Parameter Optimization for Climate-Resilient IEQ Assessment: Validating Essential Metrics in the PICSOU Framework Across Divergent Climate Zones

Abstract

To enhance the climate adaptability and diagnostic precision of university sustainability frameworks, this study presents a critical advancement to the PICSOU (Performance Indicators for Core Sustainability Objectives of Universities) framework’s Indoor Environmental Quality (IEQ) module. The research employs a comparative approach across two distinct climate zones: the campus of Chengdu Jincheng College in a humid subtropical climate (CDJCC; Köppen Cwa) with natural ventilation, and the campus of Tallinn University of Technology in a temperate climate (TalTech; Köppen Dfb) with mechanical ventilation. A key innovation at CDJCC was the deployment of a novel, integrated sensor that combines a Frequency-Modulated Continuous Wave (FMCW) radar module for real-time occupancy detection with standard IEQ sensor suite (CO₂, PM_2.5, temperature, humidity), enabling unprecedented analysis of occupant-IEQ dynamics. At TalTech, comprehensive IEQ monitoring was conducted using standard sensors. Results demonstrated that mechanical ventilation (TalTech) effectively decouples indoor conditions from external fluctuations. In contrast, natural ventilation (CDJCC) exhibits strong seasonal coupling, reflected by a Seasonal Ventilation Efficacy Coefficient (${\lambda }_{\mathit{season}}$), indicating that seasonal differences in effective ventilation are present but vary by indoor space type under occupied conditions. Consistent with this stronger indoor–outdoor linkage, PM_2.5 infiltration was also pronounced in naturally ventilated spaces, as evidenced by a high infiltration factor ($I/O$ ratio) that remained consistently elevated. This work conclusively validates a conditional, climate-resilient workflow for PICSOU’s IEQ category, integrating these empirical coefficients to transform its IEQ assessment into a dynamic and actionable tool for optimizing campus sustainability strategies globally.

1. Introduction

1.1. Current Status of Sustainability Assessment in University Campuses

University campuses, as core sites for knowledge innovation and talent development, hold significant strategic importance in global sustainable development and carbon emission reduction. University campuses typically contain large building stocks and complex end-use energy profiles [1,2,3]. Combined with high population density, these characteristics can make campuses a non-trivial contributor to urban carbon emissions [4,5,6,7]. Higher education institutions (HEIs) have increasingly contributed to carbon-neutrality agendas through systematic campus sustainability actions [5,8,9]. These actions commonly include green-campus programs, energy management, renewable energy deployment, and low-carbon mobility measures [1,2,3,6]. Beyond operational decarbonization, universities exert broader influence through education, research, and public engagement [7,10,11]. This institutional influence supports the diffusion of sustainability concepts and can shape pro-environmental social behaviors [5,6,12,13]. From a broader perspective, HEIs are now recognized as ideal testbeds for implementing the UN Sustainable Development Goals (SDGs), and are expected to embed sustainability principles into core strategies, curricula, and organizational culture rather than treating them as peripheral add-ons [7,12]. As forerunners of transformation, universities shape campus populations’ habits and societal mindsets in ways that can either accelerate or impede sustainable development [7].

From an energy-systems perspective, university campuses operate as diversified, high-intensity loads within urban grids, combining laboratory baselines, extended-hours teaching spaces, residential services, and shared infrastructures (e.g., district heating). This end-use portfolio produces pronounced schedule-driven peaks alongside persistent baseloads, while long asset lifecycles and centralized procurement create path dependencies in HVAC, envelopes, and controls that lock in energy-use intensity and emissions trajectories for decades. At the same time, campuses offer distinctive potential for demand flexibility—load shifting, demand response, and progressive electrification of heat—provided interventions respect academic calendars and IEQ constraints. Consequently, credible governance must move beyond design intent toward measured performance: end-use disaggregation, weather-normalized baselines, occupancy-aware setpoints, and lifecycle cost analysis that links operational decisions to verifiable energy and carbon outcomes [14,15,16]. Recent zero-carbon campus action plans reinforce this perspective by explicitly treating campuses as a small-scale city model and proposing scalable frameworks for energy and emissions accounting that can be applied across campus types, geographies, and climatic contexts [17].

Assessment systems for university campuses have evolved from green building rating schemes such as LEED (Leadership in Energy and Environmental Design) and BREEAM (Building Research Establishment Environmental Assessment Method) [10,18,19] to sustainable campus ranking systems like UI GreenMetric and STARS (Sustainability Tracking, Assessment and Rating System) [20,21,22] and further to post-occupancy evaluation (POE) and real-time monitoring [23,24,25,26,27]. However, many frameworks rely on overly complex indicator sets and suffer from data inconsistency and limited comparability across institutions [4,28,29]. In addition, their adaptability across diverse space types and climatic contexts remains insufficient, which constrains generalizability [30,31,32]. Importantly, IEQ is often underweighted or treated in a largely static manner, limiting support for climate-resilient operations [10,33]. Among these, LEED and BREEAM primarily focus on individual building design, often falling short in covering overall campus operations and activities, and involve high implementation and data collection costs [10,34]. UI GreenMetric and STARS employ extensive indicator sets suffering from issues of data inconsistency and poor comparability, making it difficult to quantify actual carbon reduction contributions [20,21,22]. Moreover, their IEQ-related items are largely static and policy-oriented (e.g., the existence of guidelines or certifications) and provide little explicit guidance on how assessment criteria, thresholds, or weights should be adapted across different climate zones, which limits their practicality for climate-resilient campus operation [17]. General corporate or supply chain frameworks like ISO 14001 and Global Reporting Initiative (GRI) are not optimized for the unique structure and functions of HEIs [14,31]. Recent campus case studies in arid and Mediterranean climates further highlight that sustainability and IEQ strategies must be explicitly tailored to local climatic and cultural conditions rather than relying on globally uniform indicators, with social infrastructure, green space, and hybrid ventilation strategies all needing climate-responsive design [35,36].

Aiming to resolve the limitations from the above-mentioned tools, the diagnostic framework of PICSOU (Performance Indicators for Core Sustainability Objectives of Universities) was identified to measure carbon footprint and socio-economic performance through six categories comprising approximately 20 core indicators while balancing universality and local adaptability, thereby facilitating cross-institutional comparison and cost–benefit analysis. Its structure enhances comprehension and implementation for university administrators and enables the quantification of improvement measures. By prioritizing areas with significant greenhouse gas emissions and social impacts, the framework improves the targeting and effectiveness of emission reduction strategies, serving as a robust tool for advancing campus carbon neutrality and sustainable development [37].

1.2. The Core Role of IEQ in University Sustainability

With the global advancement of sustainable development and healthy campus initiatives, IEQ in university settings has become an interdisciplinary research focus spanning architecture, environmental science, and public health. University students and staff spend more than 80% of their daily time indoors, including dormitories, classrooms, laboratories, and libraries [38,39,40]. In these settings, air quality, thermal comfort, lighting, and acoustics can directly affect health, well-being, and learning-related outcomes [14,24,41,42]. Improving IEQ is not only a matter of building design but also a critical component of educational equity, public health, and sustainable development [14,39,43]. Recent studies have further strengthened this evidence base by explicitly linking IEQ conditions to cognitive performance, productivity, and health outcomes. Experimental and review work in offices shows that combinations of thermal, acoustic, visual, and air-quality parameters can significantly affect attention, task performance, creativity, and perceived comfort, and that moving conditions toward high-performance IEQ ranges yields measurable gains in cognitive functioning [44,45,46,47,48]. Qualitative and survey-based studies in HEIs likewise report that suboptimal thermal, acoustic, and air-quality conditions undermine concentration, emotional state, and learning processes, whereas better IEQ and human-centered or biophilic design strategies are associated with higher satisfaction, fewer symptoms, and enhanced perceived academic performance [49,50,51,52].

In parallel, dynamic and model-based IEQ assessment frameworks have emerged that treat the indoor environment as a time-varying system rather than a static condition. A representative example is the ALDREN TAIL scheme, which operationalizes IEQ rating by linking category levels to the fraction of (occupied) time that measured conditions remain within the corresponding boundaries. In this approach, higher-quality categories are awarded when exceedances beyond the target boundaries are sufficiently rare, providing a transparent bridge between time-resolved monitoring data and categorical compliance statements [53]. Deep learning and other advanced time series methods, initially developed for environmental and industrial process monitoring, have shown strong capabilities in capturing nonlinear dynamics and forecasting multi-variable quality indicators [54,55,56,57]. In the educational context, human-centric AIoT-based IEQ modeling has been proposed to integrate dense sensor networks with deep learning algorithms and multimodal occupant feedback, enabling the prediction of multidimensional IEQ conditions and their potential impacts on occupant wellbeing in real time [25]. These developments illustrate a broader shift towards data-driven, predictive, and adaptive IEQ management, and further emphasize the need for campus sustainability frameworks to better reflect dynamic, occupancy-aware, and climate-responsive IEQ performance rather than relying solely on static indicators.

Given the significant influence of IEQ on health, academic performance, and sustainable development of university campuses, the IEQ module of the PICSOU framework naturally constitutes a key focus of current research. In the EU policy context, the Commission’s Technical Guidance for Technical Building Systems and Indoor Environmental Quality (EPBD recast; Articles 13, 23 and 24) emphasizes IEQ as an operational outcome supported by monitoring and inspection. The recommended core IEQ scope includes thermal conditions (e.g., indoor temperature), humidity, ventilation-related indicators (e.g., CO₂), and exposure to pollutants such as particulate matter (e.g., PM_2.5). Accordingly, the IEQ indicator set and reference categories adopted in this study (see Section 2.4.5) are designed to be consistent with this EU guidance, improving interpretability across building types and climate contexts [58].

