3D city models for urban farming site identification in buildings
Ankit Palliwal, Shuang Song, Hugh Tiang Wah Tan, Filip Biljecki
33D city models for urban farming site identification in buildings
Ankit Palliwal a , Shuang Song b , Hugh Tiang Wah Tan b and Filip Biljecki c,d , ∗ a Department of Geography, National University of Singapore, Singapore b Department of Biological Sciences, National University of Singapore, Singapore c Department of Architecture, National University of Singapore, Singapore d Department of Real Estate, National University of Singapore, Singapore
A R T I C L E I N F O
Keywords :3D GISfood securityOpenStreetMapsolar exposuretropical climateVI-Suite
A B S T R A C T
Studies have suggested that there is farming potential in urban residential buildings. However, thesestudies are limited in scope, require field visits and time-consuming measurements. Furthermore, theyhave not suggested ways to identify suitable sites on a larger scale let alone means of surveying numer-ous micro-locations across the same building. Using a case study area focused on high-rise buildingsin Singapore, this paper examines a novel application of three-dimensional (3D) city models to iden-tify suitable farming micro-locations (level and orientation) in residential buildings. We specificallyinvestigate whether the vertical spaces of these buildings comprising outdoor corridors, façades andwindows receive sufficient photosynthetically active radiation (PAR) for growing food crops and doso at a high resolution. We also analyze the spatio-temporal characteristics of PAR, and the impact ofshadows and different weather conditions on PAR in the building. Environmental simulations on the3D model of the study area indicated that the cumulative daily PAR or Daily Light Integral (DLI) at alocation in the building was dependent on its orientation and shape, sun’s diurnal and annual motion,weather conditions, and shadowing effects of the building’s own façades and surrounding buildings.The DLI in the study area generally increased with building’s levels and, depending on the particularmicro-location, was found suitable for growing moderately light-demanding crops such as lettuce andsweet pepper. These variations in DLI at different locations of the same building affirmed the need forsuch simulations. The simulations were validated with field measurements of PAR, and correlationcoefficients between them exceeded 0.5 in most cases thus, making a case that 3D city models offera promising practical solution to identifying suitable farming locations in residential buildings, andhave the potential for urban-scale applications.
1. Introduction
Over the years, farming in and around urban buildings,particularly residential buildings, has gained popularity inhigh-density and high-rise environments (Lim and Kishnani,2010; Khan et al., 2018; Kim et al., 2018; Song et al., 2018;Kosorić et al., 2019). This is primarily because with limitedland available for agriculture, these buildings offer under-utilized horizontal and vertical spaces that may have farmingpotential (Figure 1). In addition, improvement of emotional,mental and physical well-being of the occupants (Tan and Is-mail, 2015), mitigation of the urban heat island effect (Diehlet al., 2020), creation of job opportunities (Tablada andZhao, 2016), and reduction in carbon emissions associatedwith transportation of food (Lim and Kishnani, 2010) arecounted among the other benefits of farming in these build-ings. In Singapore, urban farming carries special signifi-cance as it has been adopted as one of the ‘Grow Local’strategies to achieve the ‘30 by 30’ vision of the SingaporeFood Agency (SFA) (Zulkifli, 2019). This vision aims to lo-cally produce 30% of Singapore’s nutritional needs by 2030.Situated at 1 ◦ North of equator, Singapore is an arableland-scarce and densely populated island city-state whichaccommodates a population of about 5.7 million (Singa- ∗ Corresponding author [email protected] (F. Biljecki)
ORCID (s): (A. Palliwal); (S. Song); (H.T.W. Tan); (F.Biljecki) pore Department of Statistics, 2019a) over its land areaof 722.5km (Singapore Department of Statistics, 2019b).Having only 1% of land set aside for agriculture (Diehlet al., 2020), Singapore meets 90% of its food requirementsthrough imports (Kosorić et al., 2019) leaving it vulnerableto external food price fluctuations and disruptions in the foodsupply chain. To reduce this heavy reliance on food importsand inline with SFA’s vision, Singapore’s numerous high-rise residential buildings, which accommodate almost its en-tire population, emerge as promising sites for urban farm-ing. These buildings not only offer under-utilized spaces butalso provide an opportunity to its occupants, who are consid-ered a key stakeholder to drive urban agriculture in Singa-pore, the ease of engaging in farming while at home (Kosorićet al., 2019). The vast majority of these buildings are publichousing buildings, whose number surpassed 10 thousandsand accommodate more than 80% of the nation (Housing& Development Board, 2020; Kosorić et al., 2019). Theyhave standardised designs and due to their public nature of-fer prospects for government-supported initiatives.Besides the optimal crop growth conditions such as thelevel of carbon dioxide, nutrients, humidity, temperature,water, among others – which are fairly constant at the build-ing scale – successful identification of soil-based farmingmicro-locations (particular site at a specific level) in a build-ing requires an assessment of its exposure to photosynthet-ically active radiation (PAR). PAR is the portion of solarspectrum, in the 400 to 700 nm wavelength range, that isutilized by plants for photosynthesis, and its amount is a key Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 1 of 17 a r X i v : . [ c s . OH ] J u l D city models for urban farming site identification in buildings (a) Corridors (b) Façades(c) Rooftops (d) Staircases(e) Void decks (f) Window ledges
Figure 1:
Open spaces in urban and high-rise residential buildings that may have farmingpotential. In this paper, we mainly focused on outdoor vertical spaces (Figures 1a, 1b,and 1f) since they are the sites that are most suitable for use in our study area (e.g. accessto rooftops is usually restricted or reserved for other purposes), but our methodologyis sufficiently generic that it can be applied to other parts of buildings and in differentgeographic locations. factor to understand whether there is a potential for farmingand what kind of crops can be grown at a specific site be-cause different crops require different PAR conditions for itsoptimal growth (Song et al., 2018). Conventionally, PAR as-sessment in different urban forms is accomplished throughfield surveys that involves placing PAR sensors at selectedlocations such as a few spots in a building (Tan and Ismail,2014, 2015; Song et al., 2018). The findings of these sur-veys suggest that, in a high density urban environment, dif- ferent urban forms are exposed to different levels of PARdue to their shape, orientation, self-shadowing effects, andshadowing effects of surrounding objects (Tan and Ismail,2014). Furthermore, PAR at a given micro-location alsovaries due to changes in the sun’s position in the sky and dif-ferent weather conditions (Song et al., 2018), thus, highlight-ing the need to understand the spatio-temporal characteris-tics of PAR at the building’s micro-locations . While thesestudies confirm that there is farming potential in residential
Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 2 of 17D city models for urban farming site identification in buildings buildings, they are: (i) constrained – only a limited num-ber of locations can be covered with sensors, and as this pa-per will show, there might be a large variation of PAR evenat the same side of a building between different levels; and(ii) arduous – as the surveys require field visits and the sen-sors have to be installed for an extended time period suchas weeks rather than enabling instantaneous measurements.Furthermore, they have not suggested ways on how to pos-sibly go about estimating the potential at the urban scale.This paper investigates whether three-dimensional (3D)city models can be used to assess the suitability of partic-ular micro-locations in high-rise buildings for urban farm-ing, leading to bypassing building visits and measurementswhile taking into account the peculiarities associated withsunlight availability in built environments, and calculatingthe potential of unlocking under-utilized spaces in residen-tial buildings for urban farming. Our work capitalizes onthe rich body of knowledge on using 3D city models for un-derstanding the benefit of installing solar panels in buildings(Section 2) and adapts the work to enable simulations suitedfor gathering the potential of urban farming (Section 3).The study has run environmental simulations to analyze thespatio-temporal characteristics of PAR, assess adequacy ofPAR received for growing crops, and understand the influ-ence of different weather conditions, self-shadowing, andshadowing effects of nearby urban forms. Unlike the vastmajority of papers dealing with simulations in 3D GIS, weconduct field measurements to verify the veracity of the sim-ulations and conclude that 3D city models are a viable in-strument for calculating the potential of spaces in buildingsfor urban farming (Section 4). To the best of our knowl-edge, 3D city models have not been used for this purposebefore. Our results are important because conducting fieldvisits and undertaking PAR measurements to identify all lo-cations in these buildings that receive adequate sunlight forgrowing crops can prove to be a difficult task, and the workcan lead to estimations of the urban farming potential at theprecinct or at the urban scale, enabling future studies consid-ering thousands of buildings at once, similar to other appli-cations in (3D) GIS. In this complete process of employing3D city models from acquisition all the way to analysis andextraction of insights, this paper also presents an alternativemethod of estimating building heights in the absence of con-ventional data by measuring staircases, which has not beendocumented in the existing academic literature.
