Input data requirements for daylight simulations in urban densifications

Abstract

One of the biggest challenges in urban densifications is securing adequate daylight access. This study examines the potential of using semantic 3D city models as input to daylight simulations. It is focusing on investigating input data requirements to these simulations from a geodata, 3D city model specification and measuring guideline perspective. To achieve this, geodata simulation input requirements for the most common daylight metrics are documented. Next, 3D city model data from two Swedish municipalities along with 3D data constructed by ourselves in CAD- and GIS-environments are utilized to explore the impact of using 3D city models of different levels of detail (LOD) and positional accuracy in daylight simulations linked to Swedish and European laws and recommendations. Similarly, the measuring guidelines and 3D city model specification requirements related to balconies and other façade accessories are also evaluated along with the utilization of façade reflectance properties and colour. It is found that LOD1 is sufficient for the obstruction angle metric for most roof types but for e.g., gabled roofs LOD2 should be used. A positional accuracy on a decimetre-level is accurate enough for the aforementioned metric. Daylight factor simulations require that balconies and façade accessories protruding more than a couple of decimetres must be represented in the 3D city model along with information on façade material and colour. The outcome of the study is expressed in the form of a list of recommendations for the creation of national profiles of 3D city models supporting daylight simulations.

1 Introduction

Cities are becoming increasingly denser as a direct effect of urbanization (United Nations, 2018) and there seems to be a political will in many countries favouring urban densification over urban expansion. This is mainly attributed to the fact that urban densifications are considered as a more sustainable solution compared to urban expansion (Pelczynski & Tomkowicz, 2019). These densifications are not problem-free and pose a challenge for urban planners to design solutions that address the increased demands on housing and commercial buildings while maintaining good living conditions through i.a., adequate access to daylight and sunlight (Jellinek et al., 2019). Limited access to daylight or sunlight has negatively been linked to increased energy demands for lighting (Lechner, 2014), poor health (Lam & Ripman, 1992; Osterhaus, 2005; Lee et al., 2022) as well as decreased productivity in workplaces and educational buildings (Vischer, 2008). National and international daylight legislation and recommendations are there to guide urban planners and real estate developers in making informed decisions. However, the complexity of modern urban settings and daylight metrics (i.e., methods for determining the availability & quality of daylight) hampers their task and calls for the application of systematic analyses that consider larger geographic areas. Consequently, a need for establishing a standardized way of applying daylight-simulations in the planning process is identified.

In Sweden, compliance with legislation regulating daylight indoor conditions is checked late in the process (at building permit stage). Because of this, real estate developers may often struggle to fulfil the established requirements (Bournas, 2021). Decisions made by urban planners during phases prior to the building design phase have a significant impact on the probability of a building meeting the requirements (Kanters & Wall, 2014; Cederström et al., 2020). Daylight simulations can support urban planners in making more informed decisions when it comes to designing a detail development plan for e.g., densifications. In addition, these simulations are essential in the building design phase; a recent study highlights the fact that simulation tools should provide quick feedback to iterate several design alternatives (Kanters et al., 2021). A large number of metrics have been developed to support daylight simulations (Ayoub, 2019). Some of these metrics are applicable at early stages, e.g., the detailed development plan phase, while others are more suitable for the building design and building permit phases.

The choice of daylight metrics in the simulations affects the choice of simulation software, which, in turn, dictates the type of input data. Some metrics only require data on building geometries while others also require knowledge about reflectance properties of walls, window properties, climate, and sun positions, etc. One factor that limits the use of daylight simulations by urban planners is the difficulty to collect and tailor input data. This especially concerns geospatial data describing the neighbourhood near the densification project area. For example, Swedish municipalities maintain 2D base maps and most commonly 3D building models, but these data are not tailored for daylight simulations. Therefore, urban planners need to spend considerable time on data manipulations before being able to execute the actual simulations.

Geospatial data and especially 3D city models have the potential to function as input data to daylight simulations. However, this requires that the 3D city models contain all relevant information for computing daylight metrics, and, that the level of detail and data quality is sufficient. In practice, we not only need 3D city models to compute daylight metrics, but also e.g., sky models, climate models and, for some metrics, also indoor information; however, the focus of this study is solely set on 3D city models and other geospatial data.

The general aim is to ensure that the upcoming Swedish 3D city model standard (currently under development) supports the execution of daylight simulations at early stages of urban densification projects. To achieve this, we start by identifying relevant daylight metrics that are explicitly or implicitly referred to by European and Swedish legislations and recommendations. Then we state the following two research objectives:

• What geospatial data are needed for these daylight metrics?

• How should 3D city model specifications and measuring guidelines be specified to facilitate accurate simulations of these metrics?

In this paper, we focus on identifying requirements for the Swedish specifications for 3D city models by evaluating how well the current specifications support daylight simulations and how they should be adjusted to better satisfy the simulation requirements, especially in the context of urban densifications. Even though the study is designed in a Swedish context, it has a broader interest since the specifications used are mainly based on international standards and the general theory and methods are universally applicable.

The paper is organized as follows. Section 2 describes related work regarding daylight metrics, standardization of geospatial data, studies on daylight simulations as well as legal requirements on daylight conditions in Sweden and Europe. These laws and recommendations are based on daylight metrics which are described in Appendix 1. Section 3 is devoted to investigating the geospatial input data requirements for the most commonly used daylight metrics and how these can be incorporated in geospatial data specifications. Section 4 presents two case studies examining the level of detail and quality requirements on geospatial data for daylight simulations. The paper concludes with discussion and conclusions.

2 Related work

This section includes descriptions on topics relevant to the framework of the case studies. More in particular, it is dedicated to presenting a brief introduction to the field of daylight metrics, discussing what has been done to date to standardize geospatial data with a special focus on 3D city models, and bringing to notice relevant studies combining semantic 3D city models and daylight/sunlight metrics.

2.1 Daylight metrics

Daylight is defined as the infrared, visible and ultraviolet part of the global solar radiation (direct or indirect sunlight: reflected or diffused) that enters the room of a building through a window or other kind of opening (EN 12665:2018, 2018, 3.4.7). Sunlight (infrared, visual, and ultraviolet light) is the light of the sun that enters a room directly without previously being reflected, while skylight (diffuse light) refers to sunlight that has been scattered in the atmosphere. Overall, in countries located at higher latitudes (e.g., Norway), about 60% of the daylight is attributed to skylight (Obradovic & Matusiak, 2020).

There are several methods for assessing the quantity and quality of available day- and sunlight which below are denoted as metrics. A solar performance metric can be described as a mathematical combination of measurements and/or dimensions and/or conditions that are represented on a continuous scale and used for evaluating e.g., the daylight performance in buildings (Mardaljevic et al., 2009). The metrics are computed based on input data such as sunlight conditions and building data.

Solar performance metrics include both daylight- and sunlight-related metrics. The difference being that daylight-related metrics take into consideration both direct and reflected sunlight, while sunlight-related metrics only consider direct sunlight. For simplicity reasons, we will use the term daylight metrics when referring to both categories throughout this paper.

Daylight metrics require different types of sky models. Sky models are mathematical formulas describing the variations in the luminance distribution of direct and indirect visible light originating from different regions of a hemispherical sky dome or sky vault (Enarun & Littlefair, 1995). Different sky models simulate different sky conditions (e.g., clear, overcast, intermediate). The overcast sky is a completely cloud-covered sky condition often required by several daylight metrics. Patches are subdivisions of a hemispherical sky dome and are the equivalent of cells in a raster. The resolution of a sky dome is defined by the number of its patches.