However, the IEQ component of the PICSOU framework currently exhibits certain limitations. The IEQ section faces four main constraints:

• In the selection of pollutant indicators, the PICSOU framework mainly emphasizes ventilation rates and thermal comfort, overlooking critical pollutants such as CO₂ and PM_2.5. Extensive reviews and empirical studies have demonstrated that these pollutants are closely linked to health, cognition, and comfort, and often exceed standards in university spaces, including dormitories and classrooms [14,15,18,24,59,60].
• The use of static indicators in the PICSOU framework fails to capture dynamic processes such as actual occupancy, window-opening behaviors, and fluctuating occupant numbers. Research indicates that indoor air quality and thermal comfort are highly dependent on dynamic occupancy, ventilation behavior, and real-time control, particularly in high-density spaces like classrooms and dormitories where CO₂ concentration and PM_2.5 levels can fluctuate rapidly [14,16,25,59,61].
• Developed initially for a temperate climate, the PICSOU framework has limited adaptability across diverse climate zones. Similar to many existing studies, IEQ research has predominantly focused on single regions, lacking multi-climatic zone validation and adaptability analysis [24,26,61,62].
• The omission of distinctions among space types—such as dormitories, classrooms, and offices—in the PICSOU framework makes it difficult to address differentiated IEQ issues. Studies indicate that dormitories are prone to elevated CO₂ and PM_2.5; classrooms often experience CO₂ accumulation and inadequate thermal comfort. Different spaces therefore present distinct problems and optimization needs, requiring category-specific management and evaluation [23,24,27,43,62].

This study aims to address the current limitations of the PICSOU framework in IEQ assessment by: (1) expanding beyond the narrow range of pollutant indicators to comprehensively include key health and comfort factors such as thermal comfort, CO₂ and PM_2.5; (2) moving beyond static evaluation methods through the introduction of dynamic monitoring and real-time occupancy analysis; (3) transcending its temperate climate origins by validating its applicability in subtropical climatic conditions; and (4) refining space-type classifications such as dormitories, classrooms, and offices to better reflect campus environmental diversity.

Targeting these four issues, the study conducts a specific enhancement and optimization of the IEQ module within the PICSOU framework, incorporating multi-pollutant monitoring, dynamic assessment methods, cross-climate applicability, and space-type differentiation, thereby providing a more scientific, universally adaptable, and operationally viable pathway for sustainable campus development. The methodological workflow, from monitoring design to IEQ optimization, is illustrated in Figure 1

2. Materials and Methods

To establish the universality of the PICSOU framework across divergent climatic and cultural contexts, particularly in its IEQ category, a parallel IEQ monitoring campaign was conducted during the 2024–2025 academic year at two university campuses: Chengdu Jincheng College (CDJCC), China, and Tallinn University of Technology (TalTech), Estonia. The distinct environmental settings and methodological approaches employed at each site are summarized in

Table 1. Overview of the experimental design.

Aspect	CDJCC Campus	TalTech Campus
Site Location	Chengdu, China	Tallinn, Estonia
Climate Zone	Subtropical monsoon climate	Temperate oceanic/continental climate
Monitoring Period	15 calendar days in autumn semester, 2024; 15 calendar days in spring semester, 2025	90 calendar days in autumn semester, 2024; 51 calendar days in spring semester, 2025
Space Types	Dormitories, classrooms, offices	Classrooms/auditoriums/meeting rooms, offices
Occupancy Tracking Method	FMCW radar module for real-time human presence detection	N/A
IEQ Parameters Measured	CO2, PM2.5, temperature, relative humidity	CO2, PM2.5, temperature, relative humidity
Data Resolution	1 min intervals for all parameters	10 min intervals for all parameters
Primary Focus	Validating IEQ under subtropical conditions, focusing on thermal comfort and PM2.5 infiltration	Validating IEQ under temperate conditions, establishing baseline performance without occupancy influence
Connection to PICSOU	Field validation of the PICSOU’s IEQ category in a non-Nordic context, highlighting need for climate-specific adaptations	Baseline validation of PICSOU’s IEQ metrics under temperate climate in a pure mechanically ventilated context
Contribution to PICSOU	Demonstrated necessity for modular adjustments in PICSOU to account for regional challenges	Provided reference datasets for temperate climate zone, reinforcing PICSOU’s core IEQ metrics without occupancy complexities
Unique Challenges	High humidity (70–80%), winter PM2.5 peaks, lack of mechanical heating leading to cold stress	Standardized monitoring in controlled environments, but with potential gaps due to lack of occupancy correlation

, providing a foundational overview of the experimental design for the subsequent analysis.

2.1. Site Selection for Two Climate Zones

This study utilized locally available IEQ sensors to assess IEQ across the spring and autumn semesters at two sites—CDJCC and TalTech. The campuses lie in distinct climate zones, enabling cross-climatic conclusions of broader applicability.

2.1.1. CDJCC

Chengdu Jincheng College (CDJCC) is located in Pidu District, Chengdu—the fourth largest city in China, with a population of 21.4 million as of Q2 2025. Located in a subtropical monsoon climate (Cwa in the Köppen climate classification), the metropolis of Chengdu has distinct seasons characterized by hot, rainy summers, mild, damp winters and persistently high humidity. The CDJCC campus harbors 31,000 students and 1771 teaching staff, covers approximately 1.4 km² and includes a variety of indoor environments such as dormitories, classrooms, and offices. Ventilation operates in mixed-mode, consisting mostly of natural ventilation with supplemental air-conditioning running during summer and winter months when thermal comfort cannot be maintained.

During the autumn semester, we deployed 13 sensors across 6 dormitories, 2 offices, 4 classrooms, and 1 outdoor location. During the spring semester, we adopted a mostly identical sensor deployment scheme (changed only two office locations due to limited accessibility during the testing period) with 1 additional outdoor location for enhanced verification of outdoor air quality (Figure 2). In both the spring and autumn semesters, data were sampled at 1 min intervals, capturing real-time IEQ parameters including CO₂, PM_2.5, temperature, and relative humidity. Each semester comprised 15 days of measurements, accompanied by synchronized records of occupant presence detected by the same IEQ sensor at each location.

2.1.2. TalTech

Tallinn University of Technology (TalTech) is located in Tallinn, the capital of the Republic of Estonia. Situated on the country’s north coast, Tallinn has a population of approximately 440,000, the largest in the nation. The city sits in the transitional zone between temperate oceanic climate and temperate continental climate (Dfb in the Köppen climate classification)—mild and humid year-round, with evenly distributed precipitation, and significant differences in daylight hours between seasons thanks to its high latitude.

The university occupies 29 buildings with a total floor area of 108,312 m², serving 9100 students, 1240 faculty members and 1002 staff as of Q4 2024. Mechanical ventilation is used year-round across the campus; systems operate continuously, and district heating is activated during cold months (normally lasts from 1 October to 31 March; exact start and end dates may vary depending on weather conditions).

Leveraging the widespread deployment of permanently installed sensors on campus, we compiled a dataset from 13 sensors—7 classrooms/auditoriums/meeting rooms, 4 offices, and 2 outdoor locations (Figure 3) over extensive periods throughout the academic year. Measurements were logged at 10 min intervals, capturing core IEQ parameters: CO₂, PM_2.5, temperature and relative humidity. Due to the absence of an occupancy detection module within the sensors, no real-time records of occupant presence were documented on the TalTech site.

2.2. Sensors Specifications

2.2.1. Sensors at CDJCC

On the CDJCC site, we custom-built sensor units by integrating dedicated modules for IEQ measurement with human presence detection based on frequency-modulated continuous-wave (FMCW) radar into a single device. All sensors shared identical parameters and specifications (

Table 2. CDJCC sensor specifications.

Device	Model	Module	Accuracy	Resolution	Range
	air quality monitor AN-PCT (Rmikey PCB Co. Ltd., Shenzhen, China)	CO2	≤± 40 ppm, ±3% of reading	1 ppm	0–9999 ppm
		PM2.5	±10 μg @ 0–100 μg/m3 ±10% @ 101–500 μg/m3	1 μg/m3	0–999 μg/m3
		Temperature	≤± 0.2 °C	0.1 °C	0–65 °C
		Relative Humidity	≤± 3%	0.10%	0–99%
		Occupancy (FMCW)	N/A	N/A	±60 degrees, 6 m

). Installation locations in representative space types (4-person dormitory, 2-person dormitory, office and classroom) are shown in Figure 4.

2.2.2. Sensors at TalTech

On the TalTech site, IEQ monitoring is outsourced to specialized industrial partner. All sensors shared identical parameters and specifications; sensor specifications are summarized in

Table 3. TalTech sensor specifications.

Device	Model	Module	Accuracy	Resolution	Range
	airPurity indoor climate monitoring sensor (Thinnect OÜ, Tallinn, Estonia)	CO2	≤± 40 ppm, ± 3% of reading	1 ppm	10–40,000 ppm
		PM2.5	±10 μg @ 0–100 μg/m3 ±10% @ 101–1000 μg/m3	1 µg/m3	0–1000 μg/m3
		Temperature	≤± 0.8 °C	0.1 °C	−10–60 °C
		Relative Humidity	≤± 6%	0.10%	0–100%

, and installation locations in representative space types (meeting room/classroom/auditorium and office) are shown in Figure 5.

2.3. Data Collection and Preparation

2.3.1. Data-Period Designation

This study compares campus IEQ performance under different ventilation modes within various climate zones. Accordingly, we analyzed monitoring data from representative autumn and spring semesters at two universities.

At CDJCC, the typical autumn semester (September–January) and spring semester (February–June) were each monitored for 15 days (during the autumn semester: 17 December–31 December; during the spring semester: 9 May–23 May). In total, we collected over 3.8 million records, comprising 13 datasets in autumn semester and 14 datasets in spring semester, with each dataset containing 21,600 time-steps and 5 attributes (CO₂, PM_2.5, temperature, relative humidity and real-time occupancy status).

At TalTech, the typical autumn semester runs September–December and the spring semester February–June. For analysis, in order to fully overlap with teaching activities and the district-heating run time, we defined a winter period of 22 September–20 December, we also defined a summer period of 2 May–21 June to coincide with teaching and mechanical cooling, the period between January and April was omitted as it is the extended heating season during the spring semester. We compiled over 1.52 million records in total, spanning 55 spring datasets and 65 autumn datasets, where each dataset contained a single-indicator time series.

2.3.2. Data Processing

Prior to conducting the comparative analysis, we subjected all raw monitoring data from the two campuses to systematic cleaning and structural preprocessing to ensure temporal completeness and the reliability of subsequent statistical procedures.

