An empirical study on the compound effects of extreme weather and UHIon building energy consumption under local climate

An empirical study on the compound effects of extreme weather and UHIon building energy consumption under local climate | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article An empirical study on the compound effects of extreme weather and UHIon building energy consumption under local climate Shuyang Zhang, Nianxiong Liu, Xiyu Wu, Yichen Han, Wenwen Li, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7763869/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Due to climate change, extreme weather (EW) events like heatwaves and cold snaps are becoming more frequent, challenging urban buildings and energy systems. Urban heat island (UHI) effects—where city centers are significantly warmer than suburbs at night—further impact heating and cooling demands of urban buildings. However, there is still a lack of systematic empirical studies linking meteorological data to building energy use, especially regarding the compound effects of UHI and extreme weather on urban building energy consumption and load during winter and summer. To address future complex climate conditions, we propose using localized weather data (LWD) that fully accounts for both background EW and UHI effects. Driven by suburban meteorological observations and high-resolution land cover data, the data is generated using the UWG urban canopy model and the UMEP tool on the QGIS platform to capture realistic local weather conditions around buildings. It can be directly input into urban building energy model (UBEM) for the corresponding local climate zones to simulate building energy use. Our study shows that LWD better captures seasonal building energy use and the effects of external and internal factors. Compared to suburban weather station data, accuracy improves by 29.6% in summer and 36.6% in winter during the typical year, and by 35.1% and 30.1% during the extreme weather year, respectively. Local air temperature (Ta) has the greatest impact on actual energy use, followed by solar radiation (Rad)—especially during summer heatwaves, when Rad may exceed Ta in influence. Internal disturbances have a greater impact in summer, but their influence lessens during extreme weather due to stronger external climatic effects. This method supports refined assessment and control of UBEM across climates and seasons, helping manage energy peaks during heatwaves and prevent overheating in winter, ultimately aiding real-weather-based energy system optimization and urban design. Civil Engineering Localized weather data Extreme weather Urban heat island UWG Energy use intensity Hourly load Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 1. Introduction In recent years, worsening global climate has led to more frequent extreme weather (EW) events [ 1 – 4 ]. In June 2023, Beijing faced a record-breaking heatwave with temperatures over 40°C for three days, with a peak of 42°C. That December, an eight-day cold snap brought daily lows below − 12°C [ 5 , 6 ]. In the U.S., a February 2021 cold wave hit Texas, causing ice-covered power lines and blackouts [ 7 , 8 ]. In summer 2023, a heatwave triggered massive air conditioning (AC) loads, again leading to widespread outages and rising electricity prices. The extreme weather events raised serious concerns about the resilience of buildings and energy systems under severe climate conditions. Beyond EW, urban heat island (UHI)—driven by heat-retaining surfaces and anthropogenic heat (AH) emissions from AC—leads to higher nighttime air temperatures in dense city center compared to suburban regions. Research indicates that UHI can increase cooling energy use by 10–120%, and reduce heating demand by 3–45% [ 9 ]. Ignoring UHI effects can lead to overestimating heating demand in winter and underestimating cooling demand in summer, which leads to misestimation of operational carbon assessments [ 10 , 11 ]. The compound effect of EW and the UHI has significantly altered urban wind speed and temperature patterns. In summer, due to the UHI effects, dense urban forms further amplify the impact of extreme heat on cities [ 12 ]. A study in Seoul, South Korea, found that during the 2012 and 2016 heatwaves, peak UHI intensity was 3.3°C and 4.5°C higher than during non-heatwave periods [ 13 ], significantly increasing peak cooling loads and annual energy demand [ 12 ]. In San Francisco, USA, research showed outdoor air temperature differences of up to 11°C between coastal and downtown areas during heatwaves, with heating energy use varying by over 100%, cooling by 65% across sites, and peak cooling demand differed by as much as 30% [ 14 ]. A study in Tianjin, China, found that the compound effect of heatwaves and UHI increased nighttime residential building loads, with peaks occurring between 2:00 and 4:00 a.m. For each 1°C increase in UHI intensity, hourly cooling loads increased by 0.5 W/m² [ 15 ]. With ongoing urbanization, the climate risks arising from the interaction between heatwaves and UHI are expected to intensify [ 16 ]. Some studies also highlight UHI’s potential benefits in cold weather. A study in Tokyo showed that, when both cooling and heating energy use were considered, UHI increased overall energy use in commercial buildings but reduced it in residential ones, resulting in lower total urban energy demand [ 17 ]. In London, UHI raised annual urban cooling loads by 25% while reducing heating loads by 22% compared to rural areas [ 18 ]. Building energy use is driven by external and internal factors. External factors include meteorological conditions (e.g., air temperature (Ta), solar radiation (Rad), relative humidity (RH), wind speed (WS)) and the surrounding urban context [ 19 ]. Meteorological conditions include both urban background climate and localized weather. Nearby buildings can obstruct solar radiation and wind flow, altering the target building’s local thermal conditions [ 20 ]. In addition, dense urban forms often intensify the local UHI effect, raising AC loads. AC systems, in turn, emit AH into the surrounding area, further reinforcing local warming [ 21 ]. These factors jointly affect the building’s thermal environment and significantly influence its energy use [ 22 ]. Internal factors involve occupancy, space use, equipment, and lighting. Differences in occupancy patterns and equipment use lead to varied energy use across building types. Residential and office buildings typically show cyclical patterns due to regular schedules [ 23 ]. Weighted linear regression of the ASHRAE Global Thermal Comfort Database II shows that for every 1°C rise in outdoor temperature, indoor temperatures increase by 0.1°C in AC buildings and 0.4°C in naturally ventilated ones [ 24 ]. Studies indicate that a 1°C increase in outdoor temperature leads to an approximate 14% increase in cooling energy use intensity (EUI)—defined as the annual cooling load per unit floor area of a building—and about a 10% decrease in heating EUI [ 25 ]. Especially during heatwaves, AH emissions may rise by 20% due to higher cooling demand, with AC accounting for more than 85% [ 26 ]. Therefore, beyond background meteorological conditions, local climates, shaped by urban form and human activity [ 27 , 28 ], significantly affect building energy demand. The Urban Building Energy Model (UBEM) is a tool for modeling and simulating the performance of multiple buildings within a city, supporting urban-scale building energy assessments, policy development, AH estimation, and urban planning and redevelopment [ 29 ]. Unlike the dynamic energy use simulation of individual buildings, UBEM accounts for inter-building interactions (e.g., shading, radiation effects) and the coupling between building energy use and the urban microclimate (e.g., AH emissions) [ 29 ]. Due to the heterogeneity of the urban local environment, local climates vary significantly across different buildings, directly impacting building load estimates in UBEM [ 30 ]. To address this, localized weather data (LWD) should be integrated into UBEM to enable local climate-energy coupling [ 31 , 32 ]. Incorporating local factors such as solar radiation, temperature, wind, and humidity—particularly the compound effects of EW and UHI conditions [ 16 , 33 ]—enables more accurate predictions and better understanding of how local climates affect building energy use [ 34 ]. This approach supports climate-responsive urban design, including building form, envelopes, and external spatial configurations [ 22 ]. In recent years, Urban Canopy Models (UCM) have become key tools for modeling energy balance and surface temperatures across urban surfaces, such as roads, roofs, and walls [ 35 ]. Driven by boundary meteorological data, UCMs can simulate long-term local urban weather conditions. Integrating UCM with UBEM enables more accurate representation of local climates and building energy use. Compared to geometry-intensive CFD models like Envi-met, UCM offers faster, parameterized LWD simulations over wider urban areas and longer timeframes (e.g., full-year, 8760-hour weather data), making it more suitable for large-scale UBEM applications. Current studies on LWD and building energy simulations mostly rely on Typical Meteorological Year (TMY) data. TMY is synthesized from long-term weather records to represent average climate conditions, excluding EW and spatial variability [ 36 , 37 ]. As a result, it cannot capture extreme events or local climates—such as heat islands, wind fields, longwave radiation between buildings, or local Ta and RH [ 32 ]. Thus, while TMY-based simulations can assess UHI impacts on energy use, they fall short in evaluating the accuracy of UCM or the effect of LWD on building energy simulation results. They also cannot reflect real energy use under extreme weather combined with UHI effects. In practice, validating simulations with measured local climate and building energy consumption data is essential, especially for large-scale seasonal UBEM calculations [ 38 ]. To address this, we conduct an empirical study using real suburban weather station data, proposing a method for localized weather-driven UBEM to explore three key research questions: First, how much does suburban-station-driven LWD improve the accuracy of simulation results compared to nearby stations or TMY data? Compared to the sole impact of UHI, how do the compound effects of extreme events like heatwaves or cold spells influence building energy use and design-day loads? Second, what are the differences and patterns between hourly simulated and actual loads under the individual and compound effects of EW and UHI conditions? Finally, to what extent do internal factors and local climate variations influence hourly energy loads in summer and winter? This study utilizes real weather and hourly building energy data to compare simulation accuracy across three aspects: weather data, annual energy use, and hourly profiles. It highlights discrepancies between simulated and actual hourly loads under the compound effects of extreme weather and urban heat islands. The analysis also identifies the relative importance of microclimate factors under different climatic conditions, aiming to pinpoint the key factors for effective climate risk mitigation. 2. Literature review of localized weather data Generally, models for generating LWD can be categorized into two approaches: physics-based and data-driven (Table 1 ). Physics-based models follow general physical laws, offer broad applicability, but require complex computations. Data-driven models are faster and more accurate within their training range, but they are limited to specific urban contexts. Physics-based models vary by input type. Models using environmental parameters are computationally efficient but lack detailed spatial characteristics. Models using geometric inputs—typically through Computational Fluid Dynamics (CFD)—can simulate detailed urban forms and produce high-resolution outputs, though requiring higher computational costs. These physics-based models operate at different spatial scales, each suited to specific resolution and duration needs. LWD are often generated using UCM, CFD, or mesoscale meteorological models coupled with UCM, with the last approach requiring multi-scale integration [ 39 – 41 ]. The LWD with high spatiotemporal resolution is critical for UBEM applications. The meteorological data used to drive the model—often at an hourly resolution—determines the simulation’s temporal resolution and duration. The driving meteorological data can come from mesoscale meteorological model outputs, global reanalysis datasets, boundary-layer data, or suburban ground station observations. In this study, we use the Urban Weather Generator (UWG), a UCM-based tool driven by suburban weather station data, to generate LWD. Table 1 Summary of localized weather data and building energy use calculation methods across different spatial scales [ 41 ] [ 30 ]. Methods Approach Climate tools Energy models Input data type Spatial resolution Time resolution Time span Advantages Limitations Reference Mesoscale meteorological model Physical-based model WRF, Meso-NH, COSMO-CML EnergyPlus (CityBES), CitySim, TRNSYS Parameter Meso-scale: >1km 1h Several months Suitable for large-scale applications; can run with UCM or independently High computational cost; limited accuracy in handling urban microclimates, such as the urban canopy effect [ 42 – 49 ] Urban canopy model (UCM) Physical-based model TEB (UWG), BEP, CAT, CIM, CLMU, UCP EnergyPlus (DesignBuilder, Dragonfly), SUNtool (UMI), TRNSYS, CitySim, BEP-BEM Parameter Local Scale: 0.1-1km 1h 1 year Running fast; supporting high spatiotemporal resolution simulations; includes AH and UHI Requires detailed local surface, building, and vegetation parameters input; limited spatial resolution [ 10 , 12 , 26 , 27 , 38 , 50 – 69 ] Microscale geometry model (CFD) Physical-based model Envi-met (RANS), ANSYS Fluent, Openfoam, SOLENE-Microclimate EnergyPlus (DesignBuilder), TRNSYS, ESP-r, BuildSysPro, HTB2, EnviBatE Geometry Micro-scale: 1m-10m 1h Several days High spatiotemporal resolution; ideal for microclimate analysis of airflow, ventilation, and pollutant diffusion between buildings Requires detailed geometry data input; high computational demand; limited to small areas and short periods (hourly or daily simulations) [ 34 , 70 – 79 ] Weather station & mobile observation Data-driven model Urban/suburban stations, mobile sensors EnergyPlus (DesignBuilder), TRNSYS, CitySim, CERMA, PREDISE / / ≤ 1h Varies High measurement accuracy Limited spatial coverage; data discontinuity [ 16 , 28 , 80 – 83 ] Weather station-based spatial interpolation Data-driven model Meteonorm 8 CitySim, TRNSYS / 0.1-1km 1h Varies Data with spatial continuity Accuracy depends on weather station density; interpolation errors exist [ 37 , 84 – 89 ] Remote sensing (empirical formula) Data-driven model MODIS land surface temperature (LST), Landsat LST / / Urban scale:0.1-1km 1 month Several years Data with spatial continuity; useful for night temperature estimation Low temporal resolution; sensitive to cloud cover; weak correlation with daytime Ta [ 85 , 90 – 92 ] Design-day temperature profile Data-driven model STEVE EnergyPlus, IES-VE / / 1h / Easy to obtain; low data cost Statistical model that cannot reflect the impacts of complex urban contexts on microclimates [ 93 – 95 ] Machine learning (ML) Data-driven model CNN, RNN, XGBoost, LSTM EnergyPlus (Dragonfly) / Local Scale: 0.1-1km 1h Varies Fast prediction, high spatial accuracy and precision Requiring large training datasets; relying on patterns within existing data; limited accuracy in predicting extreme conditions [ 32 , 96 , 97 ] 3. Materials and method 3.1. Study area and data collection This study focuses on a 27-story student dormitory in Beijing’s Haidian District (Fig. 1 c), located at the southwest corner of a residential complex (Fig. 1 b), with an average floor height of about 3 meters. LWD for 2021 (a typical year) and 2023 (an extreme year) were driven by data from a suburban weather station in Haidian (Fig. 1 a). On-site Ta and RH data for validation were collected in 2021 from a rooftop station in a nearby residential complex (Fig. 1 d) [ 98 ]. Actual hourly electricity use data covering full-year periods in 2021 and 2023 were provided by the building management. Winter energy use data for both years were estimated based on records from the district’s heating plant. EUI measurements were taken from rooms occupied during both summer and winter in 2021 and 2023. 3.2. Method This section focuses on the methods and technical details related to local weather and building energy simulation, hourly heating energy consumption calculation, the development of the LSTM prediction model, and the SHAP explanation algorithm. Sections 3.3.1 and 3.3.2 respectively describe the simulation of localized weather data and building energy use. Section 3.2.3 explains how measured heating energy consumption is converted into equivalent “electricity consumption” to facilitate comparison. Sections 3.2.4 and 3.2.5 cover the LSTM model architecture and key parameters, as well as the logic of the SHAP algorithm for interpreting time series prediction models. This enables the assessment of the relative contributions of internal and external microclimate factors on actual energy loads under different climatic conditions. Local weather and energy simulations were conducted using UWG and EnergyPlus, respectively. ① LWD was simulated using the UWG model within UMEP 4.1.1 integrated in QGIS 3.28 [ 99 ], ② providing input for UBEM and allowing flexible grid resolution according to the study area size. Building energy modeling was performed using Dragonfly and Honeybee modules within the Grasshopper, which support flexible 3D UBEM construction and simplify the modeling process [ 38 ]. ③ Building footprint data with height attributes from QGIS were imported into Rhino 7.31, converted into 3D geometry, ④ and then assembled into UBEM using Dragonfly and Honeybee (Fig. 1 e). ⑤ The UBEM was run in EnergyPlus to output hourly heating and cooling loads per unit area for all 8,760 hours, which was then compared with measured data. As heating and cooling loads are more sensitive to external conditions than lighting or plug loads, the analysis focused on the impact of LWD on these two end uses. For consistency, all EUI and hourly load results are reported in kWh/(m²·year) and W/m², respectively. 3.2.1. UWG model simulation of UHI The UWG is based on the Town Energy Balance (TEB), a single-layer UCM [ 53 ], and simulates long-term Ta, RH, and WS in simplified 2D urban canyons [ 55 ]. It offers high resolution, flexibility, and fast computation, and has been validated in cities like Singapore, Basel, Toulouse, Rome, Barcelona, Boston, and Guangzhou [ 100 ]. UWG can generate high-resolution LWD (tens to hundreds of meters) and allows parameter adjustments by urban area, building type, and season to improve accuracy [ 101 ]. Anthropogenic heat released from AC in dense urban areas, especially at night, slows the cooling rate compared to suburban areas and intensifies the nighttime UHI effect. UWG integrates a rural station model (RSM), vertical diffusion model (VDM), urban boundary layer model (UBLM), and a TEB and building energy model (TEB-BEM) module [ 101 ], which is coupled with EnergyPlus to account for canyon AH emissions from buildings and to capture the impact of building energy use on the local climate. By inputting suburban weather station data with urban background meteorological information into the UCM, the compound effects of EW and UHI are effectively captured. Unlike other UCMs that need top-canopy meteorological data, UWG incorporates RSM, VDM, and UBLM, and requires only suburban weather station data as inputs. Its open-source code has been integrated into platforms like QGIS (UMEP) and Rhino (Dragonfly), making it highly adaptable for integration with UBEM. We used the UWG module within the UMEP toolkit in QGIS to simulate gridded LWD. Driven by suburban weather data and 2D raster inputs, the tool enables fast calculation of year-round LWD (Fig. 2 ) without the complex 3D modeling required for large-scale urban vegetation and buildings. UWG’s temporal flexibility also allows users to define custom simulation periods. Required inputs include: suburban weather data, gridded local weather zones with building function and construction year classifications (as defined by UWG), land cover data, and urban morphology within each grid cell (Fig. 2 ). The output localized weather data supports both EPW and TXT formats. EPW files can be used directly in EnergyPlus, while TXT files support weather data analysis. The land cover data, used as UCM parameter input in UWG, includes six categories: Paved, Building, Tree, Grass, Bare Soil, and Water. This data has a resolution of 1 meter and is derived from GF-7 and Sentinel-2 satellite imagery. Building morphology within each grid is preprocessed in UMEP, which requires a Digital Elevation Model (DEM) containing ground elevation and a Digital Surface Model (DSM) with surface object heights. 3.2.2. Building energy model setup Before running EnergyPlus, the BEM must be configured with key parameters, including geometry, construction, occupancy and equipment schedules, window-to-wall ratio (WWR), and heating, ventilation, and air conditioning (HVAC) settings (Table 2 ). Surrounding context buildings are also defined to account for shading. To further address urban-scale scenarios, we use Dragonfly for modeling. Dragonfly can convert 3D geometry into BEM or UBEM. The target building is modeled with 3-m floor heights and divided into thermal zones by floor. We used the “Solid to DF Model” module to account for variations in roof height within the building. HVAC is set as an ideal air system with a default COP of 1. All surrounding obstructions within a 200-m radius are included. With the BEM and LWD prepared, EnergyPlus is used to simulate building energy performance under localized weather conditions. This study localizes the building’s program, construction, and WWR settings based on ASHRAE 90.1 classifications, aligned with China’s GB 55015 − 2021 “General Code for Energy Efficiency and Renewable Energy application in buildings”[ 102 ]. Detailed parameters are listed in Table 2 . The program is set to Highrise Apartment , and the construction follows ASHRAE 90.1 2016–4A Mixed Humid-Mass , aligning with Beijing’s ASHRAE climate zone classification (4A Mixed Humid) [ 103 ]. Table 2 Urban building energy model parameter settings [ 102 ]. Parameter Setting Weather data Epw file Scenario 1: Beijing TMY_CSWD data (Station number: 545110) Scenario 2: Haidian (HD) Station data (Station number: 54399) Scenario 3: Localized weather data (UWG) Time Simulation period Annual (8760h) Resolution Hourly Geometry Floor number 15 floors (N–S), 27 floors (E–W) Thermal zone Floor-based zone Gross floor area (m 2 ) 38253.65 Construction Window glass U-value (W/m 2 ·K) 1.8 Window solar heat gain coefficient (SHGC) 0.3 Exterior wall U-value (W/m 2 ·K) 0.45 Roof U-value (W/m 2 ·K) 0.3 Program Heating season indoor temperature setpoint (°C) 18 Cooling season indoor temperature setpoint (°C) 26 Occupant Density 25 m 2 /person Lighting 6 W/m 2 Equipment 10 W/m 2 Schedule Highrise_Apartment schedule (Fig. 6 -(a)-i) HVAC Maximum heating supply Ta (°C) 50 Minimum cooling supply Ta (°C) 20 Ventilation Rate 30 m 3 / (h·person) Infiltration 0.0003 m 3 /s per m 2 facade COP 1 WWR East / South / West / North 0.35 / 0.5 / 0.35 / 0.3 3.2.3. Method for calculating hourly heating demand and "electricity demand" in winter To estimate hourly winter heating demand, we used hourly building energy use data from a nearby district heating plant. Based on the total serviced building area and accounting for thermal distribution losses of heating network, we calculated the building’s hourly heating energy use per unit area. As the heating plant provides continuous centralized heating to surrounding residential buildings, the resulting EUI reflects actual building energy use under local weather conditions. The calculation method is as follows: (1) The hourly heating energy consumption per unit area from the heating plant can be calculated using specific heat capacity. $$\:{q}_{station}=\frac{c\cdot\:m\cdot\:{\Delta\:}T}{3600·A}=\frac{c\cdot\:m\cdot\:({T}_{s}-{T}_{r})}{3600·A}\left(1\right)$$ Here, \(\:{q}_{station}\) is the hourly heating supply per unit area (kWh/m²); c is the specific heat capacity of water, typically 4.186 kJ/(kg·°C); m is the hourly mass flow rate of water in the heating network (kg/h); ΔT is the temperature difference between supply water temperature (Ts) and return water temperature (Tr); A is the total heated floor area served by the heating plant (m²). (2) With \(\:{q}_{station}\) known, the building’s hourly heating demand per unit area in winter, \(\:{q}_{heating}\) , can be estimated accordingly [ 104 ]. $$\:{q}_{heating}=\eta\:\cdot\:{q}_{station}\cdot\:\left(\frac{1}{1+\alpha\:}\right)\times\:\beta\:\:\left(2\right)$$ Here, η is the heat loss rate of the courtyard distribution network, typically 2%–10%, this study uses a correction factor of 0.98. α is the overheat factor due to lack of terminal control, set at 20% for district heating systems. β is the weather correction factor, used to adjust heating demand based on actual weather conditions. $$\:\beta\:=\frac{HD{D}_{0}}{HDD}\:\left(3\right)$$ Here, HDD₀ is the standard heating degree days (°C·d) based on an 18°C baseline, set at 2699°C·d for Beijing. HDD refers to the actual heating degree days for the year, calculated using the same baseline—2535°C·d in 2021 and 2577°C·d in 2023 based on Haidian station data [ 105 ]. (3) To enable comparison across winter, summer, and annual energy use, the actual hourly heating demand per unit area \(\:{q}_{heating}\) is converted to equivalent electricity demand \(\:{q}_{electricity}\) under simulated HVAC conditions. $$\:{q}_{heating}={q}_{\text{e}\text{l}\text{e}\text{c}\text{t}\text{r}\text{i}\text{c}\text{i}\text{t}\text{y}}\cdot\:\:{COP}_{heating}\:\left(4\right)$$ Here, \(\:{COP}_{heating}\) is the coefficient of performance for residential HVAC systems, set to 2.6 [ 102 ]. 3.2.4. LSTM-based building energy consumption prediction model To analyze the impact of internal and external factors on real-time building energy loads and to predict energy use under extreme weather, this study develops a Long Short-Term Memory (LSTM) deep learning model. LSTM is a type of recurrent neural network (RNN) designed for sequence prediction tasks, such as time-series forecasting and speech recognition. LSTM is trained by inputting existing multistep sequences of variables to predict their values in other or future time periods [ 106 , 107 ]. The LSTM model, built using the TensorFlow framework, includes an LSTM layer with 128 units, a Dropout layer to prevent overfitting, a Flatten layer to reshape the output, and a Dense layer for the final prediction. The model uses the Adam optimizer with a default learning rate of 0.001, combining momentum and adaptive learning rate techniques to improve convergence efficiency. By applying adaptive learning rates to each parameter, the update step is adjusted by assigning smaller rates to large gradients and larger rates to small gradients. Key parameter settings are shown in the Table 3 . Table 3 Key parameters and settings of the LSTM prediction model. LSTM model parameter Setting Explanation and function Dropout 0.1 Prevents overfitting by randomly deactivating neurons during training to improve generalization. Learning Rate 0.001 Controls step size in weight updates; too high may cause instability, too low may slow training. Units 128 Number of hidden neurons in the LSTM layer; more units improve model capacity but also raise the risk of overfitting and computation time. Time Steps 10 Number of past time points used as training input; short sequences may miss context, while longer ones increase computational cost. Epochs 100 Number of full training cycles; each epoch processes all training data with one weight update. Too few epochs may underfit, while too many may overfit. Batch Size 256 Number of samples per training batch; small batches train faster but are less stable, while large batches are more stable but resource-intensive. 3.2.5. SHAP interpretation model To analyze the influence of internal and external microclimate variables on actual load at different times in the LSTM model, as well as how these influences vary under different climate effects, the study used the SHapley Additive exPlanations (SHAP) explanation algorithm. SHAP enhances the interpretability of LSTM models by applying game theory to provide both global and local explanations. The Shapley value \(\:{\varphi\:}_{i}\left(f\right)\) for an input feature \(\:{x}_{i}\) is calculated using the following formula: $$\:{\varphi\:}_{i}\left(f\right)=\sum\:_{S\subseteq\:N\setminus\:\left\{i\right\}}\:\frac{\left|S\right|!\left(\left|N\right|-\left|S\right|-1\right)!}{\left|N\right|!}\left[{f}_{S\cup\:\left\{i\right\}}\left(S\cup\:\left\{i\right\}\right)-{f}_{S}\left(S\right)\right]\:\left(5\right)$$ Here, \(\:{\varphi\:}_{i}\left(f\right)\) is the Shapley value of feature \(\:{x}_{i}\) ; N is the full set of features; S is any subset excluding \(\:{x}_{i}\) ; ∣ N ∣ and ∣ S ∣ are the sizes of the full set and subset, respectively. \(\:{f}_{S}\left(S\right)\) is the model prediction using subset S , and \(\:{f}_{S\cup\:\left\{i\right\}}\left(S\cup\:\left\{i\right\}\right)\) is the prediction after adding \(\:{x}_{i}\) . In LSTM models, features span multiple time steps. Each feature at each time step has a specific Shapley value, allowing the contribution of time-dependent inputs to the target variable to be assessed. To evaluate a feature's overall influence across a past time window, we sum the absolute Shapley values of that feature at all time steps. SHAP not only quantifies the importance of each input feature but also captures how its influence on energy use changes over time. This helps reveal both long-term and periodic dependencies between input features and the target variable [ 108 ]. 