2. Literature review
Geospatial technologies have been applied for longacross diverse agricultural and allied activities such as pre-cision farming (Wilson, 2005), assessment of land suitabil-ity for agriculture (Bandyopadhyay et al., 2009), suitabil-ity analysis for beekeeping sites (Estoque and Murayama,2011), and more recently quantification of potential greencover on rooftops (Santos et al., 2016). However, hitherto3D geoinformation has not been used to identify urban farm-ing sites despite their wide usage to assess the availability of solar energy in built environments for installing photo-voltaic panels on buildings (Redweik et al., 2013; Catitaet al., 2014; Freitas et al., 2015; Martínez-Rubio et al., 2016;Saretta et al., 2020). These simulations, which are primar-ily focused on rooftops of buildings, are used to determinewhether a particular part of a building/rooftop receives suf-ficient solar exposure to warrant the installation of a solarpanel. The amount of solar exposure is primarily influencedby the geographic location, orientation, and nearby objectsthat cause (self-)shadowing. The fact that 3D city modelsprovide sufficient information for such simulations catalyzedthe development of this long-standing research line.Since urban farming much depends on the available levelof light, which directly dictates the suitability of particulartypes of crops and influences the agricultural yield of crops(as much as it drives the energy yield of solar panels), ourwork takes advantage of the developments in the energy de-partment, and seeks into leveraging them for a different pur-pose essentially establishing a new research line marryingurban agriculture and 3D GIS.Given that an integral component of our work is gen-erating a 3D model of the study area, it is worthwhile toprovide a short literature review of the process. Most 3Dcity models are generated by extrusion, combining build-ing footprints and data on building heights, usually ob-tained from lidar point clouds (Dukai et al., 2019). Thisprocess results in block building models (or LOD1 as perCityGML/CityJSON (Gröger and Plümer, 2012; Ledouxet al., 2019)), which despite their coarse nature have provenuseful in scores of simulations such as predictions of the im-pact of noise in the built enviroment (Stoter et al., 2020).However, point clouds, a reliable but expensive sourceof building heights, are often unavailable, as it is the casefor our study area. To counter this gap, alternative methodsof deriving building heights in absence of direct elevationmeasurements have emerged. Biljecki et al. (2017) reviewseveral of them concluding that the most common unortho-dox approach is using the number of levels of a buildingas a proxy for its height, which continues to be engaged inmany studies, such as for energy simulations (Cheng et al.,2020). Another method, recently published in this journal,demonstrates that heights can be estimated from a singlephotograph captured through smartphones (Bshouty et al.,2020). While for our study area we have at disposal a 3D citymodel generated using open data on building levels, whichis reasonably accurate, we take advantage of the fact that allbuildings there are publicly accessible (being public hous-ing buildings) and count the number of stairsteps across theirvertical extent. As trivial as this approach appears, we be-lieve that it is powerful, presenting another contribution ofours, which may warrant attention for future investigations,especially in the context of crowdsourcing building heights.
Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 3 of 17D city models for urban farming site identification in buildings
3. Materials and methods
The study area consisted of the buildings situated on Ju-rong West Street 65 in Singapore (Figure 2). Jurong West isa residential town in the West region of Singapore. For theanalyses, we focused on the façades having corridors andwindow ledges of public housing building ‘Block 633’. Ac-cording to Kosorić et al. (2019), these micro-locations arethe most preferred for farming among the occupants, pre-senting an appropriate focus. The construction of ‘Block633’ was completed in 2000. This residential building has138 dwelling units spread over 16 levels (Housing & Devel-opment Board, 2020), and it shares the same design as manyother public housing buildings across Singapore, renderingour study generic and not constrained to a particular build-ing. Three of its sides have other residential buildings anda multi-storey car park adjoining them. The fourth side hasa school and another residential building across the street(Figure 3). The dense built form inevitably results in shad-owing, compounding the uncertainty of the amount of solarexposure required for urban farming and cultivating partic-ular crops.
Figure 2:
Study area: Block 633, Jurong West Street 65,Singapore.
A 3D city model of the study area is available as opendata (Biljecki, 2020). This dataset has been generated com-bining building footprints available in OpenStreetMap withthe number of levels released by the public housing agency.In order to double down on the accuracy of the data, we haveinvestigated whether there are alternative approaches to es-timate the building heights, which would be somewhat moreaccurate than using the number of levels as a proxy. To esti-mate building heights, the number of stairsteps from groundlevel/floor to the top level were counted, for each buildingaround the block in focus. The floor-to-floor height was ob-tained by multiplying the stairstep count between two con-secutive levels with the measured stairstep height. The de-rived floor-to-floor height was in consonance with the typical level one and floor-to-floor heights of 3.6m and 2.8m respec-tively in these buildings (Housing & Development Board,2014). Finally, the building height was estimated by sum-ming the floor-to-floor heights across all levels in the build-ing (Housing & Development Board, 2020). The generated3D city model of the study area is shown in Figure 4.
