Evaluating the Future Effectiveness of Cooling Strategies in Subtropical Terrace Houses Under Climate Change

Abstract

Due to increasing overheating risk from the global warming, it is crucial to evaluate building performance, especially that of a subtropical terrace house with limited façade exposure, to ensure its future livability. The effectiveness of current cooling strategies has not yet been thoroughly analyzed by considering specific zone position in a terrace house and varying future weather scenarios in existing studies. This study quantifies livability of six typical zones in two types of terrace houses employing two cooling strategies during summer under historical and future weather scenarios. It considers two key outputs: annual heat discomfort hours and cooling Energy Use Intensity (EUI). Additionally, a global sensitivity analysis using Standard Regression Coefficients (SRCs) identifies the key variables. The results indicate that the future annual cooling EUI is projected to increase by around 150 kWh/m², with discomfort hours reaching 80% during summertime. Furthermore, window-to-wall ratio, total length, and solar heat gain coefficients (SHGCs) of windows significantly reduce cooling EUI by 50–165 kWh/m². However, most input variables have a minimal impact on heat discomfort hours, resulting only in a 0.7–7.2% reduction. This study provides an analytical framework for assessing the future livability and sustainability of subtropical terrace houses, facilitating the development of robust strategies during the early design stage.

1. Introduction

The impact of climate change on human well-being has become a global concern over the past few decades. One significant issue is the increasing intensity and frequency of heat waves, which lead to extended periods of high outdoor temperatures [1]. Heat waves have resulted in substantial public health challenges. The threats posed by heatwaves to humans include not only physical issues but also psychological issues. Chique et al. [2] found the rate of heat exposure leading to depression is 24%, while it accounts for approximately 17% for both Post-Traumatic Stress Disorder (PTSD) and anxiety, by investigating the correlation between extreme weather events and psychological health. However, a previous study indicates that people have a limited perception of overheating risk due to a lack of knowledge about heat-related issues, and the perceived heat-related psychological impacts are more severe than the physiological impacts [3].

Given that people spend 90% of their time indoors, a livable and healthy indoor environment is essential for occupants. The indoor thermal environment is one of the critical factors influencing residents’ well-being [4]. Therefore, existing research evaluates the impact of current and projected future heatwaves on building thermal performance to provide insights for enhancing thermal performance. For instance, a study predicts that, under future Representative Concentration Pathway (RCP) 8.5 scenarios, subtropical urban village houses will experience 4325 overheating hours and an overheating degree of 10.9 °C [5]. Additionally, the worst simulation case in San Pedro Sula projects that 97% of hours will be overheated, with an overheating degree of 3.7 °C, under future 2050 A2 scenarios (The A scenarios, such as A1 and A2, emphasize economic growth and material wealth, with varying global cooperation and technological progress) [6]. Ashrafian et al. [7] analyzed the impact derived from future climate change on the energy, cost, and thermal comfort of a Turkish school building. Habitzreuter et al. [8] assessed overheating risk in a high-rise residential building in London by considering the impact from urban context and heatwaves, which indicates exterior shading contributes to reducing overheating occurrence by 74%.

Additionally, as the building sector accounts for 36% of final energy use, with heating, ventilation, and air conditioning comprising 50% of building energy consumption in developed countries [9], heat waves are expected to drive up energy consumption due to increased cooling demand [10]. A study on the influence of urban heat island on building cooling load indicated the cooling load of buildings increased significantly during heat waves, and for every 1 °C rise in Urban Heat Island Intensity (UHII), the hourly cooling load of urban residential or office buildings increases by 0.5 × 10³ kWh/m² [11]. Building energy consumption in China is anticipated to rise by 80% compared to the current situation under the 13th Five-Year Plan strategy and future scenarios [12]. Heatwaves not only affect the indoor thermal environment but also significantly impact building energy systems. It is crucial to improve indoor thermal comfort and enhance building energy efficiency.

1.1. Terrace House

Terrace houses, a building type characterized by shorter width and longer length, are typically joined by one or two adjacent buildings. This construction style saves space and simplifies planning, making it common in various climate regions worldwide [13]. In particular, a large number of terrace houses have been built in the subtropical region of China [14]. However, in contrast to high-rise residential buildings, terrace houses are more likely to face increased overheating risks and higher cooling energy consumption, primarily due to their constrained interface area for heat exchange with outdoor air. For example, Zou et al. [15] found that rowhouses, which are the same type as terrace houses, experienced high overheating hours, accounting for 69.1% in 2016, and a high overheating degree of 1.48 °C in 2010 among recent years. Furthermore, Wu et al. [16] analyzed the effectiveness of passive cooling strategies by investigating various natural ventilation operations in a corner terraced house in Kuala Lumpur. Bugenings et al. [17] investigated the overheating risk in six typical Danish residential buildings with various adiabatic façade configurations adjacent to other buildings, considering both current and future climate scenarios. As indicated by these studies, variations in thermal performance among different types of houses may be attributed to differences in envelope thermal materials or the area exposed to the sun.

However, current analyses of the thermal performance of independent houses often consider the entire building or several predominantly used rooms as the evaluation objects, rather than each individual room or zone [18]. Different zones in various terrace houses along a street may receive varying amounts of solar radiation. This variation is likely to result in some zones being less adaptable to mitigate overheating risk and reduce energy consumption with retrofit strategies based on the entire house. To establish an analysis framework for assessing thermal performance tailored to specific zones within terrace houses, it is essential to consider the unique features of each zone based on their different positions within the house.