According to Alenius et al. (2019), the most important factors affecting a building’s access to daylight are its typology/shape, façade materials and structure, orientation, latitude, solar angle, extent of the sky view along with the size and location of its windows and the façade material of surrounding buildings. By considering photometric properties of façade materials and windows, it is possible to estimate the quantity and pattern of reflected and transmitted sunlight (Arnesen et al., 2011).

A recent review study by Czachura et al. (2022) identified the most commonly used solar performance metrics and categorized them according to the following taxonomy:

• Geometric

  metrics that are based on geometric properties of a 3D city model and can therefore also be characterized as static metrics

• Latitudinal

  metrics that take into consideration the position of the sun at a given latitude, when simulating solar irradiance at a location

• External climate

  metrics that besides the geometry and the latitude also take into consideration the climate of a location (usually require a weather data file to run)

• Internal climate

  metrics that in addition to all aforementioned requirements also require 3D city models characterized by a high level of external and internal modelling detail.

Current national legislation and international recommendations regulating daylight access often include certain daylight metrics. The daylight metrics used in this paper; obstruction angle (OA), daylight factor (DF) and sunlight exposure (SE), belong to the geometric and latitudinal categories and were selected on the basis of either being explicitly mentioned in the Swedish legislation or implicitly referred to in European and Swedish recommendations or standard practices. More information on the daylight-related legal framework and recommendations is found in Appendix 2, while a detailed description of the selected metrics is included in Appendix 1.

2.2 Standardization of geospatial data

2.2.1 International standard for 3D city models - CityGML

The most comprehensive standard today for semantic 3D city models is the CityGML (City Geography Markup Language) standard by the Open Geospatial Consortium. Currently the most used version is CityGML 2.0 (Gröger et al., 2012; Biljecki et al., 2018; Noardo et al., 2020), but in 2021 a conceptual model for CityGML 3.0 was approved (Kolbe et al., 2021); CityGML 3.0 data schemas for XML and possibly other formats are to be published in the near future. The description below follows CityGML 2.0 since the current Swedish standardisation, as well as most other national standards, are still based on this version.

The main focus of CityGML is to represent the geometric and semantic aspects of features in a city. For the geometric part, CityGML is based on Geography Markup Language (GML) version 3.1.1 with the limitation that CityGML only handles planar surfaces (which implies that e.g., curved walls must be represented by several smaller planar surfaces, in technical terms denoted multi-surfaces). CityGML contains an information model divided into several thematic modules. For computing daylight metrics, the most interesting modules are buildings and vegetation.

CityGML supports multiresolution modelling by including different LODs where the geometry, topology, and semantics are described with varying levels of complexity. CityGML 2.0 defines five LODs, which for building exteriors is reaching from footprint/roof print (LOD0) to detailed representations including roof structure, windows, and doors (LOD3) and finally building interior information being added in LOD4 (Gröger et al., 2006). Some researchers and national organisations have extended this LOD model with sub-types to specify in greater detail how a building should be represented. The Swedish specifications mainly follow the LOD refinements in Biljecki et al. (2016) (Fig. 1).

Fig. 1

A refined version of LOD definitions in CityGML 2.0. LOD4 indoor representations are not included. (Biljecki et al., 2016, p. 28)

The CityGML information model can be extended to support various applications (Biljecki et al., 2018). In technical terms, this is done by introducing an Application Domain Extension (ADE) (Kolbe et al., 2021). To the authors’ knowledge, there has not been any ADE especially developed for daylight simulations but from an information modelling perspective there are some related ADEs such as the one for urban energy simulations, which does include a DaylightIlluminance attribute in its WeatherData object (Nouvel et al., 2015; Agugiaro et al., 2018).

2.2.2 Development of geospatial data specifications in Sweden

The main data source for the Swedish urban planning is currently the municipality 2D maps. These maps include information about buildings and transport infrastructure and sometimes also about vegetation and land use; the level of detail corresponds to a map scale of around 1:1000. These 2D maps have not been standardized on a national level, which has resulted in municipalities using different information models. Around year 2000, municipalities started to create 3D models of buildings in urban areas. These 3D models were originally mainly designed and used for visualisation purposes and hence denoted presentation models. To enhance the possibility of conducting analyses and simulations, semantic 3D city models have started to emerge (e.g., based on the CityGML standard).

During the last decades several countries have defined national profiles for semantic 3D city models based on CityGML (van den Brink et al., 2013; Gruber et al., 2014). In Sweden, there has been some earlier studies (e.g., Eriksson et al., 2020) and there is ongoing work initiated by the three main cities (Stockholm, Gothenburg, and Malmö) together with Lund University to create a national CityGML profile denoted the 3CIM model (3D City Information Modelling) (Uggla et al., 2023). The vision is that the 3CIM model will function as a 3D base map of the cities in the future, including linkages between 3CIM data and city databases/operational systems (such as building permit systems). In 3CIM, the building theme is based on CityGML 2.0 with the addition of some specific attributes (added as an ADE). The 3CIM model has also linkages to the national specifications for building data NS building (Lantmäteriet, 2023) jointly produced by the Swedish National Mapping Agency (NMA; Swe: Lantmäteriet), the Swedish National Board of Housing, Building and Planning (Swe: Boverket) and other national/regional/municipal actors. NS building includes descriptions on the physical divisions of a building (e.g., building, building part, building accessories) along with the administrative divisions of a building (e.g., residential, commercial, educational, etc.).

The specifications for CityGML, 3CIM and NS building do not specify in detail what information should be stored in the datasets. Therefore, they must be complemented with measuring guidelines that provide detailed specifications of the datasets for both producers and users. A measuring guideline which is the basis for several other guidelines is created by SIG3D Quality Working Group (2017). In Sweden, there is ongoing work to create measuring guidelines as part of the National specifications for buildings (Lantmäteriet, 2023) that will also be used by the 3CIM specifications. Below these are denoted as NS measuring guidelines.

It is not yet settled how the specifications above will influence the future Swedish 3D city models and digital twins (Ketzler et al., 2020) but a likely scenario is as follows. The NS building – together with other national specifications – will specify the content of the municipality production geospatial databases along with the measuring guidelines. Then the 3CIM model (or its successor) will specify how a subset of the data in the national specifications as well as links to other data sources will form the 3D city model. In this perspective, it is vital that NS building, 3CIM and the NS measuring guidelines support all necessary simulations (e.g., daylight, noise, flood, etc.) for sustainable urban planning.

2.3 Daylight studies

Though a number of studies have been conducted on the topic of semantic 3D city models (e.g., CityGML) and sunlight from a solar energy perspective (Catita et al., 2014; Biljecki et al., 2015; Saran et al., 2015; Prieto et al., 2019; Malhotra et al., 2021; Han et al., 2022), there are limited studies on semantic 3D city models from a daylight perspective. An indicative example of research combining solar energy and CityGML is the development of the Solar Energy ADE for performing energy analyses at an urban scale (Agugiaro et al., 2018). While different ways of computing solar irradiation on roofs to determine their level of suitability for installing photovoltaic solar panels have been investigated extensively, this is less the case for building facades (Desthieux et al., 2018). Saretta et al. (2020) present an indicatory list of studies dedicated to investigating the solar energy potential of façades at the urban/district scale. They also provide an example of how solar radiation analysis performed over LOD2 CityGML models can be matched with LOD3 façade archetypes to produce more realistic estimates regarding the potential of installing Building-Integrated Photovoltaics (BIVPs) on facades.