For the CDJCC dataset, where each exported record contains multiple monitored parameters associated with a single timestamp, a complete 1 min time series was first generated for each designated monitoring period. All collected records were then Boolean-matched to this full-period sequence to ensure accurate alignment at every time point. Based on the aligned data, we further quantified the overall data coverage separately for the spring (corresponding to the summer monitoring period) and autumn (corresponding to the winter monitoring period) semesters. In the spring semester, 302,400 theoretical timestamps were expected, among which, 273,461 valid entries were recorded and 28,939 were missing, yielding a loss rate of 9.57%. In the autumn semester, 280,800 timestamps were expected, with 263,175 valid entries and 17,625 missing, corresponding to a loss rate of 6.28%. Most lost points were attributed to accidental sensor power interruptions or operational mishandling. To avoid artificial smoothing effects that could bias the interpretation of tail behavior in duration-curve analysis, all missing values were retained as NaN without interpolation or model-based reconstruction.

The TalTech dataset differs fundamentally from CDJCC in structure, as each timestamp corresponds to only one IEQ parameter, forming a set of single-indicator time series. Based on the three key parameters required for subsequent analysis (CO₂, PM_2.5, and temperature), all relevant records were extracted from the operator-provided database for both the winter (corresponding to the autumn semester) and summer (corresponding to the spring semester) monitoring windows, and the seasonal data availability was quantified accordingly. The results reveal substantial seasonal and inter-indicator variation in data coverage. During the winter monitoring period, the loss rates were 26.82% for PM_2.5, 22.34% for CO₂, and 22.33% for temperature. In contrast, during the summer monitoring period, the loss rates were significantly lower, amounting to 4.87% for PM_2.5, 4.99% for CO₂, and 4.99% for temperature. These missing values reflect the intrinsic recording limitations of the single-indicator monitoring system across seasons and parameters; therefore, all missing points were retained as NaN, without interpolation, so as to preserve the true coverage profile of each indicator.

Regarding extreme-value handling, all extreme observations were retained in the dataset rather than being removed. High or low IEQ values typically correspond to meaningful transient conditions—such as CO₂ accumulation under high occupancy, PM_2.5 ingress during window-opening events, or short-term ventilation or infiltration fluctuations—and thus contain valuable diagnostic information. Because the duration curve is a central analytical tool in this study, and its interpretive power relies on retaining the full distribution, especially the upper and lower tails, discarding extreme values would directly weaken its ability to characterize the frequency and persistence of exceedances. It should also be noted that box plots were constructed using the 1.5 × IQR rule, which suppresses the display of outliers; therefore, retaining extreme values does not affect box-plot readability but is essential for ensuring accurate duration-curve-based assessments.

2.4. Analytical Method

2.4.1. Duration Curve

A duration curve (also called a duration–exceedance curve) orders a variable from high to low over the observation period and plots the proportion (or probability) of time the variable exceeds a given value, thereby providing an intuitive view of its distribution across time using Equation (1).

$$P\left(X\ge x_i\right)={\displaystyle \frac{m}{N+1}}$$

(1)

Here, $X$ denotes the monitored indicator (e.g., PM_2.5 level or CO₂ concentration); $x_i$ is the ith observation after sorting; $m$ is the number of samples greater than or equal to that observation; and $N$ is the total number of observations. Using $N+1$ avoids boundary probabilities of 0 or 1 and yields a more reasonable exceedance distribution. This probability reflects the share of time during which the indicator exceeds a given value and is used to plot the duration curve, thereby revealing the occurrence frequency of different concentration levels and their persistence over the monitoring period.

In this study, we applied the duration curve method to analyze three IEQ parameters—CO₂ concentration, PM_2.5 level, and temperature—separately for the spring and autumn semesters to reveal seasonal differences. For each parameter, observations collected during occupied periods were first filtered and then sorted from high to low; the corresponding exceedance probabilities were calculated to construct the duration curve. This approach statistically characterizes the relationship between a variable’s magnitude and the percentage of time it is equaled or exceeded, providing an intuitive view of the distribution and persistence of indoor air quality over time and facilitating the identification and comparison of extremes.

2.4.2. Occupancy Probability

The custom-built sensor units deployed at CDJCC are capable of monitoring IEQ data and real-time human presence simultaneously thanks to their multi-modular configuration. Since there has been no widely documented precedent of a single, integrated device that combines an FMCW radar module with a standard IEQ sensor suite (CO₂, PM_2.5, temperature, relative humidity) for simultaneous monitoring, it is crucial to conduct a comprehensive verification of the reliability of the FMCW radar module within the sensors. To this aim, in addition to examining the continuity of the time-stamped occupancy data, we also adopted the dimensionless quantity of occupancy probability to be used together with duration curves to better observe the correlation between the average IEQ values within a space type and the overall occupancy of the same space type. At any given timestamp, occupancy probability is designated as the proportion of the total number of occupied rooms over the total number of rooms within the same space type. For example, a 0.5 occupancy probability in classrooms denotes that human presence is detected in half of all the classrooms at the timestamp of observation. To solve for occupancy probability, we adopted Equation (2):

$$O_{drm/off/cls}={\displaystyle \frac{\sum _{i=1}^n{O}_i}{N_{drm/off/cls}}}$$

(2)

in which, $O_{drm/off/cls}$ is the occupancy probability at a given timestamp within one of the 3 space types at CDJCC (dormitory/office/classroom), n is the number of rooms with detected human presence within a specific space type, $O_i$ is a binary value designated as the Occupancy Detection Contributor, when $O_i$ = 1, human presence is detected in a room, when $O_i$ = 0, no human presence is detected in a room. $N_{drm/off/cls}$ is the total number of sensors dedicated to a specific space type. At CDJCC, $N_{drm}$ = 6, $N_{off}$ = 2, $N_{cls}$ = 4.

To solve for the average IEQ value within a space type at a given timestamp, we adopted Equation (3):

$$\overline{X}={\displaystyle \frac{\sum _{i=1}^{N_{drm/off/cls}}(C_i/P_i/T_i/H_i)\ast D_i}{\sum _{i=1}^{N_{drm/off/cls}}{D}_i}}$$

(3)

in which $\overline{X}$ is the average IEQ value from one of the 4 IEQ indicators (CO₂, PM_2.5, temperature and relative humidity) at a given timestamp within one of the 3 space types (dorm/office/classroom) represented by C_i, P_i, T_i and H_i. $N_{drm/off/cls}$ has the same definition as in Equation (2). $D_i$ is a binary value introduced as an error-proofing mechanism designated as the Denominator Contributor. When $D_i$ = 1, a sensor is functioning as intended and IEQ data is documented, whereas $D_i$ = 0 denotes a mishap with no data documented. Verification of reliability of the FMCW radar module was carried out during the autumn semester’s monitoring period at CDJCC, from which, a documented 6.28% of the total data were lost due to reasons such as mis-operation or/and accidental/unauthorized unplugging, in the event of an unpredicted data loss, $D_i$ helps automatically adjust the number of properly functioning sensors so that the average IEQ value is always correctly solved for.

2.4.3. Box Plot

Coined in 1977, box plot is a nonparametric visualization based on the five-number summary. It displays the minimum, maximum, median, and the lower and upper quartiles. It is used to describe the central tendency and dispersion of numerical data and to identify outliers [63].

For IEQ indicators such as PM_2.5 level and CO₂ concentration, box plots require no normality assumption and, in the presence of skewness and short-lived peaks, provide a robust way to compare typical levels and anomalous fluctuations across different spaces, periods, or operating conditions. Relevant equations are as follows (see Equations (4)–(11)):

$$x_{\left(1\right)} \le x_{\left(2\right)} \le \dots \le x_{\left(n\right)}$$

(4)

$$Q_1=quantile\left(x,0.25\right)$$

(5)

$$Q_2=median\left(x\right)$$

(6)

$$Q_3=quantile\left(x,0.75\right)$$

(7)

$$\mathit{IQR}=Q_3-Q_1$$

(8)

$$L_{\mathit{fence}}=Q_1-1.5\hspace{0.17em}\mathit{IQR}$$

(9)

$$U_{\mathit{fence}}=Q_3+1.5\hspace{0.17em}\mathit{IQR}$$

(10)

$$\mathit{outlier} \mathit{if} x_i<L_{\mathit{fence}} \mathit{or} x_i>U_{\mathit{fence}}$$

(11)

here, $Q_1$ is the lower quartile (25%), the median $Q_2$ is at the 50%, and the upper quartile $Q_3$ is at the 75%. The interquartile range $\mathit{IQR}=Q_3-Q_1$ describes the dispersion of the middle 50% of the data. $L_{\mathit{fence}}$ and $U_{\mathit{fence}}$ denote the whisker ends, the minimum/maximum observations. Outside the range of ${[L}_{\mathit{fence}},U_{\mathit{fence}}]$ is classified as an outlier.

To characterize IEQ across climate zones, seasons, and space types and to enable macro-level comparisons, this study employs box plots using Tukey’s 1.5 × $\mathit{IQR}$ rule—with whiskers limited to $Q_1 - 1.5\mathit{IQR}$, $Q_3 + 1.5 \mathit{IQR}$—to estimate typical levels and thereby reveal overall differences among spaces. Applying the 1.5 × $\mathit{IQR}$ criterion provides a more precise view of central tendency and spread for the overall picture. In view of inevitable extremes, outliers are not displayed to improve readability; these conditions are instead depicted using duration curves.

2.4.4. Occupation Determination

In this study, occupied periods at CDJCC were identified using the instruments’ built-in presence-detection module, which labeled the occupant-presence status at the corresponding timestamps.

In the campus buildings of TalTech, occupant presence was primarily inferred from the CO₂ time series. Outdoor CO₂ background levels typically fluctuate slightly within a narrow range; indoors, since human respiration is the main CO₂ source, a pronounced rise in CO₂ concentration within a given time window generally indicates that the space is occupied. Conversely, if the concentration remains stable or declines, the space is considered unoccupied. Indoor CO₂ is also affected by mechanical-ventilation operation and outdoor-background variability, and the collection of time-series data is subject to sensor acquisition delays, leading to a lag in the data timestamps relative to actual occupancy.

Based on the above principles, the collected time-series data can be analyzed using Equations (12) and (13) to determine whether occupants were present at each time step.

$$\Delta C\left(t\right)=C\left(t\right)-C(t-1)$$

(12)

$$O\left(t\right)=\left\{\begin{array}{cc}1,& \mathrm{if} \Delta C\left(t\right)>0 \mathit{and} C\left(t\right)>550\\ 0,& \mathit{otherwise}\end{array}\right.$$

(13)

In this formulation, let $\Delta C\left(t\right)$ denote the CO₂ concentration at time $\mathrm{t}$. The increment between adjacent time steps is defined as $\Delta C\left(t\right)=C\left(t\right)-C(t-1)$. On this basis, we construct an occupancy indicator $O\left(\mathrm{t}\right)$: assign “$O\left(t\right) = 1$” if and only if $\Delta C\left(t\right) > 0$ and $C\left(t\right) > 550$; otherwise assign “$O\left(t\right) = 0$”. This rule emphasizes increases relative to the preceding time step and incorporates an absolute threshold to reduce the risk of misclassification.

