1 Introduction e energy demand and consumption in urban areas are increasing due to the expanding urban population and rapid urbanization (Gu et al., 2020;Zhao et al., 2019). Cities account for approximately two-thirds of the total primary energy consumption of the world, and urban buildings account for approximately 40% of the energy use in cities (Ali et al., 2021). Climate change significantly affects the requirements for heating and cooling, potentially resulting in either higher or lower energy consumption for buildings, depending on their geographic position (Wang & Chen, 2014). Besides, buildings constructed in different years demonstrate varying energy performance profiles due to the evolution of construction standards, materials, and technologies. Climate change and aging infrastructure have further jeopardized the sustainability and resilience goals of cities by demanding more energy consumption. erefore, there is an urgent need for better urban planning and design to cope with the increase in urban energy consumption and optimize urban energy use.

To simulate the urban building energy use, numerous methodologies and tools have been devised by scholars and, with two distinct approaches in urban building energy models (UBEMs) being top-down and bottom-up approaches (Nesbakken, 1999;Arnfield, 2003;Swan & Ugursal, 2009;Li et al., 2017). e top-down approach is to determine trends in building energy consumption based on econometric (e.g., income and gross domestic product) and technological (e.g., building type and energy standards) data by addressing all buildings as a single energy entity (Ghiassi & Mahdavi, 2017). On the other hand, the bottomup approaches simulate urban energy consumption at a disaggregated level where the models can address the energy use of individual end-uses in the buildings, which can also be aggregated to a larger city scale (Swan & Ugursal, 2009;Li et al., 2017). Researchers tend to prefer bottom-up approaches due to their advantages in handling different scales and providing flexibility when collecting necessary parameters. In cases where certain parameters are impractical to obtain, researchers utilize building standards or prototype models as substitutes. However, existing research in bottom up UBEM encounters two major challenges in terms of data collection and scenarios analysis. For large scale simulation, some necessary parameters are impractical to collect for each building (e.g., window-to-wall ratio (WWR)), which require an efficient way of large-scale building auditing. For scenario analysis, existing research has considered various retrofit measures in urban energy use but fewer of them consider the effect of retrofitting under the impact of climate change. is study proposes a research framework for urban building energy use simulation by integrating various datasets to collect the necessary input parameters. Specifically, we estimate the WWR from street view images by segmenting walls and windows using image segmentation models. To account for the impact of climate change, we also consider different retrofit plans of increasing and decreasing the U value of building envelopes under current and future weather conditions for a costal neighborhood in Galveston, Texas, USA.

Bottom-up UBEMs can be categorized as a physically based approach, a reduced-order approach, or a data-driven approach (Ali et al., 2021;Hong et al., 2020). e physically based approach utilizes information from weather conditions, building characteristics, and construction features in urban/neighborhood scale simulation tools to simulate the end-use energy consumption of each building, with consideration of the interactions between buildings (such as shading and solar reflection). e reduced order approach provides a way to quickly assess building energy performance by using normative model parameters as inputs into computational tools for urban building energy use, reducing data demands compared to the other two approaches. Unlike the previous two approaches based on simulation, the data-driven approach uses regression or machine learning methods to estimate the energy consumption of the building based on its relationship with other buildings and environmental variables such as building characteristics and socioeconomic variables. Among the three approaches, the physically based approach estimates the energy consumption of each building using a simulation engine (such as EnergyPlus) with actual building sample data and considers the thermal influence of the environment, which shows great flexibility and predictive power (Ali et al., 2021). Various computational tools have been developed in UBEM from the physically based approach, including SEMANCO (Madrazo et al., 2012), CitySim (Vermeulen et al., 2013), UMI (Reinhart et al., 2013), CityBES (Hong et al., 2016), UrbanOPT (Polly et al., 2016), and CityBEUM (Li et al., 2018). ese tools run on the Web, as standalone desktop applications, or as plugins in other software.