Few studies utilise city models in daylight metric computations and even fewer examine the corresponding input data requirements. This is the research gap this paper attempts to shed some light over. One such study was performed by Brasebin et al. (2012), who evaluated the impact of 3D geometric modelling on the Sky View Factor (SVF) corresponding to data stored in two separate databases including 3D city models represented in different levels of detail (LOD). Another research output example combining city models and daylight metrics is the Pyliburo Python library, which was developed to provide optimization capabilities to CAD-user workflows (Chen & Norford, 2017). The library has adopted CityGML as its internal data processing format and has been tested by computing the Floor to Area Ratio (FAR) metric over large geographic areas. Even though the number of studies in this field is limited, it is likely that the interest in city models will grow as the sustainable urban densification trend requires daylight-related input information on both planned and existing buildings (where, CAD¹/ BIM² data are often missing and only city model data are available).

3 Geospatial input data requirements for daylight metrics

Simulations of daylight metrics in e.g., densification projects require data for existing and planned buildings as well as some other objects (e.g., vegetation). In general, information on existing buildings is found in geospatial datasets, e.g., 3D city models, while information on planned buildings is found in CAD/BIM datasets. In Section 3.1, we investigate the need for geospatial data in the computation of daylight metrics, and in Section 3.2 we link this need to the geospatial data specifications.

3.1 Geospatial data requirements of daylight metrics

To investigate the need for geospatial data in daylight simulations, we made an inventory of the geospatial data requirements of several metrics (Tables 1, 2, 3 and 4). The selection of metrics is based on current national and European legislations and recommendations as well as on established practices by domain experts and research literature. A detailed description of these metrics can be found in Czachura et al. (2022).

Table 1 Data requirements - geometric metrics for daylight simulations

Table 2 Data requirements—latitudinal metrics for daylight simulations

Table 3 Data requirements - external climate metrics for daylight simulations

Table 4 Building and vegetation data requirements - internal climatic metrics for daylight simulations

3.2 Geospatial data specifications in the context of daylight metrics

To support the need for geospatial data in daylight simulations (summarized in Tables 1, 2, 3 and 4) it is important that the data specifications are adequate. In this section, we evaluate the 3CIM, the NS building, and the NS measuring guideline specifications for computing daylight metrics. It should be noted that there are currently only preliminary versions of the specifications (3CIM - Version 1.0; NS Building – Version 1.0 – Test 5; NS measuring guidelines – Version 1.0 – Test 4), and likely some details in these specifications will be changed. This is an additional motive for our study since we aim to influence the development of these specifications so that they can support daylight metric computations. A summary of the specifications for the building and vegetation data types that are most relevant when computing daylight metrics (cf. Tables 1, 2, 3 and 4) is presented below. Specifications for CAD/BIM data are not treated.

3.2.1 3D model – exterior

The allowed LODs in NS Building are 0.1, 2.1, 2.2 and 2.3 (Fig. 1). All buildings in Sweden should be represented by at least one of these LODs, and there is also a requirement that the building data must be updated. NS building state that if the modelling is done in LOD 0.1, it must include an attribute that defines which definition of height is used. Furthermore, the NS measuring guidelines specify that roof overhang (of at least 0.2 m) must be represented in LOD 2.3 and is allowed to be represented in LOD 2.2. The roof types are classified according to the INSPIRE building theme (Fig. 2).

Fig. 2

Selection of common roof type categories featured in the INSPIRE data specification for buildings (v 3.0)

The building geometries in NS building consist of surfaces with id-numbers (and not only solids). This entails that it is possible to add specific attributes to individual surfaces. The measuring guidelines specify that a wall should be represented by a vertical surface (but an attribute could be used for non-vertical walls) and that curved walls are represented by multi-surfaces.

There is no detailed requirement on positional accuracy of building data in any of the specifications. But the national specifications refer to guidelines in other geospatial data quality documents. In short, these guidelines state that (municipal) geospatial data in urban environments should have a positional accuracy corresponding to a root mean square of a decimetre or better. To guarantee this, the guidelines state an expected positional accuracy for some measuring techniques. NS building also requires that the measuring methods are specified as part of the metadata on the object level.

3.2.2 3D model – interior

NS building does not include indoor modelling of buildings but allows references to external geometries in CAD/BIM models. Currently, it is hard to estimate to which degree CAD/BIM models from e.g., the building permit process will be saved and linked to the future NS building datasets. There is no requirement that these models should be updated. The 3CIM model allows all LODs in CityGML 2.0 including LOD4. This implies that 3CIM does allow indoor data, but it will probably be rare for municipalities to use city models higher than LOD2. And most likely they will strive to use corresponding LODs to NS building i.e., LOD 0.1, 2.1, 2.2 and 2.3, as well as LOD 1.x that easily could be transformed to LOD 0.1.

3.2.3 Window information

As noted above, 3CIM does allow all LODs in CityGML 2.0, but the higher LODs are likely not to be used often in practice. There is a need for window information from many applications, including daylight simulations, but the specifications do not provide an easy way to add this information (unless a CityGML LOD3 or CAD/BIM model are created with all their complexities).

3.2.4 Building facade reflectance properties

The situation for reflectance properties is similar to the one previously described for windows. That is, though there are several applications that would benefit from them (e.g., daylight and noise simulations), the current specifications do not support storing these properties unless a higher LOD (or CAD/BIM) is used.

3.2.5 Balconies & other building accessories on facades

The NS measuring guidelines state that all new measurements of buildings and other building façade accessories should be in LOD 2.1 or higher. These accessories could be represented as either surface (Fig. 3) or volume objects depending on LOD. In the measuring guidelines it is stated that in urban areas all balconies and other accessories on facades should be represented only if: (1) they protrude at least 0.2 m from the façade and (2) their total area is at least 2m².

Fig. 3

Surface representation of a balcony (NS measuring guidelines Version 1.0, Test 4, p. 48)

3.2.6 Lifecycle handling of building information

NS building specify buildings (and building parts) at several stages in the building lifecycle. To identify which phase a building is in, there is a status attribute with the following allowed values: preliminary, planned, under construction, current, dilapidated, ruin, and teared down. There is a requirement that all (planned) buildings should be represented in 3D at least 6 months after the building permit is approved. Before the building permit is approved some information about possible future buildings can be extracted from the detailed development plans. The 3CIM building theme has a status attribute on the building parts with the same attribute values as the NS buildings.

3.2.7 Vegetation

The 3CIM model contains a theme called Vegetation that includes both plant cover and solitary vegetation objects (following the CityGML standard). However, though Swedish municipalities most commonly maintain a record of vegetation (trees) on municipality-owned land, this is not always the case for trees on private land. These tree records often consist of a 2D point-layer describing the name, main type (deciduous, coniferous) and year a tree was planted with no further information on its height or crown diameter. Building a 3D-layer based on this information on a municipal-level is not trivial, but a method of doing so in different LODs is presented by Ortega-Córdova (2018). Vegetation is not included in any of the use cases as a big part of our study area consists of privately-owned land where this information is unavailable.

4 Case studies

Two case studies were performed to evaluate how well the geospatial data (as defined by the currently available specifications) meet the requirements of daylight metrics that are regulated by Swedish and European laws and recommendations (see Appendix 2): obstruction angle (OA), sunlight exposure, and daylight factor (DF). Case study I implements the OA metric, which, from a geospatial data perspective, is mainly affected by the level of detail (LOD) and positional accuracy of buildings which are, therefore, the focus of the study. Case study II consists of two parts. The first part concerns the sunlight exposure metric for which knowledge of obstructions of the direct sunlight is vital. In this part, we study the effect of building accessories (e.g., balconies and balcony railings) on the modelling of sunlight exposure. The second part concentrates on the DF metric. One important aspect of correct modelling of this metric, especially in a dense urban area, is the reflection of sunlight on building façades. Hence, this part of the case study focuses on the possible need for modelling building façade materials with different optical reflection properties. It should be noted that vegetation is not included in any of the use cases since the studies are designed to address the question of how building objects should be represented.

