Prediction models of bioaerosols inside office buildings: A field study investigation

Abstract

Bioaerosols formed by microorganisms in the air directly affect people’s health. The air quality in an office building in Shenzhen, China, is investigated and pollutant levels measured on 36 occasions; six times for each of six indoor spaces. A relationship between indoor bioaerosols and environmental factors was determined using both linear regression analysis and Poisson regression analysis. Our results and analysis indicate that linear regression is a poor predictor for the concentration of bioaerosols based on a single indicator. In contrast, Poisson regression can better predict the concentration of bioaerosols, and PM₁₀ may be the indicator with the greatest impact on bioaerosols. As a result, a simple, fast, and low-cost online monitoring method for monitoring indoor bioaerosols is developed and reported. Our paper provides first-hand basic data to predict the indoor bioaerosol concentration and helps to formulate appropriate monitoring guidelines. The proposed method offers more practical values compared to existing studies as our prediction model facilitates estimation of the concentration of bioaerosols at low cost. Additionally, due to the current maturity and low cost of indoor environmental sensors, the proposed method is suitable for large-scale deployment for most buildings.

Practical application

Based on measurement data from a real office building, our investigation explores the relationship between indoor microorganisms and building environmental indicators through a combination of probability analysis and actual measurement. We establish a novel indoor microbial prediction model using the Poisson regression model. Our work presents an effective, low-cost, method for estimating the concentration of bioaerosols and discusses the possibility for large-scale deployment of microbial monitoring equipment inside buildings which may then support real-time monitoring of indoor microbial concentration to provide healthy indoor environments for personnel.

Keywords

Office building bioaerosols microorganism forecast model linear regression poisson regression

Highlights

• A simple, fast and low-cost method for predicting bioaerosol concentrations is developed through monitoring data.

• The effect of several factors affecting indoor bioaerosol concentration are examined in depth.

• PM₁₀ is the most significant influential indicator of indoor bioaerosol concentration.

• The Poisson regression is more effective in explaining bioaerosol concentration compared with linear regression.

Introduction

People spend >90% of their time indoors,^1,2 which means that indoor air quality (IAQ) can significantly affect human health.³ Cases of COVID-19 have continued to occur worldwide and continue to occur in Shenzhen, China. The outbreak of COVID-19, monkeypox, influenza, and other epidemics has focussed many people's attention on the quality of indoor air and the indoor environment, particularly in relation their health and wellbeing.²

The office building is a complex place with a concentrated population and frequent communication making it a key location for harmful microbial infections. The complex indoor environment contains many factors that affect indoor microbial concentration, including particles of different particle sizes, carbon dioxide (CO₂), temperature, humidity, ventilation rate, season, height, etc.^4–13

Epidemiological evidence shows that there is a relationship between indoor pollutants and health risks,¹⁴ and indoor pollutants of concern include particulate matter (PM), biological organisms (fungal spores, bacteria, and viruses), allergens, over 400 different chemical compounds, mainly volatile organic compounds (VOCs) such as benzene, toluene, methanol, ethylbenzene and xylene, inorganic compounds (ICs) such as carbon monoxide (CO), carbon dioxide (CO₂), nitrogen oxides (NO_x) and ozone (O₃), among others.^15,16

Given that 5%–34% of particulate matter (PM) in indoor air is in the form of bioaerosols (i.e., bacteria),¹⁷ these bioaerosols are accompanied by many harmful microorganisms and viruses, such as SARS-CoV-2 and other viruses, which are spread through bioaerosols.^2,18,19 So they are gaining increasing research attention. Indoor bioaerosol concentration is a direct factor affecting indoor air quality and human health, and also a direct cause of the spread of influenza and other epidemic diseases. The composition and concentration of bioaerosols in the indoor air of buildings reported in the literature indicate that it is important to further study the relationship between indoor bioaerosols and various measurable environmental factors to predict the concentration of indoor bioaerosols and prevent indoor infection. In this case, the quantitative estimation of microorganisms in the air is used to monitor the concentration of indoor microbial aerosol and improve the safety of indoor air.

