Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The economic value of urban landscapes in a suburban city of Tokyo, Japan: A semantic segmentation approach using Google Street View images

The economic value of urban landscapes in a suburban city of Tokyo, Japan: A semantic... JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING https://doi.org/10.1080/13467581.2022.2070492 The economic value of urban landscapes in a suburban city of Tokyo, Japan: A semantic segmentation approach using Google Street View images a b c b Masatomo Suzuki , Junichiro Mori , Takashi Nicholas Maeda and Jun Ikeda a b Center for the Promotion of Social Data Science Education and Research, Hitotsubashi University, Tokyo, Japan; Graduate School of Information Science and Technology, The University of Tokyo, Tokyo, Japan; Department of Information System Engineering, School of System Design and Technology, Tokyo Denki University, Tokyo, Japan ABSTRACT ARTICLE HISTORY Received 2 February 2022 Visual impression of urban landscape has been investigated in detail through behavioral Accepted 22 April 2022 experiments and questionnaire surveys in the field of architecture. However, in order to give an incentive to build and maintain a good residential environment, an economic consideration KEYWORDS of the urban landscape across space is also an important aspect. Employing a semantic Hedonic price model; segmentation approach using Google Street View images, we investigate the relationship landscape; machine learning; between urban landscapes and property prices in low-rise residential areas in a suburban Google Street View images; city of Tokyo, Japan. Such visual images are used to represent the landscape of both surround- Japan ing districts in general and the street-level landscape; more specifically, the latter is derived after controlling for the district fixed effect. We first show that greenery, openness and visual enclosure are positively correlated with property price at street level. We also investigate the value of urban landscapes commonly seen in Japan: (i) the presence of a power pole is negatively correlated with property price at district level; and (ii) the presence of a road shoulder or farmland, either of which may disrupt the continuity of a residential area, does not exhibit negative correlations with property price at either district or street level. time to nearest station), and land use around the prop- 1. Introduction erties. The hedonic regression analysis separates the Visual impression of urban landscape has been widely contribution of each component to property prices. investigated in detail through behavioral experiments The economic value of a landscape is captured using and questionnaire surveys in the field of architecture. a landscape index, which provides additional explana- Urban landscape impressions are basically based on tory variables to the basic characteristics. The land- components related to greenery, openness, and visual scape index has traditionally been created through enclosure (e.g., Hirate and Yasuoka 1986; Ishikawa et al. a field survey, whereby criteria are set and scored by 1995; Nishio and Ito 2015, 2020; Takei and Fukushima a researcher (Gao and Asami 2007). Alternatively, 1983). The other elements such as power poles (elec- researchers collect detailed land use information (Gao trical wires) also form the urban landscape impression and Asami 2001), advertising information or 3D geo- (Oku 1985). These urban landscapes may be recog- graphical information on whether a property offers nized as a street level landscape or as a landscape a scenic view, such as of the ocean (Jim and Chen having a certain spatial unit such as a district (Koura 2009; Yamagata et al. 2016), to create the landscape and Kamino 1995, 1996). However, whether these index. impressions are reflected in economic valuation has This paper further extends these methods by cap- been less intensively investigated. In order to give an turing urban landscape factors using Google Street incentive to build and maintain a good residential View images. Recent research has recognized that by environment, an economic consideration of the combining Street View images and artificial intelli- urban landscape across space is also an important gence (AI) technology for image recognition, there is aspect. the potential to greatly improve urban landscape ana- The quality of surrounding landscapes has lyses (Biljecki and Ito 2021). Yang et al. (2021) have a potential impact on real estate price formation quantified the amount of greenery in Street View because landscapes represent elements that make up images based on the percentage of pixels recognized the quality of the surrounding neighborhood. There as greenery in the image. Using open-source image are many factors forming property prices, including segmentation methods that employ deep learning building characteristics (e.g., property age and floor (e.g., SegNet), recent studies have decomposed Street area), land characteristics (e.g., land area and walking View images into each landscape element and have CONTACT Masatomo Suzuki suzuki.m@r.hit-u.ac.jp Center for the Promotion of Social Data Science Education and Research, Hitotsubashi University, 2-1, Naka, Kunitachi, Tokyo, 186-8601, Japan © 2022 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group on behalf of the Architectural Institute of Japan, Architectural Institute of Korea and Architectural Society of China. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. 2 M. SUZUKI ET AL. created indexes typically based on the pixel ratio (Chen space within a private lot – is a typical characteristic of et al. 2020; Fu et al. 2019; Ito and Biljecki 2021; Tang Japanese cities. Given that residential lots tend to be small and Long 2019; Ye et al. 2019a, 2019b; Zhang and in Japanese cities, a typical parking space is not hidden by Dong 2018). Using these new approaches, recent lit- fences or greenery (as will be shown in Section 4) and erature investigates the relationship between property disrupts the continuity of a residential area. Intuitively the price and landscape elements, while the landscape presence of such a car park would have a negative impact type has been largely restricted to the proportions of on neighboring property prices, but as this situation is greenery, sky and buildings present (Chen et al. 2020; common in Japan, its specific impact is one of the issues Fu et al. 2019; Yang et al. 2021; Ye et al. 2019a; Zhang examined in this paper. and Dong 2018). In Western contexts, other factors like In low-rise residential areas in a suburban city of “curb appeal” (that is, the attractiveness of the exterior Tokyo, this paper investigates the relationship of a property when viewed from a public space, such as between common Japanese urban landscape and a street or sidewalk), architectural style, and urban property prices via the following three steps. First, we design elements have been gradually investigated collect Google Street View images in front of detached using new image recognition technologies (Johnson, houses that have been sold. Second, we extract com- Tidwell, and Villupuram 2020; Lindenthal and Johnson ponents of the landscape using AI technology for 2021). image recognition, and create what we call the Greenery, openness, and visual enclosure are basic “urban landscape index.” Specifically, we apply the elements of urban landscapes in Japan and in other recent semantic segmentation approach proposed by countries as well. However, other elements of urban Tao, Sapra, and Catanzaro (2020), which as of landscapes differ greatly in Japan from those in August 2021 has achieved state-of-the-art results in Europe, the United States and even other Asian coun- two commonly used, large-scale open data sets: tries. Specifically, the existence of power poles above Mapillary and Cityscapes. This approach is impressive the ground, the presence of urban farmland within in its segmentation accuracy and label diversity com- residential areas, and private parking space facing the pared to previous studies. Third, we construct a model road, are landscape factors specific to Japan that can to explain the extent to which these components con- be tested from the Street View image data. First, the tribute to the transaction price. After controlling for undergrounding of power lines is not at all common in a sufficient range of characteristics of the properties Japanese cities. Whereas major cities in Europe (such and the neighborhoods considered, the strength of the as London and Paris) and Asia (for instance, Hong Kong correlation between the urban landscape index and and Singapore) have all become pole-free, Japan lags property price is derived from the “landscape” channel. behind, this goal being achieved in only 8% of Tokyo’s The component captured here as the urban landscape wards area and 6% of the city of Osaka. The existence index may represent the landscape of the surrounding of power poles at ground level in residential areas district in general (e.g., the landscape of a well- potentially lowers the value of properties, as the over- developed residence), or the street-level landscape head-to-underground conversion of electricity distri- more specifically. Thus, the landscape channel is dis- bution networks has been perceived positively in tinguished as district-level and street-level, the latter of Western countries, based on stated preferences which is derived after further controlling for the district (McNair et al. 2011; Tempesta, Vecchiato, and Girardi fixed effect. 2014). The rest of the paper is organized as follows. Section Second, farmland remains within residential areas on 2 describes the study area and property transaction the fringes of Japanese cities. Whereas in Western cities, data. Section 3 presents the methodology. Section 4 urban planning concepts, such as zoning and greenbelt presents examples of semantic segmentation of the additions, have been applied to encourage controlled Japanese landscape and develops hypotheses. urban growth, Asian cities have historically placed land Section 5 presents the variables and summary statistics use patterns of urban and rural character next to each for the following hedonic regression analysis. Section 6 other (Yokohari et al. 2000). Although this type of land- shows the empirical results. Section 7 concludes the scape potentially gives us a positive impression through paper. the provision of rural environments for urban residents, the presence of farmland may disrupt the continuity of an urban residential area; its specific impact is one of the 2. Study area and property transaction data issues examined in this paper. Third, unlike the parking To investigate the economic value of urban landscape space on road shoulders seen in Western cities, private in typical low-rise residential areas commonly seen in parking space facing the road – that is, paved parking the suburbs of metropolitan areas in Japan, the Ministry of Land, Infrastructure, Transport and Tourism. “Status of development of non-pole systems (domestic and overseas).” Available at: https://www. mlit.go.jp/road/road/traffic/chicyuka/chi_13_01.html (in Japanese; accessed 9 December 2021). JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 3 analysis is conducted in Hachioji city, a suburban city Table 1. Summary statistics. (a) Property characteristics of Tokyo’s metropolitan area. The city is located more Mean S.D. than 30 km away from Shinjuku Station, a terminus in Dependent variable central Tokyo. The population is around 577,000 Price [10,000 JPY] 2466.3 1115.9 according to the 2015 Census and has remained Building characteristics Newly built 0.250 almost stable since 2005, and a relatively stable hous- Age [years] 17.5 14.4 ing demand is expected in the analysis period. We 2 Floor area [m ] 107.3 27.9 employ samples in areas with designated zoning Non-timbered 0.109 Land characteristics (that is, we exclude mountainous areas in which zon- Land area [m ] 169.1 63.2 ing is not applied), especially focusing on low-rise Walking time to station 5.9 7.4 [min] residential areas (i.e., category I, exclusively low-rise Bus to station 0.530 residential zones) in the main analysis. In our transac- With parking space 0.755 Distance to junior high 0.777 0.401 tion samples of detached houses, 61% are located in school [km] low-rise residential areas. Floor-area ratio (FAR) [%] 75.638 13.823 Front road width We employ the transaction data of detached Less than 4 m 0.034 houses collected by an association of real estate 4–5 m 0.179 agents through the Real Estate Information 5–6 m 0.196 2 6–7 m 0.254 Network System (REINS). Information on real 7–10 m 0.053 estate transactions is recorded in these data, in 10 m or more 0.093 Unknown 0.193 the same way as in the Multiple Listing Service Land use within 500 m (MLS) in the United States. The REINS is the only radius [ratio] High-rise building 0.020 0.038 real estate transaction system designated by the Dense low-rise building 0.021 0.081 Ministry of Land, Infrastructure, Transport and Low-rise building 0.630 0.168 Vacant land 0.016 0.030 Tourism (MLIT) in Japan, and is a representative Park and green 0.005 0.014 system for real estate companies to register trans- Farmland 0.035 0.057 action information. The registration of data is con- Number of observations 800 sidered to have sufficient coverage, as it is (b) Urban landscape index created through sematic segmentation mandatory for a real estate agent or company to of the Google Street View images register the information on transacted properties Variable Mean Min Max under exclusive brokerage service agreement, Mean S.D. Mean S.D. Mean S.D. although not for all properties. Indeed, transaction Vegetation 0.108 0.080 0.236 0.170 Sky 0.237 0.059 0.346 0.053 volumes in each regional submarket (mainly at Building 0.200 0.088 0.068 0.059 prefecture level) based on the REINS data have Building + Wall + Fence 0.348 0.102 been announced to the public as market reports. Pole 0.007 0.006 0.027 0.025 Sidewalk 0.029 0.023 0.094 0.057 In this context, the number of samples and their Sidewalk + Car 0.042 0.030 property characteristics studied in this paper is Terrain 0.025 0.039 0.091 0.120 likely to be representative of the submarket of Hachioji city. We employ newly built and resold detached houses 3. Methodology transacted from 2016 to 2019. There are 800 observa- tions in total in the low-rise residential areas consid- Figure 1 shows the flow of our analyses. First, using Street ered (and 1,307 samples in the entire area of the View Static API, 12 Google Street View images (i.e., every Hachioji city), after truncating samples with missing 30 degrees) are collected in front of each transacted or atypical information. The addresses are geocoded property. To be more specific, for each property, the using Geocoding API, provided by Google. The trans- nearest point on the road is chosen. Although the land- action prices and other property characteristics (which scape impression may change depending on the retriev- will be shown later in Table 1(a) are recorded in the ing point and on the angle, the effects are reduced when data. As we describe in the next section, variables we take the average of the multiple images at the retriev- constructed from the Google Street View images will ing point. We collect the most recent images uploaded to be linked to the property transaction data. Google Maps as of August 2021. For details, see: http://www.reins.or.jp/ (in Japanese; accessed 22 December 2021). For details of the market data created from the REINS database, see: http://www.reins.or.jp/library/2019.html (accessed 16 February 2022; in Japanese). Further, for condominiums in the Tokyo metropolitan area, Shimizu, Nishimura, and Watanabe (2016) compare the nature of the REINS data and other related data sources: properties listed in a web portal provided by one of the largest private vendors of residential information, and transacted properties (a part of registered properties) whose transaction prices are collected by the MLIT. It is shown that the regression coefficients in the hedonic regression analyses are at least consistent even though there may be differences in property characteristics. In this low-rise residential area, we truncate 118 samples for inaccurate addresses, 26 samples for having a property age older than 50 years, and 27 samples for other missing information. 