The methodology consists of conducting solar expo-sure estimations adapted for urban agriculture and verticalspaces, and carrying out conventional measurements to ver-ify the results.For the simulations, Blender v2.79a and VI-Suite v0.4have been used. VI-Suite is a free open source add-on pack-age for Blender, a popular 3D computer graphics software.It consists of building environment performance simulationmodules that allow 3D geospatial data analysis and visual-ization (Southall and Biljecki, 2017). It has the ability to (1)process large 3D geospatial datasets through a user-friendlyinterface, (2) integrate customized Python scripts, and (3)export the simulation results in comma separated values for-mat for subsequent analysis; which made it suitable for thisresearch and which we have done transporting the results toR v3.6.3. It is important to note that these tools are free andopen-source, so together with the open data (Section 3.2), itmeans that our work relies entirely on open sources, facili-tating its reproducibility.This study has used VI-Suite modules for sun path,shadow map, and lighting analysis. While sun path analysisdisplays the sun’s position and its trajectory relative to the3D model at any date, time and location; shadow mapping,on the other hand, calculates the percentage of time of thesimulation period a location was exposed to direct sunlighton a sunny day. For these analyses, VI-Suite uses some ofthe in-built functionalities of Blender (Southall and Biljecki,2017). With lighting analysis, it is possible to calculate theirradiance values at discrete moments in time (referred to asbasic lighting in VI-Suite) as well as the cumulative solar ra-diation received at a location over the simulation period (alsoknown as Climate Based Daylight Modelling (CBDM)). Forlighting analysis, VI-Suite uses Radiance lighting simula-tion suite in the background. Radiance is based on a back-ward ray-tracing daylight simulation method and is consid-ered among the best freeware available for daylighting anal-ysis (Ward, 1994; Freitas et al., 2015).This research mainly focused on environmental simula-tions during three periods: (I) 02 Mar 2020 6am – 11 Mar2020 6am, (II) 15 Mar 2020 6am – 25 Mar 2020 6am, and(III) 27 Mar 2020 6am – 06 Apr 2020 6am, accompaniedby PAR surveys conducted in the study area. InstantaneousPAR was measured at several locations along the corridors ofthe residential building (I, II) and one of the window ledgesof a dwelling unit (III). PAR was measured using Onset PARSmart Sensor (S-LIA-M003) with a sampling interval of onesecond and mean values logged at five-minute intervals inHOBO Micro Station (H21-USB and H21-002) data logger.Prior to conducting survey, these sensors were calibrated
Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 4 of 17D city models for urban farming site identification in buildings
Figure 3:
Buildings surrounding the public housing block in focus (Google Earth - We-bAssembly, 2020a,b).
Figure 4:
3D model of the study area, generated from opendata and using open-source software. against an Odyssey PAR logger according to the manufac-turer’s manual . These sensors were placed on various levelsof the building along the corridor railings of façades A, B,and C and on one of the window ledges of façade W (Figures5a, 5b, and 5c). Sky conditions/daily weather forecasts werealso monitored during these periods (The Weather Channel;Weather).The precise location of each sensor placed in the buildingwas determined using a measuring tape. To match these lo-cations in the 3D model, each façade of the 3D model in Fig-ure 5c was converted into a 16 ×
16 grid (Figure 5e ). Match-ing the positions of sensors to the corresponding grid cellsin the 3D model is essential to enable comparisons. Heightand width of the façades in the 3D model were determinedfrom the local coordinates of their vertices. Based on pro-portionality and determined height and width of the façades,each sensor location in the building (Figure 5d) was mappedto the corresponding grid cell in the 3D model (Figure 5e).To illustrate, during survey period I, sensors were placed ingrid cells 6 and 12 at levels 2, 5, 8, 12, and 16 of façade B. The respective calibration equations for the sensors along with coef-ficients of determination (R ) can be found in Table 1 in Appendix A. The numbers 1-16 on the vertical and horizontal axes in this figurerepresent the 𝑖 𝑡ℎ level and grid cell respectively. Running simulations in VI-Suite requires one or more ofthe following as inputs: (1) latitude/longitude informationof the study area, (2) weather data of the study area in En-ergyPlus Weather (EPW) format, (3) sky/weather condition(called skytype in VI-Suite) namely, sunny, partly cloudy,and cloudy during simulation period, and (4) start date/hourand end date/hour of the simulation period. This study usedthe weather data of Singapore available on EnergyPlus web-site (American Society of Heating, Refrigerating and Air-Conditioning Engineers, 2001) which has also been used inTan and Ismail (2014, 2015). Skytype during survey periods(Figure 6) were based on National Environment Agency’s24-hr weather forecast for the West region (Weather). Inthese forecasts, ’Fair’ and ’Fair & Warm’ sky conditionswere classified as ’Sunny’ and ’Showers’ and ’ThunderyShowers’ sky conditions were classified as ’Cloudy’.Start date/hour and end date/hour of the simulations weredecided based on the objectives of this study. Sun-pathswere generated for the first (02 Mar) and the last day (05Apr) of the PAR survey. Rendered images from solar illumi-nation at 10am and 1pm were also produced for these days toshow the movements of shadows casted by buildings duringand between these days. Shadow maps of façades were gen-erated for a sunny day (17 Mar) of the survey period. Mapswere produced at discrete moments in time (10am, 1pm, and4pm) and for different time periods (7am–1pm, 1pm–7pm,and 7am–7pm) to highlight the shadowing effects during theday. PAR simulations were carried out, using basic light-ing analysis, at 10am, 1pm, and 4pm on a sunny (17 Mar),partly cloudy (09 Mar), and cloudy (03 Apr) day to analyzethe spatio-temporal distribution of PAR on the façades underdifferent weather conditions. PAR simulations without theground plane were also carried out for these time instanceson the sunny day to demonstrate the effects of ground re-flections. Simulated PAR (in 𝜇 mol m −2 s −1 ( Ψ )) were ob-tained by multiplying the solar irradiance (in W m −2 ) fromVI-suite by 2.02 (Mavi and Tupper, 2004, p.36; Foken, 2017,p.257). Measurement units are one of the principal differ-ence in comparison to studies focused on assessing the suit-ability of installing photovoltaic panels. For assessing thelevel of sunlight availability, the results from such simula- Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 5 of 17D city models for urban farming site identification in buildings (a) PAR sensor placement along the corridors. (b) PAR sensor placement on the window ledge.(c) Façade description in the study area on the 3D model. (d) Sensors placed on the building façades mapped to the correspond-ing grid cells in the 3D model based on proportionality.(e) Sensor locations on the gridded façades of the 3D model during survey periods.
Figure 5:
Façade description and sensor placement in the study area during survey periods. tions cannot be used directly, but have to be converted toother units, which are not supported by simulation software.While PAR refers to the instantaneous amount of solarradiation, according to Song et al. (2018), the adequacy ofsunlight for crop growth is expressed in terms of DLI whichis defined as the cumulative instantaneous PAR over 24-hourperiod. Thus, DLI (in mol m −2 day −1 ( Φ )) can be calculatedfor each sensor location in the study area using the equation: DLI = ∑ 𝑖 =1 ( 𝑖 𝑡ℎ mean logged instantaneous PAR ( in Ψ)×5 × 60) × 10 −6 where, 𝑖 depicts the 𝑖 𝑡ℎ five-minute interval in 24-hourperiod. For each location, average DLI can also be derivedfrom these DLI for each survey period. The DLI equivalentin VI-Suite was obtained by using CBDM analysis whereinthe hourly beam and diffuse solar radiation data of Singa-pore was taken in EPW format (Southall and Biljecki, 2017). Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 6 of 17D city models for urban farming site identification in buildings
Figure 6:
Observed skytype during survey periods.