Culture-based methods have traditionally been used to determine the concentration of bioaerosols^20,21 but as these require offsite processing in remote laboratories, they cannot supply real-time information and cannot be applied on a large scale, due to strict cultivation conditions, long cycles, and high costs.²² Fortunately, there are correlations between indoor bioaerosols and other physical parameters, supplying the possibility to explore a simple, fast and low-cost bioaerosol concentration prediction method.²³

Different environmental factors can affect the distribution and survival ability of indoor microorganisms, such as the survival ability of SARS-CoV-2 and its distribution in the environment.²⁴ Research has shown that air pollutants, especially particulate matter, and other meteorological factors, are important carriers of infectious microorganisms and play a crucial role in the spread of diseases.²⁵ The ability of microorganisms to survive and maintain their pathogenicity in ambient air depends on the following conditions: relative humidity (RH), temperature (T), ventilation, or chemical composition.^19,26–29 The survival rate of microorganisms under low relative humidity is low, and the ventilation airflow is directly related to the relative humidity,^19,30,31 and the ventilation airflow affects the distribution of bioaerosols in the indoor environment.^19,32 RH levels between 40% and 70% can help reduce the spread of the virus, and engineering controls need to be implemented by adjusting the HVAC system to ensure optimal humidity in the indoor environment.^26,30,33,34 At the same time, temperatures above 30°C and RH below 78% will strongly weaken the spread of SAS-CoV-2.³⁴ Moreover, more researchers insist that carbon dioxide (CO₂) can reflect the trend of bioaerosol concentration, and strongly recommend that CO₂ concentration be used for infection risk-based monitoring.^35–39

Research has shown that particulate matter pollution is a serious global risk factor to human health among all key pollutants,⁴⁰ and PM_2.5 levels are positively correlated with COVID-19 exposure risk,²⁵ as fine particles (FPs) and ultrafine particles (UFPs) have small particle sizes and large specific surface areas, making them easy to adsorb metals, microorganisms, and other pollutants.⁴⁰ Particles may play a role in the differential distribution and propagation rate of SARS-CoV-2.⁴¹ Smaller particles can be used as tracers to estimate the air exchange rate, and the real-time air exchange rate of buildings can be predicted,^42,43 so the real-time and recent concentrations of bioaerosols and particles (PM) can be determined with relatively high accuracy.²² However, they can only be used as agents and cannot accurately reflect the distribution and quantity of indoor bioaerosols in real-time.²²

The above studies show the relationship between bioaerosols in indoor air of buildings and various environmental parameters, but these studies do not involve integrating these measurable parameters with the prediction of bioaerosols. However, in reality, environmental factors may lead to significant changes in space and time.³¹ Therefore, it is necessary to consider the spatial changes of temperature, relative humidity, ventilation, particulate matter, and other environmental factors in the whole room to accurately determine the real indoor bioaerosol concentration in real-time. Because of the importance of preventing the spread of bioaerosols in the indoor environment, low-cost technologies that can accurately monitor indoor bioaerosols, especially the concentration of bioaerosols in the air,⁴⁴ are urgently needed.

Particle concentration, temperature, and relative humidity (RH) can be fitted into a multi-linear regression model, enabling it to reasonably and accurately predict the concentration of bacteria and fungi in indoor air from culture-based data.⁴⁵ Real-time sensor data can also be used to monitor key environmental indicators (including indoor temperature, carbon dioxide concentration, relative humidity, and water consumption). The large amount of data from these real-time sensors can help to model and predict indoor air pollution, to determine the real-time and recent concentrations of indoor bioaerosols with relatively high accuracy.^{22,44,46–48}

The prediction models of indoor bioaerosol concentration mainly include linear regression, nonlinear regression, empirical equations, and other prediction models,^9,12,49,50 and the use of linear regression models is better than the use of nonlinear regression models,^49,50 and the study of bioaerosol models is better in predicting single office buildings.⁴⁹

So far, although the indoor bioaerosol concentration is clearly specified in China’s Standards for indoor air quality GB/T 18883-2022,⁵¹ there is no detailed regulation for different office sections in the office building. In addition, there needs to be a more detailed analysis of the factors affecting the bioaerosol concentration in indoor environments, such as office buildings, and the development of prediction models.

The study will further explore the qualitative and quantitative relationship between indoor bioaerosol concentration and measurable environmental factors, and try to develop a simple, fast and low-cost bioaerosol concentration prediction method, as a simple alarm tool for indoor air quality in office buildings.

