A GIS-Based Approach for Urban Building Energy Modeling under Climate Change with High Spatial and Temporal Resolution

Abstract

The energy demand and associated greenhouse gas (GHG) emissions of buildings are significantly affected by the characteristics of the building and local climate conditions. While energy use datasets with high spatial and temporal resolution are highly needed in the context of climate change, energy use monitoring data are not available for most cities. This study introduces an approach combining building energy simulation, climate change modeling, and GIS spatial analysis techniques to develop an energy demand data inventory enabling assessment of the impacts of climate change on building energy consumption in Shanghai, China. Our results suggest that all types of buildings exhibit a net increase in their annual energy demand under the projected future (2050) climate conditions, with the highest increase in energy demand attributed to Heating, Ventilation, and Cooling (HVAC) systems. Variations in building energy demand are found across building types. Due to the large number of residential buildings, they are the main contributor to the increases in energy demand and associated CO₂ emissions. The hourly residential building energy demand on a typical hot summer day (29 July) under the 2050 climate condition at 1 p.m. is found to increase by more than 40%, indicating a risk of energy supply shortage if no actions are taken. The spatial pattern of total annual building energy demand at the individual building level exhibited high spatial heterogeneity with some hotspots. This study provides an alternative method to develop a building energy demand inventory with high temporal resolution at the individual building scale for cities lacking energy use monitoring data, supporting the assessment of building energy and GHG emissions under both current and future climate scenarios at minimal cost.

1. Introduction

Over the past 20 years, global energy consumption and greenhouse gas (GHG) emissions have increased by 49% and 30%, respectively [1]. China has been the world’s largest annual GHG emitter since 2008 [2]. To mitigate this problem, China has pledged to achieve the goals of carbon peaking by 2030 and carbo–n neutrality by 2060 [3]. About 70% of GHG emissions in China come from urban areas [4] and, according to the China Association of Building Energy Efficiency [5], GHG emissions related to buildings accounted for a significant proportion (38.2%) in 2021. GHG emissions during the building operation stage reached 2.3 billion tons in 2021, which is more than half (56.6%) of the total building-related GHG emissions in China, and has increased each year since 2014 [5]. Compared to the construction, rehabilitation, and demolition stages, GHG emissions during the building operation stage are more difficult to manage, as they are determined by many factors such as climate conditions [6,7,8], building envelope characteristics [9,10], and occupancy behaviors [11,12]. Moreover, climate change can affect the energy demand of existing buildings during their operation stage through changes in the amount of required cooling and heating energy [6,13,14], which brings many uncertainties and challenges to achieving the “carbon peaking and carbon neutrality” goals. This is because electricity—the primary energy used during the building operation stage—is still mainly generated from traditional fossil fuels [6].

As the speed of urbanization in Chinese cities is declining, policymakers have shifted their focus on urban development from constructing new buildings to rehabilitating [15] and renovating [16] existing building stocks, with an emphasis on enhancing their energy efficiency during the operational phase [17,18], thereby promoting urban sustainability [19,20,21,22]. This requires knowledge regarding the influences of different factors on buildings with different characteristics, such as building envelopes, functions, and locations [23,24]. Researchers have proposed various approaches to study the energy use of urban buildings, which can be divided into two categories: top–down statistics-based and bottom–up simulation-based approaches [25]. Top–down statistics-based approaches estimate the energy demand of buildings based on statistical data and numerical models [26,27], which combine various driving factors such as per capita Gross Domestic Product (GDP) [28], industrial structure [29], population density [30], and the numbers of heating and cooling degree days [31]. For example, Ürge-Vorsatz et al. [32] decomposed the cooling and heating energy demand of buildings into critical drivers, including household numbers, population size, floor space per capita, and GDP. They predicted the future trends of building thermal energy use based on the historical data of these factors at the global and regional (11 world regions) scale. Afaifia et al. [33] constructed an energy consumption model for residential buildings in 48 provinces in Algeria based on regression analysis and hierarchical clustering. Their results suggested an increase in residential building energy consumption related to the size of the residential building stock and the connection rates of electricity and natural gas. However, top–down statistical approaches rely on the availability of statistical data on socio-economic factors. Therefore, they only focus on building energy demand estimation at broader spatial (global, national, and province/state) and coarser temporal (annual) scales.

Bottom–up simulation-based approaches were developed to capture the characteristics of building energy consumption at finer scales. Such an approach adopts software (e.g., EnergyPlus 24.1.0, DOE-2.3, and eQuest 3.65) including different models to simulate the energy use of building prototypes using envelope attributes and thermal properties combined with meteorological data (e.g., temperature, relative humidity, and solar irradiance). As hourly meteorological data are available for many cities, a bottom–up simulation-based approach can be used to simulate building energy consumption at the individual building level at an hourly temporal scale. These approaches have been widely used to study the impacts of climate change on building energy demand in cities for many existing buildings [34,35,36,37], using the future climate projections obtained through certain models. The effects of climate change can then be assessed by comparing the simulated future building energy demands with the current building energy demands. Huang and Gurney [34] used a similar method to simulate the energy demand of 18 prototype buildings in different climate zones in the United States, which were developed by the U.S. Department of Energy (DOE) to assess the impacts of climate change. Their results indicated that the variation in building energy use within the same climate zone caused by different building types can be more significant than the variation between different climate zones. Zou et al. [35] provided a comprehensive evaluation of the future energy resilience of the urban residential sector in hot–humid regions of China in the 21st century (from 2025 to 2099) under different climate change scenarios. Their results suggested that, under the most severe scenario, the average annual energy demand of the residential sector may increase to more than twice the energy demand under the current typical climate. Besides the studies that have focused on discussing building energy characteristics in different climate zones using representative building prototypes, many existing studies have combined Geographic Information Systems (GIS) with building energy simulation to visualize the building energy distribution pattern at the urban scale. The GIS modeling technique allows for the integration, management, and analysis of geographic data at different spatial resolutions, which is helpful in urban resources management [38,39] and hotspot identification [40,41]. Yu et al. [42] combined the GIS technique with building energy modeling to investigate the effects of eight urban planning factors on building energy use. Song et al. [43] utilized multi-source GIS data to develop different building prototypes that can readily be used in urban building energy simulation. Camporeale et al. [44] presented a GIS-based approach to building clusters at the district scale. They further evaluated the potential of energy demand reduction through PV production using the spatial analysis method in GIS. García-Pérez et al. [45] combined life cycle assessment (LCA) and GIS techniques to examine the environmental implications of building retrofits at different scales (building and urban scale). Their results indicated that, even though construction systems and thermal insulation could lead to energy savings for buildings, they are still affected by urban morphologies.