e physically based UBEMs have powerful simulation abilities and high spatial and temporal resolution in estimating urban building energy use, but the need for large scale and detailed input data remains a challenge (Kavgic et al., 2010). e fundamental task of building a physically based UBEMs is to collect and integrate large scale building datasets including geometric data (e.g., building footprint, number of floors, building envelope, roofs, and windows) and non-geometric data (e.g., U-value of the envelope, land use type of building, occupants, lights, and other equipment) (Ali et al., 2021;Hong et al., 2020). e availability of large-scale, 3D building datasets spanning numerous cities establishes the necessary foundation for obtaining geometric data for preparing a UBEM. However, most nongeometric data still necessitates acquisition through survey of the specific research site. Several cities and organizations have provided building footprint and building height information publicly which makes it easy to build a 3D urban building models for energy use simulation. For example, the Department of information Technology & Telecommunications (DoITT) in New York City released the 3D building model for every building present in the 2014 aerial survey; this model contains Level of Detail (LOD) 1 and 2 information in CityGML format (NYC DoITT, 2022). OpenStreet Map and Microsoft also provide open building footprint and building heigh information for users to download in rich GIS format (Openstreet Map, 2022;Microsoft, 2022). However, these datasets are limited in their spatial and temporal scale and lack other necessary information, such as year built and land use codes. To cope with this limitation, studies have used Light Detection and Ranging (Lidar) data to extract building footprint and building height to construct high-resolution 3D urban models (Bizjak et al., 2021) and to link building information with land use parcel data to determine other attributes (e.g. land use type and built year) of the building within the parcel (Li et al., 2018). Besides, computer vision technology has demonstrated potential capabilities to estimate the non-geometric information from street view imagery of the building (Gao et al., 2024;Ning et al., 2022). Drawn on the characteristics of street view imagery, this non-geometric information usually related to the building façade, such as window-toall ratio (Szcześniak et al., 2022) and architecture age (Sun et al., 2022). Using non-geometric data derived through the utilization of a computer vision model on street view images offers the potential to construct a UBEM that overcomes certain data limitations efficiently.

UBEMs have been widely used for operational energy analysis, energy use optimization, energy use predictions, and retrofit analysis (Ali et al., 2021;Reinhart & Cerezo Davila, 2016). Retrofitting is taken as the most common and feasible strategy to reduce the building energy use and increase thermal comfort for the occupants. According to the review article by Ali et al. (2021), 2363 articles leveraged a physically based method for urban building energy simulations between 2001 and 2020. Among them, 163 articles (6.9%) employed urban building energy models for retrofit analysis. Retrofit plans normally entail updates or replacements to building envelopes or equipment (e.g., heating, ventilation, and air conditioning (HVAC), or lighting) to improve the energy efficiency in buildings (Shen et al., 2019).

In empirical research, the application of retrofit analysis with the utilization of UBEMs across various retrofit scenarios has been employed to estimate the energy performance of urban buildings and optimize their energy usage. For instance, Lam et al. (2008) investigated the potential electricity savings for 10 high-rise office buildings in Hong Kong with retrofit plans improving the building envelope and HVAC system. In addition, Chen et al. (2017) used CityBES to conduct retrofit analysis for 940 office and retail buildings in San Francisco with 6 retrofit plans: replacing lighting with light emitted diodes (LEDs), upgrading the cooling system and heating system, adding an air economizer, replacing windows, or replacing lighting with LEDs and adding air economizers. Ben and Steemers (2020) developed a modelling approach accounting for household archetypes and occupants' behavior variations to provide more efficient retrofit strategies in the UK. Mohammadiziazi et al. (2021) leveraged a UBEM to evaluate the effectiveness of retrofit plans for 209 commercial buildings in Pittsburgh, Pennsylvania and found that upgrading lighting systems and plug and process load reduction increased the average heating energy use intensity (EUI) by 3% and 1%, respectively. ese studies provide valuable references for architects and urban planners by establishing various optimization schemes and simulating energy usage under these schemes, offering meaningful insights for constructive purposes.

