a) Stockholm

b) Minneberg

c) HammarbySjöstad

**Figure 8.** The two urban areas in Stockholm (**a**) considered in this study—(**b**) Minneberg and (**c**) Hammarby Sjöstad. The buildings analysed are marked in red.

### **4. Results**

In this section, the results of parametric simulations are presented to analyse the impact of the level of detail (LoD) (Section 4.1, the thermal zoning impact (Section 4.2), and the shadowing impact of the surrounding environment (Section 4.3). Two different districts in Stockholm municipality (Section 3) were used for the purpose of illustration. For all

simulations, the climate of Stockholm Arlanda airport was used from the IWEC typical year database from ASHRAE [39].

#### *4.1. The Impact of Level of Detail (LoD)*

Minneberg district was used to investigate the impact of the level of detail (LoD) of the building geometry. Two levels of detail were analysed: LoD 1.2 and LoD 1.3. For this analysis, 23 out of 33 buildings were used as other buildings were not available in the LoD 1.2 format. The application of the two LoDs resulted in two different geometries generated for each building, as depicted for the two sample buildings in Figure 9.

**Figure 9.** Two sample buildings represented with (**a**) LoD 1.2 and (**b**) LoD 1.3 models in comparison to (**c**) a satellite view.

The calculated thermal energy demand was normalised using the heated area to compensate for the difference in total heated area. However, the changes in the surface of the external envelope and the solar gains emerging from the choice of LoD still led to different heat gains and losses. Figure 10 presents the deviation between the thermal energy demand intensity (TEDI) for LoD 1.3 and the reference of LoD 1.2 versus the change in shape factor (the ratio between the envelope surface area facing outwards and its volume) induced by upgrading from LoD 1.2 to LoD 1.3. The results show that even though most discrepancies remained below 4%, some buildings demonstrated more than 10% greater heat needs for LoD 1.3 than LoD 1.2. The two largest changes were observed in the case of buildings 9 and 10 where the shape factor increase was nearly 20%. Thus, at the UBEM scale, keeping LoD 1.2 could lead to a 10% extra discrepancy of TEDI for some buildings. However, for the overall considered district (23 buildings), the difference remained below 1% (ΔTEDI = 0.76%). Hence, using a higher level of detail might be irrelevant for some larger scale UBEMs targeted at lower spatial resolution. At the same time, making the extra effort by using LoD 1.3 can be worth it in the case of building calibration or analysing the impact of ECMs, as in this case, the identified 10% extra TEDI would result in a skewed definition of calibrated building parameters or wrongly estimated energy savings.

ƄShapeFactor, %

**Figure 10.** Relative change of shape factor and thermal energy demand intensity (TEDI) from the upgrade of the level of detail (LoD) for buildings in the Minneberg district, from *LoD 1.2* to *LoD 1.3* (LoD 1.2 serves as a reference).

#### *4.2. Impact of Thermal Zoning*

This subsection presents, for the two districts described above, the impact of different *thermal zoning* resolutions. Figure 11 presents the different options available in the UBEM for a simple building: (**a**) single zone for heated and non-heated volumes, (**b**) single zone per floor, (**c**) core-perimeter zones for heated and non-heated volumes, and (**d**) core-perimeter zones per floor.

**Figure 11.** Four thermal zoning approaches, applied to a sample building consisting of three regular floors and two basement floors: (**a**) single zone for heated and non-heated volumes, (**b**) single zone per floor, (**c**) core-perimeter zones for heated and non-heated volumes, and (**d**) core-perimeter zones per floor.

The paradigm of floor multiplier was applied for options (**a**) and (**c**). The core and perimeter (**c**, **d**) zone definition followed the algorithm presented above (Section 2.3.3). All elements other than the thermal zones remained the same within the different simulation setups presented below. The impact of thermal zoning is characterised by the change in TEDI. Figure 12 presents the distributions of absolute *(left)* and relative *(right)* discrepancies along the four zoning options, with (**b**) *(single zone per floor)* as the reference. The same trends were observed for the three geometry cases (one district with LoD 1.2 and two with LoD 1.3). The configuration with *single zone* (**a**) remained close, with a minor underestimation of TEDI, to the configuration (**b**) with *single zone per floor*. The *core and perimeter zone* approach (**c**, **d**) increases TEDI by a small amount, keeping the same difference between configuration (**c**) and (**d**) as between (**a**) and (**b**), with a very minor underestimation of TEDI when aggregating the different floors into one volume (**c**). These results match the findings of

similar studies reported earlier [26]. The highest *relative* difference applies for the buildings with the lowest consumption, while the highest *absolute* difference was observed for the buildings with the highest consumption.