In recent years, the concept of “digital twin” has become increasingly popular in urban studies, providing virtual maps of actual cities in the digital world. In urban building studies, digital twin approaches integrating multiple techniques, such as GIS and Building Information Models (BIMs), have been used to simulate energy use on an urban scale from different sectors based on various data sources [46]. As such an approach can comprehensively describe the current energy use of buildings and predict future energy use within a city, it has excellent prospects in urban planning and management [47,48]. Fonseca et al. [49] provided a 3D visualization of the spatial and temporal energy consumption patterns in communities and urban areas using a GIS framework. Rossknecht et al. [50] predicted heating demand using a 3D city model of Helsinki, Finland based on the CityGML Energy Application Domain Extension (Energy ADE) and different scenarios. They predicted a 4% reduction in heating demand per decade under climate change and an 82% reduction in heating CO₂ emissions by 2035 compared to 1990. Mylonas et al. [51] developed digital twin scenarios based on machine learning and data-driven methods including real-time energy data monitoring.

As top–down statistical approaches only focus on broader spatial (global, national, and province/state) and temporal (annual) scales, they cannot provide useful guidance for city governments, such as identifying regions that are vulnerable to energy shortages within a city during extremely hot or cold days under future climate scenarios. Although the concept of the digital twin is emerging in building energy studies at the urban scale, it is still in its infancy in terms of predicting future building energy demand at the urban scale. Furthermore, prediction of the future energy consumption of buildings is typically based on a large amount of historical building energy use data captured by sensors in individual buildings, which are not readily available to the public in most cities worldwide. Due to the limitations mentioned above, studies assessing the impacts of climate change on building energy demand at the sub-city scale with high temporal (monthly and hourly) scales are lacking. To fulfill the needs of city governments, in terms of assessing potential power outage hazards and managing GHG emissions from the building sector to adapt to climate change, there is a critical need for an alternative method for the assessment of building energy demands under different climate scenarios. In summary, there are two key research gaps in the relevant existing literature: (1) a very limited number of studies focused on comparing building energy demand under current and future climate scenarios at the individual building scale within urban areas and (2) a lack of methods to identify the potentially vulnerable regions that might suffer energy supply shortages under the impacts of climate change within the city. To fill these gaps, this study introduces a novel approach combining a bottom–up building energy demand simulation method, climate change modeling, and a GIS spatial analysis technique in order to assess the impact of climate change on the energy demand for buildings in Shanghai, China. The specific research objectives are (1) to develop a building energy demand dataset with high spatial and temporal resolution that can be incorporated into urban GHG emissions inventories and (2) to provide an approach enabling identification of the areas most vulnerable to climate change, regarding energy supply shortage, within the city. Therefore, two hypotheses are put forth: (1) the majority of building types show an obvious annual increase in energy demand under future climate conditions, and (2) the variation in energy increases across different building types will be larger at finer temporal scales. The results of this study are expected to guide city governments in developing the most effective mitigation strategies for different administrative districts and townships, in order to reduce GHG emissions and adapt to climate change.

2. Methods

Figure 1 presents the workflow of the methodology. First, an Extreme Gradient Boosting (XGBoost) version 2.0 model was used to estimate the types of existing buildings in the study area. Second, building prototypes were used to simulate the energy demand of different types of buildings and the energy use intensity (EUI), representing the energy use per square meter, was determined for six building prototypes under 2020 and 2050 climate conditions. Furthermore, a method was developed to calibrate the simulation results. Finally, the calibrated building EUI values were combined with building footprints, along with information on building types, areas, and floor numbers, in order to calculate the energy demand of individual buildings at different temporal (annual, monthly, and hourly) scales. A GIS spatial analysis technique was used to identify the spatial patterns of building energy consumption at the sub-city scale. Each step is presented in more detail in the following sections.

2.1. Study Area and Data

This study selected two administrative districts adjacent to each other—Xuhui District and Minhang District—in the City of Shanghai, China as the study area (Figure 2). Shanghai is located on the eastern coast of China and falls under the “hot summer and cold winter” (HSCW) climate category [52], with an annual average temperature of 17.9 °C. The percentage of building GHG emissions from the HSCW zone has increased yearly, from 26% in 2015 to 29% in 2021 [5]. The two districts have a total area of approximately 426.61 km², with Xuhui District covering 54.93 km² and Minhang District covering 371.68 km² (Figure 2). At the end of 2022, the gross domestic product (GDP) of Xuhui District reached 255.79 billion Chinese Yuan, similar to that of Minhang District (288.01 billion Chinese Yuan). Xuhui District has a population of 1.09 million, while Minhang District has a population of 2.69 million during the same year [53,54]. Despite being spatially adjacent, these two districts differ significantly in their population, economic, and industrial structures, which might lead to different building energy consumption patterns between them.

Table 1. Sources of primary datasets used in this study.

