Data-Driven Life-Cycle Assessment of Household Air Conditioners: Identifying Low-Carbon Operation Patterns Based on Big Data Analysis

Abstract

Air conditioners are a critical adaptation measure against heat- and cold-related risks under climate change. However, their electricity use and refrigerant leakage increase greenhouse gas (GHG) emissions. This study developed a data-driven life-cycle assessment (LCA) framework for residential room air conditioners in Japan by integrating large-scale field operation data with life-cycle climate performance (LCCP) modeling. We aggregated 1 min records for approximately 4100 wall-mounted split units and evaluated the 10-year LCCP across nine climate regions. Using the annual operating hours and electricity consumption, we classified the units into four behavioral quadrants and quantified the life-cycle GHG emissions and parameter sensitivities for each. The results show that the use-phase electricity dominated the total emissions, and that even under the same climate and capacity class, the 10-year per-unit emissions differed by roughly a factor of two between the high- and low-load quadrants. The sensitivity analysis identified the heating hours and the setpoint–indoor temperature difference as the most influential drivers, whereas the grid CO₂ intensity, equipment lifetime, and refrigerant assumptions were of secondary importance. By replacing a single assumed use scenario with empirical profiles and behavior-based clusters, the proposed framework improves the representativeness of the LCA for air conditioners. This enabled the design of cluster-specific mitigation strategies.

1. Introduction

Climate change is driving increases in both the frequency and intensity of heatwaves and extremely high temperatures worldwide. The number of people exposed to heat conditions that exceed the body’s thermoregulatory capacity is projected to grow rapidly [1,2]. The Sixth Assessment Report of the Inter-governmental Panel on Climate Change (IPCC) concluded that increases in heatwaves and high-temperature extremes substantially increase the risk of heat-related mortality and morbidity among urban populations, particularly among older adults and people with pre-existing health conditions [3].

At the same time, although warming winters are expected to reduce the health risks associated with extreme cold, cold-related health risks remain substantial, and several studies report that most temperature-related mortality is still attributable to cold rather than heat [4,5]. Under these circumstances, air-conditioning systems that ensure indoor thermal comfort are becoming increasingly widespread, particularly in urban areas and emerging economies, and have become a major driver of the final energy consumption and greenhouse gas (GHG) emissions in the building sector [6,7]. According to the International Energy Agency (IEA), the cooling energy demand driven primarily by the residential sector is projected to more than triple by 2050, at which point space cooling will account for up to 16% of global electricity use [8]. Long-term scenario analyses for the residential sector indicate that although rising winter temperatures will reduce the heating demand, space heating will remain a major contributor to building energy use, even by 2050 [9,10]. In Japan, future climate projections suggest decreasing heating and increasing cooling demands; however, heating energy use is expected to remain higher than cooling energy use [11]. Thus, although air-conditioning is indispensable as an adaptation measure against climate change, its energy consumption and refrigerant-related emissions create a non-negligible tradeoff from a mitigation perspective [12].

Life-Cycle Assessment (LCA) and life-cycle climate performance (LCCP) have been widely used as quantitative approaches to evaluate this tradeoff. LCA provides a framework for assessing the environmental impacts of products and systems across the entire life cycle, from resource extraction to manufacturing, use, end-of-life treatment, and recycling. LCCP has been applied as a climate impact indicator that integrates direct emissions from refrigerants and indirect emissions from energy use in air-conditioning systems. Previous LCA/LCCP studies on residential air-conditioning systems have consistently shown that most climate change impacts are attributable to electricity consumption during the usage phase. For example, a study of room air conditioners in Indonesia found that more than 90% of 10-year life-cycle GHG emissions arose from the use phase, with manufacturing, transport, and end-of-life contributing only a small percent each [13]. An LCCP study of packaged air conditioners likewise showed that indirect emissions from electricity use during operation dominate the LCCP, exceeding direct emissions from refrigerant leakage [14]. Case studies of residential heating and cooling systems in four U.S. regions and residential air-conditioning in Saudi Arabia have also reported that operational electricity consumption accounts for the largest share of climate impacts, primarily due to the type of fuel used for power generation [15,16].

At the same time, previous LCA studies of air conditioning have reported that LCA results can vary widely depending on the underlying scenario assumptions. A systematic review of air-conditioning LCAs showed that the global warming potential (GWP) of grid-connected vapor-compression systems ranges from 3402 to 104,210 kg-CO₂ eq per system, depending on the assumed functional unit, system boundary, and electricity mix [7]. In particular, assumptions regarding the annual energy consumption have been identified as a key issue, and several studies have shown that variability in energy use driven by occupant behavior can be a significant source of uncertainty in the results [17,18,19,20,21]. However, only a limited number of studies have explicitly separated and evaluated cooling and heating usage during the operational phase [22]. Moreover, as summarized in

Table 1. Summary of environmental assessment studies on residential air-conditioning systems.