4. Results 4.1. Comparison of suburban and local air temperature around the target building We compared UWG-simulated local Ta around the target building with actual readings from the Haidian (HD) suburban weather station during the 2023 heatwave (around June 17–23) and cold wave (around December 21), referencing the same periods in 2021 for comparison (Fig. 3 ). Using the daily average temperature of the TMY as a baseline, significant differences can be observed between the extreme weather year (2023) and the normal year (2021). (i) 24-hour temperature profiles: In summer 2021, the average maximum UHI intensity stayed below 8°C. During the 2023 heatwave (around June 23), it exceeded 10°C, an increase of 2°C that can significantly impact building energy consumption. Studies suggest that a 1°C rise in temperature can increase building peak electric loads by up to 4.6% and total energy use by up to 8.5% [ 55 ]. In winter 2021, UHI intensity was around 5–6°C, while the 2023 cold wave reduced it by only 0.5°C. UHI intensity also peaks at different times of the day. In summer, it is strongest between 2 and 6 a.m. and weakest at noon [ 15 ]. At dawn, before solar radiation reaches the ground, residential areas retain heat due to overnight AC use, slowing cooling, while suburban sites like the HD weather station reach their lowest Ta of the day. In winter, UHI intensity is also lowest around noon but peaks around midnight, as heating systems remain on throughout the day while internal heat gains from occupants and equipment decreases during nighttime. (ii) Continuous temperature curves: During the summer heatwave, amplified UHI pushed nighttime ΔTa above 10°C, sharply increasing residential cooling loads. In winter, UHI-driven ΔTa stayed around 6°C, but the nighttime warming effect was significantly reduced during the cold wave, with ΔTa dropping below 3°C. (iii) Temperature deviation scatter plots: Local temperatures were generally higher than those at the suburban station due to UHI. During the 2023 heatwave, the compound effect of heatwave and UHI led to a high RMSE of 6.47°C between Ta_UWG and Ta_HD. During the winter cold wave, the RMSE dropped to 3.83°C. Extreme weather affects the temperature gap between local and suburban areas, amplifying it in summer and narrowing it in winter. 4.2. Meteorological data validation We compared 2021 rooftop weather station data from a nearby residential building with Ta_HD, Ta_UWG, RH_HD, and RH_UWG to assess differences between suburban (HD) weather data, UWG-simulated, and measured LWD, and to validate the accuracy of UWG outputs (Fig. 4 ). Figure 4 (i) and (ii) highlight the 24-hour and daytime differences in Ta and RH between LWD and suburban station data after accounting for local weather. Compared with the HD station, UWG-simulated LWD aligned closely with real local measurements, with R² values of 81.45% for summer Ta and 82.12% for winter Ta (Fig. 4 (a, c)-iii). For RH, R² reached 86.95% in summer and 74.82% in winter (Fig. 4 (e, g)-iii). Based on the validation results, although UWG slightly overestimated nighttime UHI intensity in some periods, it effectively captured anthropogenic heat effects and explained local climate patterns caused by local UHI effects. Therefore, UWG-simulated Ta and RH data are suitable for building energy simulations. In contrast, suburban station data differed significantly from real local data due to the UHI effect. In summer, the temperature difference peaked at 8.3°C at midnight, and 6.8°C at 6 a.m. The largest daytime temperature difference occurred at 3 a.m. on June 19, with UHI reaching 11.6°C (Fig. 4 (b)-ii). In winter, the temperature difference decreased, with the largest 24-hour difference at 11 p.m. (5.0°C) and the largest daytime difference (6.6°C) at 7 a.m. on December 18 (Fig. 4 (d)-ii). These measurements show that nighttime Ta gaps between suburban and local rooftop data were significant in both seasons, with UHI peaking around midnight or just before sunrise. Relative humidity differences in summer peaked at 33% at 6 a.m. and 52% at 6 a.m. on June 19. In winter, the largest 16% difference occurred at 11 p.m., aligning with the temperature differences. Daytime differences peaked at 32% at 7 a.m. on December 15. Higher relative humidity in suburban areas are attributed to abundant vegetation and stronger transpiration [ 109 ], which leads to lower temperatures and higher moisture content in the air [ 110 ]. 4.3. Accuracy of annual energy use and peak load simulation Keeping the target building constant, this study used LWD to simulate annual heating and cooling energy use for 2021 and 2023. The simulated EUI was compared with actual measured data. Changes in weather conditions were also analyzed for their impact on heating/cooling EUI and design day loads (Load_DDY). Here, EUI_UWG represents energy use intensity under UHI-influenced conditions, with the 2023 data reflecting extreme weather scenarios. Comparing Fig. 5 (a), the 2021 values of EUI_HD and EUI_UWG show that under the UHI effect alone, the target building’s cooling demand increased by 8.38 kWh/(m²·year) and heating demand decreased by 10.76 kWh/(m²·year), resulting in a net annual energy reduction of 2.38 kWh/(m²·year). In 2023, under compound influence of UHI and extreme weather, comparisons between EUI_HD and EUI_UWG show that the UHI effect led to a summer cooling demand rose by 10.33 kWh/(m²·year), a winter heating demand dropped by 10.56 kWh/(m²·year), and a slight reduction in total energy use by 0.23 kWh/(m²·year). Comparisons of 2021 and 2023 EUI under HD data show that extreme weather alone increased total energy use by 8.01 kWh/(m²·year), with cooling up by 5.25 kWh/(m²·year) and heating up by 2.76 kWh/(m²·year). Further comparisons of 2021 and 2023 EUI_UWG show that under the compound influence of EW and UHI, total energy use rose by 10.16 kWh/(m²·year), with a 7.20 kWh/(m²·year) increase in summer cooling and 2.96 kWh/(m²·year) in winter heating. UHI significantly raised summer EUI. Comparisons of 2021 EUI_HD with 2023 EUI_UWG reveal that the compound warming effect of heatwaves and UHI sharply increased summer cooling EUI by 15.58 kWh/(m²·year), posing a significant challenge to the urban energy system [ 16 ]. In contrast, the opposing influences of cold waves and UHI lowered winter heating EUI by 7.80 kWh/(m²·year). UHI’s long-term influence proved more dominant, while extreme weather events were short-term and abrupt, leading to a total energy increase of 7.78 kWh/(m²·year). Additionally, Fig. 5 (a) shows that simulations using LWD tend to slightly underestimate EUI, while HD and TMY data tend to overestimate heating EUI and underestimate cooling EUI, especially in TMY, which does not account for climate change and exaggerates heating demand while underestimating cooling. When comparing percentage deviations from actual values, LWD yields more accurate estimates of individual heating and cooling EUI than using suburban weather station data. Using LWD improves simulation accuracy for cooling and heating EUI by 36.6% and 29.6% in a typical year, and by 30.1% and 35.1% in an extreme year. Although total EUI from HD data may appear closer to actual values due to offsetting errors (overestimated heating and underestimated cooling), this balance is coincidental. Results calculated using suburban weather station data still carry considerable uncertainty, especially in estimating seasonal heating and cooling EUI or equipment load. Whenever possible, LWD is strongly preferred. We further compared the 2021 and 2023 Load_DDY using both suburban weather station data (HD) and LWD (UWG). Subplots (b) and (c) show that under UHI influence alone, the 0.4% design day cooling load (Cooling Load_DDY 0.4%) in summer decreased by 0.31 W/m² in 2021 and by 0.11 W/m² in 2023. Similarly, the 99.6% design day heating load (Heating Load_DDY 99.6%) in winter dropped by 1.54 W/m² and 3.40 W/m², respectively. This indicates that UHI notably offset the impact of the 2023 winter cold wave on heating loads. Comparing Load_DDY HD between 2021 and 2023 shows that under the influence of a heatwave alone, the Cooling Load_DDY 0.4% increased by 1.37 W/m², while under the influence of a cold wave alone, the Heating Load_DDY 99.6% increased by 3.68 W/m². Further comparing Load_DDY UWG between 2021 and 2023 reveals that under the compound effects of UHI and a heatwave, the Cooling Load_DDY 0.4% increased by 1.57 W/m², and under the compound effects of UHI and a cold wave, the Heating Load_DDY 99.6% increased by 1.82 W/m². The presence of UHI significantly reduced winter peak heating loads. Comparing Load_DDY HD in 2021 with Load_DDY UWG in 2023, we see that the joint effect of a heatwave and UHI increased the Cooling Load_DDY 0.4% by 1.26 W/m², while the compound effect of a cold wave and UHI increased the Heating Load_DDY 99.6% by just 0.28 W/m². In summary, as organized in Table 4 , we examined the individual and compound impacts of UHI and EW on heating and cooling EUI and design day loads. For target residential buildings, EW increases both heating and cooling EUI and design day loads, placing additional strain on building energy systems. Although UHI increases summer cooling EUI [ 111 ], it reduces winter heating demand, total annual EUI, and both cooling and heating design day loads [ 112 ]. This aligns with the findings of Y. Hirano and T. Fujita regarding the UHI's impact on heating and cooling energy consumption in Tokyo's residential buildings [ 17 ]. Erell et al. also found that increased summer cooling demand from UHI may be offset by winter heating energy savings [ 50 ]. Notably, UHI raises winter temperatures, reducing heating load fluctuations during cold waves [ 113 ]. Similarly, in summer, UHI moderates local temperature fluctuations compared to suburban areas, also reducing variations in cooling loads. Table 4 Impact of EW, UHI, and their compound effects on building heating/cooling EUI and design day loads. Scenario Urban heat island only (2021 HD, 2021 UWG) Extreme weather only (2021 HD, 2023 HD) UHI + extreme weather (2021 HD, 2023 UWG) Cooling EUI Increase (8.38 kWh/(m²·year), 57.8%) Increase (5.25 kWh/(m²·year), 36.2%) Increase (15.58 kWh/(m²·year), 107.4%) Heating EUI Decrease (10.76 kWh/(m²·year), 34.8%) Increase (2.76 kWh/(m²·year), 8.9%) Decrease (7.80 kWh/(m²·year), 25.2%) Total EUI Decrease (2.38 kWh/(m²·year), 5.2%) Increase (8.01 kWh/(m²·year), 17.6%) Increase (7.78 kWh/(m²·year), 17.1%) Cooling Load_DDY Decrease (0.31 W/m², 1.8%) Increase (1.37 W/m², 8.1%) Increase (1.26 W/m², 7.4%) Heating Load_DDY Decrease (1.54 W/m², 6.4%) Increase (3.68 W/m², 15.4%) Increase (0.28 W/m², 1.2%) * The values in parentheses represent the change amount and its proportion relative to the original values before local climate effects. 4.4. Correlation between measured loads, simulated loads, and weather parameters We analyzed the correlation between hourly measured loads and simulated loads or weather parameters during extreme weather, using three datasets: UWG, HD, and TMY (Table 5 ). This helps assess how closely each dataset reflects real conditions. Results show that Ta, RH, and simulated loads based on both UWG and HD data correlate significantly with actual loads, but the LWD based on UWG performs best. Ta, RH, and simulated loads from UWG data show the strongest correlations with measured loads (e.g., 2021 cooling load: 0.51***; 2023 cooling load: 0.56***; 2021 heating load: 0.65***; 2023 heating load: 0.57***), outperforming both HD and TMY data. During heatwaves, simulated cooling loads and Ta are positively correlated with actual cooling load, while RH shows a negative correlation. In cold waves, simulated heating loads are positively correlated with actual heating loads, while Ta and RH show negative correlations. In the same year, load and temperature correlations with actual values are stronger in winter than in summer. Table 5 Pearson correlation between hourly measured heating and cooling loads and UWG, HD, and TMY weather data and simulated loads during extreme weather period in 2021 and 2023. 2021 measured cooling load (kWh/m 2 ) 2023 measured cooling load (kWh/m 2 ) 2021 measured heating load (kWh/m 2 ) 2023 measured heating load (kWh/m 2 ) Simulated load Load_UWG (kWh/m 2 ) 0.51*** 0.56*** 0.65*** 0.57*** Load_HD (kWh/m 2 ) 0.16** 0.42*** 0.55*** 0.49*** Load_TMY (kWh/m 2 ) -0.06 0.25*** -0.23*** 0.14** UWG meteorological parameters Ta_UWG (°C) 0.55*** 0.58*** -0.72*** -0.66*** RH_UWG (%) -0.39*** -0.52*** -0.25*** -0.46*** WS_UWG (m/s) 0.11 0.14* 0.05 0.16** Rad_UWG (W/m 2 ) -0.12 -0.08 -0.10 -0.08 HD meteorological parameters Ta_HD (°C) 0.18** 0.39*** -0.60*** -0.58*** RH_HD (%) -0.28*** -0.46*** -0.22*** -0.32*** WS_HD (m/s) 0.05 0.15* 0.05 0.17*** Rad_HD (W/m 2 ) -0.12 -0.08 -0.10 -0.08 TMY meteorological parameters Ta_TMY (°C) -0.06 0.25*** 0.20*** -0.17*** RH_TMY (%) 0.21*** -0.07 0.46*** -0.40*** WS_TMY (m/s) 0.00 0.18** -0.19*** 0.29*** Rad_TMY (W/m 2 ) -0.22*** -0.19** -0.15** -0.07 * The p-value indicates statistical significance; the smaller the p-value, the more significant the correlation. Results with p > 0.05 are considered not statistically significant (*** p < 0.001, ** p < 0.01, * p < 0.05). 5. Discussion This section compares the numerical differences between simulated and measured hourly heating and cooling loads under different climate conditions using three weather datasets (UWG, HD, and TMY), and analyzes the discrepancies caused by extreme weather, urban heat islands, occupant behavior, and limitations of HVAC models based on ideal air systems. It also introduces an LSTM-based calibration method to improve simulation accuracy. Additionally, it examines how the influence of internal and external disturbances on actual energy loads varies over time under extreme weather, highlights the overall impact weight of key factors, and identifies critical microclimate elements for climate risk mitigation. 5.1. Comparison of 24-hour and multi-day trends between simulated and measured loads during extreme weather We compared the relationships among simulated load, air temperature, scheduled occupancy patterns, and actual load in winter and summer under two scenarios: the typical year with isolated UHI and the extreme year with combined extreme weather and local heat islands. We also examined the multi-day variations and 24-hour patterns in the discrepancies between simulated and actual loads (Fig. 6 and Fig. 7 ), and explored the underlying causes. 5.1.1. Comparison of cooling load in summer Under the influence of isolated UHI during typical summer days, the simulated load closely follows the diurnal Ta changes (Fig. 6 (a–c)-i), showing clear peaks and troughs aligning with sunset (around 18:00) and sunrise, respectively. This indicates that simulated loads are strongly influenced by rising surface Ta and Rad gains, which together increase building energy loads. Among the three weather datasets, LWD reflects nighttime UHI effects, producing smoother load curves that align better with actual data than HD or TMY. However, discrepancies between the assumed design schedules and actual occupant and device usage still lead to differences in simulated energy loads. The real load is heavily influenced by occupant behavior, particularly daily routines. In summer, this results in a noticeable shift, with actual peak loads occurring later—around midnight—compared to the earlier peaks. Based on the multi-day summer trends (Fig. 6 (a-c)-ii), unlike simulated loads that closely follow Ta changes, actual loads are influenced by occupants' AC use habits. For example, on cooler days like June 17 and 24, 2021, and June 20, 2023, simulated loads dropped with lower Ta, but some users still ran air conditioning out of habit, leading to a mismatch between simulated and actual energy use. Using LWD, the simulated load tends to slightly overestimate the peak values compared to actual load, whereas HD captures peak values more accurately but underestimates minimum loads due to the omission of UHI effects. Compared to suburban station and TMY data, simulations using LWD align more closely with actual loads along the 45° line (Fig. 6 (a–c)-iii), with the lowest RMSE and highest Pearson correlation. Under the compound effects of heatwave and UHI during the extreme summer days in 2023, the simulated cooling load peaks still aligned with the Ta peak at 6 p.m., while the actual load peaks remain at midnight (Fig. 6 (d–f)-i). Peak demand reached 23 W/m² (Fig. 6 (d)-ii), nearly double that of the same period in a typical year—an increase of 11 W/m²—placing significant short-term stress on the urban summer power grid. Although LWD slightly overestimated loads in a normal year due to occupancy assumptions, it underestimated the peak cooling load during the heatwave by about 5 W/m². HD and TMY data, which ignore UHI, further underestimated cooling load during heatwave (Fig. 6 (d–f)-ii) [ 38 ]. TMY also fails to account for long-term climate warming. Relying on these datasets could lead to significant underestimation of peak demand during heatwaves, posing greater potential risks. Hourly load scatter plots (Fig. 6 (d–f)-iii) show that during the heatwave, LWD data consistently produced load estimates closer to actual values than those from HD or TMY. A comparison of actual daytime loads in 2021 and 2023 shows that heatwaves significantly increased both peak and cumulative loads (Fig. 6 (g)-ii). During the 2023 heatwave days, the actual daily peak load rose to 17.5 W/m²—an increase of 8.5 W/m² over 2021—while daily fluctuations reached 10 W/m², 5 W/m² higher than in 2021 (Fig. 6 (g)-i). In summary, the following conclusions can be drawn: (1) Summer load variations driven by actual occupancy patterns differ from those simulated by ideal air systems, which respond sensitively to weather conditions. This leads to mismatched peak times and an underestimation of actual peak loads during heatwaves. (2) User behavior is habitual—once air conditioning is turned on, some users continue using it even after weather cools down. As a result, the timing of the first heatwave can significantly influence total seasonal cooling demand. Overall, while external factors like weather affect energy loads, they are only part of the picture. Despite some limitations—such as mismatched peak timing, underestimation of peak loads, and reliance on outdoor Ta rather than user behavior—LWD-based simulations still align more closely with real load trends during the heatwave than other datasets, helping improve prediction accuracy. 5.1.2. Comparison of heating load in winter Under the influence of isolated UHI during typical winter days in 2021, the 24-hour daily load profile (Fig. 7 (a–c)-i) shows that actual heating loads remain highly stable, with minimal response to outdoor air temperature changes. Central heating, unaffected by user behavior or manual activation, maintained a steady hourly load with no distinct peaks. In contrast, simulated loads—based on ideal air systems—fluctuate with outdoor Ta, showing clear peaks and troughs. However, the peaks and troughs within the daily cycle balance out, resulting in a similar overall intensity to measured daily loads. From the multi-day view (Fig. 7 (a)-ii), although winter heating loads simulated using LWD differ significantly from actual hourly patterns, both follow similar daily trends, rising and falling noticeably with Ta changes. The scatter plot in Fig. 7 (a)-iii reveals much greater variation in simulated loads, creating a distorted distribution compared to real loads. HD and TMY simulations show similar issues, but with one key difference: winter UHI reduces heating demand. Since HD and TMY don’t account for UHI and have lower, more fluctuating Ta, they lead to substantial overestimation and increased volatility in heating load predictions. Under the compound effects of cold wave and UHI during the extreme winter days in 2023, actual heating loads remained relatively stable throughout the day (Fig. 7 (d)-i). Unlike during heatwaves, simulated loads—based on ideal air systems—overestimated actual demand due to differences in HVAC system behavior, supply temperature settings, and reduced COP under extreme cold. LWD overestimated the 24-hour daily average and peak heating loads by approximately 3 W/m² and 16 W/m², respectively (Fig. 7 (d)-i,ii), while HD data resulted in even larger overestimations—around 7 W/m² for the average and 18 W/m² for the peak (Fig. 7 (e)-i,ii). LWD is suitable for estimating winter loads under normal conditions but less reliable during extreme cold. Relying on simulated loads for heating system control may lead to actual heating exceeding real demand, resulting in over-supply, thermal discomfort, and energy waste. Additionally, because simulated loads are based on ideal air systems, they respond sensitively to temperature drops during cold waves, whereas actual district heating systems adjust based on end-point temperature feedback. The thermal inertia of the building’s walls causes indoor temperature changes to lag behind outdoor temperature variations, resulting in a delayed response due to thermal inertia and control lag (Fig. 7 (d)-ii). A comparison of actual daytime loads in 2021 and 2023 (Fig. 7 (g)-ii) show that cold wave significantly raised both peak and cumulative heating demand (represented by the area between the curve and the x-axis). Unlike in summer, actual winter hourly heating loads—whether in typical or cold wave periods—remain relatively stable throughout the day without sharp peaks or troughs (Fig. 7 (g)-i). However, during cold waves, average hourly heating EUI increases by about 2 W/m². In summary, winter heating load simulations have several limitations: (1) Due to the flexible temperature control settings of the ideal air system, the simulated load is overly sensitive to temperature variations, which amplifies daily heating load fluctuations. This is inconsistent with the relatively stable load variations of central heating systems in northern regions, resulting in peaks and troughs not observed in the actual daily cycle. (2) Due to the end-point temperature feedback-dependent nature of district heating systems, actual heating load responses are delayed. Despite these limitations, applying detailed district heating HVAC models in UBEM is often impractical, due to the complexity of simulating plants, distribution, substations, and terminals for every building making large-scale modeling difficult. To improve winter heating load estimation in UBEM, machine learning models are needed to better capture and predict realistic heating load patterns. To address the over-sensitivity of ideal air system simulations to temperature changes, we applied an LSTM time-series model aimed to predict actual winter energy use under EW conditions. The model uses LWD (Ta, RH, WS, Rad), simulated loads, occupancy schedules, and local time as input features, with actual heating load as the prediction target. Using LWD improves the accuracy of meteorological inputs and reduces unnecessary temperature fluctuations, enhancing winter load prediction under EW. The model was trained on data from 2021 winter and validated using data from 2023 winter (Fig. 8 ). The model achieved an R² of 78.38% (Fig. 8 (iii)), with optimal epochs selected based on the loss curve (Fig. 8 (iv)). In the prediction results, the model did not fully capture the lag in daily peak heating demand and slightly underestimated intraday variation, which was about 3 W/m² on December 21, 2023, but it reduced daily fluctuations and captured overall trends well (Fig. 8 (ii)). 5.2. Impact of internal and external factors on winter and summer building loads To better understand the impact of internal and external factors on hourly building loads under extreme weather, we developed LSTM models using 2021 and 2023 LWD and local time as inputs, with actual energy use as the target. We compared models using only external inputs with those including both external and internal factors to assess their ability to explain actual energy use (Table 6 ). External factors refer to microclimate variables, while internal factors, such as occupancy and equipment use, are difficult to measure but follow daily patterns, making local time (LT) a useful proxy. Results show that energy prediction accuracy was similar across extreme and typical year for the same season. External factors had a greater impact on heating loads than on cooling loads, indicating that winter energy use is more sensitive to outdoor weather than summer energy use. Including internal factors (via local time) improved cooling load prediction by up to 9.8% and heating load prediction by up to 2.7%, suggesting that internal influences are less significant in winter. Table 6 Test set R² of LSTM load prediction models (100 epochs). Scenario LWD_summer_2021 LWD _summer_2023 LWD _winter_2021 LWD _winter_2023 External factors only 0.6101 0.5852 0.7767 0.8163 Internal + external factors 0.7079 0.6070 0.8038 0.8360 5.2.1 Hourly variation patterns of internal and external impacts on winter and summer actual building loads To further identify the most influential microclimate factors affecting actual load under isolated UHI and combined UHI–extreme weather conditions, as well as the temporal variation patterns in the influence weights of microclimate factors and occupant behavior on actual load, we applied the SHAP algorithm to interpret the key drivers in the LSTM load prediction model. This allows for a comparison of the temporal variation patterns in the influence weights of five factors—air temperature (Ta), relative humidity (RH), wind speed (WS), solar radiation (Rad), and local time (LT)—under different local climate conditions in typical and extreme years in both winter and summer, including the hourly variations and peak timings of each factor’s SHAP effect. 5.2.1.1. Variation of factor weights in summer In summer (Fig. 9 ), Ta effect closely follows the trends of both Ta and mean load , peaking after 10 p.m., shortly after the Ta peak and aligning with the mean load peak. Its lowest point occurs around 5 a.m. (Fig. 9 (a)-Ta (i)). During heatwaves in 2023, Ta effect peaks earlier—around 5 p.m.—when temperatures are highest (Fig. 9 (b)-Ta (i)). Ta effect aligns more closely with hourly load peaks during high load periods than at other times from daily variation, both in typical summers and during heatwaves (Fig. 9 (a, b)-Ta (ii)). RH and WS effects are weaker and show poor alignment with load peaks in both normal and extreme years (Fig. 9 (a, b)-RH,WS(i, ii)). Rad effect peaks around 5 p.m. in normal year, when radiation drops rapidly (Fig. 9 (a)-Rad(i)). During heatwaves, it shows two peaks (Fig. 9 (b)-Rad(i)): one at midday when Rad is highest, and another around 8 p.m. when it drops rapidly. Local time effect displays a strong diurnal pattern, closely matching hourly load trends (Fig. 9 (a,b)-LT (ii)) and perfectly aligning with peak mean load (Fig. 9 (a,b)-LT (i)), indicating internal factors dominate summer peak demand, while microclimate elements are secondary. Overall, Local time dominates during the night and early morning, while weather factors are more influential in the midday and afternoon. This reflects typical residential occupancy patterns: low daytime occupancy coincides with higher outdoor Ta, while nighttime occupancy is higher when outdoor Ta is relatively lower [ 114 ]. 5.2.1.2. Variation of factor weights in winter In winter (Fig. 10 ), whether in typical year or during cold waves, the Ta effect peaks around 3 p.m., aligning with the daily high in Ta, and drops to its lowest point around 4 a.m., when Ta is relatively low and indoor occupancy is highest (Fig. 10 (a, b)-Ta(i)). During colder periods—especially the cold wave— Ta effect aligns closely with peak hourly load , indicating stronger temperature influence on heating demand (Fig. 10 (a, b)-Ta(ii)). RH effect and WS effect show weaker correlations with mean load and hourly load peaks (Fig. 10 (a, b)- RH, WS (i, ii)). However, during cold wave days, the WS effect aligns more closely with the Hourly load peak, as higher wind speeds under low temperature accelerate heat loss from building surfaces and increase heating load demand (Fig. 10 (b)–WS(ii)) [ 115 , 116 ]. Interestingly, despite lower winter Rad (about half of summer levels), the Rad effect still shows two daytime peaks during cold waves: one at noon (solar peak) and another around 8 p.m., when Rad is zero (Fig. 10 (b)-Rad(i)). The evening peak of Rad effect closely matches the mean load peak, likely due to Rad dropping to zero while other factors contribute less at that time. Local time effect differs significantly from summer. It peaks around 5 a.m. (Fig. 10 (a,b)-LT(i)), but this does not align with the mean load peak at 8 p.m., indicating that higher nighttime occupancy does not increase heating loads. This is due to buildings using centralized district heating from thermal stations in winter, which is not directly controlled by occupant behavior. 5.2.2. Impact weights of internal and external factors on winter and summer actual building loads To further clarify the dominant microclimate factors influencing building energy consumption under different climate conditions in both typical and extreme years, we visualized the overall influence weight proportions of different factors (Fig. 11 -i), as well as their 24-hour diurnal and daytime variations patterns (Fig. 11 -ii, iii). We also analyzed the temporal changes in the influence weight proportions of microclimate factors and occupant behavior on actual heating and cooling loads. From the diurnal perspective (Fig. 11 (iii)), whether in typical or extreme years, during heatwaves or cold waves, Local time has the highest influence around 4 a.m. [ 15 ]—a period of peak occupancy, no solar radiation, and stable outdoor microclimate before sunrise. The lowest Local time influence varies by season: in summer, it occurs around 2 p.m. when outdoor Ta peaks and occupancy is lower (Fig. 11 (a,b)-iii); in winter, it drops lowest point around 4 p.m. (Fig. 11 (c)-iii) as occupancy decreases and microclimate factors like Ta and Rad drop sharply before sunset. During cold waves, as rapid Ta changes occur later, this low point of the influence of local time shifts to 7–8 p.m. (Fig. 11 (d)-iii). Rad and Ta show typical intraday patterns—lower influence in the early morning and late at night, and higher in the afternoon and evening. Radiation, in particular, varies strongly with solar angle throughout the day. Comparing seasons, Local time has a greater influence in summer, while outdoor air temperature becomes more dominant in winter. From the daily variation (Fig. 11 (ii)), around the summer heatwave (Fig. 11 (b)-ii, June 17 and 23, 2023), the influence of Ta and RH drops significantly, while Rad becomes the dominant factor. During the winter cold wave (Fig. 11 (d)-ii, December 21, 2023), the influence of Ta increases sharply, while the impact of Local time —representing internal factors—decreases noticeably. From the overall importance (Fig. 11 (i)), during typical summer days, Local time has the greatest impact on actual load (Fig. 11 (a)-i), followed by Ta and Rad, with WS having the least effect. During heatwaves (Fig. 11 (b)-i), Rad becomes the most influential factor, surpassing both Local time and Ta. During typical winter days, Ta dominates due to lower solar angles and the continuous operation of central heating unaffected by user behavior (Fig. 11 (c)-i), followed by local time and then solar radiation. During cold waves (Fig. 11 (d)-i), the influence of Ta increases further, while the weight of Local time decreases, though the overall factor ranking remains similar to that of a typical year. In summary, compared to summer, Local time —representing occupancy and equipment use patterns—has less impact in winter due to continuous heating, while the influence of outdoor weather, especially air temperature, on actual load increases. During extreme weather events in both seasons, external factors gain importance, while internal factors become less influential. 