For each grid cell, cumulative solar radiation (expressed inkWh m −2 ) was obtained for the whole year as well as for themonths of March, June, September, and December. Aver-age values were obtained for each grid cell by dividing thesecumulative values by the corresponding number of days inthe month/year. Finally, simulated DLI (in Φ ) was obtainedby multiplying the averaged values with 7.272 (1 kWh m −2 day −1 = 3.6 ÃŮ 2.02 Φ ). These simulated DLI were roundedoff to the nearest integer and then compared with the knownDLI of different crops (Faust, 2002; Song et al., 2018) toidentify the suitable crops for a given location. Lastly, hourlymeasured and simulated PAR from 7am to 7pm were ob-tained for each sensor location during the survey periods.Spearman’s correlation coefficient ( 𝜌 ) was determined be-tween them due to their non-normal distributions. They werealso compared based on mean absolute error (MAE) and rootmean square error (RMSE). To enable reproducibility of thework, other parameter values used in the simulations of sunpath, shadow map, and lighting analysis of VI-Suite havebeen included in this paper (see Table 2 in Appendix A).
4. Results and discussion
The main result of the work is that 3D city models ap-pear to be a promising tool for assessing the potential of ur-ban farming in high-rise buildings and for identifying suit-able sites. While using them to assess the solar exposure ofparts of buildings is not our invention, the results indicatethat such assessments can be adapted for use in urban farm-ing (e.g. using different units, considering different spacesand time frames). Unlike simulations for assessing the suit-ability of installing photovoltaic panels, growing crops hasits own particularities, requiring special attention and settingthis use case apart from the previous body of knowledge fo-cused on solar panels. For example, different crops requiredifferent levels of solar exposure. Unlike solar panels, onthe other end, some crops are sensitive to too little light, es-sentially setting a range of acceptable values for the solarexposure. The same goes for the amount of direct solar ex-posure — some crops are more suitable to be grown in shade,and not to be exposed directly to the sun, requiring a modestamount of indirect light. In contrast to solar panels, which the more exposed to the sun they are, they harness more en-ergy so applications are focused on identifying the brightestspot on the rooftop, a beneficial particularity of this use caseis that less exposed and shaded parts of buildings are usefulas well — in that case they are rather suitable for particularcrops which do not prefer an excess amount of light. Simu-lations using 3D city models can help to understand whichcrop is best suited for a particular site in a building. Thissection elaborates on the results in details.
Figure 7 shows the hourly sun path diagrams, which il-lustrates the sun’s movement at different hours, on the firstand last day of the PAR survey. The convoluted rings ( ∞ )depict the hours of the day from dawn to dusk with centralring representing noon. The points ( ∙ ) depict the sun’s posi-tion at the hour represented by the ring. In agreement withFigure 2, this figure also indicates that ‘Block 633’ is ori-ented in the North West - South East direction. As a re-sult, the sunlight distribution is uneven on different façadesof the building owing to the sun’s movement from East toWest from dawn to dusk respectively. While façades A andB that face North East and South East respectively receiveddirect sunlight from morning till afternoon, façades C andW that both face South West received direct sunlight duringthe afternoon and evening hours.It can also be seen from the figure that the sun movedin the northern direction along the convoluted rings from 02Mar to 05 Apr. In a typical year, the sun traverses from thesouthern extreme of this ring to the northern extreme duringthe first half (i.e. January–June) and in the reverse directionduring the second half (i.e. July–December). This move-ment of the sun suggests a variation in PAR received at agiven location. However, this variation during the year maynot be significant in the context of this research due to thehigher solar elevation given the equatorial position of Sin-gapore (Tan and Ismail, 2015; Tablada and Zhao, 2016).Figure 8 shows the role played by shadowing effects on aday-to-day basis. It can be seen from this figure that PAR re-ceived on different façades of the buildings was affected bythe shadows casted by the building’s own façades and sur-rounding buildings. Further, the size of these shadows var-ied at different hours of the day owing to the sun’s diurnalmotion from East to West. For example, façade B is partlyshadowed by façade C and shadow casted by a nearby build-ing in the morning and as the day progresses, these shadow-ing effects recede. In addition, the size and the orientation ofthese shadows are affected by the annual motion of the sunwhich can be observed from the buildings’ shadows in thefigure on the first and last day of the PAR survey. For exam-ple, the size of the shadow casted by façade C on façade B at10am is relatively larger on 05 Apr as compared to 02 Mar.Moreover, the shadows casted also depend on the shape ofthe building. For instance, shadow casted by ‘Block 633’ onitself and nearby buildings differs from the shadows castedby nearby buildings of different shapes in the study area. Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 7 of 17D city models for urban farming site identification in buildings
Figure 7:
Hourly sun path diagrams on first and last day of the PAR survey.
Figure 8:
Solar illumination at different hours of the first and last day of survey.
Shadow maps (Figure 9) indicate that different areas offaçades A, B, C, and W receive direct sunlight for differentdurations on a sunny day. Façade A received direct sunlightonly for a small duration till noon which has been affectedby shadowing effects of adjacent façade and nearby building.The direct sunlight received on Façade B from 7am–1pm and1pm-7pm is affected by the shadowing effects of adjacentfaçades (including façade C) and buildings, and by the sun’sdiurnal motion respectively. Although façades C and W facein the same direction and received direct sunlight for sameduration from 7am–1pm, the percent of time they receiveddirect sunlight from 1pm–7pm varied due to the shadowingeffects of façade B and nearby building respectively. It isalso observed that the duration of time a façade is exposedto direct sunlight increased with the height of the building.However, except for façade W, there is only a small area offaçades B and C which received direct sunlight for more thanequal to 50% of the daytime (i.e. 7am–7pm). While façadeB received most of this sunlight during the first half of the day, façades C and W received it during the second half. Fur-ther, these areas are mostly located at level 11 & above forfaçade B and level 15 & above for façade C. Façade A re-ceived direct sunlight for about 40% of the daytime only onlevel 16. For rest of the daytime, all these façades receivedindirect sunlight. As different crops have different light re-quirements, they may only achieve optimum growth whenplaced at certain façades that can meet the crops’ light re-quirements.