There are several tools for daylight simulations (see e.g., overviews in Freitas et al., 2015 and Jakica, 2018) based on CAD, BIM, and GIS platforms. For case study I we used a GIS-based tool and for case study II a CAD-based tool.

4.1 Case study I - study of LOD and positional accuracy requirements of the OA metric

4.1.1 Background

In this study, we evaluate how the OA metric is influenced by (1) different roof modelling strategies and LODs, and (2) the positional accuracy of the building footprint. For the first part of the study, we evaluate the OA metric by comparing its generated simulation output when executed over 3D city models of varying LODs; LOD2.2 and LOD1.3 models; where the building heights of the latter are expressed based on three different definitions of roof height (Fig. 4): roof edge of building (RE), roof ridge (RR), and mean roof height (MRH) computed as the mean of RE and RR.

Fig. 4

Roof height definitions (upper row; defined according to NS Building – Version 1.0 – Test 4) and their corresponding LOD1.3 models (lower row)

For the second part of the study, we examine the effect of positional accuracy (of the building footprints) on OA computations by comparing the results when computed over LOD1.3 models of the same roof height created using building footprints; from two sources: (1) the National Mapping Agency (NMA) and (2) the Lund Municipality 3D city model (LM). An illustration of limited positional accuracy, and variation in building footprint representation, is given in Fig. 5.

Fig. 5

Example of differences in building footprints between datasets from the National Mapping Agency (NMA) and Lund Municipality 3D city model (LM). The positional accuracy of the NMA data is on dm-level and so are the on average differences between the two footprint representations

4.1.2 Study area

The study area is situated in the central part of Lund, southern Sweden (at a latitude of 55 degrees) (Fig. 6). It is divided into 15 neighbourhoods, where each neighbourhood is treated separately in every test case. The purpose of the test case is to simulate the OA on one or several facades of a building. To enable these simulations the neighbourhood must contain the buildings that affect the obstruction angle. The study area is carefully chosen to ensure a great variety of building typologies and roof types (reflecting both modern and older architectural trends) in the OA simulations. It also includes some examples of densifications that have taken place at different points in time (e.g., 1899, 1960, 1980, 2012). The study area consists of roughly one third gabled roofs, and a large amount of flat and monopitch roofs.

Fig. 6

Study area in Lund, southern Sweden with examples of some of the building neighbourhoods

4.1.3 Data collection and tailoring

The case study utilizes building objects from the Lund municipality 3D city model which was initially constructed by a consultant company. This model was generated from aerial images (0.1 m spatial resolution) and represents roof constructions on an LOD2.2 level of detail. The model also uses building footprints obtained from the municipality base map based on terrestrial measurements. The building footprints had their elevation either measured or derived from a digital elevation model (DEM). The building volumes are an intersection between the photogrammetrically measured roof model and an extrusion of the building footprints. Buildings were split into different building parts when the roof had a vertical shift of 0.5 m or more, continuously through the entire building. Buildings smaller than 12m² were not included in the 3D city model and therefore not used in this study.

To support the aim of the study we had to: (1) change the LM model so that it followed the NS building measuring guidelines, (2) add window information, (3) divide the buildings into building parts depending on their roof structure or overhangs, and (4) define the height values for each building part (Fig. 7).

Fig. 7

Example of a building part with multiple candidate roof edge heights of different values. Screenshot for Karl XII Gatan 1, Lund, Sweden. The image from the left is from Google Earth

The OA should be measured for all windows of a target building in every neighborhood. To enable that, we collected window information (photos) and modeled these images as part of the 3D model. The majority of the window 3D information was created by taking photographs of building facades at right angles so that façade elements were not distorted. Additionally, a 2 m spatial resolution DEM was downloaded from the NMA.

For the second part of the case study – where we examine the positional accuracy of the footprint – we used the 2D building footprints from the NMA map in scale 1:10,000 (Lantmäteriet’s Fastighetskarta). The level of detail for the building footprint in this dataset is similar to the level of detail in the municipality map and the difference in absolute position of the footprints is on decimetre-level. Based on the NMA data we constructed LOD1.3 building objects using the same heights as used for the LM model (Fig. 4). Then we computed and compared the difference in OA for RE LOD1.3 buildings created using NMA and LM building footprints. The same process was applied for RR LOD1.3 buildings.

4.1.4 Compute obstruction angle (OA)

A separate tool was developed in ArcPy³ to compute the general case of a window’s OA. The tool requires three input files with information on:

• windows (Multipatch feature class)

• Digital Surface Model - DSM (Raster dataset)

• 2D building footprints (Polygon feature class).

The window-datafile contains 3D information on rectangular windows placed on vertical façade surfaces. Building-ID information is also stored for every window. The 3D buildings of every neighbourhood are converted to rasters (0.1 m spatial resolution) and added to a DEM (resampled to 0.1 m) to produce the DSM. The 3D buildings are converted to 2D building footprints and every building is uniquely defined by an ID (same building-IDs that are used in the windows-datafile).

The tool⁴ computes the orientation and centre point of each window. A viewshed analysis over the DSM raster is executed for every centre point to identify which cells are seen within a horizontal scan range from 85—95° (vertical distance to obstruction). From the two produced viewshed rasters, the one that is empty or overlaps with the building footprint is deleted. The cells of the remaining viewshed raster are converted to points and the OA is calculated for all points according to the following equation. The max OA is added as a result in the corresponding column of the window layer’s attribute table.

$$OA=\mathrm{arctan}(\frac{\left|{window}_{{point}_{z}}- {viewshed}_{{point}_{z}}\right|}{\sqrt{{\left({window}_{{point}_{x}}-{viewshed}_{{point}_{x}}\right)}^{2}+{\left({window}_{{point}_{y}}-{viewshed}_{{point}_{y}}\right)}^{2}} })$$

Where ${point}_{z}$, ${point}_{x}$, and ${point}_{y}$ refer to the window centroid and viewshed point’s height, easting, and northing values correspondingly.

4.1.5 Result

The OA-results for every window are post-processed to compute the difference between the generated OA-values per 3D city model (LM, RE, MRH & RR). The analysis is performed separately for the different roof types (Fig. 2). Only results for the gabled roof types are reported as they are the ones showing the largest differences indicating that this is a roof type with great need for LOD2 modelling. Though cone and sawtooth roofs are showing similar results, there are too few buildings of that roof type to draw definite conclusions.

Figure 8 illustrates the difference in OA-values per window between the LM LOD2 3D city model (LM) and the corresponding OA-values of the LOD1.3 models (RE, MRH & RR). The difference in OA-values between the models depends on the inclination of the gabled roof, the height of the window centroid, and the angle at which a window is facing the opposite gabled-roof building. It should be noted that the LOD1.3-RE is problematic for small obstruction angles. However, in practice, small angles are not important in daylight simulations. What is most interesting is to view the effect of the roof modelling on the larger OA-values which usually correspond to windows of the ground floor. From Fig. 8, it becomes evident that though there are some similarities, there are also cases where the 4 models produce very different OA-values.