6. Conclusion UHI effects vary locally and persist long-term, while extreme weather is unpredictable, with strong interannual and seasonal variation. Under global warming, with widespread local UHI, and more frequent extreme events, TMY data can no longer accurately represent real urban local weather conditions. To address the compound impact of UHI and EW on building energy use and loads, this study introduces LWD and conducts an empirical analysis on a high-rise student dormitory in a cold region. We examine how external local climate conditions and internal factors (occupant and equipment use) affect real heating and cooling loads and EUI under the compound effects. This study also provides empirical support for scaling the method to broader urban applications. The study highlights the reliability and advantages of LWD in building energy simulation. Compared to suburban weather stations and TMY data, it more accurately captures the weather conditions around the target building, providing more precise external inputs for large-scale UBEM [ 38 ]. This improves simulation accuracy and strengthens the model’s ability to explain actual heating and cooling intensity, as well as daily load variations. 6.1 Key research findings and actionable insights Findings show that occupancy schedules heavily influence simulation results, and LWD is key to improving accuracy [ 50 , 55 ]. This study focuses on three main questions: (1) Compound effects of EW and the UHI on local weather and annual building energy use simulation accuracy. (2) Hourly differences and patterns between simulated and actual loads under individual and compound effects of EW and the UHI. (3) The temporal variation in the influence weights of localized weather and internal factors on the actual loads. The key conclusions are summarized as follows: (1) For the target residential building, localized weather data improved accuracy by 29.6% in summer and 36.6% in winter during typical years, and by 35.1% and 30.1% during extreme years. Unlike EW, which increases both EUI and load, UHI raises summer cooling EUI but reduces winter and annual EUI, as well as both heating and cooling design-day loads. (2) Hourly load simulations using ideal air systems may underestimate or overestimate load fluctuations during heatwaves or cold spells. In particular, for winter heating, the ideal air system shows oversensitivity to outdoor temperature changes, leading to unrealistic load fluctuations and significant overestimation of energy demand during cold waves. To address this, simulation results needed to be calibrated using real district heating load data combined with LSTM-based models. (3) Occupancy significantly affects summer cooling loads but has less impact on winter district heating. Unlike summer cooling loads, which are strongly influenced by occupant behavior, winter district heating loads are less sensitive to occupancy in diurnal cycles. During extreme weather in both summer and winter, the influence of external factors increases, while the weight of internal factors decreases. Based on the key findings, feasible strategies can be proposed for differentiated building design, building energy system control, urban energy system planning, and policy development tailored to localized urban climates. (1) Climate-responsive building design and energy system control: During heatwaves and cold spells, external factors such as solar radiation and air temperature become more dominant, while internal factors have less impact. Design strategies should consider the time-varying weights of external and internal factors and improve the local climate through coordinated design at both the individual building and urban cluster levels. In summer, the solar heat gains can be reduced through reflective roofs [ 117 , 118 ], movable shading devices, or self-shading building forms during extreme heat events. In winter, enhanced thermal insulation of building envelopes and leveraging UHI through urban form (e.g., slowing heat loss in street canyons through cluster design) can help mitigate cold wave impacts and reduce climate-related risks. Heating systems can also be preloaded ahead of cold spells based on accurate forecasts to avoid delays in heat supply. (2) Urban energy system planning and policy support based on local climate: Traditional TMY data no longer suffice for building energy consumption assessment and urban design in dense high-density areas under the complex future climate conditions caused by compound impact of EW and UHI conditions. This study recommends incorporating LWD into building energy simulations—particularly in areas with strong UHI effects and frequent EW events, as well as for evaluating individual heating and cooling demands. This approach enables more accurate assessments of urban building energy use and carbon emissions. LWD can also be used to develop urban local climate zone (LCZ) maps and inform climate-specific design strategies, enabling differentiated design based on LCZ. It also supports the refinement of climate-responsive building energy efficiency standards and carbon accounting protocols, and helps develop urban risk zones and energy capacity planning models to optimize the deployment of power grids and district heating networks [ 119 ]. 6.2 Methodological strengths, limitations, and future outlook This study proposes a coupled UCM + UBEM framework that incorporates LWD into the energy simulation workflow—enhancing conventional UBEM, which typically relies only on building program and construction classifications, by incorporating more accurate weather data inputs to improve simulation accuracy. Compared to the chained CFD (Envi-met) + BEM approach, this method supports longer time spans (full-year, hourly simulations) and larger spatial scales (from neighborhoods to entire cities). It captures the compound impact of UHI and EW on annual building energy intensity. The parameterized UCM approach avoids the complex modeling of vegetation and building geometry. Using the UWG tool, suburban weather station data can replace meteorological inputs at the top of the urban canopy, and its integration with BEM allows for the impact of AC heat emissions to be fully considered. Additionally, based on QGIS platform, the resolution of localized weather grids can be flexibly defined, enabling finer analysis of environmental influences on local climates. This approach supports large-scale block or city-level modeling with LWD support. Its application improves UBEM accuracy, enhances understanding of external influences, and provides decision-making support for renewable energy–based urban energy planning under future climate change to enhance urban and building climate resilience [ 34 ]. The study also applies an LSTM time-series model and SHAP interpretability, using Local time to reflect internal factors such as occupancy and equipment use, capturing their time-varying impact on energy loads. However, limitations remain, including arbitrary grid definitions, the inability of the UWG tool to account for vegetation evapotranspiration [ 120 , 121 ], and a fragmented workflow across QGIS and Rhino platforms for LWD generation and building energy simulation. In the future, the workflow can be streamlined by integrating LWD calculation, urban building modeling, and UBEM simulation into a unified tool within Rhino or GIS, enabling more efficient use by urban planners and architects. Understanding the interaction between local climate and energy use is critical for long-term [ 69 ], climate-sensitive urban energy planning. Accurate local climate data and baseline models are essential for informed building design and post-construction evaluation [ 22 ]. Climate-integrated design tools enable better prediction and response to climate–building interactions, supporting climate-adaptive architecture design. This approach fosters deeper integration of climate science with building and urban design, and provides reliable data for more accurate calculation of operational carbon emissions, providing robust data support for achieving carbon neutrality in the built environment. Declarations Data availability The data that has been used is confidential. Declaration of generative AI in scientific writing During the preparation of this work the authors used ChatGPT 4.0 for language refinement. After using this service, the authors reviewed and edited the content as needed and take full responsibility using this service, the authors reviewed and edited the content as needed and take full responsibility Acknowledgements Funding: This work was funded by the National Key R&D Program of China [2022YFC3803801 in 2022YFC3803800], National Natural Science Foundation of China [NO.42171337], Beijing Natural Science Foundation [grant No.8232030]. References Jay O, Capon A, Berry P, Broderick C, de Dear R, Havenith G, Honda Y, Kovats RS, Ma W, Malik A (2021) Reducing the health effects of hot weather and heat extremes: from personal cooling strategies to green cities. Lancet 398(10301):709–724 Benestad RE, Lussana C, Dobler A (2024) Global record-breaking recurrence rates indicate more widespread and intense surface air temperature and precipitation extremes. Sci Adv 10(45):eado3712 Zhu Z, Shao L, Lu R, Hua W (2025) Two contrasting tropical convection modes from the eastern Pacific to northern Africa that drive Eurasian teleconnections in boreal summer. npj Clim Atmospheric Sci 8(1):56 Zhu Z, Yang Y, Liu B (2025) Physics-based predictions of the month-by-month summer western North Pacific anomalous anticyclone. J Clim 38(10):2187–2203 Zhang W, Clark R, Zhou T, Li L, Li C, Rivera J, Zhang L, Gui K, Zhang T, Li L (2023) : Weather and climate extremes hitting the globe with emerging features, Springer, 2024 Xiao H, Xu P, Wang L (2024) The unprecedented 2023 North China heatwaves and their S2S predictability. Geophys Res Lett 51(6) e2023GL107642. Entity NFERCR (2021) The february 2021 cold weather outages in texas and the south central united states, FERC and NERC and Regional Entity Staff, Tech. Rep Magness B (2021) Review of February 2021 extreme cold weather event–ERCOT presentation. Urgent Board of Directors Meeting, ERCOT Santamouris M, Cartalis C, Synnefa A, Kolokotsa D (2015) On the impact of urban heat island and global warming on the power demand and electricity consumption of buildings—A review. Energy Build 98:119–124 Hashemi F, Mills G, Poerschke U, Iulo LD, Pavlak G, Kalisperis L (2024) A novel parametric workflow for simulating urban heat island effects on residential building energy use: Coupling local climate zones with the urban weather generator a case study of seven US cities. Sustainable Cities Soc 110:105568 Palme M, Inostroza L, Villacreses G, Lobato-Cordero A, Carrasco C (2017) From urban climate to energy consumption. Enhancing building performance simulation by including the urban heat island effect. Energy Build 145:107–120 Perera ATD, Javanroodi K, Mauree D, Nik VM, Florio P, Hong T, Chen D (2023) Challenges resulting from urban density and climate change for the EU energy transition. Nat Energy 8(4):397–412 Ngarambe J, Nganyiyimana J, Kim I, Santamouris M, Yun GY (2020) Synergies between urban heat island and heat waves in Seoul: The role of wind speed and land use characteristics. PLoS ONE 15(12):e0243571 Hong T, Xu Y, Sun K, Zhang W, Luo X, Hooper B (2021) Urban microclimate and its impact on building performance: A case study of San Francisco. Urban Clim 38:100871 Meng F, Zhang L, Ren G, Zhang R (2023) Impacts of UHI on variations in cooling loads in buildings during heatwaves: A case study of Beijing and Tianjin, China. Energy 273:127189 Yang X, Yao L, Li M, Cao J, Zhong Q, Peng W, Wu W, Zhou J (2025) How heat waves and urban microclimates affect building cooling energy demand? Insights from fifteen eastern Chinese cities. Appl Energy 384:125424 Hirano Y, Fujita T (2012) Evaluation of the impact of the urban heat island on residential and commercial energy consumption in Tokyo. Energy 37(1):371–383 Kolokotroni M, Zhang Y, Watkins R (2007) The London Heat Island and building cooling design. Sol Energy 81(1):102–110 Li W, Zhou Z, Wang C, Han Y (2024) Impact of outdoor microclimate on the performance of high-rise multi-family dwellings in cold areas and optimization of building passive design. Build Environ 248:111038 Hui Y, Tang Y, Yang Q, Mochida A (2024) Numerical study on influence of surface vegetation on aerodynamics of high-rise buildings. Sustainable Cities Soc 107:105407 Yuan C, Zhu R, Tong S, Mei S, Zhu W (2022) Impact of anthropogenic heat from air-conditioning on air temperature of naturally ventilated apartments at high-density tropical cities. Energy Build 268:112171 Yang L, Liu Y, Cao Q (2024) Toward interdisciplinary integration of architecture and climatology, The Innovation Energy 1(4) 100054-1-100054-2 Bianchi C, Zhang L, Goldwasser D, Parker A, Horsey H (2020) Modeling occupancy-driven building loads for large and diversified building stocks through the use of parametric schedules. Appl Energy 276:115470 Nazarian N, Krayenhoff E, Bechtel B, Hondula D, Paolini R, Vanos J, Cheung T, Chow W, de Dear R, Jay O (2022) Integrated assessment of urban overheating impacts on human life, Earth's Future 10(8) e2022EF002682 Wang C, Song J, Shi D, Reyna JL, Horsey H, Feron S, Zhou Y, Ouyang Z, Li Y, Jackson RB (2023) Impacts of climate change, population growth, and power sector decarbonization on urban building energy use. Nat Commun 14(1):1–16 Luo X, Vahmani P, Hong T, Jones A (2020) City-scale building anthropogenic heating during heat waves. Atmosphere 11(11):1206 Tehrani AA, Sobhaninia S, Nikookar N, Levinson R, Sailor DJ, Amaripadath D (2025) Data-driven approach to estimate urban heat island impacts on building energy consumption. Energy 316:134508 Salvati A, Roura HC, Cecere C (2017) Assessing the urban heat island and its energy impact on residential buildings in Mediterranean climate: Barcelona case study. Energy Build 146:38–54 Hong T, Chen Y, Luo X, Luo N, Lee SH (2020) Ten questions on urban building energy modeling. Build Environ 168:106508 Jie P, Su M, Gao N, Ye Y, Kuang X, Chen J, Li P, Grunewald J, Xie X, Shi X (2023) Impact of urban wind environment on urban building energy: A review of mechanisms and modeling. Build Environ 245:110947 Pan X, Xu Y, Hong T (2024) Surrogate modelling for urban building energy simulation based on the bidirectional long short-term memory model. J Build Perform Simul 1–19 Han JM, Ang YQ, Malkawi A, Samuelson HW (2021) Using recurrent neural networks for localized weather prediction with combined use of public airport data and on-site measurements. Build Environ 192:107601 Li K, Chen Y (2023) Characterizing the indicator-based, day-and-night, and climate-based variations in response of surface urban heat island during heat wave across global 561 cities. Sustainable Cities Soc 99:104877 Xu L, Tong S, He W, Zhu W, Mei S, Cao K, Yuan C (2022) Better understanding on impact of microclimate information on building energy modelling performance for urban resilience. Sustainable Cities Soc 80:103775 Lipson MJ, Grimmond S, Best M, Abramowitz G, Coutts A, Tapper N, Baik JJ, Beyers M, Blunn L, Boussetta S (2024) Evaluation of 30 urban land surface models in the Urban-PLUMBER project: Phase 1 results. Q J R Meteorol Soc 150(758):126–169 Nik VM, Perera A, Chen D (2021) Towards climate resilient urban energy systems: a review. Natl Sci Rev 8(3):nwaa134 Khan HS, Paolini R, Caccetta P, Santamouris M (2022) On the combined impact of local, regional, and global climatic changes on the urban energy performance and indoor thermal comfort—The energy potential of adaptation measures. Energy Build 267:112152 Yang Y, Gu Q, Wei H, Liu H, Wang W, Wei S (2023) Transforming and validating urban microclimate data with multi-sourced microclimate datasets for building energy modelling at urban scale. Energy Build 295:113318 Du S, Zhang X, Jin X, Zhou X, Shi X (2022) A review of multi-scale modelling, assessment, and improvement methods of the urban thermal and wind environment. Build Environ 213:108860 Frayssinet L, Merlier L, Kuznik F, Hubert J-L, Milliez M, Roux J-J (2018) Modeling the heating and cooling energy demand of urban buildings at city scale. Renew Sustain Energy Rev 81:2318–2327 Lauzet N, Rodler A, Musy M, Azam M-H, Guernouti S, Mauree D, Colinart T (2019) How building energy models take the local climate into account in an urban context–A review. Renew Sustain Energy Rev 116:109390 Ding X, Zhao Y, Strebel D, Fan Y, Ge J, Carmeliet J (2024) A WRF-UCM-SOLWEIG framework for mapping thermal comfort and quantifying urban climate drivers: Advancing spatial and temporal resolutions at city scale. Sustainable Cities Soc 112:105628 Dorer V, Allegrini J, Orehounig K, Moonen P, Upadhyay G, Kampf J, Carmeliet J (2013) Modelling the urban microclimate and its impact on the energy demand of buildings and building clusters, Building Simulation IBPSA, 2013, pp. 3483–3489 Liu S, Kwok YT, Ren C (2023) Investigating the impact of urban microclimate on building thermal performance: A case study of dense urban areas in Hong Kong. Sustainable Cities Soc 94:104509 Haddad S, Zhang W, Paolini R, Gao K, Altheeb M, Al Mogirah A, Bin Moammar A, Hong T, Khan A, Cartalis C (2024) Quantifying the energy impact of heat mitigation technologies at the urban scale. Nat Cities 1(1):62–72 Adilkhanova I, Santamouris M, Yun GY (2024) Green roofs save energy in cities and fight regional climate change. Nat Cities 1(3):238–249 Ciancio V, Falasca S, Golasi I, Curci G, Coppi M, Salata F (2018) Influence of input climatic data on simulations of annual energy needs of a building: EnergyPlus and WRF modeling for a case study in Rome (Italy). Energies 11(10):2835 Takane Y, Nakajima K, Kikegawa Y, Yamaguchi K Enhancing urban canopy building energy models through the integration of social big data: Improvement and application, International Association for Urban Climate (IAUC) Urban Climate News 89 17–21 Xu H, Zhao Y, Dajun Z, Duan Y, Xu X (2025) Exploring the typhoon intensity forecasting through integrating AI weather forecasting with regional numerical weather model. npj Clim Atmospheric Sci 8(1):38 Erell E, Zhou B (2022) The effect of increasing surface cover vegetation on urban microclimate and energy demand for building heating and cooling. Build Environ 213:108867 Krayenhoff ES, Jiang T, Christen A, Martilli A, Oke TR, Bailey BN, Nazarian N, Voogt JA, Giometto MG, Stastny A (2020) A multi-layer urban canopy meteorological model with trees (BEP-Tree): Street tree impacts on pedestrian-level climate. Urban Clim 32:100590 Du R, Liu C-H, Li X-X (2024) A new method for detecting urban morphology effects on urban-scale air temperature and building energy consumption under mesoscale meteorological conditions. Urban Clim 53:101775 Bueno B, Norford L, Hidalgo J, Pigeon G (2013) The urban weather generator. J Build Perform Simul 6(4):269–281 Cui P, Lu J, Wu Y, Tang J, Jiang J (2024) Effect of urban morphology on microclimate and building cluster energy consumption in cold regions of China. Sustainable Cities Soc 115:105838 Li R, Huang C, Xin W, Ye J, Zhang X, Qu R, Wang J, Yuan L, Yao J (2025) Data-driven optimization reveals the impact of Urban Heat Island effect on the retrofit potential of building envelopes. Build Environ 269:112367 Xu S, Wang S, Li G, Zhou H, Meng C, Qin Y, He B-J (2024) Performance-based design of residential blocks for the co-benefits of building energy efficiency and outdoor thermal comfort improvement. Build Environ 264:111926 Bian C, Cheung KL, Chen X, Lee CC (2025) Integrating microclimate modelling with building energy simulation and solar photovoltaic potential estimation: The parametric analysis and optimization of urban design. Appl Energy 380:125062 Redon EC, Lemonsu A, Masson V, Morille B, Musy M (2017) Implementation of street trees within the solar radiative exchange parameterization of TEB in SURFEX v8. 0. Geosci Model Dev 10(1):385–411 Oleson K, Feddema J (2020) Parameterization and surface data improvements and new capabilities for the Community Land Model Urban (CLMU), Journal of Advances in Modeling Earth Systems 12(2) e2018MS001586 Karimi A, Moreno-Rangel D, García-Martínez A (2025) Granular Mapping of UHI and Heatwave Effects: Implications for Building Performance and Urban Resilience, Building and Environment 112705 Deng X, Cao Q, Wang L, Wang W, Wang S, Wang S, Wang L (2023) Characterizing urban densification and quantifying its effects on urban thermal environments and human thermal comfort. Landsc Urban Plann 237:104803 Bueno B, Pigeon G, Norford L, Zibouche K, Marchadier C (2012) Development and evaluation of a building energy model integrated in the TEB scheme. Geosci Model Dev 5(2):433–448 Simpson CH, Brousse O, Taylor T, Grellier J, Taylor J, Fleming LE, Davies M, Heaviside C (2024) Modeled temperature, mortality impact and external benefits of cool roofs and rooftop photovoltaics in London. Nat Cities 1(11):751–759 Tan H, Kotamarthi R, Wang J, Qian Y, Chakraborty T (2023) Impact of different roofing mitigation strategies on near-surface temperature and energy consumption over the Chicago metropolitan area during a heatwave event. Sci Total Environ 860:160508 Tarasova M, Varentsov M, Stepanenko V (2023) Parameterization of the Interaction between the Atmosphere and the Urban Surface: Current State and Prospects. Izv Atmospheric Ocean Phys 59(2):111–130 Xu X, Chen F, Shen S, Miao S, Barlage M, Guo W, Mahalov A (2018) Using WRF-urban to assess summertime air conditioning electric loads and their impacts on urban weather in Beijing. J Geophys Research: Atmos 123(5):2475–2490 Pappaccogli G, Giovannini L, Zardi D, Martilli A (2020) Sensitivity analysis of urban microclimatic conditions and building energy consumption on urban parameters by means of idealized numerical simulations. Urban Clim 34:100677 Perera A, Coccolo S, Scartezzini J-L, Mauree D (2018) Quantifying the impact of urban climate by extending the boundaries of urban energy system modeling. Appl Energy 222:847–860 Li XC, Zhao L, Qin Y, Oleson K, Zhang Y (2024) Elevated urban energy risks due to climate-driven biophysical feedbacks. Nat Clim Change 14(10):1056–1063 Liu J, Gao H, Jia R, Wang R, Han D, Liu L, Xu X, Qiao Z (2025) A downscaling framework with WRF-UCM and LES/RANS models for urban microclimate simulation strategy: Validation through both measurement and mechanism model. Build Environ 269:112361 Mei S-J, Yuan C (2021) Analytical and numerical study on transient urban street air warming induced by anthropogenic heat emission. Energy Build 231:110613 Zhang M, Gao Z (2021) Effect of urban form on microclimate and energy loads: Case study of generic residential district prototypes in Nanjing, China. Sustainable Cities Soc 70:102930 Mangan SD, Oral GK, Sozen I, Kocagil IE (2020) Evaluation of settlement textures in terms of building energy, economic performance, and outdoor thermal comfort. Sustainable Cities Soc 56:102110 Tian X, Zhang H, Liu L, Huang J, Liu L, Liu J (2024) Establishment of LCZ-based urban building energy consumption dataset in hot and humid subtropical regions through a bottom-up method. Appl Energy 368:123491 Xi C, Wang D, Cao S-J (2023) Impacts of trees-grass area ratio on thermal environment, energy saving, and carbon benefits. Urban Clim 47:101393 Allegrini J, Dorer V, Carmeliet J (2012) Influence of the urban microclimate in street canyons on the energy demand for space cooling and heating of buildings. Energy Build 55:823–832 Miller C, Thomas D, Kämpf J, Schlueter A (2018) Urban and building multiscale co-simulation: case study implementations on two university campuses. J Build Perform Simul 11(3):309–321 Bouyer J, Inard C, Musy M (2011) Microclimatic coupling as a solution to improve building energy simulation in an urban context. Energy Build 43(7):1549–1559 Gros A, Bozonnet E, Inard C, Musy M (2016) Simulation tools to assess microclimate and building energy–A case study on the design of a new district. Energy Build 114:112–122 Liu L, Du Z, Wang S, Liu J, Hang J, Yang W (2025) Modeling the impact of heat rejection from split air conditioner on outdoor air temperature in high-density residential areas. Build Environ 112685 Yang X, Yao L, Peng LL (2024) Impacts of urban air temperature and humidity on building cooling and heating energy demand in 15 cities of eastern China. Energy 288:129887 Yang X, Peng LL, Jiang Z, Chen Y, Yao L, He Y, Xu T (2020) Impact of urban heat island on energy demand in buildings: Local climate zones in Nanjing. Appl Energy 260:114279 Tian L, Hao T, He X, Chan I, Niu J, Chan P, Ng W, Huang J (2024) Examining the non-linear relationship between urban form and air temperature at street level: A case of Hong Kong. Build Environ 264:111884 Xiang Y, Zheng B, Bedra KB, Ouyang Q, Liu J, Zheng J (2023) Spatial and seasonal differences between near surface air temperature and land surface temperature for Urban Heat Island effect assessment. Urban Clim 52:101745 Venter ZS, Brousse O, Esau I, Meier F (2020) Hyperlocal mapping of urban air temperature using remote sensing and crowdsourced weather data. Remote Sens Environ 242:111791 Yin H, Zhao X (2024) Urban heat island analysis based on high resolution measurement data: A case study in Beijing. Sustainable Cities Soc 106:105389 Han J, Chong A, Lim J, Ramasamy S, Wong NH, Biljecki F (2024) Microclimate spatio-temporal prediction using deep learning and land use data. Build Environ 253:111358 Chen G, Hua J, Shi Y, Ren C (2023) Constructing air temperature and relative humidity-based hourly thermal comfort dataset for a high-density city using machine learning. Urban Clim 47:101400 Ding X, Zhao Y, Fan Y, Li Y, Ge J (2023) Machine learning-assisted mapping of city-scale air temperature: Using sparse meteorological data for urban climate modeling and adaptation. Build Environ 234:110211 Benali A, Carvalho A, Nunes J, Carvalhais N, Santos A (2012) Estimating air surface temperature in Portugal using MODIS LST data. Remote Sens Environ 124:108–121 Colaninno N, Morello E (2022) Towards an operational model for estimating day and night instantaneous near-surface air temperature for urban heat island studies: outline and assessment. Urban Clim 46:101320 Chen F, Li X, Wang Y (2025) A knowledge-augmented deep fusion method for estimating near-surface air temperature. Remote Sens Environ 326:114819 Adrian CZM, Hien WN, Marcel I, Kardinal JS (2013) Predicting the envelope performance of commercial office buildings in Singapore. Energy Build 66:66–76 Ignatius M, Wong NH, Jusuf SK (2016) The significance of using local predicted temperature for cooling load simulation in the tropics. Energy Build 118:57–69 Hsiao Y-C (2024) Hourly temperature downscaling method based on clustering and linear transformation: Utilizing mean, maximum, and minimum temperatures. Energy Build 325:114975 Ye J, Huang C, Zhong Z, Shen Y, Ao X, Su Y, Cao J, Duan H, Yao J (2025) Improving the accuracy of microclimate coupled urban building energy modeling using convolutional neural networks. Build Environ 277:112923 Sampaio RJ, Rodriguez DA, Espíndola RP, da Silva FP (2025) Spatiotemporal Assessment and Monitoring of Urban Heat Islands in Metropolitan Areas Using Machine Learning and Downscaling. Sustainable Cities Soc 106365 Li W, Zhou Z, Zhang S, Feng J (2022) Study on relationship between shading and outdoor air temperature based on the comparison of two high-rise residential estates with field measurements. Buildings 12(11):1813 Lindberg F, Sun T, Grimmond S, Tang Y (2023) UMEP Manual Documentation Bueno B, Nakano A, Norford L Urban weather generator: a method to predict neighborhood-specific urban temperatures for use in building energy simulations, ICUC9-9th International Conference on Urban Climate Jointly with 12th Symposium on the Urban Environment, 2015 Liu Y, Chu C, Zhang R, Chen S, Xu C, Zhao D, Meng C, Ju M, Cao Z (2024) Impacts of high-albedo urban surfaces on outdoor thermal environment across morphological contexts: A case of Tianjin, China. Sustainable Cities Soc 100:105038 Housing Mo (2021) U.-R.D.o.t.P.s.R.o. China, General Code for Energy Efficiency and Renewable Energy Application in Buildings (GB 55015 – 2021). China Architecture and Building Press Beijing, China ASHRAE A (2020) Standard 169–2020. Climatic Data for Building Standards, Atlanta, GA Housing Mo (2016) U.-R.D.o.t.P.s.R.o. China, Standard for energy consumption of building (GB 51161 – 2016) Housing Mo (2014) U.-R.D.o.t.P.s.R.o. China, Standard for weather data of building energy efficiency Wang S, Yan S, Zhang D, Wan S, Lv H, Wang L (2024) A real-time indoor temperature and relative humidity prediction method for exhibition hall based on Long Short-Term Memory network. J Building Eng 98:111492 Hochreiter S, Schmidhuber J (1997) Long short-term memory. Neural Comput 9(8):1735–1780 Lundberg SM, Lee S-I (2017) A unified approach to interpreting model predictions. Adv Neural Inf Process Syst 30 Yang Y, Lv Y, Zhou D (2025) The impact of urban parks on the thermal environment of built-up areas and an optimization method. PLoS ONE 20(3):e0318633 Thierry J, Herpin S, Levi R, Canonne D, Demotes-Mainard S, Cannavo P, Lemesle D, Brialix L, Rodriguez F, Bournet P (2024) Impact of a water restriction on the summer climatic benefits of trees inside an outdoor street canyon scale model. Build Environ 261:111722 Sosa MB, Correa EN, Cantón MA (2017) Urban grid forms as a strategy for reducing heat island effects in arid cities. Sustainable Cities Soc 32:547–556 Wang C, Wang Z-H, Wang C, Myint SW (2019) Environmental cooling provided by urban trees under extreme heat and cold waves in US cities. Remote Sens Environ 227:28–43 Wang S, Zhan W, Zhou B, Tong S, Chakraborty T, Wang Z, Huang K, Du H, Middel A, Li J (2025) Dual impact of global urban overheating on mortality. Nat Clim Change 1–8 Barbour E, Davila CC, Gupta S, Reinhart C, Kaur J, González MC (2019) Planning for sustainable cities by estimating building occupancy with mobile phones. Nat Commun 10(1):3736 Mirsadeghi M, Costola D, Blocken B, Hensen JL (2013) Review of external convective heat transfer coefficient models in building energy simulation programs: Implementation and uncertainty. Appl Therm Eng 56(1–2):134–151 Ma K, Yu Y, Yang B, Yang J (2019) Demand-side energy management considering price oscillations for residential building heating and ventilation systems. IEEE Trans Industr Inf 15(8):4742–4752 Chen Y, Sha A, Jiang W, Lu Q, Du P, Hu K, Li C (2025) Eco-friendly bismuth vanadate/iron oxide yellow composite heat-reflective coating for sustainable pavement: Urban heat island mitigation. Constr Build Mater 470:140645 Huang X, Bou-Zeid E, Pigliautile I, Pisello AL, Mandal J (2024) Optimizing retro-reflective surfaces to untrap radiation and cool cities. Nat Cities 1(4):275–285 Li Y, Li H, Miao R, Qi H, Zhang Y (2023) Energy–Environment–Economy (3E) Analysis of the Performance of Introducing Photovoltaic and Energy Storage Systems into Residential Buildings: A Case Study in Shenzhen, China. Sustainability 15(11):9007 Rahman MN (2024) Seasonal and annual trends in reference evapotranspiration and prediction using machine learning models across seven climatic zones of Bangladesh, Geology, Ecology, and Landscapes 1–16 Meili N, Zheng X, Takane Y, Nakajima K, Yamaguchi K, Chi D, Zhu Y, Wang J, Qiu Y, Paschalis A (2025) Modeling the effect of trees on energy demand for indoor cooling and dehumidification across cities and climates. J Adv Model Earth Syst 17(3) e2024MS004590. Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7763869","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":523642127,"identity":"cfae1ee6-5c04-4b7f-8405-189fdedf3ff3","order_by":0,"name":"Shuyang Zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABVUlEQVRIie2RT0vDMBTAUwrZJWPXlDr2FQKFqijbV1nYYZdOBoJ4jAyyS9FrxaFfYV52NfLAXfYBNrbDhtDTDoI6ivRgVv/AagceBfsjhLz38iPJC0I5OX8RmswKlUxTIcT2PsP1+K5uUayurGtFh/i3ChuN2EeIvw7aolSuOuHiJZ5xMfZe+1Gb1lClu4D224xfCPN+SlD1KKUYvYddpyxDfha0BhOfUS4wYXB5HvJA4cYBQY3jlGLSumtbAniHtgZjwmgdYYKg6AMXirg2QYqLTQXT5sqmMXBJvXASM30xXJgnyo0qrbIUQj3XesLAfTLCU32KITBiQCLgfUVwlkKpd2IbEpygIN3pjn6LxB6DogDnFrCz32ONtFIJmgMriqF8B2Y4WcaHtZI5fHwmOnM97CzGy9NqWklaQDZelzRSrvPrFcv4GV2PfubizJ05OTk5/5N3cLV7CtpgCr4AAAAASUVORK5CYII=","orcid":"","institution":"Tsinghua University","correspondingAuthor":true,"prefix":"","firstName":"Shuyang","middleName":"","lastName":"Zhang","suffix":""},{"id":523642128,"identity":"c066c965-1b43-47cc-bb97-a72ae342029d","order_by":1,"name":"Nianxiong Liu","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Nianxiong","middleName":"","lastName":"Liu","suffix":""},{"id":523642129,"identity":"4d1d9746-4525-4a1f-a635-7e889b36b089","order_by":2,"name":"Xiyu Wu","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Xiyu","middleName":"","lastName":"Wu","suffix":""},{"id":523642130,"identity":"e03beaa0-0d01-4037-8377-8fd2f9f8e474","order_by":3,"name":"Yichen Han","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Yichen","middleName":"","lastName":"Han","suffix":""},{"id":523642131,"identity":"525e57a7-8d27-4ed0-9a68-98891bf92f88","order_by":4,"name":"Wenwen Li","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Wenwen","middleName":"","lastName":"Li","suffix":""},{"id":523642132,"identity":"83ed4cab-d766-47db-ae6e-3767cbcb7426","order_by":5,"name":"Fang Guo","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Fang","middleName":"","lastName":"Guo","suffix":""},{"id":523642133,"identity":"08c92c46-743d-4f5b-b09b-b74e4d110706","order_by":6,"name":"Xiujiao Song","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Xiujiao","middleName":"","lastName":"Song","suffix":""},{"id":523642134,"identity":"bf7d2435-7300-497f-b8cb-ce8e2830bfc6","order_by":7,"name":"Chao Yuan","email":"","orcid":"","institution":"National University of Singapore","correspondingAuthor":false,"prefix":"","firstName":"Chao","middleName":"","lastName":"Yuan","suffix":""}],"badges":[],"createdAt":"2025-10-02 05:27:50","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7763869/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7763869/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":92694442,"identity":"27be022d-356f-43bf-8bf7-2c51f7d41ddc","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":28601643,"visible":true,"origin":"","legend":"","description":"","filename":"00Manuscript0924MarkedRS.docx","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/28c3f1377fcd3507d981379f.docx"},{"id":92694432,"identity":"b0069edb-ef4f-4019-b421-d8127af0c476","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs7763869.json","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/0bbde467c4b40256318e876e.json"},{"id":92694436,"identity":"c92c9c6b-eaee-4896-ab9d-66507b0bebae","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":283223,"visible":true,"origin":"","legend":"","description":"","filename":"rs77638690enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/db5e6714f1345a9a53063b7b.xml"},{"id":92695212,"identity":"0c6919e9-ffc3-46c0-996c-7cdf982b9aa0","added_by":"auto","created_at":"2025-10-03 06:41:50","extension":"jpeg","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1964699,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/36ed7bfd19cd0001ceff2db1.jpeg"},{"id":92694451,"identity":"7fc6ba9f-cdb2-45b7-b930-a84138c57fab","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4780148,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/ae391a3125f8e5b301306706.jpeg"},{"id":92694456,"identity":"8e497834-941a-4dc4-bf1b-7ce9abec0efc","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":734219,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/1b735bb0e4162316ade6a06a.png"},{"id":92694463,"identity":"117ddb2d-d589-4466-aa0e-77319a6a13cc","added_by":"auto","created_at":"2025-10-03 06:33:51","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":151350,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/330552ae51584f20941e9a5f.png"},{"id":92695219,"identity":"185bf24f-8daa-4a48-9900-f3715e22be21","added_by":"auto","created_at":"2025-10-03 06:41:50","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":592801,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/85b99615d33f60d4393da456.png"},{"id":92694453,"identity":"ba4c38e9-84a9-4ab6-84c0-612358f97c50","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"jpeg","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1998292,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/969ec4b3f69af3b7ebc4e0c0.jpeg"},{"id":92694437,"identity":"e8fe42b7-c24a-4da9-83fe-33ce72197568","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":325612,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/4507770d425c10026d4770a3.png"},{"id":92695222,"identity":"c9fda949-e730-46c2-9fb7-8818a70edc97","added_by":"auto","created_at":"2025-10-03 06:41:51","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6720383,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/c163a5b2117d2fd827767e30.jpeg"},{"id":92695221,"identity":"eeb9d3bd-08b3-450f-aa36-5960deb1b737","added_by":"auto","created_at":"2025-10-03 06:41:51","extension":"jpeg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6758507,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/ec171b1ad971a4d9d5e8a7e8.jpeg"},{"id":92694444,"identity":"553a53ad-ce17-472f-a134-a0895c9e35d0","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":492423,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/9437edd2af2785de8009f2e7.png"},{"id":92694459,"identity":"f83a6340-3ea5-4a68-9ed6-707c83fc22d4","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"jpeg","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3905401,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/bec35238e4edff0c119b3d29.jpeg"},{"id":92695214,"identity":"7517fd29-1fe0-405c-9a52-b4f7472790d8","added_by":"auto","created_at":"2025-10-03 06:41:50","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1303642,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/8914c570fda33ead078f38ff.png"},{"id":92695215,"identity":"b893e070-e3fe-4d35-a08c-96f6a10f6d7f","added_by":"auto","created_at":"2025-10-03 06:41:50","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":848203,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/fcd0dc2612d7f64aab23e64b.png"},{"id":92695220,"identity":"1a386010-e51f-4a9c-a287-5f58a12ef5a6","added_by":"auto","created_at":"2025-10-03 06:41:50","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":148035,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/a7c65ad49d642b80fda1d886.png"},{"id":92694445,"identity":"20f336dd-6bbb-4b52-ab6e-93593ee0bb90","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":54040,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/107f1c8df2b1e96f00881abd.png"},{"id":92694441,"identity":"87cf6d3a-f57e-497e-a419-84b12cd2d23e","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":104586,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/5d5403b5408f4bdd141411d0.png"},{"id":92694448,"identity":"3e95a847-4bd7-476d-bb24-92f1fd4a7862","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":872959,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/43e63560a1d427d37ec6ed2b.png"},{"id":92694449,"identity":"10303e99-4c6d-40c8-8adb-a35fb0d6f75c","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":65135,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/f0cf8a28fbc2d99ed74b7df1.png"},{"id":92695216,"identity":"ec5c4a69-afbe-4f48-8da8-7d774ced495f","added_by":"auto","created_at":"2025-10-03 06:41:50","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1211152,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/110887c9740eb29c942e7a83.png"},{"id":92694466,"identity":"0a67a0c6-c9f1-4d55-a545-16a0ac389be3","added_by":"auto","created_at":"2025-10-03 06:33:51","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1158925,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/faa4faee38af5fcd01facaf2.png"},{"id":92694457,"identity":"87f32fb1-f79b-4315-a35e-39459563f46e","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":89557,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/de7503cd77b3a594452ef382.png"},{"id":92694464,"identity":"162b8551-e583-4143-8ece-9f3f1edfe343","added_by":"auto","created_at":"2025-10-03 06:33:51","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":620558,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/2059bafb669fb42ed689db59.png"},{"id":92694461,"identity":"ec8351ea-1408-4d9b-a72e-029048ea0edf","added_by":"auto","created_at":"2025-10-03 06:33:51","extension":"xml","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":282338,"visible":true,"origin":"","legend":"","description":"","filename":"rs77638690structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/b8484789e947b49d70502298.xml"},{"id":92694460,"identity":"902f5655-6332-44d7-88e6-b3848ce3d785","added_by":"auto","created_at":"2025-10-03 06:33:51","extension":"html","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":296388,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/a480fe50353686c0aecb9c2f.html"},{"id":92694430,"identity":"674f7450-08c4-4d3f-8855-9bc931ac381e","added_by":"auto","created_at":"2025-10-03 06:33:49","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1964699,"visible":true,"origin":"","legend":"\u003cp\u003eStudy area, target building and validation weather station.\u003c/p\u003e","description":"","filename":"image1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/8f0007352908e089df171391.jpeg"},{"id":92694431,"identity":"bd042638-bae9-4c99-bb5d-daa74a4a68fe","added_by":"auto","created_at":"2025-10-03 06:33:49","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":799533,"visible":true,"origin":"","legend":"\u003cp\u003eThe LWD-based UBEM simulation platform includes three core modules: the UCM model, building geometry model, and UBEM modeling—corresponding to localized weather data processing, 3D building modeling, and energy consumption simulation. ① Input local environmental data into UWG; ② feed the simulated LWD into UBEM; ③process 3D building geometry with attributes in Rhino; ④ convert the geometry into UBEM; ⑤simulate hourly heating and cooling loads per unit area for the target building over a full year.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/a571cf27c34793df8b326533.png"},{"id":92694433,"identity":"69ba796f-4b15-4e22-b639-5b3b22de8e1e","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":756850,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of Ta_HD and Ta_UWG during 2023 extreme weather events and the same periods in 2021. (i) 24-hour average temperature profiles during the heatwave and cold wave (lines), with ±1 standard deviation (shaded area); (ii) Multi-day temperature trends during extreme events; (iii) Scatter plots of temperature differences (ΔTa, or UHI intensity), where ΔTa is the difference between Ta_UWG and Ta_HD.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/be9571e9fd82875c0fad2055.png"},{"id":92694434,"identity":"7bc76e38-c787-40db-86c6-928ae3056544","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1998292,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of Ta_HD, Ta_UWG, RH_HD, and RH_UWG against[ZS1] measured values for winter and summer 2021. (i) 24-hour hourly average temperature (lines) with ±1 standard deviation (shaded area); (ii) Multi-day continuous comparison; (iii) Scatter plot comparison.\u003c/p\u003e\n\u003cp\u003e[ZS1]2\u003c/p\u003e","description":"","filename":"image4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/928c366adfde24fafac333ce.jpeg"},{"id":92694439,"identity":"441f261e-e085-4afe-813e-d99d97fb98e9","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":547865,"visible":true,"origin":"","legend":"\u003cp\u003e(a) Comparison of heating and cooling EUI simulated using UWG, HD, and TMY weather data vs. measured values for different years (percentage on bars indicates deviation from actual); (b) Comparison of 0.4% design-day cooling loads using UWG and HD data; (c) Comparison of 99.6% design-day heating loads using UWG and HD data. UWG, HD, and TMY represent simulation results using LWD, suburban station data, and TMY, respectively. 2021 reflects a typical year; 2023 represents extreme weather conditions.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/a6adeecc1cbdc888c817941d.png"},{"id":92694462,"identity":"b178b1c9-15cb-4fde-89ae-7bbb429891f5","added_by":"auto","created_at":"2025-10-03 06:33:51","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":6720383,"visible":true,"origin":"","legend":"\u003cp\u003e(a–f)[ZS1] Comparison of hourly cooling loads simulated using UWG, HD, and TMY data with measured loads for summer 2021 and 2023. (i) Diurnal variation of cooling load: 24-hour hourly average cooling load between measured and simulated data (lines) ±1 standard deviation (shaded area); (ii) Multi-day variation of cooling load: continuous load comparison during the heatwave; (iii) Scatter plots of simulated compared with measured loads. (g) Comparison of cooling loads during the 2023 heatwave with the same period in 2021.\u003c/p\u003e\n\u003cp\u003e[ZS1]4\u003c/p\u003e","description":"","filename":"image6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/d3428f5bf671deff33c49072.jpeg"},{"id":92694446,"identity":"2f95e2ab-6865-4b26-8311-823c48621060","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":6758507,"visible":true,"origin":"","legend":"\u003cp\u003e(a–f) Comparison of hourly heating loads simulated using UWG, HD, and TMY data with measured loads for winter 2021 and 2023. (i) Diurnal variation of heating load: 24-hour hourly average heating load between measured and simulated data (lines) ±1 standard deviation (shaded area); (ii) Multi-day variation of cooling load: continuous load comparison during the cold wave; (iii) Scatter plots of simulated compared with measured loads. (g) Comparison of heating loads during the 2023 cold wave with the same period in 2021.\u003c/p\u003e","description":"","filename":"image7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/eaa6e387ae8faf868eeee131.jpeg"},{"id":92695218,"identity":"640ff975-e399-48d8-b0ac-e78f9fb9e309","added_by":"auto","created_at":"2025-10-03 06:41:50","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":609405,"visible":true,"origin":"","legend":"\u003cp\u003eOptimizing winter LWD-simulated heating load using LSTM prediction model. (i) Comparison of predicted and actual values during 24-hour diurnal variation; (ii) Comparison of predicted and actual values over multiple days; (iii) Scatter plot of predicted and actual values; (iv) Model training loss and R².\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/10087e89da8f1f377266a046.png"},{"id":92695217,"identity":"bc0b3419-6566-46be-9cb3-86494dfee453","added_by":"auto","created_at":"2025-10-03 06:41:50","extension":"jpeg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":3905401,"visible":true,"origin":"","legend":"\u003cp\u003eTime-dependent influence of climate factors and internal disturbances on peak energy load in summer during typical and heatwave conditions. (i) Diurnal variation of factor impacts; (ii) Daily variation of factor impacts.\u003c/p\u003e","description":"","filename":"image9.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/996d93bcc8e5afce7932006c.jpeg"},{"id":92694454,"identity":"00154c21-e435-4cb8-9831-9567fc0eec40","added_by":"auto","created_at":"2025-10-03 06:33:50","extension":"jpeg","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":4780148,"visible":true,"origin":"","legend":"\u003cp\u003eTime-dependent influence of climate factors and internal disturbances on peak energy load in winter during typical and cold wave conditions. (i) Diurnal variation of factor impacts; (ii) Daily variation of factor impacts.\u003c/p\u003e","description":"","filename":"image10.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/44a3f4282c6d9e2732944f97.jpeg"},{"id":92695213,"identity":"2bf848b9-a055-4f26-b3df-985dbccfdbd0","added_by":"auto","created_at":"2025-10-03 06:41:50","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":950539,"visible":true,"origin":"","legend":"\u003cp\u003eImpact of internal and external factors on actual heating and cooling loads in extreme and typical years: (i) Overall importance of each factor; (ii) Daily variation in factor importance; (iii) 24-hour diurnal variation in factor importance. \u003cem\u003eNote: In addition to microclimate variables, \"Local time\" represents potential internal influences.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"image11.png","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/07664f94f282900184491427.png"},{"id":92695489,"identity":"7f723340-9c53-47c1-a62d-8ca8e349a875","added_by":"auto","created_at":"2025-10-03 06:50:01","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":31462692,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7763869/v1/7852818c-4868-43a8-9318-e428ee5df485.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eAn empirical study on the compound effects of extreme weather and UHIon building energy consumption under local climate\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eIn recent years, worsening global climate has led to more frequent extreme weather (EW) events [\u003cspan additionalcitationids=\"CR2 CR3\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. In June 2023, Beijing faced a record-breaking heatwave with temperatures over 40\u0026deg;C for three days, with a peak of 42\u0026deg;C. That December, an eight-day cold snap brought daily lows below \u0026minus;\u0026thinsp;12\u0026deg;C [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. In the U.S., a February 2021 cold wave hit Texas, causing ice-covered power lines and blackouts [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. In summer 2023, a heatwave triggered massive air conditioning (AC) loads, again leading to widespread outages and rising electricity prices. The extreme weather events raised serious concerns about the resilience of buildings and energy systems under severe climate conditions. Beyond EW, urban heat island (UHI)\u0026mdash;driven by heat-retaining surfaces and anthropogenic heat (AH) emissions from AC\u0026mdash;leads to higher nighttime air temperatures in dense city center compared to suburban regions. Research indicates that UHI can increase cooling energy use by 10\u0026ndash;120%, and reduce heating demand by 3\u0026ndash;45% [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Ignoring UHI effects can lead to overestimating heating demand in winter and underestimating cooling demand in summer, which leads to misestimation of operational carbon assessments [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe compound effect of EW and the UHI has significantly altered urban wind speed and temperature patterns. In summer, due to the UHI effects, dense urban forms further amplify the impact of extreme heat on cities [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. A study in Seoul, South Korea, found that during the 2012 and 2016 heatwaves, peak UHI intensity was 3.3\u0026deg;C and 4.5\u0026deg;C higher than during non-heatwave periods [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], significantly increasing peak cooling loads and annual energy demand [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. In San Francisco, USA, research showed outdoor air temperature differences of up to 11\u0026deg;C between coastal and downtown areas during heatwaves, with heating energy use varying by over 100%, cooling by 65% across sites, and peak cooling demand differed by as much as 30% [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. A study in Tianjin, China, found that the compound effect of heatwaves and UHI increased nighttime residential building loads, with peaks occurring between 2:00 and 4:00 a.m. For each 1\u0026deg;C increase in UHI intensity, hourly cooling loads increased by 0.5 W/m\u0026sup2; [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. With ongoing urbanization, the climate risks arising from the interaction between heatwaves and UHI are expected to intensify [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Some studies also highlight UHI\u0026rsquo;s potential benefits in cold weather. A study in Tokyo showed that, when both cooling and heating energy use were considered, UHI increased overall energy use in commercial buildings but reduced it in residential ones, resulting in lower total urban energy demand [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. In London, UHI raised annual urban cooling loads by 25% while reducing heating loads by 22% compared to rural areas [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eBuilding energy use is driven by external and internal factors. External factors include meteorological conditions (e.g., air temperature (Ta), solar radiation (Rad), relative humidity (RH), wind speed (WS)) and the surrounding urban context [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Meteorological conditions include both urban background climate and localized weather. Nearby buildings can obstruct solar radiation and wind flow, altering the target building\u0026rsquo;s local thermal conditions [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. In addition, dense urban forms often intensify the local UHI effect, raising AC loads. AC systems, in turn, emit AH into the surrounding area, further reinforcing local warming [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. These factors jointly affect the building\u0026rsquo;s thermal environment and significantly influence its energy use [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Internal factors involve occupancy, space use, equipment, and lighting. Differences in occupancy patterns and equipment use lead to varied energy use across building types. Residential and office buildings typically show cyclical patterns due to regular schedules [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Weighted linear regression of the ASHRAE Global Thermal Comfort Database II shows that for every 1\u0026deg;C rise in outdoor temperature, indoor temperatures increase by 0.1\u0026deg;C in AC buildings and 0.4\u0026deg;C in naturally ventilated ones [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Studies indicate that a 1\u0026deg;C increase in outdoor temperature leads to an approximate 14% increase in cooling energy use intensity (EUI)\u0026mdash;defined as the annual cooling load per unit floor area of a building\u0026mdash;and about a 10% decrease in heating EUI [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Especially during heatwaves, AH emissions may rise by 20% due to higher cooling demand, with AC accounting for more than 85% [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Therefore, beyond background meteorological conditions, local climates, shaped by urban form and human activity [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], significantly affect building energy demand.\u003c/p\u003e\u003cp\u003eThe Urban Building Energy Model (UBEM) is a tool for modeling and simulating the performance of multiple buildings within a city, supporting urban-scale building energy assessments, policy development, AH estimation, and urban planning and redevelopment [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Unlike the dynamic energy use simulation of individual buildings, UBEM accounts for inter-building interactions (e.g., shading, radiation effects) and the coupling between building energy use and the urban microclimate (e.g., AH emissions) [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Due to the heterogeneity of the urban local environment, local climates vary significantly across different buildings, directly impacting building load estimates in UBEM [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. To address this, localized weather data (LWD) should be integrated into UBEM to enable local climate-energy coupling [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Incorporating local factors such as solar radiation, temperature, wind, and humidity\u0026mdash;particularly the compound effects of EW and UHI conditions [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u0026mdash;enables more accurate predictions and better understanding of how local climates affect building energy use [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. This approach supports climate-responsive urban design, including building form, envelopes, and external spatial configurations [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn recent years, Urban Canopy Models (UCM) have become key tools for modeling energy balance and surface temperatures across urban surfaces, such as roads, roofs, and walls [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. Driven by boundary meteorological data, UCMs can simulate long-term local urban weather conditions. Integrating UCM with UBEM enables more accurate representation of local climates and building energy use. Compared to geometry-intensive CFD models like Envi-met, UCM offers faster, parameterized LWD simulations over wider urban areas and longer timeframes (e.g., full-year, 8760-hour weather data), making it more suitable for large-scale UBEM applications.\u003c/p\u003e\u003cp\u003eCurrent studies on LWD and building energy simulations mostly rely on Typical Meteorological Year (TMY) data. TMY is synthesized from long-term weather records to represent average climate conditions, excluding EW and spatial variability [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. As a result, it cannot capture extreme events or local climates\u0026mdash;such as heat islands, wind fields, longwave radiation between buildings, or local Ta and RH [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Thus, while TMY-based simulations can assess UHI impacts on energy use, they fall short in evaluating the accuracy of UCM or the effect of LWD on building energy simulation results. They also cannot reflect real energy use under extreme weather combined with UHI effects. In practice, validating simulations with measured local climate and building energy consumption data is essential, especially for large-scale seasonal UBEM calculations [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. To address this, we conduct an empirical study using real suburban weather station data, proposing a method for localized weather-driven UBEM to explore three key research questions:\u003c/p\u003e\u003cp\u003eFirst, how much does suburban-station-driven LWD improve the accuracy of simulation results compared to nearby stations or TMY data? Compared to the sole impact of UHI, how do the compound effects of extreme events like heatwaves or cold spells influence building energy use and design-day loads?\u003c/p\u003e\u003cp\u003eSecond, what are the differences and patterns between hourly simulated and actual loads under the individual and compound effects of EW and UHI conditions?\u003c/p\u003e\u003cp\u003eFinally, to what extent do internal factors and local climate variations influence hourly energy loads in summer and winter?\u003c/p\u003e\u003cp\u003eThis study utilizes real weather and hourly building energy data to compare simulation accuracy across three aspects: weather data, annual energy use, and hourly profiles. It highlights discrepancies between simulated and actual hourly loads under the compound effects of extreme weather and urban heat islands. The analysis also identifies the relative importance of microclimate factors under different climatic conditions, aiming to pinpoint the key factors for effective climate risk mitigation.\u003c/p\u003e"},{"header":"2. Literature review of localized weather data","content":"\u003cp\u003eGenerally, models for generating LWD can be categorized into two approaches: physics-based and data-driven (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Physics-based models follow general physical laws, offer broad applicability, but require complex computations. Data-driven models are faster and more accurate within their training range, but they are limited to specific urban contexts. Physics-based models vary by input type. Models using environmental parameters are computationally efficient but lack detailed spatial characteristics. Models using geometric inputs\u0026mdash;typically through Computational Fluid Dynamics (CFD)\u0026mdash;can simulate detailed urban forms and produce high-resolution outputs, though requiring higher computational costs. These physics-based models operate at different spatial scales, each suited to specific resolution and duration needs. LWD are often generated using UCM, CFD, or mesoscale meteorological models coupled with UCM, with the last approach requiring multi-scale integration [\u003cspan additionalcitationids=\"CR40\" citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. The LWD with high spatiotemporal resolution is critical for UBEM applications. The meteorological data used to drive the model\u0026mdash;often at an hourly resolution\u0026mdash;determines the simulation\u0026rsquo;s temporal resolution and duration. The driving meteorological data can come from mesoscale meteorological model outputs, global reanalysis datasets, boundary-layer data, or suburban ground station observations. In this study, we use the Urban Weather Generator (UWG), a UCM-based tool driven by suburban weather station data, to generate LWD.