Data	Source	Year
Building footprint	Gaode Map	2020
Point of Interest (POI)	Gaode Map	2022
Land use Data	Essential Urban Land Use Categories (EULUC) China Data	2018
Building prototypes	Tsinghua University	N/A
Chinese Standard Weather Data	National Solar Radiation Data Base (NSRDB)	2020
Landsat 8 satellite imagery	https://www.usgs.gov/, accessed on 16 April 2024	2020

lists the primary datasets used in this study. The 2020 building footprint data for the two districts and the 2022 Points of Interest (POIs) datasets were obtained from the Gaode Map. The building footprint data provided information on the area, the number of floors, and the height of each existing building in the study area. Information on building types and functions were determined according to the Essential Urban Land Use Categories (EULUC) China data (http://data.ess.tsinghua.edu.cn, accessed on 20 January 2024), high-resolution orthophoto, and the POI data. Building prototypes were collected from Tsinghua University, China (https://cal.dest.net.cn/building_model/building/download, accessed on 24 November 2023), representing various types of existing buildings in different climate zones in China. The Chinese Standard Weather Data (CSWD) of Shanghai, which was developed by the National Solar Radiation Data Base (NSRDB), was also collected (https://energyplus.net/weather-location/asia_wmo_region_2/CHN/CHN_Shanghai.Shanghai.583620_CSWD, accessed on 27 November 2023), which compiles weather data for Shanghai from multiple years at the hourly scale.

2.2. Building Prototypes

The building prototypes dataset developed by Tsinghua University [55] was used for energy demand simulation. These prototypes were built according to the properties of existing buildings in different climatic zones in China, and they include the characteristics of the building envelope properties, occupancy behaviors, and heating, ventilation, and air-conditioning (HVAC) systems based on the Chinese design standard of the public buildings (GB50189-2015) [56] and residential buildings (JGJ134-2010) [57]. Prototypes for the HSCW climate zone—where Shanghai is located—were selected. These included five types of public buildings (commercial, school, hospital, office, and hotel), as well as apartment buildings. These five types of public buildings and apartments account for 92.3% and 93.6% of public and residential buildings in Shanghai, according to the Shanghai Statistical Yearbook 2023 [58]. Therefore, they were considered to be representative of existing buildings in the study area. Fundamental thermal properties of residential and public buildings were set according to An et al. [55]. For residential buildings, the shape coefficient, roof U factor, and exterior wall U factor were set to 0.31, 0.8 W/(m²·K), and 1 W/(m²·K), respectively [55,57]. For public buildings, the roof U factor was set to 0.6 W/(m²·K), the exterior wall U factor was set to 0.7 W/(m²·K), and the shape coefficient was varied from 0.13 to 0.36 by type [56].

2.3. Determination of Building Types

The approach provided by Deng et al. [59] was adopted to determine the types of each building footprint. POI datasets are digital representations of distinct locations within urban geospatial contexts, which are commonly featured in contemporary online maps. Due to their high accuracy, precision, and frequent updates, POIs are widely used in research on topics such as urban spatial configurations and land-use changes. In this study, 116,923 records of POIs, which were updated in December 2020, were collected from Gaode Map (https://lbs.amap.com/, accessed on 18 March 2024). The collected POIs included 14 building function categories combined into 7 significant functions: office, school, commercial (shopping mall), hotel, hospital, residential apartment, and industrial. A connection was built between POIs and building footprints by examining their spatial relationships. Buildings that contained only one type of POI were assigned the corresponding type. If residential and non-residential POIs were found simultaneously within a building, this building was designated as residential. Many buildings in Chinese cities have mixed functions, with the first floor used as shops and the rest of the floors used for residential purposes. POIs that were not located within any building footprint were assigned to the nearest buildings. After the initial POIs examination, it was found that 86,958 out of 102,250 (85%) buildings did not contain any residential POIs, which does not match the actual situation in cities [59]. Therefore, an XGBoost model was used to estimate the types of these 86,958 buildings. The XGBoost model is a machine learning classifier that has been widely used for hazard assessment [60,61,62], image classification [63,64,65], and urban building function identification [66]. Based on the combination of remote sensing imagery, high-resolution orthophotos, Baidu Street View, and field survey, 4400 samples were selected, including 2200 residential and 2200 non-residential buildings (Figure 3). About 70% were used as training samples, while the other 30% were used as reference samples for accuracy assessment of building type classification. In the next step, the XGBoost model was used to classify the 86,958 buildings into residential and non-residential buildings based on morphological parameters, road distance, and number of POIs. The morphological parameters, which include the area, the perimeter, the ratio between area and perimeter, the ratio between area and floor number, and the ratio between perimeter and floor number, were calculated for each building according to its attributes in the building footprint dataset. After the classification, 52,654 of 86,958 buildings were classified as residential buildings with 83% classification accuracy. There were 11,818 buildings that remained unclassified as they did not contain any POIs. Therefore, another XGBoost model was applied to classify them based on remote sensing imagery and land use data into five types of public buildings (i.e., hospitals, hotels, shopping malls, schools, and offices).

2.4. Building Energy Simulation and Calibration

This study used the Designer’s Simulation Toolkit (DesT) 2.0 software to simulate the energy consumption of five public and one residential building prototypes under 2020 and 2050 climate conditions. The DesT software is a famous building energy simulation software developed by Tsinghua University, China, which takes building prototypes and local climate data as inputs to simulate the hourly building energy consumption through jointly considering factors such as building envelopes, HVAC systems, and the energy-related behaviors of occupants. As all the building prototypes were designed based on Chinese building standards, they have been widely used in building energy dynamic simulation [67,68], energy savings analysis [69,70], and the design of building energy-related retrofits [71,72] in Chinese cities.