At present, no method has been developed that can continuously and accurately determine the real-time and near-future indoor microbial bioaerosol concentration. Therefore, in this study, the qualitative and quantitative relationship between indoor bioaerosol concentration and measurable environmental factors will be further explored, and a simple, fast and low-cost bioaerosol concentration prediction method will be developed based on linear regression and Poisson regression to accurately predict the continuous real-time and recent bioaerosol concentration using the environmental parameters monitored in real-time by indoor environmental monitoring sensors. The ability to predict the real-time and different durations of recent bioaerosol concentrations is very important for taking mitigation measures to deal with imminent and potential health risks.⁵²

Methodology

An office building in Shenzhen, China, was completed in 1996. The building has an area of 32,000 m² and a total of six floors, as shown in Figure 1. Shenzhen is located in an area where summer is hot, and winter is warm with the average yearly temperature is 23.3°C.⁵³ The average temperature in January of the year is the lowest, with an average of 15.7°C. The average temperature in July is the highest, with an average of 29.0°C. The summer lasts for over 6 months (with an average summer length of 202 days).⁵³ Representative space in the office building is selected for on-site monitoring, including one office, one meeting room, two toilets and two corridors. With different environmental factor data as independent variables and microbial concentration as dependent variables, we analyse the correlation between microbial concentration and environmental factors and establish corresponding prediction models. The model with the best prediction accuracy is then selected as most representative of the real office building indoor microbial environment.

Sampling plan and field measurement

Field sampling was conducted in an office building in Shenzhen, China, from July to August 2022, and bioaerosol concentration, particulate matter (including PM₁, PM_2.5, PM₄, PM₁₀, and Total Particulate Matter (TPM)), CO₂ concentration, air exchange rate (ACH), temperature (T), and relative humidity (RH) were measured in each measurement space. During the measurement, the staff worked normally. Only during the sampling period did they leave and turn off the air conditioner to reduce the impact of personnel and air conditioner on the concentration of environmental microorganisms. The location and time of measurement are shown in Table 1.

Table 1.

The location and time of measurement.

Location	Office (2^nd F)	Meeting room (2^nd F)	Corridor (2^nd F)	Toilet (2^nd F)	Toilet (1^st F)	Corridor (1^st F)
Time	9:00	9:00	9:00	9:00	9:00	9:00
	10:00	10:00	10:00	10:00	10:00	10:00
	11:00	11:00	11:00	11:00	11:00	11:00
	14:00	14:00	14:00	14:00	14:00	14:00
	15:00	15:00	15:00	15:00	15:00	15:00
	16:00	16:00	16:00	16:00	16:00	16:00

Using different air quality monitoring sensors to monitor indoor air pollutants in six different locations in Shenzhen, including bioaerosol concentration, particulate matter (including PM₁, PM_2.5, PM₄, PM₁₀, and Total Particulate Matter (TPM)), CO₂ concentration, the air exchange rate (ACH), temperature (T), and relative humidity (RH).

Two QY-6 impinging air microbial samplers are used for bioaerosol sampling, two US TSI8532 aerosol monitors are used to monitor the concentration of particles corresponding to PM₁, PM_2.5, PM₄ and PM₁₀, one Telaire-7001 CO₂ Sensor is used to measure indoor CO₂ concentration, and one testo 440 probe system is used to measure indoor temperature and humidity. The sampling points follow the principles of diagonal and centreline layout, as shown in Figure 2. Two QY-6 and two TSI8532 aerosol monitors were placed diagonally in the measurement room, while Telaire 7001 and Testo 440 were placed at the centreline of the measurement room. The on-site measurement location diagram is shown in Figures 3 and 4.

The microbiological sampling method is the Chinese Examination methods for public places-Part 3:Airborne microorganism (GB/T 18204.3-2013), and the QY-6 impinging air microbial sampler is used for bioaerosol sampling. The sampler can capture six levels of particle range (Level 1: ≥ 7.0 μm; Level 2: 4.7–7.0 μm; Level 3: 3.3–4.7 μm; Level 4: 2.1–3.3 μm; Level 5: 1.1–2.1 μm; Level 6: 0.65–1.1 μm), capture rate:>98%, sampling flow rate: 28.3 L/min (0–45 L adjustable), directly capturing microorganisms in the air onto the culture dish containing nutrient agar. After sampling, put the sample in a 37°C constant temperature incubator for 48 h, and then assign a special person to record and calculate the total number of bacterial colonies.