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eSummary of localized weather data and building energy use calculation methods across different spatial scales [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e] [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"11\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMethods\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eApproach\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eClimate tools\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eEnergy models\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eInput data type\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eSpatial resolution\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eTime resolution\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e\u003cp\u003eTime span\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c9\"\u003e\u003cp\u003eAdvantages\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c10\"\u003e\u003cp\u003eLimitations\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c11\"\u003e\u003cp\u003eReference\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMesoscale meteorological model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePhysical-based model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eWRF, Meso-NH, COSMO-CML\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eEnergyPlus (CityBES), CitySim, TRNSYS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eParameter\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eMeso-scale: \u0026gt;1km\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e1h\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003eSeveral months\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003eSuitable for large-scale applications; can run with UCM or independently\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003eHigh computational cost; limited accuracy in handling urban microclimates, such as the urban canopy effect\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e[\u003cspan additionalcitationids=\"CR43 CR44 CR45 CR46 CR47 CR48\" citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eUrban canopy model (UCM)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePhysical-based model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTEB (UWG), BEP, CAT, CIM, CLMU, UCP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eEnergyPlus (DesignBuilder, Dragonfly), SUNtool (UMI), TRNSYS, CitySim, BEP-BEM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eParameter\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLocal Scale: 0.1-1km\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e1h\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e1 year\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003eRunning fast; supporting high spatiotemporal resolution simulations; includes AH and UHI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003eRequires detailed local surface, building, and vegetation parameters input; limited spatial resolution\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan additionalcitationids=\"CR51 CR52 CR53 CR54 CR55 CR56 CR57 CR58 CR59 CR60 CR61 CR62 CR63 CR64 CR65 CR66 CR67 CR68\" citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMicroscale geometry model (CFD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePhysical-based model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eEnvi-met (RANS), ANSYS Fluent, Openfoam, SOLENE-Microclimate\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eEnergyPlus (DesignBuilder), TRNSYS, ESP-r, BuildSysPro, HTB2, EnviBatE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eGeometry\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eMicro-scale: 1m-10m\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e1h\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003eSeveral days\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003eHigh spatiotemporal resolution; ideal for microclimate analysis of airflow, ventilation, and pollutant diffusion between buildings\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003eRequires detailed geometry data input; high computational demand; limited to small areas and short periods (hourly or daily simulations)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e, \u003cspan additionalcitationids=\"CR71 CR72 CR73 CR74 CR75 CR76 CR77 CR78\" citationid=\"CR70\" class=\"CitationRef\"\u003e70\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e79\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWeather station \u0026amp; mobile observation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eData-driven model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eUrban/suburban stations, mobile sensors\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eEnergyPlus (DesignBuilder), TRNSYS, CitySim, CERMA, PREDISE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e\u0026le;\u0026thinsp;1h\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003eVaries\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003eHigh measurement accuracy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003eLimited spatial coverage; data discontinuity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan additionalcitationids=\"CR81 CR82\" citationid=\"CR80\" class=\"CitationRef\"\u003e80\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR83\" class=\"CitationRef\"\u003e83\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWeather station-based spatial interpolation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eData-driven model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMeteonorm 8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eCitySim, TRNSYS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.1-1km\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e1h\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003eVaries\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003eData with spatial continuity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003eAccuracy depends on weather station density; interpolation errors exist\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan additionalcitationids=\"CR85 CR86 CR87 CR88\" citationid=\"CR84\" class=\"CitationRef\"\u003e84\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR89\" class=\"CitationRef\"\u003e89\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRemote sensing (empirical formula)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eData-driven model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMODIS land surface temperature (LST), Landsat LST\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eUrban scale:0.1-1km\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e1 month\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003eSeveral years\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003eData with spatial continuity; useful for night temperature estimation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003eLow temporal resolution; sensitive to cloud cover; weak correlation with daytime Ta\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e[\u003cspan citationid=\"CR85\" class=\"CitationRef\"\u003e85\u003c/span\u003e, \u003cspan additionalcitationids=\"CR91\" citationid=\"CR90\" class=\"CitationRef\"\u003e90\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e92\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDesign-day temperature profile\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eData-driven model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSTEVE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eEnergyPlus, IES-VE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e1h\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003eEasy to obtain; low data cost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003eStatistical model that cannot reflect the impacts of complex urban contexts on microclimates\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e[\u003cspan additionalcitationids=\"CR94\" citationid=\"CR93\" class=\"CitationRef\"\u003e93\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR95\" class=\"CitationRef\"\u003e95\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMachine learning (ML)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eData-driven model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCNN,\u003c/p\u003e\u003cp\u003eRNN, XGBoost, LSTM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eEnergyPlus (Dragonfly)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e/\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLocal Scale: 0.1-1km\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e1h\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003eVaries\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003eFast prediction, high spatial accuracy and precision\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003eRequiring large training datasets; relying on patterns within existing data; limited accuracy in predicting extreme conditions\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR96\" class=\"CitationRef\"\u003e96\u003c/span\u003e, \u003cspan citationid=\"CR97\" class=\"CitationRef\"\u003e97\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e"},{"header":"3. Materials and method","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e3.1. Study area and data collection\u003c/h2\u003e\u003cp\u003eThis study focuses on a 27-story student dormitory in Beijing\u0026rsquo;s Haidian District (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec), located at the southwest corner of a residential complex (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb), with an average floor height of about 3 meters. LWD for 2021 (a typical year) and 2023 (an extreme year) were driven by data from a suburban weather station in Haidian (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea). On-site Ta and RH data for validation were collected in 2021 from a rooftop station in a nearby residential complex (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ed) [\u003cspan citationid=\"CR98\" class=\"CitationRef\"\u003e98\u003c/span\u003e]. Actual hourly electricity use data covering full-year periods in 2021 and 2023 were provided by the building management. Winter energy use data for both years were estimated based on records from the district\u0026rsquo;s heating plant. EUI measurements were taken from rooms occupied during both summer and winter in 2021 and 2023.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e3.2. Method\u003c/h2\u003e\u003cp\u003eThis section focuses on the methods and technical details related to local weather and building energy simulation, hourly heating energy consumption calculation, the development of the LSTM prediction model, and the SHAP explanation algorithm. Sections 3.3.1 and 3.3.2 respectively describe the simulation of localized weather data and building energy use. Section \u003cspan refid=\"Sec8\" class=\"InternalRef\"\u003e3.2.3\u003c/span\u003e explains how measured heating energy consumption is converted into equivalent \u0026ldquo;electricity consumption\u0026rdquo; to facilitate comparison. Sections \u003cspan refid=\"Sec9\" class=\"InternalRef\"\u003e3.2.4\u003c/span\u003e and \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e3.2.5\u003c/span\u003e cover the LSTM model architecture and key parameters, as well as the logic of the SHAP algorithm for interpreting time series prediction models. This enables the assessment of the relative contributions of internal and external microclimate factors on actual energy loads under different climatic conditions.\u003c/p\u003e\u003cp\u003eLocal weather and energy simulations were conducted using UWG and EnergyPlus, respectively. ① LWD was simulated using the UWG model within UMEP 4.1.1 integrated in QGIS 3.28 [\u003cspan citationid=\"CR99\" class=\"CitationRef\"\u003e99\u003c/span\u003e], ② providing input for UBEM and allowing flexible grid resolution according to the study area size. Building energy modeling was performed using Dragonfly and Honeybee modules within the Grasshopper, which support flexible 3D UBEM construction and simplify the modeling process [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. ③ Building footprint data with height attributes from QGIS were imported into Rhino 7.31, converted into 3D geometry, ④ and then assembled into UBEM using Dragonfly and Honeybee (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ee). ⑤ The UBEM was run in EnergyPlus to output hourly heating and cooling loads per unit area for all 8,760 hours, which was then compared with measured data. As heating and cooling loads are more sensitive to external conditions than lighting or plug loads, the analysis focused on the impact of LWD on these two end uses. For consistency, all EUI and hourly load results are reported in kWh/(m\u0026sup2;\u0026middot;year) and W/m\u0026sup2;, respectively.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\u003ch2\u003e3.2.1. UWG model simulation of UHI\u003c/h2\u003e\u003cp\u003eThe UWG is based on the Town Energy Balance (TEB), a single-layer UCM [\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e], and simulates long-term Ta, RH, and WS in simplified 2D urban canyons [\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e]. It offers high resolution, flexibility, and fast computation, and has been validated in cities like Singapore, Basel, Toulouse, Rome, Barcelona, Boston, and Guangzhou [\u003cspan citationid=\"CR100\" class=\"CitationRef\"\u003e100\u003c/span\u003e]. UWG can generate high-resolution LWD (tens to hundreds of meters) and allows parameter adjustments by urban area, building type, and season to improve accuracy [\u003cspan citationid=\"CR101\" class=\"CitationRef\"\u003e101\u003c/span\u003e]. Anthropogenic heat released from AC in dense urban areas, especially at night, slows the cooling rate compared to suburban areas and intensifies the nighttime UHI effect. UWG integrates a rural station model (RSM), vertical diffusion model (VDM), urban boundary layer model (UBLM), and a TEB and building energy model (TEB-BEM) module [\u003cspan citationid=\"CR101\" class=\"CitationRef\"\u003e101\u003c/span\u003e], which is coupled with EnergyPlus to account for canyon AH emissions from buildings and to capture the impact of building energy use on the local climate. By inputting suburban weather station data with urban background meteorological information into the UCM, the compound effects of EW and UHI are effectively captured. Unlike other UCMs that need top-canopy meteorological data, UWG incorporates RSM, VDM, and UBLM, and requires only suburban weather station data as inputs. Its open-source code has been integrated into platforms like QGIS (UMEP) and Rhino (Dragonfly), making it highly adaptable for integration with UBEM.\u003c/p\u003e\u003cp\u003eWe used the UWG module within the UMEP toolkit in QGIS to simulate gridded LWD. Driven by suburban weather data and 2D raster inputs, the tool enables fast calculation of year-round LWD (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) without the complex 3D modeling required for large-scale urban vegetation and buildings. UWG\u0026rsquo;s temporal flexibility also allows users to define custom simulation periods. Required inputs include: suburban weather data, gridded local weather zones with building function and construction year classifications (as defined by UWG), land cover data, and urban morphology within each grid cell (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The output localized weather data supports both EPW and TXT formats. EPW files can be used directly in EnergyPlus, while TXT files support weather data analysis. The land cover data, used as UCM parameter input in UWG, includes six categories: Paved, Building, Tree, Grass, Bare Soil, and Water. This data has a resolution of 1 meter and is derived from GF-7 and Sentinel-2 satellite imagery. Building morphology within each grid is preprocessed in UMEP, which requires a Digital Elevation Model (DEM) containing ground elevation and a Digital Surface Model (DSM) with surface object heights.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section3\"\u003e\u003ch2\u003e3.2.2. Building energy model setup\u003c/h2\u003e\u003cp\u003eBefore running EnergyPlus, the BEM must be configured with key parameters, including geometry, construction, occupancy and equipment schedules, window-to-wall ratio (WWR), and heating, ventilation, and air conditioning (HVAC) settings (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Surrounding context buildings are also defined to account for shading. To further address urban-scale scenarios, we use Dragonfly for modeling. Dragonfly can convert 3D geometry into BEM or UBEM. The target building is modeled with 3-m floor heights and divided into thermal zones by floor. We used the \u0026ldquo;Solid to DF Model\u0026rdquo; module to account for variations in roof height within the building. HVAC is set as an ideal air system with a default COP of 1. All surrounding obstructions within a 200-m radius are included. With the BEM and LWD prepared, EnergyPlus is used to simulate building energy performance under localized weather conditions.\u003c/p\u003e\u003cp\u003eThis study localizes the building\u0026rsquo;s program, construction, and WWR settings based on ASHRAE 90.1 classifications, aligned with China\u0026rsquo;s GB 55015\u0026thinsp;\u0026minus;\u0026thinsp;2021 \u0026ldquo;General Code for Energy Efficiency and Renewable Energy application in buildings\u0026rdquo;[\u003cspan citationid=\"CR102\" class=\"CitationRef\"\u003e102\u003c/span\u003e]. Detailed parameters are listed in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The program is set to \u003cem\u003eHighrise Apartment\u003c/em\u003e, and the construction follows \u003cem\u003eASHRAE 90.1 2016\u0026ndash;4A Mixed Humid-Mass\u003c/em\u003e, aligning with Beijing\u0026rsquo;s ASHRAE climate zone classification (4A Mixed Humid) [\u003cspan citationid=\"CR103\" class=\"CitationRef\"\u003e103\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eUrban building energy model parameter settings [\u003cspan citationid=\"CR102\" class=\"CitationRef\"\u003e102\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eParameter\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSetting\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eWeather data\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eEpw file\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eScenario 1: Beijing TMY_CSWD data\u003c/p\u003e\u003cp\u003e(Station number: 545110)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eScenario 2: Haidian (HD) Station data\u003c/p\u003e\u003cp\u003e(Station number: 54399)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eScenario 3: Localized weather data (UWG)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eTime\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSimulation period\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAnnual (8760h)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eResolution\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eHourly\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eGeometry\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFloor number\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e15 floors (N\u0026ndash;S), 27 floors (E\u0026ndash;W)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eThermal zone\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFloor-based zone\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGross floor area (m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e38253.65\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003eConstruction\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eWindow glass U-value (W/m\u003csup\u003e2\u003c/sup\u003e\u0026middot;K)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.8\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eWindow solar heat gain coefficient (SHGC)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExterior wall U-value (W/m\u003csup\u003e2\u003c/sup\u003e\u0026middot;K)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.45\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRoof U-value (W/m\u003csup\u003e2\u003c/sup\u003e\u0026middot;K)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e\u003cp\u003eProgram\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eHeating season indoor temperature setpoint (\u0026deg;C)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e18\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCooling season indoor temperature setpoint (\u0026deg;C)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e26\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOccupant Density\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e25 m\u003csup\u003e2\u003c/sup\u003e/person\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLighting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6 W/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEquipment\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e10 W/m\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSchedule\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eHighrise_Apartment schedule (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e-(a)-i)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"4\" rowspan=\"5\"\u003e\u003cp\u003eHVAC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMaximum heating supply Ta (\u0026deg;C)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e50\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMinimum cooling supply Ta (\u0026deg;C)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e20\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eVentilation Rate\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e30 m\u003csup\u003e3\u003c/sup\u003e/ (h\u0026middot;person)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eInfiltration\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.0003 m\u003csup\u003e3\u003c/sup\u003e/s per m\u003csup\u003e2\u003c/sup\u003e facade\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCOP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWWR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eEast / South / West / North\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.35 / 0.5 / 0.35 / 0.3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section3\"\u003e\u003ch2\u003e3.2.3. Method for calculating hourly heating demand and \"electricity demand\" in winter\u003c/h2\u003e\u003cp\u003eTo estimate hourly winter heating demand, we used hourly building energy use data from a nearby district heating plant. Based on the total serviced building area and accounting for thermal distribution losses of heating network, we calculated the building\u0026rsquo;s hourly heating energy use per unit area. As the heating plant provides continuous centralized heating to surrounding residential buildings, the resulting EUI reflects actual building energy use under local weather conditions. The calculation method is as follows:\u003c/p\u003e\u003cp\u003e(1) The hourly heating energy consumption per unit area from the heating plant can be calculated using specific heat capacity.\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:{q}_{station}=\\frac{c\\cdot\\:m\\cdot\\:{\\Delta\\:}T}{3600\u0026middot;A}=\\frac{c\\cdot\\:m\\cdot\\:({T}_{s}-{T}_{r})}{3600\u0026middot;A}\\left(1\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eHere, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{q}_{station}\\)\u003c/span\u003e\u003c/span\u003e is the hourly heating supply per unit area (kWh/m\u0026sup2;); \u003cem\u003ec\u003c/em\u003e is the specific heat capacity of water, typically 4.186 kJ/(kg\u0026middot;\u0026deg;C); \u003cem\u003em\u003c/em\u003e is the hourly mass flow rate of water in the heating network (kg/h); \u003cem\u003eΔT\u003c/em\u003e is the temperature difference between supply water temperature (Ts) and return water temperature (Tr); \u003cem\u003eA\u003c/em\u003e is the total heated floor area served by the heating plant (m\u0026sup2;).\u003c/p\u003e\u003cp\u003e(2) With \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{q}_{station}\\)\u003c/span\u003e\u003c/span\u003e known, the building\u0026rsquo;s hourly heating demand per unit area in winter, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{q}_{heating}\\)\u003c/span\u003e\u003c/span\u003e, can be estimated accordingly [\u003cspan citationid=\"CR104\" class=\"CitationRef\"\u003e104\u003c/span\u003e].\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:{q}_{heating}=\\eta\\:\\cdot\\:{q}_{station}\\cdot\\:\\left(\\frac{1}{1+\\alpha\\:}\\right)\\times\\:\\beta\\:\\:\\left(2\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eHere, \u003cem\u003eη\u003c/em\u003e is the heat loss rate of the courtyard distribution network, typically 2%\u0026ndash;10%, this study uses a correction factor of 0.98. \u003cem\u003eα\u003c/em\u003e is the overheat factor due to lack of terminal control, set at 20% for district heating systems. \u003cem\u003eβ\u003c/em\u003e is the weather correction factor, used to adjust heating demand based on actual weather conditions.\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\\:\\beta\\:=\\frac{HD{D}_{0}}{HDD}\\:\\left(3\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eHere, HDD₀ is the standard heating degree days (\u0026deg;C\u0026middot;d) based on an 18\u0026deg;C baseline, set at 2699\u0026deg;C\u0026middot;d for Beijing. HDD refers to the actual heating degree days for the year, calculated using the same baseline\u0026mdash;2535\u0026deg;C\u0026middot;d in 2021 and 2577\u0026deg;C\u0026middot;d in 2023 based on Haidian station data [\u003cspan citationid=\"CR105\" class=\"CitationRef\"\u003e105\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e(3) To enable comparison across winter, summer, and annual energy use, the actual hourly heating demand per unit area \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{q}_{heating}\\)\u003c/span\u003e\u003c/span\u003e is converted to equivalent electricity demand \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{q}_{electricity}\\)\u003c/span\u003e\u003c/span\u003e under simulated HVAC conditions.\u003cdiv id=\"Equd\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e\n$$\\:{q}_{heating}={q}_{\\text{e}\\text{l}\\text{e}\\text{c}\\text{t}\\text{r}\\text{i}\\text{c}\\text{i}\\text{t}\\text{y}}\\cdot\\:\\:{COP}_{heating}\\:\\left(4\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eHere, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{COP}_{heating}\\)\u003c/span\u003e\u003c/span\u003e is the coefficient of performance for residential HVAC systems, set to 2.6 [\u003cspan citationid=\"CR102\" class=\"CitationRef\"\u003e102\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section3\"\u003e\u003ch2\u003e3.2.4. LSTM-based building energy consumption prediction model\u003c/h2\u003e\u003cp\u003eTo analyze the impact of internal and external factors on real-time building energy loads and to predict energy use under extreme weather, this study develops a Long Short-Term Memory (LSTM) deep learning model. LSTM is a type of recurrent neural network (RNN) designed for sequence prediction tasks, such as time-series forecasting and speech recognition. LSTM is trained by inputting existing multistep sequences of variables to predict their values in other or future time periods [\u003cspan citationid=\"CR106\" class=\"CitationRef\"\u003e106\u003c/span\u003e, \u003cspan citationid=\"CR107\" class=\"CitationRef\"\u003e107\u003c/span\u003e]. The LSTM model, built using the TensorFlow framework, includes an LSTM layer with 128 units, a Dropout layer to prevent overfitting, a Flatten layer to reshape the output, and a Dense layer for the final prediction. The model uses the Adam optimizer with a default learning rate of 0.001, combining momentum and adaptive learning rate techniques to improve convergence efficiency. By applying adaptive learning rates to each parameter, the update step is adjusted by assigning smaller rates to large gradients and larger rates to small gradients. Key parameter settings are shown in the Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eKey parameters and settings of the LSTM prediction model.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLSTM model parameter\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSetting\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExplanation and function\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDropout\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePrevents overfitting by randomly deactivating neurons during training to improve generalization.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLearning Rate\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.001\u0026thinsp;\u0026lt;\u0026thinsp;default learning rate\u0026gt;\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eControls step size in weight updates; too high may cause instability, too low may slow training.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eUnits\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e128\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNumber of hidden neurons in the LSTM layer; more units improve model capacity but also raise the risk of overfitting and computation time.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTime Steps\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNumber of past time points used as training input; short sequences may miss context, while longer ones increase computational cost.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEpochs\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNumber of full training cycles; each epoch processes all training data with one weight update. Too few epochs may underfit, while too many may overfit.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBatch Size\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e256\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNumber of samples per training batch; small batches train faster but are less stable, while large batches are more stable but resource-intensive.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section3\"\u003e\u003ch2\u003e3.2.5. SHAP interpretation model\u003c/h2\u003e\u003cp\u003eTo analyze the influence of internal and external microclimate variables on actual load at different times in the LSTM model, as well as how these influences vary under different climate effects, the study used the SHapley Additive exPlanations (SHAP) explanation algorithm. SHAP enhances the interpretability of LSTM models by applying game theory to provide both global and local explanations. The Shapley value \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\varphi\\:}_{i}\\left(f\\right)\\)\u003c/span\u003e\u003c/span\u003e for an input feature \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{i}\\)\u003c/span\u003e\u003c/span\u003e is calculated using the following formula:\u003cdiv id=\"Eque\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Eque\" name=\"EquationSource\"\u003e\n$$\\:{\\varphi\\:}_{i}\\left(f\\right)=\\sum\\:_{S\\subseteq\\:N\\setminus\\:\\left\\{i\\right\\}}\\:\\frac{\\left|S\\right|!