Source	Region	Evaluation Target	Product Lifetime (Years)	Energy Consumption Calculation Type	Method for Calculating Energy Consumption	Main Findings
Gheewala & Nielsen (2003) [23]	Thailand	Central vs. split AC 1 systems	10	Measured and standards-based	Central systems: measured data; split systems: company-average data	Results are highly dependent on the electricity generation technology mix.
Shah et al. (2008) [16]	United States (4 regions)	Residential heating/cooling systems	35	Simulation and standards-based	Operation simulated using efficiency indicators (SEER 2)	Most environmental impacts arise from the system’s operational phase.
Joudi & Qusay (2014) [24]	Iraq	Residential AC	15	Standard and measurement-based	Measurements conducted under ASHRAE 3 Standard 37 test conditions [30]	Direct impacts from refrigerants account for approximately 5–9% of total impacts; the remainder is due to electricity use during operation.
Almutairi et al. (2015) [15]	Saudi Arabia	Residential window and split AC units	10/15	Standard-based	Calculated from efficiency indicators (EER 4) and prescribed operating hours	Environmental impacts are driven by electricity consumption in the use phase.
Li (2015) [14]	United States (7 cities)	Residential packaged AC	15	Standard and modeling-based	Annual energy consumption calculated from SEER under prescribed operating hours based on ANSI 5/AHRI 6 Standard 210/240 [31]	More than 90% of LCCP 7 emissions are attributable to indirect emissions during operation.
Nishijima (2017) [25]	Japan	Residential AC	12.6	Standard and modeling-based	Calculated from catalogue values under conditions specified by JIS 8 C 9612 [32]	Improvements in energy efficiency are more important than extending product lifetime for reducing CO2 emissions.
Choi et al. (2017) [22]	South Korea (5 cities)	Residential heating/cooling systems	15	Standard and modeling-based	Calculated from temperature distributions under AHRI 210/240 Standard conditions [31]	Heating energy use accounts for approximately half of total emissions, and cooling energy use accounts for approximately 30%.
Karkour et al. (2021) [13]	Indonesia (Jakarta)	Residential AC	10	Standard-based	Calculated from manufacturer-specified operating hours	More than 90% of GWP 9 is due to electricity consumption during the use phase.
Wan et al. (2021) [26]	Worldwide (11 cities)	Unitary AC	15	Standard and simulation-based	Calculated based on AHRI Standard 210/240 [33] test conditions and ASHRAE Standard 34-2019 [34]	In countries with high grid emission factors, efficiency is critical; in countries with low grid emission factors, refrigerant leakage becomes the dominant contributor.
Andrade et al. (2024) [27]	Colombia	Variable and fixed-type AC	15	Standard and measurement-based	Measurements conducted under ISO 5151 [35] and ISO 16358-1 [36] test conditions	The choice of calculation method for energy consumption results in significant variations in TEWI 10 values.
Aljolani et al. (2024) [28]	Europe (3 cities)	Residential AC	15	Simulation and standards-based	Cooling demand simulated with ASPEN PLUS v10 and HAP v5.11, and annual consumption calculated according to EN 14825:2013 [37]	Electricity consumption accounts for most life-cycle GHG 11 emissions from air-conditioner use.
Marcinkowski & Levchenko (2025) [29]	Ukraine (Sumy)	IREC 12 vs. VCRM 13 system	10	Simulation-based	Calculated from summer weather data and performance indicators (COP 14)	Electricity consumption during operation accounts for 88–95% of total life-cycle impacts.

¹: Air conditioner, ²: Seasonal Energy Efficiency Ratio, ³: American Society of Heating, Refrigerating, and Air-Conditioning Engineers, ⁴: Energy Efficiency Ratio, ⁵: American National Standards Institute, ⁶: Air-Conditioning, Heating, and Refrigeration Institute, ⁷: life-cycle climate performance, ⁸: Japanese Industrial Standard, ⁹: global warming potential, ¹⁰: Total Equivalent Warming Impact, ¹¹: greenhouse gas, ¹²: Indirect Regenerative Evaporative Cooling, ¹³: Vapor Compression Refrigerating Machine, ¹⁴: Coefficient of Performance.

, many existing LCAs of air-conditioning systems modeled the use phase using simplified representative scenarios based on industrial standards that specify the annual operating hours, load factors, and weather conditions. Although such approaches facilitate comparability between products, their representativeness under real-world usage conditions has been questioned [6,7,13,14,15,16,22,23,24,25,26,27,28,29].

In the building energy field, occupant behaviors, such as air conditioner operation and thermostat setpoints, are also often simplified as fixed schedules in building performance simulations, and this has been reported to cause discrepancies between the predicted and measured energy uses [38,39]. However, in building environmental and heating, ventilation, and air-conditioning (HVAC) engineering, studies that exploit long-term, high-frequency data from sensors and smart meters to analyze air-conditioning usage and energy demand in a data-driven manner have been advancing rapidly. For example, studies that apply clustering analysis to measured cooling load data in multi-family buildings to extract multiple typical types of cooling intensity and operating patterns, as well as studies that model air conditioner on/off behavior in relation to indoor and outdoor environmental conditions based on monitoring data, have demonstrated the effectiveness of behavior typologies and demand forecasting derived from empirical data [40,41,42]. Furthermore, reviews of building energy performance have highlighted that the stochastic and diverse nature of occupant behavior is a significant cause of the performance gap between predicted and measured performances and have emphasized the need to integrate behavioral models with data-driven approaches [43]. However, to date, insights into air-conditioning operating patterns derived from such big data analyses have not been fully integrated into LCA frameworks that cover the entire life cycle. Big data studies in the HVAC and building domains have mainly focused on kWh-based operational efficiency, peak load reduction, and demand–response control, and have rarely evaluated life-cycle environmental impacts, including equipment manufacturing and refrigerants [40,42,43].

Against this background, the objective of this study is to clarify regional usage patterns and develop a data-driven LCA framework that integrates typologies of operating behavior with a life-cycle environmental impact assessment using operating record big data from approximately 4100 residential room air conditioners across Japan. By combining big data analytics with LCA, we aimed to address the representativeness and uncertainty issues in the environmental assessment of air-conditioning equipment, identify low-carbon operating patterns, and provide insights into sustainable air-conditioning.

This study extended previous air-conditioner LCA/LCCP and HVAC big data research. Many LCA/LCCP studies evaluated a few standardized or modeled usage scenarios, addressing behavioral differences primarily through comparisons or sensitivity tests. Meanwhile, data-driven HVAC research has advanced in identifying behavioral typologies and demand-side insights but rarely integrates such behavior-based evidence into life-cycle impact assessments that include manufacturing and refrigerants. By linking high-resolution operational data from thousands of homes to a process-based LCA/LCCP model and introducing an interpretable quadrant typology within the same region and capacity class, this study quantified how behavioral variability affects cluster-specific life-cycle GHG values and (one-at-a-time) sensitivity results. This combined approach offers an empirical basis for targeted mitigation strategies based on operating patterns.

2. Materials and Methods

2.1. Data Collection and Preprocessing

This study focused on wall-mounted split-type residential air conditioners (cooling/heating), which are predominant in the Japanese market, and analyzed sensor data from approximately 70,000 units that employed heat pump technology and inverter control for high-efficiency operation. The data were provided by a domestic air-conditioner manufacturer and recorded at 1 min intervals over 3.5 years from August 2017 to January 2021. The dataset comprised the following three categories of variables: (i) unit attributes (model information, cooling capacity, and installation region), (ii) operational information (on/off status, power consumption, operating mode, and thermostat setpoint), and (iii) environmental information (outdoor and indoor temperatures).

