LAKE	1371	2108.75	15.72	5909.01	1315.69
LAKE_AREA	1371	32,235,165.66	8963.14	297,613,534.25	92,351,342.23
D_RIVER	1371	1898.30	17.60	11,104.78	1728.35
D_RIVER	5404	2556.32	14.75	17,679.39	3402.40

,

Table A5. Descriptive statistics of data in Shenyang.

	Number	Mean	Min.	Max.	S.D.
PRCIE	2331	10,372.38	2303	42,146	3531.86
AGE	2331	15.86	0	45	6.50
D_METRO	2331	1029.73	33.50	15,199.20	1587.51
D_PRIM	2331	2314.49	40.64	16,622.41	2414.25
D_MID	2331	2280.50	59.93	12,857.79	1862.71
D_UNI	2331	2338.61	0	17,830.52	1970.22
D_HOS	2331	2062.52	85.64	16,005.08	2186.70
OBNO_FAC	2331	0.91	0	3	0.78
SHOP_CON	2331	0.30	0	2.14	0.37
D_PAKR	2331	904.67	0	11,698.68	1328.62
PARK_AREA	2331	214,817.16	394.41	3,614,190.01	580,051.85
D_LAKE	2331	2640.90	19.14	17,067.73	1922.25
LAKE_AREA	2331	179,669.66	11,081.32	4,771,177.99	422,316.71
D_RIVER	2331	2354.33	13.43	15,812.14	2399.22
D_MRIVER	2331	5893.88	179.98	28,737.17	4353.85

,

Table A6. Descriptive statistics of data in Kunming.

	Number	Mean	Min.	Max.	S.D.
PRCIE	1682	25,860.22	4940.00	94,628.00	11,073.72
AGE	1682	15.89	0	50	7.97
D_METRO	1682	2050.41	19.00	37,202.74	4365.34
D_PRIM	1682	6203.6	66.07	51,464.21	7693.08
D_MID	1682	6126.67	64.88	47,508.82	7561.84
D_UNI	1682	5035.58	23.54	31,579.25	4146.53
D_HOS	1682	4886.81	70.17	42,433.44	5867.35
OBNO_FAC	1682	0.34	0	3	0.58
SHOP_CON	1682	0.32	0	1.81	0.31
D_PAKR	1682	1025.63	0	14,830.84	1728.18
PARK_AREA	1682	117,900.9	138.27	3,448,936.00	323,564.8
D_LAKE	1682	888.78	31.52	19,233.89	1787.66
LAKE_AREA	1682	2556.88	43.66	6174.22	1534.51
D_RIVER	1682	2.71×108	16,627.65	2.36×109	7.42×108
D_RIVER	1682	25,860.22	4940.00	94,628.00	11,073.72

,

Table A7. Descriptive statistics of data in Wuhan.

	Number	Mean	Min.	Max.	S.D.
PRCIE	2769	20,977.99	2380	57,141	6032.58
AGE	2769	18.23	0	51	7.34
D_METRO	2769	547.24	38.14	4021.35	348.54
D_PRIM	2769	2397.74	51.37	19,865.27	2138.33
D_MID	2769	1748.73	76.84	12,749.33	1208.48
D_UNI	2769	1858.43	0	16,826.68	1492.12
D_HOS	2769	1467.48	45.29	17,138.97	1241.46
OBNO_FAC	2769	0.72	0	3	0.72
SHOP_CON	2769	0.50	0	2.57	0.56
D_PAKR	2769	562.63	0	4647.28	418.51
PARK_AREA	2769	362,962.79	226.73	4,659,206.27	910,299.79
D_LAKE	2769	1638.53	16.93	6537.02	1136.58
LAKE_AREA	2769	7,166,933.74	31,372.12	45,027,924.22	11,970,837.80
D_RIVER	2769	3600.19	36.42	16,528.58	2325.22
D_MRIVER	2769	3914.42	63.98	15,066.57	2998.57

and

Table A8. Descriptive statistics of data in Chongqing.

	Number	Mean	Min.	Max.	S.D.
PRCIE	4084	12,307.95	4413	48,214	4204.70
AGE	4084	15.82	0	61	7.67
D_METRO	4084	834.55	3.64	14,985.13	1192.95
D_PRIM	4084	5077.99	54.22	29,445.41	5197.24
D_MID	4084	4014.44	52.51	20,900.39	3684.22
D_UNI	4084	3359.81	0	26,685.79	2663.94
D_HOS	4084	3156.37	5.74	23,150.91	2996.79
OBNO_FAC	4084	0.66	0	4	0.72
SHOP_CON	4084	0.44	0	3.16	0.60
D_PAKR	4084	996.40	0	11,839.42	877.70
PARK_AREA	4084	168,848.28	1578.74	3,438,314.32	424,480.96
D_LAKE	4084	1575.76	11.55	11,088.44	1433.50
LAKE_AREA	4084	69,245.46	25.06	1,350,632.02	164,264.84
D_RIVER	4084	2299.96	6.14	9141.65	1610.65
D_MRIVER	4084	2844.41	22.79	22,730.65	2668.20

in the Appendix A. The White test was used to diagnose the heteroskedasticity of the models across different cities. After confirming the presence of heteroskedasticity, a log transformation of all variables, including price, distance, and area, was performed to reduce heteroskedasticity and enhance the interpretability of the models [28].