Retrofitting is a long-term strategy to cope with the increasing energy use of buildings, but most of the existing literature is focused on analyzing retrofit plans under the current climate conditions. Weather conditions are one of the most important parameters for urban building energy simulation and long-term climate change is likely to have significant impacts on building energy consumption (Li, 2020). Failure to consider future climate change may limit the accuracy of the retrofit analysis results that urban planners and policymakers need for long-term urban energy resilience planning (Mutani et al., 2020).

Building energy consumption is highly sensitive to climate change, since the associated long-term shifts in global or regional climate patterns will alter the heating and cooling demands of buildings (Belzer et al., 2007;Xie et al., 2015). Numerous studies have investigated the impacts of climate change on building energy performance under climate change scenario (Clarke et al., 2018;Fathi et al., 2020;Fonseca et al., 2020;Larsen et al., 2020;Waddicor et al., 2016;Wan et al., 2011Wan et al., , 2012;;Zhou et al., 2013). For instance, Wan et al. (2012) investigated the effects of climate change under two emission scenario in five major cities across five climate zones in China to estimate the future trend of building energy expenditure and provides policy recommendation for mitigation and adaptation. Clarke et al. (2018) explored the effects of climate change on building energy consumption at a global level under representative concentration pathway (RCP) 4.5 and 8.0 scenarios. In addition, a study investigating the impacts of climate change on building energy performance found that US cities in hot or warm and humid climate zones would be most likely to experience climate change-related increases in building energy consumption (Fonseca et al., 2020). To effectively mitigate and adapt to the impacts of climate change, it is necessary to examine current and future urban building energy performance and develop optimal long-term energy conservation measures to enhance urban energy resilience (Mutani et al., 2020).

Only a few studies have examined the energy consumption of urban buildings within the context of combined scenarios that encompass climate change and retrofit plans (Akkose et al., 2021;Buckley et al., 2021;De Masi et al., 2021;Katal et al., 2019). Akkose et al. (2021) investigated the long-term effects of climate change and the urban heat island on the energy use performance of an educational building in Ankara, Turkey.

ey couple future weather projections for 2060 with urban microclimate simulations to modify projections of future weather conditions to account for the urban heat island effect. Several retrofit strategies are considered in their simulation including heating, ventilation, and air conditioning (HVAC) systems, green roof plans, and updates to exterior walls and glazing interventions. Buckley et al. (2021) explored the potential of UBEMs to simulate the current and future building energy demands in Dublin, Ireland with respect to retrofit scenarios.

is neighborhood-scale energy simulation provides information on the spatial and temporal building energy demands, which could be a basis for developing optimal retrofit plans. ese examples suggest that integrating retrofit analysis with climate change scenarios shows great potential for informing engineering approaches to coping with the long-term climatic effects on energy consumption. e integration of climate change and retrofit plans in a joint analysis offers a dynamic perspective that surpasses the insights obtained from individual analyses of either factor alone.

is study applies an UBEM to simulate urban energy use under climate change scenarios and retrofit plans in a costal neighborhood in Galveston, Texas. Urban Modelling Interface (UMI) was selected as the simulation platform due to its rich and user-friendly functionality. Geometric and nongeometric information of the building and weather data are needed to input to the UMI for energy use simulation. We construct a 3D urban model from building footprint geospatial data and Lidar datasets to acquire the geometric information of the buildings including their footprint and height. en we link the buildings geometry with property appraisal data to get non-geometric information including built year and land use type. Additionally, we download the street view images of each building and use an image segmentation model to segment the walls and windows of the building and calculate the window-to-wall ratio. For other non-geometric information, we assume the equipment and lighting parameters to follow ASHRAE codes and we set the occupants load of the buildings according to DOE Prototype Models. Galveston weather data from the DOE EnergyPlus website is used to define current weather conditions. Two future weather scenarios under climate change are established using future weather projection tool. Seven retrofit plans are considered, labeled retrofit plans A to G, which update the building envelope material to thermal property. e Energy Use Intensity (EUI) for status quo of the building under 2021, 2050 and 2080 weather conditions are used as the baseline to compare with the EUI under retrofit plans. e framework of this study is shown in Fig. 1.