The local climate data used in DesT were retrieved from the CSDW dataset, which provides hourly weather data for Shanghai and was utilized in DesT for building energy simulations over an entire year at the hourly temporal scale. As the CSDW dataset was constructed based on meteorological observations from at least ten years prior, the hourly weather data files for the years 2020 and 2050 were built by employing a widely used algorithm in the existing literature [6,73,74,75], which applied HadCM3 (Hadley Centre Coupled Model, version 3) and the CCWorldWeatherGen tool to the original hourly weather data for Shanghai from the CSDW dataset. The CCWorldWeatherGen tool can apply a morphing algorithm to down-scale the monthly mean weather variations in 2020 and 2050 generated by HadCM3 to hourly variations. Figure 4 presents the monthly average air temperature under 2020 and 2050 climate conditions modeled by HadCM3 in Shanghai, China. In particular, the annual average temperature was expected to increase by 1.26 °C in 2050 and the values of monthly average temperature increases range from 0.94 to 1.74 °C, with the largest increase in August, followed by March and November.

Figure 5 presents the building energy simulation modules incorporated into DeST, including building envelopes, energy systems, and occupant behavior. The modules presented were developed based on comprehensive thermodynamic analysis, numerical calculations, and algorithms for the solution of the heat and mass transfer processes of the respective elements. The accuracy of each module has been verified through comparisons between simulation results and experimental data [76]. The parameters of building envelopes, energy systems, and occupancy were set according to the Chinese design standards for residential (JGJ134-2010) [56] and public buildings (GB50189-2015) [53], as mentioned in Section 2.2. The occupant action modules include behavior modules of heating and cooling control, lighting and equipment use, window opening, and so on (Figure 5). These stochastic modules were developed based on actual observed occupant behavior data and their reliability was verified in practical applications [76].

After the simulation was completed in DesT, the energy use intensity (EUI), representing the energy use per square meter, was determined for the six building prototypes under 2020 and 2050 climate conditions. To calculate the hourly energy use for individual buildings over a year in the study area, a building energy modeling method adopted by Zheng and Weng [24] was used. The energy use for an individual building (EU) i during hour j can be calculated as follows:

$${EU}_{hour\left(i,j\right)}={EUI}_{i,j}\times A_{building\left(i\right)}\times {FN}_{building\left(i\right)},$$

(1)

where ${EUI}_{i,j}$ is the energy use of building i during hour j; $A_{building\left(i\right)}$ represents the area of building i, which can be calculated from the building footprint dataset; and ${FN}_{building\left(i\right)}$ is the floor number of building i. The annual energy consumption of an individual building i, which covers 8760 h, was calculated as follows:

$${EU}_{annual\left(i\right)}=\sum _{j=1}^{8760}{EU}_{hour\left(i,j\right)}.$$

(2)

Annual building energy use at the district spatial scale can then be calculated by aggregating the energy use from all individual buildings within each district, as follows:

$${DEU}_{annual\left(k\right)}=\sum _{t=1}^n{EU}_{annual\left(i\right)},$$

(3)

where ${DEU}_{annual\left(k\right)}$ is the simulated annual energy use of all buildings in district k; t is the type of building i; and n is the number of buildings within the district.

However, there were discrepancies between the simulation results and the actual building energy use, caused by uncertainties such as those associated with occupancy behaviors. This study used different methods to calibrate the simulated energy use in residential and public buildings. First, the monitored annual energy use data of five types of public buildings from the 2021 Shanghai State Office Buildings and Large-scale Public Buildings Energy Consumption Monitoring and Analysis Report [77,78,79,80,81], provided by the Shanghai Municipal Commission of Housing and Urban-Rural Development and Management, were used to calibrate the corresponding simulation results for public buildings:

$${EU}_{pub,cali,hour(t,j)}={\displaystyle \frac{{EU}_{pub,act,annual\left(t\right)}}{{EU}_{pub,sim,annual\left(t\right)}}}\ast {EU}_{pub,sim,hour\left(t,j\right),}$$

where ${EU}_{pub,cali,hour(t,j)}$ refers to the calibrated energy use for public buildings of type t in hour j; ${EU}_{pub,sim,hour(t,j)}$ refers to the simulated energy use for public buildings of type t in hour j; ${EU}_{pub,act,annual\left(t\right)}$ is the monitored annual energy use for public buildings of type t; and ${EU}_{pub,sim,annual\left(t\right)}$ represents the simulated annual energy use for public buildings of type t. However, monitored energy use data for residential buildings were not available. Therefore, this study used the annual energy consumption data of residential buildings, available in the Statistical Yearbook of Xuhui District and Minhang District [53,54], to calibrate the corresponding simulation results and eliminate the discrepancies:

$${EU}_{res,cali\left(k\right)}={\displaystyle \frac{{DEU}_{act,annual\left(k\right)}}{{DEU}_{sim,annual\left(k\right)}}}\ast {EU}_{res,sim\left(k\right)},$$

(4)

where ${EU}_{res,cali\left(k\right)}$ refers to the calibrated energy use for an individual residential building over a certain period (annually, monthly, hourly) in district k; ${EU}_{res,sim}$ refers to the simulated energy use for the individual residential building over the same period in district k; ${DEU}_{act,annual\left(k\right)}$ is the actual residential building energy use in district k; and ${DEU}_{sim,annual\left(k\right)}$ is the simulated residential building energy use in district k.