4 M. SUZUKI ET AL. (i) Collect Google Street View images Section 4 Semantic segmentation (ii) Create urban landscape index (iii) Hedonic Economic value of regression urban landscape analysis Section 5 Basic characteristics Section 6 collected from property Property price transaction data and land use data Figure 1. Methodological flow. Second, for the collected images, semantic segmen- characteristics of building as explanatory variables in tation is conducted, a deep learning algorithm that the hedonic regression analysis so that the building associates a label or category to every pixel in an price level is sufficiently controlled. image. We employ the recent semantic segmentation We estimate the correlation between property price approach proposed by Tao, Sapra, and Catanzaro and urban landscape in the following regression: (2020), which as of August 2021 has achieved state-of- X X the-art results in two commonly used open data sets: ln P ¼ αþ β V þ γ X þ D þ T þ e (1) it ki j t it s si k Mapillary and Cityscapes. Cityscapes is a large data set that labels semantic classes across 5,000 urban street where ln P is the log of the transaction price for unit it images in 50 German cities (Cordts et al. 2016). i in period t. α is the constant. V is the urban landscape si Mapillary Vistas is another large data set, containing index for landscape component s, and β is the corre- 25,000 high-resolution images from across the world, sponding coefficient. Specifically, the index represents with annotated semantic object categories (Neuhold the proportion of landscape component s in the et al. 2017). The semantic segmentation model is first images, and is created through semantic segmentation pre-trained on the larger Mapillary, and then trained of the Google Street View images. More precisely, 1% on Cityscapes. Using the pre-trained model, every increase in the proportion of component s leads to a β Street View image is partitioned into multiple- % increase in the transaction price. X is the control ki segment categories: road, sidewalk, building, wall, variable k shown in Table 1(a), and γ is the corre- fence, pole, traffic light, traffic sign, vegetation, terrain, sponding coefficient. D captures the district fixed sky, person, rider, car, truck, bus, train, motorcycle and effect, that is, it takes 1 for properties in district j, and bicycle. Given that some categories are almost blank, 0 otherwise. T captures the quarterly fixed effect for we create our urban landscape index using major the time of the transaction, that is, it takes 1 for categories (which are shown in Table 1(b)). a property transaction in period t, and 0 otherwise. e it Third, we conduct hedonic regression analysis to is the error term. decompose the property prices (that is, the sum of When we exclude the district fixed effect, D , the the building price and land price) into multiple com- urban landscape index V serves as a proxy for the si ponents. There are many factors forming the property district-level landscape. For instance, if the entire prices, including building characteristics (e.g., property residential area (i.e., the district) has been uniformly age and floor area), land characteristics (e.g., land area developed as a “greenery residential area,” the and walking time to nearest station), and land use greenery is likely to be the district-level landscape. around the properties. Beyond these basic character- On the other hand, when we include the district istics, this paper further extends these methods by fixed effect, D , the urban landscape index V serves j si capturing urban landscape factors using Google as a proxy for the street-level landscape. For Street View images. It is expected that the urban land- instance, even within each “greenery residential scape index would impact the land price, and the area” (i.e., within each district), openness can vary building price would be basically determined by the across streets. characteristics of building itself. Thus, we include the The economic value of urban landscape for existing properties can be captured through a property transaction price, which is the sum of building and land prices. The transaction price in our data more formally captures a market valuation than publicly available appraisals of land prices in Japan. Employing land transaction data also has difficulty in acquiring Google Street View images containing buildings that will be constructed on the land after the transactions; this may completely alter the landscape of newly developing residential areas. JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 5 Figure 2. Examples of original images and semantic segmentation.The right-hand squares show the pixel ratios of each landscape component within the images. 4. Semantic segmentation and hypotheses category captures “openness” as the sky component development (accounting for 28.3%). The dark gray category captures “visual enclosure” as the building component (account- 4.1. Example of semantic segmentation ing for 13.8%). In some cases, buildings are also captured To ensure that each category from the semantic segmen- by a wall (light blue category; accounting for 9.4%) or tation captures the intended landscape component, a fence (light pink category; accounting for 4.5%); thus, Figure 2 shows examples of original images and their we will also create an alternative indicator of the visual semantic segmentation outcomes, as well as the pixel enclosure as the sum of these three categories. The light ratios of each landscape component within the images. gray category captures “power pole” as the pole compo- Although the pre-trained model is trained on street nent (accounting for 1.5%). The purple category captures images from all around the world, we confirm that the the road component (accounting for 29.3%), and this is semantic segmentation is conducted properly for the set as a reference category in the following hedonic Japanese Street View images. regression analyses. Panel (a) shows a typical residential area. The dark Panel (b) shows a case capturing “road shoulder,” green category captures vertical “greenery” as the vege- which is an open space in the shoulder of a road that is tation component (accounting for 11.6% of the pixels in often used as private parking space or as paved vacant the entire image), including trees and plants. The blue lots. This is captured by the dark pink category as the It is true that in some cases, sidewalks are really sidewalks for wider streets. However, as in Figure 2(a), in typical residential districts with narrow streets, it is not common to see sidewalks with curbstones. Note that we controlled for the front road width in the hedonic regression analyses to partially separate this effect. 6 M. SUZUKI ET AL. Figure 3. 12 views (every 30 degrees) for the typical residential areas shown in Figure 2(a). sidewalk component (accounting for 6.1%). In some shoulder as the sum of the two components: sidewalk cases, cars are parked in the parking space, and so we and car (dark blue category; accounting for 0.6%). Panel will also create an alternative indicator of the road (c) shows a case capturing “farmland” within a residential JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 7 Table 2. Percentage of each landscape component for the 12 views of the typical residential areas shown in Figure 2(a). Building + Number Vegetation Sky Building Wall + Fence Pole Sidewalk Sidewalk + Car Terrain 1 11.64 28.21 13.79 27.72 1.52 1.60 1.60 0.02 2 8.21 28.97 18.60 22.39 0.11 2.03 13.57 0.61 3 8.68 22.79 37.28 46.83 0.00 3.60 11.02 0.13 4 3.91 14.91 32.74 56.26 0.00 0.12 11.02 9.30 5 6.93 12.58 63.58 69.44 0.00 1.19 1.19 0.07 6 2.80 16.42 39.39 45.55 0.24 0.00 0.00 0.00 7 2.89 26.12 31.07 39.09 0.16 1.15 1.15 0.00 8 3.51 26.35 17.30 38.31 0.33 1.31 1.31 0.00 9 0.75 16.91 33.18 58.58 0.00 2.29 2.29 0.00 10 2.51 7.66 45.78 74.20 0.00 0.16 0.16 0.00 11 9.47 12.92 33.16 55.71 1.55 0.55 0.55 0.00 12 8.78 26.48 20.04 33.53 1.17 0.82 0.82 0.04 Mean 5.84 20.03 32.16 47.30 0.42 1.24 3.72 0.85 Min 0.75 7.66 13.79 22.39 0.00 0.00 0.00 0.00 Max 11.64 28.97 63.58 74.20 1.55 3.60 13.57 9.30 In the hedonic regression analysis, aggregated proportions of the 12 views (shown in bold) are employed. area, which is the exposure of the ground. This is captured images, the summary statistics to be measured, and by the light green category as the terrain component a hypothesis on the direction of the correlation with (accounting for 16.9%). Note that the vertical greenery the price level. in the farmland is captured by the vegetation component. Our hypotheses on the major scene perceptions are For the typical residential area (panel (a) of Figure 2), basically consistent with previous studies on the visual Figure 3 shows 12 Street View images (i.e., every 30 impression of urban landscape in the field of architec- degrees). Table 2 shows the proportion of each land- ture (e.g., Hirate and Yasuoka 1986; Ishikawa et al. scape component for these 12 views in percentage 1995; Nishio and Ito 2015, 2020; Takei and Fukushima terms; the mean, minimum and maximum values for 1983) and on the relationship between property price these 12 views are also calculated. Although the and landscape elements using recent image recogni- semantic segmentation is conducted properly overall, tion technologies (e.g., Chen et al. 2020; Fu et al. 2019; we see that there are some misclassifications of side- Yang et al. 2021; Ye et al. 2019a; Zhang and Dong walks, walls and fences by buildings (e.g., image num- 2018). “Greenery” is captured through the mean or ber 5). It is also true that the proportion of buildings maximum proportion of the vegetation component. increases when the building is in the foreground (e.g., It is expected to have a positive relationship with image number 10). These problems can be reduced by transaction price through the provision of taking the mean value of the proportion for each a comfortable environment with abundant greenery component from the 12 views, justified by the fact and plantings. “Openness” is captured through the that the landscape is an average impression from mean or maximum proportion of the sky component. a 360-degree view at the location. The minimum or It is expected to have a positive relationship with maximum values of the proportion from the 12 views transaction price by offering sufficient sunlight to the may also be useful in capturing the strongest impres- community. “Visual enclosure” is captured through the sion in the location. For instance, if the minimum mean or minimum proportion of the building compo- proportion of the building is large, the impression nent, as well as the mean proportion of the sum of the conveyed is that the visual enclosure is high at the building, wall and fence components. Sufficient level location; by contrast, the maximum proportion of the of visual enclosure represents that the residential area building may not be useful, because the view in front is matured, and thus, it is expected to have a positive of the building is always included in the 12 views. relationship with transaction price. Similarly, if the maximum proportion of vegetation is With regard to urban landscapes specific to Japan, large, the impression conveyed is that the location has “power pole” is captured through the mean or max- plenty of greenery. imum proportion of the pole component. It is expected to have a nonpositive relationship with transaction price, as it is a messy landscape characterized by elec- 4.2. Hypotheses trical wires in the sky, but it is so common in the Japanese landscape that it may not reduce the transac- Based on the examples discussed above, Table 3 pre- tion price. “Road shoulder” is captured through the sents the hypotheses for the hedonic regression analy- mean or maximum proportion of pixels for the side- sis. For each landscape element addressed here, we walk component, as well as the mean proportion of the present the baseline and additional components of the 8 M. SUZUKI ET AL. Table 3. Hypotheses regarding correlations between the urban landscape index and property price. Landscape Urban landscape Summary Expected correlation with property price element index statistics Greenery Vegetation Mean (Max) + (Comfortable environment with abundant greenery and plantings) Openness Sky Mean (Max) + (Sufficient sunlight in the community) Visual enclosure Building (+ Wall + Mean (Min) + (Matured residential area with a certain level of enclosure by buildings) Fence) Power pole Pole Mean (Max) - (Messy Landscape characterized by electrical wires) or 0 (Common urban landscape in Japan) Road shoulder Sidewalk (+ Car) Mean (Max) ? (Loss of continuity of buildings or visual enclosures; Enhance openness; Common urban landscape in Japan) Farmland Terrain Mean (Max) ? (Loss of continuity of buildings or visual enclosures; Enhance openness; Common urban landscape in Japan) sum of the sidewalk and car components. Its relation- Information (Ministry of Land, Infrastructure, Transport ship with transaction price is unclear; although it and Tourism). The categories include: high-rise build- reduces the continuity of buildings or the visual enclo- ing; dense low-rise building (that is, a low-rise building sure, it enhances the openness. This is so common in is densely concentrated in the area); low-rise building the Japanese landscape that it may not reduce the (the most common surrounding land use type, as we transaction price. “Farmland” is captured through the employ low-rise residential area samples); vacant land; mean or maximum proportion of the terrain compo- park and green space (well-maintained park or green nent. Its relationship with transaction price is also space); and farmland. After controlling for the initial unclear; the same mechanisms as for road shoulder three categories, “openness” and “visual enclosure” apply. capture the landscape of the location. Further, the last three categories control part of the quality of the location. The existence of vacant land partly controls 5. Variables and summary statistics for the sidewalk component, the existence of a park and/or green space partly controls for the vegetation We employ low-rise residential area samples for the component (and even the pole component, as the main analysis. Table 1 shows summary statistics overall quality of the location), and the existence of (means and standard deviations) for (a) property char- farmland around the location partly controls for the acteristics and (b) urban landscape index, the latter of farmland component. which is created through sematic segmentation of the In panel (b), the means and standard deviations of Google Street View images. In panel (a), the building the urban landscape index created through sematic characteristics include: a newly built dummy (taking 1 segmentation of the Google Street View images are for a newly built house and 0 otherwise); property age; shown for the mean and minimum and maximum floor area; and a non-timbered dummy (most Japanese proportions. On average, vegetation, sky and buildings detached houses are timbered). The land characteris- account for 10.8%, 23.7%, and 20.0%, respectively, of tics include: land area; walking time to the nearest the mean proportion. On the other hand, the corre- station; a dummy variable indicating a need to take sponding values for pole, sidewalk and terrain are less a bus to the nearest station; a dummy variable for a lot than 5%. As expected, the average of the minimum with parking space; distance to a junior high school; (maximum) proportion is smaller (larger) than the and the regulation (upper limit) of the floor-area ratio mean proportion of the corresponding landscape com- (FAR). After controlling for the existence of parking ponent; these variables will be employed in the robust- space in the lot, the road shoulder captures the land- ness checks. scape of the location. Front road width is included as a categorical variable. In Japan, if the front road width is less than 4 m (and the length of the lots connected 6. Empirical results to the road is less than 2 m), rebuilding the existing 6.1. Main results house is forbidden. The “less than 4 m category” is expected to be negatively correlated with price. The Table 4 shows the correlation between property price unknown status is included so as not to delete missing and the urban landscape index based on equation (1). observation samples, which exist in non-negligible As in the following discussion, the results confirm that numbers. the quality of surrounding landscapes has a meaning- Land use within a 500 m radius around the property ful relationship with real estate price formation follow- (that is, the ratio of each land use type) is measured ing the hypotheses in Table 2. Columns (1) and (2) using a geographic information system (GIS) based on control only the very basic variables, and thus the 50 m-mesh land use data, provided as National Land urban landscape index simply serves as a locational JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 9 Table 4. Correlations between urban landscape index and property price. Measuring: Locational proxy District-level landscape Street-level landscape (1) (2) (3) (4) (5) (6) Urban landscape index Vegetation (Mean) 0.6469*** 0.5866*** 0.3454* (0.1924) (0.1892) (0.2018) Sky (Mean) 0.9836*** 0.9684*** 1.0541*** (0.2866) (0.2957) (0.2905) Building (Mean) 