Galveston is a coastal city comprised Galveston Island and Pelican Island and located in the Southeast Texas coast. Cities along the Southeast Texas coast have been constantly exposed to climate change-related threats and disasters, which makes it an ideal location for conducting research on climate change (Cai et al., 2023;Coleman et al., 2023;Liu & Mostafavi, 2023). is study uses a census block group (Census tract 7243, block group 3) in Galveston Island, Texas as a test case to simulate urban energy use under climate change scenarios. e study area covers both commercial and residential land use types with relatively diverse socio-economic and demographic characteristics. Figure 2 shows the zoning district of the study area, half of which is in the central business district and the other half is residential area with single-family houses within the historical district of the City of Galveston. ere are 136 residential buildings and 40 commercial buildings located in the study area with a median built year of 1960 and 1970 respectively, suggesting an urgent need for retrofitting to reduce energy consumption (See Fig. 3).

e research site is a mixed-land use district with a relatively uniform racial component. According to the 2017-2021 American Community Survey (ACS), 378 residents live in the study site with average household number 2.02 people.

e primary racial component of the area is White and Asian, which take 77% and 11% of the total population respectively. For housing characteristics in the research site, the median number of rooms is 5.1, 59% of the total housing units are detached single houses, 24% of the total housing units have two houses attached. Half of the occupied housing units use utility gas as the heating gas, and 47% of the occupied housing units use electricity as the house heating energy source. e median value of owner-occupied housing units is $33,3700.

e urban building energy simulation platform used in this research is Urban Modeling Interface (UMI) (Reinhart et al., 2013) which was developed by MIT. UMI is based on Rhinoceros which not only allows users to carry out assessments for operational building energy use and transportation but also for embodied energy. Besides, using UMI supports future implementation of further applications based on Grasshopper and Python scripts from Rhinoceros, which are very versatile and powerful and could be used to extend this research. UMI has expanded the simulation of single building energy simulation to larger-scale simulation by considering the daylight using a light propagation algorithm (Reinhart et al., 2013). UMI has been used in various urban building energy simulation in multiple cities around the world, including Boston, US (Cerezo et al., 2016;Nidam et al., 2023), Kuwait City, Kuwait (Cerezo et al., 2017), Dublin, Ireland (Buckley et al., 2021), etc.

e data used to build the 3D urban model of the research site include building footprint data from Houston-Galveston Area Council Data Hub (H-GAC), Light Detection and Ranging (Lidar) data from National Oceanic and Atmospheric Administration (NOAA), and property appraisal data from the Galveston Central Appraisal District (GCAD). e 2018 building footprint data for the Galveston area include linked parcel IDs and were published by H-GAC in Jan 2021. e shape and boundary of the buildings in the research site can be identified through this building footprint data, but no information for building height, built year, and property use type are provided. We use high-resolution point-cloud based Lidar data from NOAA to derive a surface height layer of the buildings in ArcGIS Pro. e height layer is calculated by subtracting a digital surface model with a digital elevation model generated from Lidar data. e built year and property use information for the buildings are acquired by using a web crawler on GCAD's property appraisal data. We use parcel ID of the building to link with property appraisal data to get the built year and property use types.

e property appraisal data records all historical buildings and improvements in the parcel, but we use the most recent buildings and improvements to define the status quo for the research site.