Before the analysis, we applied the following preprocessing steps:

• To ensure consistent calculation of the annual operating hours and annual electricity consumption with complete coverage of both cooling and heating seasons, we extracted data for 2020, which had the fewest missing records in the dataset.
• To characterize the regional usage patterns, we restricted the sample to units whose installation regions could be identified.
• To eliminate differences in equipment characteristics by rated capacity and compare patterns driven solely by actual usage, we limited the analysis to units with a cooling capacity of 5.6 kW, which was the class with the largest number of units in the dataset.
• To exclude units that were not in service, we removed air conditioners with annual total operating hours of less than one.
• To remove outliers and capture general usage tendencies, we excluded units whose annual electricity consumption laid outside the mean ± 3 standard deviations after a logarithmic transformation.

No imputation of missing values was performed in this analysis. The final analytical dataset consisted of 4092 units of the 5.6 kW class that satisfied these conditions.

2.2. Regional Classification and Climatic Characteristics

To account for Japan’s diverse climatic conditions, we divided the country into nine regions (Hokkaido, Tohoku, Kanto, Hokuriku, Tokai, Kinki, Chugoku, Shikoku, and Kyushu) based on the classification provided by the Japan Meteorological Agency (Figure 1) [44]. The number of units analyzed in each region is summarized in

Table 2. Count of units of nine regions (year 2020).

Hokkaido	Tohoku	Kanto	Hokuriku	Tokai	Kinki	Chugoku	Shikoku	Kyushu	Total
24	86	2322	140	386	777	55	82	216	4092

. The climate in Japan varied significantly between regions. The air temperature showed a transparent latitudinal gradient from the cold-temperate climate of Hokkaido to the near-subtropical conditions of Kyushu. Precipitation patterns also differed markedly between the Pacific and Japan Sea sides. Many areas on the Pacific side experienced hot and humid summers due to the rainy season and typhoons. The winters were relatively dry and sunny. In contrast, where regions facing the Sea of Japan, such as Hokuriku, experienced some of the heaviest snowfalls in the world due to winter monsoon winds, with precipitation concentrated in winter. Thus, latitude, monsoon patterns, and topography interacted to create diverse climatic conditions across the nine regions.

2.3. LCCP

In this study, the reference product was a single 5.6 kW residential room air conditioner, and the functional unit was defined as “use of one unit over 10 years.” LCCP is defined, following the guidelines of the International Institute of Refrigeration (IIR), as the sum of direct GHG emissions due to refrigerant leakage and indirect GHG emissions associated with electricity consumption, and the manufacturing and end-of-life treatment of the unit and refrigerant [45]. The system boundary was cradle-to-grave and included the manufacturing of the indoor and outdoor units and the refrigerant, operation of the air conditioner, end-of-life treatment, and refrigerant destruction (Figure 2).

The LCCP for an air conditioner installed in region r is given by Equation (1):

$${\mathrm{L}\mathrm{C}\mathrm{C}\mathrm{P}}_r=E_{\mathrm{d}\mathrm{i}\mathrm{r}}+E_{\mathrm{i}\mathrm{n}\mathrm{d},\mathrm{o}\mathrm{p},r}+E_{\mathrm{i}\mathrm{n}\mathrm{d},\mathrm{e}\mathrm{m}\mathrm{b}}$$

(1)

where $E_{dir}$(kg-CO₂ eq) is the direct emissions from refrigerant leakage and end-of-life venting, $E_{ind,op,r}$ (kg-CO₂ eq) is the indirect emissions from electricity consumption during the use phase in region r, and $E_{ind,emb}$ (kg-CO₂ eq) is the indirect embodied emissions from the manufacturing, transport, and end-of-life treatment of the unit and refrigerant.

2.3.1. Direct Emissions

The direct emissions $E_{dir}$ from refrigerant leakage and end-of-life venting are expressed as the sum of emissions due to leakage during use, $E_{leak}$, and the emissions released at the end-of-life, $E_{EoL}$ (Equation (2)):

$$E_{\mathrm{d}\mathrm{i}\mathrm{r}}=E_{\mathrm{l}\mathrm{e}\mathrm{a}\mathrm{k}}+E_{\mathrm{E}\mathrm{o}\mathrm{L}}$$

(2)

Emissions due to refrigerant leakage during use, $E_{leak}$, were calculated by assuming an annual leakage rate $f_{leak}$ over the equipment lifetime $L$ (Equation (3)):

$$E_{leak}=C\cdot L\cdot f_{leak}\cdot {GWP}_{R32}$$

(3)

Emissions due to venting at the end-of-life, $E_{EoL}$, were calculated by assuming that the non-recovered fraction $\left(1-r_{rec}\right)$ of the remaining charge after in-use leakage was released into the atmosphere (Equation (4)):

$$E_{EoL}=\left[C-C\cdot L\cdot f_{leak}\right]\left(1-r_{rec}\right)\cdot {GWP}_{R32}$$

(4)

where $C$ (kg) is the initial refrigerant charge, $L$ (years) is the equipment lifetime, $f_{leak }$ (%/year) is the annual leakage rate, $r_{rec}$ (%) is the refrigerant recovery rate at the end-of-life, and ${GWP}_{R32}$ (kg-CO₂ eq/kg) is the 100-year GWP of R32. Based on information obtained from the manufacturer, we set $L=10$ years, $f_{leak}=2\%$/year, and $r_{rec}=30\%$. The refrigerant was assumed to be R32, and we adopted a ${GWP}_{R32}$ value of 772 kg-CO₂ eq/kg-R32 from the IPCC Sixth Assessment Report [3].

2.3.2. Indirect Emissions from Operation

Indirect emissions from electricity consumption during the use phase in regions r, $E_{ind,op,r}$, were calculated using the following Equation (5):

$$E_{\mathrm{i}\mathrm{n}\mathrm{d},\mathrm{o}\mathrm{p},r}=L\cdot E_{\mathrm{e}\mathrm{l},r}\cdot {\mathrm{E}\mathrm{F}}_{\mathrm{e}\mathrm{l},r}$$

(5)

where $E_{el,r}$ (kWh/unit/year) is the annual electricity consumption per unit in region r, and $E{F}_{el,r}$ (kg-CO₂/kWh) is the CO₂ emission factor for electricity in that region. The values of $E_{el,r}$ are for the annual cooling and heating electricity consumption estimated for each of the nine regions in this study. The electricity CO₂ emission factors,$E{F}_{el,r}$, were based on location-based emission factors for electric utilities corresponding to the nine regions, as reported by the Ministry of the Environment [46].