2.4.5. Identification of Critical IEQ Reference Values

Based on commonly used Chinese standards [64,65], relevant EU standards [66], and international standards [67]—together with recent findings on occupant health and comfort—and using the summer and winter thresholds specified in these standards, we determined the following key IEQ reference values (see

Table 4. Critical IEQ reference values.

Semester	Issuer	Standard	CO2, ppm	PM2.5, μg/m3	Temperature, °C	Relative Humidity, %
Spring	China	GB/T 18883-2022 [64]	1000	75 (24 h avg.)	Summer: 22–28	Summer: 40–80
		T/ASC 02-2021 [65]	1000	35 (Max. 5 days/1 y) or 15(1 y avg.)	N/A	N/A
	EU	EN 16798-1:2019 [66]	I: 950	N/A	I: 23.5–25.5	N/A
			II: 1200		II: 23–26
			III: 1750		III: 22–27
			IV: 1750		IV: 21–28
	WHO	AQG 2021 [67]	N/A	AQG: 5	N/A	N/A
				Interim 4/3/2/1: 10/15/25/35
Autumn	China	GB/T 18883-2022 [64]	1000	75 (24 h avg.)	Winter: 16–24	Winter: 30–60
		T/ASC 02-2021 [65]	1000	35 (Max.5 days/1 y) or 15(1 y avg.)	N/A	N/A
	EU	EN 16798-1:2019 [66]	I: 950	N/A	I: 21–23	N/A
			II: 1200		II: 20–24
			III: 1750		III: 19/18–25
			IV: 1750		IV: 17–25
	WHO	AQG 2021 [67]	N/A	AQG: 5	N/A	N/A
				Interim 4/3/2/1: 10/15/25/35

). This selection of core IEQ indicators and the category-based interpretation is also consistent with the EU Commission’s technical guidance on indoor environmental quality monitoring and assessment under the EPBD recast [58]. Values highlighted in bold are the thresholds used as benchmarks for the IEQ assessment in this study.

2.4.6. T-Distribution Confidence Intervals for Key Room-Level Indicators

In this study, Student’s t-distribution (Equation (14)) [68,69] is used in an estimation framework to quantify the uncertainty of two key conclusions, rather than solely for classical null-hypothesis significance testing. For each campus–season–space-type combination, room-level indicators, namely the seasonal CO₂ ratio ${\mathrm{\lambda }}_{\mathrm{season}}$ and the room-level PM_2.5 $I/O$ ratio, are treated as independent observations. Within each group, the parameter of interest is the mean value of the room-level indicator, and its 95% confidence interval is constructed using the t-distribution.

Specifically, for a generic room-level indicator $x$, the 95% confidence interval for the group mean is given by

$$\overline{x}\pm t_{0.975,n-1}\hspace{0.17em}{\displaystyle \frac{s}{\sqrt{n}}}$$

(14)

where $\overline{x}$ is the sample mean across rooms, $s$ is the sample standard deviation between rooms, $n$ is the number of rooms (or sensor points) in the group, and $t_{0.975,n-1}$ is the two-sided 97.5th percentile of Student’s t-distribution with $n -1$ degrees of freedom. This formulation explicitly accounts for the fact that the population variance is unknown and must be estimated from the sample, which is particularly important given the relatively small number of rooms in some campus–space-type groups.

The resulting t-distribution confidence intervals are then used to examine whether the empirically observed room-level indicators are consistent with the two main conclusions of this study (regarding the seasonal CO₂ ratios ${\mathrm{\lambda }}_{\mathrm{season}}$ and the PM_2.5$I/O$ behavior), following current recommendations that emphasize estimation and interval interpretation rather than reliance on point estimates or p-values alone [69].

3. Results

3.1. Reliance Verification of CDJCC Sensors’ FMCW Radar Module

We conducted reliance verification on the FMCW radar module from custom-built sensors used at CDJCC during the autumn semester’s monitoring period by plotting the duration curve of each space type’s average CO₂ value under different occupancy probabilities (see Figure 6).

Figure 6 indicates a positive correlation between average CO₂ concentration and occupancy probability across all space types, which aligns with the established principle that CO₂ level serves as a reliable proxy for human presence, although such correlation can be coincidental and does not necessarily suggest any direct causality between increased occupancy probability and increased average CO₂ value of occupied rooms within the same space type, as each occupied room’s peak CO₂ value can greatly vary due to ventilation efficiency and occupant behavior, the fact that all non-zero occupancy probabilities’ duration curves appear above that of the 0% occupancy probability suggests that the FMCW radar module: (1) was not producing grossly incorrect occupancy detection results; (2) produced occupancy detection that was at least directionally correct (occupied rooms have higher average CO₂ value than empty ones); (3) was not systematically misclassifying occupancy in a way that violates such fundamental physical relationship. Additionally, an anomaly is observed in dormitories, where the duration curves for 83% and 67% occupancy probabilities largely overlap and exceed the curve for 100% occupancy probability. This inverse relationship can be attributed to the small spatial volume of dormitories; higher occupancy probability increases the probability of door or window opening for natural ventilation, thereby reducing CO₂ accumulation. In contrast, larger spaces like offices and classrooms exhibit a stronger positive correlation, void of such anomaly, likely due to reduced ventilation interventions stemming from social inhibition among occupants. The overall realism of the duration curves validates the synchronous data collection of the FMCW radar module and its IEQ suite counterpart, confirming that the FMCW radar module was working as intended.

To further assess the methodological comparability between the FMCW radar-based occupancy detection at CDJCC and the CO₂-based occupancy inference used at TalTech, we conducted a targeted cross-validation using the CDJCC dataset. The analysis was restricted to the same space types as at TalTech, namely offices and classrooms, and to periods in the spring and autumn semesters when both FMCW radar–derived occupancy flags and indoor CO₂ measurements were simultaneously available. On this subset of data, we reproduced the TalTech rule-based occupancy model exactly: a room is classified as “occupied” if the indoor CO₂ concentration remains above 550 ppm over a 10 min interval, and the resulting binary occupancy status is inferred solely from the CO₂ time series without using any radar information. The CO₂-based occupancy labels were then temporally aligned with the FMCW-based occupancy labels, and all time steps with both labels available were used to quantify their agreement.

The cross-validation results can be summarized in two main points. For clarity, we report the summary statistics in

Table 5. Cross-validation of CO₂-based occupancy inference against FMCW-based occupancy detection at CDJCC for spring and autumn semester datasets.

Metric	Spring-Semester Dataset	Autumn-Semester Dataset	Spring + Autumn Combined
Number of rows with CO2-based occupancy prediction	18,396	7681	26,077
Fraction of predicted rows in growing phase	0.483	0.449	0.473
Fraction of predicted rows in decaying phase	0.517	0.551	0.527
Number of matching rows (prediction = FMCW)	11,177	4765	15,942
Matching ratio	0.608	0.62	0.611
Matching in growth phase	0.752	0.776	0.758
Matching in decay phase	0.473	0.494	0.479

and illustrate two representative episodes in Figure 7.

First, during “standard growing phases” of CO₂ (i.e., periods when indoor CO₂ exhibits a sustained increase without pronounced reversals), the CO₂-based occupancy estimates derived from the TalTech rule show a high level of agreement with the FMCW-based occupancy labels at CDJCC. This indicates that, for typical situations in which occupants enter the room and remain there for some time so that indoor CO₂ accumulates monotonically, the CO₂-based inference and the radar-based detection identify occupied states in a broadly comparable manner. These phases form the core of the clearly occupied periods that underpin the cross-campus analysis in this study.

Second, once indoor CO₂ no longer increases monotonically but instead exhibits plateaus, declines, or repeated oscillations, discrepancies between the two methods become much more pronounced. Under the mixed-mode ventilation conditions at CDJCC, window opening, door opening, and short absences tend to dilute indoor CO₂, so that even when people are still present, CO₂ may temporarily decrease or fluctuate substantially; when the number of occupants is small or the occupancy level varies rapidly, indoor CO₂ can also remain at comparatively low levels for extended periods. Because the TalTech rule requires CO₂ to remain above 550 ppm and to display a rising tendency over the 10 min interval, such “occupied but CO₂-decreasing or low-CO₂” episodes are systematically classified as “unoccupied” and thus omitted from the CO₂-based occupancy record. It should be noted that, although TalTech employs a purely mechanical ventilation system and therefore tends to exhibit smoother and more nearly linear CO₂ buildup during occupancy, periods with few occupants or rapidly changing occupancy can still produce oscillations or short-term decreases in CO₂, which are likewise prone to being excluded as “unoccupied” under the same rule.

Taken together, these findings suggest that the two campuses are methodologically comparable with respect to clearly occupied periods characterized by sustained CO₂ buildup: in such intervals, the TalTech CO₂-based method and the CDJCC FMCW-based detection provide consistent “occupied” classifications, and our cross-campus comparisons are intentionally based primarily on this robust subset of occupied states. Consequently, the moderate overall matching ratio reported in Table 5 should not be interpreted as low FMCW radar accuracy, but rather as a reflection of the conservative nature of the CO₂-based inference, which systematically excludes certain occupied intervals by design. At the same time, the cross-validation also reveals that the current TalTech CO₂-based approach structurally omits part of the occupancy dynamics, specifically those situations in which people are present but indoor CO₂ decreases or remains low due to strong ventilation or low occupant density. From a methodological standpoint, complementing CO₂-based inference with non-intrusive, direct occupancy sensing technologies such as FMCW radar in TalTech-type settings would enable a more complete representation of occupied periods spanning both CO₂ accumulation and decay phases, and would further strengthen the robustness of cross-campus occupancy comparisons.

3.2. Spring–Autumn Semester Comparisons for Each Campus

In this comparative analysis, the portion from Q₁ to Q₃ in the box plots represents the middle 50% of the data for each room type, reflecting the general IEQ level under typical conditions; we refer to this as the “typical interval.” The upper and lower whiskers represent the fluctuation range of IEQ indicators under special conditions for each space type, referred to here as the “extreme-value interval.”