From the USA climate zone information, Galveston is located in Climate Zone 2A.

e building envelope's material inputs should meet the requirements of American Society of Heating and Air-Conditioning Engineers (ASHRAE) codes for that climate zone.

ere are two main types of buildings in the study area, residential and commercial. According to the building appraisal data, the year of building construction in that area ranges from 1859 to 2017.

e ASHRAE codes and associated years for this simulation include ASHRAE 90.1 (1999 and 2019) for commercial buildings and ASHRAE 90.2 (2007 and 2018) for residential buildings. Commercial buildings with the year of construction before 1999 were assigned code ASHRAE 90.1 1999, but those built after 1999 were assigned code ASHRAE 90.1 2019. Residential buildings with the year of construction before 2007 were assigned code ASHRAE 90.2 2007, but those built after 2007 were assigned code ASHRAE 90.1 2018. Detailed envelope materials and their thermal property for commercial and residential building from ASHRAE are listed in Appendix. It is obvious that the code updates for ASHRAE 90.1 and 90.2 required much higher thermal quality of construction materials including opaque facade, roof, and windows.

Window-to-wall ratio is a key parameter measuring the proportion of windows to the wall for each façade. WWR is important in physically based energy use simulation, which influences both thermal transition and daylight of the buildings. In UBEM, the WWR is commonly assumed to be a uniformed, industry-standard value of 40% because it is impractical to collect WWRs for each building for a large area. Street view images have shown great potentials in semi-or auto-extraction of WWRs for neighborhoods or cities using computer vision technique (Dogan & Knutins, 2018;Szcześniak et al., 2022). Szcześniak et al. (2022) proposed an auto-extraction method of WWR from street view images through pixel counting of grayscale image for the façade after an orthogonal transformation. eir experiment in Manhattan showed a similar result for autoextracted and manually determined WWRs for 1057 buildings with less than 10% difference for 66% of the façade.

Different from the grayscale approach, we apply semiautomated approach of WWR calculation using image segmentation approach from pre-processed street view images.

e methods follow three steps: image pre-processing, image segmentation, and WWR calculation. For image preprocessing, we first crop the orthogonal image of the street-facing façade of each building. For those images that cannot be cropped as orthogonal image, we perform an orthogonal transformation using Photoshop. We selected the least occluded one from multiple years of street view images to enhance the quality of the image dataset. Because street view images can only capture the street-facing façade of the building, we take WWR of one street-facing façade orthogonal image of the building as reference.

e WWR values of the two adjacent walls are set as half of the reference value and the WWR of the opposite façade is set the same as the reference value. Secondly, we apply an image segmentation model, lang-segment-anything to segment the walls and windows from the images (Medeiros, 2023). Lang-segment-anything is an extension of Segment Anything Model (SAM), a zero-shot segmentation model proposed by Meta (Kirillov et al., 2023). By leveraging the image segmentation capabilities of SAM with text prompts, it can effectively identify the target object for extraction.

e segmented wall and window are represented as binary image with white pixels representing them and black pixels representing the background. Finally, we count the number of pixels of wall and windows in the image and calculate the WWR as the number of pixels of windows divided by the number of pixels of the wall. Figure 4 shows the process of WWR calculation.

e load inputs including occupants, equipment, and lighting, are developed based on the requirements of DOE Prototype Models for commercial and residential buildings (Table 1). It is reasonable that occupant load and lighting of commercial buildings are bigger than the ones of residential buildings. However, the equipment load from residential buildings is larger than commercial buildings because of appliances in the kitchen. e usage schedule of the equipment and lighting is set according to the default schedule of UMI.

e standard Galveston EPW weather file available from the DOE EnergyPlus website, which is based on observations made at Scholes International Airport in Galveston, Texas, was used for historical weather data. e Climate Change World Weather File Generator for World-Wide Weather Data (CCworldWeatherGen) by Jentsch et al. (2013), was used for future climate conditions. CCworld-WeatherGen alters the Galveston EPW file in line with projections under an IPCC emissions scenario A2, for which expected global CO2 concentrations for 2050 are 575 ppm.

e current weather file downloaded from DOE Energy-Plus website is assumed to be the weather conditions in 2021. Two future weather scenarios for 2050 and 2080 are projected through CCworldWeatherGen tools.