To compare real-world operation with standard test conditions, we also computed the LCCP under the Japanese Industrial Standard “JIS C 9612:2013 Room air conditioners.” For this purpose, we used the cooling and heating season electricity consumption values listed on the national energy-efficient product information website, which were calculated according to JIS C 9612:2013, to obtain the annual electricity consumption, $E_{el,JIS}$, under JIS conditions. Substituting $E_{el,JIS}$ into Equation (5) yields the JIS-based LCCP [32,47]. The operating conditions prescribed in JIS C 9612:2013 are listed in

Table 3. Overview of operating conditions in JIS C 9612:2013 for room air conditioners [32].

Item	Cooling	Heating
Outside temperature	Modeled outside temperatures for Tokyo (Extended Automated Meteorological Data Acquisition System (AMeDAS) Weather Data 2000)
Setting temperature	27 °C	20 °C
Operation start temperature	Above 24 °C	Below 16 °C
Operation period	May 23–October 4 (135 days)	November 8–April 16 (159 days)
Operation time	6:00–24:00 (18 h/day)

.

2.3.3. Indirect Embodied Emissions

The indirect embodied emissions,${ E}_{ind,emb}$, associated with the manufacturing, transport, and end-of-life treatment of the air conditioner (indoor and outdoor units) and refrigerant were calculated using following Equation (6):

$$E_{\mathrm{i}\mathrm{n}\mathrm{d},\mathrm{e}\mathrm{m}\mathrm{b}}=E_{\mathrm{A}\mathrm{C},\mathrm{m}\mathrm{a}\mathrm{n}}+E_{\mathrm{A}\mathrm{C},\mathrm{E}\mathrm{o}\mathrm{L}}+E_{\mathrm{r}\mathrm{e}\mathrm{f},\mathrm{m}\mathrm{a}\mathrm{n}}+E_{\mathrm{r}\mathrm{e}\mathrm{f},\mathrm{d}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{r}}$$

(6)

where $E_{AC,man}$ (kg-CO₂ eq/unit) is the GHG emissions from manufacturing the indoor and outdoor units and accessories (including raw material extraction, component manufacturing, assembly, and transport), $E_{AC,EoL}$ (kg-CO₂ eq/unit) is the GHG emissions from the end-of-life treatment and recycling of the unit, $E_{ref,man}$ (kg-CO₂ eq/unit) is the GHG emissions from manufacturing R32 refrigerant, and $E_{ref,destr}$ (kg-CO₂ eq/unit) is the GHG emissions from the destruction of R32 at end-of-life.

The $E_{AC,man}$ and $E_{AC,EoL}$ values were calculated by mapping the unit mass and primary material composition (steel, copper, aluminum, plastics, etc.) provided by the manufacturer to the manufacturing and waste treatment processes in the Inventory Database for Environmental Analysis version 3.5 (IDEA v3.5) [48].

For $E_{ref,man}$ and $E_{ref,destr}$, we adopted emission factors from the LCA of the R32 manufacturing and destruction processes reported by Yasaka et al. [49]. Particularly, $E_{ref,man}$ was calculated by multiplying the refrigerant charge $C$ (kg) by 7.77 kg-CO₂ eq/kg-R32 (GHG emissions from R32 manufacturing), and $E_{ref,destr}$ was calculated by multiplying the recovered refrigerant at end-of-life by 27.62 kg-CO₂ eq/kg-R32 (GHG emissions from R32 destruction).

These four terms were assumed to be constant across regions, and together with the direct emissions $E_{dir}$ and the indirect operational emissions $E_{ind,op,r}$ defined above, were used to calculate the LCCP according to Equation (1).

2.4. Sensitivity Analysis

To identify the key drivers of the LCA results, we conducted a one-at-a-time (OAT) sensitivity analysis of the parameters listed in

Table 4. Parameters and ranges used in the one-at-a-time sensitivity analysis.

Parameter	Unit	Low Value	Baseline Value	High Value	Definition of Low/High Scenario	Source
Product lifetime	year	6	10	14	Low: statutory durable years; high: average service life of room ACs in Japan	[50,51]
Annual refrigerant leakage rate	%/year	1	2	10	Range of lower and upper leakage rates is reported for residential ACs	2006 IPCC Guidelines (2019 Refinement) [52]
Refrigerant recovery rate at end-of-life	%	0	30	80	Range of lower and upper recovery rates is reported for residential ACs
Grid CO2 emissions factor	kg-CO2/kWh	0.362	0.453	0.601	Low: minimum among nine utilities (Kansai Electric); high: maximum (Hokkaido Electric)	Utility-specific GHG intensity data [46]
Cooling/heating operating hours per unit	h/unit/year (°C)	25th percentile of observed hours	Mean of observed hours	75th percentile of observed hours	Percentile values calculated from the nationwide AC big data set	Statistical analysis of monitoring data
ΔTemperature (setting–indoor)		25th percentile of ΔT	Mean ΔT	75th percentile of ΔT	Difference between setpoint and indoor air temperature; percentile values from monitoring data

. Based on the contribution analysis and previous studies, we selected parameters that contributed substantially to the total GHG emissions and were subject to significant uncertainty, which included the equipment lifetime, annual refrigerant leakage rate, refrigerant recovery rate at end-of-life, electricity CO₂ emission factor, cooling and heating operating hours, and the temperature difference between the thermostat setpoint and indoor air during cooling and heating.

In the sensitivity analysis, we varied only one parameter at a time to its lower or upper bound, as defined in Table 4, while keeping all other parameters at their baseline values and evaluated the resulting change in the life-cycle GHG emissions.

For the equipment lifetime, we adopted 6 years as the lower bound, corresponding to the statutory service life for durable consumer goods in Japan, and 14 years as the upper bound, corresponding to the average service life of room air conditioners reported in a consumer survey [50,51]. For the annual refrigerant leakage and end-of-life recovery rates, we used the lower and upper bounds reported in the 2019 Refinement of the 2006 IPCC Guidelines for National Greenhouse Gas Inventories [52]. For the electricity CO₂ emissions factor, we used the minimum and maximum values among the nine major electric utilities in Japan, namely, 0.362 kg-CO₂/kWh for Kansai Electric Power and 0.601 kg-CO₂/kWh for Hokkaido Electric Power, as the lower and upper bounds, respectively [46]. For the cooling and heating operating hours and the thermostat–indoor temperature differences during cooling and heating, we used the 25th and 75th percentiles of the empirical distributions derived from the usage data analyzed in this study as the lower and upper bounds, respectively.

3. Results

3.1. LCA Results Reflecting Real-World Air-Conditioner Use

To examine how discrepancies between the JIS test conditions and actual usage affected the LCA results, we evaluated the life-cycle GHG emissions for three cases: a JIS-based scenario (JIS-based), an overall national average (overall), and region-specific scenarios applying cooling and heating electricity consumption and electricity emission factors for each of the nine regions. The results are shown in Figure 3.

For the annual electricity consumption (Figure 3a), the mean total under the JIS-based scenario (1962 kWh/unit/year) substantially exceeded the national mean under actual usage (overall, 729 kWh/unit/year), reaching a factor of approximately 2.7. In the JIS-based case, the heating consumption significantly exceeded the cooling consumption by approximately 2.3. By contrast, the regional means under actual usage fell within a relatively narrow range of approximately 630–790 kWh/unit/year, and no statistically significant differences were observed between the regions. In all regions, the annual electricity use for heating exceeded that for cooling. The 5th–95th percentile ranges (error bars) for each region were extensive, indicating that the intra-regional variation in usage patterns was a more dominant driver of electricity consumption than the differences in regional means.

Figure 3b shows the life-cycle GHG emissions, assuming 10 years of operation. The use phase (cooling and heating) accounted for approximately 75–85% of total emissions, whereas the contributions from material production, manufacturing, logistics, and end-of-life treatment were relatively small. Emissions from refrigerant leakage amounted to only a small percentage of the total but were non-negligible compared with those from other life-cycle stages. When location-based electricity emission factors were applied, the regional mean total emissions ranged from approximately 3500 to 5500 kg-CO₂ eq, with a maximum-to-minimum ratio of approximately 1.5 between Hokkaido and Kinki. Again, the intra-regional variability, as measured by the 5th–95th percentiles, was much larger than the differences in the regional means. In all regions, heating contributed more emissions than cooling did.

To quantify the drivers of the wide intra-regional variability observed in Figure 3, we focused on the 5.6 kW units installed in the Kanto region (2020, n = 2322), which had the largest sample size with JIS C 9612 test conditions based on Tokyo’s climate. We classified units into four quadrants using the empirical 75th percentiles (upper quartiles) of annual operating hours and annual electricity consumption to define “high” values. Quantile-based thresholds are non-parametric and are less sensitive to extreme tail values than mean–standard-deviation cutoffs, which is advantageous for skewed consumption-related variables. The upper-quartile cutoff (top 25%) provides an interpretable separation of the upper tail while retaining sufficient sample sizes in each quadrant for stable comparisons. We then evaluated the LCA results for each cluster (Figure 4).

The scatter plot in Figure 4a shows the relationship between the annual operating hours and annual electricity consumption. The regression line $y=0.34x$ approximates this relationship through the origin, with a coefficient of determination of $R^2=0.545$. Using the 75th percentiles of annual operating hours and annual electricity consumption as thresholds, we defined the four quadrants as follows: first quadrant—high use, high consumption; second quadrant—low use, high consumption; third quadrant—low use, low consumption; and fourth quadrant—high use, low consumption. This classification yielded 373 units in the first, 208 in the second, 1533 in the third, and 208 in the fourth quadrants. Even at similar levels of annual operating hours, the annual electricity consumption varied widely, and many units with similar annual consumptions exhibited very different operating hours. These results suggest that even within a single region, differences in occupant operating behavior play a significant role in driving electricity use variation.

Figure 4b presents the life-cycle GHG emissions over 10 years based on the JIS-based annual electricity consumption, regional mean consumption for Kanto (region average), and quadrant-specific annual electricity consumption for each cluster. The JIS-based GHG emissions (9795 kg-CO₂ eq) exceeded both the region average and all cluster means, yielding values roughly 1.5–2 times higher than those based on the actual usage. Among the clusters, the first and second quadrants showed the highest emissions, at 6985 and 6465 kg-CO₂ eq, respectively. The third quadrant had the lowest emissions at 2977 kg-CO₂ eq, whereas the fourth quadrant lay in between at 4480 kg-CO₂ eq.

Across all cases, the use phase, particularly the heating operation, dominated the life-cycle emissions, whereas the contributions from raw material acquisition, manufacturing, and refrigerant-related processes were relatively small. A notable feature was the apparent difference in the heating emissions between quadrants with similar operating hours (first vs. fourth quadrants and second vs. third quadrants). This suggests that differences in heating practices have a substantial impact on cluster-level GHG emissions.

3.2. Cluster-Specific Operating Profile Analysis

To characterize the detailed operating behavior of the four quadrant clusters, we analyzed the 2020 annual operating profiles based on the daily mean values (Figure 5). In each panel, the blue line represents the average daily electricity consumption. The red, green, and yellow lines represent the mean daily outdoor, indoor, and thermostat setpoint temperatures, respectively.

In the first quadrant (high use and consumption), the electricity consumption was high throughout the year, with pronounced peaks in winter (January–March and December) and summer (July–September). In winter, the daily electricity consumption reached approximately 11 kWh, indicating intensive heating under high-heating-load conditions when the mean outdoor temperature falls below 5 °C. In summer, continuous cooling was also observed when outdoor temperatures reached around 30 °C, but the peak daily consumption was somewhat lower than that in winter, with a maximum of approximately 7 kWh.

The second quadrant (low use, high consumption) exhibited high electricity consumption in winter. Daily consumption of approximately 10–12 kWh was observed in January–March and December, whereas summer cooling use was limited, and consumption in July–September was lower than that in the first quadrant. A particularly notable feature is that indoor temperature remained approximately 5 °C below the thermostat setpoint for extended periods in winter, suggesting that the combined effects of building-side characteristics, such as insulation performance, and operating patterns, such as intermittent use, lead to high-load heating operation under severe heating conditions, thereby driving up annual electricity consumption.