In this study, the large data volume produced too many outliers in the box plots, making them difficult to read. The duration curve compensates for this limitation by more completely displaying each room’s special behavior during occupied periods.

3.2.1. Spring–Autumn Semester Comparisons at CDJCC

• CO₂

In the spring semester at CDJCC (Figure 8, Figure 9 and Figure 10 and

Table 6. Occupied-period CO₂, PM_2.5 and temperature (Med, IQR) with category-compliance labels based on the 5% criterion.

Campus (Semester)	Space Type	CO2, ppm (Med; IQR; Labels)	PM2.5, μg/m3 (Med; IQR; Labels)	Temperature, °C (Med; IQR; Labels)
CDJCC (Spring)	Dormitory	913; 509; outside of EU cat.III	49; 48; outside of CN limit	26.0; 1.8; CN limit
	Office	489; 160; EU cat.I	41; 50; outside of CN limit	25.9; 1.9; CN limit
	Classroom	552; 352; EU cat.III	52; 46; outside of CN limit	27.2; 2.4; outside of CN limit
	Outdoor	419; 72; N/A	57; 60; outside of CN limit	27.9; 7.4; N/A
TalTech (Spring)	Office	604; 84; EU cat.I	0; 1; WHO AQG	23.9; 2.2; EU cat.I
	Classroom	611; 159; EU cat.III	0; 1; WHO AQG	23.3; 1.8; EU cat.I
	Outdoor	481; 72; N/A	2; 6; WHO Interim 1	11.8; 6.0; N/A
CDJCC (Autumn)	Dormitory	789; 355; EU cat.III	96; 50; outside of CN limit	17.4; 2.8; outside of CN limit
	Office	474; 74; EU cat.I	83; 54.8; outside of CN limit	22.4; 4.1; outside of CN limit
	Classroom	388; 84; CN limit	124; 64; outside of CN limit	12.6; 2.3; outside of CN limit
	Outdoor	359; 52; N/A	130; 63; outside of CN limit	7.6; 2.8; N/A
TalTech (Autumn)	Office	604; 98; EU cat.I	0; 1; WHO AQG	21.8; 1.7; EU cat.III
	Classroom	621; 191; EU cat.III	0; 1; WHO AQG	21.7; 1.7; EU cat.III
	Outdoor	496; 52; N/A	0; 4; WHO Interim 1	4.9; 5.4; N/A

), the median CO₂ concentrations for dormitories/offices/classrooms/outdoors were 913/489/552/419 ppm, with the median ranking from high to low being dormitories > classrooms > offices. The typical CO₂ concentrations (box range) in dormitories exceeded the EU cat. II summer limit of 1200 ppm in 2% of cases; by contrast, those for all other space types remained below their respective standard thresholds. Dormitories and classrooms showed relatively long box ranges, which indicates larger typical fluctuation. When interpreted in conjunction with the duration curves, the results indicate that, in the spring semester, CO₂ concentrations in dormitories and classrooms exhibit larger typical fluctuations with occasional extreme values.

In the autumn semester, the medians for dormitories/offices/classrooms/outdoors were 789/474/388/359 ppm. The recorded outdoor CO₂ concentrations dipping below the 400 ppm baseline is attributed to a combination of the campus’s ultra-high vegetation coverage, which may create a localized carbon sink effect, and a potential sensor offset in the lower range. However, for the core objective of assessing indoor IEQ, this anomaly is inconsequential. Any such low-range sensor inaccuracy diminishes non-linearly and becomes negligible at the higher CO₂ concentrations typical of occupied indoor spaces. Therefore, no compensation was applied to the outdoor data, as the indoor assessments remain valid when benchmarked against the unadjusted outdoor readings. The typical indoor CO₂ concentrations (box ranges) for all space types were below the winter limit set by the adopted CN standard (1000 ppm). The median ranking from high to low was dormitories > offices > classrooms. Dormitories had relatively longer box ranges, which indicates larger typical fluctuation. When interpreted in conjunction with the duration curves, the results indicate that, in the autumn semester, CO₂ concentrations in dormitories exhibit larger typical fluctuations with occasional extreme values.

Comparing spring and autumn across space types shows that both the median and the typical CO₂ concentrations (box ranges) were generally lower in the autumn semester than in the spring semester, with the exception that office CO₂ concentration was similar between the two semesters. For offices and classrooms, the box ranges in the autumn semester were shorter than in the spring semester, implying smaller typical fluctuations and the absence of extremely high values in the autumn semester. Dormitories showed similar patterns across the two terms. The overall drop in the autumn semester medians and the marked shortening of office/classroom boxes may reflect the combined effects of greater indoor–outdoor temperature differences (enhancing buoyancy-driven ventilation and infiltration) together with relatively lower outdoor CO₂ concentration, which dilutes extremes and tightens the typical interval. The spring–autumn similarity in offices may result from stable occupancy and operating strategies. Dormitories retained long boxes in the autumn semester, suggesting that differences in residential behavior and nighttime door/window closures persist across seasons, limiting the dampening effect of seasonal change.

• PM_2.5

In the spring semester (Figure 8, Figure 9 and Figure 10 and Table 6), the median PM_2.5 levels for dormitories/offices/classrooms/outdoors were 49/41/52/57 µg/m³, with the median ranking classroom > dormitory > office. In dormitories, 4% of the typical PM_2.5 levels (box range) exceeded the summer limit of 75 μg/m³ specified by the adopted CN standard, whereas the box ranges for all other space types remained below this limit. Across space types, the typical ranges were relatively long; classroom levels were nearly at outdoor levels and tracked outdoor trends, indicating larger typical fluctuations, likely influenced by outdoor PM_2.5 levels. When interpreted in conjunction with the duration curves, the results indicate that, in the spring semester, PM_2.5 levels exhibit larger typical fluctuations with periods of extreme values.

In the autumn semester, medians for dormitories/offices/classrooms/outdoors were 96/83/124/130 µg/m³, ranked classroom > dormitory > office. Typical PM_2.5 ranges for all space types were above the WHO Interim Target 1 limit (35 μg/m³), and exceeded the winter limit under the adopted CN standard in 92% of dormitory cases, 62% of office cases, and 100% of classroom cases. Overall, typical ranges were long; classroom values were again close to outdoor levels and mirrored outdoor trends, consistent with the spring pattern.

Comparison between the two semesters shows that both the medians and the typical PM_2.5 levels (box ranges) were higher in the autumn semester than in the spring semester. In both semesters, the ranking remained classroom > dormitory > office, and classroom values were closest to outdoor levels. Because PM_2.5 exposure is primarily governed by outdoor infiltration and ventilation rates, the seasonal rise in medians and typical ranges is likely driven by higher outdoor backgrounds and variability rather than changes in indoor sources.

• Temperature

In the spring semester (Figure 8, Figure 9 and Figure 10 and Table 6), the median temperatures for dormitories/offices/classrooms/outdoors were 26.0/25.9/27.2/27.9 °C, with the median ranking classroom > dormitory > office. Typical temperature ranges for all space types were below the summer upper limit specified by the adopted CN standard (28 °C), except in classrooms, where 17% exceeded the limit; indoor temperatures were generally lower than outdoors. Typical ranges across space types were similar and relatively short in the spring semester except for classrooms. When interpreted in conjunction with the duration curves, the results indicate that, in the spring semester, typical temperature fluctuations were modest and episodes of extreme variation were limited, except in classrooms.

In the autumn semester, the medians for dormitories/offices/classrooms/outdoors were 17.4/22.4/12.6/7.6 °C, ranked office > dormitory > classroom. Typical temperature ranges differed markedly across space types: the office box lay above the winter lower limit set by EU cat. II (20 °C), the dormitory box above the winter lower limit set by the adopted CN standard (16 °C), and the classroom box below the winter lower limit set by the adopted CN standard (16 °C). Overall, typical temperature ranges were relatively short, indicating small typical fluctuations across space types. When interpreted in conjunction with the duration curves, the results indicate that, in the autumn semester, offices exhibited generally stable temperatures, classrooms were overall colder, and both dormitories and classrooms displayed larger temperature extremes.

Comparison between spring and autumn shows that, due to outdoor temperature effects, typical ranges and medians were higher in the spring semester. Relative to spring, the autumn boxes for dormitories and offices were longer. Classrooms showed a much larger overall decrease in temperature relative to spring, with the same pattern as before and the closest alignment with outdoor conditions. This indicates larger typical fluctuations in dormitories and offices in the autumn semester, with dormitories showing larger extremes and offices showing larger low-temperature extremes, while classrooms were more influenced by outdoor temperatures. A plausible explanation is that autumn brings outdoor cooling, intermittent heating, and natural ventilation operating simultaneously, leading to mostly stable but occasionally low temperatures in offices, greater variability in dormitories, and classrooms that are overall colder and more closely track outdoor conditions.

3.2.2. Spring–Autumn Semester Comparisons at TalTech

• CO₂

In the spring semester at TalTech (Figure 8 and Figure 11 and Table 6), the median CO₂ values for offices/classrooms/outdoors were 604/611/481 ppm. There was no clear difference among space types in the medians, and the typical concentration intervals (box ranges) were similar. All space types were below the limit set by EU cat. I (950 ppm); the classroom box was relatively long with a low-positioned median, indicating relatively larger fluctuations within the typical range and occasional extremely high values.

In the autumn semester, the medians for offices/classrooms/outdoors were 604/621/496 ppm, again with no clear differences among space types. The typical concentration intervals for all spaces were below the limit set by EU cat. I (950 ppm), with no obvious change from the spring pattern.

Comparing the two semesters, the upper quartiles of the autumn semester boxes were slightly higher than in the spring semester in spite of unchanged ventilation strategies, likely due to weather-driven reduced outdoor activities leading to a higher indoor occupancy.

• PM_2.5

In the spring semester (Figure 8 and Figure 11 and Table 6), the median PM_2.5 values for offices/classrooms/outdoors were 0/0/2. There were no obvious differences among spaces, and all typical concentration intervals were below the WHO AQG guideline (5 μg/m³).

In the autumn semester, the medians for offices/classrooms/outdoors were 0/0/0, with no obvious differences among spaces; all typical concentration intervals were below the WHO AQG guideline (5 μg/m³).