\\left(\\left|N\\right|-\\left|S\\right|-1\\right)!}{\\left|N\\right|!}\\left[{f}_{S\\cup\\:\\left\\{i\\right\\}}\\left(S\\cup\\:\\left\\{i\\right\\}\\right)-{f}_{S}\\left(S\\right)\\right]\\:\\left(5\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eHere, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\varphi\\:}_{i}\\left(f\\right)\\)\u003c/span\u003e\u003c/span\u003e is the Shapley value of feature \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{i}\\)\u003c/span\u003e\u003c/span\u003e; \u003cem\u003eN\u003c/em\u003e is the full set of features; \u003cem\u003eS\u003c/em\u003e is any subset excluding \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{i}\\)\u003c/span\u003e\u003c/span\u003e; ∣\u003cem\u003eN\u003c/em\u003e∣ and ∣\u003cem\u003eS\u003c/em\u003e∣ are the sizes of the full set and subset, respectively. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{f}_{S}\\left(S\\right)\\)\u003c/span\u003e\u003c/span\u003e is the model prediction using subset \u003cem\u003eS\u003c/em\u003e, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{f}_{S\\cup\\:\\left\\{i\\right\\}}\\left(S\\cup\\:\\left\\{i\\right\\}\\right)\\)\u003c/span\u003e\u003c/span\u003e is the prediction after adding \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{i}\\)\u003c/span\u003e\u003c/span\u003e. In LSTM models, features span multiple time steps. Each feature at each time step has a specific Shapley value, allowing the contribution of time-dependent inputs to the target variable to be assessed. To evaluate a feature's overall influence across a past time window, we sum the absolute Shapley values of that feature at all time steps. SHAP not only quantifies the importance of each input feature but also captures how its influence on energy use changes over time. This helps reveal both long-term and periodic dependencies between input features and the target variable [\u003cspan citationid=\"CR108\" class=\"CitationRef\"\u003e108\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"4. Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e4.1. Comparison of suburban and local air temperature around the target building\u003c/h2\u003e\u003cp\u003eWe compared UWG-simulated local Ta around the target building with actual readings from the Haidian (HD) suburban weather station during the 2023 heatwave (around June 17\u0026ndash;23) and cold wave (around December 21), referencing the same periods in 2021 for comparison (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Using the daily average temperature of the TMY as a baseline, significant differences can be observed between the extreme weather year (2023) and the normal year (2021). (i) 24-hour temperature profiles: In summer 2021, the average maximum UHI intensity stayed below 8\u0026deg;C. During the 2023 heatwave (around June 23), it exceeded 10\u0026deg;C, an increase of 2\u0026deg;C that can significantly impact building energy consumption. Studies suggest that a 1\u0026deg;C rise in temperature can increase building peak electric loads by up to 4.6% and total energy use by up to 8.5% [\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e]. In winter 2021, UHI intensity was around 5\u0026ndash;6\u0026deg;C, while the 2023 cold wave reduced it by only 0.5\u0026deg;C. UHI intensity also peaks at different times of the day. In summer, it is strongest between 2 and 6 a.m. and weakest at noon [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. At dawn, before solar radiation reaches the ground, residential areas retain heat due to overnight AC use, slowing cooling, while suburban sites like the HD weather station reach their lowest Ta of the day. In winter, UHI intensity is also lowest around noon but peaks around midnight, as heating systems remain on throughout the day while internal heat gains from occupants and equipment decreases during nighttime. (ii) Continuous temperature curves: During the summer heatwave, amplified UHI pushed nighttime ΔTa above 10\u0026deg;C, sharply increasing residential cooling loads. In winter, UHI-driven ΔTa stayed around 6\u0026deg;C, but the nighttime warming effect was significantly reduced during the cold wave, with ΔTa dropping below 3\u0026deg;C. (iii) Temperature deviation scatter plots: Local temperatures were generally higher than those at the suburban station due to UHI. During the 2023 heatwave, the compound effect of heatwave and UHI led to a high RMSE of 6.47\u0026deg;C between Ta_UWG and Ta_HD. During the winter cold wave, the RMSE dropped to 3.83\u0026deg;C. Extreme weather affects the temperature gap between local and suburban areas, amplifying it in summer and narrowing it in winter.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e4.2. Meteorological data validation\u003c/h2\u003e\u003cp\u003eWe compared 2021 rooftop weather station data from a nearby residential building with Ta_HD, Ta_UWG, RH_HD, and RH_UWG to assess differences between suburban (HD) weather data, UWG-simulated, and measured LWD, and to validate the accuracy of UWG outputs (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e(i) and (ii) highlight the 24-hour and daytime differences in Ta and RH between LWD and suburban station data after accounting for local weather. Compared with the HD station, UWG-simulated LWD aligned closely with real local measurements, with R\u0026sup2; values of 81.45% for summer Ta and 82.12% for winter Ta (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e(a, c)-iii). For RH, R\u0026sup2; reached 86.95% in summer and 74.82% in winter (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e(e, g)-iii). Based on the validation results, although UWG slightly overestimated nighttime UHI intensity in some periods, it effectively captured anthropogenic heat effects and explained local climate patterns caused by local UHI effects. Therefore, UWG-simulated Ta and RH data are suitable for building energy simulations.\u003c/p\u003e\u003cp\u003eIn contrast, suburban station data differed significantly from real local data due to the UHI effect. In summer, the temperature difference peaked at 8.3\u0026deg;C at midnight, and 6.8\u0026deg;C at 6 a.m. The largest daytime temperature difference occurred at 3 a.m. on June 19, with UHI reaching 11.6\u0026deg;C (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e(b)-ii). In winter, the temperature difference decreased, with the largest 24-hour difference at 11 p.m. (5.0\u0026deg;C) and the largest daytime difference (6.6\u0026deg;C) at 7 a.m. on December 18 (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e(d)-ii). These measurements show that nighttime Ta gaps between suburban and local rooftop data were significant in both seasons, with UHI peaking around midnight or just before sunrise. Relative humidity differences in summer peaked at 33% at 6 a.m. and 52% at 6 a.m. on June 19. In winter, the largest 16% difference occurred at 11 p.m., aligning with the temperature differences. Daytime differences peaked at 32% at 7 a.m. on December 15. Higher relative humidity in suburban areas are attributed to abundant vegetation and stronger transpiration [\u003cspan citationid=\"CR109\" class=\"CitationRef\"\u003e109\u003c/span\u003e], which leads to lower temperatures and higher moisture content in the air [\u003cspan citationid=\"CR110\" class=\"CitationRef\"\u003e110\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e4.3. Accuracy of annual energy use and peak load simulation\u003c/h2\u003e\u003cp\u003eKeeping the target building constant, this study used LWD to simulate annual heating and cooling energy use for 2021 and 2023. The simulated EUI was compared with actual measured data. Changes in weather conditions were also analyzed for their impact on heating/cooling EUI and design day loads (Load_DDY). Here, EUI_UWG represents energy use intensity under UHI-influenced conditions, with the 2023 data reflecting extreme weather scenarios.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eComparing Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e (a), the 2021 values of EUI_HD and EUI_UWG show that under the UHI effect alone, the target building\u0026rsquo;s cooling demand increased by 8.38 kWh/(m\u0026sup2;\u0026middot;year) and heating demand decreased by 10.76 kWh/(m\u0026sup2;\u0026middot;year), resulting in a net annual energy reduction of 2.38 kWh/(m\u0026sup2;\u0026middot;year). In 2023, under compound influence of UHI and extreme weather, comparisons between EUI_HD and EUI_UWG show that the UHI effect led to a summer cooling demand rose by 10.33 kWh/(m\u0026sup2;\u0026middot;year), a winter heating demand dropped by 10.56 kWh/(m\u0026sup2;\u0026middot;year), and a slight reduction in total energy use by 0.23 kWh/(m\u0026sup2;\u0026middot;year). Comparisons of 2021 and 2023 EUI under HD data show that extreme weather alone increased total energy use by 8.01 kWh/(m\u0026sup2;\u0026middot;year), with cooling up by 5.25 kWh/(m\u0026sup2;\u0026middot;year) and heating up by 2.76 kWh/(m\u0026sup2;\u0026middot;year). Further comparisons of 2021 and 2023 EUI_UWG show that under the compound influence of EW and UHI, total energy use rose by 10.16 kWh/(m\u0026sup2;\u0026middot;year), with a 7.20 kWh/(m\u0026sup2;\u0026middot;year) increase in summer cooling and 2.96 kWh/(m\u0026sup2;\u0026middot;year) in winter heating. UHI significantly raised summer EUI. Comparisons of 2021 EUI_HD with 2023 EUI_UWG reveal that the compound warming effect of heatwaves and UHI sharply increased summer cooling EUI by 15.58 kWh/(m\u0026sup2;\u0026middot;year), posing a significant challenge to the urban energy system [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. In contrast, the opposing influences of cold waves and UHI lowered winter heating EUI by 7.80 kWh/(m\u0026sup2;\u0026middot;year). UHI\u0026rsquo;s long-term influence proved more dominant, while extreme weather events were short-term and abrupt, leading to a total energy increase of 7.78 kWh/(m\u0026sup2;\u0026middot;year).\u003c/p\u003e\u003cp\u003eAdditionally, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e (a) shows that simulations using LWD tend to slightly underestimate EUI, while HD and TMY data tend to overestimate heating EUI and underestimate cooling EUI, especially in TMY, which does not account for climate change and exaggerates heating demand while underestimating cooling. When comparing percentage deviations from actual values, LWD yields more accurate estimates of individual heating and cooling EUI than using suburban weather station data. Using LWD improves simulation accuracy for cooling and heating EUI by 36.6% and 29.6% in a typical year, and by 30.1% and 35.1% in an extreme year. Although total EUI from HD data may appear closer to actual values due to offsetting errors (overestimated heating and underestimated cooling), this balance is coincidental. Results calculated using suburban weather station data still carry considerable uncertainty, especially in estimating seasonal heating and cooling EUI or equipment load. Whenever possible, LWD is strongly preferred.\u003c/p\u003e\u003cp\u003eWe further compared the 2021 and 2023 Load_DDY using both suburban weather station data (HD) and LWD (UWG). Subplots (b) and (c) show that under UHI influence alone, the 0.4% design day cooling load (Cooling Load_DDY 0.4%) in summer decreased by 0.31 W/m\u0026sup2; in 2021 and by 0.11 W/m\u0026sup2; in 2023. Similarly, the 99.6% design day heating load (Heating Load_DDY 99.6%) in winter dropped by 1.54 W/m\u0026sup2; and 3.40 W/m\u0026sup2;, respectively. This indicates that UHI notably offset the impact of the 2023 winter cold wave on heating loads. Comparing Load_DDY \u003csub\u003eHD\u003c/sub\u003e between 2021 and 2023 shows that under the influence of a heatwave alone, the Cooling Load_DDY 0.4% increased by 1.37 W/m\u0026sup2;, while under the influence of a cold wave alone, the Heating Load_DDY 99.6% increased by 3.68 W/m\u0026sup2;. Further comparing Load_DDY \u003csub\u003eUWG\u003c/sub\u003e between 2021 and 2023 reveals that under the compound effects of UHI and a heatwave, the Cooling Load_DDY 0.4% increased by 1.57 W/m\u0026sup2;, and under the compound effects of UHI and a cold wave, the Heating Load_DDY 99.6% increased by 1.82 W/m\u0026sup2;. The presence of UHI significantly reduced winter peak heating loads. Comparing Load_DDY \u003csub\u003eHD\u003c/sub\u003e in 2021 with Load_DDY \u003csub\u003eUWG\u003c/sub\u003e in 2023, we see that the joint effect of a heatwave and UHI increased the Cooling Load_DDY 0.4% by 1.26 W/m\u0026sup2;, while the compound effect of a cold wave and UHI increased the Heating Load_DDY 99.6% by just 0.28 W/m\u0026sup2;. In summary, as organized in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, we examined the individual and compound impacts of UHI and EW on heating and cooling EUI and design day loads. For target residential buildings, EW increases both heating and cooling EUI and design day loads, placing additional strain on building energy systems. Although UHI increases summer cooling EUI [\u003cspan citationid=\"CR111\" class=\"CitationRef\"\u003e111\u003c/span\u003e], it reduces winter heating demand, total annual EUI, and both cooling and heating design day loads [\u003cspan citationid=\"CR112\" class=\"CitationRef\"\u003e112\u003c/span\u003e]. This aligns with the findings of Y. Hirano and T. Fujita regarding the UHI's impact on heating and cooling energy consumption in Tokyo's residential buildings [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Erell et al. also found that increased summer cooling demand from UHI may be offset by winter heating energy savings [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]. Notably, UHI raises winter temperatures, reducing heating load fluctuations during cold waves [\u003cspan citationid=\"CR113\" class=\"CitationRef\"\u003e113\u003c/span\u003e]. Similarly, in summer, UHI moderates local temperature fluctuations compared to suburban areas, also reducing variations in cooling loads.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eImpact of EW, UHI, and their compound effects on building heating/cooling EUI and design day loads.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eScenario\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eUrban heat island only\u003c/p\u003e\u003cp\u003e(2021 HD, 2021 UWG)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExtreme weather only\u003c/p\u003e\u003cp\u003e(2021 HD, 2023 HD)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eUHI\u0026thinsp;+\u0026thinsp;extreme weather\u003c/p\u003e\u003cp\u003e(2021 HD, 2023 UWG)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCooling EUI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eIncrease\u003c/p\u003e\u003cp\u003e(8.38 kWh/(m\u0026sup2;\u0026middot;year), 57.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eIncrease\u003c/p\u003e\u003cp\u003e(5.25 kWh/(m\u0026sup2;\u0026middot;year), 36.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eIncrease\u003c/p\u003e\u003cp\u003e(15.58 kWh/(m\u0026sup2;\u0026middot;year), 107.4%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHeating EUI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDecrease\u003c/p\u003e\u003cp\u003e(10.76 kWh/(m\u0026sup2;\u0026middot;year), 34.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eIncrease\u003c/p\u003e\u003cp\u003e(2.76 kWh/(m\u0026sup2;\u0026middot;year), 8.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eDecrease\u003c/p\u003e\u003cp\u003e(7.80 kWh/(m\u0026sup2;\u0026middot;year), 25.2%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTotal EUI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDecrease\u003c/p\u003e\u003cp\u003e(2.38 kWh/(m\u0026sup2;\u0026middot;year), 5.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eIncrease\u003c/p\u003e\u003cp\u003e(8.01 kWh/(m\u0026sup2;\u0026middot;year), 17.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eIncrease\u003c/p\u003e\u003cp\u003e(7.78 kWh/(m\u0026sup2;\u0026middot;year), 17.1%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCooling Load_DDY\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDecrease\u003c/p\u003e\u003cp\u003e(0.31 W/m\u0026sup2;, 1.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eIncrease\u003c/p\u003e\u003cp\u003e(1.37 W/m\u0026sup2;, 8.1%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eIncrease\u003c/p\u003e\u003cp\u003e(1.26 W/m\u0026sup2;, 7.4%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHeating Load_DDY\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDecrease\u003c/p\u003e\u003cp\u003e(1.54 W/m\u0026sup2;, 6.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eIncrease\u003c/p\u003e\u003cp\u003e(3.68 W/m\u0026sup2;, 15.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eIncrease\u003c/p\u003e\u003cp\u003e(0.28 W/m\u0026sup2;, 1.2%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e* The values in parentheses represent the change amount and its proportion relative to the original values before local climate effects.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003e4.4. Correlation between measured loads, simulated loads, and weather parameters\u003c/h2\u003e\u003cp\u003eWe analyzed the correlation between hourly measured loads and simulated loads or weather parameters during extreme weather, using three datasets: UWG, HD, and TMY (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). This helps assess how closely each dataset reflects real conditions. Results show that Ta, RH, and simulated loads based on both UWG and HD data correlate significantly with actual loads, but the LWD based on UWG performs best. Ta, RH, and simulated loads from UWG data show the strongest correlations with measured loads (e.g., 2021 cooling load: 0.51***; 2023 cooling load: 0.56***; 2021 heating load: 0.65***; 2023 heating load: 0.57***), outperforming both HD and TMY data. During heatwaves, simulated cooling loads and Ta are positively correlated with actual cooling load, while RH shows a negative correlation. In cold waves, simulated heating loads are positively correlated with actual heating loads, while Ta and RH show negative correlations. In the same year, load and temperature correlations with actual values are stronger in winter than in summer.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePearson correlation between hourly measured heating and cooling loads and UWG, HD, and TMY weather data and simulated loads during extreme weather period in 2021 and 2023.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2021 measured cooling load (kWh/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2023 measured cooling load (kWh/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003e2021 measured heating load (kWh/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003e2023 measured heating load (kWh/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eSimulated load\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLoad_UWG (kWh/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.51***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.56***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.65***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.57***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLoad_HD (kWh/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.16**\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.42***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.55***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.49***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLoad_TMY (kWh/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e-0.06\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.25***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e-0.23***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.14**\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003eUWG meteorological parameters\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTa_UWG (\u0026deg;C)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.55***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.58***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e-0.72***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e-0.66***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRH_UWG (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e-0.39***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e-0.52***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e-0.25***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e-0.46***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eWS_UWG (m/s)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.14*\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.05\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.16**\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRad_UWG (W/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e-0.12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e-0.08\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e-0.10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e-0.08\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003eHD meteorological parameters\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTa_HD (\u0026deg;C)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.18**\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.39***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e-0.60***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e-0.58***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRH_HD (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e-0.28***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e-0.46***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e-0.22***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e-0.32***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eWS_HD (m/s)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.05\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.15*\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.05\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.17***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRad_HD (W/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e-0.12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e-0.08\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e-0.10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e-0.08\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003eTMY meteorological parameters\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTa_TMY (\u0026deg;C)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e-0.06\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.25***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.20***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e-0.17***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRH_TMY (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.21***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e-0.07\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.46***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e-0.40***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eWS_TMY (m/s)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.18**\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e-0.19***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.29***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRad_TMY (W/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e-0.22***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e-0.19**\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e-0.15**\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e-0.07\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e* The p-value indicates statistical significance; the smaller the p-value, the more significant the correlation. Results with p\u0026thinsp;\u0026gt;\u0026thinsp;0.05 are considered not statistically significant (*** p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, ** p\u0026thinsp;\u0026lt;\u0026thinsp;0.01, * p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\u003c/div\u003e"},{"header":"5. Discussion","content":"\u003cp\u003eThis section compares the numerical differences between simulated and measured hourly heating and cooling loads under different climate conditions using three weather datasets (UWG, HD, and TMY), and analyzes the discrepancies caused by extreme weather, urban heat islands, occupant behavior, and limitations of HVAC models based on ideal air systems. It also introduces an LSTM-based calibration method to improve simulation accuracy. Additionally, it examines how the influence of internal and external disturbances on actual energy loads varies over time under extreme weather, highlights the overall impact weight of key factors, and identifies critical microclimate elements for climate risk mitigation.\u003c/p\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003e5.1. Comparison of 24-hour and multi-day trends between simulated and measured loads during extreme weather\u003c/h2\u003e\u003cp\u003eWe compared the relationships among simulated load, air temperature, scheduled occupancy patterns, and actual load in winter and summer under two scenarios: the typical year with isolated UHI and the extreme year with combined extreme weather and local heat islands. We also examined the multi-day variations and 24-hour patterns in the discrepancies between simulated and actual loads (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e), and explored the underlying causes.\u003c/p\u003e\u003cdiv id=\"Sec18\" class=\"Section3\"\u003e\u003ch2\u003e5.1.1. Comparison of cooling load in summer\u003c/h2\u003e\u003cp\u003eUnder the influence of isolated UHI during typical summer days, the simulated load closely follows the diurnal Ta changes (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e(a\u0026ndash;c)-i), showing clear peaks and troughs aligning with sunset (around 18:00) and sunrise, respectively. This indicates that simulated loads are strongly influenced by rising surface Ta and Rad gains, which together increase building energy loads. Among the three weather datasets, LWD reflects nighttime UHI effects, producing smoother load curves that align better with actual data than HD or TMY. However, discrepancies between the assumed design schedules and actual occupant and device usage still lead to differences in simulated energy loads. The real load is heavily influenced by occupant behavior, particularly daily routines. In summer, this results in a noticeable shift, with actual peak loads occurring later\u0026mdash;around midnight\u0026mdash;compared to the earlier peaks. Based on the multi-day summer trends (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e (a-c)-ii), unlike simulated loads that closely follow Ta changes, actual loads are influenced by occupants' AC use habits. For example, on cooler days like June 17 and 24, 2021, and June 20, 2023, simulated loads dropped with lower Ta, but some users still ran air conditioning out of habit, leading to a mismatch between simulated and actual energy use. Using LWD, the simulated load tends to slightly overestimate the peak values compared to actual load, whereas HD captures peak values more accurately but underestimates minimum loads due to the omission of UHI effects. Compared to suburban station and TMY data, simulations using LWD align more closely with actual loads along the 45\u0026deg; line (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e(a\u0026ndash;c)-iii), with the lowest RMSE and highest Pearson correlation.\u003c/p\u003e\u003cp\u003eUnder the compound effects of heatwave and UHI during the extreme summer days in 2023, the simulated cooling load peaks still aligned with the Ta peak at 6 p.m., while the actual load peaks remain at midnight (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e(d\u0026ndash;f)-i). Peak demand reached 23 W/m\u0026sup2; (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e(d)-ii), nearly double that of the same period in a typical year\u0026mdash;an increase of 11 W/m\u0026sup2;\u0026mdash;placing significant short-term stress on the urban summer power grid. Although LWD slightly overestimated loads in a normal year due to occupancy assumptions, it underestimated the peak cooling load during the heatwave by about 5 W/m\u0026sup2;. HD and TMY data, which ignore UHI, further underestimated cooling load during heatwave (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e(d\u0026ndash;f)-ii) [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. TMY also fails to account for long-term climate warming. Relying on these datasets could lead to significant underestimation of peak demand during heatwaves, posing greater potential risks. Hourly load scatter plots (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e(d\u0026ndash;f)-iii) show that during the heatwave, LWD data consistently produced load estimates closer to actual values than those from HD or TMY.\u003c/p\u003e\u003cp\u003eA comparison of actual daytime loads in 2021 and 2023 shows that heatwaves significantly increased both peak and cumulative loads (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e(g)-ii). During the 2023 heatwave days, the actual daily peak load rose to 17.5 W/m\u0026sup2;\u0026mdash;an increase of 8.5 W/m\u0026sup2; over 2021\u0026mdash;while daily fluctuations reached 10 W/m\u0026sup2;, 5 W/m\u0026sup2; higher than in 2021 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e (g)-i).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eIn summary, the following conclusions can be drawn: (1) Summer load variations driven by actual occupancy patterns differ from those simulated by ideal air systems, which respond sensitively to weather conditions. This leads to mismatched peak times and an underestimation of actual peak loads during heatwaves. (2) User behavior is habitual\u0026mdash;once air conditioning is turned on, some users continue using it even after weather cools down. As a result, the timing of the first heatwave can significantly influence total seasonal cooling demand. Overall, while external factors like weather affect energy loads, they are only part of the picture. Despite some limitations\u0026mdash;such as mismatched peak timing, underestimation of peak loads, and reliance on outdoor Ta rather than user behavior\u0026mdash;LWD-based simulations still align more closely with real load trends during the heatwave than other datasets, helping improve prediction accuracy.