For all residential and public building types, the ratio between simulated energy use in 2050 and 2020 at the individual building level was used to calibrate the building energy use in 2050, using the calibrated building energy use in 2020 as input:

$${EU}_{cali 2050\left(t\right)}={\displaystyle \frac{{EU}_{sim 2050\left(t\right)}}{{EU}_{sim 2020\left(t\right)}}}\ast {EU}_{cali 2020\left(t\right),}$$

(5)

where ${EU}_{cali 2050\left(t\right)}$ and ${EU}_{cali 2020\left(t\right)}$ refer to the calibrated energy use for building type t in 2050 and 2020, respectively; ${EU}_{sim 2050\left(t\right)}$ is the simulated energy use for building type t in 2050; and ${EU}_{sim 2020\left(t\right)}$ represents the simulated energy use for building type t in 2020. This study used two metrics to evaluate the change in energy demand between identical building types under 2020 and 2050 climate conditions, which were calculated using the following formulas:

$${\mathrm{A}\mathrm{D}}_{\left(t\right)}=\left({EU}_{cali 2050\left(t\right)}-{EU}_{cali 2020\left(t\right)}\right)/\mathrm{F}\mathrm{A},$$

(6)

$${\mathrm{R}\mathrm{C}}_{\left(t\right)}={\displaystyle \frac{\left({EU}_{cali 2050\left(t\right)}-{EU}_{cali 2020\left(t\right)}\right)}{{EU}_{cali 2020\left(t\right)}}}\ast 100\mathrm{\%},$$

(7)

where ${\mathrm{A}\mathrm{D}}_{\left(t\right)}$ and ${\mathrm{R}\mathrm{C}}_{\left(t\right)}$ are the absolute difference (AD) and relative change (RC) of energy demand for building type t under 2020 and 2050 climate conditions, respectively; and FA is the floor area of the corresponding building.

3. Results

3.1. Classification Results of Building Types and Their Spatial Distribution Patterns

The results of the functional identification of the building stock are presented in Figure 3. Significant variations were found in the distribution patterns of different types of existing buildings across different districts within the study area (

Table 2. Statistics of numbers and floor area of buildings in two administrative districts.

	Number			Floor Area (million m2)
Types	Xuhui	Minhang	Total	Xuhui	Minhang	Total
Commercial	489	1071	1560	17	37	54
Educational	1552	4243	5795	30	80	111
Hotel	244	582	826	11	22	33
Industrial	500	7275	7775	10	293	303
Office	1502	3176	4678	88	115	204
Residential	14,703	65,115	79,818	494	1565	2059
Hospital	717	1081	1798	28	41	69
Total	19,707	82,543	102,250	679	2153	2832

). Residential buildings accounted for 78.06% of the total number and 72.68% of the total floor area of buildings in the two districts combined and were the most numerous among all building types (Table 2). This result was in agreement with the building floor area statistics provided in the Shanghai Statistical Yearbook 2023 [57]. Residential buildings were widely distributed throughout the study area, except for the northern and southwestern parts (Figure 6). Office buildings were mainly distributed in the north part of the study area, where the Caohejing Development Zone is located, accounting for 4.58% of the buildings. Industrial buildings were primarily distributed in the southwest of Minhang District, where Xinzhuang Industrial Park is located, accounting for 4.6% of the buildings. As industrial buildings are relatively insensitive to weather variations [25], they were not included in the energy demand analysis in this study. No clustering distribution patterns were found for other types of buildings; most only accounted for small portions of the number and floor area of total existing buildings (Figure 6).

3.2. Analysis of Building Energy Demand at Annual Scale

Figure 7 shows the annual EUI for six types of buildings with the contributions of three energy use categories under 2020 and 2050 climate conditions. The three end-use energy use categories included HVAC, lighting and equipment, and other energy use (e.g., lifts, water supply, drainage). Significant variation was observed regarding annual building EUI and the proportion of energy use categories across different building types (Figure 7), which can be attributed to differences in function and operational hours; for example, hospital and commercial (shopping mall) buildings were found to have much higher annual EUI than others, due to their longer operation hours and higher occupancy density. Although their yearly EUIs were similar, the proportion contributed by different energy use categories was different. HVAC energy use accounted for 53% of total energy use in hospital buildings, as they have strict indoor temperature requirements and ventilation standards, such that air conditioners and fans usually have longer operation hours. On the other hand, lighting and equipment accounted for most of the energy used in commercial buildings.

Climate change mainly affects the energy use of buildings via the HVAC system by affecting changes in the cooling and heating energy demand to maintain the indoor temperature.

Table 3. Relative change (%) and absolute difference (kwh/m²) in annual energy demand for different building types between 2050 and 2020 in Shanghai.

Types	Total		Heating Season		Cooling Season
	AD	RC	AD	RC	AD	RC
Residential	0.81	2.12	−0.49	−2.69	1.27	7.68
Commercial	0.77	0.55	−0.15	−0.31	0.83	1.17
Office	0.66	0.74	−2.35	−7.37	2.59	5.60
Educational	0.56	1.25	−1.27	−6.35	1.56	8.70
Hotel	3.14	2.81	0.04	0.11	2.64	4.21
Hospital	4.01	2.70	−1.44	−2.68	4.99	6.42

presents the RC and AD in energy demand for six different types of buildings between 2020 and 2050 in Shanghai. Due to global warming, all buildings showed increased energy use in these two metrics in 2050 during the cooling season (May to September in Shanghai), as there were higher energy demands for space cooling. On the other hand, during the heating season (November to March in Shanghai), less energy was used for space heating, and most types of buildings showed decreased energy demand. Moreover, as the increased energy demand during the cooling season cannot be offset by the reduced energy demand in the heating season, all types of buildings will have an increased total energy demand in 2050. It was found that hospital and hotel buildings were the most sensitive to climate change, as they have much higher RC and AD in annual EUI than others (Table 3), which was caused by the more prominent ADs in their EUI during the cooling season. The possible reasons are that they both have more extended operation hours for their HVAC systems, which usually work 24 h a day to maintain the comfort of the indoor environment during the cooling season. It is worth noting that even residential buildings have lower annual EUI than all public buildings; they exhibited higher AD and RC of total energy than office, commercial, and educational buildings (Table 3). HVAC energy use accounts for a significant portion of the energy use in residential and educational buildings (Figure 7), causing them to have larger RC in energy demand change during the warmer cooling season in 2050 than others (Table 3).

Table 4. Total energy demand (10⁷ kWh) of different types of existing buildings in two districts under climate conditions of 2020 and 2050.

	2020			2050
	Xuhui	Minhang	Total	Xuhui	Minhang	Total
Com	24.3	52.1	76.5	24.5	52.4	76.9
Edu	13.6	36.2	49.9	13.8	36.7	50.5
Hotel	12.2	24.8	36.9	12.5	25.5	38.0
Off	79.3	103.5	182.8	79.9	104.3	184.2
Res	187.6	594.4	782.0	191.6	607.0	798.6
Hosp	41.8	60.6	102.3	42.9	62.2	105.1
Total	358.9	871.6	1230.4	365.2	888.0	1253.2

presents the total energy demand of existing buildings in Xuhui District and Minhang District under climate conditions of 2020 and 2050. Significant differences were observed between these two districts, caused by the composition and size of buildings. The GIS spatial statistical results suggested that, under the climate condition of 2020, the total energy demands of buildings were 87.2 × 10⁸ kWh and 35.9 × 10⁸ kWh in Minhang District and Xuhui District, respectively. Although residential buildings had relatively low EUI compared to others, they were still the most significant contributor to energy use in both districts as they were dominant in number and floor area. Energy use by residential buildings accounted for 68% of total building energy use in Minhang District. On the other hand, public buildings in Xuhui District contributed more building energy use, with residential buildings accounting for only 52%. Office buildings were the most significant contributor among public buildings, accounting for 22% of energy use. The total building energy demand was predicted to increase from 1.23 × 10¹⁰ kWh in 2020 to 1.25 × 10¹⁰ kWh in 2050, with 1.9% growth.

3.3. Time-Series Analysis of Building Energy Demand at Monthly and Hourly Scales

Figure 8 presents the energy demand for all existing buildings in the two districts combined under 2020 and 2050 climate conditions at a monthly scale. All building types (except for educational buildings) exhibited an apparent seasonality, with higher energy demand in summer and winter than in transition months (spring and fall). The most apparent increase in energy demand occurred in August, as the monthly average temperature also increased the most during this month (Figure 4). Although March and November had relatively larger monthly average temperature increases, there were no obvious building energy demand changes in these two months. This may be because, during the transition season, the increased temperature under future climate conditions might still be within the comfort level and, thus, will not trigger heating or cooling energy use. The Hospital building had the highest EUI in summer, followed by commercial and hotel buildings. In contrast, most of the classrooms in educational buildings are not used during summer and winter vacations. Moreover, all building types showed different degrees of ADs in EUI under the 2050 climate condition, with hospital and hotel buildings having the most significant EUI increases in the cooling season (Figure 8a), which is in agreement with the results presented in Table 4. Figure 8b presents the monthly variation of the energy demand for existing buildings in two districts combined, and residential buildings showed the most significant contrasts between 2020 and 2050 due to their larger number and floor area than other types of existing buildings. During the cooling and heating seasons, the office buildings showed more prominent ADs than other public buildings. A possible reason for this is that they have the largest floor area among public buildings. Moreover, their operation hours are daytime, when the building EUI is usually higher.

Building energy demand on typical cold winter and typical hot summer days in 2020 and 2050 at a diurnal (hourly) scale is presented in Figure 9 and Figure 10, respectively. 1 February was chosen as the typical cold winter day, and 29 July was selected as the typical hot summer day. As educational buildings are typically not in operation on 29 July, due to summer vacation, this study evaluated the diurnal energy demand of educational buildings on 27 June instead. The average daily temperature on 1 February was −2.1 °C, with the lowest temperature (−5.8 °C) occurring at 5 a.m. The average daily temperature on 29 July was 30.5 °C, with the highest temperature (34.1 °C) occurring at 1 p.m. In the projected 2050 hourly weather data, the lowest temperature on 1 February was −4.5 °C and the highest temperature on 29 July reached 35.5 °C simultaneously during the day. The daily peak temperature also occurred at 1 p.m. on 27 June, which was 31.7 °C in 2020 and 33 °C in 2050, respectively. Residential and office buildings were found to have higher energy demand than others, as they have larger floor areas and numbers (Figure 9 and Figure 10). As Figure 9 and Figure 10 suggest, different types of buildings exhibited different diurnal characteristics of energy demand due to the operation time of buildings. Hospitals, hotels, and residential buildings were in operation 24 h throughout the day, while energy use in office, educational, and commercial buildings was limited during non-operation hours. In general, for all types of buildings, there were higher energy demands on the typical hot summer day than on the typical cold winter day, suggesting that the energy supply would face more challenges during the summer. It can be observed that all building types showed larger RCs of energy demand in 2050 at an hourly scale than on an annual and monthly scale on typical hot summer days. For example, on 29 July, the hourly RC for hotel building energy demand increased by 15.4% at 12 p.m., while the hourly RC for residential building energy demand was found to increase by more than 40%, which were both significantly larger than their annual RCs of 2.81% and 2.12%, respectively (Table 4). It is worth noting that there were sharp increases in energy demand in residential buildings on 29 July at 9 a.m., 1 p.m., and 6 p.m., which was due to the higher occupancy rate and cooking behaviors. These factors can contribute to increased internal heat gain, forcing cooling systems to work harder and increasing energy use.

3.4. Spatial Patterns of Building Energy Consumption

3.4.1. Spatial Distribution at Individual Building Level

The GIS spatial analysis technique was used to examine the spatial distribution of building energy demand at the individual building scale. Figure 11 shows the total annual energy demand for existing buildings in Xuhui and Minhang Districts. A high degree of spatial heterogeneity was observed, with the building with the highest annual energy demand reaching 1.3 × 10⁷ kWh in the downtown area of Xuhui District. In comparison, the building with the lowest annual energy demand was only 3.8 × 10³ kWh, located in the suburban area of Minhang District. A significant gradient between urban and suburban areas was also observed in the spatial pattern of energy demand from the HVAC system during the cooling season, falling from 2.1 × 10⁶ kWh in downtown Xuhui District to 0.95 × 10³ kWh in the suburban area of Minhang District (Figure 12B). Energy use due to HVAC systems during the heating season showed a smaller urban–suburban gradient, decreasing from 3.3 × 10⁵ kWh in downtown Xuhui District to 1.1 × 10³ kWh in Minhang District (Figure 12A). Buildings with annual total energy consumption higher than 9.8 × 10⁵ kWh were mostly distributed in the northeast part of the study area, where high-rise office buildings and large commercial complexes are present. On the other hand, buildings with annual total energy consumption lower than 0.7 × 10⁵ kWh were distributed in the southwest and southeast of the study area, where low-rise residential buildings are primarily present.

At the township scale, Caohejing Town in Xuhui District had the highest annual EUI (1334 kWh/m²), followed by Xujiahui Town, which is also in Xuhui District and had EUIs exceeding 1000 kWh/m². On the other hand, Maqiao Town in Minhang District had the lowest annual EUI among all townships (93 kWh/m²).

3.4.2. Spatial Patterns of Energy Consumption in Response to Climate Change

Figure 13 presents the spatial distributions of AD and RC for annual total building energy demand in Xuhui District and Minhang District between 2050 and 2020. Spatial heterogeneity with hotspots was observed for both AD and RC, but their patterns differed. The circles with dashed lines indicate the spatial agglomeration phenomenon (hotspots) in areas with higher AD or RC, which were sensitive to energy demand change under climate change. Areas with higher AD were located in the study area’s northern part, including Xujiahui and Caohejing Town in Xuhui District and Hongqiao Town in Minhang District. The annual building energy demand increase can reach 2.0 × 10⁸ kWh in Xujiahui Town and Caohejing Town, and 1.1 × 10⁸ kWh in Hongqiao Town. These townships all have Central Business District (CBD) zones with many tall public buildings, and it is not surprising that they had higher energy use increases than other areas, due to their larger building floor areas. Areas with higher RC of annual building energy demand were distributed in the Hengshao Road–Fuxing Road historic area (Heng-Fu area) in Xuhui District and several townships in Minhang District, including Huacao Town, Qibao Town, Xinzhuang Town, Meilong Town, Zhuanqiao Town, and Maqiao Town. The RC in annual total building energy demand in these areas reached 2%.

The spatial distributions of the AD and RC for energy demand associated with HVAC systems in heating and cooling seasons between 2020 and 2050 were examined (Figure 14), as climate change affects the energy demand of buildings mainly through changes in energy use from HVAC systems. The results exhibited similar spatial distribution patterns and spatial agglomeration characteristics as the annual total building energy demand (Figure 13 and Figure 14). In the CBD zone within Xujiahui Town, the AD of building energy demand in HVAC systems during the heating season was −2.3 × 10⁷ kWh, while during the cooling season, it was 3.9 × 10⁷ kWh. A high percentage of tall office buildings contributed to the high sensitivity of building energy demand under the context of climate change, as office buildings are more sensitive to warmer climate conditions (Table 4). On the other hand, the RC hotspots of HVAC energy demand were found in townships with a higher percentage of residential buildings, including Hunan Road Town and Tianping Road Town in the northeast part of Xuhui District, and Maqiao Town and Zhuangqiao Town in the south part of Minhang District. Townships with the highest RC of HVAC energy demand exhibited a 6% decrease during the heating season but a 12% increase during the cooling season. It can be found that, no matter whether AD or RC was used as an evaluation metric, the increase in HVAC energy demand during the cooling season cannot be offset by its decrease during the heating season, contributing to a net increase in total annual building energy demand.

4. Discussion

This section presents this study’s key findings, innovation, and contribution. Hypothesis 1, that “the majority of building types show an obvious annual increase in energy demand under future climate conditions”, was supported. The results suggested that all types of buildings presented increased energy consumption in 2050, as the increased energy demand in the cooling season could not be offset by the decreased energy demand in the heating season, in agreement with the results obtained for cities with similar climate conditions presented in previous studies [34,75,82]. Moreover, our results also matched the observed trend in China in that the percentage of building GHG emissions from the HSCW zone has increased annually from 26% in 2015 to 29% in 2021 [5]. Given that electricity is the source of cooling energy and traditional fossil fuels are still used for its generation in Shanghai [83], GHG emissions from the building sector can be expected to continue to increase.

This study evaluated the impacts of climate change on building energy demand at fine spatial and temporal scales based on an existing building stock database and GIS modeling techniques. When the EUI was used as a metric, hospital and hotel buildings were found to be the most sensitive to climate change, presenting higher RC and AD in annual EUI than other building types in 2050 (Table 4); however, residential and office buildings were found to be the two most significant contributors to the annual total building energy demand increase for the two districts (Table 4), as they have larger floor areas and are more numerous than others (Table 2), which could only be revealed through the use of a database on the existing building stock. In this case, policymakers will also need to consider strategies to minimize the increase in cooling energy demand in residential and office buildings, in addition to hospital and hotel buildings. Utilizing high-efficiency cooling systems is extremely important to directly reduce the cooling energy demand in hospital and hotel buildings, as they operate 24 h a day during summer. Given that the cooling energy demand in office buildings is only high during operation hours, implementing automation systems that can adjust the cooling energy supply based on occupancy might be helpful. Moreover, installing green roofs, cool roofs, and solar panels can also be a good choice for public buildings with high EUIs, such as hospital and commercial buildings. Considering the large number of residential buildings, public education of homeowners regarding energy saving and environmental protection is necessary. The city government can benefit residents who install smart thermostats or use energy-efficient cooling systems for existing residential buildings. On the other hand, they can enforce building codes that require energy-efficient designs, such as better insulation and window–wall ratio, to be used in newly built residential buildings.

The spatial patterns of susceptible building stocks at the individual building level under climate change regarding energy demand change can allow policymakers to develop mitigation strategies tailored to a specific region. For example, Caohejing Town and Xujiahui Town in Xuhui District (Figure 13) showed the largest AD of annual building energy demand increase (reaching 2.0 × 10⁸ kWh), suggesting that local governments should reduce building energy use in these areas first to optimally reduce the total amount of building-related GHG emissions. As our results indicated that increased cooling energy consumption from large office building clusters was the primary cause, implementing systems with dynamic cooling loads and installing solar panels, green roofs, or cool roofs can be considered as suitable strategies. For regions with high RC of annual building energy demand, local governments should check the electricity supply system and ensure that it can fulfill higher energy demands, especially during summer.

The high spatial and temporal resolution of the building energy demand database allowed the most vulnerable regions to climate change to be determined, which is the novelty of this study. In this regard, Hypothesis 2, that “the variation in energy increase across different building types will be larger at finer temporal scales”, was supported. Our results suggested that the increase in hourly residential building energy demand on the typical hot summer day (29 July) under the 2050 climate condition at 1 p.m. was found to increase by more than 40%, which is significantly higher than its annual RC in building energy demand of 2.12%. These dramatic changes could be missed at the coarser temporal scale (annual and monthly), due to values averaged over many time nodes. The peak hours for electricity usage are the most critical factor in determining the balance of the regional electricity supply [84,85,86]. Therefore, regions with a high percentage of residential buildings should adopt adaptation strategies as soon as possible, in order to avoid potential power outage hazards around midday on sweltering summer days. Moreover, aside from its value in potential hazard assessment, the hourly building energy demand database at the individual building level can be combined with various weather prediction models, such as the Weather Research and Forecasting (WRF) model, to quantify the urban heat island (UHI) effect, as building energy use is highly correlated with anthropogenic heat [27]. It can also support the risk assessment of power shortages and heatwave exposure for specific regions during extremely hot summer or cold winter days.

The practical value of our study is that a building energy demand dataset with high spatial and temporal resolution was developed, which can be incorporated into the urban GHG emissions inventory. On the one hand, it allows city governments to tailor mitigation strategies to susceptible regions, such as Caohejing and Xujiahui Town, for which the total annual building energy demand may increase by as much as 2.0 × 10⁸ kWh. The total building energy demand was predicted to increase from 1.23 × 10¹⁰ kWh to 1.25 × 10¹⁰ kWh, with 1.9% growth under future climate conditions. Given that the electricity emission factor in Shanghai was 0.58 kg CO₂/kWh [86] in 2021, this change in electricity consumption can be expected to be associated with a further 116 kt of CO₂ emissions. Although the emission factor of building energy and the efficiency of HVAC systems might change over time, the building energy demand dataset proposed in this study—with frequent updates based on the newest standards (e.g., related to HVAC systems and emission factors)—can allow local governments to quantify GHG emissions from the building sector at different temporal scales (annual, monthly, daily, or diurnal), which could serve as an excellent alternative in cities without energy use monitoring data for a large number of individual buildings, supporting building energy and GHG emissions forecasting functions in systems such as digital twin models at minimal cost. The results of this study can also be integrated with weather prediction models, such as the WRF model, in UHI and “carbon peaking and carbon neutrality” studies at the urban level.

5. Conclusions

This study introduced a comprehensive approach to assessing the impacts of climate change on the energy use of the existing building stock at the city scale in Shanghai, China based on the integration of building energy simulation, climate change modeling, and GIS spatial analysis techniques. First, existing buildings in the Xuhui and Minhang administrative districts were classified into different types based on the XGBoost model and data from multiple sources. Building prototypes were then utilized to simulate the energy demand (in EUI) for various building types under both 2020 and 2050 climate conditions. A methodology was developed to calibrate the hourly simulation results by incorporating actual energy consumption data at larger scales. Subsequently, the calibrated EUI was integrated with building footprint data, including information on building types, areas, and numbers of floors, in order to estimate the energy demand of individual buildings. The approach used in this study can be applied to other cities, as all data used were publicly and freely available. The prototypes of buildings used in energy demand simulation were developed based on Chinese building standards by Tsinghua University, China, which makes them more suitable for similar studies in Chinese cities.

All types of buildings exhibited a net increase in their annual energy demand under the projected future (2050) climate condition, as the increased energy demand during the cooling season cannot be offset by the decrease during the heating season. Significant differences in building energy demand characteristics were found across building types at annual, monthly, and diurnal (hourly) temporal scales, associated with building functions and operation hours. The increase in hourly total building energy demand on typical hot summer days may be significantly higher than that at annual and monthly scales. The spatial pattern of total annual building energy demand at the individual building level exhibited high spatial heterogeneity, reflecting the influence of different types, sizes, and densities of buildings in urban areas and the spatial agglomeration characteristics of urban energy use.

Several limitations may have resulted in some uncertainties in our results. The calibration of residential building energy use was based on energy end-use data at the administrative district scale. The calibration of public building energy use was based on the average of all buildings of the same type that participated in the energy use monitoring program. Due to confidentiality concerns, it is unrealistic for metered energy use data for individual buildings in the study area to become available. Moreover, this study mostly discussed the impacts of climate change on the energy systems of buildings, especially on the cooling and heating energy demand. Certain other factors, such as the complexity of layout and orientation, were not considered. Furthermore, occupancy behavior was set as identical for the same type of building, according to the Chinese design standards for residential and public buildings. Future research can focus on several aspects to improve upon the current study. First, the relationships between energy use behaviors in residential buildings and some factors, such as household income, educational level, and age, can be examined. Second, the spatial pattern of hourly building energy demand can be mapped and compared with building AH discharge calculated using remote sensing imagery in order to promote an understanding of their correlations, which can benefit urban UHI studies. Risk assessment research regarding power shortages and heat wave exposure under climate change scenarios for specific regions can also be conducted.