Comparing spring and autumn, the outdoor typical concentration interval was slightly higher in the spring semester than in the autumn semester, whereas the indoor typical intervals were fairly constant. This suggests that filtration in the mechanical ventilation system keeps indoor pollutant levels consistently low.

• Temperature

In the spring semester (Figure 8 and Figure 11 and Table 6), the median temperatures for offices/classrooms/outdoors were 23.9/23.3/11.8 °C, with no substantial differences among spaces. Typical temperature intervals for all spaces were below the summer upper limit set by EU cat. I (25.5 °C); the typical intervals were similar and relatively short, indicating limited indoor temperature fluctuations within a narrow range.

In the autumn semester, the median temperatures for offices/classrooms/outdoors were 21.8/21.7/4.9 °C, again with no substantial differences among spaces. Typical temperature intervals were mostly above the EU cat. I winter lower limit (21 °C), except in offices, where 20% fell below the EU cat. I winter lower limit (21 °C). Differences among space types were small, indicating limited indoor temperature fluctuations within a narrow range.

Comparing the spring semester and the autumn semester, typical temperature intervals and medians were generally lower in the autumn semester than in the spring semester; outdoors, both the typical interval and the median were lower in the autumn semester, with no obvious change in the overall pattern. The outdoor difference exceeded the indoor difference, highlighting the advantage of mechanical ventilation in maintaining indoor thermal comfort.

3.2.3. Comparative Analysis of CDJCC and TalTech

• CO₂

Comparing the spring semesters at CDJCC and TalTech (Figure 8, Figure 9, Figure 10, Figure 11 and Figure 12 and Table 6), the outdoor CO₂ typical interval and median at TalTech were slightly higher than at CDJCC, with broadly similar trends. The medians for offices/classrooms at TalTech were slightly higher than at CDJCC. However, CDJCC showed longer typical intervals than TalTech, indicating greater variability within the typical range and more pronounced high-end extremes in offices/classrooms at CDJCC.

Comparing the autumn semesters, TalTech again had a slightly higher outdoor CO₂ typical interval and median than CDJCC, with similar overall trends. The classroom median at TalTech sat lower within its typical interval relative to CDJCC, indicating a tendency toward high-end fluctuations in the autumn semester.

Given that CDJCC uses mixed-mode ventilation (mostly natural ventilation with supplemental air-conditioning) while TalTech uses mechanical ventilation with district heating, these differences reflect the advantage of mechanical systems in controlling indoor CO₂ concentrations.

• PM_2.5

In the spring semester (Figure 8, Figure 9, Figure 10, Figure 11 and Figure 12 and Table 6), the outdoor PM_2.5 typical interval and median at CDJCC were markedly higher than at TalTech, and indoor spaces at both campuses followed the same trend. At both campuses, indoor typical intervals and medians were lower than outdoors.

In the autumn semester, the pattern was similar to spring. Overall, indoor PM_2.5 levels were driven primarily by outdoor infiltration; filtration in mechanical ventilation can provide supplementary reduction.

• Temperature

In the spring semester (Figure 8, Figure 9, Figure 10, Figure 11 and Figure 12 and Table 6), CDJCC’s outdoor typical temperature interval and median were significantly higher than TalTech’s; office/classroom typical intervals and medians were slightly higher at CDJCC than at TalTech. Space-type-wise temperature trends were broadly similar, but CDJCC showed a tighter coupling between indoor and outdoor temperatures.

In the autumn semester, CDJCC’s outdoor typical interval and median were slightly higher, while TalTech’s outdoor temperature fluctuated more. CDJCC offices had a slightly higher median but longer typical intervals, indicating greater variability and more low-temperature extremes. CDJCC classrooms had a markedly lower typical interval and median than TalTech, with similar trends, implying a substantially lower overall temperature range in the autumn semester at CDJCC.

Overall, TalTech exhibited more stable indoor temperature control with weaker coupling to outdoor conditions, whereas CDJCC showed weaker stability in offices and classrooms during autumn, larger inter-room differences, and stronger coupling to outdoor temperatures.

4. Discussion and Implications

4.1. Cross-Campus Summary of Empirical IEQ Findings

4.1.1. Climate Zones/Seasons and Ventilation Modes

In cross-climate and cross-ventilation-mode comparisons, this study corroborates three points: (1) under a predominantly natural-ventilation scenario, the climate zone and season primarily determine the baseline values of CO₂ and PM_2.5 in campus IEQ, whereas temperature can generally be kept thermally comfortable by air-conditioning; (2) the ventilation-system configuration chiefly determines indoor temperature variability—that is, the lengths of the whiskers and box ranges in Figure 8 and Figure 12, as well as the steepness of each room’s duration curves in Figure 9, Figure 10 and Figure 11; and (3) system configuration exerts a stronger influence on IEQ than climate differences alone.

In mechanically ventilated spaces with effective filtration, high-end CO₂ and PM_2.5 tails shrink, temperature exceedance times drop, and overall IEQ variability decreases. Under natural ventilation without dedicated outdoor-air systems, IEQ outcomes become highly sensitive to window-opening and operational strategies, and in unfavorable seasons CO₂, PM_2.5 and temperature fluctuate more strongly, highlighting the role of ventilation and filtration in controlling extremes.

Within a given climate zone, season effectively sets the IEQ baseline. Winter typically raises overall CO₂ and particulate levels, and where ventilation is weak or relies on natural ventilation, winter pollution and low temperatures together with summer overheating more readily combine to create indoor extremes.

4.1.2. Space-Type IEQ Risk and Staged Control Strategy

Across space types, poorly adapted airflow organization is the main risk, especially in high-occupancy, high-density rooms. Dormitories and classrooms often have insufficient outdoor-air supply and thus show persistently elevated CO₂ and more extremes, whereas lower-density offices exhibit more moderate IEQ conditions. This indicates that high-occupancy/high-density spaces require stronger outdoor-air provision and more responsive demand-controlled ventilation to reduce the risk of extremes.

Improvement/remedial strategies should use the overall patterns by space type in the box plots for global control and take above/below-threshold durations in the duration curves as core indicators, following the sequence “secure the median first, then control variability.” (1) First, improve overall comfort through effective airflow organization and ventilation systems so that the medians and box ranges of over-limit spaces gradually converge toward compliance. (2) At the individual-space level, implement context-specific operational optimization to make box ranges and whiskers converge, thereby reducing the frequency and magnitude of above/below-limit tail exceedances in the duration curves and stabilizing overall comfort and health.

4.2. Conditional Evaluation Workflow

Through a systematic comparative analysis of campus buildings in a temperate climate and a subtropical monsoon climate, this study developed and validated a new conditional evaluation workflow based on two empirical indicators designated as the seasonal ventilation coefficient indicator ${\lambda }_{\mathit{season}}$ and the infiltration factor ($I/O$).

4.2.1. Seasonal Ventilation Effectiveness Coefficient

The seasonal ventilation coefficient indicator ${\mathrm{\lambda }}_{\mathrm{season}}$ characterizes the relative indoor CO₂ concentration under comparable occupancy density between the spring semester and the autumn semester, thereby quantifying the “effective ventilation difference” attributable to season. It is defined as the ratio of the average indoor CO₂ concentration in the autumn semester to that in the spring semester (see Equation (15)).

$${\lambda }_{\mathit{season}}={\displaystyle \frac{{\overline{\mathrm{C}{O}_2}}_{\mathit{Autumn}}}{{\overline{\mathrm{C}{O}_2}}_{\mathit{Spring}}}}$$

(15)

in which: ${\lambda }_{\mathit{season}}$ is the “seasonal ventilation effectiveness coefficient,” representing the relative difference in effective ventilation between the spring semester and the autumn semester under comparable occupancy conditions. ${\overline{\mathrm{C}{O}_2}}_{\mathit{Spring}}$ denotes the average indoor CO₂ during occupied periods in the spring semester; ${\overline{\mathrm{C}{O}_2}}_{\mathit{Autumn}}$ denotes the average indoor CO₂ during occupied periods in the autumn semester. Ground-truth occupant counts (and per-room occupancy probabilities) were not available in this campaign; therefore, occupied periods were identified using binary presence flags (CDJCC) and CO₂-based inference (TalTech), and ${\lambda }_{\mathit{season}}$ is not normalized on a per capita basis. Mathematically, ${\lambda }_{\mathit{season}} > 1$ means that indoor CO₂ is higher (and effective ventilation is weaker) in the autumn semester than in the spring semester; conversely, ${\lambda }_{\mathit{season}} < 1$ indicates lower CO₂ (i.e., stronger effective ventilation) in the autumn semester under comparable occupancy conditions. When computing this coefficient, comparable occupancy density/activity intensity, consistent instrument calibration, and specified averaging method and time window should be ensured as prerequisites.

${\lambda }_{\mathit{season}}$ is used to quantify the change in effective ventilation attributable to seasonal differences: under comparable occupancy density and activity intensity (e.g., identical class schedules/office hours), take the autumn-to-spring ratio of the indoor CO₂ averages over corresponding time spans (preferably the median or mean during occupied periods). Under a well-mixed condition, indoor CO₂ concentration is approximately inversely proportional to the effective air-change rate; therefore, ${\lambda }_{\mathit{season}} > 1$ indicates weaker effective ventilation in the autumn semester than in the spring semester. To achieve the same indoor CO₂ target in the autumn semester as in the spring semester, the ventilation rate would need to be increased by approximately ${\lambda }_{\mathit{season}}$.

For example, for CDJCC indoor spaces we obtained a campus-level mean ${\lambda }_{\mathit{season}}$= 0.9076 (95% CI: 0.7490–1.0661), indicating a weak tendency toward lower CO₂ (i.e., slightly stronger effective ventilation) in the autumn semester relative to the spring semester under the present monitoring conditions; however, the confidence interval overlaps unity, suggesting that the campus-wide seasonal effect is not robust at the 95% confidence level under the current assumptions.