The concentration of bioaerosol in the air is calculated according to equation (1):

C o u n t o f b i o a e r o s o l (c f u / m^{3}) = \frac{C o l o n y c o u n t o f a l l p l a t e s \times 1000}{S a m p l i n g t i m e (\min) \times 28.3 (L / \min)}

(1)

Particulate matter (including PM₁, PM_2.5, PM₄, PM₁₀, and TPM), CO₂ concentration, ACH, T, and RH of indoor space were investigated.

The US TSI8532 aerosol monitor is used to monitor the particle concentration corresponding to PM₁, PM_2.5, PM₄ and PM₁₀, and can monitor the particle concentration range from 0.001 to 150 mg/m³.

The testo 440 probe system can be used for continuous measurement of the temperature and humidity, with a measuring range of 0 to 100% and −20 to + 70°C.

The Telaire-7001 CO₂ Sensor can be used to measure indoor CO₂ concentration, with a measurement range of 0–10000 ppm and an accuracy of ±50 ppm.

Some studies show that the content of bioaerosol in outdoor air is usually higher than that in indoor air.^54–56 Therefore, to reduce the impact of outdoor microorganisms, the air conditioning system is closed during measurement to reduce the impact of outdoor air aerosol. The carbon dioxide concentration is used to calculate the indoor air exchange rate ACH, and the calculation equation is shown in equation (2).^54,55,56

A C H = \frac{k n}{V (C_{i} - C_{o})} \times 10^{6}

(2)

Where ACH is the air exchange rate, h⁻¹; V is the measured space volume, m³; C_i and C_o are the indoor and outdoor CO₂ concentration, ppm; k is the CO₂ emission rate of each occupant (m³/h), and taken as 0.018 m³/h in the paper; n is the number of staff in the work area.^54–56

Data analysis and processing

The data recording of the continuous monitor should be at least 1 min for sampling the initial air quality. The sampling interval of all monitoring equipment was set as 1 s and data were averaged and logged at 10-min intervals. Microsoft Excel and SPSS 22.0 were used to process and analyse the data. Microsoft Excel was used for basic statistical analysis to quickly process raw data, including calculation of range, median, and average.⁴⁹ Then SPSS 22.0 was used for more advanced statistical analysis, and the confidence interval was set to 95% for statistical significance.

All measurement data were tabulated and stored in Excel. Before using the measurement data in the model, the following four steps are performed to organize the measurement data to reduce the impact of measurement errors. Firstly, abnormal data was removed from the dataset. Secondly, we perform a mean interpolation to fill in any missing values for all parameters. This provides three and four-time series data points for the bathroom and corridor, respectively. Thirdly, the measurement data adopts an average to reduce errors, that is, directly processing the average results of a series of measured values with internal differences.^49,57 After the bioaerosol concentration is cultivated in the laboratory, the concentration of two groups of bioaerosols at the measurement location and time is obtained by equation (1), and the average value is taken as the model application data. The US TSI8532 aerosol monitor counts and measures particle concentration every minute for a total of 10 min (obtaining 10 sets of data). The data is averaged every minute, and then the average of these 10 sets of data is taken as the model application data. Temperature, relative humidity, and CO₂ concentration were measured using an instrument that counts data every minute for a total of 10 min (resulting in 10 sets of data). The average of these 10 sets of data was taken as the model application data. Fourthly, obtain the corresponding ventilation rate according to equation (2). Finally, the model uses nine parameters (temperature, relative humidity, CO₂ concentration, ventilation rate, and five different particle size particle concentrations) as input factors and bioaerosol concentration data as target factors to study the prediction model for indoor bioaerosol concentration.

Linear regression and Poisson regression analysis were conducted with the concentration of bioaerosols as the dependent variable and particulate matter, CO₂ concentration, ACH, T, and RH as independent variables. First, the p-value of the whole model is analysed. If the value is less than 0.05, the model is valid. Otherwise, the model is invalid. Then when the p-value of an indicator is less than 0.05, it indicates that the indicator will impact the bioaerosol.

Development of prediction model for bioaerosol concentration

The indoor bioaerosol concentration and other indoor indicators may not be purely linear but more likely to show a probability correlation. Chao⁶ found in his research on the indoor air of 37 office buildings in Taipei, Taiwan, that it is better to use multiple linear regression model to predict the microbial concentration of an office building, but field measurement found that the microbial concentration and indoor physical parameters are not completely corresponding to the linear relationship, and the bioaerosol concentration relationship in the same local reaction area is more likely to be a probabilistic problem, thus realizing online monitoring. Therefore, the paper attempts to use Poisson regression analysis and compare it with linear regression analysis.

The Poisson regression model is used to describe the frequency distribution of an event found in unit time, unit area or unit volume and is usually used to describe the distribution of rare events (i.e., small probability).⁵⁸ The indoor environment is not constant, and the bioaerosol concentration will change with external changes such as air circulation. Therefore, the bioaerosol concentration does not have a long-term linear relationship with other environmental parameters but more likely conforms to the probability model of Poisson regression. The prediction model of bioaerosol concentration is developed through Poisson regression model to predict indoor microbial concentration. The Poisson regression prediction equation (3) of indoor microbial concentration is as follows:

U = C_{1} + e^{C_{2} + Σ_{i = 1}^{n} {α_{i} P}_{i}}

(3)

Where:U: bioaerosol concentration, CFU/m³, CFU (colony-forming units) is a unit used in microbiology to estimate the number of viable bacteria or fungal cells in a sample;C₁: correction constant, obtained from comprehensive analysis of actual measurement and simulation;C₂: basic value, correction constant, obtained from comprehensive analysis of actual measurement and simulation;P_i: particulate matter, CO_2, and other parameters, including PM₁, PM_2.5, PM₄, PM₁₀, TPM: particulate matter with different particle sizes, mg/m³; RH: humidity, %; CO₂: carbon dioxide concentration, ppm; ACH: indoor air exchange rate, h⁻¹; T: Temperature, °C, when the temperature is stable, the bioaerosol concentration is independent of the temperature, and the coefficient α₉ is 0;α_1∼9: coefficients of different parameters.

Results and discussion

Sample data

Table 2 lists the indoor field measurement processed data of the office building and the ACH calculated by equation (2).

Table 2.

Indoor field measurement data of an office building.

Item		Bioaerosol concentration (CFU/m³)	PM₁	PM_2.5	PM₄	PM₁₀	TPM	T (°C)	RH (%)	CO₂ (ppm)	ACH (h⁻¹)
Item		Bioaerosol concentration (CFU/m³)	(mg/m³)					T (°C)	RH (%)	CO₂ (ppm)	ACH (h⁻¹)
Office (2^nd F)	1	833.922	0.000413	0.000494	0.001118	0.004747	0.015118	25.623	58.182	657.400	1.079
	2	828.622	0.001036	0.001718	0.003097	0.006611	0.017758	25.817	58.195	626.380	1.251
	3	627.209	0.042481	0.135328	0.225827	0.276284	0.292933	26.317	56.971	644.690	1.144
	4	747.350	0.002221	0.002042	0.001301	0.003766	0.017191	25.591	53.870	637.930	1.181
	5	740.283	0.001284	0.00033	0.000642	0.004027	0.019709	25.819	54.372	647.800	1.127
	6	805.654	0.000819	0.000403	0.001115	0.006335	0.027244	25.474	55.048	591.090	1.529
Meeting room (2^nd F)	7	1466.431	0.000403	0.000503	0.001246	0.006702	0.026244	26.183	65.907	714.780	1.900
	8	1406.360	0.000592	0.001082	0.00317	0.012003	0.039075	26.273	61.505	689.810	1.813
	9	1561.837	0.000537	0.000364	0.00186	0.008005	0.023743	26.197	62.034	694.080	2.460
	10	1405.477	0.000174	0.000304	0.001243	0.007754	0.029385	26.549	62.872	735.580	1.384
	11	1404.594	0.000647	0.000866	0.002133	0.009883	0.035386	26.078	58.356	730.390	1.518
	12	855.124	0.000578	0.000748	0.001933	0.007514	0.028352	25.179	55.630	646.270	1.493
Toilet (2^nd F)	13	549.470	0.001433	0.001609	0.001935	0.003112	0.015628	28.233	58.207	590.540	1.289
	14	703.180	0.002999	0.003181	0.003750	0.007609	0.034578	28.528	53.199	659.960	1.311
	15	802.120	0.000424	0.000489	0.000798	0.003473	0.025710	28.655	50.911	705.990	1.826
	16	964.664	0.009625	0.010047	0.011027	0.016569	0.060022	28.613	51.386	734.180	2.468
	17	436.396	0.000623	0.000771	0.001321	0.004933	0.026773	28.672	49.498	750.440	1.717
	18	798.587	0.004300	0.004475	0.004755	0.005946	0.015221	28.578	48.662	715.640	1.428
Corridor (2^nd F)	19	791.519	0.000992	0.001115	0.001394	0.003050	0.010242	26.387	62.805	597.690	1.163
	20	265.018	0.000181	0.000198	0.000261	0.001266	0.011771	26.570	58.389	588.660	1.183
	21	593.640	0.000165	0.000203	0.000423	0.002856	0.022224	26.546	56.473	642.790	1.647
	22	287.986	0.001098	0.001233	0.001625	0.004744	0.030173	26.309	54.694	598.950	2.226
	23	325.088	0.000909	0.001017	0.001297	0.003169	0.015236	26.290	58.439	614.240	1.548
	24	206.714	0.000781	0.001562	0.002336	0.003980	0.013423	26.278	57.658	623.210	1.288
Corridor (1^st F)	25	456.773	0.000115	0.000147	0.000283	0.001650	0.009842	26.402	63.554	770.440	1.168
	26	469.376	0.001140	0.001793	0.002817	0.005440	0.015587	26.492	68.653	638.710	1.108
	27	438.163	0.000150	0.000193	0.000362	0.001525	0.011790	26.256	61.804	782.780	1.244
	28	378.092	0.001762	0.001181	0.001663	0.002822	0.013748	26.558	70.853	518.930	2.130
	29	492.933	0.000294	0.000368	0.000757	0.003812	0.020229	25.921	61.276	615.570	2.253
	30	668.728	0.000799	0.000343	0.000534	0.004236	0.039307	25.927	58.992	785.900	1.674
Toilet (1^st F)	31	893.993	0.004213	0.004273	0.006437	0.018326	0.072768	26.830	65.247	663.610	2.468
	32	768.551	0.001003	0.001213	0.001842	0.007681	0.059981	27.141	58.413	682.540	1.717
	33	982.332	0.001452	0.001628	0.002067	0.006789	0.068093	27.134	57.070	676.570	1.428
	34	556.537	0.000423	0.000531	0.000803	0.003325	0.029125	27.205	59.637	797.190	1.295
	35	558.3034	0.001350	0.001665	0.002402	0.006731	0.049370	27.912	55.024	733.270	1.229
	36	1199.647	0.001515	0.001727	0.003722	0.009953	0.038783	27.235	56.470	682.910	1.380