Figure 10a shows that on a sunny day PAR exceeds 500 Ψ on façades A and B in the morning hour except for locationswhich were shadowed by adjacent façade(s) and building.However, as the day progresses, PAR tend to remain below300 Ψ . A trend reversal was observed in case of façades Cand W where PAR at or below 300 Ψ was observed in themorning followed by PAR exceeding 500 Ψ in the later hoursof the day. Further, enormously high PAR observed on the Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 8 of 17D city models for urban farming site identification in buildings
Figure 9:
Shadow maps on a sunny day (17 Mar). lower levels of façades B, C, and W were due to ground re-flections (Figure 10b) which are a component of indirect sun-light. It is admitted here that such PAR levels from groundreflections are not observed in practice, can be attributed tohigh reflectance given to ground plane (Lu and Du, 2013,p.171), and is a limitation of the simulations carried out inVI-Suite. In the absence of ground reflections, PAR wasfound to increase with height on these façades during thesetimes. PAR on the same level of the façade was found to beuniform except in cases where shadows were casted by ad-jacent façade(s) and building. In such cases, on an average,PAR was reduced by 58% of the PAR observed in unshadedconditions on the same level (Table 3 in Appendix A).PAR on different façades during a partly cloudy day (Fig-ure 11a) followed similar trend as that of a sunny day that is,PAR decreased on façades A and B and increased on façadesC and W through the day. While the maximum PAR ob-served on façades A and B was about 150 Ψ and 200 Ψ re-spectively pre-noon, it remained below 50 Ψ and 100 Ψ re-spectively post-noon. PAR on façades C and W at 10am,1pm, and 4pm remained below 120 Ψ , 150 Ψ , and 200 Ψ re-spectively. PAR showed an increasing trend with height at10am for façades A and B and for all façades at 4pm. In othercases, PAR remained in a narrow range with no incrementaltrend with height. On the same level of a façade, on an av- erage, PAR under shaded conditions was reduced by 40% ofthe PAR in unshaded conditions (Table 3 in Appendix A).Abnormally high PAR on lower levels of façades were alsoobserved on partly cloudy day.Shadowing effects did not have a significant impact onPAR distribution on a cloudy day (Figure 11b) primarily dueto the cloud cover and no particular trend in PAR on façadeswas observed through the day. Further, maximum PAR ob-served for all the façades at 10am, 1pm, and 4pm was around90 Ψ , 100 Ψ , and 50 Ψ respectively. For all the façades, PARwas found to increase with height while remaining in a nar-row range.The annual average DLI on façades A, B, C, and Wranged between 5–12 Φ , 1–12 Φ , 7–13 Φ , and 12–15 Φ respectively (Figure 12). For all the façades, these valuesincreased with height. Further, average DLI above 9 Φ wasobserved in all grid cells of façade W and in some grid cellsof façades A, B, and C above levels 12, 6, and 7 respectively.The number of grid cells with these values on façades A, B,and C also increased at higher levels. The variation in av-erage DLI on a given level, if any was the cumulative resultof the shadowing effects of the adjacent façade(s) and build-ing(s) and changing weather conditions as per the EPW data.The monthly average DLI for March, June, September,and December (Figure 13) demonstrated similar trends as Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 9 of 17D city models for urban farming site identification in buildings (a) With ground reflections. (b) Without ground reflections.
Figure 10:
PAR distribution on façades on a sunny day (17 Mar). that of annual average DLI. While the range of monthly av-erage DLI for the façades remained almost same, the levelsat which the average DLI exceeded 9 Φ in the grid cells var-ied significantly across the months for all the façades exceptfaçade W (Table 4 in Appendix A). Spearman’s 𝜌 between measured and simulated PAR,considering all skytypes and sensor locations on façades to-gether (Figure 14), was determined to be 0.62. MAE andRMSE were found to be 176.9 Ψ and 365.2 Ψ respectively.Location wise analysis showed that under all skytypes, 𝜌 wasbetween 0.52 and 0.74 for all sensor locations except for 12a where it was 0.45. In addition, relatively high values of MAEand RMSE were observed for sensors placed on façade B atlevels 8, 12, and 15.Under sunny skytype, all sensor locations together ex-hibited 𝜌 , MAE, and RMSE of 0.66, 233.2 Ψ , and 407.3 Ψ respectively. Observations similar to all skytypes were alsomade for sunny skytype for individual sensor locations with 𝜌 between 0.58 and 0.82 except for 12a ( 𝜌 = 0.45) and rela- Sensors placed on the same level of a façade were labelled as ‘a’, ‘b’,and so on starting from the left in Figure 5e. tively high values of MAE and RMSE for sensors at levels 8and 12 of façade B.When all sensor locations were considered together, 𝜌 ,MAE, and RMSE under partly cloudy skytype were compa-rable to sunny and all skytypes with values of 0.70, 170.1 Ψ ,and 389.7 Ψ respectively. However, location wise analysisshowed that although MAE and RMSE were relatively highfor sensors at levels 8, 10, 12, and 15 on façade B, 𝜌 wasbetween 0.51 and 0.92 except for sensors at 5a, 12a, 15, 16a,and 16b.Under cloudy skytype, 𝜌 , MAE, and RMSE were 0.61,135.9 Ψ , and 287.5 Ψ respectively for all sensor locationsconsidered together. 𝜌 varied between 0.65 and 0.80 exceptfor sensors at façade W where it was about 0.33. MAE andRMSE remained relatively low except for sensor at level 14of façade C. Sun path analysis, shadow map analysis, and spatio-temporal analysis of PAR and DLI suggest that PAR andDLI at a location in the building are dependent on the build-ing’s shape and orientation, the sun’s diurnal and annualmotion, skytypes, and shadowing effects of the building’sown façades and nearby buildings. Further, in contrast to
Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 10 of 17D city models for urban farming site identification in buildings (a) Partly cloudy day (09 Mar). (b) Cloudy day (03 Apr).
Figure 11:
PAR distribution on façades on a partly cloudy and cloudy day.
Figure 12:
Annual average DLI distribution on façades.
Martínez-Rubio et al. (2016), it is not possible to adjudgethe significance of one factor over another as all these factorshave a cumulative effect on PAR and DLI at a given locationand their individual significance may change with change instudy area and study period.In agreement with Song et al. (2018), PAR on the façadesin the study area remained largely similar as the façades wereexposed to direct sunlight for similar durations owing to thesun’s diurnal motion. While façades A and B experienced higher PAR during the former half of the day, façades C andW experienced it during the latter half. However, this patternwas visible only for sunny and partly cloudy skytypes. Nosuch pattern was observed under cloudy skytype primarilydue to cloud cover. Further, PAR on the façades was signifi-cantly reduced moving from sunny to partly cloudy to cloudyskytypes.In addition to sunny skytype (Song et al., 2018), PARon these façades increased with height for cloudy skytypeas well. However, no such trend was discerned for partlycloudy skytype as it is characterized by bouts of sunlightand cloud cover through the day. In consonance with thefindings of Tan and Ismail (2014), on the same level of thefaçade, PAR was mainly affected due to shadowing effectsof adjacent façade(s) and/or building. Based on some testcases, PAR reduction due to shadowing effects was found tobe more under sunny skytype as compared to partly cloudyskytype.The annual as well as monthly average DLI ranged from1 to 15 Φ at different locations on these façades with largelysimilar values on façades across months. Confirming thefinding of Song et al. (2018), average DLI increased withheight. However, in real world, this trend as well as thehigher average DLI observed on some façades may be af- Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 11 of 17D city models for urban farming site identification in buildings
Figure 13:
Average DLI distribution on façades for selected months.