**Figure 12.** Absolute (**I**) and relative (**II**) change in thermal energy demand intensity (TEDI) for buildings in Minneberg (LoD 1.2 and LoD 1.3) and Hammarby Sjöstad (LoD 1.3) districts with single zone (a), single core−perimeter (c), and core-perimeter per floor (d) zoning applied. Single zone per floor (b) was used as a reference.

Table 1 provides the calculated changes of TEDI across different LoDs and thermal zoning approaches at a district scale. These values sugges<sup>t</sup> that, similarly to the varying LoDs, different thermal zoning approaches might lead to the same results and are not worthy of interest for analysis made on a district scale. As a grea<sup>t</sup> deal of extra time is required when using a *core and perimeter zone on each floor* (**d**), *one zone per floor* (**b**) zoning can be suggested as the default choice for UBEM studies.

**Table 1.** Change in the total thermal energy demand intensity (TEDI) due to different zoning approaches (Figure 11) at the district scale.


*4.3. Impact of Surrounding Shadowing Environment*

This subsection explores the impact of the threshold distance, beyond which the shadowing effect of surrounding buildings is not considered. The distance was defined as presented earlier in the model workflow (Section 2.3.4). Parametric simulations were conducted for all buildings in the two case districts with a fixed LoD (1.3) and thermal zoning *(b, one zone per floor)* configuration.

The buildings' performances, estimated as TEDI, were obtained for each building and aggregated at the district scale. The TEDI factor represents the ratio of TEDI for each shadowing distance over the maximum TEDI computed for all shadowing distances. As expected, there was an evident dependency of the shadowing effect from the surrounding environment. Figure 13 shows that on a *building level* (**i**), greater shadowing areas resulted in higher TEDIs. While this held true for both districts, the aggregated results at the district level were quite different. Only 5% TEDI difference was observed for Minneberg at the district scale, while 12% TEDI difference was computed for Hammarby Sjöstad.

**Figure 13.** Impact of shadowing environment limited by a distance threshold on thermal energy demand intensity (TEDI) for each building (top) and the entire district (bottom) in (**a**) Minneberg and (**b**) Hammarby Sjöstad districts. Maximum TEDI was used as the reference.

These results allow us to characterize the magnitude of the effect of certain thresholds for shadowing environments on TEDI. At a district scale, differences below 2% could be achieved by including all shadowing surfaces within 50 m from the building's centroid. Furthermore, surfaces farther than 150 m did not seem to have any effect at the district level. At the same time, these results sugges<sup>t</sup> that a threshold of 200 m should be kept for analysis at the building level.

#### **5. Conclusions and Discussion**

This paper has presented MUBES—a new simulation tool for urban building energy modelling (UBEM). This tool can be used for a number of applications including: (a) analysis of the current energy performance of an existing building stock at a district or city scale; (b) mapping the system effects from the large scale roll-out of retrofitting actions; (c) generation of a calibrated sample of simulations that can further be used to compensate for missing data; and (d) analysis of various operation strategies for the building stock on a district scale that could improve the overall performance of the urban energy system (including power distribution grid or district heating network).

MUBES UBEM follows a physics-based paradigm using a Python-based framework as an environment for the generation and managemen<sup>t</sup> of simulations and EnergyPlus as a core thermal engine. To enable analysis of the impact of the level of detail (LoD), the geometry definition with photogrammetric point cloud method was conducted at the data integration stage. The developed UBEM workflow generates models for building energy performance simulations building-by-building and runs simulations at the district scale in a fully automated way. Input data are provided through a GeoJSON file containing both geometric (polygons for all building's external surfaces) and non-geometric properties for each building integrated from several data sources. At its core, the workflow follows a

*shoebox* paradigm with ideal HVAC system, and in this way provides additional robustness for the further expansions required for intaking input data in other formats.