In this study, the proposed seasonal ventilation coefficient ${\lambda }_{\mathit{season}}$ was further examined using the monitoring data from the CDJCC and TalTech campuses. For each monitoring point, daily mean indoor CO₂ concentrations were first computed separately for the spring semester and the autumn semester. All spring-semester daily means were then fully crossed with all autumn-semester daily means within the same point to generate a set of inter-semester CO₂ ratio samples, from which a representative ${\lambda }_{\mathit{season}}$ was derived for each room. The resulting room-level ${\lambda }_{\mathit{season}}$ values were subsequently grouped by campus and indoor space type (classrooms, dorms/bedrooms, and offices). For each group, the number of rooms, the mean ${\lambda }_{\mathit{season}}$, its standard deviation, and the 95% confidence interval were estimated using a t-distribution–based approach (t critical value multiplied by the standard error of the mean) (Equation (14)). This room-based procedure treats rooms as independent observational units and uses the t-distribution confidence intervals as an empirical consistency check: it demonstrates that, for the present dataset, the observed seasonal ventilation coefficients across different campuses and indoor space types fall within ranges compatible with the conceptual interpretation of ${\lambda }_{\mathit{season}}$ given above. The summary statistics of ${\lambda }_{\mathit{season}}$ for each group are reported in

Table 7. t-distribution-based 95% confidence interval test results for ${\lambda }_{\mathit{season}}$ using the CDJCC and TalTech CO₂ datasets as examples.

Campus	Group	N	Mean	SD	CI95_Low	CI95_High
CDJCC	classrooms	4	0.9822	0.4291	0.2994	1.665
	dorms	6	0.8486	0.1326	0.7094	0.9877
	offices	2	0.9353	0.0039	0.9005	0.9701
	outdoors	2	0.8496	0.0005	0.8447	0.8545
CDJCC	indoor	12	0.9076	0.2495	0.749	1.0661
TalTech	classrooms	6	1.1117	0.1119	0.9943	1.2292
	office	3	1.0221	0.0164	0.9814	1.0628
	outdoors	2	1.0307	0.0042	0.9931	1.0684
TalTech	indoor	9	1.0819	0.0995	1.0053	1.1584

.

From a comparative perspective, the room-level ${\mathrm{\lambda }}_{\mathrm{season}}$ values in Table 7 reveal clear differences both within and between campuses. At CDJCC, dorms and offices have mean ${\lambda }_{\mathit{season}}$ below unity, while classrooms are close to unity; the campus-wide indoor mean is ${\lambda }_{\mathit{season}}$= 0.9076 with a 95% confidence interval of 0.7490–1.0661, indicating a weak overall tendency toward lower CO₂ (i.e., slightly stronger effective ventilation) in the autumn semester, while the confidence interval overlaps unity. Among CDJCC indoor spaces, classrooms exhibit the largest between-room variability, whereas offices show the most clustered room-level coefficients and dorms fall in between, suggesting heterogeneous operational and occupant-driven effects across room types. At TalTech, by contrast, the mean ${\lambda }_{\mathit{season}}$ for both classrooms and offices is above unity, and the campus-wide indoor confidence interval lies entirely above 1, implying weaker effective ventilation in the autumn semester than in the spring semester for this dataset. Outdoor ${\lambda }_{\mathit{season}}$ behaves differently across the two campuses: it is close to unity at TalTech, whereas it is noticeably below 1 at CDJCC in this dataset, indicating that outdoor background CO₂ seasonality may not be negligible; therefore, indoor ${\lambda }_{\mathit{season}}$ should be interpreted in conjunction with outdoor conditions (and, where relevant, the $I/O$ indicator). Overall, these patterns support the intended interpretation of ${\lambda }_{\mathit{season}}$ as a room-level indicator of seasonal ventilation differences and illustrate how the proposed coefficient behaves across different room types and campuses.

4.2.2. Infiltration Factor

The infiltration factor ($I/O$) is derived from paired indoor–outdoor monitoring and is defined as the ratio of the seasonal indoor average PM_2.5 level to the outdoor average PM_2.5 level over the same period (see Equation (16)).

$$I/O={\displaystyle \frac{{\overline{P{M}_{2.5}}}_{\mathit{indoor}}}{{\overline{P{M}_{2.5}}}_{\mathit{outdoor}}}}$$

(16)

where $I/O$ is the “infiltration factor,” measuring the coupling strength between indoor and outdoor; ${\overline{P{M}_{2.5}}}_{\mathit{indoor}}$ denotes the synchronized average indoor PM_2.5 level, and ${\overline{P{M}_{2.5}}}_{\mathit{outdoor}}$ the synchronized average outdoor PM_2.5 level.

An $I/O$ close to 1 indicates high coupling (frequent window opening, weak filtration, strong outdoor penetration); values well below 1 indicate effective filtration/enclosure (substantial attenuation of particles entering indoors); values greater than 1 typically suggest indoor sources (e.g., cooking, printing, cleaning, dust resuspension) or issues with time pairing/instrument drift and should be reviewed. As a practical guide, 0.5–0.8 indicates moderate coupling; 0.8–0.95 indicates high coupling.

Linking to ventilation strategy: when $I/O$ is high and outdoor pollution is high, reduce natural intake and/or strengthen filtration; when indoor PM_2.5 level is high but outdoor air is clean (e.g., after rain or at dawn), opportunistic ventilation can lower indoor PM_2.5 level at low cost.

In this study, the room-level PM_2.5 indoor–outdoor ratio $I/O$, as defined in the conceptual framework, is further examined using monitoring data from the CDJCC and TalTech campuses. For each monitoring point, continuous PM_2.5 observations were first aggregated into hourly mean indoor and outdoor concentrations separately for the spring semester and the autumn semester. Within each monitoring point and semester, hourly PM_2.5 $I/O$ values were then obtained by pairing indoor and outdoor hourly means at the same hour. These semester-specific hourly $I/O$ samples were subsequently aggregated to the room level by taking the arithmetic mean over all valid hours, yielding a room-level PM_2.5 $I/O$ indicator for each semester.

The resulting room-level $I/O$ values were grouped by campus, semester (spring vs. autumn), and indoor space type (classrooms, dorms and offices). For each group, the number of rooms, the mean room-level $I/O$, its standard deviation, and the 95% confidence interval were estimated using a t-distribution–based approach (t critical value multiplied by the standard error of the mean) (Equation (14)). Because PM_2.5 concentrations at TalTech are frequently at or near the sensor’s effective resolution, some groups can exhibit near-zero central tendencies; under a t-distribution CI formulation, this may yield a negative lower bound, which should be interpreted as a statistical artifact rather than a physically negative infiltration. In this context, rooms are treated as independent observational units, and the t-distribution confidence intervals are used as an empirical consistency check of the proposed $I/O$ indicator, rather than as the basis for deriving it. The summary statistics for all campus–semester–room-type combinations are reported in

Table 8. t-distribution-based 95% confidence interval test results for $I/O$ using the CDJCC and TalTech PM_2.5 datasets as examples (CI is reported as computed without truncation; negative lower bounds may occur when values are near zero).

Campus	Season	Group	N	Mean	SD	CI95_Low	CI95_High
CDJCC	spring semester	classrooms	4	1.0543	0.0519	0.9717	1.0951
		dorms	6	0.8737	0.0978	0.7711	0.9798
		offices	2	0.9942	0.032	0.7069	1.0192
		all	12	0.954	0.1124	0.8826	0.9919
	autumn semester	classrooms	4	0.9496	0.0592	0.8554	1.0266
		dorms	6	0.7997	0.091	0.7043	0.7917
		offices	2	0.7743	0.0267	0.5343	1.1910
		all	12	0.8454	0.1039	0.7794	0.8669
	spring and autumn semester		24	0.8997	0.1195	0.8493	0.9502
TalTech	spring semester	classrooms	6	0.3532	0.3869	−0.0528	1.0444
		office	3	0.379	0.3018	−0.3708	2.6886
		all	9	0.3618	0.3413	0.0994	0.9863
	autumn semester	classrooms	7	0.1681	0.0981	0.0774	0.4744
		office	4	0.2244	0.073	0.1082	0.6135
		all	11	0.1886	0.0905	0.1278	0.4034
	spring and autumn semester		20	0.2665	0.2474	0.1508	0.3823

and should be interpreted as a data-based demonstration of how the PM_2.5 $I/O$ behaves under real monitoring conditions.

From a comparative perspective, the room-level PM_2.5 $I/O$ values in Table 8 reveal systematic differences both within and between campuses. At CDJCC, indoor $I/O$ values are generally close to or moderately below unity in the spring semester, and tend to decrease further in the autumn semester, indicating that indoor PM_2.5 is on average similar to, or somewhat lower than, outdoor levels, with a modest seasonal enhancement of indoor removal or shielding in autumn. Among CDJCC room types, dorms exhibit both lower mean $I/O$ and larger between-room variability than classrooms and offices, consistent with a stronger influence of occupant-controlled window opening and other room-specific behaviors.

At TalTech, by contrast, room-level PM_2.5 $I/O$ values are substantially below unity in both semesters, and the campus-wide indoor confidence intervals lie well below 1, especially in the autumn semester. We note that ratios involving near-zero indoor PM_2.5 values can be mathematically sensitive; therefore, TalTech $I/O$ results—especially for groups with very low indoor PM_2.5—should be read primarily as evidence of strong attenuation rather than as a precise point estimate. This pattern indicates a stronger overall reduction in outdoor PM_2.5 indoors, reflecting more effective removal, filtration, or infiltration/ventilation characteristics in the TalTech building stock for the present dataset. The relatively wide confidence intervals for some spring-semester groups, particularly where the number of rooms is small, suggest notable between-room variability and the influence of episodic events, but the combined spring–autumn statistics still consistently point to $I/O$ values markedly below 1. Overall, these results illustrate that the PM_2.5 $I/O$ indicator, as proposed in this study, behaves in a physically meaningful way across campuses, semesters, and room types, and provide an empirical validation of its usefulness for comparing effective particle removal between indoor environments.

5. Conclusions

5.1. IEQ Assessment Decision Process

Upon introduction of the two empirical indicators, the IEQ assessment decision workflow can be summarized using the workflow illustrated in Figure 13.

5.2. Natural Ventilation Strategy Recommendations

For naturally ventilated buildings, our diagnostic workflow yields three pragmatic, actionable measures:

• User-centered ventilation scheduling. A quantified ${\lambda }_{\mathit{season}}$ (e.g., 2.0) provides the evidence base for structured winter ventilation intervals: briefly open windows fully during periods when outdoor PM_2.5 levels are low (guided by real-time outdoor AQI, e.g., between classes) to quickly dilute accumulated CO₂ concentration while minimizing thermal-comfort loss and particulate matter ingress.
• Supplemental filtration during low-ventilation periods. Even when outdoor air quality is generally good, a high $I/O$ indicates the need for air purifiers so that, with windows closed in winter, indoor PM_2.5 levels are decoupled from outdoor pollution—without changing the basic natural-ventilation strategy.
• Operations tuned to outdoor AQI coupling. For example, when the summer $I/O$ is about 0.95, prioritize ventilation during rainfall or immediately afterward, when outdoor PM_2.5 levels are naturally lower; window opening then provides the largest net purification benefit.