Figure 14:
Metrics for comparing measured and simulated PAR. fected due to presence of trees and other built structures atlower levels. Except for façade W, locations with averageDLI exceeding 9 Φ were mainly located at higher levels of thefaçades. This may be attributed to the self-shadowing andshadowing effects of adjacent buildings on the lower levels(Martínez-Rubio et al., 2016). The dry phase of NortheastMonsoon during January–March (Climate) and the sun’s po-sition in the southern hemisphere leading to higher PAR on the façades can be the reasons for large number of grid cellswith average DLI exceeding 9 Φ in the month of March.The range of observed average DLI correspond to theDLI requirements of crops that belong to the very low light(< 5 Φ ), low light (5–10 Φ ), and moderate light (10–20 Φ )categories. Out of these categories, only crops grown undermoderate light conditions are considered suitable for com-mercial production (Faust, 2002). Thus, crops such as sweet Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 12 of 17D city models for urban farming site identification in buildings pepper (
Capsicum annuum ) and lettuce (
Lactuca sativa ) be-longing to the moderate-light categories can be grown atlocations where average DLI exceeds 9 Φ on these façades(Song et al., 2018).Statistically significant and moderate to high values of 𝜌 (> 0.5) under different skytypes suggests that positive linearrelationship exists between measured and simulated PAR,affirming the usability of 3D city models for this use case.High values of MAE and RMSE suggest that the simulatedPAR deviate from the measured PAR. Thus, the simulatedPAR was able to capture the trend followed by measuredPAR at a sensor location but not its values. There are mainlythree reasons for this. Firstly, the 3D model used for carry-ing out simulations was LOD1 model, excluding vegetation.The present model did not take into account the architecturalelements of the façades such as cantilevered louvres withperforated sunscreen above the windows, roof overhangs onlevel 16, among others and the presence of trees causingshadows at lower levels. As a result, simulations do not ac-count for their effects on PAR at a location. Although higherdetailed (i.e. LOD3) models can take care of façade’s archi-tectural elements, past research suggests that these modelsare generally available for a small study area and their non-availability becomes a constraint when considering a largerstudy area.Secondly, hourly simulated PAR generated at varioussensor locations for comparison with measured PAR werebased on the 24-hour weather forecasts and not the actualweather conditions. Cases have emerged where PAR mea-sured through sensors on a cloudy day were found to beequivalent to those measured on a sunny day. For instance,as seen in Figure 15, PAR logged at location 9a around 3pmon 27 Mar (cloudy day) matched to those measured on 31Mar (sunny day). PAR logged around 3pm on 27 Mar werevery high as compared to those logged for another cloudyday (04 Apr) around the same time. Such differences in ac-tual and forecasted weather conditions have resulted in ex-ceptionally low 𝜌 and high MAE and RMSE for the cloudysky conditions on façade W. Relatively low 𝜌 and relativelyhigh MAE and RMSE observed at sensor locations on façadeB and level 14 of façade C for partly cloudy and cloudy skyconditions can also be attributed to the same reason. Theseoutliers have significantly impacted the performance of oth-erwise fairly accurate simulations.Lastly, the inability of these simulations to model thesharply contrasting periods of low and high PAR is anotherreason for deviation in measured and simulated PAR. Toillustrate the same, Figure 16 shows measured and simu-lated PAR at location 12a on a sunny day (02 Mar) at five-minute and half-hourly intervals respectively with PAR athalf-hourly intervals highlighted with points ( ∙ ). This par-ticular location is free from any obstacles that may possiblyaffect measured PAR. As seen in the figure, this high vari-ability in PAR is observed around noon when the sun is posi-tioned right above and moving past façade B. Consequently,such variable sunlight conditions (Smith and Berry, 2013)can be attributed to the parapet edges of the higher levels Figure 15:
Measured PAR at location 9a on sunny and cloudydays during period III. which have been captured by the PAR sensor also placedon the railing of the corridor’s parapet (Figure 5a). Samereasoning can also be applied to sensors placed at levels 8,10, and 15. On the same level of façade B, relatively bet-ter values of these metrics for location ‘b’ (e.g. 12b) thanlocation ‘a’ (e.g. 12a) is due to the fact that these locationswere shadowed by façade C for some duration (e.g. 7am-1pm in Figure 9) when this façade received direct sunlight.As a result, some of these variable sunlight episodes werenot observed in measured PAR for these locations. Giventhese reasons, simulated PAR data compare fairly well withmeasured PAR.
Figure 16:
Measured and simulated PAR for location 12a ona sunny day (02 Mar), which has the highest discrepancy.
The excessive irradiation captured by the PAR sensorover short durations around noon also contribute to the DLIat the sensor location. Consequently, these variable sunlightepisodes may result in crop’s reduced photosynthetic perfor-mance in the built environment (Tan and Ismail, 2014; Songet al., 2018) due to its mistaken selection based on inflatedDLI. By not being able to model these episodes, simula-tions eliminate locations that receive higher DLI due to thesebouts of highly variable intensity of sunlight. At the sametime, this also results in underestimation of DLI at a locationfound suitable for farming. Thus, some caution may have tobe exercised during crop selection. Only those crops having
Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 13 of 17D city models for urban farming site identification in buildings
Figure 17:
Average DLI for March on façades of the study area. These results are a criticalinsight for decision-making for high-rise urban farming and for maximizing the crop yield. threshold DLI obtained through simulations and which aretolerant to these episodes would be suitable for such loca-tions.The results in the previous section and the above discus-sion have not only corroborated the findings of the exist-ing literature but have also contributed toward an enrichedunderstanding of PAR and DLI at different micro-locationsof the same building as well as in different sky conditions.Particularly, it highlighted how 3D city models can facili-tate understanding of the sun’s diurnal and annual motionin the study area and enable estimating the shadowing ef-fects and DLI for locations that may not be easily accessi-ble to PAR surveys (e.g. windows of dwelling units). Thus,with 3D city models it is possible to assess the suitabilityof a micro-location in a building for farming; alleviating theneed to conduct PAR survey and simultaneously introducingtheir new application. One way to possibly use the results ofthese analyses on 3D city models is to map the locations ofgrid cells found suitable for farming (Figure 17) with cor-responding locations in 3D modeling tools such as Blenderand Google Earth (Figure 3) and examine whether it is prac-tically possible to do farming at those locations. Not to men-tion, while the present paper has only focused on a particularbuilding in Singapore, the methodology employed to gener-ate 3D city models and the analyses carried out herein areequally applicable to other high-rise buildings within and be-yond Singapore.
While the results and subsequent discussion suggest that3D city models can support urban farming site identifica-tion in buildings and help deciding which crop to grow atwhich site, they suffer from some limitations. Firstly, EPW data used in DLI simulations in this paper correspond tothe 1990s. As a result, the actual DLI in the present con-ditions may differ from the simulated DLI. However, signif-icant variations in DLI at the scale of months/year, whichhave been used in the analysis, are not expected given theequatorial position of Singapore. Secondly, besides the lim-itation associated with ground reflections, this version of VI-Suite does not account for leap year, 2020 being one. Whilelarge variations in simulated PAR were not observed be-tween consecutive days, it is hoped that this and other lim-itations will be rectified in its upcoming versions. Lastly,vegetation around buildings may play an important role insolar exposure assessment, which we did not have in our 3Dcity model.
5. Conclusion
This paper investigated a new application of using 3Dcity models to identify urban farming sites in buildings andunderstand their potential for growing particular crops basedon sunlight properties derived by simulations. It capitalizedon the prior work relying on 3D city models to estimate thesolar potential for assessing the suitability of installing pho-tovoltaic panels on rooftops, and adapted it for a significantlydifferent purpose and of a different nature with certain par-ticularities (urban farming) and different locations (verticalspaces of buildings). Our work includes field measurementsto verify the integrity of the simulations, which is a rarity inrelated work. The important points from this work are:• There is a large variation in the level of availablesunlight within a building, requiring understandingthe potential of different sites for urban farming at amicro-location scale.
Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 14 of 17D city models for urban farming site identification in buildings • 3D city models can be conveniently used to supporturban farming by identifying such sites in an approx-imate manner. They have an unparalleled advantageover doing field measurements when there are manymore locations to evaluate and especially when scal-ing up the estimations at the precinct or urban scaleto cover thousands of buildings, which is in practiceimpossible to carry out with field measurements.• Such analyses can be conducted using block (LOD1)models obtained from open data, and the simulationscan be run using open-source software, facilitatingreplication elsewhere and scalability to cover entirecities.We believe that this novel use case has rich potential tobe further researched, and there are several avenues for ex-panding this work. Quantifying the farming area and the pro-jected crop yield in a building is the first one. Improving sim-ulation accuracy by employing datasets including vegetationand using 3D models of higher detail such as architecturalmodels which are becoming increasingly common (Stouffset al., 2018), integrating dynamic and indoor data which is apromising research direction in 3D city modelling (Kutzneret al., 2020; Konde et al., 2018), and experimenting with dif-ferent material types for ground surfaces offer another line offuture work. Most importantly, growing crops at farming lo-cations identified through simulations would be the real testof 3D city models. It is hoped that these lines of researchwill show the path to accurately estimate the farming poten-tial in buildings and provide thrust to undertake this activityat the urban scale. As another possible future work direc-tion, it is also foreseen that once the potential is assessedand the ensuing urban farming activities in buildings com-mence, 3D city models — optionally coupled with additionaldata such as legal matters — can be used also to managethem and serve as a registry for coordination purposes, forexample, for organising the provision of subsidies and forissuing permits for farming in public buildings. Finally, itwould be interesting to investigate whether this use case canbe combined with assessing the suitability for installing so-lar panels and energy simulations, recommending the opti-mal mix and arrangement of photovoltaic installations andagricultural crops in the same building, presenting a holis-tic solution for supporting green buildings and sustainabledevelopment.
Acknowledgements
The authors appreciate the contact with Dr Ryan Southall(University of Brighton) for answering queries related to VI-Suite, Dr Siu-Kit Lau (National University of Singapore) forproviding a light meter during the preliminary analysis, andthe mapping efforts of the OpenStreetMap community. Theauthors would also like to thank Dr Puay Yok Tan and DrXiao Ping Song for their help with the PAR surveys. This re-search was conducted as part of the master’s thesis of the leadauthor while he was on sabbatical leave from Union Bank of India. It is also part of the project Large-scale 3D Geospa-tial Data for Urban Analytics, which is supported by the Na-tional University of Singapore under the Start Up Grant R-295-000-171-133.
A. Appendix
Table 1
Calibration equations for the PAR sensors used in the survey.PAR sensor ID Calibration equation R Table 2
Parameter values for simulations in VI-Suite.Analysis/Simulation type Parameter name Parameter valueSun path Suns Single or HourlyThickness 0.15Shadowmap Animation StaticResult Point FacesOffset 0.01Basiclighting Result Point FacesOffset 0.01Program GenskyGround ref 0.00Turbidity 2.75Accuracy MediumClimateBasedDaylightModelling Result Point FacesOffset 0.01Type ExposureAccuracy FinalAnkit Palliwal et al.:
Preprint submitted to Elsevier
Page 15 of 17D city models for urban farming site identification in buildings
Table 3
Percent (%) reduction in PAR under different sky conditions.Skytype Time Façade Level 𝜇 mol m −2 s −1 ) ( 𝜇 mol m −2 s −1 )Sunny (17 Mar) 10am A 13 10 568 9 259 54.4010am B 5 9 653 10 341 47.7710am B 10 12 570 13 256 55.081pm C 5 5 434 4 240 44.704pm C 13 11 606 10 84 86.13Partly Cloudy (09 Mar) 10am A 13 11 130 10 85 34.6110am B 5 10 151 11 98 35.0910am B 10 12 145 13 90 37.931pm C 5 3 90 2 56 37.774pm C 13 10 146 9 63 56.84 Table 4
Monthly average DLI range and levels with average DLI above 9 mol m −2 day −1 .Month Façade Range Levels at whichMinimum Maximum average DLI exceeds(mol m −2 day −1 ) (mol m −2 day −1 ) 9 mol m −2 day −1 March A 5 12 14 and aboveB 1 13 allC 7 13 allW 12 15 allJune A 5 14 10 and aboveB 1 10 15 and aboveC 4 10 14 and aboveW 11 13 allSeptember A 5 12 13 and aboveB 1 12 9 and aboveC 6 12 9 and aboveW 12 14 allDecember A 4 10 16B 1 15 allC 8 14 1, 2, 7 and aboveW 10 15 all
References
American Society of Heating, Refrigerating and Air-Conditioning Engi-neers, 2001. Weather Data Download - Singapore 486980 (IWEC)[Electronic data file]. energyplus.net/weather. Available from: https://energyplus . net/weather-location/southwest_pacific_wmo_region_5/SGP//SGP_Singapore . [Accessed: 12 April 2020].Bandyopadhyay, S., Jaiswal, R.K., Hegde, V.S., Jayaraman, V., 2009. As-sessment of land suitability potentials for agriculture using a remotesensing and GIS based approach. International Journal of Remote Sens-ing 30, 879–895.Biljecki, F., 2020. Exploration of open data in Southeast Asia to generate3D building models. ISPRS Ann. Photogramm. Remote Sens. SpatialInf. Sci. .Biljecki, F., Ledoux, H., Stoter, J., 2017. Generating 3D city models withoutelevation data. Computers, Environment and Urban Systems 64, 1–18.Bshouty, E., Shafir, A., Dalyot, S., 2020. Towards the generation of 3DOpenStreetMap building models from single contributed photographs.Computers, Environment and Urban Systems 79, 1–10.Catita, C., Redweik, P., Pereira, J., Brito, M., 2014. Extending solar poten- tial analysis in buildings to vertical facades. Computers & Geosciences66, 1–12.Cheng, L., Zhang, F., Li, S., Mao, J., Xu, H., Ju, W., Liu, X., Wu, J., Min,K., Zhang, X., Li, M., 2020. Solar energy potential of urban buildingsin 10 cities of China. Energy 196, 117038.Climate, 2020. Climate of Singapore. [Online]. Meteorological ServiceSingapore. Available from: . weather . gov . sg/climate-climate-of-singapore/ [Accessed: 05 July 2020].Diehl, J.A., Sweeney, E., Wong, B., Sia, C.S., Yao, H., Prabhudesai, M.,2020. Feeding cities: Singapore’s approach to land use planning forurban agriculture. Global Food Security 26, 1–11.Dukai, B., Ledoux, H., Stoter, J.E., 2019. A multi-height LoD1 model of allbuildings in the Netherlands. ISPRS Ann. Photogramm. Remote Sens.Spatial Inf. Sci. IV-4/W8, 51–57. doi: . .Estoque, R.C., Murayama, Y., 2011. Suitability Analysis for BeekeepingSites Integrating GIS & MCE Techniques, in: Murayama, Y., Thapa,R.B. (Eds.), Spatial Analysis and Modeling in Geographical Transfor-mation Process: GIS-based Applications. Springer, Dordrecht, pp. 215–233. Ankit Palliwal et al.:
Preprint submitted to Elsevier
Page 16 of 17D city models for urban farming site identification in buildings
Faust, J.E., 2002. Light Management in Greenhouses: Plant Growth Re-sponses to Daily Light Integrals. [Online] Aurora, Illinois, SpectrumTechnologies. Available from: . specmeters . com/assets/1/7/A052 . pdf [Accessed: 05 July 2020].Foken, T., 2017. Micrometeorology. 2 ed., Springer-Verlag, Berlin.Freitas, S., Catita, C., Redweik, P., Brito, M., 2015. Modelling solar poten-tial in the urban environment: State-of-the-art review. Renewable andSustainable Energy Reviews 41, 915–931.Google Earth - WebAssembly, 2020a. Jurong West Street 65, Block 633,Singapore, 1 ◦ ◦ https://earth . google . com/web/ [Accessed:29 May 2020].Google Earth - WebAssembly, 2020b. Jurong West Street 65, Block 633,Singapore, 1 ◦ ◦ https://earth . google . com/web/ [Accessed:29 May 2020].Gröger, G., Plümer, L., 2012. CityGML – Interoperable semantic 3Dcity models. ISPRS Journal of Photogrammetry and Remote Sensing71, 12 – 33. URL: . sciencedirect . com/science/article/pii/S0924271612000779 , doi: . . isprsjprs . . . .Housing & Development Board, 2014. Precast pictorial guide. Housing &Development Board, Singapore.Housing & Development Board, 2020. HDB property information [Elec-tronic data file]. Data.gov.sg. Available from: https://data . gov . sg/dataset/hdb-property-information [Accessed: 12 April 2020].Khan, R., Aziz, Z., Ahmed, V., 2018. Building integrated agriculture in-formation modelling (BIAIM): An integrated approach towards urbanagriculture. Sustainable Cities and Society 37, 594–607.Kim, H., Lee, K., Lee, J., Lee, S., 2018. Exploring outdoor solar poten-tial in high-density living: Analyzing direct sunlight duration for urbanagriculture in seoul’s residential complexes. Energies 11, 1–16.Konde, A., Tauscher, H., Biljecki, F., Crawford, J., 2018. Floor plans inCityGML. ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci.IV-4/W6, 25–32. doi: . .Kosorić, V., Huang, H., Tablada, A., Lau, S.K., Tan, H.T., 2019. Surveyon the social acceptance of the productive façade concept integratingphotovoltaic and farming systems in high-rise public housing blocks inSingapore. Renewable and Sustainable Energy Reviews 111, 197–214.Kutzner, T., Chaturvedi, K., Kolbe, T.H., 2020. CityGML 3.0: New Func-tions Open Up New Applications. PFG – Journal of Photogramme-try, Remote Sensing and Geoinformation Science , 1 – 19doi: . .Ledoux, H., Ohori, K.A., Kumar, K., Dukai, B., Labetski, A., Vitalis, S.,2019. CityJSON: A compact and easy-to-use encoding of the CityGMLdata model. Open Geospatial Data, Software and Standards 4, 4.doi: . .Lim, Y.A., Kishnani, N.T., 2010. Building Integrated Agriculture: UtilisingRooftops for Sustainable Food Crop Cultivation in Singapore. Journalof Green Building 5, 105–113.Lu, M., Du, J., 2013. Assessing the daylight and sunlight availability inhigh-density residential areas: a case in North-east China. ArchitecturalScience Review 56, 168–182.Martínez-Rubio, A., Sanz-Adan, F., Santamaría-Pe¯na, J., Martínez, A.,2016. Evaluating solar irradiance over facades in high building cities,based on LiDAR technology. Applied Energy 183, 133–147.Mavi, H.S., Tupper, G.J., 2004. Agrometeorology: Principles and Applica-tions of Climate Studies in Agriculture. The Haworth Press, Bingham-ton, New York.Redweik, P., Catita, C., Brito, M., 2013. Solar energy potential on roofsand facades in an urban landscape. Solar Energy 97, 332–341.Santos, T., José António Tenedório, José Alberto Gonçalves, 2016. Quan-tifying the city’s green area potential gain using remote sensing data.Sustainability 8, 1–16.Saretta, E., Bonomo, P., Frontini, F., 2020. A calculation method for theBIPV potential of Swiss façades at LOD2.5 in urban areas: A case fromTicino region. Solar Energy 195, 150–165.Singapore Department of Statistics, 2019a. Population Trends 2019.[Online]. Singapore Department of Statistics. Available from: . singstat . gov . sg/-/media/files/publications/population/population2019 . pdf [Accessed: 05 July 2020].Singapore Department of Statistics, 2019b. Singapore in figures 2019.[Online]. Singapore Department of Statistics. Available from: . singstat . gov . sg/-/media/files/publications/reference/sif2019 . pdf [Accessed: 05 July 2020].Smith, W.K., Berry, Z.C., 2013. Sunflecks? Tree Physiology 33, 233–237.Song, X.P., Tan, H.T., Tan, P.Y., 2018. Assessment of light adequacy forvertical farming in a tropical city. Urban Forestry & Urban Greening 29,49–57.Southall, R., Biljecki, F., 2017. The VI-Suite: a set of environmental analy-sis tools with geospatial data applications. Open Geospatial Data, Soft-ware and Standards 2, 1–13.Stoter, J., Peters, R., Commandeur, T., Dukai, B., Kumar, K., Ledoux, H.,2020. Automated reconstruction of 3D input data for noise simulation.Computers, Environment and Urban Systems 80, 101424. URL: . sciencedirect . com/science/article/pii/S0198971519302662 ,doi: https://doi . org/10 . . compenvurbsys . . .Stouffs, R., Tauscher, H., Biljecki, F., 2018. Achieving Complete and Near-Lossless Conversion from IFC to CityGML. ISPRS International Jour-nal of Geo-Information 7, 355. URL: . mdpi . com/2220-9964/7/9/355 , doi: . .Tablada, A., Zhao, X., 2016. Sunlight availability and potential food andenergy self-sufficiency in tropical generic residential districts. Solar En-ergy 139, 757–769.Tan, P.Y., Ismail, M.R.B., 2014. Building shade affects light environmentand urban greenery in high-density residential estates in Singapore. Ur-ban Forestry & Urban Greening 13, 771–784.Tan, P.Y., Ismail, M.R.B., 2015. The effects of urban forms on photosyn-thetically active radiation and urban greenery in a compact city. UrbanEcosystems 18, 937–961.The Weather Channel, 2020. The Weather Channel. [Online]. NewYork, IBM. Available from: https://weather . com/en-SG/weather/today/l/SNXX0006:1:SN [Accessed: 05 July 2020].Ward, G.J., 1994. The RADIANCE Lighting Simulation and RenderingSystem, in: Proceedings of the 21st Annual Conference on ComputerGraphics and Interactive Techniques, Association for Computing Ma-chinery, New York, NY, USA. pp. 459–472.Weather, 2020. Weather. [Online]. Singapore, National EnvironmentAgency. Available from: . nea . gov . sg/weather [Accessed: 05July 2020].Wilson, J.P., 2005. Local, national, and global applications of GIS in agri-culture, in: Longley, P., Goodchild, M., Maguire, D., Rhind, D. (Eds.),Geographical Information Systems: Volume 2, Management Issues andApplications. Wiley, Chichester, pp. 981–998.Zulkifli, M., 2019. A MEWR Trilogy: Our Food and Environmen-tal Stories Develop from Our Water Story. [Online]. Singa-pore, Ministry of the Environment and Water Resources. Avail-able from: . mewr . gov . sg/news/speech-by-mr-masagos-zulkifli--minister-for-the-environment-and-water-resources--at-the-committee-of-supply-debate-2019--on-7-march-2019 [Accessed:05 July 2020]. Ankit Palliwal et al.: