大数据与城市规划 (11).pdf

《大数据与城市规划 (11).pdf》由会员分享，可在线阅读，更多相关《大数据与城市规划 (11).pdf（18页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、RESEARCH ARTICLEHow green are the streets?An analysis forcentral areas of Chinese cities using TencentStreet ViewYing Long1*,Liu Liu21 School of Architecture and Hang Lung Center for Real Estate,Tsinghua University,Beijing,China,2 ChinaAcademy of Urban Planning and Design,Shanghai,China*AbstractExte

2、nsive evidence has revealed that street greenery,as a quality-of-life component,isimportant for oxygen production,pollutant absorption,and urban heat island effect mitiga-tion.Determining how green our streets are has always been difficult given the time andmoney consumed using conventional methods.

3、This study proposes an automatic methodusing an emerging online street-view service to address this issue.This method was used toanalyze street greenery in the central areas(28.3 km2each)of 245 major Chinese cities;this differs from previous studies,which have investigated small areas in a given cit

4、y.Sucha city-system-level study enabled us to detect potential universal laws governing streetgreenery as well as the impact factors.We collected over one million Tencent Street Viewpictures and calculated the green view index for each picture.We found the following rules:(1)longer streets in more e

5、conomically developed and highly administrated cities tended tobe greener;(2)cities in western China tend to have greener streets;and(3)the aggregatedgreen view indices at the municipal level match with the approved National Garden Cities ofChina.These findings can prove useful for drafting more app

6、ropriate policies regarding plan-ning and engineering practices for street greenery.1.IntroductionAs one of the most prominent colors in nature,green has been an everlasting beloved color ofhumans,and the“garden city”advocated by 1 is among the most famous planning theories.According to 2,green spac

7、es offer significant potential for restoration,correspond to theinnate human tendency to focus on life and lifelike processes,and promote behaviors thatboost well-being;thus,increasing the provision and utilization of urban green spaces can pro-mote stress reduction,happiness,health,and well-being a

8、mong humans.As an essential aspectof green-city implementation,green coverage at various scalesat the block level(green landarea divided by block area),for example,or citywide(total green land area divided by the citysurban land area)is a mandatory element of spatial plans for promoting a high quali

9、ty of life.As a result of partial planning implementation and the diverse composition of green spaces,PLOS ONE|DOI:10.1371/journal.pone.0171110February 14,20171/18a1111111111a1111111111a1111111111a1111111111a1111111111OPENACCESSCitation:LongY,LiuL(2017)Howgreenarethestreets?Ananalysisforcentralareas

10、ofChinesecitiesusingTencentStreetView.PLoSONE12(2):e0171110.doi:10.1371/journal.pone.0171110Editor:XiaoleiMa,BeihangUniversity,CHINAReceived:July10,2016Accepted:January15,2017Published:February14,2017Copyright:2017Long,Liu.ThisisanopenaccessarticledistributedunderthetermsoftheCreativeCommonsAttribut

11、ionLicense,whichpermitsunrestricteduse,distribution,andreproductioninanymedium,providedtheoriginalauthorandsourcearecredited.DataAvailabilityStatement:Allrelevantdataarewithinthepaper.Funding:ThefirstauthorwouldliketoacknowledgethefundingoftheNationalNaturalScienceFoundationofChina(No.51408039).Comp

12、etinginterests:Theauthorshavedeclaredthatnocompetinginterestsexist.green coverage in planning drawings does not directly correspond to the total greenery in real-ity.This is one reason why visual greenery has been extensively discussed in the research com-munity and is suggested for use in practice.

13、Although not required in spatial plans,streetgreeneryas the focus of this study and a key indicator for evaluating urban form at the city-design levelis important for citizens quality of life(especially for pedestrians in daily life);however,this has not been sufficiently studied due to a lack of fi

14、ne-scale data.Understanding how green our streets are has never been easy.Using the conventionalmethods,it is generally time consuming and expensive.To address this issue,we developed anautomatic method using a street-view service while also borrowing and modifying ideas fromexisting studies such as

15、 35.The green color ratio in street views(termed“green view index”in this paper)which reflects objective city(as well as rural in most street-view products)street(and road)landscapeswas selected as the proxy for linking with street greenery in thisstudy,which falls under the umbrella of visual green

16、ery studies.Different from online geo-tagged photos,which reflect city images captured subjectively by photographers,street viewobjectively depicts the true urban landscape.This is another reason why we chose street viewto understand street visual greenery in this study.Today in China,academic studi

17、es areincreasingly using open data from social networks,commercial websites,and official channelsto understand city systems and urban structure,as well as human mobility and activity(see 6for a review).To the best of our knowledge,this is one of the first studies to analyze streetgreenery in a large

18、 number of cities using street view.This paper is organized as follows:To illustrate the research context,section 2 reviewsrelated areas such as visual greenery and using street-view pictures for urban studies.Sections3 and 4 introduce the study area,data,and research methods.Section 5 presents the

19、researchresults in various aspectssuch as the overall pattern,intercity rankings and analysis,andintracity pattern analysisas well as the validation of the results.In the final section,we dis-cuss potential applications,academic contribution,research biases,and future plans.2.Literature review2.1 Us

20、ing street-view pictures in urban studiesSystems like Google Street View and Bing Maps Streetside enable users to virtually visit cities(on the streets or even indoors)by navigating immersive 360 panoramas.There are variousendeavors related to Google Street View(GSV)image recognition,including 3-D c

21、ity modelconstruction 7,commercial-entity identification 8,real-time text localization and recogni-tion 9,and layer interpretation for ground,pedestrians,vehicles,buildings,and sky 10.In addition to these existing studies in the field of computer science,there are related stud-ies in urban geography

22、,regional science,urban studies,and urban planning.Rundle et al.11suggest that GSV can be used to audit neighborhood environments by checking the concor-dance between GSV analysis and field surveys.Odgers et al.12 observed childrens neighbor-hoods using GSV and found it to be a reliable and cost-eff

23、ective tool.Kelly et al.13 used GSVto audit built environments and also found it to be a reliable method.Hwang and Sampson14 identified visible clues of neighborhood gentrification using GSV for systematic socialobservation.Carrasco-Hernandez 15 reconstructed building geometries and urban sky viewfa

24、ctors using the GSV image database.In general,street view has proven to be an effective andreliable tool for measuring built environments on various scales,such as streets and neighbor-hoods.The aforementioned studies were all conducted manually by looking at street-viewimages,not by automatic means

25、.This time-consuming process places constraints on usingstreet view to analyze large geographical areas.We did find an investigation 16 thatHow green are the streets in Chinese cities?PLOS ONE|DOI:10.1371/journal.pone.0171110February 14,20172/18combined crowdsourcing techniques with GSV to identify

26、street-level accessibility problems,but this still relied heavily on manual human effort.Based on our review of the use of street view in two general fields(computer science andurban studies),we found the following mismatch.Computer scientists have been developingadvanced image recognition algorithm

27、s to automatically identify specific objects,texts,or pat-terns from street view.Urban scientists,however,have employed street view manually,withoutdrawing on the latest progress made by computer scientists.Such time-consuming techniquesare not easy for urban scientists to overcome.The second author

28、 of this paper has proposed asolution that involves automatic cognitive city mapping using geotagged photos(not street-view pictures),drawing on Kevin Lynchs The Image of the City 1718.The present study aimsto further explore using street-view pictures to automatically and exhaustively analyze/visua

29、lizestreet greenery in our cities,and thus contribute to building a science of cities 19.2.2 Understanding visual greeneryThe effort to bring natural greenery into urban environments has a long history.In the 1850s,Olmsted focused on urban park reform and street design,trying to combine natural envi

30、ron-ments with urban living spaces 20.The greenway movement in the late 1980s was a large-scale concept that proposed creating a green network to give people access to open spacesclose to where they live and to link rural and urban spaces in the American landscape 21.Such urban“green constructions”a

31、re mostly valued for their economic or environmental ben-efits.A study of the cooling effects of street greenery at 11 urban sites in Tel-Aviv showed thatthe shaded area under a canopy plays a key role in alleviating the“heat island”effect 22.Another important contribution of street-level vegetation

32、 is that it improves air quality alongstreet canyons,which has been studied by many researchers like 23.However,aspects of the visual effect or aesthetic amenity of greenery have received lessattention.The ratio of greenery as a measurement of the visibility of street greenery,first pro-posed by 24,

33、calibrates the effective ratio for a variety of landscape scenes via different focaldistances.Ohno 25 further studied the measurement of ambient visual information.He sug-gested that while visible greenery can soften the negative impression of an“artificial”environ-ment,the positive impression of a“