5.3. Overall Contributions and Implications

This study addresses the common shortcoming in campus sustainability and IEQ work—static “pass/fail” assessments that do not inform operations—by proposing and validating a closed-loop pathway of assessment → decision → operation. The pathway is comparable across climates, and directly answers operational questions (when to boost airflow or open windows, whether and how much to filter, and how to calibrate winter–summer baselines), thereby turning evaluation into action.

We introduce FMCW radar module into the IEQ sensing stack as a privacy-preserving, low-power, and scalable method for presence/occupancy detection (occupied vs. unoccupied) and space-use monitoring. FMCW markedly improves the timeliness and usability of data while lowering deployment and maintenance burdens. It tightens the assessment–decision link—e.g., triggering airflow increases when occupied and shifting to energy-saving modes when unoccupied—thereby providing a practical sensing foundation for fine-grained operations in naturally ventilated buildings.

We combine box plots (to read cross-space patterns) with duration curves (to quantify exceedance probabilities for CO₂, PM_2.5, and temperature), enabling nuanced spatial comparisons. We further propose two actionable metrics: ${\lambda }_{\mathit{season}}$ to calibrate winter–summer baselines (i.e., how much to scale the ventilation rate) and $I/O$ to guide opportunistic ventilation and filter strength (i.e., indoor–outdoor coupling). Together, these translate static compliance into executable, day-to-day “how-to-optimize” rules.

For naturally ventilated existing buildings, we offer a scalable approach: replace heavy, permanent deployments with low-cost sampling of ${\lambda }_{\mathit{season}}$ and $I/O$, and adopt time-segmented operation that fuses outdoor air quality with time of day (opportunistic window opening, demand-based filtration). Within this framework, the presence signal picked up by the sensor’s FMCW radar module acts as a core trigger, strengthening the assessment–decision–operation loop, lowering data and infrastructure thresholds, and making campus-scale closed-loop control feasible.

Compared with TalTech, CDJCC shows longer box-range (typical-interval) spans and—considered with duration curves—greater within-range variability with more pronounced high-end extremes in offices and classrooms. We also identify a structural paradox: mechanically ventilated buildings often possess rich BMS (Building Management Systems) data but require less seasonal validation, whereas naturally ventilated buildings most need empirical, seasonal validation yet lack ready data. This mismatch constrains the scaling of climate-resilient IEQ assessments and highlights the urgency of baseline data collection and lightweight validation systems for naturally ventilated stock.

Overall, this work advances the PICSOU framework’s IEQ category from static compliance to a comparable, governable, and transferable decision layer, with real-time occupancy-enabled data collection as a practical catalyst for closed-loop, low-cost, and scalable operations.

6. Limitations and Future Work

6.1. Limitations of Monitoring Design and Data Coverage

This study has several limitations related to monitoring design and data coverage, particularly with respect to monitoring duration, the number of devices, and variation in teaching activities across semesters.

First, the monitoring campaigns were designed around typical spring and autumn teaching semesters at the two campuses, and were conducted over finite periods within each semester rather than as continuous, multi-year observations. The relatively limited monitoring duration means that the datasets are well suited to characterizing representative conditions for the selected weeks, but they do not capture the full range of inter-annual variability or conditions during other parts of the academic year (for example, early spring, late autumn, or summer recess). Short-term episodes such as rare pollution or heat events may also be underrepresented. In addition, the selected monitoring windows did not include holiday breaks or other atypical occupancy periods, so IEQ responses during these special periods could not be explicitly compared or evaluated. While structuring the analysis by semester improves comparability between the two campuses and between spring–autumn terms, the finite duration of each campaign inevitably constrains how far the results can be generalized in time.

Second, the spatial coverage and number of devices are constrained by practical considerations. Only a limited number of sensors could be deployed, and these were distributed across a subset of buildings and space types on each campus, with a focus on typical teaching and office spaces. Within individual buildings, sensor locations were influenced by access, installation feasibility, safety, and maintenance requirements. As a result, some space typologies—such as certain laboratories, shared learning hubs, recreational areas, and support spaces—are underrepresented or not captured, and the monitored rooms may not span the full spectrum from “best-performing” to “worst-performing” IEQ spaces. For derived indicators such as ${\lambda }_{\mathit{season}}$ and $I/O$, additional data-completeness criteria further reduce the number of usable devices, rooms, and days in the robustness analysis, introducing a further level of selection into the underlying samples.

Third, variation in teaching activities and space use across semesters introduces additional uncertainty. Although the monitored periods were selected to represent regular teaching weeks, there remain unavoidable differences between semesters and years in course timetables, classroom allocation, student numbers, and occupancy patterns (e.g., examination weeks, project weeks, or minor timetable changes). These factors influence how intensively rooms are used and how ventilation systems are operated, and thus may partly contribute to differences observed between semesters and between campuses. The present analysis cannot fully disentangle the effects of climate and ventilation configuration from those of changing teaching activities. In addition, building-envelope performance (e.g., airtightness and thermal properties) was not measured or harmonized across buildings/campuses, so its influence cannot be separated from the observed climate- and ventilation-mode effects.

Taken together, these limitations mean that the present results should be viewed as indicative for the studied campuses and their monitored spring and autumn semesters, rather than as an exhaustive characterization of all campus spaces, seasons, and operational conditions. They also motivate future extensions with longer monitoring periods, a larger number of devices and monitored rooms, and more systematic documentation of teaching and occupancy patterns, which would help to further strengthen the robustness and generality of the conclusions.

6.2. Limitations of Occupancy Representation and Cross-Campus Comparability

A further limitation of this study lies in the different ways in which occupancy is represented at the two campuses. As discussed in Section 3.1, these differences have been examined empirically; here, they are summarized as constraints on the occupancy information used in the analysis.

At CDJCC, all custom-built IEQ sensor units are integrated with an FMCW radar module, and the present study uses radar-based presence as the occupancy signal. This provides a direct, privacy-preserving indication of whether a room is occupied, and forms the basis for linking occupancy to IEQ patterns and for illustrating the assessment–decision–operation loop. At TalTech, by contrast, no dedicated occupancy sensor was available, and occupancy had to be inferred solely from CO₂ time-series patterns. The two campuses therefore differ not only in sensing hardware but also in how “human presence” enters the analysis: one site uses a direct signal, the other relies on an indirect proxy.

The CO₂-based occupancy proxy at TalTech is inherently less specific than radar-based detection. CO₂ dynamics reflect the combined influence of human emissions and air exchange, so elevated or declining concentrations cannot be uniquely attributed to changes in occupancy. Section 3.1 uses CDJCC data as a reference to compare FMCW-derived occupancy with CO₂-based inference, and shows that systematic deviations can occur in certain periods and space types. These findings indicate that a simplified CO₂-only method may misclassify some occupied and unoccupied intervals, and that such biases are difficult to quantify or correct at TalTech, where no independent ground truth is available.

Taken together, these factors mean that the occupancy information used in this study should be regarded as uneven in strength across the two campuses: CDJCC benefits from direct radar-based presence detection, whereas TalTech relies on a CO₂-driven proxy. Although Section 3.1 demonstrates that the two approaches are broadly comparable for the purposes of this study, the remaining discrepancies highlight a limitation in cross-campus comparisons that condition on occupancy, and motivate future work toward more consistent, privacy-preserving occupancy sensing and integrated use of CO₂ and direct presence signals.

6.3. Future Directions for PICSOU’s IEQ Category

Based on this study, the PICSOU framework’s IEQ module will continue to develop along three main directions.

To keep PICSOU’s IEQ category aligned with the evolving EU technical guidance and practice, future updates can formalize a clearly defined indicator set that reflects both health relevance and operational measurability. At minimum, this would include thermal conditions (e.g., operative/air temperature), humidity (RH), ventilation adequacy (e.g., CO₂ and/or ventilation-rate proxies), and key exposure-related pollutants (e.g., PM_2.5; with VOCs where feasible), while allowing optional extensions for acoustics and lighting when instrumentation is available. Methodologically, adopting an exceedance-based, time-fraction interpretation (as exemplified by TAIL) would allow PICSOU to translate monitoring time series into concise category-level compliance statements, improving robustness across climates and space types [58].

First, the indicator set within the CO₂/PM_2.5/temperature triad will be expanded and better weighted against learning- and health-relevant outcomes. Additional IEQ dimensions (e.g., humidity, noise, or lighting) can be incorporated in a modular way, allowing the IEQ module to evolve from a minimal, three-indicator set toward a broader, but still interpretable, indicator system that reflects both regulatory limits and outcome-oriented priorities.

Second, the IEQ module will move from largely static interpretations toward more dynamic evaluation, explicitly treating indoor air as a time-varying system rather than as a collection of averages. Building on the duration-curve and box-plot approach used in this study, future work will place greater emphasis on temporal patterns—such as the timing, frequency, and persistence of exceedances—and on how these patterns coincide with occupancy and operation. This opens the door to a tighter coupling between PICSOU’s IEQ category, occupancy sensing (e.g., FMCW radar–based presence signals), and operational data, so that the framework can support near-real-time feedback, event detection (e.g., persistent under-ventilation), and rule-based or data-driven control strategies.

Third, PICSOU’s IEQ category will be strengthened in terms of cross-climate and cross-space robustness. The present ${\lambda }_{\mathit{season}}$ and $I/O$ indicators provide an initial step toward comparable baselines and “how-much-to-adjust” rules across seasons and ventilation modes; future work will extend these concepts by building parameter libraries and baseline ranges across multiple climate zones, campus types, and space categories (e.g., classrooms, dormitories, offices, laboratories). As more campuses and building types are monitored, these libraries can be iteratively refined into a generalizable evaluation system that allows both within-campus benchmarking and cross-campus comparison, while acknowledging local operational constraints.

Overall, these developments aim to advance PICSOU’s IEQ category from a compact, proof-of-concept implementation to a more complete, results-oriented and transferable decision-making workflow, one that is capable of integrating more nuanced occupancy information, supporting dynamic operation, while remaining scalable for campus-wide deployment.