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section3\"\u003e\u003ch2\u003e5.1.2. Comparison of heating load in winter\u003c/h2\u003e\u003cp\u003eUnder the influence of isolated UHI during typical winter days in 2021, the 24-hour daily load profile (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e(a\u0026ndash;c)-i) shows that actual heating loads remain highly stable, with minimal response to outdoor air temperature changes. Central heating, unaffected by user behavior or manual activation, maintained a steady hourly load with no distinct peaks. In contrast, simulated loads\u0026mdash;based on ideal air systems\u0026mdash;fluctuate with outdoor Ta, showing clear peaks and troughs. However, the peaks and troughs within the daily cycle balance out, resulting in a similar overall intensity to measured daily loads. From the multi-day view (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e(a)-ii), although winter heating loads simulated using LWD differ significantly from actual hourly patterns, both follow similar daily trends, rising and falling noticeably with Ta changes. The scatter plot in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e(a)-iii reveals much greater variation in simulated loads, creating a distorted distribution compared to real loads. HD and TMY simulations show similar issues, but with one key difference: winter UHI reduces heating demand. Since HD and TMY don\u0026rsquo;t account for UHI and have lower, more fluctuating Ta, they lead to substantial overestimation and increased volatility in heating load predictions.\u003c/p\u003e\u003cp\u003eUnder the compound effects of cold wave and UHI during the extreme winter days in 2023, actual heating loads remained relatively stable throughout the day (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e(d)-i). Unlike during heatwaves, simulated loads\u0026mdash;based on ideal air systems\u0026mdash;overestimated actual demand due to differences in HVAC system behavior, supply temperature settings, and reduced COP under extreme cold. LWD overestimated the 24-hour daily average and peak heating loads by approximately 3 W/m\u0026sup2; and 16 W/m\u0026sup2;, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e(d)-i,ii), while HD data resulted in even larger overestimations\u0026mdash;around 7 W/m\u0026sup2; for the average and 18 W/m\u0026sup2; for the peak (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e(e)-i,ii). LWD is suitable for estimating winter loads under normal conditions but less reliable during extreme cold. Relying on simulated loads for heating system control may lead to actual heating exceeding real demand, resulting in over-supply, thermal discomfort, and energy waste. Additionally, because simulated loads are based on ideal air systems, they respond sensitively to temperature drops during cold waves, whereas actual district heating systems adjust based on end-point temperature feedback. The thermal inertia of the building\u0026rsquo;s walls causes indoor temperature changes to lag behind outdoor temperature variations, resulting in a delayed response due to thermal inertia and control lag (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e(d)-ii).\u003c/p\u003e\u003cp\u003eA comparison of actual daytime loads in 2021 and 2023 (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e(g)-ii) show that cold wave significantly raised both peak and cumulative heating demand (represented by the area between the curve and the x-axis). Unlike in summer, actual winter hourly heating loads\u0026mdash;whether in typical or cold wave periods\u0026mdash;remain relatively stable throughout the day without sharp peaks or troughs (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e(g)-i). However, during cold waves, average hourly heating EUI increases by about 2 W/m\u0026sup2;.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eIn summary, winter heating load simulations have several limitations: (1) Due to the flexible temperature control settings of the ideal air system, the simulated load is overly sensitive to temperature variations, which amplifies daily heating load fluctuations. This is inconsistent with the relatively stable load variations of central heating systems in northern regions, resulting in peaks and troughs not observed in the actual daily cycle. (2) Due to the end-point temperature feedback-dependent nature of district heating systems, actual heating load responses are delayed. Despite these limitations, applying detailed district heating HVAC models in UBEM is often impractical, due to the complexity of simulating plants, distribution, substations, and terminals for every building making large-scale modeling difficult. To improve winter heating load estimation in UBEM, machine learning models are needed to better capture and predict realistic heating load patterns.\u003c/p\u003e\u003cp\u003eTo address the over-sensitivity of ideal air system simulations to temperature changes, we applied an LSTM time-series model aimed to predict actual winter energy use under EW conditions. The model uses LWD (Ta, RH, WS, Rad), simulated loads, occupancy schedules, and local time as input features, with actual heating load as the prediction target. Using LWD improves the accuracy of meteorological inputs and reduces unnecessary temperature fluctuations, enhancing winter load prediction under EW. The model was trained on data from 2021 winter and validated using data from 2023 winter (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e). The model achieved an R\u0026sup2; of 78.38% (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e(iii)), with optimal epochs selected based on the loss curve (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e(iv)). In the prediction results, the model did not fully capture the lag in daily peak heating demand and slightly underestimated intraday variation, which was about 3 W/m\u0026sup2; on December 21, 2023, but it reduced daily fluctuations and captured overall trends well (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e(ii)).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003ch2\u003e5.2. Impact of internal and external factors on winter and summer building loads\u003c/h2\u003e\u003cp\u003eTo better understand the impact of internal and external factors on hourly building loads under extreme weather, we developed LSTM models using 2021 and 2023 LWD and local time as inputs, with actual energy use as the target. We compared models using only external inputs with those including both external and internal factors to assess their ability to explain actual energy use (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). External factors refer to microclimate variables, while internal factors, such as occupancy and equipment use, are difficult to measure but follow daily patterns, making local time (LT) a useful proxy. Results show that energy prediction accuracy was similar across extreme and typical year for the same season. External factors had a greater impact on heating loads than on cooling loads, indicating that winter energy use is more sensitive to outdoor weather than summer energy use. Including internal factors (via local time) improved cooling load prediction by up to 9.8% and heating load prediction by up to 2.7%, suggesting that internal influences are less significant in winter.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eTest set R\u0026sup2; of LSTM load prediction models (100 epochs).\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eScenario\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLWD_summer_2021\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLWD _summer_2023\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLWD _winter_2021\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eLWD _winter_2023\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eExternal factors only\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.6101\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.5852\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.7767\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.8163\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInternal\u0026thinsp;+\u0026thinsp;external factors\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.7079\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.6070\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.8038\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.8360\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cdiv id=\"Sec21\" class=\"Section3\"\u003e\u003ch2\u003e5.2.1 Hourly variation patterns of internal and external impacts on winter and summer actual building loads\u003c/h2\u003e\u003cp\u003eTo further identify the most influential microclimate factors affecting actual load under isolated UHI and combined UHI\u0026ndash;extreme weather conditions, as well as the temporal variation patterns in the influence weights of microclimate factors and occupant behavior on actual load, we applied the SHAP algorithm to interpret the key drivers in the LSTM load prediction model. This allows for a comparison of the temporal variation patterns in the influence weights of five factors\u0026mdash;air temperature (Ta), relative humidity (RH), wind speed (WS), solar radiation (Rad), and local time (LT)\u0026mdash;under different local climate conditions in typical and extreme years in both winter and summer, including the hourly variations and peak timings of each factor\u0026rsquo;s SHAP effect.\u003c/p\u003e\u003cdiv id=\"Sec22\" class=\"Section4\"\u003e\u003ch2\u003e5.2.1.1. Variation of factor weights in summer\u003c/h2\u003e\u003cp\u003eIn summer (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e), \u003cem\u003eTa effect\u003c/em\u003e closely follows the trends of both \u003cem\u003eTa\u003c/em\u003e and \u003cem\u003emean load\u003c/em\u003e, peaking after 10 p.m., shortly after the \u003cem\u003eTa\u003c/em\u003e peak and aligning with the \u003cem\u003emean load\u003c/em\u003e peak. Its lowest point occurs around 5 a.m. (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e (a)-Ta (i)). During heatwaves in 2023, \u003cem\u003eTa effect\u003c/em\u003e peaks earlier\u0026mdash;around 5 p.m.\u0026mdash;when temperatures are highest (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e (b)-Ta (i)). \u003cem\u003eTa effect\u003c/em\u003e aligns more closely with hourly load peaks during high load periods than at other times from daily variation, both in typical summers and during heatwaves (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e (a, b)-Ta (ii)). \u003cem\u003eRH\u003c/em\u003e and \u003cem\u003eWS effects\u003c/em\u003e are weaker and show poor alignment with load peaks in both normal and extreme years (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e (a, b)-RH,WS(i, ii)). \u003cem\u003eRad effect\u003c/em\u003e peaks around 5 p.m. in normal year, when radiation drops rapidly (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e (a)-Rad(i)). During heatwaves, it shows two peaks (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e (b)-Rad(i)): one at midday when Rad is highest, and another around 8 p.m. when it drops rapidly. \u003cem\u003eLocal time effect\u003c/em\u003e displays a strong diurnal pattern, closely matching hourly load trends (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e (a,b)-LT (ii)) and perfectly aligning with peak \u003cem\u003emean load\u003c/em\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e (a,b)-LT (i)), indicating internal factors dominate summer peak demand, while microclimate elements are secondary. Overall, \u003cem\u003eLocal time\u003c/em\u003e dominates during the night and early morning, while weather factors are more influential in the midday and afternoon. This reflects typical residential occupancy patterns: low daytime occupancy coincides with higher outdoor Ta, while nighttime occupancy is higher when outdoor Ta is relatively lower [\u003cspan citationid=\"CR114\" class=\"CitationRef\"\u003e114\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec23\" class=\"Section4\"\u003e\u003ch2\u003e5.2.1.2. Variation of factor weights in winter\u003c/h2\u003e\u003cp\u003eIn winter (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e), whether in typical year or during cold waves, the \u003cem\u003eTa effect\u003c/em\u003e peaks around 3 p.m., aligning with the daily high in Ta, and drops to its lowest point around 4 a.m., when Ta is relatively low and indoor occupancy is highest (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e(a, b)-Ta(i)). During colder periods\u0026mdash;especially the cold wave\u0026mdash;\u003cem\u003eTa effect\u003c/em\u003e aligns closely with peak \u003cem\u003ehourly load\u003c/em\u003e, indicating stronger temperature influence on heating demand (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e(a, b)-Ta(ii)). \u003cem\u003eRH effect\u003c/em\u003e and \u003cem\u003eWS effect\u003c/em\u003e show weaker correlations with \u003cem\u003emean load\u003c/em\u003e and \u003cem\u003ehourly load\u003c/em\u003e peaks (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e(a, b)- RH, WS (i, ii)). However, during cold wave days, the \u003cem\u003eWS effect\u003c/em\u003e aligns more closely with the \u003cem\u003eHourly load\u003c/em\u003e peak, as higher wind speeds under low temperature accelerate heat loss from building surfaces and increase heating load demand (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e(b)\u0026ndash;WS(ii)) [\u003cspan citationid=\"CR115\" class=\"CitationRef\"\u003e115\u003c/span\u003e, \u003cspan citationid=\"CR116\" class=\"CitationRef\"\u003e116\u003c/span\u003e]. Interestingly, despite lower winter Rad (about half of summer levels), the \u003cem\u003eRad effect\u003c/em\u003e still shows two daytime peaks during cold waves: one at noon (solar peak) and another around 8 p.m., when Rad is zero (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e(b)-Rad(i)). The evening peak of \u003cem\u003eRad effect\u003c/em\u003e closely matches the \u003cem\u003emean load\u003c/em\u003e peak, likely due to Rad dropping to zero while other factors contribute less at that time. \u003cem\u003eLocal time effect\u003c/em\u003e differs significantly from summer. It peaks around 5 a.m. (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e(a,b)-LT(i)), but this does not align with the \u003cem\u003emean load\u003c/em\u003e peak at 8 p.m., indicating that higher nighttime occupancy does not increase heating loads. This is due to buildings using centralized district heating from thermal stations in winter, which is not directly controlled by occupant behavior.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec24\" class=\"Section3\"\u003e\u003ch2\u003e5.2.2. Impact weights of internal and external factors on winter and summer actual building loads\u003c/h2\u003e\u003cp\u003eTo further clarify the dominant microclimate factors influencing building energy consumption under different climate conditions in both typical and extreme years, we visualized the overall influence weight proportions of different factors (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e-i), as well as their 24-hour diurnal and daytime variations patterns (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e-ii, iii). We also analyzed the temporal changes in the influence weight proportions of microclimate factors and occupant behavior on actual heating and cooling loads.\u003c/p\u003e\u003cp\u003eFrom the diurnal perspective (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e (iii)), whether in typical or extreme years, during heatwaves or cold waves, \u003cem\u003eLocal time\u003c/em\u003e has the highest influence around 4 a.m. [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u0026mdash;a period of peak occupancy, no solar radiation, and stable outdoor microclimate before sunrise. The lowest \u003cem\u003eLocal time\u003c/em\u003e influence varies by season: in summer, it occurs around 2 p.m. when outdoor Ta peaks and occupancy is lower (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e(a,b)-iii); in winter, it drops lowest point around 4 p.m. (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e(c)-iii) as occupancy decreases and microclimate factors like Ta and Rad drop sharply before sunset. During cold waves, as rapid Ta changes occur later, this low point of the influence of local time shifts to 7\u0026ndash;8 p.m. (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e(d)-iii). Rad and Ta show typical intraday patterns\u0026mdash;lower influence in the early morning and late at night, and higher in the afternoon and evening. Radiation, in particular, varies strongly with solar angle throughout the day. Comparing seasons, Local time has a greater influence in summer, while outdoor air temperature becomes more dominant in winter.\u003c/p\u003e\u003cp\u003eFrom the daily variation (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e (ii)), around the summer heatwave (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e(b)-ii, June 17 and 23, 2023), the influence of Ta and RH drops significantly, while Rad becomes the dominant factor. During the winter cold wave (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e(d)-ii, December 21, 2023), the influence of Ta increases sharply, while the impact of \u003cem\u003eLocal time\u003c/em\u003e\u0026mdash;representing internal factors\u0026mdash;decreases noticeably.\u003c/p\u003e\u003cp\u003eFrom the overall importance (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e (i)), during typical summer days, \u003cem\u003eLocal time\u003c/em\u003e has the greatest impact on actual load (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e(a)-i), followed by Ta and Rad, with WS having the least effect. During heatwaves (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e(b)-i), Rad becomes the most influential factor, surpassing both \u003cem\u003eLocal time\u003c/em\u003e and Ta. During typical winter days, Ta dominates due to lower solar angles and the continuous operation of central heating unaffected by user behavior (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e(c)-i), followed by local time and then solar radiation. During cold waves (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e(d)-i), the influence of Ta increases further, while the weight of \u003cem\u003eLocal time\u003c/em\u003e decreases, though the overall factor ranking remains similar to that of a typical year.\u003c/p\u003e\u003cp\u003eIn summary, compared to summer, \u003cem\u003eLocal time\u003c/em\u003e\u0026mdash;representing occupancy and equipment use patterns\u0026mdash;has less impact in winter due to continuous heating, while the influence of outdoor weather, especially air temperature, on actual load increases. During extreme weather events in both seasons, external factors gain importance, while internal factors become less influential.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eUHI effects vary locally and persist long-term, while extreme weather is unpredictable, with strong interannual and seasonal variation. Under global warming, with widespread local UHI, and more frequent extreme events, TMY data can no longer accurately represent real urban local weather conditions. To address the compound impact of UHI and EW on building energy use and loads, this study introduces LWD and conducts an empirical analysis on a high-rise student dormitory in a cold region. We examine how external local climate conditions and internal factors (occupant and equipment use) affect real heating and cooling loads and EUI under the compound effects. This study also provides empirical support for scaling the method to broader urban applications. The study highlights the reliability and advantages of LWD in building energy simulation. Compared to suburban weather stations and TMY data, it more accurately captures the weather conditions around the target building, providing more precise external inputs for large-scale UBEM [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. This improves simulation accuracy and strengthens the model\u0026rsquo;s ability to explain actual heating and cooling intensity, as well as daily load variations.\u003c/p\u003e\u003cdiv id=\"Sec26\" class=\"Section2\"\u003e\u003ch2\u003e6.1 Key research findings and actionable insights\u003c/h2\u003e\u003cp\u003eFindings show that occupancy schedules heavily influence simulation results, and LWD is key to improving accuracy [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e, \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e]. This study focuses on three main questions: (1) Compound effects of EW and the UHI on local weather and annual building energy use simulation accuracy. (2) Hourly differences and patterns between simulated and actual loads under individual and compound effects of EW and the UHI. (3) The temporal variation in the influence weights of localized weather and internal factors on the actual loads. The key conclusions are summarized as follows:\u003c/p\u003e\u003cp\u003e(1) For the target residential building, localized weather data improved accuracy by 29.6% in summer and 36.6% in winter during typical years, and by 35.1% and 30.1% during extreme years. Unlike EW, which increases both EUI and load, UHI raises summer cooling EUI but reduces winter and annual EUI, as well as both heating and cooling design-day loads.\u003c/p\u003e\u003cp\u003e(2) Hourly load simulations using ideal air systems may underestimate or overestimate load fluctuations during heatwaves or cold spells. In particular, for winter heating, the ideal air system shows oversensitivity to outdoor temperature changes, leading to unrealistic load fluctuations and significant overestimation of energy demand during cold waves. To address this, simulation results needed to be calibrated using real district heating load data combined with LSTM-based models.\u003c/p\u003e\u003cp\u003e(3) Occupancy significantly affects summer cooling loads but has less impact on winter district heating. Unlike summer cooling loads, which are strongly influenced by occupant behavior, winter district heating loads are less sensitive to occupancy in diurnal cycles. During extreme weather in both summer and winter, the influence of external factors increases, while the weight of internal factors decreases.\u003c/p\u003e\u003cp\u003eBased on the key findings, feasible strategies can be proposed for differentiated building design, building energy system control, urban energy system planning, and policy development tailored to localized urban climates.\u003c/p\u003e\u003cp\u003e(1) Climate-responsive building design and energy system control: During heatwaves and cold spells, external factors such as solar radiation and air temperature become more dominant, while internal factors have less impact. Design strategies should consider the time-varying weights of external and internal factors and improve the local climate through coordinated design at both the individual building and urban cluster levels. In summer, the solar heat gains can be reduced through reflective roofs [\u003cspan citationid=\"CR117\" class=\"CitationRef\"\u003e117\u003c/span\u003e, \u003cspan citationid=\"CR118\" class=\"CitationRef\"\u003e118\u003c/span\u003e], movable shading devices, or self-shading building forms during extreme heat events. In winter, enhanced thermal insulation of building envelopes and leveraging UHI through urban form (e.g., slowing heat loss in street canyons through cluster design) can help mitigate cold wave impacts and reduce climate-related risks. Heating systems can also be preloaded ahead of cold spells based on accurate forecasts to avoid delays in heat supply.\u003c/p\u003e\u003cp\u003e(2) Urban energy system planning and policy support based on local climate: Traditional TMY data no longer suffice for building energy consumption assessment and urban design in dense high-density areas under the complex future climate conditions caused by compound impact of EW and UHI conditions. This study recommends incorporating LWD into building energy simulations\u0026mdash;particularly in areas with strong UHI effects and frequent EW events, as well as for evaluating individual heating and cooling demands. This approach enables more accurate assessments of urban building energy use and carbon emissions.\u003c/p\u003e\u003cp\u003eLWD can also be used to develop urban local climate zone (LCZ) maps and inform climate-specific design strategies, enabling differentiated design based on LCZ. It also supports the refinement of climate-responsive building energy efficiency standards and carbon accounting protocols, and helps develop urban risk zones and energy capacity planning models to optimize the deployment of power grids and district heating networks [\u003cspan citationid=\"CR119\" class=\"CitationRef\"\u003e119\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec27\" class=\"Section2\"\u003e\u003ch2\u003e6.2 Methodological strengths, limitations, and future outlook\u003c/h2\u003e\u003cp\u003eThis study proposes a coupled UCM\u0026thinsp;+\u0026thinsp;UBEM framework that incorporates LWD into the energy simulation workflow\u0026mdash;enhancing conventional UBEM, which typically relies only on building program and construction classifications, by incorporating more accurate weather data inputs to improve simulation accuracy. Compared to the chained CFD (Envi-met)\u0026thinsp;+\u0026thinsp;BEM approach, this method supports longer time spans (full-year, hourly simulations) and larger spatial scales (from neighborhoods to entire cities). It captures the compound impact of UHI and EW on annual building energy intensity. The parameterized UCM approach avoids the complex modeling of vegetation and building geometry. Using the UWG tool, suburban weather station data can replace meteorological inputs at the top of the urban canopy, and its integration with BEM allows for the impact of AC heat emissions to be fully considered. Additionally, based on QGIS platform, the resolution of localized weather grids can be flexibly defined, enabling finer analysis of environmental influences on local climates. This approach supports large-scale block or city-level modeling with LWD support. Its application improves UBEM accuracy, enhances understanding of external influences, and provides decision-making support for renewable energy\u0026ndash;based urban energy planning under future climate change to enhance urban and building climate resilience [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe study also applies an LSTM time-series model and SHAP interpretability, using \u003cem\u003eLocal time\u003c/em\u003e to reflect internal factors such as occupancy and equipment use, capturing their time-varying impact on energy loads. However, limitations remain, including arbitrary grid definitions, the inability of the UWG tool to account for vegetation evapotranspiration [\u003cspan citationid=\"CR120\" class=\"CitationRef\"\u003e120\u003c/span\u003e, \u003cspan citationid=\"CR121\" class=\"CitationRef\"\u003e121\u003c/span\u003e], and a fragmented workflow across QGIS and Rhino platforms for LWD generation and building energy simulation. In the future, the workflow can be streamlined by integrating LWD calculation, urban building modeling, and UBEM simulation into a unified tool within Rhino or GIS, enabling more efficient use by urban planners and architects.\u003c/p\u003e\u003cp\u003eUnderstanding the interaction between local climate and energy use is critical for long-term [\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e], climate-sensitive urban energy planning. Accurate local climate data and baseline models are essential for informed building design and post-construction evaluation [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Climate-integrated design tools enable better prediction and response to climate\u0026ndash;building interactions, supporting climate-adaptive architecture design. This approach fosters deeper integration of climate science with building and urban design, and provides reliable data for more accurate calculation of operational carbon emissions, providing robust data support for achieving carbon neutrality in the built environment.\u003c/p\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data that has been used is confidential.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of generative AI in scientific writing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDuring the preparation of this work the authors used ChatGPT 4.0 for language refinement. After using this service, the authors reviewed and edited the content as needed and take full responsibility using this service, the authors reviewed and edited the content as needed and take full responsibility\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFunding: This work\u0026nbsp;was funded by the National Key R\u0026amp;D Program of China [2022YFC3803801 in 2022YFC3803800], National Natural Science Foundation of China [NO.42171337], Beijing Natural Science Foundation [grant No.8232030].\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eJay O, Capon A, Berry P, Broderick C, de Dear R, Havenith G, Honda Y, Kovats RS, Ma W, Malik A (2021) Reducing the health effects of hot weather and heat extremes: from personal cooling strategies to green cities. Lancet 398(10301):709\u0026ndash;724\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBenestad RE, Lussana C, Dobler A (2024) Global record-breaking recurrence rates indicate more widespread and intense surface air temperature and precipitation extremes. Sci Adv 10(45):eado3712\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhu Z, Shao L, Lu R, Hua W (2025) Two contrasting tropical convection modes from the eastern Pacific to northern Africa that drive Eurasian teleconnections in boreal summer. npj Clim Atmospheric Sci 8(1):56\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhu Z, Yang Y, Liu B (2025) Physics-based predictions of the month-by-month summer western North Pacific anomalous anticyclone. J Clim 38(10):2187\u0026ndash;2203\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang W, Clark R, Zhou T, Li L, Li C, Rivera J, Zhang L, Gui K, Zhang T, Li L (2023) : Weather and climate extremes hitting the globe with emerging features, Springer, 2024\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXiao H, Xu P, Wang L (2024) The unprecedented 2023 North China heatwaves and their S2S predictability. Geophys Res Lett 51(6) e2023GL107642.