1.2. Performance Evaluation

Building performance evaluation contributes to analyzing building thermal performance in response to uncertain climate changes and provides valuable references for retrofit decision-making. There are three widely used methods to assess the building performance, including survey questionnaire, on-site measurement, and performance simulation. Since simulation methods can contribute to projecting building performance under future climate conditions and facilitate comparisons of the pros and cons of different passive strategies at the early design stage, they are widely employed in existing research. For example, Berardi et al. [10] used OpenStudio 2.8.0 to simulate the heating and cooling demand of 16 building prototypes in the city of Toronto under climate change with the CCWorldWeatherGen tool for generating future climate data. Nurlybekova et al. [19] evaluated the thermal and energy performance of different Phase Change Material (PCM)-integrated buildings located in a monsoon-influenced humid subtropical climate zone considering current and future climate scenarios. Furthermore, Sobhy et al. [20] conducted a simulation analysis to evaluate the effectiveness of nighttime ventilation in residential buildings located in hot and arid climate zones under current and future climate scenarios.

Nevertheless, different retrofit strategies show varying degrees of optimization in zones with different positions within a building. For instance, a roof with lower U-value material may enhance the thermal resistance of rooftop zones against hot weather [21], while walls with low U-value materials or a lower window-to-wall ratio are likely to maintain thermal comfort on the middle floor [22]. The lack of a comprehensive multi-zone analysis is likely to introduce bias into the assessment results, affecting the judgment of retrofitting strategies. Previous studies have utilized sensitivity analysis methods to identify significant factors influencing building performance, including energy efficiency and thermal comfort [23]. For instance, Wu et al. [24] used two global sensitivity analysis methods, Standard Regression Coefficient (SRC) and Targeted Global Perturbation (TGP), to quantify the impact of uncertain simulation inputs on thermal discomfort hours and energy consumption in rural houses. In addition, Shen et al. [25] analyzed the significance of different building parameters on commercial buildings’ energy efficiency and adequate structural behavior through conducting a global sensitivity analysis method.

Therefore, a sensitivity analysis of zones located in various positions within a building should be conducted to evaluate their performance under future climate conditions and offer insights for further optimization studies.

1.3. Aims and Contributions

Due to the high frequency of heatwaves posing threats to residents, existing research focuses on indoor overheating risks by evaluating building energy and thermal performance, particularly in terraced housing, which has limited façades for ventilation. In hot and humid regions, prolonged periods of extreme heat are expected to become more severe in the future compared to other climates. However, there are limited studies evaluating the effectiveness of cooling strategies in current subtropical terraced housing under both current and future weather conditions. To protect residents from unsafe living conditions with high overheating potential and to reduce building energy consumption for environmental sustainability, it is essential to evaluate the performance of specific zones in terraced houses located in subtropical regions characterized by high temperatures and relative humidity.

This study aims to identify the key parameters influencing thermal comfort and cooling energy in each type of zone within terrace houses in the subtropical region of China, providing a reference for further retrofit strategy analysis. This analysis contributes to proposing a framework for architects and engineers to make better decisions on retrofit strategies or new designs for subtropical terrace houses, aiming to reduce thermal discomfort hours and energy consumption.

To comprehensively present the evaluation framework, this research is organized into five sections. Section 1 introduces the background of severe heatwaves and the potential overheating risks in terraced houses within the subtropical regions of China. Section 2 describes the materials and methods used to evaluate cooling strategies, including the chosen site and building, input parameters, a simulation workflow, a sensitivity analysis method, and evaluation criteria. Section 3 presents the effectiveness of cooling strategies in three parts: an assessment of current and projected overheating risks with baseline strategies; a ranking of the most influential parameters for overheating reduction; and a comparison of overheating outcomes between baseline and optimized strategies. Section 4 discusses the limitations and future directions of this research, and Section 5 summarizes the main findings of the evaluation.

2. Materials and Methods

The workflow for evaluating energy and comfort of multi-zone terrace houses located in a subtropical region involves four steps, as illustrated in Figure 1.

First, 11 input variables influencing building thermal performance were collected, including building geometry (width, length, window-to-wall ratio), orientation, overhang depth, distance from neighboring buildings, and thermal-related variables (U-value of walls, window glazing, roof, solar heat gain coefficient (SHGC), and visual transmittance of window glazing). A set of baseline parameters was selected based on the Chinese design standard for energy efficiency in rural residential buildings (GB/T 50824-2013) [26]. Additionally, since terraced houses have one or two walls adjacent to neighboring buildings, we considered two building types: the middle type, with only the front and back walls exposed to the air, and the side type, with one side wall, the front wall, and the back wall exposed. In this study, 6 zone types consisting of two building types and three floors were selected. To analyze the impact of these variables on these zones under current and future weather conditions, Typical Meteorological Year (TMY) data and Representative Concentration Pathways (RCPs) 2.6, 4.5, and 8.5 data spanning 2030 to 2090 were used in the simulation.

Following this, since Latin Hypercube Sampling (LHS) is a statistical technique that efficiently samples the input space by evenly distributing sampling points, making it ideal for analyzing how various parameters influence complex systems like building energy performance, we generated 200 simulation samples from 11 variables using the Design Space Exploration (DSE) plugin with the LHS function. In addition to the input samples, six zone types and ten weather datasets were included, resulting in a total of 12,000 simulation cases in this evaluation. Subsequently, we used the Ladybug and Honeybee tools, along with OpenStudio version 2.9.1, to simulate the study cases. Two assessment criteria were applied to evaluate building energy efficiency and comfort level, Cooling Energy Use Intensity (EUI) and hours of heat discomfort, as cooling solutions are of primary concern locally due to consistently high temperatures throughout the year.

To identify the influential variables affecting energy and comfort levels in terraced housing, sensitivity analysis using Standard Regression Coefficients (SRCs) was conducted. SRCs are effective for identifying influential variables in building energy and comfort analysis, as it quantifies the relative impact of each input variable on energy reduction and comfort improvement within a regression model framework. Before applying the SRC method, we tested the normal distribution of the cooling EUI output model and the comfort output model by considering all simulation inputs and outputs to ensure the reliability and effectiveness of the SRC method. Subsequently, 11 variables were ranked based on SRC values, taking into account zone types and weather conditions.