Retrofit is a feasible way to cope with the increasing urban energy consumption expected under the impacts of climate change. e current building status under the 2021 weather condition is established as the baseline scenario. Seven retrofit plans are set to test the energy saving effects of retrofit and to consider how much retrofit will be needed to alleviate the increasing energy demand induced by climate change. Retrofit Plan A is the assumes that the envelope materials of both commercial and residential buildings will be updated to the newest standards and that the Retrofit Plan B, C, and D assume the U value of envelope materials will decrease by 50%, 30%, and 10% respectively. And Retrofit Plan E, F, and G assume that the U value of envelope materials will increase by 10%, 30%, and 50%. e U value of envelope materials for baseline and retrofit plan A to G are shown in Table 2. 3 Results 3.1 Impact of climate change on temperature e cooling and heating demand will be dramatically changed because of the increase temperature due to climate change in the scenario of 2050 and 2080. e hourly temperatures for the current year, 2050, and 2080 as estimated through CCworldWeatherGen are shown in Fig. 4.

e hourly average temperature of Galveston in 2021 is 21.3 °C. Under the influence of climate change, the average temperature will increase by 2.7 °C and 4.6 °C in 2050 and 2080 respectively, which implies the energy use for cooling in this area will dramatically increase under these two future scenarios. However, the increasing temperatures under the effect of climate change not only affect energy use for heating as well as for cooling. We set the starting temperatures for heating and cooling as 15.5 °C and 22 °C according to conventional practice in urban energy use simulation (Suppa & Ballarini, 2023). e total number of heating hours will be 1849 in 2021, 1331 in 2050, and 905 in 2080. e total number of cooling hours in 2021, 2050, and 2080 will be 4345, 5093, and 5776, respectively. Compared to present-day temperatures, the warmer conditions projected in 2050 and 2080 decrease the number of heating hours by 515 and 944 but increase the number of cooling hours by 748 and 1431 h per year.

e dramatic change in cooling and heating hours and the counter effect of cooling and heating on energy use will reflect on energy consumption in the study area.

Climate change will dramatically increase the operational energy use in our 2050 and 2080 scenarios for buildings in the study area (see Fig. 5).

Table 3 shows the simulated average Energy Use Intensity (EUI) for residential and commercial buildings for different built year under three scenarios. e average EUI of total area is 108.1 kWh/m 2 / year in the research neighborhood in 2021 but it increases to 111.4 kWh/m 2 /year and 114.1 kWh/m 2 /year under the influence of 2050 and 2080 weather conditions, which is 3.0% and 5.5% increase in percentage respectively. For residential buildings, the current EUIs for residential buildings built before and after 2007 are 91.2 kWh/m 2 /year and 94.1 kWh/m 2 /year respectively. Under the 2050 scenario, the EUI of residential buildings built before 2007 and after 2007 increased by 2.1% and 1.8%. In the scenario for 2080, climate change will increase EUI by 4.2% and 3.7% for residential buildings built before and after 2007. For commercial buildings built before and after 1999, the current EUI is 165.8 kWh/m 2 /year and 166.2 kWh/m 2 /year. Under the 2050 scenario, the EUI of commercial buildings built before 2007 and after 2007 increased by 5.0% and 4.4%. In the scenario for 2080, the EUI of commercial buildings built before 1999 and after 1999 will increase by 8.2% and 7.4%. e proportion of increased energy usage in commercial buildings is approximately 2 to 4 percentage points higher than that in residential buildings, indicating the differential impact of climate change on buildings for different purposes. Besides, older residential and commercial buildings are more significantly affected by climate change in their energy use.