2.5. Model Training

2.5.1. Hedonic Price Model

Residents’ perception of marginal implicit prices in UBS can be inferred from the hedonic price model proposed by Rosen [57]. This method argues that the value of a good depends on the number of attributes it contains and can also be used to explore the hidden prices of its influencing factors based on house prices, as follows:

$$\begin{array}{c} P\left(c\right)=f\left({\mathrm{Z}}_1,Z_2,\cdots ,{\mathrm{Z}}_n\right)+\epsilon \end{array}$$

(1)

where $P$ is the market price of a good; ${\mathrm{Z}}_1,Z_2,\cdots ,{\mathrm{Z}}_n$ represents various characteristics that affect the price of the property; and $\epsilon$ is the error term associated with uncertainty in the measurement of variables, unexplained variables, and the personal preferences of homebuyers.

The ordinary least squares (OLS) model is the basic form of the hedonic price model, known for its high interpretability [42,58]. However, several studies have shown that when the OLS model is used in the presence of spatial correlations, it can lead to biased marginal implied prices and inconsistent or inefficient estimates [59,60]. In addition, multicollinearity is a common issue in hedonic studies of house prices and must be considered. To address this, we tested the collinearity of the variables using variance inflation factors (VIFs). Furthermore, Lagrange multiplier (LM) diagnostics were applied to determine the spatial correlations in the models [61].

2.5.2. Spatial Autoregressive Combined Model

After confirming that there was no significant issue of multicollinearity between variables and that both spatial lag and spatial error dependence were present in our models, we chose to use the Spatial Autoregressive Combined (SAC) model instead of the ordinary least squares (OLS) model. The SAC model, also known as the Spatial Autoregressive with Autoregressive Disturbance or SARAR model, was first proposed by Kelejian and Prucha [62] and takes into account both spatial endogenous variables and spatial interactions, examining the spatial autocorrelation of both explanatory variables and error terms. The equation for the SAC model is as follows:

$$\begin{gathered} \begin{array}{c}{P}_c=\rho W{P}_c+X\beta +{\mu }_c \end{array} \\ \mathrm{and}{\mu }_c=\lambda \mathrm{W}{\mu }_c+{\epsilon }_c \end{gathered}$$

(2)

where $P_c$ is the logarithm of the average house price of the community (CNY/m²). The coefficient $\beta$ is the vector of the building, neighborhood, and environmental characteristics in hedonic modeling. The error term, ${\mu }_c$, for spatial dependence in the error terms is calculated using $\lambda$, the coefficient of disturbances, and ${\epsilon }_c$, the uncorrelated error terms. $\mathrm{W}$ is an N × N spatial weight matrix. Considering that our research is a cross-city analysis, the spatial weights matrix is calculated as a distance-based matrix with a 1 km specified bandwidth, which is specified as [45]:

$$\begin{array}{c} W=\{\begin{array}{c}{b}_{ij}/1 if b_{ij}\le 1\\ 0 if b_{ij}>1\end{array}\end{array}$$

(3)

All SAC models were performed using the “sacsarlm” function in the “spdep” package of R, and the maximum likelihood method was used for estimation.

2.5.3. Model Specification

Two models were adopted to answer the second research question. Model I, applied to all eight cities, included all variables except D_MRIVER and was used to investigate the impact of various sizes of lakes and small rivers on house prices. Model II, which builds upon Model I, was applied to the four cities with the main river and incorporated D_MRIVER as an additional explanatory variable to examine the influence of the main river on house prices.

2.5.4. Marginal Implicit Price Estimation

The preference of house buyers is implied in the regression coefficient, that is, the change in the dependent variable (condominium transaction price) caused by changes in the independent variables. Following previous empirical research [47,63], the percentage premium of each variable on house prices can be expressed as follows, after the spatial autocorrelation correction:

$$\begin{array}{c}\%\Delta PRICE=100\times \left[{\beta }_k{\left(1-\rho \right)}^{-1}\right]\end{array}$$

(4)

where ${\beta }_k$ is the estimated coefficient of the explanatory variable ${\mathrm{X}}_{\mathrm{k}}$, and $\rho$ is the estimated coefficient of the spatial autocorrelation.

2.6. Tiao–Goldberger Test

The eight cities selected for the study can be considered as eight distinct housing submarkets, given differences in factors such as location, population, income levels, and government policies. To assess variability among the submarkets, we employed the Tiao–Goldberger test. This F-test, which is widely used in hedonic studies of submarkets [64,65], tests the equality of regression coefficients in multiple regression models. The Tiao–Goldberger test assumes that the coefficients of the $j$ th explanatory variable (i.e., the implied price) are equal in each model, and the F statistic is calculated as follows:

$$\begin{array}{c}{F}_j^{TG}=\frac{{\sum }_{i=1}^L\frac{{\left({\widehat{b}}_{ij}-{\overline{b}}_j\right)}^2}{P_{ij}}}{{\sum }_{i=1}^{L}SS{R}_i}\times \frac{{\sum }_{i=1}^L\left(T_i-K_i\right)}{\left(L-1\right)}\end{array}$$

(5)

where:

$$\begin{array}{c} {\overline{b}}_j=\frac{{\sum }_{i=1}^L\frac{{\widehat{b}}_{ij}}{P_{ij}}}{{\sum }_{i=1}^L\frac{1}{P_{ij}}}\end{array}$$

(6)

$L$ is the number of housing submarket models; $b_{ij}$ is the estimated value of the $j$ th parameter in model $i$; $P_{ij}$ is the diagonal of the ${\left(X^{\prime }X\right)}_i^{-1}$ th parameter in model $i$; $SS{R}_i$ is the residual sum of squares of model $i$; $T_i$ is the number of samples in model $i$; $K_i$ is the number of explanatory variables in model $i$ (including the constant term). The test statistic has the F-distribution with degrees of freedom $\left(L-1\right)$ and $\sum _{i=1}^L\left(T_i-K_i\right)$. A large value of $F_j^{TG}$ indicates that there is a significant difference in the $j$ th implied price among the submarkets.

2.7. Abundance of Urban Blue Spaces

Before commencing our study, it was necessary to consider how to define the abundance of UBS. To this end, we focused on several hedonic studies that involved abundance independent variables. For example, van Dijk et al. defined water abundance as the percentage of water within a 1 km radius around each house in their study of valuing water resources in Switzerland [53]. Meanwhile, Netusil et al. set buffers in 200-foott, 1/4-mile, and 1/2-mile around houses in Portland and classified cells into high-structure vegetation [66], low-structure vegetation, impervious surface, or water to determine the abundance of vegetation.

Given that residents of cities with different UBS abundances may have different perceptions of UBS values, we defined abundance in terms of cities rather than houses. UBS is defined as all forms of natural and manmade surface water in a city area [67]. Although blue space and urban water bodies are not the same concepts, we believe that the proportion of urban water bodies reflects the abundance of UBS. To quantify the UBS abundance, we collected land cover data for 2020 provided by Esri, which was derived from ESA Sentinel-2 imagery at 10 m resolution with an accuracy of 0.99 for water [68]. Therefore, we counted the number of pixels of water in each of the eight cities and defined UBS abundance as the percentage of water area to the total area in the study areas.

2.8. Variance Partitioning Analysis

We performed a variance partitioning analysis to assess the relative importance of the UBS variables in determining house prices in metropolitan areas [69,70]. This analytical approach partitions the total variance and calculates the fraction of variance explained by different groups of explanatory variables. As illustrated in Figure 4, the dependent variable (Y) is separated into four scores based on the contribution of the two groups of explanatory variables (X1 and X2): (1) the fraction based on both X1 and X2, as shown in [b]; (2) the separate contributions of X1 and X2, depicted in [a] and [c]; and (3) the unexplained variance, represented in [d]. The mathematical calculation behind this analysis can be found in Anderson and Gribble [71].

In our study, the variables were grouped into three categories: building characteristics, neighborhood characteristics, and UBS characteristics. We calculated the relative importance of UBS characteristics in Model I for eight cities and in Model II for four cities located along major rivers. This information was compared with the UBS richness of each city to investigate the differences in the contribution of UBS features to house prices in cities with varying UBS abundances.

3. Results

3.1. Results of Hedonic Price Models and Tiao–Goldberger Test

We employed a spatial econometric model in our analysis, starting with ordinary least squares (OLS) regression, as shown in

Table A9. OLS estimation results for Model I.

	Beijing	Chengdu	Guangzhou	Kunming	Shenyang	Suzhou	Wuhan	Chongqing
Intercept	14.470 ***	12.107 ***	13.866 ***	11.184 ***	11.382 ***	13.481 ***	11.697 ***	10.292 ***
AGE	0.000	−0.017 ***	−0.020 ***	−0.005 ***	−0.017 ***	−0.008 ***	−0.004 ***	−0.023 ***
D_METRO	−0.047 ***	−0.077 ***	−0.047 ***	−0.041 ***	−0.055 ***	−0.064 ***	−0.045 ***	−0.035 ***
D_PRIM	−0.081 ***	−0.121 ***	−0.086 ***	−0.041 ***	−0.055 ***	0.082 ***	−0.109 ***	−0.001
D_MID	−0.096 ***	−0.026 ***	−0.071 ***	−0.003	−0.028 ***	−0.059 ***	−0.058 ***	−0.104 ***
D_UNI	−0.042 ***	0.021 ***	−0.103 ***	−0.014 ***	−0.010	−0.079 ***	−0.018 ***	0.030 ***
D_HOS	−0.113 ***	0.002	−0.089 ***	−0.048 ***	−0.019 ***	−0.121 ***	−0.017 ***	0.012 ***
OBNO_FAC	−0.019 **	−0.021 ***	0.025 ***	−0.026 ***	−0.015 **	−0.065 ***	−0.026 ***	−0.068 ***
SHOP_CON	0.021 *	0.040 ***	0.070 ***	0.041 ***	−0.002	0.288 ***	0.036 ***	0.026 ***
D_PAKR	0.006	−0.029 ***	−0.018 **	0.000	−0.022 ***	−0.049 ***	−0.016 ***	−0.017 ***
PARK_AREA	−0.008 ***	−0.005 **	0.006	0.003	−0.011 ***	−0.010 **	−0.001	0.001
D_LAKE	−0.110 ***	−0.064 ***	−0.069 ***	−0.090 ***	−0.021 **	−0.101 ***	0.001	−0.012 **
LAKE_AREA	0.023 ***	0.003	−0.003	0.006 ***	0.003	−0.011 ***	0.011 ***	0.012 ***
D_RIVER	0.003	−0.023 ***	0.056 ***	0.001	−0.048 ***	−0.029 ***	−0.003	0.033 ***
N	4031	5404	3325	1371	2331	1682	2769	4084
R2	0.411	0.440	0.465	0.298	0.291	0.471	0.295	0.392
AIC	1137.74	920.29	2590.31	−384.96	311.37	709.93	−395.40	393.03

Note: * p < 0.10, ** p < 0.05, *** p < 0.01. Where N is the number of observations. AIC = Akaike information criterion.

and

Table A10. OLS estimation results for Model II.

	Guangzhou	Shenyang	Wuhan	Chongqing
Intercept	13.829 ***	11.812 ***	11.918 ***	10.415 ***
AGE	−0.020 ***	−0.017 ***	−0.004 ***	−0.023 ***
D_METRO	−0.018 *	−0.039 ***	−0.044 ***	−0.038 ***
D_PRIM	−0.072 ***	−0.045 ***	−0.099 ***	0.012 *
D_MID	−0.063 ***	−0.025 **	−0.055 ***	−0.101 ***
D_UNI	−0.105 ***	−0.010	−0.024 ***	0.029 ***
D_HOS	−0.082 ***	−0.024 **	−0.017 **	0.016 **
OBNO_FAC	0.038 ***	−0.010	−0.018 **	−0.066 ***
SHOP_CON	0.092 ***	0.009	0.036 ***	0.031 ***
D_PAKR	−0.019 **	−0.007	−0.009 *	−0.013 **
PARK_AREA	0.006 *	−0.013 ***	−0.002	0.001
D_LAKE	−0.057 ***	−0.027 **	−0.010 *	−0.0224 ***
LAKE_AREA	−0.004	0.016 **	0.013 ***	0.013 ***
D_RIVER	0.087 ***	−0.052 ***	0.002	0.031 ***
M_RIVER	−0.088 ***	−0.088 ***	−0.039 ***	−0.027 ***
N	3325	2331	2769	4084
R2	0.493	0.340	0.307	0.396
AIC	2412.29	147.5	−437.43	366.86

Note: * p < 0.10, ** p < 0.05, *** p < 0.01. Where N is the number of observations. AIC = Akaike information criterion.

. To assess the presence of spatial dependence effects and multicollinearity, we conducted LM diagnostics and calculated variance inflation factors (VIFs). The results in

Table A11. VIFs for Model I.

Variables	Beijing	Chengdu	Guangzhou	Kunming	Shenyang	Suzhou	Wuhan	Chongqing
AGE	1.075	1.291	1.219	1.087	1.168	1.225	1.575	1.116
D_METRO	1.132	1.434	1.228	1.269	1.171	1.551	2.203	1.109
D_PRIM	1.198	2.751	2.789	2.134	1.804	1.615	4.916	1.543
D_MID	1.231	2.135	2.219	1.876	1.497	1.664	3.869	1.261
D_UNI	1.114	1.262	1.396	1.163	1.676	1.205	2.017	1.394
D_HOS	1.205	1.412	2.034	1.596	1.756	2.014	4.544	1.371
OBNO_FAC	1.061	1.068	1.120	1.282	1.158	1.060	1.080	1.078
SHOP_CON	1.162	1.238	1.798	1.494	1.914	1.326	1.225	1.309
D_PAKR	1.090	1.136	1.218	1.155	1.427	1.573	1.645	1.199
PARK_AREA	1.126	1.231	1.260	1.168	1.245	1.207	1.229	1.331
D_LAKE	1.351	1.331	1.438	1.531	1.669	1.668	1.568	1.271
LAKE_AREA	1.084	1.665	1.163	1.068	1.211	1.164	1.141	1.460
D_RIVER	1.148	1.457	1.279	1.177	1.656	1.651	1.922	1.230

,

Table A12. VIFs for Model II.

Variables	Guangzhou	Shenyang	Wuhan	Chongqing
AGE	1.088	1.225	1.123	1.222
D_METRO	1.335	1.585	1.109	1.235
D_PRIM	2.172	1.635	1.651	3.277
D_MID	1.885	1.665	1.266	2.229
D_UNI	1.165	1.205	1.455	1.399
D_HOS	1.601	2.019	1.372	2.063
OBNO_FAC	1.296	1.062	1.115	1.127
SHOP_CON	1.521	1.329	1.309	1.826
D_PAKR	1.155	1.639	1.245	1.242
PARK_AREA	1.168	1.210	1.338	1.260
D_LAKE	1.545	1.672	1.394	1.672
LAKE_AREA	1.068	1.213	1.489	1.170
D_RIVER	1.340	1.655	1.253	1.286
D_MRIVER	1.358	1.197	1.461	1.875

,

Table A13. LM tests for Model I.

	Beijing	Chengdu	Guangzhou	Kunming	Shenyang	Suzhou	Wuhan	Chongqing
Moran’s I (error)	80.704 ***	63.271 ***	48.297 ***	19.635 ***	28.505 ***	23.315 ***	27.854 ***	54.593 ***
Lagrange Multiplier (lag)	160.543 ***	68.241 ***	25.711 ***	16.169 ***	28.854 ***	13.305 ***	41.361 ***	107.645 ***
Robust LM (lag)	12.209 ***	4.119 **	0.003	5.763 **	6.504	2.057	6.711 ***	24.063 ***
Lagrange Multiplier (error)	5744.473 ***	3512.861 ***	1992.569 ***	297.234 ***	687.013 ***	472.337 ***	636.883 ***	2573.780 ***
Robust LM (error)	5596.139 ***	3448.738 ***	1966.861 ***	286.827 ***	664.663 ***	461.089 ***	602.234 ***	2490.208 ***
Lagrange Multiplier (SARMA)	5756.682 ***	3516.980 ***	1992.572 ***	302.996 ***	693.517 ***	474.394 ***	643.594 ***	2597.852 ***

** p < 0.05, *** p < 0.01.

and

Table A14. LM tests for Model II.

Variables	Guangzhou	Shenyang	Wuhan	Chongqing
Moran’s I (error)	42.597 ***	23.737 ***	26.852 ***	54.472 ***
Lagrange Multiplier (lag)	17.556 ***	28.692 ***	40.860 ***	104.015 ***
Robust LM (lag)	0.040	9.694 ***	7.531 ***	22.712 ***
Lagrange Multiplier (error)	1536.768 ***	468.815 ***	585.103 ***	2552.276 ***
Robust LM (error)	1519.252 ***	449.817 ***	551.764 ***	2470.974 ***
Lagrange Multiplier (SARMA)	1536.808 ***	478.508 ***	592.634 ***	2574.988 ***

*** p < 0.01.

indicated that all VIFs for the dependent variables in the OLS models were less than 5, and nearly all LM tests were statistically significant. This suggests that there were no significant issues with multicollinearity and that the spatial error and spatial lag were present in the dependent variables. Finally, we employed a spatial autoregressive conditional heteroskedasticity (SAC) model to estimate the coefficients.

The results of SAC estimation for Model I are displayed in

Table 3. Spatial autoregressive combined (SAC) estimation and marginal implicit price results for Model I.

	Beijing	Chengdu	Guangzhou	Kunming	Shenyang	Suzhou	Wuhan	Chongqing	Tiao–Goldberger F Statistic
Intercept	11.850 ***	11.235 ***	12.364 ***	10.783 ***	10.854 ***	12.817 ***	10.000 ***	9.843 ***
AGE	−0.002 *** (−0.21)	−0.013 *** (−1.35)	−0.016 *** (−1.59)	−0.006 *** (−0.55)	−0.016 *** (−1.61)	−0.007 *** (−0.68)	−0.005 *** (−0.53)	−0.017 *** (−1.75)	957.090 ***
D_METRO	−0.007 (N.S.)	−0.036 *** (−3.59)	0.003 (N.S.)	−0.022 ** (−2.20)	−0.015 (N.S.)	−0.035 *** (−3.52)	−0.025 *** (−2.56)	−0.014 ** (−1.46)	491.895 ***
D_PRIM	−0.038 *** (−3.92)	−0.067 *** (−6.79)	−0.085 *** (−8.55)	−0.057 *** (−5.76)	−0.035 *** (−3.61)	−0.011 (N.S.)	−0.079 *** (−8.07)	−0.023 ** (−2.30)	2305.907 ***
D_MID	−0.024 *** (−2.53)	−0.012 (N.S.)	−0.027 ** (−2.76)	−0.022 (N.S.)	−0.026 * (−2.66)	−0.053 *** (−5.34)	−0.051 *** (−5.28)	−0.059 *** (−6.02)	1161.473 ***
D_UNI	0.004 (N.S.)	0.007 (N.S.)	−0.049 *** (−4.98)	−0.003 (N.S.)	−0.007 (N.S.)	−0.094 *** (−9.50)	−0.010 * (−1.03)	0.020 *** (2.07)	4093.883 ***
D_HOS	−0.015 * (−1.56)	−0.017 ** (−1.89)	−0.045 *** (−4.52)	−0.022 * (−2.27)	−0.034 *** (−3.48)	−0.073 *** (−7.34)	−0.007 (N.S.)	0.019 * (1.89)	2256.853 ***
OBNO_FAC	−0.013 ** (−1.39)	0.003 (N.S.)	−0.027 ** (−2.70)	−0.016 (N.S.)	−0.011 (N.S.)	−0.023 (N.S.)	−0.018 ** (−1.82)	−0.034 *** (−3.46)	11.343 ***
SHOP_CON	0.034 (N.S.)	0.037 *** (3.70)	0.058 ** (5.87)	0.051 ** (5.18)	0.026 (N.S.)	0.151 *** (15.29)	0.036 * (3.69)	0.023 (N.S.)	16.418 ***
D_PAKR	0.006 (N.S.)	−0.011 ** (−1.13)	−0.002 (N.S.)	−0.017 ** (−1.77)	−0.008 (N.S.)	−0.014 * (−1.45)	−0.011 * (−1.17)	−0.004 (N.S.)	118.932 ***
PARK_AREA	0.001 (N.S.)	−0.001 (N.S.)	0.006 (N.S.)	0.008 (N.S.)	−0.002 (N.S.)	0.004 (N.S.)	−0.001 (N.S.)	0.005 (N.S.)	77.113 ***
D_LAKE	−0.053 *** (−5.46)	−0.054 *** (−5.42)	−0.058 *** (−5.90)	−0.070 *** (−7.13)	−0.037 ** (−3.72)	−0.079 *** (−7.96)	0.005 (N.S.)	−0.020 *** (−2.07)	1965.046 ***
LAKE_AREA	0.004 (N.S.)	−0.002 (N.S.)	0.000 (N.S.)	0.001 (N.S.)	−0.008 (N.S.)	−0.004 (N.S.)	0.004 (N.S.)	0.001 (N.S.)	136.186 ***
D_RIVER	−0.010 (N.S.)	−0.019 *** (−1.92)	0.030 ** (3.01)	0.015 (N.S.)	−0.038 *** (−3.90)	−0.001 (N.S.)	−0.006 (N.S.)	0.028 *** (2.83)	1857.718 ***
ρ	0.033 ***	0.009 ***	0.010	0.016 ***	0.019 ***	0.013 ***	0.027 ***	0.017 ***
λ	0.866 ***	0.795 ***	0.787 ***	0.652 ***	0.663 ***	0.664 ***	0.694 ***	0.770 ***
N	4031	5404	3325	1371	2331	1682	2769	4084
Pseudo-R2	0.683	0.581	0.600	0.393	0.414	0.595	0.389	0.563
AIC	−1359	−642.39	1633.20	−580.12	−127.98	265.77	−788.16	−946.87

Note: * p < 0.10, ** p < 0.05, *** p < 0.01. The marginal implicit price calculation results are shown in parentheses. N.S., not statistically significant. Where N is the number of observations. Pseudo-R² is Nagelkerke pseudo-R². AIC = Akaike information criterion.

and are consistent with our prior expectations. The presence of spatial lag and spatial error is further evidenced by the fact that the values of $\rho$ in seven models and $\lambda$ in eight models were positive and statistically significant at the 1% level. The variable “AGE” had a significant negative impact on house prices in all cities, with reductions ranging from 0.21% to 1.75%. In the cities where variables “metros,” “middle schools,” and “primary schools” were significant, they had a positive effect on nearby house prices. “Universities” and “hospitals” also had a positive effect in most cities, with the exception of Chongqing. “Obnoxious facilities” decreased house prices as expected, with reductions ranging from 1.39% to 3.46% for each additional obnoxious facility, as reported by Kim and Goldsmith [72]. On the other hand, “shopping convenience” had a positive influence on house prices. In four cities, “parks” had a significant positive impact on home prices, with a 1.17% to 1.77% increase for every 1 km closer to the nearest park, according to Kim et al. [27]. The variable “PARK_AREA” was not statistically significant in any of the cities.

For the Tiao–Goldberger test, although the F-statistics differed significantly among the variables, all variables exceeded the critical value at the 0.01 level of significance, which implied that the coefficients of all variables in the eight models cannot be considered equal.

In this study, the effect of distance to the lake on house prices was found to be significant in seven out of the eight selected cities, with prices decreasing by 2.07% to 7.96% for every 1 km increase in distance to the nearest lake. However, the size of the lake was not found to have a significant association with house prices in the hedonic analysis. The variable D_RIVER produced peculiar results. The river was found to have a significant influence on house prices in four of the eight study cities; however, the direction of this influence varied. In Shenyang and Chengdu, house prices increased by 1.92% and 3.90% for every 1 km closer to the river, while in Guangzhou and Chongqing, they decreased by 3.01% and 2.83%, respectively. In Model II, the variable M_RIVER was added and estimated for four cities with a main river. The results and corresponding house prices are shown in

Table 4. SAC estimation and marginal implicit price results for Model II.

	Guangzhou	Shenyang	Wuhan	Chongqing	Tiao–Goldberger F Statistic
Intercept	12.798 ***	11.463 ***	11.311 ***	10.002 ***
AGE	−0.016 *** (−1.57)	−0.016 *** (−1.61)	−0.005 *** (−0.53)	−0.017 *** (−1.74)	596.380 ***
D_METRO	0.004 (N.S.)	−0.014 (N.S.)	−0.027 *** (−2.76)	−0.016 *** (−1.67)	358.225 ***
D_PRIM	−0.078 *** (−7.81)	−0.030 *** (−3.03)	−0.073 *** (−7.48)	−0.016 (N.S.)	3194.705 ***
D_MID	−0.031 ** (−3.15)	−0.026 ** (−2.65)	−0.051 *** (−5.27)	−0.054 *** (−5.43)	571.594 ***
D_UNI	−0.052 *** (−5.21)	−0.009 (N.S.)	−0.014 ** (−1.41)	0.020 ** (2.04)	3403.865 ***
D_HOS	−0.046 *** (−4.63)	−0.030 *** (−3.07)	−0.007 (N.S.)	0.022 ** (2.20)	2981.845 ***
OBNO_FAC	−0.022 ** (−4.63)	−0.010 (N.S.)	−0.015 * (−1.51)	−0.032 *** (−3.25)	6.091 ***
SHOP_CON	0.073 *** (7.30)	0.032 (N.S.)	0.039 ** (4.02)	0.028 (N.S.)	16.980 ***
D_PAKR	0.001 (N.S.)	−0.004 (N.S.)	−0.008 (N.S.)	−0.001 (N.S.)	31.664 ***
PARK_AREA	0.007 * (0.66)	−0.004 (N.S.)	−0.002 (N.S.)	0.006 (N.S.)	170.285 ***
D_LAKE	−0.057 *** (−5.72)	−0.036 *** (−3.65)	−0.004 (N.S.)	−0.026 *** (−2.64)	1367.763 ***
LAKE_AREA	−0.001 (N.S.)	0.005 (N.S.)	0.007 * (0.76)	0.002 (N.S.)	105.664 ***
D_RIVER	0.046 *** (4.59)	−0.040 *** (−4.05)	0.000 (N.S.)	0.030 *** (3.04)	3777.203 ***
M_RIVER	−0.071 *** (−7.14)	−0.099 *** (−10.07)	−0.045 *** (−4.66)	−0.036 *** (−3.66)	2599.007 ***
ρ	0.004	0.020 ***	0.026 ***	0.016 ***
λ	0.768 ***	0.601 ***	0.675 ***	0.7710 ***
N	3325	2331	2769	4084
Pseudo-R2	0.603	0.427	0.392	0.565
AIC	1603.40	−179.01	−800.08	−963.83

Note: * p < 0.10, ** p < 0.05, *** p < 0.01. The marginal implicit price calculation results are shown in parentheses. N.S., not statistically significant. Where N is the number of observations. Pseudo-R² is Nagelkerke pseudo-R². AIC = Akaike information criterion.

, and it was found that main rivers attracted a positive premium in all four cities, with price increases ranging from 3.66% to 10.07% for every 1 km increase in distance to main rivers.

3.2. Results of the Variance Partitioning Analysis

The results of the variance partitioning analysis for Model I and Model II are presented in Figure A1 and Figure A2, respectively. The relative contribution of the UBS variables in both models can be seen in Figure 5, along with the ranking statistics of the UBS abundance values for the eight cities. It was observed that the addition of main river variables in the four cities resulted in an increased contribution of UBS features, highlighting the significance of the M_RIVER variable in the UBS characteristics. Despite having a small sample size, our findings did not indicate a significant relationship between UBS abundance and the relative contribution of UBS, as confirmed by the absence of a significant trend in Pearson’s correlation coefficient. This suggests that the abundance of UBS does not affect the valuation of UBS.

4. Discussion

4.1. Premiums from Non-UBS Characteristics

Non-UBS characteristics are not the focus of our study, but some interesting results could be discussed. Age of housing had negative effects on house prices in all cities, except that it was slightly less sensitive to the variable age in cities with a long history and a high concentration of older homes, such as Beijing [42]. In Beijing, house prices decreased by only 0.21% for each additional year of housing age, while in Chongqing, they decreased by 1.75%. The metro station distance variable was significant in all cities except Beijing, Guangzhou, and Shenyang. We suspect that the prevalence of numerous metro stations throughout these two megacities, Beijing and Guangzhou, makes people less sensitive to their presence, whereas in the case of Shenyang, the smaller size of the city may also result in a reduction in the importance of metro stations [54,73,74].

In all cities, house prices were positively influenced by the proximity of key primary schools or key middle schools, providing further validation of China’s “school district housing” policy [56,75]. The effect of university proximity (D_UNI) was significant in Guangzhou, Suzhou, Wuhan, and Chongqing. However, universities in Guangzhou, Suzhou, and Wuhan had a positive impact on house prices, which may be attributed to the positive cultural atmosphere, environment, and sports facilities they provide [64,76,77]. In contrast, in Chongqing, the influence of universities was negative. Upon examining the spatial distribution of universities in Chongqing, known as “the mountain city,” we found that they were mostly located in suburban areas away from residential areas, making it difficult to drive up house prices [48]. Similarly, the negative impact on house prices imposed by 3A hospitals may also be due to the complex topography of Chongqing [78].

As expected, the presence of obnoxious facilities reduced house prices, with a decrease ranging from 1.39% to 3.46% for each additional obnoxious facility [72]. On the other hand, shopping convenience had a positive influence on house prices. In several cities, parks had a significant positive impact on home prices, with a 1.17% to 1.77% increase for every 1 km distance from the nearest park [27]. The PARK_AREA variable was insignificant in all cities, which is consistent with the findings of Morancho [79], who also found that park size does not significantly affect house prices in Castellón, Spain.

4.2. Premiums from Lake Characteristics

Almost all cities exhibited a significant positive impact on house prices due to nearby lakes, with the exception of Wuhan—known as the “city of thousands of lakes” [80]. Liu et al. suggested that an inverted U-shaped relationship exists between the distance from wetlands and house prices in Wuhan [28], which they attribute to the city’s high rainfall, arguing that proximity to wetlands increases the risk of flooding. However, our findings indicate that the positive impact of lakes on house prices is significant in Suzhou, where the average annual rainfall is similar to Wuhan, but the abundance of UBS is higher.

Upon reviewing the spatial distribution and descriptive statistics of our data, we determined that while Suzhou had a higher UBS abundance, the UBS in Wuhan was more evenly distributed. Our data showed that the median distance from communities to lakes was 1.4 km in Wuhan and 2.3 km in Suzhou, with most of the lake area in Suzhou concentrated in Taihu Lake, which is far from the urban area. Therefore, we speculate that the distribution pattern of lakes could explain Wuhan’s insensitivity to the proximity of lakes. Jiao and Liu conducted a separate analysis of East Lake (the most famous wetland park in Wuhan) in their Wuhan hedonic study and found that East Lake had a positive influence on the selling price of apartments located within an 800 m buffer, while other lakes did not [81]; therefore, the purpose for which the lake was built (e.g., as a wetland park) may also influence its effect on house prices. Additionally, the fact that not all lakes in Wuhan are built as wetland parks may also contribute to the insignificance of proximity variables for lakes in Wuhan.

The variable “LAKE_AREA” was not found to be significant in any of the cities studied. This aligns with the findings of previous research, which have concluded that lake size does not significantly impact lake values and may even lead to a decrease in some developing countries [82,83,84]. In urban areas, small lakes are sufficient to meet the needs of residents in terms of ecosystem cultural services, such as fishing, boating, swimming, camping, and sightseeing. As lake size increases, the marginal benefits of the services provided by most lakes may decrease, while potential risks such as pollution and flooding may increase. Additionally, large lakes are often fragmented or disappear entirely during urban development, making them uncommon in most urban areas. As a result, lake size is deemed to be of negligible significance in determining house prices from an urban perspective [85].

4.3. Premiums from River Characteristics

The results of the variable D_RIVER were opposite across the four cities, and there may be several reasons for this phenomenon. One possibility is the level of river pollution. Li et al. found that in Guangzhou, homebuyers displayed a strong negative preference for rivers with black and foul-smelling water [63]. A study in Milan also showed that polluted streams negatively impacted real estate values, while artificial canals with good water quality were seen as a positive feature [46]. Rivers with high levels of pollution lack aesthetic value and pose potential health risks associated with disease, making them an unattractive feature for homebuyers. Conversely, pristine or restored rivers offer attractive natural landscapes and recreational spaces for nearby residents and are highly valued by homebuyers [86].

Topographical factors may also play a role in the value of rivers. A study in Chongqing found that river landscapes did not affect house prices [87], while mountain views did. In high-density cities, residents are often able to see mountains but not rivers. Münch et al. found that riparian buffers in Denmark increased the aesthetic value of nature/landscape for nearby residents and positively influenced property values [88]. Similar findings were reported by Mei et al., suggesting that buffer zones may be important in determining the value of rivers [44].