In the third quadrant (low use, low consumption), electricity consumption was low throughout the year. In both winter and summer, the average daily consumption remained at approximately 2–5 kWh, and during the shoulder seasons (April–May and October), it was close to zero. This period broadly corresponds to the non-operating season assumed in the JIS standard (mid-April to late May and early October to early November), suggesting that users in this cluster may operate their units in a manner similar to the usage period envisaged by the JIS standard. Although some deviation of the indoor temperature from the thermostat setpoint was observed in both winter and summer, the lower electricity consumption compared with that of the second quadrant suggests that factors such as wider comfort-tolerance ranges or the use of alternative equipment may also influence behavior.

In the fourth quadrant (high use, low consumption), continuous operation over long periods was observed, particularly in the summer. However, the daily electricity consumption was maintained at approximately 6 kWh, and winter consumption was clearly lower than those in the first and second quadrants. In contrast to the second quadrant, the gap between the indoor temperature and thermostat setpoint was small, and abrupt fluctuations in electricity consumption were rare. Taken together, these characteristics suggest that this group maintains relatively low annual electricity use through long-duration, low-load operations throughout the year.

3.3. Sensitivity Analysis of LCA Results

Figure 6 presents tornado charts showing the sensitivity of 10-year GHG emissions to variations in the key parameters for each quadrant cluster.

Except for the third quadrant, the heating operating hours exhibited the most significant sensitivity. In the high-usage first and fourth quadrants, the sensitivities were enormous, where the life-cycle GHG emissions varied by approximately −97% to +110% relative to the baseline. This indicates that even within each quadrant cluster, the variability in operating hours can substantially affect the LCA results, underscoring the need for mitigation measures tailored to each cluster’s usage characteristics.

The temperature difference between the thermostat setpoint and indoor air during cooling and heating also showed significant sensitivities across all quadrants, where relative changes in the GHG emissions ranged from −70% to +106%. This suggests that even with the same annual operating hours, maintaining stable indoor temperatures can significantly reduce long-term GHG emissions. In the second quadrant (low use, high consumption), the sensitivity of ΔT during heating (ΔT heating) was higher than that during cooling, unlike in the other quadrants. This implies that operating under high heating loads, when the indoor temperature falls far below the setpoint, increases electricity consumption per unit time, thereby increasing annual GHG emissions.

By contrast, the sensitivity to the grid CO₂ emissions factor was moderate in all quadrants. When the factor was varied within a realistic range for current technology (0.362–0.601 kg-CO₂/kWh, the life-cycle GHG emissions changed by approximately −13% to +36%. This confirms previous findings that differences in the generation mix and penetration of renewable energy influence the LCA results for air conditioners, while also showing that the sensitivity is lower than that associated with occupant behavior and operating conditions (operating hours and ΔT). Therefore, in the short term, improvements in usage behavior offer greater mitigation potential.

The sensitivities to the equipment lifetime, refrigerant leakage rate, and refrigerant recovery rate were comparatively small in all quadrants. Varying the lifetime within the range of 6–14 years changed the GHG emissions by only approximately −13% to +10%, and varied the refrigerant leakage rate (1–10%) and recovery rate (0–80%) results by roughly −9% to +5%. This is reflected in Figure 4b, which shows that contributions from upstream processes, such as material production and unit manufacturing, as well as refrigerant-related processes, were small compared with those from electricity-related emissions during the use phase.

4. Discussion

4.1. Comparison Between JIS Values and Big-Data-Based Measurements

In this study, we compared the standard usage assumptions defined in JIS C 9612:2013 (JIS-based scenario) with the empirical usage distributions derived from measurements of more than 4100 residential air conditioners. JIS C 9612:2013 defines annual electricity consumption and the annual performance factor (APF) based on Tokyo weather conditions, an average wooden house, and a load profile, with fixed operating hours as specified in Table 2, without explicitly accounting for regional variation or occupant behavioral diversity [32].

Figure 2 and Figure 3b show that the 10-year GHG emissions under the JIS-based scenario are consistently higher than the regional averages and quadrant-specific representative values obtained in this study, reaching approximately twice the regional average for the 5.6 kW class in the Kanto region. This is because the JIS scenario assumes long operating hours and high loads for cooling and heating, along with a relatively high heating demand, uniformly across all cases [32]. Similarly, Wan et al. noted that many LCA studies assumed steady-state design conditions and did not adequately reflect the empirical distribution of operating hours observed in the field [21].

Therefore, our results highlight the need to clearly distinguish between the JIS-based scenario and the distributions of actual usage derived from big data measurements and to use them selectively according to the assessment purpose. The JIS-based scenario remains appropriate for policy evaluations such as equipment labeling and setting performance standards. From a policy perspective, using only a standardized scenario can mask regional and behavioral heterogeneity and can misestimate the mitigation potential of demand-side measures for specific user segments. Empirically derived usage distributions can complement standards by supporting region- and segment-specific program design and by enabling an ex post evaluation of whether interventions shift operating patterns toward lower-emission profiles. Therefore, standards-based scenarios and big-data-based profiles should be treated as complementary tools for compliance-oriented assessments and mitigation planning. However, to estimate a realistic mitigation potential and design behavioral interventions, using quadrant- and region-specific representative profiles, as developed in this study, allows practitioners to capture population-level mean impacts and their variability in a way that faithfully reflects real-world usage.

4.2. Comparison with Previous LCA Studies of Air Conditioners

In the JIS-based scenario and quadrant-specific LCA results of this study, the use phase (cooling and heating operations) accounted for most of the total life-cycle GHG emissions. In particular, in the high-load quadrants 1 and 2, the 10-year emissions ranged from 6000 to 7000 kg-CO₂ eq (Figure 2 and Figure 3b), which fell within the mid-range of the GWP values reported by Litardo et al. for grid-connected systems [7].

As summarized in Table 1, many previous studies considered only one or a small number of usage scenarios based on test standards or building energy simulations, and occupant behavioral heterogeneity was treated merely through scenario comparisons or sensitivity analyses [6,13,14,15,16,22,26,28]. Moreover, although numerous studies have shown that electricity consumption during the use phase dominates the environmental impacts of air-conditioning systems, only Chou et al. have explicitly analyzed the breakdown between cooling and heating use [22]. Studies that evaluate the renewable energy supply or recycling benefits, such as those by Longo et al. and Byrd et al., also assumed a single design condition [53,54]. By contrast, this study defined four quadrants within the same region and the capacity class based on annual operating hours and electricity consumption. It presents quadrant-specific representative values and sensitivity structures for the life-cycle GHG emissions. Consequently, we quantitatively demonstrated that even under common generation-mix and equipment-performance conditions, differences in usage behavior can cause per-unit 10-year GHG emissions to vary by approximately a factor of two. We also showed that the main drivers were heating hours and thermostat–indoor temperature differences, underscoring the importance of occupant behavioral changes for mitigation.

4.3. Influence of Occupant Behavior and Operating Patterns

As shown in Figure 3a and Figure 5, big data analysis revealed substantial variations in annual operating hours and electricity consumption, even within the same region and capacity class. Quadrant 1 corresponds to typical high-load households that use cooling and heating for long hours throughout the year, with the daily electricity consumption in winter reaching approximately 11 kWh. In quadrant 2, a similarly high daily consumption of 10–12 kWh was concentrated during the heating season, whereas the summer cooling use was limited. Quadrant 3 comprised low-use households that used their air conditioners mainly during the peak summer and winter periods, with operations in the shoulder seasons being close to zero. Quadrant 4 was characterized by long, continuous operation primarily in summer, yet daily consumption was generally kept around 6 kWh, and winter consumption was clearly lower than that in quadrants 1 and 2; this group can be interpreted as practicing “long-duration, low-load operation” to suppress annual electricity use.

These diverse operating patterns are consistent with existing research on occupant behavior. Field-monitoring studies in Japan and other regions have reported significant household-to-household differences in on/off control and thermostat setpoints, even under similar outdoor temperature conditions, driven by factors such as time spent at home, energy-saving attitudes, and the presence of alternative heating equipment [17,20,38,41,55,56,57,58,59,60]. For example, a monitoring study of AC operation in 20 Japanese dwellings during the cooling season showed that on/off decisions depend not only on the indoor temperature but also on daily activities and individual comfort tolerance [55]. Furthermore, a big data study of typical residential AC-usage patterns in China showed that operating characteristics, such as thermostat setpoints and operating hours, differ by installation location, including bedrooms versus living rooms [41]. Therefore, it is likely that some units classified in quadrant 3 in our dataset corresponded to bedroom units used only at night, resulting in shorter annual operating hours despite substantial cooling or heating during these limited periods.

4.4. Operating-Pattern-Specific Low-Carbon Strategies

The sensitivity analysis identified the heating operating hours as the most influential factor that affected GHG emissions (Figure 6). In other words, as a short- to medium-term mitigation option, it is more rational to prioritize controlling the heating duration and the temperature difference ΔT than to focus on refrigerant management or eco-design. However, mechanically shortening heating hours can excessively lower indoor temperatures and increase heat- and cold-related health risks. Therefore, simple reductions in operating time alone are not sufficient.

Against this background, it becomes necessary to prioritize mitigation options by quadrant, focusing on “how to control heating hours and ΔT.” In quadrant 3, air conditioner use is low year-round and GHG emissions are minimal, which is desirable from an energy-saving perspective. Nevertheless, if the thermal environment becomes overly severe, health risks may increase. Therefore, the goal is not to shift all users uniformly into quadrant 3. Instead, it is important to appropriately widen the acceptable comfort range while avoiding heatstroke- and cold-related events, such as heat shock.

Quadrant 1 is characterized by high loads in both summer and winter. Hence, the priority is to suppress extreme winter peaks while lowering the heating setpoint temperature within a range that preserves thermal comfort. Concentrating operations during occupied periods using timers and scheduling functions, combined with envelope retrofits, such as improved insulation and reduced air leakage, can slow the rate of indoor temperature decay. This allows the same comfort level to be maintained with a smaller ΔT and shorter peak loads. In the winter-dominated quadrant 2, the key is to avoid “high-load heating under conditions where indoor temperature has dropped far below the thermostat setpoint,” as observed in Figure 5. Passive measures, such as upgrading to high-insulation windows, installing interior storm windows, and improving curtain and floor insulation, can improve both the warm-up behavior and stability of the indoor temperature. As a result, ΔT and the required heating duration can be reduced [61,62,63]. Quadrant 4, by contrast, was characterized by long continuous operation with a small ΔT and stable indoor temperatures and can be regarded as one of the target operating modes for the high-load quadrants. Shifting toward continuous operation with low heating and cooling loads, achieved through improved insulation and the use of a building’s thermal mass, can be a rational strategy for maintaining comfort while reducing GHG emissions. Quadrant 4 includes cases in which units operate continuously under mild outdoor conditions or during unoccupied periods. Given that future warming is projected to reduce the heating demand while potentially increasing the cooling demand [64], cooling-side measures that further streamline operation to continuous use only during necessary periods are also important, for example, through window opening and natural ventilation or the use of fans and circulators.

Overall, the quadrant classification and sensitivity analysis in this study clarified a typology of priorities: in quadrants 1 and 2, shifting from high-peak loads to low-load continuous operation; in quadrant 4, optimizing operating hours; and in quadrant 3, ensuring appropriate use by considering health risks. In the future, demand-side measures that combine thermal retrofits, tariff designs, and behavioral interventions should be designed for each quadrant. The quadrant-specific recommendations discussed in this section are derived from the observed differences in operating profiles and the associated life-cycle GHG emissions across the quadrants, as well as the sensitivity analysis results. We did not conduct a dedicated thermal comfort assessment or occupant surveys; therefore, we do not claim that any specific thermostat setpoint or operating mode is optimal for comfort. The term “comfort” in this section should be interpreted as maintaining acceptable indoor environmental conditions while avoiding heatstroke- and cold-related events, rather than the result of quantified comfort optimization.

4.5. Limitations and Future Directions

This study had several methodological limitations. First, we used region-specific annual-average CO₂ emission factors for electricity and did not account for temporal variations in the generation mix across seasons or times of day. In the LCCP literature and building LCA studies, dynamic assessment approaches that treat the carbon intensity of the electricity grid at a temporal resolution have been proposed. These studies have highlighted the importance of evaluating whether peak heating coincides with system-wide emission peaks [21,65]. Incorporating dynamic emission factors into the quadrant-specific profiles developed in this study would allow us to evaluate additional differences in LCA results arising from “when” air conditioners are used, as well as the effects of the demand response.

Second, the refrigerant model assumed constant annual leakage and recovery rates, and therefore, did not sufficiently reflect actual failures, improper handling, or regional and model-specific differences. Litardo et al., Wan et al., and others have noted that assumptions about refrigerant management introduce substantial uncertainty into LCA/LCCP results [7,14,21,49,66], and specifying leakage and recovery rates based on empirical data remains an important task. After conducting an LCA that compared refrigerant reclamation and destruction, Yasaka et al. reported that the reclamation scenario results in approximately 5.7–15.9 kg-CO₂ eq lower emissions per kilogram of refrigerant [49]. As operational reductions in the use phase are achieved and refrigerant-related emissions become relatively more important, optimization of the entire refrigerant life cycle, in combination with product LCA, should also be considered. In practice, annual leakage and end-of-life recovery rates can vary with maintenance practices, improper handling, and failure events, as well as with regional and model-specific conditions, whereas our model applies uniform values. Therefore, refrigerant-related contributions should be interpreted as scenario-based estimates under the stated assumptions and are not intended as precise predictions for individual units.

Third, the detailed behavioral analysis in this study focused on data from a specific year and a single air-conditioner manufacturer; thus, the generalizability of the findings across different countries, years, and equipment types is not fully verified. In particular, 2020 coincided with COVID-19-related lifestyle changes (e.g., increased time spent at home), which may have influenced patterns of residential HVAC operation [67,68]. Moreover, differences in product design, model lineups, and embedded control strategies across manufacturers may influence operating patterns and electricity use; therefore, the generalizability of the absolute distributions and quadrant boundaries to other brands and model generations is not fully verified. Previous studies have shown that the climate zone, building insulation performance, air-conditioner penetration, and electricity generation mix strongly affect LCA results [8,16,55,65]. In addition, the present results should be interpreted as evidence of within-class behavioral variability and its life-cycle implications for the 5.6 kW segment, while the transferability of absolute levels and quadrant boundaries to other capacity classes remains to be tested. Therefore, the next step toward generalizing big-data-driven LCA is to apply the quadrant classification and sensitivity analysis framework developed here to other regions, capacity classes, years, and manufacturers, and to examine the extent to which distributional patterns and sensitivity structures are shared across them.

Fourth, the quadrant-based typology is a rule-based segmentation defined by a fixed percentile cutoff applied to two annual summary variables. While interpretable, it may not capture complex usage patterns, such as the heating–cooling balance, setpoint–indoor temperature gaps, or seasonal operation. Group assignments can vary with feature selection, cutoff choice, and clustering method. Future work should compare this typology with alternative unsupervised methods, such as k-means, hierarchical clustering, and model-based clustering. Because the quadrant boundaries depend on the chosen cutoff, assessing robustness across a reasonable range of percentile cutoffs, for example by comparing the 70th, 75th, and 80th percentiles, is an important direction for future work.

Finally, the sensitivity analysis used an OAT design, varying each parameter independently within the ranges in Table 4 while keeping others fixed. This approach is useful for screening and emphasizing cluster-specific priorities but explores only a limited portion of the input space and cannot fully capture non-linearities or interactions. Using variance-based global sensitivity analysis tools, such as Sobol indices, would offer a more comprehensive uncertainty assessment and is a valuable future extension.

5. Conclusions

This study developed a data-driven LCA framework that explicitly reflects the real-world use of residential room air conditioners in Japan using measured operational data from approximately 4100 units nationwide. By comparing this framework with a standard scenario based on the JIS test values, we examined the representativeness of JIS-based assessments and the importance of incorporating actual usage patterns. For the 5.6 kW class, the annual electricity consumption calculated from JIS C 9612:2013 was approximately 2.7 times higher than the nationwide measured average, and the 10-year GHG emissions were consistently higher than both the regional averages and quadrant-based representative values. These results indicate that JIS-based evaluations function as conservative upper bounds but do not adequately capture the actual mitigation potential or behavior-driven variability. While the regional mean values fell within a relatively narrow range, usage distributions within each region were wide, demonstrating that intra-regional behavioral diversity rather than inter-regional differences dominated the uncertainty in the LCA results.

Focusing on the Kanto region, we further clustered the units into four quadrants based on annual operating hours and electricity consumption. Even within the same region and capacity class, the 10-year GHG emissions per unit differed by a factor of up to two. The results showed that electricity use during cooling and heating operations was the dominant contributor to life-cycle GHG emissions and that differences in heating practices across clusters had a measurable impact on emissions. The sensitivity analysis revealed that the heating operating hours and temperature difference between the thermostat setpoint and indoor air during cooling and heating exhibited the highest sensitivities. In contrast, the influences of grid emission factors, product lifetime, refrigerant leakage, and recovery rates were relatively small. These findings concretize the commonly cited conclusion of previous LCA/LCCP studies that “the use phase is dominant” by expressing it as cluster-specific variability and sensitivity structures based on big data. They suggested that reducing heating hours and optimizing thermostat settings in high-load quadrants are key levers for low-carbon operations that do not require equipment replacement.

This study highlights the importance of using JIS-based scenarios and region- and behavior-specific profiles derived from big data in a complementary manner. The former is well suited for policy applications, such as inter-product comparisons, energy labeling, and setting performance standards. In contrast, the latter is more appropriate for designing demand-side measures and evaluating the mitigation potential through behavioral change, accounting for population-level mean impacts and variability. In practice, the quadrant framework can be implemented using annual operating hours and annual electricity consumption that are often available from smart meters or device logs. This enables stakeholders to segment heterogeneous users into a few interpretable groups and to prioritize interventions, such as targeted guidance for high-load heating segments and retrofit support for households with large setpoint–indoor temperature gaps. The same framework can also be used to track whether programs shift operating patterns toward lower-emission profiles over time.

Extending the proposed big-data-driven LCA framework to incorporate dynamic emission factors, measured data on refrigerant management, and case studies in other countries is expected to provide practical guidance for sustainable air-conditioning use that contributes to both climate change adaptation and mitigation.