Fig. 8

Difference between computed OA for the LM LOD2 3D city model and the LM LOD1.3 3D models (RE, MRH & RR) for windows whose obstructing point is located on a gabled roof. The models are ordered with increasing OA values (using the LM model), i.e., the window-ID with the lowest OA is the leftmost

The biggest deviations in OA-values are observed between the LOD2-LM and LOD1.3-RR models. This is expected as, in most cases, the position of the obstruction point maintains its height but is transferred much closer to the window centroid, generating a higher OA. Fluctuations in the RR-LM values could be attributed to windows of different floors (i.e., the OA-value of windows of higher floors is expected to increase less in comparison to windows of the ground floor). The average RR-LM OA-value difference is 4.6° with a standard deviation of approximately 4°, meaning that the difference can potentially exceed 10°. On the contrary, the LM-LOD2 and LOD1.3-RE models seem to generate more similar OA-values (RE-LM curve in Fig. 8). This is attributed to the fact that the top of a gabled roof is often not visible from windows in opposite buildings that are located on floors closer to the ground floor. Consequently, there is a high probability the obstruction points of those windows being placed at the same location (roof edge) in both LM-LOD2 and LOD1.3-RE OA simulations. Other possible reasons explaining why the RE-LM curve is so smooth, could be the low inclination of the obstructing gabled roof as well as the low height of buildings facing gabled roof buildings (i.e., buildings in neighbourhoods dominated by gabled roof types rarely have more than 4 floors). In general, the values for RR-LM and MRH-LM are higher than those of RE-LM. The only exception occurs when the obstruction point of a window remains on the same building for LOD2-LM and LOD1.3-RE but shifts to another building for the LOD1.3-RR and LOD1.3-MRH. This is the result of the combined effect of the roof type, height, and placement of the buildings opposite that window as well as the design of the OA-tool, which is programmed to select the point generating the highest OA-value as the obstruction point within a 10° horizontal radius from the window centroid.

Figure 9 presents the regression analysis results of the difference between OA-values computed using LOD2 (LM) and LOD1.3 (RE, MRH & RR) models when the obstruction point of the LOD2 model is located on a gabled roof. The results show a significant positive correlation between the increase in measured OA-value for the LOD2-LM model and its corresponding difference from the LOD1.3 models. The difference in OA-values between these two groups of models generally increases for larger obstruction angles (especially in the LOD2-LM and LOD1.3-RR case, where the difference is often in the range of 10°), pointing towards the fact that the bigger the computed LOD2-LM obstruction angle, the larger the level of uncertainty as to the OA-value computed for its corresponding LOD1.3 3D model.

Fig. 9

Regression analysis over computed OA for the LM LOD2 3D city model and the LM LOD 1.3 models (RE, MRH & RR) for windows whose obstructing point is located on a gabled roof

Regarding what effect the positional accuracy has on the computed OA, we find that decimetre size -differences in the building footprints cause the largest shift of the OA-result for RR & RE LOD1.3 buildings. This shift is less than 1° and on average about 0.2°.

4.2 Case study II - influence of building accessories and façade reflection properties on the sunlight exposure and daylight factor metrics

4.2.1 Background

In case study II, we evaluate the sunlight exposure and daylight factor metrics by examining two things: (1) the effect of balconies and balcony railings in sunlight exposure simulations and (2) the effect of including true reflectance properties for different combinations of façade materials and colours of surrounding buildings in the computation of the DF metric.

4.2.2 Study area

The study area is located in the newly built Hyllie-neighbourhood of Malmö municipality, southern Sweden and is characterized by modern high-rise office, commercial and residential buildings with relatively flat roofs (Fig. 10). It is an area that was recently densified, with a high building density for Swedish standards. Therefore, studying the effect of different reflectance properties from the most common building façade materials seemed more appropriate. The high building density was also the reason why we decided that it would be preferable to test the effect of balconies on direct sunlight hours in Malmö (an attempt to, in a way, examine the worst-case scenarios).

Fig. 10

Solkvarteret Hyllie, Malmö, southern Sweden. The buildings in focus of the study are highlighted

4.2.3 Input data

The surrounding buildings are modelled in the Malmö city 3D building model, corresponding to LOD1.3 and LOD 2.2. The focus buildings consist of a CAD-model corresponding to LOD4. A total number of 56 balconies are modelled in the focus building. Balcony areas range from 3.4 to 10.5m², while balcony depth’s range from 1.25 to 2.26 m distance from the façade. Take note that some of the balconies are built into the façade, meaning that they protrude less from the roof edge than they do from the façade (cf. Fig. 11). A total number of 285 windows and 40 glazed balcony doors are modelled with areas ranging from 0.9–3.15m² and 1.6–1.8m².

Fig. 11

DF simulations over living room areas at different floors of the focus building

4.2.4 Methodology

The simulations of sunlight exposure⁵ and daylight factor⁶ were conducted in the Rhinoceros⁷ CAD-environment using the HoneyBee,⁸ LadyBug,⁹ and Climate Studio¹⁰ plugins to the Grasshopper¹¹ scripting environment.

The output of the sunlight exposure simulation was a regular grid of points with 0.1 m spatial resolution covering aperture areas (i.e., windows and glazed balcony doors). The grid registered total direct sunlight hours at a 30 min temporal resolution on March 21^st at the latitude of Malmö. The output was further processed using ArcGIS Pro to extract the grid-point value closest to an aperture’s centroid, according to the EU-recommendation (mid-width of window and 0.3 m above window’s lowest point). Statistics were produced to compare the direct sunlight hours registered at the aperture centroids for simulations executed with different obstructions:

• Surrounding buildings (SB)

• Surrounding buildings & balconies (SB + B)

• Surrounding buildings, balconies & balcony railings (SB + B + BR)

For the DF-simulation of one apartment within the building, a horizontal grid of points (0.2 m resolution) was set at a height of 0.8 m from the floor and 0.5 m away from the interior walls of a living room. The simulation was executed at different floors for a room with the same layout location (Fig. 11) using different obstructions (including or excluding balconies and balcony railings) and reflectance properties of various combinations of façade materials and colours (Table 5).

Table 5 Physical characteristics of façade materials used in the DF-simulations

Specular and Diffuse refer to the percentage of the incident solar radiation that is reflected by specular or diffuse reflection, while Reflectance is expressed as the sum of the two. Roughness¹² describes the material’s smoothness and is specified as the root mean square facet slope of a Gaussian surface, where 0.0 corresponds to smooth and 1.0 to rough surfaces. Colour is represented by its corresponding HSV-value, which is a measure of brightness ranging from 0 to 100%. An HSV-value close to 0% will be totally black (dark) while values near 100% are white (bright). The HSV-value was obtained from the façade colour by extracting the max of its RGB-values (expressed with values ranging from 0–1 instead of 0–255).

In total, 156 simulations were carried out. The output was processed to produce the DF_median and ADF as well as the DF range and the % of room area fulfilling the BBR-requirement (DF > 1%) (cf. Appendix 1).

4.2.5 Result

Sunlight exposure (SE) simulations

Figure 12 shows an example of a window (see highlighted window) that fulfils the EU-recommendation of 1.5-h of direct sunlight, when obstructing buildings are presented in lower LODs, but seizes to do so when balcony and balcony railing geometries are included as obstructions, proving how incoming sunlight may be restricted by balconies in densely built environments. From the left subfigure of Fig. 12 it is evident that windows on lower floors receive the least amount of direct sunlight. The inclusion of balconies as obstructions, as depicted in the central subfigure of Fig. 12, limits the direct sunlight access of not only windows placed exactly below them but also of nearly located windows (see windows that do not have balconies on the 1^st and 2^nd floor) in the same or neighbouring building. When also balcony railings are considered, the direct sunlight access is further restricted even for windows on slightly higher floors (see window with no balcony on 2^nd floor in the right subfigure of Fig. 12). Again, the effect is observed in both windows belonging to the same building as the balcony railings as well as neighbouring buildings.

Fig. 12

Example of a window to a room that fails to meet the 1.5 direct sunlight hour EU-requirement when balcony and balcony railing geometries are included as obstructing surfaces in the Sunlight Exposure simulation

Table 6 illustrates the effect of including balconies and balcony railings in sunlight exposure simulations. Not all windows are affected (38/285 or 49/285), but of those that are affected about 30% fail to meet the 1.5-h of direct sunlight requirement.

Table 6 Difference in sunlight exposure simulation results depending on level of detail of obstructing buildings

Figure 13 presents the dispersion of the differences in direct sunlight hours for windows that are affected by adding balconies and balcony railing geometries as obstructions. A total of 38 windows were affected by the inclusion of balconies as obstructing geometries, while 49 windows showcased a decrease in direct sunlight hours when both balconies and balcony railings were considered. In both cases, the number of windows, whose access to direct sunlight hours decreased between 1.5 to 3.5 h, seems to have remain unchanged. However, when balcony railings are added as obstructions, then the number of windows previously receiving 0.5 to 1.0 h less direct sunlight drops while the corresponding number of windows receiving a decrease of 1.0 to 1.5 direct sunlight hours increases.

Fig. 13

Difference in direct sunlight hours between simulations executed over 3D city models with and without balcony and balcony railing geometries (SB + B or SB + B + BR)

Daylight factor (DF_median)

including balconies and façade colour information seems to have a considerable effect on the results of the DF-simulations and the likelihood of rooms in the lower floors meeting the Swedish (SE) and European (EU) daylight requirements (Figs. 14 and 15). The inclusion of balcony geometries as obstructions, limits the cases where rooms in lower floors meet the daylight requirements (Fig. 14). In both cases, the European recommendations are only met by rooms located at higher floors. A bright façade colour may positively influence daylight access for rooms at lower floors in a densely built urban environment (Fig. 15). For rooms at higher floors, daylight access is less influenced by the colour of neighbouring facades.

Fig. 14

Percentage of rooms for which the Swedish (SE) or European (EU) DF-requirements are met depending on whether balconies and balcony railing geometries are included as obstructions or not

Fig. 15

Percentage of rooms for which the Swedish (SE) or European (EU) DF-requirements are met depending on whether the colour of the obstructing building facades is dark (HSV-value < 50) or light (HSV-value >  = 50)

Including information of façade material in the DF-simulations may have a substantial effect on the possibility of rooms in the lower floors meeting the corresponding SE & EU requirements (Fig. 16).

Fig. 16

Percentage of rooms for which the Swedish (SE) or European (EU) DF-requirements are met depending on the reflectance properties of the materials used to build the obstructing building facades

Table 7 presents the relationships between simulation input variables and results that determine whether a European or a Swedish daylight DF-requirement/recommendation has been met for a living room on the ground floor. The ground floor is chosen (Figs. 14, 15 and 16) as lower floors have less access to direct diffused sunlight and are therefore more likely to fail in meeting the requirements. The Pearson correlation results show that there is a strong positive correlation between using lighter façade colours and materials with higher reflectance and meeting the Swedish DF-requirement for rooms on the ground level. On the contrary, using façade materials with high roughness and including balcony geometries in the DF-simulations had a statistically significant negative correlation to meeting the DF-requirement.

Table 7 Pearson correlation (P_corr) results & regression p-value (p) for façade attributes & DF-criteria on the ground floor

As Reflectance shows high multicollinearity (variance inflation factor > 5) with other optical properties and the HSV-value, it is treated separately. The linear regression models for all variables in Table 7 are expressed as:

$$\begin{array}{cc}{\mathbf Y}_{\mathbf1\mathbf A}\boldsymbol=\mathbf0\boldsymbol.\mathbf{91}\boldsymbol\ast{\mathbf X}_{\mathbf1\mathbf A}\boldsymbol-\mathbf{19}\boldsymbol.\mathbf{26}\boldsymbol\ast{\mathbf X}_{\mathbf2\mathbf A}\boldsymbol-\mathbf{61}\boldsymbol.\mathbf{61}\boldsymbol\ast{\mathbf X}_{\mathbf3\mathbf A}\boldsymbol+\mathbf{26}\boldsymbol.\mathbf{05}&\boldsymbol(\mathbf R^{\mathbf2}\boldsymbol=\mathbf{94}\boldsymbol\%\boldsymbol)\\{\mathbf Y}_{\mathbf1\mathbf B}\boldsymbol=\mathbf0\boldsymbol.\mathbf{98}\boldsymbol\ast{\mathbf X}_{\mathbf1\mathbf B}\boldsymbol-\mathbf{19}\boldsymbol.\mathbf{26}\boldsymbol\ast{\mathbf X}_{\mathbf2\mathbf B}\boldsymbol+\mathbf{12}\boldsymbol.\mathbf{75}&\boldsymbol(\mathbf R^{\mathbf2}\boldsymbol=\mathbf{98}\boldsymbol\%\boldsymbol)\\\begin{array}{c}{\mathbf Y}_{\mathbf2\mathbf A}\boldsymbol=\mathbf0\boldsymbol.\mathbf{02}\boldsymbol\ast{\mathbf X}_{\mathbf1\mathbf A}\boldsymbol-\mathbf0\boldsymbol.\mathbf{23}\boldsymbol\ast{\mathbf X}_{\mathbf2\mathbf A}\boldsymbol-\mathbf1\boldsymbol.\mathbf{39}\boldsymbol\ast{\mathbf X}_{\mathbf3\mathbf A}\boldsymbol+\mathbf0\boldsymbol.\mathbf{16}\\{\mathbf Y}_{\mathbf2\mathbf B}\boldsymbol=\mathbf0\boldsymbol.\mathbf{02}\boldsymbol\ast{\mathbf X}_{\mathbf1\mathbf B}\boldsymbol-\mathbf0\boldsymbol.\mathbf{23}\boldsymbol\ast{\mathbf X}_{\mathbf2\mathbf B}\boldsymbol-\mathbf0\boldsymbol.\mathbf{17}\end{array}&\begin{array}{c}\boldsymbol(\mathbf R^{\mathbf2}\boldsymbol=\mathbf{68}\boldsymbol\%\boldsymbol)\\\boldsymbol(\mathbf R^{\mathbf2}\boldsymbol=\mathbf{71}\boldsymbol\%\boldsymbol)\end{array}\end{array}$$

where:

Y_1A, Y_1B = Percentage of indoor floor area meeting the DF > 1% requirement

Y_2A, Y_2B = Room meeting the SE DF-criterion

X_1A = Colour (HSV-value)

X_2A, X_2B = Balcony & balcony railing geometries¹³

X_3A = Roughness

X_1B = Reflectance

5 Discussion

5.1 Lessons learned from the case studies

Semantic 3D city models have the potential to meet all geospatial data input requirements of daylight simulations (cf. Tables 1, 2, 3 and 4). Which input requirements are met depends on the available LOD. Information on window locations, optical properties of façade materials and the building interior (e.g., room floor geometry) is important for conducting daylight simulations in densification projects. One problem for incorporating this information into a CityGML2 model is that a LOD4 model is required, which implies a very detailed and expensive model. There are some methods developed for making e.g., window information more easily available (Fan et al, 2021), but work is still needed to ensure that building and window objects are well aligned. Some researchers (e.g., Tang et al., 2020) argue that we should use a more flexible LOD concept that is more application-driven (e.g., allowing objects previously related to higher LODs to be combined with objects of lower LODs). This has already been implemented in the CityGML3 standard (Kolbe et al, 2021), but it remains to be seen how it will be utilized in future city models, and if this will influence the Swedish geospatial data specifications. Nevertheless, it is clearly interesting from a daylight simulation perspective.

From part 1 of case study 1, it becomes apparent that it is possible to use an LOD1.3 3D city model for geometric daylight metrics, such as the obstruction angle, since the difference in degrees with its corresponding LOD2 3D city model rarely exceeds 1° for most roof types. However, neighbourhoods characterized by buildings with gabled roofs should preferably be handled utilizing 3D city models whose LOD allows for a more detailed representation of the roof geometry as the difference in OA degrees sometimes exceeds 10° (e.g., Figs. 17 and 18). The regression analysis (Fig. 9) shows a significant positive correlation between the increase in measured OA-value for the LOD2 LM 3D city model and its corresponding difference from the RR NMA LOD1.3 3D model, pointing towards the fact that the bigger the computed LM obstruction angle, the larger the level of uncertainty as to the OA-value computed for its corresponding LOD1.3 3D model. Besides gabled roofs, this is also true for roof types such as cone roof and pyramidal broach roof, where replacing the LOD2 city model with a LOD1.3 3D model either overestimates (roof ridge) or underestimates (highest roof edge) the building/building part height and obstructing capacities.

Fig. 17

Example of roof with large difference between roof edge heights on Stora Fiskaregatan 13A, Lund, Sweden

Fig. 18

Examples of how lower LOD versions of the same building may generate higher OA than their corresponding LOD2 counterpart, when the roof type is gabled roof

Part 2 of case study 1 evaluates the positional accuracy of building footprints that are extruded to produce 3D city model objects. The case study further investigates any possible effect this may have on the results of geometric daylight metrics such as OA. Most of the results show that both LOD2 and LOD1.3 yield similar OA-value results that don’t differ more than 1°, verifying that if 3D buildings & NMA building footprints follow the positional accuracy specifications for measuring buildings, this will be enough for estimating the OA metric. The results are in-line with findings in Biljecki et al. (2015) who examined the effect of positional deviations on estimations of solar irradiation on building roofs and found that a positional uncertainty within the range of 0.3–0.6 m yielded an error of 10% in the estimated results.

Part 1 of case study 2 examines the effect different LODs have on the direct sunlight exposure simulation. When computing direct sunlight hours, to test whether the European recommendation of a room (that is frequently visited) receiving at least 1.5 h of direct sunlight during one of the days between February 1^st and March 21^st is met, we can see that façade elements such as balconies and balcony railings may play an important role in determining whether a room conforms to the requirement, especially when the buildings are located in a more densely built urban environment. The results show that though the number of windows being affected by the addition of the balcony and balcony railing geometries is comparatively low (13%-17%), the risk of the affected ones not fulfilling the requirement is not negligible (30%) (Table 6). It should be noted that for windows that are affected, the difference in direct sunlight hours for 50% of them (22/38) ranges from 1.5–3.5 h (SB + B) while the difference for 25% of them (22/49) ranges from 1.5–3.5 h (SB + B + BR) (Fig. 13). Though including balcony railings leads to an increase of the windows getting restricted sunlight, only additionally two fail to meet the requirement. However, there is a clear shift in the total number of windows showing a difference between 1.0 and 1.5 h (Fig. 13).

Part 2 of case study 2 examines the effect of including information on the optical properties of the surrounding buildings on the output of the DF simulation. The results prove the importance of including this information in the simulation of dense urban environments and the value of storing and maintaining this information in semantic 3D city models. The effect is significant for rooms located on the lower floors and the ground floor (Figs. 14 and 15). Including façade material and colour information can make all the difference in estimating the amount of property unit land area that can be built (regulated in the detailed development plan) and be a determining factor in deciding the actual number of apartments that could be built to accommodate the increased demand in housing solutions for an ever-growing urban population when valuable agricultural land in the suburbs needs to be preserved. This is shown by the strong positive correlations between the colour, reflectance, and roughness properties of the façade and the corresponding Swedish and European daylight-criteria (Table 7). By using bright colours on materials with high reflectance and uneven surfaces that cause the light to reflect diffusively, even ground floor rooms in dense urban environments have the potential to meet the currently existing daylight requirements (Fig. 16). Other research confirms that façade material and colours are indeed important for increasing access to daylight, while emphasising that they may also cause dangerous implications such as glare (Danks et al., 2016).

Balconies have been found to significantly impact daylight access in building interiors (Loche et al., 2021), so here we investigate the extent of this impact in densely built environments. The Pearson correlation results show a strong negative relationship between DF-results and including balcony geometries that protrude more than 1.25 m from the building façade in the DF-simulations (Table 7). This relationship is statistically significant when examining compliance to the Swedish daylight-requirement (Table 7) and influences more rooms located in lower floors, where not including balcony geometries in the DF-simulation may increase the probability of meeting the Swedish criterion by 13.7% (Fig. 14). Similar trends are observed for the DF-related EU daylight-recommendation (Fig. 16).

Another factor that could potentially affect daylight simulation output is the resolution of DSMs representing 3D geometries of buildings as well as the resolution of sensor grids. Desthieux et al. (2018) conducted a solar irradiation analysis on building facades and roofs for the city of Geneva using a DSM with 0.5 m resolution. The results for selected facades were evaluated by comparison to those produced by standard commercially available PV-simulation software (PVSyst) and found to correlate well. In our case, the resolution of the DSM (0.1 m) used in this paper’s first case study (OA metric) was high enough to represent all roof structures and façade accessories available in the LM LOD2.2. city model. Additionally, the sensor grids used in case study 2 were 0.1 m for the Sunlight Exposure metric and 0.2 m for the DF metric. Peronato et al. (2018) performed a solar irradiation analysis on building envelopes using sensor grids of different resolutions (0.5 m, 1 m, 2 m, 3 m and 4 m). A sensitivity analysis was performed to evaluate the simulated results and it was proven that sensor grids with a resolution up to 2 m fitted well with the real measured values. Consequently, the choice of sensor grid resolution in both examples of case study II should not affect the simulation output, considering that the grid resolutions are much higher than those used in the aforementioned related study.

5.2 Limitations of the study

There are, in terms of appropriate daylight simulations, several limitations to this study. One such limitation is that we are only focusing on the geospatial data input to daylight metrics, omitting effects of e.g., climate and sky models. Another issue that is disregarded is the limitations in the definition of the daylight metrics (cf. Bournas, 2021; Eriksson et al., 2019). There are also limitations in the design of this study. In Section 3.1 we made a quite extensive investigation of the need for geospatial data for a large and representative selection of daylight metrics. But, on the other hand, we did not go into detail regarding what level of detail was necessary for the geospatial data per metric. For the two case studies the situation was the opposite. In that part, we studied more details about the requirement of level of detail and quality, but only for a limited number of daylight metrics. Even though these are metrics that are recommended (also regulated by law in some countries), there are many other daylight metrics (e.g., vertical sky component) used that potentially have other requirements of geospatial data. Furthermore, in the case studies we identified neighborhood areas that were suitable for the estimation of the metrics; but it should be noted that the areas are limited in size and in e.g., the building typologies present. To sum up, we acknowledge that the detail of the study is affected by the study design, but we claim that that the general conclusions are valid.