0.2578 0.2528 0.5360*** (0.1743) (0.1750) (0.1980) Pole (Mean) −3.6017* −3.4615* −1.2807 (1.9449) (1.7929) (1.8650) Sidewalk (Mean) 1.3820*** 0.7494 0.1757 (0.5338) (0.5243) (0.5769) Terrain (Mean) −0.4643 −0.2295 −0.0182 (0.3085) (0.2788) (0.2567) Building characteristics Newly built 0.0114 0.0111 0.0644* 0.0578 0.1072*** 0.1038*** (0.0337) (0.0351) (0.0350) (0.0360) (0.0346) (0.0345) Age [years] −0.0180*** −0.0180*** −0.0169*** −0.0170*** −0.0169*** −0.0167*** (0.0013) (0.0013) (0.0013) (0.0013) (0.0013) (0.0012) Floor area [m ] 0.0064*** 0.0063*** 0.0051*** 0.0050*** 0.0041*** 0.0041*** (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) Non-timbered 0.0384 0.0395 0.0415 0.0437 (0.0418) (0.0413) (0.0413) (0.0408) Land characteristics Land area [m ] 0.0019*** 0.0017*** 0.0019*** 0.0018*** 0.0017*** 0.0016*** (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) Walking time to station [min] −0.0130*** −0.0139*** −0.0157*** −0.0162*** −0.0098*** −0.0095*** (0.0028) (0.0027) (0.0030) (0.0030) (0.0037) (0.0035) Bus to station −0.5044*** −0.5210*** −0.4926*** −0.5077*** −0.2076*** −0.2138*** (0.0396) (0.0389) (0.0449) (0.0459) (0.0792) (0.0751) With parking space 0.0474* 0.0462* 0.0462* 0.0491* (0.0280) (0.0280) (0.0254) (0.0250) Distance to junior high school [km] −0.1268*** −0.1238*** −0.0099 −0.0238 (0.0277) (0.0278) (0.0558) (0.0544) Floor-area ratio (FAR) [%] −0.0010 −0.0011 0.0018* 0.0015 (0.0008) (0.0008) (0.0010) (0.0009) Front road width Less than 4 m −0.1234** −0.1421*** −0.1656*** −0.1651*** (0.0515) (0.0549) (0.0546) (0.0581) 4–5 m −0.1595*** −0.1614*** −0.1363*** −0.1430*** (0.0350) (0.0355) (0.0340) (0.0345) 5–6 m (reference) (reference) (reference) (reference) 6–7 m 0.0433 0.0344 −0.0213 −0.0265 (0.0280) (0.0277) (0.0291) (0.0285) 7–10 m 0.0352 0.0099 0.0082 −0.0062 (0.0481) (0.0469) (0.0492) (0.0475) 10 m or more 0.0217 −0.0091 −0.0276 −0.0394 (0.0487) (0.0482) (0.0488) (0.0456) Unknown −0.0357 −0.0370 −0.0606* −0.0579* (0.0343) (0.0340) (0.0337) (0.0328) Land use within 500 m radius [ratio] High-rise building 0.3648 0.2703 0.3789 0.1972 (0.2777) (0.2817) (0.5165) (0.5177) Dense low-rise building 0.4992*** 0.5111*** 0.5392* 0.2361 (0.1279) (0.1322) (0.2855) (0.2908) Low-rise building 0.1109 0.0854 0.1773 0.1054 (0.0789) (0.0844) (0.1200) (0.1198) Vacant land 0.9247*** 0.7879*** 0.1512 −0.0726 (0.3043) (0.2929) (0.5705) (0.5676) Park and green 2.9514*** 2.3671*** −1.0923 −1.3626 (0.7220) (0.7411) (1.0142) (1.0067) (Continued) 10 M. SUZUKI ET AL. Table 4. (Continued). Farmland −0.2368 −0.3831** 0.0651 −0.0975 (0.1917) (0.1836) (0.2975) (0.2872) Constant 7.3556*** 7.0257*** 7.5285*** 7.2457*** 7.2960*** 6.9458*** (0.0799) (0.1387) (0.1272) (0.1528) (0.1989) (0.2100) Time fixed effect (quarterly) Yes Yes Yes Yes Yes Yes District fixed effect Yes Yes Number of observations 800 800 800 800 800 800 R 0.635 0.652 0.695 0.705 0.794 0.801 Adjusted R 0.624 0.638 0.678 0.687 0.751 0.757 The dependent variable is the log of the transaction price. White-corrected robust standard errors are in parentheses. Significance level: *** p < 0.01, ** p < 0.05, * p < 0.1. proxy. Columns (3) and (4) add control variables, but The coefficient of sky does not vary much even when still do not include the district fixed effect, hence col- we control much fewer control variables in columns (2) umn (4) measures the district-level landscape. Columns and (4), demonstrating that openness is heterogeneous (5) and (6) now include the district fixed effect, and across districts and across the streets within a district. The thus column (6) measures the street-level landscape. coefficient of building diminishes and loses statistical Comparing R between columns (1) and (2) (0.635 and significance once we control for much fewer control 0.652), between columns (3) and (4) (0.695 and 0.705), variables in columns (2) and (4). This shows that visual and between columns (5) and (6) (0.794 and 0.801), enclosure is heterogeneous across the streets within reveals that the landscape index partly improves the a district, but does not have a clear trend across districts. explanatory power of the model. However, the degree Second, the presence of a power pole is negatively of improvement is much larger when we include addi- correlated with property price at district level. In col- tional control variables. umn (4) without controlling for the district fixed effect, the coefficient of pole is negative with statistical sig- First, greenery, openness, and visual enclosure are nificance; a 1% increase in the proportion of pole leads positively correlated with property price at street level. to a 3.46% decrease in transaction price. However, In column (6) including the district fixed effect, the once we control for the district fixed effect in column coefficients of the mean proportions of vegetation, (6), the size of the coefficient diminishes and statistical sky, and building are positive with statistical signifi - significance is lost. This shows that power poles serve cance. A 1% increase in the proportion of vegetation, as a district-level landscape. The negative valuation at sky, and building, respectively, lead to 0.34%, 1.05%, district-level means that it is recognized as a messy and 0.53% increases in transaction price. landscape characterized by electrical wires in the sky, The coefficient of vegetation is larger when we use and that the residential area achieving high valuation much fewer control variables, showing that greenery is likely to have less power poles in the whole district. partly serves as a locational proxy. Specifically, the park Let us roughly discuss the size of the coefficient of and green space ratio within 500 m is positive and statis- the pole component at district level, 3.46. Three power tically significant in column (4); this and the other control poles are recognized in the semantic segmentation in variables explain the decrease in the size of the coefficient the typical residential area in Figure 2(a), and this is for the mean proportion of the vegetation component summarized as a mean proportion of 0.42% in Table 1. from column (2) to column (4). However, greenery still Thus, one power pole lowers property prices by captures the street-level landscape; that is, even within (3.46 × 0.42)/3 = 0.48 [%], compared to a pole-free a district, the level of greenery is different on each street. landscape. Given that each power pole lowers non- Specifically, the coefficient of vegetation is still positive negligible numbers of surrounding houses, the over- with 10% statistical significance in column (6), which head-to-underground conversion of electricity distri- controls for the district fixed effect. This is in contrast to bution networks is likely to improve property values the fact that the park and green space ratio within 500 m widely in the residential area. now becomes negative without statistical significance; Third, the presence of either a road shoulder or the effect of the existence of a well-maintained park or farmland does not exhibit negative correlations with green space, which is likely to capture the characteristics property price even at district level. In both columns (4) of the district, is absorbed by the district fixed effect. and (6), the coefficients of sidewalk are positive but without statistical significance. The coefficient becomes larger as fewer control variables are included; The variance inflation factor (VIF) for the urban landscape index in Table 4 (columns (1), (4), and (6)) is less than 5, not a level that would cause much of a multicollinearity problem. At variable level, the correlation between the “vegetation” and the “park and green space ratio” is weak (correlation coefficient of 0.03), implying that the greenery differs on each street, as a landscape. JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 11 Table 5. Subsample analysis of land use type. Sample: Entire area Other residential area Loosely regulated area District-level Street-level District-level Street-level District-level Street-level Measuring: landscape landscape landscape landscape landscape landscape (1) (2) (3) (4) (5) (6) Urban landscape index Vegetation 0.4505*** 0.1060 0.2889 0.1646 0.1436 −0.0573 (Mean) (0.1582) (0.1695) (0.3808) (0.4132) (0.5192) (0.6095) Sky (Mean) 0.6148*** 0.7087*** 0.3485 0.4885 0.0111 0.4896 (0.2009) (0.2046) (0.4133) (0.5209) (0.5656) (0.8693) Building (Mean) 0.1893 0.3214** −0.0310 0.5410 0.3167 0.3383 (0.1392) (0.1497) (0.3956) (0.4030) (0.3864) (0.5824) Pole (Mean) −1.5049 0.0444 1.6424 7.8979 2.9837 3.9114 (1.3403) (1.5129) (3.5051) (5.9158) (2.6071) (4.2802) Sidewalk (Mean) 0.3459 −0.0141 −0.1771 1.6259 −0.3943 −1.2080 (0.4646) (0.4866) (0.9767) (1.2077) (1.2012) (1.8886) Terrain (Mean) −0.4776* −0.1830 −0.8464** 0.1083 −0.6307 −0.6099 (0.2498) (0.2389) (0.4156) (0.4695) (0.5149) (0.9942) Building Yes Yes Yes Yes Yes Yes characteristics, Land characteristics, Land use within 500 m radius, Zoning category, Time fixed effect (quarterly) and a Constant District fixed Yes Yes Yes effect Number of 1307 1307 297 297 210 210 observations R 0.699 0.795 0.742 0.894 0.756 0.875 Adjusted R 0.686 0.757 0.693 0.806 0.684 0.748 The dependent variable is the log of the transaction price. White-corrected robust standard errors are in parentheses. Significance level: *** p < 0.01, ** p < 0.05, * p < 0.1. thus, the sidewalk component partly serves as a proxy 6.2. Subsample on land use type of locational quality. In both columns (4) and (6), the Table 5 shows the entire area sample (columns (1) coefficients of terrain are negative but without statis- and (2)) and subsamples of other residential areas tical significance. The coefficient becomes larger in (excluding residential areas with a category absolute terms as fewer control variables are included; I exclusively low-rise residential zone; columns (3) thus, the terrain component partly serves as a proxy of and (4)) and loosely regulated areas (quasi- locational quality. The insignificant or neutral valua- industrial zones, neighborhood commercial zones, tions of road shoulder and farmland mean that these and commercial zones; columns (5) and (6)). types of landscape have two sides: They reduce the Columns (1), (3) and (5) capture the district-level continuity of buildings or the visual enclosure, while landscape without controlling for the district fixed they enhance the openness. It is also true that the effect (corresponding to column (4) of Table 4), landscapes are so common in Japanese residential while columns (2), (4) and (6) capture the street- areas that they may not be reflected in the valuations. level landscape by controlling for the district fixed With regard to the control variables, a narrow front effect (corresponding to column (6) of Table 4). road width (less than 5 m) reduces property price with The nonexclusive zoning system used in Japan statistical significance. Most of the land use ratio within restricts specific types of land use in a location, 500 m radius of a transacted property exhibits statis- and does not intend to realize a single (pure) tical significance when we do not control for the dis- land use therein. “Low-rise residential area” is the trict fixed effect (column (4)). However, it loses strictest type of zoning, literally allowing only low- statistical significance once we control for the district rise residential houses to be situated in an area. In fixed effect (column (6)). This confirms that the district an “other residential area,” low-rise and high-rise fixed effect control is effective in accounting for the residential houses (and even other facilities) can heterogeneity in land use across districts. 12 M. SUZUKI ET AL. Table 6. Robustness checks. (a) Greenery, openness and visual enclosure Measuring: District-level Street-level District-level Street-level District-level Street-level District-level Street-level landscape landscape landscape landscape landscape landscape landscape landscape (1) (2) (3) (4) (5) (6) (7) (8) Urban landscape index Alternative indicator Vegetation (Max) 0.2312*** 0.1781** (0.0842) (0.0886) Sky (Max) 0.6096** 0.7201*** (0.2603) (0.2541) Building (Min) 0.3150 0.6047** (0.2181) (0.2417) Building + Wall + 0.1377 −0.1104 Fence (Mean) (0.2400) (0.2400) Baseline indicator Vegetation (Mean) 0.3657** 0.1219 0.5445*** 0.2394 0.5503** −0.0936 (0.1728) (0.1735) (0.1727) (0.1798) (0.2614) (0.2682) Sky (Mean) 0.8598*** 1.0606*** 0.8706*** 0.8275*** 0.9398** 0.4881 (0.2905) (0.2782) (0.2695) (0.2515) (0.3971) (0.3722) Building (Mean) 0.1612 0.5294*** 0.0390 0.3121* (0.1663) (0.1850) (0.1556) (0.1615) Pole (Mean) −3.6499** −1.2459 −3.1002* −0.9070 −3.7286** −2.0014 −3.4045* −1.7044 (1.7957) (1.8669) (1.7543) (1.8524) (1.8196) (1.9027) (1.7803) (1.8727) Sidewalk (Mean) 0.7678 0.1951 0.9873* 0.4614 0.7256 0.1422 0.9189* 0.1610 (0.5217) (0.5753) (0.5092) (0.5609) (0.5280) (0.5842) (0.5239) (0.5981) Terrain (Mean) −0.2364 −0.0334 −0.2364 −0.0302 −0.1958 0.0276 −0.1925 −0.1518 (0.2756) (0.2545) (0.2780) (0.2637) (0.2894) (0.2712) (0.2981) (0.2826) Building Yes Yes Yes Yes Yes Yes Yes Yes characteristics, Land characteristics, Land use within 500 m radius, Time fixed effect (quarterly) and a Constant District fixed effect Yes Yes Yes Yes Number of 800 800 800 800 800 800 800 800 observations R 0.705 0.801 0.702 0.798 0.705 0.800 0.704 0.798 Adjusted R 0.686 0.758 0.683 0.754 0.687 0.756 0.686 0.753 (b) Power pole, road shoulder, and farmland Measuring: District-level Street-level District-level Street-level District-level Street-level District-level Street-level landscape landscape landscape landscape landscape landscape landscape landscape (1) (2) (3) (4) (5) (6) (7) (8) Urban landscape index Alternative indicator Pole (Max) −0.7583* −0.3847 (0.4217) (0.4510) Sidewalk (Max) 0.1081 0.0442 (0.1977) (0.2030) Sidewalk + Car 0.5241 0.2099 (Mean) (0.3816) (0.3497) Terrain (Max) −0.0931 0.0017 (0.0954) (0.0906) Baseline indicator Vegetation (Mean) 0.5967*** 0.3453* 0.5783*** 0.3434* 0.5965*** 0.3532* 0.6010*** 0.3439* (0.1890) (0.2019) (0.1896) (0.2017) (0.1893) (0.2009) (0.1911) (0.2038) Sky (Mean) 0.9552*** 1.0531*** 0.9987*** 1.0584*** 0.9716*** 1.0473*** 0.9759*** 1.0524*** (0.2928) (0.2890) (0.2903) (0.2874) (0.2941) (0.2852) (0.2945) (0.2903) Building (Mean) 0.2565 0.5360*** 0.2601 0.5369*** 0.2413 0.5285*** 0.2530 0.5372*** (0.1747) (0.1980) (0.1751) (0.1987) (0.1757) (0.1992) (0.1746) (0.1981) Pole (Mean) −3.6663** −1.2998 −3.5458** −1.2658 −3.3911* −1.2803 (1.7483) (1.8453) (1.7694) (1.8492) (1.7998) (1.8701) Sidewalk (Mean) 0.7509 0.1687 0.7269 0.1860 (0.5234) (0.5776) (0.5198) (0.5726) Terrain (Mean) −0.2359 −0.0189 −0.3090 −0.0290 −0.2540 −0.0110 (0.2783) (0.2567) (0.2780) (0.2575) (0.2784) (0.2635) (Continued) JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 13 Table 6. (Continued). Building Yes Yes Yes Yes Yes Yes Yes Yes characteristics, Land characteristics, Land use within 500 m radius, Time fixed effect (quarterly) and a Constant District fixed effect Yes Yes Yes Yes Number of 800 800 800 800 800 800 800 800 observations R 0.705 0.801 0.704 0.801 0.705 0.801 0.705 0.801 Adjusted R 0.686 0.757 0.686 0.757 0.686 0.757 0.687 0.757 The dependent variable is the log of the transaction price. White-corrected robust standard errors are in parentheses. Significance level: *** p < 0.01, ** p < 0.05, * p < 0.1. coexist. In a “loosely regulated area,” residential landscape without controlling for the district fixed effect houses and commercial/industrial uses can coexist. (corresponding to column (4) of Table 4), while columns In this context, the economic value of landscape is (2), (4), (6) and (8) capture the street-level landscape by more likely to be observed in pure low-rise resi- controlling for the district fixed effect (corresponding to dential areas, and may be lost in more loosely column (6) of Table 4). We employ alternative indicators regulated areas where multiple types of land use one by one, and the other baseline indicators are coexist. included as control variables. For instance, when we employ the maximum proportion of vegetation (instead In columns (1) and (2), the entire area sample basi- of the mean proportion of vegetation), we include the cally confirms the same results as those yielded from mean proportions of sky, building, pole, sidewalk, and our main analysis using low-rise residential areas terrain. The results are basically consistent with those (Table 4), except for the loss of statistical significance presented in Table 4. for the vegetation component at street level. This may be due to the fact that the other residential area and Panel (a) presents the results regarding greenery, loosely regulated area subsamples do not show openness, and visual enclosure. In columns (1) and (2), a correlation between vegetation and property price we employ the maximum proportion of the vegetation (columns (3)–(6)). In other words, street-level greenery component, whose coefficient is