Finally, the top five influential variables for reducing energy usage and discomfort hours were identified based on the ranking results. Additionally, optimized parameters for each influential variable were selected based on the four rules introduced in Section 3.4. The effectiveness of the cooling strategy was then evaluated by comparing overheating results between the baseline strategy and the retrofit strategy.

2.1. Case Study

Given that Guangzhou, China, is characterized by long hot periods and short winters [27,28], a typical terrace houses in Guangzhou was selected as the study object. Terrace houses, a type of terraced housing attached to one or two adjacent terrace houses, feature interior zones with different positions that may face varying thermal discomfort risks under climate change. We selected two typical terrace houses located in Liwan District, Guangzhou by considering side type and middle type. The site is shown in Figure 2, and the building plan is presented in Figure 3. Side-type houses are typically located at the corner or end of a street with a roof and two walls exposed to the air, and middle-type houses are situated in the middle of a street with a roof and only the front wall exposed to the air. Given that terrace houses are typically limited to two to four floors due to their construction frame, we selected three representative floors within the side-type and middle-type houses (the ground floor, middle floor, and rooftop floor) to cover most floor positions. Thus, six zone types were considered in this study. The selection of zone types is shown in Figure 4.

Previous research indicates that modifying the thermal materials of envelope and adjusting the window-to-wall ratio (WWR) positively impacts reducing overheating hours in buildings by comparing the performance of different passive strategies [29,30]. However, retrofit strategies developed for a whole house may be inadequate for improving the thermal performance of zones in different positions within a terrace house. To analyze the impact of commonly analyzed variables on the energy and thermal performance of each zone in a terrace house, we considered exterior and interior variables. Exterior variables include width, length, orientation, overhang depth, and the distance between the selected zone and its opposite zone. Interior variables include the U-value of the roof, walls, and windows, as well as the solar heat gain coefficient (SHGC) and visual transmittance (VT) of the windows. All these variables with baseline value were input into simulation models constructed on the Grasshopper platform using Ladybug and Honeybee tools, and are shown in

Table 1. The input variables of each zone in the terrace house.

Variable	Based Value	Unit	Description	Range	Interval
Width	4	m	-	$x_1\in [4,6]$	1
Length	10	m	-	$x_2\in [10,12]$	1
Orientation	0	-	South	$x_3\in \left[0,3\right]$ 0 represents south, 1 denotes west, 2 represents north, 3 denotes east.	1
Overhang depth	0.8	m	Single vertical overhang	$x_4\in [0.0,1.0]$	0.1
Shade distance	4	m	Distance between the evaluated zone and its opposite zone	$x_5\in [2,10]$	1
Window-to-wall ratio	0.3	-	The area of window occupying in a whole wall	$x_6\in [0.1,0.4]$	0.1
Wall U-value	1.5	W/m2·K	The U-value of the front wall or the side wall, which is built based on a brick-concrete structure frame	$x_7\in [1.0,2.0]$	0.1
Roof U-value	1	W/m2·K	The U-value of the roof built based on concrete	$x_8\in [0.5,1.5]$	0.1
Window U-value	5	W/m2·K	The U-value of the window made of single-layer flat transparent glass	$x_9\in [1,5]$	1
Window solar heat gain coefficient (SHGC)	0.8	W/m2·K	The SHGC of the window made of single-layer flat transparent glass	$x_{10}\in [0.4,0.8]$	0.1

.

Since annual cooling energy consumption and indoor thermal comfort were used as evaluation metrics for building performance in this study, two building operation modes were established within the platform: the cooling device operation and the natural ventilation operation. These modes represent two scenarios: cooling with active devices and natural ventilation through open windows. For the scenario of cooling with active devices, an ideal air loads system was utilized by employing the “HB_Assign HVAC System (HVAC System)” component on the modeling platform, with a cooling setpoint of 26 °C. Furthermore, three schedules for lighting, equipment, and occupancy, which are recommended by the General Code for Energy Efficiency and Renewable Energy Application in Buildings (GB 55015-2021) [31], were used in this study and are shown in Figure 5. For the scenario of cooling with opening windows, natural ventilation was facilitated by a wind-driven calculation model. The window opening schedule aligned with occupancy patterns. Given that most windows in the subtropical terrace houses are casement windows, the operable fraction of glazing area was set at 80%.

2.2. Weather Datasets

The outdoor climate conditions, such as outdoor air temperature, relative humidity, and solar radiation, play a key role in the performance evaluation. Theses meteorological parameters significantly affect the indoor thermal environment, influencing occupants’ thermal sensations. To comprehensively analyze the climate-adaptive strategy performance, current and future weather were considered in this study.

For current climate, TMY weather dataset was derived from the period of 1994–2003 and developed from the Chinese Standard Weather Dataset. Additionally, Representative Concentration Pathways (RCPs) are the scenarios projecting the potential future concentration of greenhouse gases in the atmosphere and their associated radiative forcing [32]. Four RCP scenarios, RCP2.6, RCP4.5, RCP6.0, and RCP8.5, are indicated in the Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC). They represent projected scenarios where radiative forcing reaches 2.6 W/m², 4.5 W/m², 6.0 W/m², and 8.5 W/m² by 2100, respectively.

This study considers the RCP2.6, RCP4.5, and RCP8.5 scenarios and three years for each scenario, including 2030 (early stage), 2050 (mid stage), and 2090 (late stage), for the climate-adaptive strategy performance evaluation. Hence, we employed Meteonorm software version 8.0 to generate the future weather datasets, which were developed by using 10 General Circumstance Models (GCMs) posed on AR5 of IPCC with a statistical downscaling method. The annual air temperature, relative humidity, and solar radiation under TMY and the three future RCP scenarios between 2010 and 2019 are presented in

Table 2. The annual meteorological parameter from 10 weather datasets.