Retrofit programs are essential measures in mitigating the negative impact of climate change on building energy use. Figure 6 presents the percentage change of the EUI for total and each type of building under baseline and various assumed retrofit plans (A to G) in three weather conditions. e estimated EUI without retrofit in three weather conditions is used as a baseline for comparison with the four different retrofit plans. e percentage within parentheses represents the change relative to the baseline scenario. Retrofit plan A represents all the building envelopes upgrading to the newest standard and it will result in a 0.09%, 0.29%, and 0.28% decrease on the EUI of the three weather scenarios. is suggests that newer building standards exhibit greater energy-saving characteristics compared to their older counterparts. Retrofit plan B to D assumes a decrease in the U value of the newest building standards. In general, these three retrofit plans will result in reduced energy consumption across all three weather scenarios according to the simulation results. Retrofit plan B is the most energy-saving plan among the three plans, which will lead to 1.73%, 1.2%, and 0.98% decrease of the total area under 2021, 2050, and 2080 weather scenarios.

e decreasing percentage change from 2021 to 2080 also indicates that climate change will alleviate the impact of energy-saving measures through retrofitting. Retrofit plan E to G assumes an increase in the U value of the newest building standards, whose results show an increase in the energy use of the total area under three weather scenarios. Under the 2021, 2050, and 2080 weather scenarios, retrofit plan G is projected to result in an increase in building energy usage by 1.10%, 0.67%, and 0.45% respectively. e simulated EUI exhibits varying degrees of decrease of all the retrofit scenarios relative to the baseline scenario. is demonstrates that retrofit measures can be a feasible way of mitigating the impact of climate change. While both increasing and decreasing the U value of building envelop can significantly impact the energy performance of the building, in our Galveston case, decreasing the U value tends to be a more feasible way to save building energy use.

e effects of climate extremes on urban systems are a global concern for urban residents and coupled physical infrastructure (Ye & Niyogi, 2022). e temperature variation induced by climate change has dramatic impacts on urban energy consumption. is study simulates the larger scale building energy use intensity under scenarios of current weather and projected 2050 and 2080 weather and tests the effect of four retrofit plans to alleviate the increasing energy use intensity. By integrating various dataset including building footprint geospatial data, Lidar, property m 2 /year) EUI of 2050 (kWh/m 2 /year) Impact of 2050 weather EUI of 2080 (kWh/m 2 /year) Impact of 2080 weather Residential building (before 2007) 53833.2 91.2 93.1 2.1% 95.0 4.2% Residential building (after 2007) 1133.9 94.1 95.7 1.8% 97.5 3.7% Commercial building (before 1999) 75652.8 165.8 174.1 5.0% 179.2 8.2% Commercial building (after 1999) 39214.5 166.2 173.5 4.4% 178.5 7.4% Total area 169834.4 108.1 111.4 3.0% 114.1 5.5%

appraisal data, current and projected weather, street view images, we perform a scenarios analysis of the energy use using urban building use model for a neighborhood in the City of Galveston, Texas. Building footprint and Lidar data are used to generate a 3D urban model, which further linked with land appraisal data for its built year and land use type. Street view images are used to calculate the windowto-wall ratio using an image segmentation model. e retrofit analysis is based on the baseline condition and seven retrofit plans under three weather scenarios in 2021, 2050, and 2080. e weather projection shows that there will be a 2.7 °C and 4.6 °C increase in the yearly average temperature in the study area in 2050 and 2080, which will result in a decrease of 515 and 644 heating hours but an increase of 748 and 1431 cooling hours for the two future weather scenarios. e results show that there will be an increase of 3.0% and 5.5% in the EUI of the total area in 2050 and 2080 if the housing condition keeps the current status. Commercial buildings are more significantly impacted by climate change compared to residential buildings. And older commercial and residential buildings are impacted more than the newer ones. Adaptation measures for climate change are needed to cope with increasing urban energy consumption. Retrofit of buildings, especially updating envelope materials, can be one of the major climate change adaptation measures. Seven retrofit plans are tested in the research and are compared with a baseline in which no retrofit plan is performed. All the retrofit scenarios demonstrate varying degrees of decrease of EUI for total area relative to the baseline scenario. e result shows that retrofit plan B, which assumes a 50% decrease on the envelope material, shows the largest energy use savings that reduce EUI by 1.73%, 1.2% and 0.98% for the whole area in three weather scenarios. e efficacy of retrofit plans diminishes in warmer weather conditions, suggesting a juxtaposition between retrofitting endeavors and the influence of climate change. Furthermore, for buildings in Galveston neighborhoods, reducing the U-value of the building envelope tends to be a viable approach for curbing building energy consumption amid rising temperatures from the simulation results. is result can point a direct for location like Galveston to update the building standard and implement retrofit measures.