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEntity NFERCR (2021) The february 2021 cold weather outages in texas and the south central united states, FERC and NERC and Regional Entity Staff, Tech. Rep\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMagness B (2021) Review of February 2021 extreme cold weather event\u0026ndash;ERCOT presentation. Urgent Board of Directors Meeting, ERCOT\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSantamouris M, Cartalis C, Synnefa A, Kolokotsa D (2015) On the impact of urban heat island and global warming on the power demand and electricity consumption of buildings\u0026mdash;A review. Energy Build 98:119\u0026ndash;124\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHashemi F, Mills G, Poerschke U, Iulo LD, Pavlak G, Kalisperis L (2024) A novel parametric workflow for simulating urban heat island effects on residential building energy use: Coupling local climate zones with the urban weather generator a case study of seven US cities. Sustainable Cities Soc 110:105568\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePalme M, Inostroza L, Villacreses G, Lobato-Cordero A, Carrasco C (2017) From urban climate to energy consumption. Enhancing building performance simulation by including the urban heat island effect. Energy Build 145:107\u0026ndash;120\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePerera ATD, Javanroodi K, Mauree D, Nik VM, Florio P, Hong T, Chen D (2023) Challenges resulting from urban density and climate change for the EU energy transition. Nat Energy 8(4):397\u0026ndash;412\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNgarambe J, Nganyiyimana J, Kim I, Santamouris M, Yun GY (2020) Synergies between urban heat island and heat waves in Seoul: The role of wind speed and land use characteristics. PLoS ONE 15(12):e0243571\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHong T, Xu Y, Sun K, Zhang W, Luo X, Hooper B (2021) Urban microclimate and its impact on building performance: A case study of San Francisco. Urban Clim 38:100871\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMeng F, Zhang L, Ren G, Zhang R (2023) Impacts of UHI on variations in cooling loads in buildings during heatwaves: A case study of Beijing and Tianjin, China. Energy 273:127189\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang X, Yao L, Li M, Cao J, Zhong Q, Peng W, Wu W, Zhou J (2025) How heat waves and urban microclimates affect building cooling energy demand? Insights from fifteen eastern Chinese cities. Appl Energy 384:125424\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHirano Y, Fujita T (2012) Evaluation of the impact of the urban heat island on residential and commercial energy consumption in Tokyo. Energy 37(1):371\u0026ndash;383\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKolokotroni M, Zhang Y, Watkins R (2007) The London Heat Island and building cooling design. Sol Energy 81(1):102\u0026ndash;110\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi W, Zhou Z, Wang C, Han Y (2024) Impact of outdoor microclimate on the performance of high-rise multi-family dwellings in cold areas and optimization of building passive design. Build Environ 248:111038\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHui Y, Tang Y, Yang Q, Mochida A (2024) Numerical study on influence of surface vegetation on aerodynamics of high-rise buildings. Sustainable Cities Soc 107:105407\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYuan C, Zhu R, Tong S, Mei S, Zhu W (2022) Impact of anthropogenic heat from air-conditioning on air temperature of naturally ventilated apartments at high-density tropical cities. Energy Build 268:112171\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang L, Liu Y, Cao Q (2024) Toward interdisciplinary integration of architecture and climatology, The Innovation Energy 1(4) 100054-1-100054-2\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBianchi C, Zhang L, Goldwasser D, Parker A, Horsey H (2020) Modeling occupancy-driven building loads for large and diversified building stocks through the use of parametric schedules. Appl Energy 276:115470\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNazarian N, Krayenhoff E, Bechtel B, Hondula D, Paolini R, Vanos J, Cheung T, Chow W, de Dear R, Jay O (2022) Integrated assessment of urban overheating impacts on human life, Earth's Future 10(8) e2022EF002682\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang C, Song J, Shi D, Reyna JL, Horsey H, Feron S, Zhou Y, Ouyang Z, Li Y, Jackson RB (2023) Impacts of climate change, population growth, and power sector decarbonization on urban building energy use. Nat Commun 14(1):1\u0026ndash;16\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLuo X, Vahmani P, Hong T, Jones A (2020) City-scale building anthropogenic heating during heat waves. Atmosphere 11(11):1206\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTehrani AA, Sobhaninia S, Nikookar N, Levinson R, Sailor DJ, Amaripadath D (2025) Data-driven approach to estimate urban heat island impacts on building energy consumption. Energy 316:134508\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSalvati A, Roura HC, Cecere C (2017) Assessing the urban heat island and its energy impact on residential buildings in Mediterranean climate: Barcelona case study. Energy Build 146:38\u0026ndash;54\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHong T, Chen Y, Luo X, Luo N, Lee SH (2020) Ten questions on urban building energy modeling. Build Environ 168:106508\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJie P, Su M, Gao N, Ye Y, Kuang X, Chen J, Li P, Grunewald J, Xie X, Shi X (2023) Impact of urban wind environment on urban building energy: A review of mechanisms and modeling. Build Environ 245:110947\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePan X, Xu Y, Hong T (2024) Surrogate modelling for urban building energy simulation based on the bidirectional long short-term memory model. J Build Perform Simul 1\u0026ndash;19\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHan JM, Ang YQ, Malkawi A, Samuelson HW (2021) Using recurrent neural networks for localized weather prediction with combined use of public airport data and on-site measurements. Build Environ 192:107601\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi K, Chen Y (2023) Characterizing the indicator-based, day-and-night, and climate-based variations in response of surface urban heat island during heat wave across global 561 cities. Sustainable Cities Soc 99:104877\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu L, Tong S, He W, Zhu W, Mei S, Cao K, Yuan C (2022) Better understanding on impact of microclimate information on building energy modelling performance for urban resilience. Sustainable Cities Soc 80:103775\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLipson MJ, Grimmond S, Best M, Abramowitz G, Coutts A, Tapper N, Baik JJ, Beyers M, Blunn L, Boussetta S (2024) Evaluation of 30 urban land surface models in the Urban-PLUMBER project: Phase 1 results. Q J R Meteorol Soc 150(758):126\u0026ndash;169\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNik VM, Perera A, Chen D (2021) Towards climate resilient urban energy systems: a review. Natl Sci Rev 8(3):nwaa134\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKhan HS, Paolini R, Caccetta P, Santamouris M (2022) On the combined impact of local, regional, and global climatic changes on the urban energy performance and indoor thermal comfort\u0026mdash;The energy potential of adaptation measures. Energy Build 267:112152\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang Y, Gu Q, Wei H, Liu H, Wang W, Wei S (2023) Transforming and validating urban microclimate data with multi-sourced microclimate datasets for building energy modelling at urban scale. Energy Build 295:113318\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDu S, Zhang X, Jin X, Zhou X, Shi X (2022) A review of multi-scale modelling, assessment, and improvement methods of the urban thermal and wind environment. Build Environ 213:108860\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFrayssinet L, Merlier L, Kuznik F, Hubert J-L, Milliez M, Roux J-J (2018) Modeling the heating and cooling energy demand of urban buildings at city scale. Renew Sustain Energy Rev 81:2318\u0026ndash;2327\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLauzet N, Rodler A, Musy M, Azam M-H, Guernouti S, Mauree D, Colinart T (2019) How building energy models take the local climate into account in an urban context\u0026ndash;A review. Renew Sustain Energy Rev 116:109390\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDing X, Zhao Y, Strebel D, Fan Y, Ge J, Carmeliet J (2024) A WRF-UCM-SOLWEIG framework for mapping thermal comfort and quantifying urban climate drivers: Advancing spatial and temporal resolutions at city scale. Sustainable Cities Soc 112:105628\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDorer V, Allegrini J, Orehounig K, Moonen P, Upadhyay G, Kampf J, Carmeliet J (2013) Modelling the urban microclimate and its impact on the energy demand of buildings and building clusters, Building Simulation IBPSA, 2013, pp. 3483\u0026ndash;3489\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu S, Kwok YT, Ren C (2023) Investigating the impact of urban microclimate on building thermal performance: A case study of dense urban areas in Hong Kong. Sustainable Cities Soc 94:104509\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHaddad S, Zhang W, Paolini R, Gao K, Altheeb M, Al Mogirah A, Bin Moammar A, Hong T, Khan A, Cartalis C (2024) Quantifying the energy impact of heat mitigation technologies at the urban scale. Nat Cities 1(1):62\u0026ndash;72\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAdilkhanova I, Santamouris M, Yun GY (2024) Green roofs save energy in cities and fight regional climate change. Nat Cities 1(3):238\u0026ndash;249\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCiancio V, Falasca S, Golasi I, Curci G, Coppi M, Salata F (2018) Influence of input climatic data on simulations of annual energy needs of a building: EnergyPlus and WRF modeling for a case study in Rome (Italy). Energies 11(10):2835\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTakane Y, Nakajima K, Kikegawa Y, Yamaguchi K Enhancing urban canopy building energy models through the integration of social big data: Improvement and application, International Association for Urban Climate (IAUC) Urban Climate News 89 17\u0026ndash;21\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu H, Zhao Y, Dajun Z, Duan Y, Xu X (2025) Exploring the typhoon intensity forecasting through integrating AI weather forecasting with regional numerical weather model. npj Clim Atmospheric Sci 8(1):38\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eErell E, Zhou B (2022) The effect of increasing surface cover vegetation on urban microclimate and energy demand for building heating and cooling. Build Environ 213:108867\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKrayenhoff ES, Jiang T, Christen A, Martilli A, Oke TR, Bailey BN, Nazarian N, Voogt JA, Giometto MG, Stastny A (2020) A multi-layer urban canopy meteorological model with trees (BEP-Tree): Street tree impacts on pedestrian-level climate. Urban Clim 32:100590\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDu R, Liu C-H, Li X-X (2024) A new method for detecting urban morphology effects on urban-scale air temperature and building energy consumption under mesoscale meteorological conditions. Urban Clim 53:101775\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBueno B, Norford L, Hidalgo J, Pigeon G (2013) The urban weather generator. J Build Perform Simul 6(4):269\u0026ndash;281\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCui P, Lu J, Wu Y, Tang J, Jiang J (2024) Effect of urban morphology on microclimate and building cluster energy consumption in cold regions of China. Sustainable Cities Soc 115:105838\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi R, Huang C, Xin W, Ye J, Zhang X, Qu R, Wang J, Yuan L, Yao J (2025) Data-driven optimization reveals the impact of Urban Heat Island effect on the retrofit potential of building envelopes. Build Environ 269:112367\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu S, Wang S, Li G, Zhou H, Meng C, Qin Y, He B-J (2024) Performance-based design of residential blocks for the co-benefits of building energy efficiency and outdoor thermal comfort improvement. Build Environ 264:111926\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBian C, Cheung KL, Chen X, Lee CC (2025) Integrating microclimate modelling with building energy simulation and solar photovoltaic potential estimation: The parametric analysis and optimization of urban design. Appl Energy 380:125062\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRedon EC, Lemonsu A, Masson V, Morille B, Musy M (2017) Implementation of street trees within the solar radiative exchange parameterization of TEB in SURFEX v8. 0. Geosci Model Dev 10(1):385\u0026ndash;411\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOleson K, Feddema J (2020) Parameterization and surface data improvements and new capabilities for the Community Land Model Urban (CLMU), Journal of Advances in Modeling Earth Systems 12(2) e2018MS001586\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKarimi A, Moreno-Rangel D, Garc\u0026iacute;a-Mart\u0026iacute;nez A (2025) Granular Mapping of UHI and Heatwave Effects: Implications for Building Performance and Urban Resilience, Building and Environment 112705\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDeng X, Cao Q, Wang L, Wang W, Wang S, Wang S, Wang L (2023) Characterizing urban densification and quantifying its effects on urban thermal environments and human thermal comfort. Landsc Urban Plann 237:104803\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBueno B, Pigeon G, Norford L, Zibouche K, Marchadier C (2012) Development and evaluation of a building energy model integrated in the TEB scheme. Geosci Model Dev 5(2):433\u0026ndash;448\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSimpson CH, Brousse O, Taylor T, Grellier J, Taylor J, Fleming LE, Davies M, Heaviside C (2024) Modeled temperature, mortality impact and external benefits of cool roofs and rooftop photovoltaics in London. Nat Cities 1(11):751\u0026ndash;759\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTan H, Kotamarthi R, Wang J, Qian Y, Chakraborty T (2023) Impact of different roofing mitigation strategies on near-surface temperature and energy consumption over the Chicago metropolitan area during a heatwave event. Sci Total Environ 860:160508\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTarasova M, Varentsov M, Stepanenko V (2023) Parameterization of the Interaction between the Atmosphere and the Urban Surface: Current State and Prospects. Izv Atmospheric Ocean Phys 59(2):111\u0026ndash;130\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu X, Chen F, Shen S, Miao S, Barlage M, Guo W, Mahalov A (2018) Using WRF-urban to assess summertime air conditioning electric loads and their impacts on urban weather in Beijing. J Geophys Research: Atmos 123(5):2475\u0026ndash;2490\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePappaccogli G, Giovannini L, Zardi D, Martilli A (2020) Sensitivity analysis of urban microclimatic conditions and building energy consumption on urban parameters by means of idealized numerical simulations. Urban Clim 34:100677\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePerera A, Coccolo S, Scartezzini J-L, Mauree D (2018) Quantifying the impact of urban climate by extending the boundaries of urban energy system modeling. Appl Energy 222:847\u0026ndash;860\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi XC, Zhao L, Qin Y, Oleson K, Zhang Y (2024) Elevated urban energy risks due to climate-driven biophysical feedbacks. Nat Clim Change 14(10):1056\u0026ndash;1063\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu J, Gao H, Jia R, Wang R, Han D, Liu L, Xu X, Qiao Z (2025) A downscaling framework with WRF-UCM and LES/RANS models for urban microclimate simulation strategy: Validation through both measurement and mechanism model. Build Environ 269:112361\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMei S-J, Yuan C (2021) Analytical and numerical study on transient urban street air warming induced by anthropogenic heat emission. Energy Build 231:110613\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang M, Gao Z (2021) Effect of urban form on microclimate and energy loads: Case study of generic residential district prototypes in Nanjing, China. Sustainable Cities Soc 70:102930\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMangan SD, Oral GK, Sozen I, Kocagil IE (2020) Evaluation of settlement textures in terms of building energy, economic performance, and outdoor thermal comfort. Sustainable Cities Soc 56:102110\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTian X, Zhang H, Liu L, Huang J, Liu L, Liu J (2024) Establishment of LCZ-based urban building energy consumption dataset in hot and humid subtropical regions through a bottom-up method. Appl Energy 368:123491\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXi C, Wang D, Cao S-J (2023) Impacts of trees-grass area ratio on thermal environment, energy saving, and carbon benefits. Urban Clim 47:101393\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAllegrini J, Dorer V, Carmeliet J (2012) Influence of the urban microclimate in street canyons on the energy demand for space cooling and heating of buildings. Energy Build 55:823\u0026ndash;832\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMiller C, Thomas D, K\u0026auml;mpf J, Schlueter A (2018) Urban and building multiscale co-simulation: case study implementations on two university campuses. J Build Perform Simul 11(3):309\u0026ndash;321\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBouyer J, Inard C, Musy M (2011) Microclimatic coupling as a solution to improve building energy simulation in an urban context. Energy Build 43(7):1549\u0026ndash;1559\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGros A, Bozonnet E, Inard C, Musy M (2016) Simulation tools to assess microclimate and building energy\u0026ndash;A case study on the design of a new district. Energy Build 114:112\u0026ndash;122\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu L, Du Z, Wang S, Liu J, Hang J, Yang W (2025) Modeling the impact of heat rejection from split air conditioner on outdoor air temperature in high-density residential areas. Build Environ 112685\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang X, Yao L, Peng LL (2024) Impacts of urban air temperature and humidity on building cooling and heating energy demand in 15 cities of eastern China. Energy 288:129887\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang X, Peng LL, Jiang Z, Chen Y, Yao L, He Y, Xu T (2020) Impact of urban heat island on energy demand in buildings: Local climate zones in Nanjing. Appl Energy 260:114279\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTian L, Hao T, He X, Chan I, Niu J, Chan P, Ng W, Huang J (2024) Examining the non-linear relationship between urban form and air temperature at street level: A case of Hong Kong. Build Environ 264:111884\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXiang Y, Zheng B, Bedra KB, Ouyang Q, Liu J, Zheng J (2023) Spatial and seasonal differences between near surface air temperature and land surface temperature for Urban Heat Island effect assessment. Urban Clim 52:101745\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVenter ZS, Brousse O, Esau I, Meier F (2020) Hyperlocal mapping of urban air temperature using remote sensing and crowdsourced weather data. Remote Sens Environ 242:111791\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYin H, Zhao X (2024) Urban heat island analysis based on high resolution measurement data: A case study in Beijing. Sustainable Cities Soc 106:105389\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHan J, Chong A, Lim J, Ramasamy S, Wong NH, Biljecki F (2024) Microclimate spatio-temporal prediction using deep learning and land use data. Build Environ 253:111358\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen G, Hua J, Shi Y, Ren C (2023) Constructing air temperature and relative humidity-based hourly thermal comfort dataset for a high-density city using machine learning. Urban Clim 47:101400\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDing X, Zhao Y, Fan Y, Li Y, Ge J (2023) Machine learning-assisted mapping of city-scale air temperature: Using sparse meteorological data for urban climate modeling and adaptation. Build Environ 234:110211\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBenali A, Carvalho A, Nunes J, Carvalhais N, Santos A (2012) Estimating air surface temperature in Portugal using MODIS LST data. Remote Sens Environ 124:108\u0026ndash;121\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eColaninno N, Morello E (2022) Towards an operational model for estimating day and night instantaneous near-surface air temperature for urban heat island studies: outline and assessment. Urban Clim 46:101320\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen F, Li X, Wang Y (2025) A knowledge-augmented deep fusion method for estimating near-surface air temperature. Remote Sens Environ 326:114819\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAdrian CZM, Hien WN, Marcel I, Kardinal JS (2013) Predicting the envelope performance of commercial office buildings in Singapore. Energy Build 66:66\u0026ndash;76\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIgnatius M, Wong NH, Jusuf SK (2016) The significance of using local predicted temperature for cooling load simulation in the tropics. Energy Build 118:57\u0026ndash;69\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHsiao Y-C (2024) Hourly temperature downscaling method based on clustering and linear transformation: Utilizing mean, maximum, and minimum temperatures. Energy Build 325:114975\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYe J, Huang C, Zhong Z, Shen Y, Ao X, Su Y, Cao J, Duan H, Yao J (2025) Improving the accuracy of microclimate coupled urban building energy modeling using convolutional neural networks. Build Environ 277:112923\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSampaio RJ, Rodriguez DA, Esp\u0026iacute;ndola RP, da Silva FP (2025) Spatiotemporal Assessment and Monitoring of Urban Heat Islands in Metropolitan Areas Using Machine Learning and Downscaling. Sustainable Cities Soc 106365\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi W, Zhou Z, Zhang S, Feng J (2022) Study on relationship between shading and outdoor air temperature based on the comparison of two high-rise residential estates with field measurements. Buildings 12(11):1813\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLindberg F, Sun T, Grimmond S, Tang Y (2023) UMEP Manual Documentation\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBueno B, Nakano A, Norford L Urban weather generator: a method to predict neighborhood-specific urban temperatures for use in building energy simulations, ICUC9-9th International Conference on Urban Climate Jointly with 12th Symposium on the Urban Environment, 2015\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu Y, Chu C, Zhang R, Chen S, Xu C, Zhao D, Meng C, Ju M, Cao Z (2024) Impacts of high-albedo urban surfaces on outdoor thermal environment across morphological contexts: A case of Tianjin, China. Sustainable Cities Soc 100:105038\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHousing Mo (2021) U.-R.D.o.t.P.s.R.o. China, General Code for Energy Efficiency and Renewable Energy Application in Buildings (GB 55015\u0026thinsp;\u0026ndash;\u0026thinsp;2021). China Architecture and Building Press Beijing, China\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eASHRAE A (2020) Standard 169\u0026ndash;2020. Climatic Data for Building Standards, Atlanta, GA\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHousing Mo (2016) U.-R.D.o.t.P.s.R.o. China, Standard for energy consumption of building (GB 51161\u0026thinsp;\u0026ndash;\u0026thinsp;2016)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHousing Mo (2014) U.-R.D.o.t.P.s.R.o. China, Standard for weather data of building energy efficiency\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang S, Yan S, Zhang D, Wan S, Lv H, Wang L (2024) A real-time indoor temperature and relative humidity prediction method for exhibition hall based on Long Short-Term Memory network. J Building Eng 98:111492\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHochreiter S, Schmidhuber J (1997) Long short-term memory. Neural Comput 9(8):1735\u0026ndash;1780\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLundberg SM, Lee S-I (2017) A unified approach to interpreting model predictions. Adv Neural Inf Process Syst 30\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang Y, Lv Y, Zhou D (2025) The impact of urban parks on the thermal environment of built-up areas and an optimization method. PLoS ONE 20(3):e0318633\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eThierry J, Herpin S, Levi R, Canonne D, Demotes-Mainard S, Cannavo P, Lemesle D, Brialix L, Rodriguez F, Bournet P (2024) Impact of a water restriction on the summer climatic benefits of trees inside an outdoor street canyon scale model. Build Environ 261:111722\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSosa MB, Correa EN, Cant\u0026oacute;n MA (2017) Urban grid forms as a strategy for reducing heat island effects in arid cities. Sustainable Cities Soc 32:547\u0026ndash;556\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang C, Wang Z-H, Wang C, Myint SW (2019) Environmental cooling provided by urban trees under extreme heat and cold waves in US cities. Remote Sens Environ 227:28\u0026ndash;43\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang S, Zhan W, Zhou B, Tong S, Chakraborty T, Wang Z, Huang K, Du H, Middel A, Li J (2025) Dual impact of global urban overheating on mortality. Nat Clim Change 1\u0026ndash;8\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBarbour E, Davila CC, Gupta S, Reinhart C, Kaur J, Gonz\u0026aacute;lez MC (2019) Planning for sustainable cities by estimating building occupancy with mobile phones. Nat Commun 10(1):3736\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMirsadeghi M, Costola D, Blocken B, Hensen JL (2013) Review of external convective heat transfer coefficient models in building energy simulation programs: Implementation and uncertainty. Appl Therm Eng 56(1\u0026ndash;2):134\u0026ndash;151\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMa K, Yu Y, Yang B, Yang J (2019) Demand-side energy management considering price oscillations for residential building heating and ventilation systems. IEEE Trans Industr Inf 15(8):4742\u0026ndash;4752\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen Y, Sha A, Jiang W, Lu Q, Du P, Hu K, Li C (2025) Eco-friendly bismuth vanadate/iron oxide yellow composite heat-reflective coating for sustainable pavement: Urban heat island mitigation. Constr Build Mater 470:140645\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang X, Bou-Zeid E, Pigliautile I, Pisello AL, Mandal J (2024) Optimizing retro-reflective surfaces to untrap radiation and cool cities. Nat Cities 1(4):275\u0026ndash;285\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi Y, Li H, Miao R, Qi H, Zhang Y (2023) Energy\u0026ndash;Environment\u0026ndash;Economy (3E) Analysis of the Performance of Introducing Photovoltaic and Energy Storage Systems into Residential Buildings: A Case Study in Shenzhen, China. Sustainability 15(11):9007\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRahman MN (2024) Seasonal and annual trends in reference evapotranspiration and prediction using machine learning models across seven climatic zones of Bangladesh, Geology, Ecology, and Landscapes 1\u0026ndash;16\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMeili N, Zheng X, Takane Y, Nakajima K, Yamaguchi K, Chi D, Zhu Y, Wang J, Qiu Y, Paschalis A (2025) Modeling the effect of trees on energy demand for indoor cooling and dehumidification across cities and climates. J Adv Model Earth Syst 17(3) e2024MS004590.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Tsinghua University","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Localized weather data, Extreme weather, Urban heat island, UWG, Energy use intensity, Hourly load","lastPublishedDoi":"10.21203/rs.3.rs-7763869/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7763869/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDue to climate change, extreme weather (EW) events like heatwaves and cold snaps are becoming more frequent, challenging urban buildings and energy systems. Urban heat island (UHI) effects\u0026mdash;where city centers are significantly warmer than suburbs at night\u0026mdash;further impact heating and cooling demands of urban buildings. However, there is still a lack of systematic empirical studies linking meteorological data to building energy use, especially regarding the compound effects of UHI and extreme weather on urban building energy consumption and load during winter and summer. To address future complex climate conditions, we propose using localized weather data (LWD) that fully accounts for both background EW and UHI effects. Driven by suburban meteorological observations and high-resolution land cover data, the data is generated using the UWG urban canopy model and the UMEP tool on the QGIS platform to capture realistic local weather conditions around buildings. It can be directly input into urban building energy model (UBEM) for the corresponding local climate zones to simulate building energy use. Our study shows that LWD better captures seasonal building energy use and the effects of external and internal factors. Compared to suburban weather station data, accuracy improves by 29.6% in summer and 36.6% in winter during the typical year, and by 35.1% and 30.1% during the extreme weather year, respectively. Local air temperature (Ta) has the greatest impact on actual energy use, followed by solar radiation (Rad)\u0026mdash;especially during summer heatwaves, when Rad may exceed Ta in influence. Internal disturbances have a greater impact in summer, but their influence lessens during extreme weather due to stronger external climatic effects. This method supports refined assessment and control of UBEM across climates and seasons, helping manage energy peaks during heatwaves and prevent overheating in winter, ultimately aiding real-weather-based energy system optimization and urban design.\u003c/p\u003e","manuscriptTitle":"An empirical study on the compound effects of extreme weather and UHIon building energy consumption under local climate","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-03 06:33:45","doi":"10.21203/rs.3.rs-7763869/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"287139ee-2d1d-4c50-bd35-4c74f516ad8b","owner":[],"postedDate":"October 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":55658808,"name":"Civil Engineering"}],"tags":[],"updatedAt":"2025-10-03T06:33:45+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-03 06:33:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7763869","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7763869","identity":"rs-7763869","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}