Scenario	Year	Mean Air Temperature (°C)	Max. Air Tem-perature (°C)	Mean Relative Humidity (%)	Max. Relative Humidity (%)	Mean Solar Radiation (W/m2)	Max. Solar Radiation (W/m2)
TMY	-	22.23	36.60	76.89	100	67.19	797
RCP2.6	2030	23.73	38.50	71.1	100	73.30	901
	2050	24.27	38.60	70.98	100	75.69	878
	2090	24.37	38.30	71.05	100	81.85	818
RCP4.5	2030	23.70	38.70	71.10	100	68.50	853
	2050	24.39	38.90	71.22	100	73.92	846
	2090	25.56	39.60	70.85	100	76.33	895
RCP8.5	2030	23.69	38.40	71.24	100	66.90	826
	2050	24.96	39.40	70.88	100	70.65	888
	2090	27.39	41.50	71.09	100	78.57	848

.

2.3. Assessment Criterion

Thermal design for terrace houses in the subtropical region of China predominantly emphasizes cooling strategies, given the extended hot periods throughout the year compared to other seasons [33]. Existing research on building energy efficiency in these regions primarily focuses on annual cooling demand [34]. Studies on thermal comfort typically consider Predicted Mean Vote (PMV) based on steady-state environments controlled by HVAC systems [35], adaptive thermal comfort models, or modified PMV models adjusted for dynamic environments influenced by occupants’ adaptive behaviors [36].

Terrace houses typically accommodate two types of occupants: low-income tenants and middle-aged or elderly couples who have inherited their homes. As a result, some terrace houses lack installed cooling devices. In these cases, residents rely on strategies like window-opening for cooling or adjusting clothing levels. Conversely, other terrace houses have been renovated to include active cooling devices.

To evaluate the impact of various climate-adaptive strategies on terrace houses’ energy and thermal performance, two metrics are considered for each scenario: annual cooling demand assesses energy performance in houses with active cooling devices, while annual heat discomfort hours evaluate thermal performance in houses without active devices, relying instead on natural ventilation.

2.3.1. Energy Use Intensity

To calculate building energy consumption, previous research employed Energy Use Intensity (EUI) as the evaluation criterion, measured in kWh/m²/year [37]. Typically, total energy consumed by a building over one year (including fuel, gas, electricity, or other energy sources) is converted into kWh, then divided by the building’s area to measure EUI [38]. However, given that Guangzhou is a cooling-dominated region, cooling energy consumption constitutes a significant portion of total energy use in the building sector [39]. Therefore, our study primarily focuses on cooling energy usage. For scenarios involving active cooling devices in terrace houses, we calculated the total cooling energy used per zone in an entire terrace houses per year, measured in kWh/m²/year. The annual cooling demand was determined using the Ladybug (0.0.69 version) and Honeybee (0.0.66 version) simulation tool.

2.3.2. aPMV

Predict Mean Vote (PMV), an assessment metric developed by P.O. Fanger [40], is widely used to predict the average thermal comfort of people in a steady-state condition. This metric ranges from −3 to 3, with 0 representing neutral thermal sensation based on the American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) thermal sensation scale. However, some studies indicate that the PMV index may be inaccurate for evaluating thermal comfort in naturally ventilated buildings due to its lack of considering some factors of a building’s climate adaptation [41], such as the adaptive behavior of occupants. To increase the credibility of thermal comfort evaluation in naturally ventilated buildings, researchers make efforts in establishing adaptive thermal comfort metrics. Modified PMV indices for free-running buildings have been developed, such as adaptive Predict Mean Vote (aPMV). The aPMV index was developed by Yao et al. [42] through field measurements and questionnaires conducted in five climate zones of China, and it considers behavioral, psychological, and physiological adaptation with developing adaptive coefficients (λ) for different seasons at various climate zones in China. This coefficient derives the theoretical relationship between aPMV and PMV. The equation is shown below:

$$\alpha PMV={\displaystyle \frac{PMV}{1+\lambda \times PMV}}$$

(1)

where PMV is an index representing individuals’ responses on a 7-point thermal sensation scale. λ represents the adaptive coefficient. The value of the coefficient varies depends on the season and the climate zones due to the different climate adaptive habits and expectations of people living in different climate zones. The coefficient for specific climate zones in China is shown in

Table 3. The adaptive coefficient for different seasons and climate zones.

Climate Zone	Season	Value of λ
Serious cold, cold region	PMV ≥ 0	0.24
	PMV < 0	−0.50
Hot summer and cold winter, Hot summer and warm winter, Warm region	PMV ≥ 0	0.21
	PMV < 0	−0.49

. Since Guangzhou is located in a hot summer and warm winter climate zone of China, we considered the λ coefficient of 0.21 to evaluate the residents’ thermal comfort during summertime.

There are two comfort levels based on the aPMV index: Level 1, with a range between −0.5 and 0.5, and Level 2, with a range between −1 and 1. Considering the wider thermal acceptable range for residents living in Guangzhou, Level 2 with the wider range was considered in this study. If the hourly simulated aPMV in the zones of the terrace houses exceeds 1, the hour is regarded as an overheating hour.

2.4. Sensitivity Analysis

Global sensitivity analysis, an approach used for assessing the significance of various input to the output, is extensively employed with a regression method in the existing research on building performance analysis [23]. However, the regular regression does not allow to compare between variables due to their different scales. Alternatively, Standardized Regression Coefficients (SRCs) enable inputs to be transformed into unitless standardized values [43]. This process normalizes the scale across all variables, facilitating direct comparisons of their SRC. The regression model is shown below:

$$Y={\beta }_0+{\beta }_1{x}_1+{\beta }_2{x}_2+\dots +{\beta }_{11}{x}_{11}+\epsilon$$

(2)

where $Y$ is the dependent variable, ${\beta }_0$ = 0 denotes the y-intercept, ${\beta }_1$, ${\beta }_2\dots$ ${\beta }_{11}$ represent the standardized beta coefficients, $x_1$, $x_2\cdots x_{11}$ denote the independent variables, and $\epsilon$ is the error prediction.

To achieve reasonably accurate results in regression-based sensitivity analysis, it is essential to use a sampling method that adequately represents the entire search space. Latin Hypercube Sampling (LHS) is a statistical method used to generate a distribution of plausible collections of parameter values from a multidimensional distribution [44]. Research on building performance assessment utilized a combination of LHS and SRC for sensitivity analysis [45]. As indicated in previous studies, a sample size of 1.5 to 10 times variables in LHS might provide convincing result [45]. The LHS sampling process was performed using the Grasshopper platform along with the Design Space Exploration (DSE) plugin, while the SRC calculation was conducted using Python 3.9.

2.5. Model Calibration

Before conducting the simulation, model calibration is essential for reducing the bias between the simulation output and actual indoor conditions. To ensure the feasibility and reliability of the simulation model, we selected the thermal parameters recommended by the Chinese Design Standard for Energy Efficiency in Rural Residential Buildings (GB/T 50824-2013) [26]. The models were then further calibrated based on the comparison between simulated and measured indoor air temperatures, using two calibration metrics: Mean Absolute Error (MAE) and the correlation coefficient (r) [46]. MAE was used to quantify the average absolute difference between the simulated and observed values, with lower values indicating better model accuracy. The correlation coefficient (r) was used to evaluate the strength and direction of the linear relationship between the simulation outputs and the observed data, providing insight into how well the model captures the overall trends and patterns.

Since the typical summer day in Guangzhou is on 1 July (based on the standard for weather data of building energy efficiency JGJ/T 346-2014 [47]), we conducted on-site measurements to record the hourly indoor air temperature in the room on the middle floor of a middle-type terrace house on 1 July, 2024. These measurements were primarily aimed at identifying the uncertain parameters for the selected terrace houses, which were constructed in the same period. The measurements were taken using TianJianHuaYi air temperature data loggers which is sourced from Beijing Tianjian Huayi Technology Development Co., Ltd., located in Beijing, China, with an accuracy of ±0.5 °C and ±3%. Additionally, we collected the hourly outdoor air temperature for the same day from the Guangzhou weather station. The calibration results, shown in Figure 6, indicate that both the MAE and r are within the acceptable ranges (MAE < 2 °C, and r > 0.5) suggested in previous studies [46]. This suggests that the modified simulation model accurately reflects the actual indoor conditions with minimal bias error.

3. Results

3.1. Performance with Baseline Variables

The results of the cooling EUI for two terrace houses types using the base strategy under TMY and future years are shown in Figure 7. They indicate that the side type has higher cooling energy consumption than the middle type under both historical and future weather conditions. The cooling EUI for the middle type ranges between 100 kWh/m² and 200 kWh/m² under TMY, RCP2.6, and the past five decades of RCP4.5 and RCP8.5. Meanwhile, the cooling EUI for the side type ranges between 125 kWh/m² and 250 kWh/m². Additionally, as the severity of future scenarios increases, the difference in average cooling EUI between the two types becomes more pronounced, growing from approximately 12.5 kWh/m² under TMY conditions to around 60 kWh/m² in 2100 under the RCP8.5 scenario.

Figure 8 shows the heat discomfort hours for the two types of townhouses across TMY and all future scenarios. Compared to the cooling EUI results for the two types, there is minimal discrepancy between the discomfort hours in the two types under TMY and RCP scenarios. Except for the discomfort hours ranging from 35% to 60% under TMY for both house types, most of the discomfort hours range from 60% to 80% across all selected years of RCP scenarios. However, the heat discomfort hours in the side type are more concentrated than in the middle type. Similar to the cooling EUI results, as the severity of future scenarios increases, the rise in discomfort hours is greater with each passing decade.

3.2. Sensitivity Analysis of Input Variables

The linear regression method assesses the significance of each input on the output by fitting a linear model to nonlinear functions in building performance analysis. However, the reliability of this method depends on assumptions of three factors: residual normality, determination coefficient R², and p-value. Additionally, the residual analysis may not reveal issues with the assumption of normality. When only the regression is successful, as indicated by an R² value of 0.7 or higher, the SRC can function as sensitivity coefficient [48]. And the p-value aids in determining whether the observed relationships apply to the broader population within the parameter range. A p-value lower than the significance level of 0.05 means that there is less than a 5% probability that the observed results are due to chance [49].

Figure 9 and Figure 10 demonstrate a robust linear pattern in the normal distribution plots of annual cooling energy demand and thermal comfort. Both models exhibit values surpassing the 0.7 threshold and p-values below the 0.05 significance level, indicating that they effectively capture the correlation between variables and outputs, thus facilitating further analysis.

Table 4. Cooling EUI-based output regression analysis at 95% confidence level (N/A and 0.00 indicate p-value exceeding 0.05 and below 0.05, respectively).

Weather Type	Zone Type	p-Value of 11 Variables
		${\mathit{x}}_1$	${\mathit{x}}_2$	${\mathit{x}}_3$	${\mathit{x}}_4$	${\mathit{x}}_5$	${\mathit{x}}_6$	${\mathit{x}}_7$	${\mathit{x}}_8$	${\mathit{x}}_9$	${\mathit{x}}_{10}$	${\mathit{x}}_{11}$
TMY	M_F1	N/A	0.00	0.00	0.00	0.00	0.00	N/A	0.00	N/A	0.00	N/A
RCP2.6	M_F1	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A
RCP4.5	M_F1	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	0.00	N/A
RCP8.5	M_F1	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	0.00	N/A
TMY	M_F2	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	N/A	0.00	N/A
RCP2.6	M_F2	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A
RCP4.5	M_F2	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	0.00	N/A
RCP8.5	M_F2	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	0.00	N/A
TMY	M_F3	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	N/A
RCP2.6	M_F3	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A
RCP4.5	M_F3	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A
RCP8.5	M_F3	N/A	0.00	N/A	N/A	0.00	0.00	N/A	0.00	N/A	0.00	N/A
TMY	S_F1	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	0.00	N/A
RCP2.6	S_F1	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A
RCP4.5	S_F1	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	0.00	N/A
RCP8.5	S_F1	0.00	0.00	0.00	N/A	0.00	0.00	0.00	N/A	N/A	0.00	N/A
TMY	S_F2	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	N/A	0.00	N/A
RCP2.6	S_F2	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A
RCP4.5	S_F2	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	0.00	N/A
RCP8.5	S_F2	0.00	0.00	0.00	N/A	0.00	0.00	0.00	N/A	N/A	0.00	N/A
TMY	S_F3	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	N/A
RCP2.6	S_F3	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A
RCP4.5	S_F3	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	N/A
RCP8.5	S_F3	0.00	0.00	0.00	N/A	N/A	0.00	0.00	0.00	N/A	0.00	N/A

and

Table 5. Thermal comfort-based output regression analysis at 95% confidence level (N/A and 0.00 indicate p-value exceeding 0.05 and below 0.05, respectively).

Weather Type	Zone Type	p-Value of 11 Variables
		${\mathit{x}}_1$	${\mathit{x}}_2$	${\mathit{x}}_3$	${\mathit{x}}_4$	${\mathit{x}}_5$	${\mathit{x}}_6$	${\mathit{x}}_7$	${\mathit{x}}_8$	${\mathit{x}}_9$	${\mathit{x}}_{10}$	${\mathit{x}}_{11}$
TMY	M_F1	N/A	0.00	N/A	0.00	0.00	0.00	N/A	N/A	N/A	0.00	N/A
RCP2.6	M_F1	N/A	0.00	N/A	0.00	0.00	0.00	0.00	N/A	N/A	0.00	N/A
RCP4.5	M_F1	N/A	0.00	N/A	0.00	0.00	0.00	N/A	N/A	N/A	0.00	N/A
RCP8.5	M_F1	N/A	N/A	N/A	N/A	N/A	0.00	N/A	N/A	N/A	0.00	N/A
TMY	M_F2	N/A	0.00	N/A	0.00	0.00	0.00	N/A	N/A	0.00	0.00	N/A
RCP2.6	M_F2	N/A	0.00	N/A	0.00	0.00	0.00	N/A	N/A	0.00	0.00	N/A
RCP4.5	M_F2	N/A	N/A	N/A	0.00	0.00	0.00	N/A	N/A	N/A	0.00	N/A
RCP8.5	M_F2	N/A	N/A	N/A	N/A	N/A	0.00	N/A	N/A	N/A	0.00	N/A
TMY	M_F3	N/A	0.00	N/A	0.00	0.00	0.00	N/A	0.00	0.00	0.00	N/A
RCP2.6	M_F3	N/A	0.00	0.00	0.00	0.00	0.00	N/A	0.00	0.00	0.00	N/A
RCP4.5	M_F3	N/A	N/A	N/A	0.00	N/A	0.00	N/A	N/A	N/A	0.00	N/A
RCP8.5	M_F3	N/A	N/A	N/A	N/A	N/A	0.00	N/A	N/A	N/A	N/A	N/A
TMY	S_F1	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	N/A	0.00	N/A
RCP2.6	S_F1	0.00	0.00	0.00	0.00	0.00	N/A	0.00	0.00	N/A	0.00	N/A
RCP4.5	S_F1	N/A	N/A	N/A	0.00	0.00	N/A	0.00	N/A	N/A	0.00	N/A
RCP8.5	S_F1	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
TMY	S_F2	N/A	0.00	N/A	0.00	0.00	0.00	N/A	N/A	0.00	0.00	N/A
RCP2.6	S_F2	N/A	0.00	0.00	0.00	0.00	0.00	N/A	N/A	0.00	0.00	N/A
RCP4.5	S_F2	N/A	N/A	N/A	0.00	0.00	0.00	N/A	N/A	N/A	0.00	N/A
RCP8.5	S_F2	N/A	N/A	N/A	N/A	N/A	0.00	N/A	N/A	N/A	N/A	N/A
TMY	S_F3	N/A	0.00	N/A	0.00	0.00	0.00	N/A	0.00	0.00	0.00	N/A
RCP2.6	S_F3	N/A	0.00	0.00	0.00	0.00	0.00	N/A	0.00	0.00	0.00	N/A
RCP4.5	S_F3	N/A	N/A	N/A	N/A	N/A	0.00	N/A	N/A	N/A	N/A	N/A
RCP8.5	S_F3	N/A	N/A	N/A	N/A	N/A	0.00	N/A	N/A	N/A	N/A	N/A

present the regression analysis at a 95% confidence level based on cooling EUI output and thermal comfort, respectively. In these tables, “N/A” denotes that the p-value of the variable exceeds 0.05, indicating that the variable lacks correlation and significance to the output. Conversely, a p-value of 0.00 for a variable indicates a strong correlation with the output. As seen in Table 4, most variables from $x_1$ to $x_{10}$ in each type of zone demonstrate significance correlation with the cooling EUI, except for the VT of windows ($x_{11}$). Additionally, the analysis results based on thermal comfort output, presented in Table 5, reveal that most variables have a weak correlation with the output, particularly variables $x_1$, $x_3$, $x_7$, $x_8$, $x_9$, and $x_{11}$. Moreover, the p-value for variables in RCP4.5 and RCP8.5 indicate weaker significance to the output compared to those in RCP2.6 and TMY.

3.3. Ranking the Influence of Input Variables

To provide insight into whether each variable presents a different contribution to cooling energy consumption and indoor thermal comfort under various outdoor weather conditions, the SRCs of 11 variables within six zone types of terrace houses for cooling EUI and heat discomfort hours are presented in Figure 11 and Figure 12. In these figures, an SRC with a positive or negative value represents that the variable has a positive or negative correlation with the output. Additionally, the number on the top of each bar denotes the variable’s rank of contribution to the output based on SRC, with only the top five considered.