In all four cities, the main river brought a significant positive premium to house prices on both sides, unlike smaller rivers. These large rivers, similar to the Thames in London, the Mississippi in New Orleans, and the Nile in Cairo, hold significant symbolic significance for the cities they run through, in addition to the expected landscape attributes. As cities have developed, a considerable number of scenic river parks have been built along the main river, which has indirectly elevated the recreational and landscape value of the rivers and driven up property prices on both sides [39,89].

4.4. The Impact and Abundance of Urban Blue Spaces

The results of the Tiao–Goldberger test indicate the variability of the impact of the UBS variable on house prices across cities. However, variance partitioning results show that the abundance of UBS is not significantly correlated with the relative contribution of UBS variables. This implies that the abundance of UBS does not impact people’s assessment of UBS values. The services provided by UBS are reflected in house prices as implicit costs, but their value may be influenced by various factors, such as the water quality, pollution status, and topography of the city [55,81]. Additionally, the impact of UBS may be limited to the periphery of the UBS, as shown in studies by Netusil et al. and van Dijk et al. [53,66], which found a significant positive impact on house value for a percentage of water bodies within 1/2 mile and 1 km buffers of a house. However, no relevant studies support the impact of UBS at greater distances. It can be concluded that house buyers take into account the UBS surrounding a house, but the abundance of UBS in the city is not considered. Hence, the abundance of UBS at the urban scale does not affect people’s evaluation of UBS value.

Finally,

Table 5. Hypotheses, previous studies, and our conclusions and contribution.

Hypotheses	Previous Studies		Our Conclusions
Lakes attract a positive impact on house prices	Support	[43,49,53,55,84,85,87]	Support in most cities
	Contradiction	-
	Ambiguity	[81]
The size of the nearest lake has a positive effect on house prices	Support	[43,49]	Contradiction
	Contradiction	[84,85]
	Ambiguity	-
The main river attracts a positive impact on house prices	Support	[41,48,49,53,85,87]	Support
	Contradiction	-
	Ambiguity	-
Small rivers attracts a positive impact on house prices	Support	[85]	Ambiguity
	Contradiction	[33,53]
	Ambiguity	[63,65,86,88]
The impact of UBS on house prices varies across cities	-	-	Support
The variation in the impact of UBS on house prices across the cities is related to the abundance of UBS.	-	-	Contradiction

provides a comprehensive summary of our study’s findings, including how they either support or differ from previous studies, as well as our final conclusions regarding the three hypotheses tested.

4.5. Implications for Planning and Management of Urban Blue Spaces

A key approach to reducing the loss of urban ecosystem services is to incorporate more shared blue and green spaces 30. The equalization of public services in urban blue and green spaces has long been an important government goal. For urban planners and managers, creating public river parks and wetland parks based on urban lakes and the main river, instead of constructing residential houses and designating them as exclusive landscapes for a few people, cannot only increase nearby property values but also can help optimize urban ecosystem services. For smaller rivers with substantial inter-urban differences, riparian greening can also be effective in enhancing the services they provide, but addressing river pollution may be a more pressing concern 47. The value of UBS does not fluctuate like commodities in response to changes in supply and demand, and the factors influencing their value are more centered around the characteristics of the individual UBS. Therefore, the planning and management of UBS in different cities should be tailored to the characteristics of each individual UBS, rather than being undifferentiated.

5. Conclusions

The examination of the impact of types and abundance of UBS on its value can facilitate tailored planning and management of UBS. In this study, we analyzed eight megacities in China, utilizing a hedonic pricing model incorporating spatial lag and spatial error to explore the effects of different types of UBS on house prices in each city. The Tiao–Goldberger test was employed to verify variations in these variables across cities. Variance partitioning analysis was then used to determine the relative contribution of UBS characteristics in the eight cities, which was compared to UBS abundance to determine if residents in cities with varying UBS abundance value UBS differently.

The results indicate that lakes, rivers, and the main river can all exert a positive influence on surrounding house prices to some extent, with lakes and major rivers driving positive premiums. Meanwhile, the impact of common rivers remains uncertain. The impact of UBS variables was found to significantly vary across cities, but these differences were not caused by UBS abundance. Furthermore, the observed differences in the impact of university and hospital distance variables between Chongqing and other cities, and the impact of lake distance variable between Wuhan and other cities, suggest that city topography and land use patterns may also play a role in individuals’ assessments of the value of these characteristics.

Several limitations exist in our study. First, we utilized community-based house price data, which can exclude the interference of factors such as household type and house size. However, in reality, individual houses with river or lake views may have significantly different prices than those without these views in the same community 33, which may introduce some estimation errors. Second, our findings suggest that the abundance of UBS at the urban scale does not affect people’s assessment of its value. In contrast, at the scale of individual houses, some studies reach the opposite conclusion [53,66], indicating that the impacts of UBS are influenced by the scale of the study. Future research should explore the distance or marginal effects of UBS impacts. Third, due to the lack of historical data, we relatively overlooked the variation in the influence of UBS in the time dimension. The abundance of UBS and other characteristics may vary due to changes in land use and land cover. Combining panel data and time series analysis can provide a more comprehensive explanation of the influence of UBS on house prices and people’s valuation of UBS. Additionally, our study is based on the Chinese real estate market only. Given the differences in real estate markets across different countries, our findings may not be applicable universally. In the future, conducting multi-country hedonic price studies will be more appropriate for a better understanding of the global real estate market. Finally, our results suggest that topography and landscape patterns may affect the coefficient estimates of the variables in the model. Therefore, further investigation of the effects of these factors is warranted in future studies.