6 Conclusions

To estimate day- and sunlight metrics, in e.g., an urban densification context, there is a demand on information regarding 3D building exterior (building envelope) and interior (e.g., room and apartment floor geometries), windows, balconies, façade accessories, building footprint, and vegetation. Much of this information could be included in a 3D city model but it is important that the specifications and measuring guidelines ensure that the information is present and of good quality. The 3D city model specifications for 3CIM, being based on CityGML, have the potential of meeting these requirements with some difficulties related to acquiring detailed window, balcony, façade material/colour, and building interior information on existing buildings (for planned buildings this information is available from the CAD/BIM models that are submitted during the building permit process).

Vegetation was not treated in this study, as more research is needed to identify ways of creating 3D single vegetation objects from existing planar information in an automated way, maintaining it (this is a layer that changes in time), assuring good quality and finding the appropriate LOD it should be represented in per daylight metric.

Based on the case studies we would recommend the following remarks to city model specifications in general and the Swedish specifications in particular:

• NS building and 3CIM allow several LODs for building objects. Our studies show that LOD2.2 is to be recommended for daylight estimations. LOD1.3 with height defined at the highest roof edge could be an option for several roof types (flat roofs, monopitch roofs), but there are limitations for gabled roofs. The OA-value difference between LOD2.2 and LOD1.3 for gabled roofs often exceeds five degrees.

• The current recommended positional accuracy of the city model data (dm level) is satisfying.

• Balconies should be included according to the existing specifications. In the case of the sunlight exposure metric, including balconies in the simulations did not affect the direct sunlight access of many windows, but for those that it did, there was a 32% possibility of the window not meeting the 1.5-h of direct sunlight requirement.

• From the DF-simulations we derive that information on the façade material, colour, and spectral properties is essential and should be included in the 3CIM specifications. 3CIM being a semantic 3D city model, supports the extension of its syntax to include these parameters.

• Maintaining indoor building information for living rooms and bedrooms is necessary, if DF-simulations for existing buildings is to be tested in early stages of the urban planning process of densification projects.

Appendices

Appendix 1: Daylight metrics

This appendix includes a more detailed description of the daylight metrics used in this paper.

1.1 Geometric daylight metrics

1.1.1 Daylight factor (DF)

The DF (Fig. 19) is a daylight metric that corresponds to a percentage expressed as the ratio of the illuminance in a room compared to the illuminance on a horizontal plane under an unobstructed hemisphere with an overcast sky (Dubois et al., 2019).

Fig. 19

Daylight factor

It should be noted that the DF is only calculated for a single point based on diffuse light and does, therefore, not take into consideration orientation. Daylighting is very much climate-, time- and orientation-dependent. The metric’s failure to account for how climate-, latitude- and orientation-related variations affect the estimated daylight conditions, is rather a significant deficit (Mardaljevic et al., 2009).

Amongst the advantages of this metric is that its results can easily be described and understood by a design team (Reinhart et al., 2006). Moreover, it expresses the lighting potential of a room and its windows as well as giving a good indication of the contrast between indoor and outdoor lighting conditions (Love & Navvab, 1994).

1.1.2 Point daylight factor (DFp)

DF_p is described in the Swedish Building Regulations (BBR), where it is calculated as a single DF measurement taken at a single timepoint at a point located 1 m away from the darkest wall, in the middle of the room’s depth at 0.8 m above the floor’s surface (Fig. 20a).

Fig. 20

a Point daylight factor (DF_p) & b Median daylight factor (DF_median)

Recent research has shown that DF_p fails to provide representative estimations of indoor daylight conditions in cases where rooms have windows on different facades or when the shape of the room is irregular (Bournas & Dubois, 2018). Due to the limitations linked to the DF_p, researchers have come up with a different version of DF_p called Median Daylight factor (DF_median), which calculates the DF of a grid of points in a room and then returns the median of those DF-values (Fig. 20b) (Mardaljevic & Christoffersen, 2017). DF_p has been shown to correlate strongly with the DF_median across the room area, which indicates that complying with this criterion is equivalent to achieving a DF_median of 1% over 50% of the room area (Bournas, 2020). European recommendations regarding appropriate internal illuminance levels have been found to be equivalent to a DF_median of 2.1%¹⁴ (Mardaljevic et al., 2013) or an average daylight factor¹⁵ (ADF; Lynes, 1979) of 3.2% (BSI, 2008).

1.1.3 Obstruction angle (OA)

The BBR are imposing a set of requirements regarding what properties different proposed building typologies should have in relation to direct daylight access (see Appendix 2). As these requirements are often not checked before later stages of the urban planning process, to increase the probabilities of real estate developers meeting them, it is imperative to make use of the available technology when estimating daylight access potential at earlier stages of the urban planning process.

One way of achieving this, is by calculating the obstruction angle between neighbouring 3D/planar representations of the planned as well as existing buildings and vegetation already at the detailed development plan stage (Alenius et al., 2019). Obstruction angle is defined as the angle between the horizontal plane and a line connecting, e.g., the center point of a window and the highest point of an obstruction (e.g., neighbouring building) (Fig. 21).

Fig. 21

Obstruction angle

Obstruction angles determine the potential of direct access to skylight (Fig. 22). As seen in Fig. 22, apartments in lower floors have less access to skylight that has not been reflected on urban objects.

Fig. 22

Obstructing angle between neighbouring buildings in an urban setting. The highlighted sections correspond to apartment areas that do not have direct access to skylight

1.2 Latitudinal daylight metrics

1.2.1 Sunlight exposure (SS EN 17037)

This is the Swedish realisation of the European standard recommendation in EN 17037:2018 stating that at least one room in an apartment should receive at minimum 1.5 h of direct sunlight on any date between February 1^st and March 21^st (CEN, 2018).

Appendix 2: Legal requirements & recommendations for daylight

In Sweden, daylight is regulated in section 6:322 of the building regulations (BBR) by the National Board of Housing, Building and Planning (Swe: Boverket) (Boverket, 2011). The regulations are relevant for rooms or parts of rooms that are visited more than frequently and concern good access to direct sunlight or daylight. It is recommended that there should be at least one window in every room occupied more than occasionally, from which it is possible to follow the course of the day as well as seasonal changes. The requirement is specified by the following metrics (cf. Section 3.1, Appendix 1 for more details):

The daylight metrics used for regulating daylight access are:

• Point Daylight Factor (DF_p) >  = 1%

• Simplified window-to-floor-area method (AF-method) – equivalent to DF_p >  = 1%:

  • ◦ Obstruction angle within the range of 0–30°

  • ◦ Window-to-floor area ratio as:

$${A}_{window}\ge {f}_{(x)} \times {A}_{floor}$$

where:

• A_window = total window area of a room

• A_floor = total floor area of a room

• ${f}_{(x)}=\left\{\begin{array}{c}0.075 , x \in [0, 10]\\ 0.0025 \times x+0.05 , x \in [10, 30]\end{array}\right.$

  where x = window obstruction angle

The European Standard (EN 17037:2018 for Daylight in buildings) includes recommendations to achieve comfortable levels of indoor light by utilizing natural light (CEN, 2018). These are:

• Minimum requirement of 1.5-h direct sunlight in a frequently visited room during a day between February 1^st and March 21^st.

• Half of the surface in rooms occupied more than occasionally, should reach an internal illuminance of 300 lx during half of the time when daylight is present within a year. Otherwise, an illuminance level of approximately 100 lx should be reached for over 95% of the space for at least half of the daylight hours of a year.