As depicted in Figure 11, the findings across the four distinct weather conditions show consistency. It suggests that width, length, orientation, and overhang depth exhibit a negative correlation with the terrace houses’ cooling energy consumption, whereas the remaining variables show a positive correlation. According to the contribution ranking, the top three variables under different weather conditions include the width and length of the zone, orientation, U-value of the roof, WWR, and SHGC of the window. Among these variables, WWR and the length of the zone have a significant influence on the cooling EUI output in every zone type. However, the influence of variables on cooling EUI can vary across zones on different floors. For instance, the U-value of the roof has a greater impact on the cooling energy consumption of zones on floor 3 compared to the other floors. Furthermore, the influence of variables on cooling EUI may differ based on the positions of the terrace house. For example, the width of the zone has a more significant impact on the cooling energy consumption of side-type terrace houses compared to middle-type terrace houses. With the increasing severity of future weather scenarios, the contribution of the variable length becomes greater than that of WWR in each zone type.

The SRC results for heat discomfort hours are shown in Figure 12. The top three variables influencing thermal comfort in six zone types across four different weather conditions include the length of the zone, overhang depth, shade distance, WWR, and SHGC of windows. Among these variables, WWR and length exhibit different correlations depending on the zone types. Shade distance and SHGC of windows show a positive correlation with the thermal comfort, while overhang depth shows a negative correlation. Additionally, the SRCs of 11 variables for heat discomfort hours vary significantly between different zone types. For example, the SHGC of windows significantly influences thermal comfort in zones on the first floor, whereas the WWR of windows has a greater impact on zones on the second and third floors. Conversely, different house types with different positions have minimal impact on the influence of variables on thermal comfort. Most variables show similar impact with increasing emissions under different future scenarios. However, under RCP4.5 and RCP8.5 scenarios, the impact of the SHGC of windows on thermal comfort for the zones on the top floor are weakened, while the length variable’s impact on thermal comfort significantly increases.

3.4. Baseline and Retrofit Performance Comparison

Given that there are two outputs, cooling energy consumption and thermal discomfort, influenced by the 11 input variables, it is possible that some variables may have a positive correlation with cooling energy but a negative correlation with thermal discomfort. To select the optimal parameters within the provided range of each variable, as shown in Table 1, for these two outputs, the selection is based on four rules:

(1)

Choose the maximum parameter value when a variable in the top five has a negative correlation with both outputs or has a negative correlation with one output and no correlation with the other;

(2)

Choose the minimum parameter value when a variable in the top five has a positive correlation with both outputs or has a positive correlation with one output and no correlation with the other;

(3)

Choose the middle parameter value when a variable in the top five has a positive correlation with one output but a negative correlation with the other;

(4)

Choose the baseline parameter value when a variable has no correlation with either output.

The retrofit parameters for 11 variables in six zones under four weather conditions are shown in

Table 6. Retrofit parameters of 11 variables for cooling EUI and thermal comfort in TMY.

Variable	Unit	Zone Type
		M_F1	M_F2	M_F3	S_F1	S_F2	S_F3
Width	m	4	4	4	6	6	6
Length	m	12	11	11	12	11	11
Orientation	-	East	East	East	South	East	East
Overhang depth	m	1	1	1	1	1	1
Shade distance	m	2	2	2	2	2	2
WWR	%	0.1	0.3	0.3	0.3	0.3	0.3
Wall U-value	W/m2∙K	1.5	1.5	1.5	1	1.5	1.5
Roof U-value	W/m2∙K	1	1	1	1	1	0.5
Glaz U-value	W/m2∙K	5	3	3	5	5	5
SHGC	-	0.4	0.4	0.4	0.4	0.4	0.4
VT	-	0.8	0.8	0.8	0.8	0.8	0.8

,

Table 7. Retrofit variable of 11 variables for cooling EUI and thermal comfort under RCP2.6.

Variable	Unit	Zone Type
		M_F1	M_F2	M_F3	S_F1	S_F2	S_F3
Width	m	4	4	4	6	6	6
Length	m	12	11	11	12	11	11
Orientation	-	South	East	South	South	South	South
Overhang depth	m	1	1	1	1	1	1
Shade distance	m	2	2	2	2	2	2
WWR	%	0.1	0.3	0.3	0.1	0.3	0.3
Wall U-value	W/m2∙K	1.5	1.5	1.5	1	1	1.5
Roof U-value	W/m2∙K	1	1	1	1	1	0.5
Glaz U-value	W/m2∙K	5	3	5	5	5	5
SHGC	-	0.4	0.4	0.4	0.4	0.4	0.4
VT	-	0.8	0.8	0.8	0.8	0.8	0.8

,

Table 8. Retrofit variable of 11 variables for cooling EUI and thermal comfort under RCP4.5.

Variable	Unit	Zone Type
		M_F1	M_F2	M_F3	S_F1	S_F2	S_F3
Width	m	4	4	4	6	6	5
Length	m	12	11	11	12	11	11
Orientation	-	North	North	South	South	South	South
Overhang depth	m	1	1	1	1	1	1
Shade distance	m	2	2	2	2	2	2
WWR	%	0.1	0.3	0.3	0.1	0.3	0.3
Wall U-value	W/m2∙K	1.5	1.5	1.5	1	1	1.5
Roof U-value	W/m2∙K	1	1	0.5	1	1	0.5
Glaz U-value	W/m2∙K	3	5	5	5	5	5
SHGC	-	0.4	0.4	0.4	0.4	0.4	0.4
VT	-	0.8	0.8	0.8	0.8	0.8	0.8

and

Table 9. Retrofit variable of 11 variables for cooling EUI and thermal comfort under RCP8.5.

Variable	Unit	Zone Type
		M_F1	M_F2	M_F3	S_F1	S_F2	S_F3
Width	m	4	4	4	6	6	5
Length	m	12	11	11	12	11	11
Orientation	-	North	North	East	South	East	South
Overhang depth	m	1	1	1	1	1	1
Shade distance	m	2	2	2	2	2	2
WWR	%	0.1	0.3	0.3	0.1	0.3	0.3
Wall U-value	W/m2∙K	1.5	1.5	1.5	1	1	1.5
Roof U-value	W/m2∙K	1	1	0.5	1	1	0.5
Glaz U-value	W/m2∙K	1	5	5	1	1	5
SHGC	-	0.4	0.4	0.4	0.4	0.4	0.4
VT	-	0.8	0.8	0.8	0.8	0.8	0.8

.

After collecting retrofit parameters for the 11 variables aimed at reducing cooling energy consumption and heat discomfort hours, a comparison analysis for building performance based on the two outputs between the baseline parameters and the retrofit parameters was conducted. As shown in Figure 13, the reduction trend of cooling EUI is projected to increase over the years under different future scenarios ranging from 50 kWh/m² to 165 kWh/m², with more severe emission scenarios resulting in even higher reductions. Under RCP2.6, the fluctuation in reduction between 2030 and 2090 for middle-type and side-type terrace houses is around 10 kWh/m². Under RCP4.5, the fluctuation for a side-type terrace house is approximately 30 kWh/m², which is 10 kWh/m² higher than that for middle-type terrace houses. Under RCP8.5, the fluctuation increases to around 40–50 kWh/m² for middle-type buildings and 50–60 kWh/m² for side-type buildings. The retrofit strategy is effective in reducing cooling EUI in both house types, with a greater impact in a side-type terrace house compared to a middle-type terrace house.

The results in Figure 14 suggest that the trend in reducing heat discomfort hours is expected to stabilize or decline over the years under RCP scenarios. Moreover, in areas with higher floors, the reduction in heat discomfort hours with the retrofit strategy is less noticeable for both types of terrace houses. Specifically, the decrease in discomfort hours for a side-type terrace house on the first floor under RCP scenarios varies between 3% and 7%, while for a middle-type terrace house on the same floor, it ranges from 4% to 7%. However, both types exhibit similar reductions in discomfort hours on the second and third floors, ranging from 2% to 4% and 0% to 3.5%, respectively.

4. Discussion

While much of the relevant research on building performance focuses on assessments of entire buildings or regional clusters, limited attention has been given to various zone types within individual buildings. This gap may lead to retrofit strategies that are not universally applicable across all zones within a building. This research aims to provide insights into building performance optimization from a multi-zone-type perspective. The evaluation considers two common types of terrace housing (middle and end types) with three floors (ground, middle, and top floors), as well as current and future weather datasets.

According to the sensitivity analysis results presented in Table 4 and Table 5, under both current and future climate conditions, most variables are not statistically significant in influencing discomfort hours. Conversely, the results for cooling EUI show the opposite trend, indicating that while many commonly considered variables may improve energy efficiency, they contribute minimally to enhancing comfort levels. Therefore, innovative cooling strategies to improve comfort in terrace housing should be prioritized in future studies. Additionally, in the ranking results of the 11 variables affecting heat discomfort hours shown in Figure 10, the positive and negative impacts vary by floor. Future studies on retrofitting for indoor thermal comfort should consider zone types located on different floors separately, rather than assessing only the entire housing.

However, since most terraced houses were originally constructed in historical periods and renovated in recent years, obtaining precise thermal parameters is challenging. The thermal parameter details are derived from existing regulations, and since the analyzed houses are occupied by local residents, it is difficult to conduct long-term measurements to collect actual data. The calibration of the simulation model is based on short-term recorded indoor air temperature data during hot period. A longer measurement period would be beneficial for the assessment in future studies. Additionally, considering more thermal parameters, such as relative humidity, in future studies of subtropical regions would make the evaluation of existing cooling strategies more comprehensive.

5. Conclusions

There are three main findings in this evaluation. During TMY and most years of RCP scenarios, the cooling EUI in the side-type terrace house ranges from 125 kWh/m² to 250 kWh/m², which is higher than the range between 100 kWh/m² and 200 kWh/m² observed in the middle-type terrace house. However, the heat discomfort hours in both types of terrace house exhibit a similar scope across all weather conditions, spanning from 40% to 85%. Additionally, results based on both outputs indicate that the increase under RCP8.5 is significantly higher than under the other two RCP scenarios.

The sensitivity analysis results indicate that most variables have p-values greater than 0.05 based on the result of heat discomfort hours, suggesting they are not statistically significant for this output. In contrast, the results for cooling EUI reveal a different pattern, with several variables showing significant influence. Specifically, window-to-wall ratio (WWR) and zone length significantly affect cooling EUI across all zones. Additionally, the significance of variables affecting heat discomfort hours varies between zones on different floors, likely due to the effects of natural ventilation, which causes the indoor thermal environment to fluctuate in response to outdoor climate conditions.

With the retrofit parameters for each variable aimed at reducing cooling energy and discomfort hours, the reduction of cooling EUI ranges from 50 kWh/m² to 165 kWh/m² across all zones, and the reduction of heat discomfort hours ranges from 0.7% to 7.2%. In addition, under RCP scenarios, the reduction level for cooling EUI increases over the years, and the reduction level for thermal discomfort hours decreases over time. Additionally, the retrofit strategy shows weaker performance in mitigating heat discomfort hours on the upper floors.

This analysis provides insights into the thermal performance assessment of subtropical terrace houses from the perspective of different zone types. The findings indicate that the effectiveness of retrofit strategies varies across zones with different positions within the house. These results can inform architects, engineers, and other stakeholders in developing more tailored and effective retrofit strategies for each zone, thereby enhancing the overall thermal performance of terrace houses in subtropical regions.