For model development, this article integrates diverse types of data, notably highlighting the potential of using street view images to estimate WWR by image segmentation technique. By segmenting the wall and window from the pre-processed image of the façade, it is feasible to calculate the WWRs of the building in a neighborhood or a city automatically or semi-automatically. is is a more practical and efficient way than manually auditing the buildings. Street view images have also shown great potential for estimating other key parameters in building energy use simulation, such as built year (Nachtigall et al., 2023) or estimating the building energy efficiency (Mayer et al., 2023).

ere are several limitations for this study. e retrofit plans are determined by assumption, not based on realworld projects, which limits the further analysis of the results. With no real-world practice of retrofitting, it is hard to cost-effective analysis and provide a threshold of the optimal updating of the building envelope based on economic criteria. Second, the impact of urban microclimates and occupants' behavior on building energy use is not considered in the simulation due to the unavailability of relevant data and functional limitations of the UMI. Coupling UBEMs with other urban microclimate models could be a feasible way to account for the effects of the microclimate and occupants' behavior. Finally, the future weather data generated by CCworldWeatherGen is not based on scenarios from the most recent IPCC report. However, the methodological framework developed by this paper could be used to support follow-up studies that address these limitations by incorporating real-world project of retrofitting practice. Construction type Material/layer (outer-inner) Conductivity W/ (mK) Thickness m U value W/ (m^2 k) Designed assembly U/F value (W/ (m^2 k)) Max U value (W/ (m^2 k)) or F code requirement (2019) for climate zone 2A Facade clay-brick 0.410 0.050 8.200 0.839 0.857 xps-board 0.037 0.010 3.700 concrete_block 1.770 0.150 11.800 fiberglass batts 0.043 0.020 2.150 gypsum boards 0.160 0.040 4.000 Ground floor xps-board 0.037 0.050 0.740 0.712 0.730 concrete RC dense 1.750 0.150 11.667 concrete MC light 1.650 0.100 16.500 Cement mortar 0.800 0.030 26.667 ceramic tile 0.800 0.020 40.000 Partition wall (interior) gypsum plaster 0.420 0.020 21.000 4.015 NA softwood general 0.130 0.020 6.500 gypsum plaster 0.420 0.020 21.000 Roof xps-board 0.037 0.150 0.247 0.217 0.221 concrete MC light 1.650 0.150 11.000 concrete RC dense 1.750 0.200 8.750 air floor 15 cm 0.700 0.150 4.667 gypsum boards 0.160 0.020 8.000 Interior floor carpet 0.045 0.020 2.250 0.600 0.608 Cement mortar 0.800 0.020 40.000 concrete RC dense 1.750 0.250 7.000 air floor 15 cm 0.700 0.300 2.333 gypsum boards 0.160 0.100 1.600 Windows glass 0.900 0.007 128.571 2.540 2.550 argon air 0.006 5.400 glass 0.900 0.007 128.571 argon air 0.006 5.400 glass 0.900 0.007 128.571