The developed simulation tool was used to investigate the impact of three aspects that can affect the performance of UBEMs on a district/urban scale: (1) the level of detail (LoD) for input building geometries; (2) thermal zoning approach; and (3) the shadowing effect of the surrounding environment. Following the analysis of these phenomena for the two case districts in Stockholm, the subsequent conclusions can be drawn:

Level of detail (LoD). A change in the LoD from 1.2 to 1.3 resulted into quite distinctive shape factors (0–20%) for some buildings, leading to a noticeable (0–13%) impact on the thermal energy demand intensity (TEDI) for space heating at the building scale. At the same time, for a district scale analysis, given a certain level of homogeneity of the analysed district, a more detailed LoD 1.3 might not be required. For instance, in the case of the studied district of Minneberg, the overall TEDI difference ( ΔTEDI) at the district scale remained below 1%, despite a change of over 10% for some buildings. Hence, as use of LoD 1.3 may require extra effort in data collection, LoD 1.2 could be seen as sufficient for district scale analysis. On the other hand, bottom-up physical models are required to accurately compute the impact of energy conservation measures that are to be estimated. Thus, as these impacts might be less accurately estimated with LoD 1.2, it would still be recommended to use LoD 1.3 if available.

Thermal zoning. The analysis of various thermal zoning approaches has mostly confirmed earlier studies. Particularly, the overall ΔTEDI at the district scale has remained below 5%, despite a more pronounced effect for some buildings. The analysis showed that a single zone option for heated and non-heated volumes should be avoided, which is in line with recommendations from existing standards. At the same time, a compromise of having one zone per floor was still found to be acceptable. For higher buildings, the merging of middle floor zones while keeping bottom and top floor zones separate could be worthy of further investigation.

Surrounding shadowing environment. Up to 12% of ΔTEDI could be attributed to the change in the shadowing environment in the case of two districts with quite different types of building geometries, with a monotone increase in TEDI along with the increase in the shadowing distance threshold. At the district scale, limited effects (below 2%) were observed for the nearest shadowing environment (up to 50 m). Furthermore, surfaces farther than 100 m did not have any profound effect at the district scale for both studied areas. At the building scale, the limited effects' threshold rose to 150 m. However, as extra computing time is negligible, the authors would advise keeping 200 m for all simulations.

We conclude that the analysed modeller assumptions embedded in UBEMs have a distinct impact on the UBEMs' outcome and sugges<sup>t</sup> promoting more explicit documentation of these choices in upcoming UBEM studies.

**Author Contributions:** Conceptualisation, X.F., T.J. and O.P.; Methodology, X.F., T.J. and O.P.; Software, X.F. and T.J.; Validation, X.F.; Formal analysis, X.F., T.J. and O.P.; Investigation, X.F., T.J. and O.P.; Resources, X.F., T.J. and O.P.; Data curation, X.F. and T.J.; Writing—original draft preparation, X.F.; Writing—review and editing, X.F., T.J. and O.P.; Visualisation, X.F. and O.P.; Supervision, X.F.; Funding acquisition, O.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Swedish Energy Agency (Energimyndigheten) via the E2B2 research programme, project nos. 40846-2 and 46896-1.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The presented UBEM platform and sample input dataset for the district of Minneberg are provided open source under MIT license at https://github.com/KTH-UrbanT/ mubes-ubem (accessed on 19 January 2022). The raw data utilised in the study were obtained from Swedish public bodies (Boverket and Lantmäteriet) and are limited to use within particular research projects.

**Acknowledgments:** We express gratitude to Boverket and Lantmäteriet for providing the data extracts from the EPC database 'Gripen', building and property cadastre data and photogrammetric building data, respectively. The computations were tested using resources provided by the Swedish National Infrastructure for Computing (SNIC) at SNIC Science Cloud (SSC) partially funded by the Swedish Research Council through gran<sup>t</sup> agreemen<sup>t</sup> no. 2018-05973.

**Conflicts of Interest:** The authors declare no conflict of interest.