34、natural”environment is not enhanced when the ratioexceeds 15%.In a later study,Ohno 26 emphasized auto-centricity in vision,which relates topeoples feelings and their experience of pleasure.Nasar 27 investigated visual environmentpreferences for urban street scenes among 46 students from Japan and t

35、he US.The preferencescores showed that students from both countries preferred foreign scenes to native ones andthat preference was associated with the prominence of nature,and the absence of vehicles.Photos have been regarded as promising data sources for measuring visual greenery.Threemethods shoul

36、d be noted.First,Yang et al.3 developed the green view index to evaluate thevisibility of urban forests through a combination of field surveys and manual photographinterpretation.They found a strong correlation between the green view index and the canopycover of trees/shrubs.This method relies heavi

37、ly on human labor.Second,Google Street Viewprovides opportunities for automatically gathering a large number of street pictures.Li et al.4 proposed an automatic framework for assessing street-level urban greenery using GSV andemployed it to study the East Village in Manhattan.The quality of this met

38、hodology has beenvalidated through manually derived street greenery in Photoshop.The identification methodfor street greenery is pixel-based color recognition.Li et al.5 further analyzed the relation-ship between automatically calculated street greenery using GSV and residents socioeconomiccharacter

39、istics in Hartford,Connecticut.Li et al.28 also explored environmental inequitiesamong different types of urban greenery;since automatically recognizing street greenery usingHow green are the streets in Chinese cities?PLOS ONE|DOI:10.1371/journal.pone.0171110February 14,20173/18street-view pictures

40、is time consuming,the study area was small.Third,another method forunderstanding visual greenery is pattern recognition using semantic pixel-wise segmentationand scene labeling.Group labeling based on a combination of RGB-and depth-based cuescould obtain higher accuracy,according to several experime

41、ntal studies.Farabet et al.29achieved real-time scene parsing and image labeling to best explain the scene.Gupta et al.30recognized indoor scenes using boundary detection and hierarchical grouping.By addingglobal appearance features,Gatta and Romero 31 conducted spatial scene parsing.Badrinar-ayanan

42、 et al.32 developed SegNet for more accurate outdoor and indoor element recogni-tion,by which 11 classes,including trees,were labeled.Compared to the pixel-based color-recognition approach,this approach requires more time and techniques.Based on our review of these three methods,we used the approach

43、 developed by Li,withmodifications,to study a large number of cities in China,aiming to gain a holistic understand-ing of street greenery at the city-system level.A detailed comparison of Lis method and ours isdiscussed in Section 6.3.3.Study area and data3.1 Study areaIn China,there are 288 cities

44、at or above the prefecture level.There are 5 administrative levelsin China,and county-level cities were not included in this study.Since 43 of the cities did nothave street-view data,our study included 245 cities.Among these,there are 4 municipalitiesdirectly governed by the central government,15 su

45、bprovincial cities,17 other provincial capi-tal cities,and 209 prefecture-level cities(Fig 1).Considering the crawling,processing,andcomputation load of street-view pictures(SVPs),we only focused on the downtown areas ofeach city.Since the downtown or live-work-play center boundaries are not availab

46、le for mostChinese cities,we buffered each city center(namely,the central business district CBD)witha radius of 3 km to derive the study areas.These are regarded as proxies for Chinas CBDs andhave a total area of 8,143 km2.We acknowledge potential bias in this translation process andwill address it

47、in future studies.3.2 Chinas roads/streets in 2014We used the 2014 roads/streets of China obtained from a local road-navigation firm based inBeijing for crawling SVPs.Almost all detailed road/street networks at various levels,includingstreets and regional roads,are included in this dataset as of the

48、 end of 2014,according to com-parisons with Google Maps and Baidu Maps(a main online map service provider and popularsearch engine in China;http:/).The total road/street length is 4,249,419 km for11,562,709 street segments(367.5 m in average).Note that a street segment denotes a part or allof a stre

49、et divided by intersections.We clipped the national dataset with the study area(245 poly-gons in a circle)and derived the street segments within the study area,including 748,430 streetsegments with a total length of 66,041 km(88.2 m in average).As shown in Fig 2,some of theraw streets are dual lines

50、 or bridges,which made it necessary to preprocess the data for crawlingSVPs.We also removed highways and bridges to avoid the street-level images captured in them,which are not suitable for studying street greenery.After the data preprocessing,there were372,186 street segments with a total length of