positive with statistical is evaluated only within low-rise residential areas. Note significance at street level. Thus, we confirm the robust- that terrain exhibits a negative correlation with prop- ness on the fact that greenery is positively correlated erty price at district level. This may capture the strong with property price at street level. Columns (3) and (4) heterogeneity across districts when we employ the employ the maximum proportion of the sky compo- entire sample. That is, areas with farmland may have nent, whose coefficient is positive with statistical signifi - been facing housing development only recently, and cance at street level. Thus, we confirm the robustness of thus still retain low land prices. the fact that openness is positively correlated with prop- In the other residential areas (columns (3) and (4)) erty price at street level. Columns (5) and (6) employ the and in loosely regulated areas (columns (5) and (6)), we minimum proportion of the building component, do not observe a correlation between the urban land- whose coefficient is positive with statistical significance scape index and transaction price with statistical sig- at street level. Thus, we confirm the robustness of the nificance (except that terrain exhibits a negative fact that visual enclosure is positively correlated with correlation with property price at district level in property price at street level. Note, however, that the other residential areas, as in the entire area). This mean proportion of the sum of the three components implies that the urban landscape is basically reflected (building, wall, and fence) does not exhibit a correlation in transaction price only in low-rise residential areas, in with transaction price in columns (7) and (8). This implies which the landscape is valued by market participants. that the building component – but not the wall or fence In the remaining areas, other factors like transportation components – is correlated with property price. and shopping convenience matter, whereas the urban Panel (b) displays the results for power pole, road landscape does not influence the transaction price. shoulder, and farmland. In columns (1) and (2), we employ the maximum proportion of the pole compo- nent, whose coefficient is negative with statistical signifi - 6.3. Robustness checks cance only at district level. Thus, we confirm the robustness of the fact that the presence of a power Table 6 conducts robustness checks by using alternative pole is negatively correlated with property price at district indicators of the urban landscape index. The samples level. In columns (3) and (4) (columns (5) and (6)), we are properties in low-rise residential areas, as in Table 4. employ the maximum proportion of the sidewalk Columns (1), (3), (5) and (7) capture the district-level 14 M. SUZUKI ET AL. component (mean proportion of the sum of the sidewalk Our results also show that the semantic segmenta- and the car components). In neither regression are the tion approach, with high segmentation accuracy and coefficients statistically significant, confirming the robust- label diversity, enables researchers to quantify micro- ness of the fact that the presence of a road shoulder does level landscapes that have previously proven difficult not exhibit a negative correlation with property price at to obtain at large scale, and that this methodology is either district or street level. In columns (7) and (8), we also useful in analyzing urban landscapes specific to employ the maximum proportion of the terrain compo- Japan. The Hachioji city investigated in this paper is nent. The coefficients are not statistically significant, con- a typical low-rise residential area in the suburbs, and firming the robustness of the fact that the presence of the conclusions of this study are likely to be a common farmland does not exhibit a negative correlation with feature of at least the suburban areas of major metro- property price at either district or street level. politan areas. Although the methodology is easily applicable to other types of real estate in different regions, subsample 7. Conclusion analysis of land use type indicates that any correlation between landscape and property price may be hetero- While visual impression of urban landscape has geneous (e.g., central area and other suburban areas), been investigated in detail in the field of architec- and thus a functional form of hedonic regression should ture, an economic consideration of the urban land- be tailored to each local context. To fulfill this aim, auto- scape across space is also an important aspect to mating functional form to identify the elements that give an incentive to build and maintain a good formulate property prices (e.g., He, Páez, and Liu 2017; residential environment. Employing a novel seman- Law, Paige, and Russell 2019; Law et al. 2020) is worthy of tic segmentation approach using Google Street investigation. View images, we have investigated the relationship between urban landscapes and property prices in low-rise residential areas in a suburban city of Acknowledgments Tokyo, Japan. Such visual images have been used to represent the landscape of both surrounding We would like to thank two anonymous referees, Yasushi districts in general and the street-level landscape; Asami, Masayoshi Hayashi, Sachio Muto, Toshihiko more specifically, the latter has been derived after Yamasaki, Noriyuki Yanagawa, participants in the CREI work- controlling for the district fixed effect. We first shop at the University of Tokyo, and members of Japan Real Estate Institute for their insightful comments. We also thank showed that greenery, openness and visual enclo- the Real Estate Information Network System (REINS) and sure are positively correlated with property price at Joint Research Program No. 1075 at the Center for Spatial street level. We then investigated the value of com- Information Science, The University of Tokyo, for providing mon Japanese urban landscapes: (i) the presence of the data. a power pole is negatively correlated with property price at district level; and (ii) the presence of a road shoulder and farmland, either of which may disrupt Disclosure statement the continuity of a residential area, does not exhibit No potential conflict of interest was reported by the negative correlations with property price at either author(s). district or street level. Thus, the quality of surround- ing landscapes, as captured through Street View images, has a meaningful relationship with real Funding estate price formation. The authors gratefully acknowledge the support received Our results suggest that basic landscape elements, from JSPS KAKENHI Grant Numbers 20K14896, 20H00082, such as greenery, openness and visual enclosure, are and 17H00988, and a Google Cloud Platform Credit Coupon important in maintaining the value of residential areas; at The University of Tokyo. thus, protection of the residential environment through residential area planning and/or building agreement is meaningful. Furthermore, the value of residential areas Notes on contributors can be increased by promoting the undergrounding of Masatomo Suzuki is a specially appointed associate profes- power poles. Although road shoulder does not lead to sor at Center for the Promotion of Social Data Science a uniform decline in economic value, the value of the Education and Research, Hitotsubashi University. His landscape is potentially improved if it is well designed, research interests include housing, real estate, city planning, such as a parking lot covered with greenery. Since farm- and urban economics. land also does not lead to a uniform decline in economic Junichiro Mori is an associate professor at Graduate School of value, preservation of urban farmland within residential Information Science and Engineering, The University of area, in other words, coexistence of the two types of Tokyo. His research interests include artificial intelligence, land use, may be justified. massive data analysis, and social network analysis. JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 15 Takashi Nicholas Maeda is an associate professor at the Jim, C. Y., and W. Y. Chen. 2009. “Value of Scenic Views: School of System Design and Technology of Tokyo Denki Hedonic Assessment of Private Housing in Hong Kong.” University. His research interests include causal discovery Landscape and Urban Planning 91 (4): 226–234. and urban computing. doi:10.1016/j.landurbplan.2009.01.009. Johnson, E. B., A. Tidwell, and S. V. Villupuram. 2020. “Valuing Jun Ikeda is a former graduate student at Graduate School of Curb Appeal.” The Journal of Real Estate Finance and Information Science and Technology, The University of Economics 60 (1): 111–133. doi:10.1007/s11146-019-09713-z. Tokyo. Koura, H., and K. Kamino. 1995. “Study on Structure of Integrated Setting in Townscape of Urban Environment.” Journal of the City Planning Institute of Japan 30: 223–228. ORCID [in Japanese]. doi:10.11361/journalcpij.30.223. Koura, H., and K. Kamino. 1996. “Structure of Integrated Masatomo Suzuki http://orcid.org/0000-0002-2713-0695 Setting in Townscape as Pedestrian Environment.” Junichiro Mori http://orcid.org/0000-0002-9787-3857 Journal of Architecture and Planning (Transactions of AIJ) Takashi Nicholas Maeda http://orcid.org/0000-0003-2419- 484: 167–174. [in Japanese]. doi:10.3130/aija.61.167_3. Law, S., B. Paige, and C. Russell. 2019. “Take a Look Around: Using Street View and Satellite Images to Estimate House Prices.” ACM Transactions on Intelligent Systems and Technology (TIST) 10 (5): 1–19. References doi:10.1145/3342240. Biljecki, F., and K. Ito. 2021. “Street View Imagery in Urban Law, S., C. I. Seresinhe, Y. Shen, and M. Gutierrez-Roig. 2020. Analytics and GIS: A Review.” Landscape and Urban “Street-Frontage-Net: Urban Image Classification Using Planning 215: 104217. Deep Convolutional Neural Networks.” International Journal of Geographical Information Science 34 (4): Chen, L., X. Yao, Y. Liu, Y. Zhu, W. Chen, X. Zhao, and T. Chi. 2020. 681–707. doi:10.1080/13658816.2018.1555832. “Measuring Impacts of Urban Environmental Elements on Lindenthal, T., and E. B. Johnson. 2021. “Machine Learning, Housing Prices Based on Multisource Data: A Case Study of Architectural Styles and Property Values.” The Journal of Shanghai, China.” ISPRS International Journal of Geo- Real Estate Finance and Economics. forthcoming. Information 9 (2): 106. doi:10.3390/ijgi9020106. doi:10.1007/s11146-021-09845-1. Cordts, M., M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, McNair, B. J., J. Bennett, D. A. Hensher, and J. M. Rose. 2011. R. Benenson, U. Franke, S. Roth, and B. Schiele. 2016. “Households’ Willingness to Pay for Overhead-to- “The Cityscapes Dataset for Semantic Urban Scene underground Conversion of Electricity Distribution Understanding.” In Proceedings of the IEEE Conference Networks.” Energy Policy 39 (5): 2560–2567. doi:10.1016/j. on Computer Vision and Pattern Recognition (CVPR), Las enpol.2011.02.023. Vegas, NV, USA. 3213–3223. Neuhold, G., T. Ollmann, S. Rota Bulò, and P. Kontschieder. Fu, X., T. Jia, X. Zhang, S. Li, Y. Zhang, and S. Fu. 2019. “Do 2017. “The Mapillary Vistas Dataset for Semantic Street-Level Scene Perceptions Affect Housing Prices in Understanding of Street Scenes.” In Proceedings of the Chinese Megacities? An Analysis Using Open Access IEEE International Conference on Computer Vision (ICCV), Datasets and Deep Learning.” PloS ONE 14 (5): e0217505. Venice, Italy. 4990–4999. doi:10.1371/journal.pone.0217505. Nishio, S., and F. Ito. 2015. “The Effect of Sky Factor and Gao, X., and Y. Asami. 2001. “The External Effects of Local the Change on Impression of Townscape: With the Attributes on Living Environment in Detached Residential Experiment for Evaluation Impression while Walking.” Blocks in Tokyo.” Urban Studies 38 (3): 487–505. Journal of Architecture and Planning (Transactions of doi:10.1080/00420980120027465. AIJ) 710: 907–914. [in Japanese]. doi:10.3130/ Gao, X., and Y. Asami. 2007. “Effect of Urban Landscapes on aija.80.907. Land Prices in Two Japanese Cities.” Landscape and Urban Nishio, S., and F. Ito. 2020. “Statistical Validation of Utility of Planning 81 (1–2): 155–166. doi:10.1016/j. Head-mounted Display Projection-based Experimental landurbplan.2006.11.007. Impression Evaluation for Sequential Streetscapes.” He, L., A. Páez, and D. Liu. 2017. “Built Environment and Violent Environment and Planning B: Urban Analytics and City Crime: An Environmental Audit Approach Using Google Science 47 (7): 1167–1183. Street View.” Computers, Environment and Urban Systems Oku, T. 1985. “The Influence of Townscape Elements on the 66: 83–95. doi:10.1016/j.compenvurbsys.2017.08.001. Estimation of Townscape: Experimental Study on Visual Hirate, K., and M. Yasuoka. 1986. “A Study on the Evaluation of Street-Vistas with Roadside Trees: The Experiments with Characteristics and Feelings of Townscape Part 2.” Journal Using Black and White Photomontage Slidefilms.” Journal of Architecture, Planning and Environmental Engineering of Architecture, Planning and Environmental Engineering (Transactions of AIJ) 351: 27–37. [in Japanese]. (Transactions of AIJ) 362: 35–43. [in Japanese]. doi:10.3130/aijax.351.0_27. doi:10.3130/aijax.362.0_35. Shimizu, C., K. G. Nishimura, and T. Watanabe. 2016. “House Ishikawa, T., A. Okabe, Y. Sadahiro, and S. Kakumoto. 1995. Prices at Different Stages of the Buying/Selling Process.” “A Study on the Cognition of Open Space Using 3-D Stereo Regional Science and Urban Economics 59: 37–53. doi:10.1016/j.regsciurbeco.2016.04.001. GIS.” Journal of Architecture and Planning (Transactions of Takei, M., and T. Fukushima. 1983. “A Study on Effect of Tree AIJ) 475: 149–154. [in Japanese]. doi:10.3130/aija.60.149_5. Planning Which Mitigates the Sense of Oppression by Ito, K., and F. Biljecki. 2021. “Assessing Bikeability with Street A Building: A Study on Visual Environmental Planning of View Imagery and Computer Vision.” Transportation Urban District.” Journal of Architecture and Planning Research Part C: Emerging Technologies 132: 103371. (Transactions of AIJ) 332: 102–110. [in Japanese]. doi:10.1016/j.trc.2021.103371. 16 M. SUZUKI ET AL. Tang, J., and Y. Long. 2019. “Measuring Visual Quality of Ye, Y., H. Xie, J. Fang, H. Jiang, and D. Wang. 2019a. “Daily Street Space and Its Temporal Variation: Methodology Accessed Street Greenery and Housing Price: Measuring and Its Application in the Hutong Area in Beijing.” Economic Performance of Human-scale Streetscapes via Landscape and Urban Planning 191: 103436. doi:10.1016/ New Urban Data.” Sustainability 11 (6): 1741. doi:10.3390/ j.landurbplan.2018.09.015. su11061741. Tao, A., K. Sapra, and B. Catanzaro. 2020. “Hierarchical Ye, Y., W. Zeng, Q. Shen, X. Zhang, and Y. Lu. 2019b. Multi-scale Attention for Semantic Segmentation.” arXiv “The Visual Quality of Streets: A Human-centred preprint arXiv:2005.10821. Continuous Measurement Based on Machine Learning Tempesta, T., D. Vecchiato, and P. Girardi. 2014. “The Landscape Algorithms and Street View Images.” Environment and Benefits of the Burial of High Voltage Power Lines: A Study in Planning B: Urban Analytics and City Science 46 (8): Rural Areas of Italy.” Landscape and Urban Planning 126: 1439–1457. 53–64. doi:10.1016/j.landurbplan.2014.03.003. Yokohari, M., K. Takeuchi, T. Watanabe, and S. Yokota. 2000. Yamagata, Y., D. Murakami, T. Yoshida, H. Seya, and S. Kuroda. “Beyond Greenbelts and Zoning: A New Planning Concept 2016. “Value of Urban Views in a Bay City: Hedonic Analysis for the Environment of Asian Mega-Cities.” Landscape and with the Spatial Multilevel Additive Regression (SMAR) Urban Planning 47 (3–4): 159–171. doi:10.1016/S0169- Model.” Landscape and Urban Planning 151: 89–102. 2046(99)00084-5. doi:10.1016/j.landurbplan.2016.02.008. Zhang, Y., and R. Dong. 2018. “Impacts of Street-visible Yang, J., H. Rong, Y. Kang, F. Zhang, and A. Chegut. 2021. “The Greenery on Housing Prices: Evidence from a Hedonic Financial Impact of Street-level Greenery on New York Price Model and a Massive Street View Image Dataset in Commercial Buildings.” Landscape and Urban Planning Beijing.” ISPRS International Journal of Geo-Information 214: 104162. doi:10.1016/j.landurbplan.2021.104162. 7 (3): 104. doi:10.3390/ijgi7030104. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Asian Architecture and Building Engineering Taylor & Francis

The economic value of urban landscapes in a suburban city of Tokyo, Japan: A semantic segmentation approach using Google Street View images

Download PDF
16 pages

Loading next page...
 
/lp/taylor-francis/the-economic-value-of-urban-landscapes-in-a-suburban-city-of-tokyo-ripj7MNVc7

References (35)

Publisher
Taylor & Francis
Copyright
© 2022 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group on behalf of the Architectural Institute of Japan, Architectural Institute of Korea and Architectural Society of China.
ISSN
1347-2852
eISSN
1346-7581
DOI
10.1080/13467581.2022.2070492
Publisher site
See Article on Publisher Site

Abstract

JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING https://doi.org/10.1080/13467581.2022.2070492 The economic value of urban landscapes in a suburban city of Tokyo, Japan: A semantic segmentation approach using Google Street View images a b c b Masatomo Suzuki , Junichiro Mori , Takashi Nicholas Maeda and Jun Ikeda a b Center for the Promotion of Social Data Science Education and Research, Hitotsubashi University, Tokyo, Japan; Graduate School of Information Science and Technology, The University of Tokyo, Tokyo, Japan; Department of Information System Engineering, School of System Design and Technology, Tokyo Denki University, Tokyo, Japan ABSTRACT ARTICLE HISTORY Received 2 February 2022 Visual impression of urban landscape has been investigated in detail through behavioral Accepted 22 April 2022 experiments and questionnaire surveys in the field of architecture. However, in order to give an incentive to build and maintain a good residential environment, an economic consideration KEYWORDS of the urban landscape across space is also an important aspect. Employing a semantic Hedonic price model; segmentation approach using Google Street View images, we investigate the relationship landscape; machine learning; between urban landscapes and property prices in low-rise residential areas in a suburban Google Street View images; city of Tokyo, Japan. Such visual images are used to represent the landscape of both surround- Japan ing districts in general and the street-level landscape; more specifically, the latter is derived after controlling for the district fixed effect. We first show that greenery, openness and visual enclosure are positively correlated with property price at street level. We also investigate the value of urban landscapes commonly seen in Japan: (i) the presence of a power pole is negatively correlated with property price at district level; and (ii) the presence of a road shoulder or farmland, either of which may disrupt the continuity of a residential area, does not exhibit negative correlations with property price at either district or street level. time to nearest station), and land use around the prop- 1. Introduction erties. The hedonic regression analysis separates the Visual impression of urban landscape has been widely contribution of each component to property prices. investigated in detail through behavioral experiments The economic value of a landscape is captured using and questionnaire surveys in the field of architecture. a landscape index, which provides additional explana- Urban landscape impressions are basically based on tory variables to the basic characteristics. The land- components related to greenery, openness, and visual scape index has traditionally been created through enclosure (e.g., Hirate and Yasuoka 1986; Ishikawa et al. a field survey, whereby criteria are set and scored by 1995; Nishio and Ito 2015, 2020; Takei and Fukushima a researcher (Gao and Asami 2007). Alternatively, 1983). The other elements such as power poles (elec- researchers collect detailed land use information (Gao trical wires) also form the urban landscape impression and Asami 2001), advertising information or 3D geo- (Oku 1985). These urban landscapes may be recog- graphical information on whether a property offers nized as a street level landscape or as a landscape a scenic view, such as of the ocean (Jim and Chen having a certain spatial unit such as a district (Koura 2009; Yamagata et al. 2016), to create the landscape and Kamino 1995, 1996). However, whether these index. impressions are reflected in economic valuation has This paper further extends these methods by cap- been less intensively investigated. In order to give an turing urban landscape factors using Google Street incentive to build and maintain a good residential View images. Recent research has recognized that by environment, an economic consideration of the combining Street View images and artificial intelli- urban landscape across space is also an important gence (AI) technology for image recognition, there is aspect. the potential to greatly improve urban landscape ana- The quality of surrounding landscapes has lyses (Biljecki and Ito 2021). Yang et al. (2021) have a potential impact on real estate price formation quantified the amount of greenery in Street View because landscapes represent elements that make up images based on the percentage of pixels recognized the quality of the surrounding neighborhood. There as greenery in the image. Using open-source image are many factors forming property prices, including segmentation methods that employ deep learning building characteristics (e.g., property age and floor (e.g., SegNet), recent studies have decomposed Street area), land characteristics (e.g., land area and walking View images into each landscape element and have CONTACT Masatomo Suzuki suzuki.m@r.hit-u.ac.jp Center for the Promotion of Social Data Science Education and Research, Hitotsubashi University, 2-1, Naka, Kunitachi, Tokyo, 186-8601, Japan © 2022 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group on behalf of the Architectural Institute of Japan, Architectural Institute of Korea and Architectural Society of China. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. 2 M. SUZUKI ET AL. created indexes typically based on the pixel ratio (Chen space within a private lot – is a typical characteristic of et al. 2020; Fu et al. 2019; Ito and Biljecki 2021; Tang Japanese cities. Given that residential lots tend to be small and Long 2019; Ye et al. 2019a, 2019b; Zhang and in Japanese cities, a typical parking space is not hidden by Dong 2018). Using these new approaches, recent lit- fences or greenery (as will be shown in Section 4) and erature investigates the relationship between property disrupts the continuity of a residential area. Intuitively the price and landscape elements, while the landscape presence of such a car park would have a negative impact type has been largely restricted to the proportions of on neighboring property prices, but as this situation is greenery, sky and buildings present (Chen et al. 2020; common in Japan, its specific impact is one of the issues Fu et al. 2019; Yang et al. 2021; Ye et al. 2019a; Zhang examined in this paper. and Dong 2018). In Western contexts, other factors like In low-rise residential areas in a suburban city of “curb appeal” (that is, the attractiveness of the exterior Tokyo, this paper investigates the relationship of a property when viewed from a public space, such as between common Japanese urban landscape and a street or sidewalk), architectural style, and urban property prices via the following three steps. First, we design elements have been gradually investigated collect Google Street View images in front of detached using new image recognition technologies (Johnson, houses that have been sold. Second, we extract com- Tidwell, and Villupuram 2020; Lindenthal and Johnson ponents of the landscape using AI technology for 2021). image recognition, and create what we call the Greenery, openness, and visual enclosure are basic “urban landscape index.” Specifically, we apply the elements of urban landscapes in Japan and in other recent semantic segmentation approach proposed by countries as well. However, other elements of urban Tao, Sapra, and Catanzaro (2020), which as of landscapes differ greatly in Japan from those in August 2021 has achieved state-of-the-art results in Europe, the United States and even other Asian coun- two commonly used, large-scale open data sets: tries. Specifically, the existence of power poles above Mapillary and Cityscapes. This approach is impressive the ground, the presence of urban farmland within in its segmentation accuracy and label diversity com- residential areas, and private parking space facing the pared to previous studies. Third, we construct a model road, are landscape factors specific to Japan that can to explain the extent to which these components con- be tested from the Street View image data. First, the tribute to the transaction price. After controlling for undergrounding of power lines is not at all common in a sufficient range of characteristics of the properties Japanese cities. Whereas major cities in Europe (such and the neighborhoods considered, the strength of the as London and Paris) and Asia (for instance, Hong Kong correlation between the urban landscape index and and Singapore) have all become pole-free, Japan lags property price is derived from the “landscape” channel. behind, this goal being achieved in only 8% of Tokyo’s The component captured here as the urban landscape wards area and 6% of the city of Osaka. The existence index may represent the landscape of the surrounding of power poles at ground level in residential areas district in general (e.g., the landscape of a well- potentially lowers the value of properties, as the over- developed residence), or the street-level landscape head-to-underground conversion of electricity distri- more specifically. Thus, the landscape channel is dis- bution networks has been perceived positively in tinguished as district-level and street-level, the latter of Western countries, based on stated preferences which is derived after further controlling for the district (McNair et al. 2011; Tempesta, Vecchiato, and Girardi fixed effect. 2014). The rest of the paper is organized as follows. Section Second, farmland remains within residential areas on 2 describes the study area and property transaction the fringes of Japanese cities. Whereas in Western cities, data. Section 3 presents the methodology. Section 4 urban planning concepts, such as zoning and greenbelt presents examples of semantic segmentation of the additions, have been applied to encourage controlled Japanese landscape and develops hypotheses. urban growth, Asian cities have historically placed land Section 5 presents the variables and summary statistics use patterns of urban and rural character next to each for the following hedonic regression analysis. Section 6 other (Yokohari et al. 2000). Although this type of land- shows the empirical results. Section 7 concludes the scape potentially gives us a positive impression through paper. the provision of rural environments for urban residents, the presence of farmland may disrupt the continuity of an urban residential area; its specific impact is one of the 2. Study area and property transaction data issues examined in this paper. Third, unlike the parking To investigate the economic value of urban landscape space on road shoulders seen in Western cities, private in typical low-rise residential areas commonly seen in parking space facing the road – that is, paved parking the suburbs of metropolitan areas in Japan, the Ministry of Land, Infrastructure, Transport and Tourism. “Status of development of non-pole systems (domestic and overseas).” Available at: https://www. mlit.go.jp/road/road/traffic/chicyuka/chi_13_01.html (in Japanese; accessed 9 December 2021). JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 3 analysis is conducted in Hachioji city, a suburban city Table 1. Summary statistics. (a) Property characteristics of Tokyo’s metropolitan area. The city is located more Mean S.D. than 30 km away from Shinjuku Station, a terminus in Dependent variable central Tokyo. The population is around 577,000 Price [10,000 JPY] 2466.3 1115.9 according to the 2015 Census and has remained Building characteristics Newly built 0.250 almost stable since 2005, and a relatively stable hous- Age [years] 17.5 14.4 ing demand is expected in the analysis period. We 2 Floor area [m ] 107.3 27.9 employ samples in areas with designated zoning Non-timbered 0.109 Land characteristics (that is, we exclude mountainous areas in which zon- Land area [m ] 169.1 63.2 ing is not applied), especially focusing on low-rise Walking time to station 5.9 7.4 [min] residential areas (i.e., category I, exclusively low-rise Bus to station 0.530 residential zones) in the main analysis. In our transac- With parking space 0.755 Distance to junior high 0.777 0.401 tion samples of detached houses, 61% are located in school [km] low-rise residential areas. Floor-area ratio (FAR) [%] 75.638 13.823 Front road width We employ the transaction data of detached Less than 4 m 0.034 houses collected by an association of real estate 4–5 m 0.179 agents through the Real Estate Information 5–6 m 0.196 2 6–7 m 0.254 Network System (REINS). Information on real 7–10 m 0.053 estate transactions is recorded in these data, in 10 m or more 0.093 Unknown 0.193 the same way as in the Multiple Listing Service Land use within 500 m (MLS) in the United States. The REINS is the only radius [ratio] High-rise building 0.020 0.038 real estate transaction system designated by the Dense low-rise building 0.021 0.081 Ministry of Land, Infrastructure, Transport and Low-rise building 0.630 0.168 Vacant land 0.016 0.030 Tourism (MLIT) in Japan, and is a representative Park and green 0.005 0.014 system for real estate companies to register trans- Farmland 0.035 0.057 action information. The registration of data is con- Number of observations 800 sidered to have sufficient coverage, as it is (b) Urban landscape index created through sematic segmentation mandatory for a real estate agent or company to of the Google Street View images register the information on transacted properties Variable Mean Min Max under exclusive brokerage service agreement, Mean S.D. Mean S.D. Mean S.D. although not for all properties. Indeed, transaction Vegetation 0.108 0.080 0.236 0.170 Sky 0.237 0.059 0.346 0.053 volumes in each regional submarket (mainly at Building 0.200 0.088 0.068 0.059 prefecture level) based on the REINS data have Building + Wall + Fence 0.348 0.102 been announced to the public as market reports. Pole 0.007 0.006 0.027 0.025 Sidewalk 0.029 0.023 0.094 0.057 In this context, the number of samples and their Sidewalk + Car 0.042 0.030 property characteristics studied in this paper is Terrain 0.025 0.039 0.091 0.120 likely to be representative of the submarket of Hachioji city. We employ newly built and resold detached houses 3. Methodology transacted from 2016 to 2019. There are 800 observa- tions in total in the low-rise residential areas consid- Figure 1 shows the flow of our analyses. First, using Street ered (and 1,307 samples in the entire area of the View Static API, 12 Google Street View images (i.e., every Hachioji city), after truncating samples with missing 30 degrees) are collected in front of each transacted or atypical information. The addresses are geocoded property. To be more specific, for each property, the using Geocoding API, provided by Google. The trans- nearest point on the road is chosen. Although the land- action prices and other property characteristics (which scape impression may change depending on the retriev- will be shown later in Table 1(a) are recorded in the ing point and on the angle, the effects are reduced when data. As we describe in the next section, variables we take the average of the multiple images at the retriev- constructed from the Google Street View images will ing point. We collect the most recent images uploaded to be linked to the property transaction data. Google Maps as of August 2021. For details, see: http://www.reins.or.jp/ (in Japanese; accessed 22 December 2021). For details of the market data created from the REINS database, see: http://www.reins.or.jp/library/2019.html (accessed 16 February 2022; in Japanese). Further, for condominiums in the Tokyo metropolitan area, Shimizu, Nishimura, and Watanabe (2016) compare the nature of the REINS data and other related data sources: properties listed in a web portal provided by one of the largest private vendors of residential information, and transacted properties (a part of registered properties) whose transaction prices are collected by the MLIT. It is shown that the regression coefficients in the hedonic regression analyses are at least consistent even though there may be differences in property characteristics. In this low-rise residential area, we truncate 118 samples for inaccurate addresses, 26 samples for having a property age older than 50 years, and 27 samples for other missing information. 4 M. SUZUKI ET AL. (i) Collect Google Street View images Section 4 Semantic segmentation (ii) Create urban landscape index (iii) Hedonic Economic value of regression urban landscape analysis Section 5 Basic characteristics Section 6 collected from property Property price transaction data and land use data Figure 1. Methodological flow. Second, for the collected images, semantic segmen- characteristics of building as explanatory variables in tation is conducted, a deep learning algorithm that the hedonic regression analysis so that the building associates a label or category to every pixel in an price level is sufficiently controlled. image. We employ the recent semantic segmentation We estimate the correlation between property price approach proposed by Tao, Sapra, and Catanzaro and urban landscape in the following regression: (2020), which as of August 2021 has achieved state-of- X X the-art results in two commonly used open data sets: ln P ¼ αþ β V þ γ X þ D þ T þ e (1) it ki j t it s si k Mapillary and Cityscapes. Cityscapes is a large data set that labels semantic classes across 5,000 urban street where ln P is the log of the transaction price for unit it images in 50 German cities (Cordts et al. 2016). i in period t. α is the constant. V is the urban landscape si Mapillary Vistas is another large data set, containing index for landscape component s, and β is the corre- 25,000 high-resolution images from across the world, sponding coefficient. Specifically, the index represents with annotated semantic object categories (Neuhold the proportion of landscape component s in the et al. 2017). The semantic segmentation model is first images, and is created through semantic segmentation pre-trained on the larger Mapillary, and then trained of the Google Street View images. More precisely, 1% on Cityscapes. Using the pre-trained model, every increase in the proportion of component s leads to a β Street View image is partitioned into multiple- % increase in the transaction price. X is the control ki segment categories: road, sidewalk, building, wall, variable k shown in Table 1(a), and γ is the corre- fence, pole, traffic light, traffic sign, vegetation, terrain, sponding coefficient. D captures the district fixed sky, person, rider, car, truck, bus, train, motorcycle and effect, that is, it takes 1 for properties in district j, and bicycle. Given that some categories are almost blank, 0 otherwise. T captures the quarterly fixed effect for we create our urban landscape index using major the time of the transaction, that is, it takes 1 for categories (which are shown in Table 1(b)). a property transaction in period t, and 0 otherwise. e it Third, we conduct hedonic regression analysis to is the error term. decompose the property prices (that is, the sum of When we exclude the district fixed effect, D , the the building price and land price) into multiple com- urban landscape index V serves as a proxy for the si ponents. There are many factors forming the property district-level landscape. For instance, if the entire prices, including building characteristics (e.g., property residential area (i.e., the district) has been uniformly age and floor area), land characteristics (e.g., land area developed as a “greenery residential area,” the and walking time to nearest station), and land use greenery is likely to be the district-level landscape. around the properties. Beyond these basic character- On the other hand, when we include the district istics, this paper further extends these methods by fixed effect, D , the urban landscape index V serves j si capturing urban landscape factors using Google as a proxy for the street-level landscape. For Street View images. It is expected that the urban land- instance, even within each “greenery residential scape index would impact the land price, and the area” (i.e., within each district), openness can vary building price would be basically determined by the across streets. characteristics of building itself. Thus, we include the The economic value of urban landscape for existing properties can be captured through a property transaction price, which is the sum of building and land prices. The transaction price in our data more formally captures a market valuation than publicly available appraisals of land prices in Japan. Employing land transaction data also has difficulty in acquiring Google Street View images containing buildings that will be constructed on the land after the transactions; this may completely alter the landscape of newly developing residential areas. JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 5 Figure 2. Examples of original images and semantic segmentation.The right-hand squares show the pixel ratios of each landscape component within the images. 4. Semantic segmentation and hypotheses category captures “openness” as the sky component development (accounting for 28.3%). The dark gray category captures “visual enclosure” as the building component (account- 4.1. Example of semantic segmentation ing for 13.8%). In some cases, buildings are also captured To ensure that each category from the semantic segmen- by a wall (light blue category; accounting for 9.4%) or tation captures the intended landscape component, a fence (light pink category; accounting for 4.5%); thus, Figure 2 shows examples of original images and their we will also create an alternative indicator of the visual semantic segmentation outcomes, as well as the pixel enclosure as the sum of these three categories. The light ratios of each landscape component within the images. gray category captures “power pole” as the pole compo- Although the pre-trained model is trained on street nent (accounting for 1.5%). The purple category captures images from all around the world, we confirm that the the road component (accounting for 29.3%), and this is semantic segmentation is conducted properly for the set as a reference category in the following hedonic Japanese Street View images. regression analyses. Panel (a) shows a typical residential area. The dark Panel (b) shows a case capturing “road shoulder,” green category captures vertical “greenery” as the vege- which is an open space in the shoulder of a road that is tation component (accounting for 11.6% of the pixels in often used as private parking space or as paved vacant the entire image), including trees and plants. The blue lots. This is captured by the dark pink category as the It is true that in some cases, sidewalks are really sidewalks for wider streets. However, as in Figure 2(a), in typical residential districts with narrow streets, it is not common to see sidewalks with curbstones. Note that we controlled for the front road width in the hedonic regression analyses to partially separate this effect. 6 M. SUZUKI ET AL. Figure 3. 12 views (every 30 degrees) for the typical residential areas shown in Figure 2(a). sidewalk component (accounting for 6.1%). In some shoulder as the sum of the two components: sidewalk cases, cars are parked in the parking space, and so we and car (dark blue category; accounting for 0.6%). Panel will also create an alternative indicator of the road (c) shows a case capturing “farmland” within a residential JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 7 Table 2. Percentage of each landscape component for the 12 views of the typical residential areas shown in Figure 2(a). Building + Number Vegetation Sky Building Wall + Fence Pole Sidewalk Sidewalk + Car Terrain 1 11.64 28.21 13.79 27.72 1.52 1.60 1.60 0.02 2 8.21 28.97 18.60 22.39 0.11 2.03 13.57 0.61 3 8.68 22.79 37.28 46.83 0.00 3.60 11.02 0.13 4 3.91 14.91 32.74 56.26 0.00 0.12 11.02 9.30 5 6.93 12.58 63.58 69.44 0.00 1.19 1.19 0.07 6 2.80 16.42 39.39 45.55 0.24 0.00 0.00 0.00 7 2.89 26.12 31.07 39.09 0.16 1.15 1.15 0.00 8 3.51 26.35 17.30 38.31 0.33 1.31 1.31 0.00 9 0.75 16.91 33.18 58.58 0.00 2.29 2.29 0.00 10 2.51 7.66 45.78 74.20 0.00 0.16 0.16 0.00 11 9.47 12.92 33.16 55.71 1.55 0.55 0.55 0.00 12 8.78 26.48 20.04 33.53 1.17 0.82 0.82 0.04 Mean 5.84 20.03 32.16 47.30 0.42 1.24 3.72 0.85 Min 0.75 7.66 13.79 22.39 0.00 0.00 0.00 0.00 Max 11.64 28.97 63.58 74.20 1.55 3.60 13.57 9.30 In the hedonic regression analysis, aggregated proportions of the 12 views (shown in bold) are employed. area, which is the exposure of the ground. This is captured images, the summary statistics to be measured, and by the light green category as the terrain component a hypothesis on the direction of the correlation with (accounting for 16.9%). Note that the vertical greenery the price level. in the farmland is captured by the vegetation component. Our hypotheses on the major scene perceptions are For the typical residential area (panel (a) of Figure 2), basically consistent with previous studies on the visual Figure 3 shows 12 Street View images (i.e., every 30 impression of urban landscape in the field of architec- degrees). Table 2 shows the proportion of each land- ture (e.g., Hirate and Yasuoka 1986; Ishikawa et al. scape component for these 12 views in percentage 1995; Nishio and Ito 2015, 2020; Takei and Fukushima terms; the mean, minimum and maximum values for 1983) and on the relationship between property price these 12 views are also calculated. Although the and landscape elements using recent image recogni- semantic segmentation is conducted properly overall, tion technologies (e.g., Chen et al. 2020; Fu et al. 2019; we see that there are some misclassifications of side- Yang et al. 2021; Ye et al. 2019a; Zhang and Dong walks, walls and fences by buildings (e.g., image num- 2018). “Greenery” is captured through the mean or ber 5). It is also true that the proportion of buildings maximum proportion of the vegetation component. increases when the building is in the foreground (e.g., It is expected to have a positive relationship with image number 10). These problems can be reduced by transaction price through the provision of taking the mean value of the proportion for each a comfortable environment with abundant greenery component from the 12 views, justified by the fact and plantings. “Openness” is captured through the that the landscape is an average impression from mean or maximum proportion of the sky component. a 360-degree view at the location. The minimum or It is expected to have a positive relationship with maximum values of the proportion from the 12 views transaction price by offering sufficient sunlight to the may also be useful in capturing the strongest impres- community. “Visual enclosure” is captured through the sion in the location. For instance, if the minimum mean or minimum proportion of the building compo- proportion of the building is large, the impression nent, as well as the mean proportion of the sum of the conveyed is that the visual enclosure is high at the building, wall and fence components. Sufficient level location; by contrast, the maximum proportion of the of visual enclosure represents that the residential area building may not be useful, because the view in front is matured, and thus, it is expected to have a positive of the building is always included in the 12 views. relationship with transaction price. Similarly, if the maximum proportion of vegetation is With regard to urban landscapes specific to Japan, large, the impression conveyed is that the location has “power pole” is captured through the mean or max- plenty of greenery. imum proportion of the pole component. It is expected to have a nonpositive relationship with transaction price, as it is a messy landscape characterized by elec- 4.2. Hypotheses trical wires in the sky, but it is so common in the Japanese landscape that it may not reduce the transac- Based on the examples discussed above, Table 3 pre- tion price. “Road shoulder” is captured through the sents the hypotheses for the hedonic regression analy- mean or maximum proportion of pixels for the side- sis. For each landscape element addressed here, we walk component, as well as the mean proportion of the present the baseline and additional components of the 8 M. SUZUKI ET AL. Table 3. Hypotheses regarding correlations between the urban landscape index and property price. Landscape Urban landscape Summary Expected correlation with property price element index statistics Greenery Vegetation Mean (Max) + (Comfortable environment with abundant greenery and plantings) Openness Sky Mean (Max) + (Sufficient sunlight in the community) Visual enclosure Building (+ Wall + Mean (Min) + (Matured residential area with a certain level of enclosure by buildings) Fence) Power pole Pole Mean (Max) - (Messy Landscape characterized by electrical wires) or 0 (Common urban landscape in Japan) Road shoulder Sidewalk (+ Car) Mean (Max) ? (Loss of continuity of buildings or visual enclosures; Enhance openness; Common urban landscape in Japan) Farmland Terrain Mean (Max) ? (Loss of continuity of buildings or visual enclosures; Enhance openness; Common urban landscape in Japan) sum of the sidewalk and car components. Its relation- Information (Ministry of Land, Infrastructure, Transport ship with transaction price is unclear; although it and Tourism). The categories include: high-rise build- reduces the continuity of buildings or the visual enclo- ing; dense low-rise building (that is, a low-rise building sure, it enhances the openness. This is so common in is densely concentrated in the area); low-rise building the Japanese landscape that it may not reduce the (the most common surrounding land use type, as we transaction price. “Farmland” is captured through the employ low-rise residential area samples); vacant land; mean or maximum proportion of the terrain compo- park and green space (well-maintained park or green nent. Its relationship with transaction price is also space); and farmland. After controlling for the initial unclear; the same mechanisms as for road shoulder three categories, “openness” and “visual enclosure” apply. capture the landscape of the location. Further, the last three categories control part of the quality of the location. The existence of vacant land partly controls 5. Variables and summary statistics for the sidewalk component, the existence of a park and/or green space partly controls for the vegetation We employ low-rise residential area samples for the component (and even the pole component, as the main analysis. Table 1 shows summary statistics overall quality of the location), and the existence of (means and standard deviations) for (a) property char- farmland around the location partly controls for the acteristics and (b) urban landscape index, the latter of farmland component. which is created through sematic segmentation of the In panel (b), the means and standard deviations of Google Street View images. In panel (a), the building the urban landscape index created through sematic characteristics include: a newly built dummy (taking 1 segmentation of the Google Street View images are for a newly built house and 0 otherwise); property age; shown for the mean and minimum and maximum floor area; and a non-timbered dummy (most Japanese proportions. On average, vegetation, sky and buildings detached houses are timbered). The land characteris- account for 10.8%, 23.7%, and 20.0%, respectively, of tics include: land area; walking time to the nearest the mean proportion. On the other hand, the corre- station; a dummy variable indicating a need to take sponding values for pole, sidewalk and terrain are less a bus to the nearest station; a dummy variable for a lot than 5%. As expected, the average of the minimum with parking space; distance to a junior high school; (maximum) proportion is smaller (larger) than the and the regulation (upper limit) of the floor-area ratio mean proportion of the corresponding landscape com- (FAR). After controlling for the existence of parking ponent; these variables will be employed in the robust- space in the lot, the road shoulder captures the land- ness checks. scape of the location. Front road width is included as a categorical variable. In Japan, if the front road width is less than 4 m (and the length of the lots connected 6. Empirical results to the road is less than 2 m), rebuilding the existing 6.1. Main results house is forbidden. The “less than 4 m category” is expected to be negatively correlated with price. The Table 4 shows the correlation between property price unknown status is included so as not to delete missing and the urban landscape index based on equation (1). observation samples, which exist in non-negligible As in the following discussion, the results confirm that numbers. the quality of surrounding landscapes has a meaning- Land use within a 500 m radius around the property ful relationship with real estate price formation follow- (that is, the ratio of each land use type) is measured ing the hypotheses in Table 2. Columns (1) and (2) using a geographic information system (GIS) based on control only the very basic variables, and thus the 50 m-mesh land use data, provided as National Land urban landscape index simply serves as a locational JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 9 Table 4. Correlations between urban landscape index and property price. Measuring: Locational proxy District-level landscape Street-level landscape (1) (2) (3) (4) (5) (6) Urban landscape index Vegetation (Mean) 0.6469*** 0.5866*** 0.3454* (0.1924) (0.1892) (0.2018) Sky (Mean) 0.9836*** 0.9684*** 1.0541*** (0.2866) (0.2957) (0.2905) Building (Mean) 0.2578 0.2528 0.5360*** (0.1743) (0.1750) (0.1980) Pole (Mean) −3.6017* −3.4615* −1.2807 (1.9449) (1.7929) (1.8650) Sidewalk (Mean) 1.3820*** 0.7494 0.1757 (0.5338) (0.5243) (0.5769) Terrain (Mean) −0.4643 −0.2295 −0.0182 (0.3085) (0.2788) (0.2567) Building characteristics Newly built 0.0114 0.0111 0.0644* 0.0578 0.1072*** 0.1038*** (0.0337) (0.0351) (0.0350) (0.0360) (0.0346) (0.0345) Age [years] −0.0180*** −0.0180*** −0.0169*** −0.0170*** −0.0169*** −0.0167*** (0.0013) (0.0013) (0.0013) (0.0013) (0.0013) (0.0012) Floor area [m ] 0.0064*** 0.0063*** 0.0051*** 0.0050*** 0.0041*** 0.0041*** (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) Non-timbered 0.0384 0.0395 0.0415 0.0437 (0.0418) (0.0413) (0.0413) (0.0408) Land characteristics Land area [m ] 0.0019*** 0.0017*** 0.0019*** 0.0018*** 0.0017*** 0.0016*** (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) Walking time to station [min] −0.0130*** −0.0139*** −0.0157*** −0.0162*** −0.0098*** −0.0095*** (0.0028) (0.0027) (0.0030) (0.0030) (0.0037) (0.0035) Bus to station −0.5044*** −0.5210*** −0.4926*** −0.5077*** −0.2076*** −0.2138*** (0.0396) (0.0389) (0.0449) (0.0459) (0.0792) (0.0751) With parking space 0.0474* 0.0462* 0.0462* 0.0491* (0.0280) (0.0280) (0.0254) (0.0250) Distance to junior high school [km] −0.1268*** −0.1238*** −0.0099 −0.0238 (0.0277) (0.0278) (0.0558) (0.0544) Floor-area ratio (FAR) [%] −0.0010 −0.0011 0.0018* 0.0015 (0.0008) (0.0008) (0.0010) (0.0009) Front road width Less than 4 m −0.1234** −0.1421*** −0.1656*** −0.1651*** (0.0515) (0.0549) (0.0546) (0.0581) 4–5 m −0.1595*** −0.1614*** −0.1363*** −0.1430*** (0.0350) (0.0355) (0.0340) (0.0345) 5–6 m (reference) (reference) (reference) (reference) 6–7 m 0.0433 0.0344 −0.0213 −0.0265 (0.0280) (0.0277) (0.0291) (0.0285) 7–10 m 0.0352 0.0099 0.0082 −0.0062 (0.0481) (0.0469) (0.0492) (0.0475) 10 m or more 0.0217 −0.0091 −0.0276 −0.0394 (0.0487) (0.0482) (0.0488) (0.0456) Unknown −0.0357 −0.0370 −0.0606* −0.0579* (0.0343) (0.0340) (0.0337) (0.0328) Land use within 500 m radius [ratio] High-rise building 0.3648 0.2703 0.3789 0.1972 (0.2777) (0.2817) (0.5165) (0.5177) Dense low-rise building 0.4992*** 0.5111*** 0.5392* 0.2361 (0.1279) (0.1322) (0.2855) (0.2908) Low-rise building 0.1109 0.0854 0.1773 0.1054 (0.0789) (0.0844) (0.1200) (0.1198) Vacant land 0.9247*** 0.7879*** 0.1512 −0.0726 (0.3043) (0.2929) (0.5705) (0.5676) Park and green 2.9514*** 2.3671*** −1.0923 −1.3626 (0.7220) (0.7411) (1.0142) (1.0067) (Continued) 10 M. SUZUKI ET AL. Table 4. (Continued). Farmland −0.2368 −0.3831** 0.0651 −0.0975 (0.1917) (0.1836) (0.2975) (0.2872) Constant 7.3556*** 7.0257*** 7.5285*** 7.2457*** 7.2960*** 6.9458*** (0.0799) (0.1387) (0.1272) (0.1528) (0.1989) (0.2100) Time fixed effect (quarterly) Yes Yes Yes Yes Yes Yes District fixed effect Yes Yes Number of observations 800 800 800 800 800 800 R 0.635 0.652 0.695 0.705 0.794 0.801 Adjusted R 0.624 0.638 0.678 0.687 0.751 0.757 The dependent variable is the log of the transaction price. White-corrected robust standard errors are in parentheses. Significance level: *** p < 0.01, ** p < 0.05, * p < 0.1. proxy. Columns (3) and (4) add control variables, but The coefficient of sky does not vary much even when still do not include the district fixed effect, hence col- we control much fewer control variables in columns (2) umn (4) measures the district-level landscape. Columns and (4), demonstrating that openness is heterogeneous (5) and (6) now include the district fixed effect, and across districts and across the streets within a district. The thus column (6) measures the street-level landscape. coefficient of building diminishes and loses statistical Comparing R between columns (1) and (2) (0.635 and significance once we control for much fewer control 0.652), between columns (3) and (4) (0.695 and 0.705), variables in columns (2) and (4). This shows that visual and between columns (5) and (6) (0.794 and 0.801), enclosure is heterogeneous across the streets within reveals that the landscape index partly improves the a district, but does not have a clear trend across districts. explanatory power of the model. However, the degree Second, the presence of a power pole is negatively of improvement is much larger when we include addi- correlated with property price at district level. In col- tional control variables. umn (4) without controlling for the district fixed effect, the coefficient of pole is negative with statistical sig- First, greenery, openness, and visual enclosure are nificance; a 1% increase in the proportion of pole leads positively correlated with property price at street level. to a 3.46% decrease in transaction price. However, In column (6) including the district fixed effect, the once we control for the district fixed effect in column coefficients of the mean proportions of vegetation, (6), the size of the coefficient diminishes and statistical sky, and building are positive with statistical signifi - significance is lost. This shows that power poles serve cance. A 1% increase in the proportion of vegetation, as a district-level landscape. The negative valuation at sky, and building, respectively, lead to 0.34%, 1.05%, district-level means that it is recognized as a messy and 0.53% increases in transaction price. landscape characterized by electrical wires in the sky, The coefficient of vegetation is larger when we use and that the residential area achieving high valuation much fewer control variables, showing that greenery is likely to have less power poles in the whole district. partly serves as a locational proxy. Specifically, the park Let us roughly discuss the size of the coefficient of and green space ratio within 500 m is positive and statis- the pole component at district level, 3.46. Three power tically significant in column (4); this and the other control poles are recognized in the semantic segmentation in variables explain the decrease in the size of the coefficient the typical residential area in Figure 2(a), and this is for the mean proportion of the vegetation component summarized as a mean proportion of 0.42% in Table 1. from column (2) to column (4). However, greenery still Thus, one power pole lowers property prices by captures the street-level landscape; that is, even within (3.46 × 0.42)/3 = 0.48 [%], compared to a pole-free a district, the level of greenery is different on each street. landscape. Given that each power pole lowers non- Specifically, the coefficient of vegetation is still positive negligible numbers of surrounding houses, the over- with 10% statistical significance in column (6), which head-to-underground conversion of electricity distri- controls for the district fixed effect. This is in contrast to bution networks is likely to improve property values the fact that the park and green space ratio within 500 m widely in the residential area. now becomes negative without statistical significance; Third, the presence of either a road shoulder or the effect of the existence of a well-maintained park or farmland does not exhibit negative correlations with green space, which is likely to capture the characteristics property price even at district level. In both columns (4) of the district, is absorbed by the district fixed effect. and (6), the coefficients of sidewalk are positive but without statistical significance. The coefficient becomes larger as fewer control variables are included; The variance inflation factor (VIF) for the urban landscape index in Table 4 (columns (1), (4), and (6)) is less than 5, not a level that would cause much of a multicollinearity problem. At variable level, the correlation between the “vegetation” and the “park and green space ratio” is weak (correlation coefficient of 0.03), implying that the greenery differs on each street, as a landscape. JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 11 Table 5. Subsample analysis of land use type. Sample: Entire area Other residential area Loosely regulated area District-level Street-level District-level Street-level District-level Street-level Measuring: landscape landscape landscape landscape landscape landscape (1) (2) (3) (4) (5) (6) Urban landscape index Vegetation 0.4505*** 0.1060 0.2889 0.1646 0.1436 −0.0573 (Mean) (0.1582) (0.1695) (0.3808) (0.4132) (0.5192) (0.6095) Sky (Mean) 0.6148*** 0.7087*** 0.3485 0.4885 0.0111 0.4896 (0.2009) (0.2046) (0.4133) (0.5209) (0.5656) (0.8693) Building (Mean) 0.1893 0.3214** −0.0310 0.5410 0.3167 0.3383 (0.1392) (0.1497) (0.3956) (0.4030) (0.3864) (0.5824) Pole (Mean) −1.5049 0.0444 1.6424 7.8979 2.9837 3.9114 (1.3403) (1.5129) (3.5051) (5.9158) (2.6071) (4.2802) Sidewalk (Mean) 0.3459 −0.0141 −0.1771 1.6259 −0.3943 −1.2080 (0.4646) (0.4866) (0.9767) (1.2077) (1.2012) (1.8886) Terrain (Mean) −0.4776* −0.1830 −0.8464** 0.1083 −0.6307 −0.6099 (0.2498) (0.2389) (0.4156) (0.4695) (0.5149) (0.9942) Building Yes Yes Yes Yes Yes Yes characteristics, Land characteristics, Land use within 500 m radius, Zoning category, Time fixed effect (quarterly) and a Constant District fixed Yes Yes Yes effect Number of 1307 1307 297 297 210 210 observations R 0.699 0.795 0.742 0.894 0.756 0.875 Adjusted R 0.686 0.757 0.693 0.806 0.684 0.748 The dependent variable is the log of the transaction price. White-corrected robust standard errors are in parentheses. Significance level: *** p < 0.01, ** p < 0.05, * p < 0.1. thus, the sidewalk component partly serves as a proxy 6.2. Subsample on land use type of locational quality. In both columns (4) and (6), the Table 5 shows the entire area sample (columns (1) coefficients of terrain are negative but without statis- and (2)) and subsamples of other residential areas tical significance. The coefficient becomes larger in (excluding residential areas with a category absolute terms as fewer control variables are included; I exclusively low-rise residential zone; columns (3) thus, the terrain component partly serves as a proxy of and (4)) and loosely regulated areas (quasi- locational quality. The insignificant or neutral valua- industrial zones, neighborhood commercial zones, tions of road shoulder and farmland mean that these and commercial zones; columns (5) and (6)). types of landscape have two sides: They reduce the Columns (1), (3) and (5) capture the district-level continuity of buildings or the visual enclosure, while landscape without controlling for the district fixed they enhance the openness. It is also true that the effect (corresponding to column (4) of Table 4), landscapes are so common in Japanese residential while columns (2), (4) and (6) capture the street- areas that they may not be reflected in the valuations. level landscape by controlling for the district fixed With regard to the control variables, a narrow front effect (corresponding to column (6) of Table 4). road width (less than 5 m) reduces property price with The nonexclusive zoning system used in Japan statistical significance. Most of the land use ratio within restricts specific types of land use in a location, 500 m radius of a transacted property exhibits statis- and does not intend to realize a single (pure) tical significance when we do not control for the dis- land use therein. “Low-rise residential area” is the trict fixed effect (column (4)). However, it loses strictest type of zoning, literally allowing only low- statistical significance once we control for the district rise residential houses to be situated in an area. In fixed effect (column (6)). This confirms that the district an “other residential area,” low-rise and high-rise fixed effect control is effective in accounting for the residential houses (and even other facilities) can heterogeneity in land use across districts. 12 M. SUZUKI ET AL. Table 6. Robustness checks. (a) Greenery, openness and visual enclosure Measuring: District-level Street-level District-level Street-level District-level Street-level District-level Street-level landscape landscape landscape landscape landscape landscape landscape landscape (1) (2) (3) (4) (5) (6) (7) (8) Urban landscape index Alternative indicator Vegetation (Max) 0.2312*** 0.1781** (0.0842) (0.0886) Sky (Max) 0.6096** 0.7201*** (0.2603) (0.2541) Building (Min) 0.3150 0.6047** (0.2181) (0.2417) Building + Wall + 0.1377 −0.1104 Fence (Mean) (0.2400) (0.2400) Baseline indicator Vegetation (Mean) 0.3657** 0.1219 0.5445*** 0.2394 0.5503** −0.0936 (0.1728) (0.1735) (0.1727) (0.1798) (0.2614) (0.2682) Sky (Mean) 0.8598*** 1.0606*** 0.8706*** 0.8275*** 0.9398** 0.4881 (0.2905) (0.2782) (0.2695) (0.2515) (0.3971) (0.3722) Building (Mean) 0.1612 0.5294*** 0.0390 0.3121* (0.1663) (0.1850) (0.1556) (0.1615) Pole (Mean) −3.6499** −1.2459 −3.1002* −0.9070 −3.7286** −2.0014 −3.4045* −1.7044 (1.7957) (1.8669) (1.7543) (1.8524) (1.8196) (1.9027) (1.7803) (1.8727) Sidewalk (Mean) 0.7678 0.1951 0.9873* 0.4614 0.7256 0.1422 0.9189* 0.1610 (0.5217) (0.5753) (0.5092) (0.5609) (0.5280) (0.5842) (0.5239) (0.5981) Terrain (Mean) −0.2364 −0.0334 −0.2364 −0.0302 −0.1958 0.0276 −0.1925 −0.1518 (0.2756) (0.2545) (0.2780) (0.2637) (0.2894) (0.2712) (0.2981) (0.2826) Building Yes Yes Yes Yes Yes Yes Yes Yes characteristics, Land characteristics, Land use within 500 m radius, Time fixed effect (quarterly) and a Constant District fixed effect Yes Yes Yes Yes Number of 800 800 800 800 800 800 800 800 observations R 0.705 0.801 0.702 0.798 0.705 0.800 0.704 0.798 Adjusted R 0.686 0.758 0.683 0.754 0.687 0.756 0.686 0.753 (b) Power pole, road shoulder, and farmland Measuring: District-level Street-level District-level Street-level District-level Street-level District-level Street-level landscape landscape landscape landscape landscape landscape landscape landscape (1) (2) (3) (4) (5) (6) (7) (8) Urban landscape index Alternative indicator Pole (Max) −0.7583* −0.3847 (0.4217) (0.4510) Sidewalk (Max) 0.1081 0.0442 (0.1977) (0.2030) Sidewalk + Car 0.5241 0.2099 (Mean) (0.3816) (0.3497) Terrain (Max) −0.0931 0.0017 (0.0954) (0.0906) Baseline indicator Vegetation (Mean) 0.5967*** 0.3453* 0.5783*** 0.3434* 0.5965*** 0.3532* 0.6010*** 0.3439* (0.1890) (0.2019) (0.1896) (0.2017) (0.1893) (0.2009) (0.1911) (0.2038) Sky (Mean) 0.9552*** 1.0531*** 0.9987*** 1.0584*** 0.9716*** 1.0473*** 0.9759*** 1.0524*** (0.2928) (0.2890) (0.2903) (0.2874) (0.2941) (0.2852) (0.2945) (0.2903) Building (Mean) 0.2565 0.5360*** 0.2601 0.5369*** 0.2413 0.5285*** 0.2530 0.5372*** (0.1747) (0.1980) (0.1751) (0.1987) (0.1757) (0.1992) (0.1746) (0.1981) Pole (Mean) −3.6663** −1.2998 −3.5458** −1.2658 −3.3911* −1.2803 (1.7483) (1.8453) (1.7694) (1.8492) (1.7998) (1.8701) Sidewalk (Mean) 0.7509 0.1687 0.7269 0.1860 (0.5234) (0.5776) (0.5198) (0.5726) Terrain (Mean) −0.2359 −0.0189 −0.3090 −0.0290 −0.2540 −0.0110 (0.2783) (0.2567) (0.2780) (0.2575) (0.2784) (0.2635) (Continued) JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 13 Table 6. (Continued). Building Yes Yes Yes Yes Yes Yes Yes Yes characteristics, Land characteristics, Land use within 500 m radius, Time fixed effect (quarterly) and a Constant District fixed effect Yes Yes Yes Yes Number of 800 800 800 800 800 800 800 800 observations R 0.705 0.801 0.704 0.801 0.705 0.801 0.705 0.801 Adjusted R 0.686 0.757 0.686 0.757 0.686 0.757 0.687 0.757 The dependent variable is the log of the transaction price. White-corrected robust standard errors are in parentheses. Significance level: *** p < 0.01, ** p < 0.05, * p < 0.1. coexist. In a “loosely regulated area,” residential landscape without controlling for the district fixed effect houses and commercial/industrial uses can coexist. (corresponding to column (4) of Table 4), while columns In this context, the economic value of landscape is (2), (4), (6) and (8) capture the street-level landscape by more likely to be observed in pure low-rise resi- controlling for the district fixed effect (corresponding to dential areas, and may be lost in more loosely column (6) of Table 4). We employ alternative indicators regulated areas where multiple types of land use one by one, and the other baseline indicators are coexist. included as control variables. For instance, when we employ the maximum proportion of vegetation (instead In columns (1) and (2), the entire area sample basi- of the mean proportion of vegetation), we include the cally confirms the same results as those yielded from mean proportions of sky, building, pole, sidewalk, and our main analysis using low-rise residential areas terrain. The results are basically consistent with those (Table 4), except for the loss of statistical significance presented in Table 4. for the vegetation component at street level. This may be due to the fact that the other residential area and Panel (a) presents the results regarding greenery, loosely regulated area subsamples do not show openness, and visual enclosure. In columns (1) and (2), a correlation between vegetation and property price we employ the maximum proportion of the vegetation (columns (3)–(6)). In other words, street-level greenery component, whose coefficient is positive with statistical is evaluated only within low-rise residential areas. Note significance at street level. Thus, we confirm the robust- that terrain exhibits a negative correlation with prop- ness on the fact that greenery is positively correlated erty price at district level. This may capture the strong with property price at street level. Columns (3) and (4) heterogeneity across districts when we employ the employ the maximum proportion of the sky compo- entire sample. That is, areas with farmland may have nent, whose coefficient is positive with statistical signifi - been facing housing development only recently, and cance at street level. Thus, we confirm the robustness of thus still retain low land prices. the fact that openness is positively correlated with prop- In the other residential areas (columns (3) and (4)) erty price at street level. Columns (5) and (6) employ the and in loosely regulated areas (columns (5) and (6)), we minimum proportion of the building component, do not observe a correlation between the urban land- whose coefficient is positive with statistical significance scape index and transaction price with statistical sig- at street level. Thus, we confirm the robustness of the nificance (except that terrain exhibits a negative fact that visual enclosure is positively correlated with correlation with property price at district level in property price at street level. Note, however, that the other residential areas, as in the entire area). This mean proportion of the sum of the three components implies that the urban landscape is basically reflected (building, wall, and fence) does not exhibit a correlation in transaction price only in low-rise residential areas, in with transaction price in columns (7) and (8). This implies which the landscape is valued by market participants. that the building component – but not the wall or fence In the remaining areas, other factors like transportation components – is correlated with property price. and shopping convenience matter, whereas the urban Panel (b) displays the results for power pole, road landscape does not influence the transaction price. shoulder, and farmland. In columns (1) and (2), we employ the maximum proportion of the pole compo- nent, whose coefficient is negative with statistical signifi - 6.3. Robustness checks cance only at district level. Thus, we confirm the robustness of the fact that the presence of a power Table 6 conducts robustness checks by using alternative pole is negatively correlated with property price at district indicators of the urban landscape index. The samples level. In columns (3) and (4) (columns (5) and (6)), we are properties in low-rise residential areas, as in Table 4. employ the maximum proportion of the sidewalk Columns (1), (3), (5) and (7) capture the district-level 14 M. SUZUKI ET AL. component (mean proportion of the sum of the sidewalk Our results also show that the semantic segmenta- and the car components). In neither regression are the tion approach, with high segmentation accuracy and coefficients statistically significant, confirming the robust- label diversity, enables researchers to quantify micro- ness of the fact that the presence of a road shoulder does level landscapes that have previously proven difficult not exhibit a negative correlation with property price at to obtain at large scale, and that this methodology is either district or street level. In columns (7) and (8), we also useful in analyzing urban landscapes specific to employ the maximum proportion of the terrain compo- Japan. The Hachioji city investigated in this paper is nent. The coefficients are not statistically significant, con- a typical low-rise residential area in the suburbs, and firming the robustness of the fact that the presence of the conclusions of this study are likely to be a common farmland does not exhibit a negative correlation with feature of at least the suburban areas of major metro- property price at either district or street level. politan areas. Although the methodology is easily applicable to other types of real estate in different regions, subsample 7. Conclusion analysis of land use type indicates that any correlation between landscape and property price may be hetero- While visual impression of urban landscape has geneous (e.g., central area and other suburban areas), been investigated in detail in the field of architec- and thus a functional form of hedonic regression should ture, an economic consideration of the urban land- be tailored to each local context. To fulfill this aim, auto- scape across space is also an important aspect to mating functional form to identify the elements that give an incentive to build and maintain a good formulate property prices (e.g., He, Páez, and Liu 2017; residential environment. Employing a novel seman- Law, Paige, and Russell 2019; Law et al. 2020) is worthy of tic segmentation approach using Google Street investigation. View images, we have investigated the relationship between urban landscapes and property prices in low-rise residential areas in a suburban city of Acknowledgments Tokyo, Japan. Such visual images have been used to represent the landscape of both surrounding We would like to thank two anonymous referees, Yasushi districts in general and the street-level landscape; Asami, Masayoshi Hayashi, Sachio Muto, Toshihiko more specifically, the latter has been derived after Yamasaki, Noriyuki Yanagawa, participants in the CREI work- controlling for the district fixed effect. We first shop at the University of Tokyo, and members of Japan Real Estate Institute for their insightful comments. We also thank showed that greenery, openness and visual enclo- the Real Estate Information Network System (REINS) and sure are positively correlated with property price at Joint Research Program No. 1075 at the Center for Spatial street level. We then investigated the value of com- Information Science, The University of Tokyo, for providing mon Japanese urban landscapes: (i) the presence of the data. a power pole is negatively correlated with property price at district level; and (ii) the presence of a road shoulder and farmland, either of which may disrupt Disclosure statement the continuity of a residential area, does not exhibit No potential conflict of interest was reported by the negative correlations with property price at either author(s). district or street level. Thus, the quality of surround- ing landscapes, as captured through Street View images, has a meaningful relationship with real Funding estate price formation. The authors gratefully acknowledge the support received Our results suggest that basic landscape elements, from JSPS KAKENHI Grant Numbers 20K14896, 20H00082, such as greenery, openness and visual enclosure, are and 17H00988, and a Google Cloud Platform Credit Coupon important in maintaining the value of residential areas; at The University of Tokyo. thus, protection of the residential environment through residential area planning and/or building agreement is meaningful. Furthermore, the value of residential areas Notes on contributors can be increased by promoting the undergrounding of Masatomo Suzuki is a specially appointed associate profes- power poles. Although road shoulder does not lead to sor at Center for the Promotion of Social Data Science a uniform decline in economic value, the value of the Education and Research, Hitotsubashi University. His landscape is potentially improved if it is well designed, research interests include housing, real estate, city planning, such as a parking lot covered with greenery. Since farm- and urban economics. land also does not lead to a uniform decline in economic Junichiro Mori is an associate professor at Graduate School of value, preservation of urban farmland within residential Information Science and Engineering, The University of area, in other words, coexistence of the two types of Tokyo. His research interests include artificial intelligence, land use, may be justified. massive data analysis, and social network analysis. JOURNAL OF ASIAN ARCHITECTURE AND BUILDING ENGINEERING 15 Takashi Nicholas Maeda is an associate professor at the Jim, C. Y., and W. Y. Chen. 2009. “Value of Scenic Views: School of System Design and Technology of Tokyo Denki Hedonic Assessment of Private Housing in Hong Kong.” University. His research interests include causal discovery Landscape and Urban Planning 91 (4): 226–234. and urban computing. doi:10.1016/j.landurbplan.2009.01.009. Johnson, E. B., A. Tidwell, and S. V. Villupuram. 2020. “Valuing Jun Ikeda is a former graduate student at Graduate School of Curb Appeal.” The Journal of Real Estate Finance and Information Science and Technology, The University of Economics 60 (1): 111–133. doi:10.1007/s11146-019-09713-z. Tokyo. Koura, H., and K. Kamino. 1995. “Study on Structure of Integrated Setting in Townscape of Urban Environment.” Journal of the City Planning Institute of Japan 30: 223–228. ORCID [in Japanese]. doi:10.11361/journalcpij.30.223. Koura, H., and K. Kamino. 1996. “Structure of Integrated Masatomo Suzuki http://orcid.org/0000-0002-2713-0695 Setting in Townscape as Pedestrian Environment.” Junichiro Mori http://orcid.org/0000-0002-9787-3857 Journal of Architecture and Planning (Transactions of AIJ) Takashi Nicholas Maeda http://orcid.org/0000-0003-2419- 484: 167–174. [in Japanese]. doi:10.3130/aija.61.167_3. Law, S., B. Paige, and C. Russell. 2019. “Take a Look Around: Using Street View and Satellite Images to Estimate House Prices.” ACM Transactions on Intelligent Systems and Technology (TIST) 10 (5): 1–19. References doi:10.1145/3342240. Biljecki, F., and K. Ito. 2021. “Street View Imagery in Urban Law, S., C. I. Seresinhe, Y. Shen, and M. Gutierrez-Roig. 2020. Analytics and GIS: A Review.” Landscape and Urban “Street-Frontage-Net: Urban Image Classification Using Planning 215: 104217. Deep Convolutional Neural Networks.” International Journal of Geographical Information Science 34 (4): Chen, L., X. Yao, Y. Liu, Y. Zhu, W. Chen, X. Zhao, and T. Chi. 2020. 681–707. doi:10.1080/13658816.2018.1555832. “Measuring Impacts of Urban Environmental Elements on Lindenthal, T., and E. B. Johnson. 2021. “Machine Learning, Housing Prices Based on Multisource Data: A Case Study of Architectural Styles and Property Values.” The Journal of Shanghai, China.” ISPRS International Journal of Geo- Real Estate Finance and Economics. forthcoming. Information 9 (2): 106. doi:10.3390/ijgi9020106. doi:10.1007/s11146-021-09845-1. Cordts, M., M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, McNair, B. J., J. Bennett, D. A. Hensher, and J. M. Rose. 2011. R. Benenson, U. Franke, S. Roth, and B. Schiele. 2016. “Households’ Willingness to Pay for Overhead-to- “The Cityscapes Dataset for Semantic Urban Scene underground Conversion of Electricity Distribution Understanding.” In Proceedings of the IEEE Conference Networks.” Energy Policy 39 (5): 2560–2567. doi:10.1016/j. on Computer Vision and Pattern Recognition (CVPR), Las enpol.2011.02.023. Vegas, NV, USA. 3213–3223. Neuhold, G., T. Ollmann, S. Rota Bulò, and P. Kontschieder. Fu, X., T. Jia, X. Zhang, S. Li, Y. Zhang, and S. Fu. 2019. “Do 2017. “The Mapillary Vistas Dataset for Semantic Street-Level Scene Perceptions Affect Housing Prices in Understanding of Street Scenes.” In Proceedings of the Chinese Megacities? An Analysis Using Open Access IEEE International Conference on Computer Vision (ICCV), Datasets and Deep Learning.” PloS ONE 14 (5): e0217505. Venice, Italy. 4990–4999. doi:10.1371/journal.pone.0217505. Nishio, S., and F. Ito. 2015. “The Effect of Sky Factor and Gao, X., and Y. Asami. 2001. “The External Effects of Local the Change on Impression of Townscape: With the Attributes on Living Environment in Detached Residential Experiment for Evaluation Impression while Walking.” Blocks in Tokyo.” Urban Studies 38 (3): 487–505. Journal of Architecture and Planning (Transactions of doi:10.1080/00420980120027465. AIJ) 710: 907–914. [in Japanese]. doi:10.3130/ Gao, X., and Y. Asami. 2007. “Effect of Urban Landscapes on aija.80.907. Land Prices in Two Japanese Cities.” Landscape and Urban Nishio, S., and F. Ito. 2020. “Statistical Validation of Utility of Planning 81 (1–2): 155–166. doi:10.1016/j. Head-mounted Display Projection-based Experimental landurbplan.2006.11.007. Impression Evaluation for Sequential Streetscapes.” He, L., A. Páez, and D. Liu. 2017. “Built Environment and Violent Environment and Planning B: Urban Analytics and City Crime: An Environmental Audit Approach Using Google Science 47 (7): 1167–1183. Street View.” Computers, Environment and Urban Systems Oku, T. 1985. “The Influence of Townscape Elements on the 66: 83–95. doi:10.1016/j.compenvurbsys.2017.08.001. Estimation of Townscape: Experimental Study on Visual Hirate, K., and M. Yasuoka. 1986. “A Study on the Evaluation of Street-Vistas with Roadside Trees: The Experiments with Characteristics and Feelings of Townscape Part 2.” Journal Using Black and White Photomontage Slidefilms.” Journal of Architecture, Planning and Environmental Engineering of Architecture, Planning and Environmental Engineering (Transactions of AIJ) 351: 27–37. [in Japanese]. (Transactions of AIJ) 362: 35–43. [in Japanese]. doi:10.3130/aijax.351.0_27. doi:10.3130/aijax.362.0_35. Shimizu, C., K. G. Nishimura, and T. Watanabe. 2016. “House Ishikawa, T., A. Okabe, Y. Sadahiro, and S. Kakumoto. 1995. Prices at Different Stages of the Buying/Selling Process.” “A Study on the Cognition of Open Space Using 3-D Stereo Regional Science and Urban Economics 59: 37–53. doi:10.1016/j.regsciurbeco.2016.04.001. GIS.” Journal of Architecture and Planning (Transactions of Takei, M., and T. Fukushima. 1983. “A Study on Effect of Tree AIJ) 475: 149–154. [in Japanese]. doi:10.3130/aija.60.149_5. Planning Which Mitigates the Sense of Oppression by Ito, K., and F. Biljecki. 2021. “Assessing Bikeability with Street A Building: A Study on Visual Environmental Planning of View Imagery and Computer Vision.” Transportation Urban District.” Journal of Architecture and Planning Research Part C: Emerging Technologies 132: 103371. (Transactions of AIJ) 332: 102–110. [in Japanese]. doi:10.1016/j.trc.2021.103371. 16 M. SUZUKI ET AL. Tang, J., and Y. Long. 2019. “Measuring Visual Quality of Ye, Y., H. Xie, J. Fang, H. Jiang, and D. Wang. 2019a. “Daily Street Space and Its Temporal Variation: Methodology Accessed Street Greenery and Housing Price: Measuring and Its Application in the Hutong Area in Beijing.” Economic Performance of Human-scale Streetscapes via Landscape and Urban Planning 191: 103436. doi:10.1016/ New Urban Data.” Sustainability 11 (6): 1741. doi:10.3390/ j.landurbplan.2018.09.015. su11061741. Tao, A., K. Sapra, and B. Catanzaro. 2020. “Hierarchical Ye, Y., W. Zeng, Q. Shen, X. Zhang, and Y. Lu. 2019b. Multi-scale Attention for Semantic Segmentation.” arXiv “The Visual Quality of Streets: A Human-centred preprint arXiv:2005.10821. Continuous Measurement Based on Machine Learning Tempesta, T., D. Vecchiato, and P. Girardi. 2014. “The Landscape Algorithms and Street View Images.” Environment and Benefits of the Burial of High Voltage Power Lines: A Study in Planning B: Urban Analytics and City Science 46 (8): Rural Areas of Italy.” Landscape and Urban Planning 126: 1439–1457. 53–64. doi:10.1016/j.landurbplan.2014.03.003. Yokohari, M., K. Takeuchi, T. Watanabe, and S. Yokota. 2000. Yamagata, Y., D. Murakami, T. Yoshida, H. Seya, and S. Kuroda. “Beyond Greenbelts and Zoning: A New Planning Concept 2016. “Value of Urban Views in a Bay City: Hedonic Analysis for the Environment of Asian Mega-Cities.” Landscape and with the Spatial Multilevel Additive Regression (SMAR) Urban Planning 47 (3–4): 159–171. doi:10.1016/S0169- Model.” Landscape and Urban Planning 151: 89–102. 2046(99)00084-5. doi:10.1016/j.landurbplan.2016.02.008. Zhang, Y., and R. Dong. 2018. “Impacts of Street-visible Yang, J., H. Rong, Y. Kang, F. Zhang, and A. Chegut. 2021. “The Greenery on Housing Prices: Evidence from a Hedonic Financial Impact of Street-level Greenery on New York Price Model and a Massive Street View Image Dataset in Commercial Buildings.” Landscape and Urban Planning Beijing.” ISPRS International Journal of Geo-Information 214: 104162. doi:10.1016/j.landurbplan.2021.104162. 7 (3): 104. doi:10.3390/ijgi7030104.

Journal

Journal of Asian Architecture and Building Engineering – Taylor & Francis

Published: May 4, 2023

Keywords: Hedonic price model; landscape; machine learning; Google Street View images; Japan

There are no references for this article.