The distribution of bioaerosol concentrations in different rooms is presented in Table 3 and Figure 5. The quartile diagram of building bioaerosol concentration is shown in Figure 5. Each box represents the building bioaerosol concentration with the black box marker in the middle representing the average, and the outliers are plotted individually using the symbols ×. The overall average bioaerosol concentration of the office building is 757.519 CFU/m³, and the highest average value is 1349.971 CFU/m³ in the conference room, 411.661 CFU/m³ minimum for 1^st F toilet, lower than the index 1500 CFU/m³ in the Chinese standard GB/T 18883-2022.⁵¹ Only the bioaerosol concentration in the 2^nd F conference room was on the high side, exceeding the national standard and reaching 1561.837 CFU/m³ at 11:00 a.m.

Table 3.

Bioaerosol concentration in the office.

Location	Sample size	Minimum value	Maximum value	Mean value	Standard deviation	Median value
Location	Sample size	(CFU/m³)
Office (2^nd F)	6	627.208	833.922	763.840	77.912	776.502
Meeting room (2^nd F)	6	855.124	1561.837	1349.971	250.027	1405.919
Toilet (2^nd F)	6	436.396	964.664	709.069	190.713	750.883
Toilet (1^st F)	6	206.714	791.519	411.661	229.655	306.537
Corridor (2^nd F)	6	378.092	668.728	484.011	98.465	463.074
Corridor (1^st F)	6	556.537	1199.647	826.561	251.493	831.272
All	36	206.714	1561.837	757.519	357.318	743.816

In addition, the measured microbial concentration is higher than in the study of Tseng et al.⁴⁹ and Li⁵⁹ but similar to the data obtained in the study of Wu et al.⁶⁰

Development and analysis of different prediction models

Table 2 presents the data from indoor measurements in this office building. The bioaerosol concentration was used as the dependent variable, and the particles with different particle sizes, CO₂ concentration, ventilation rate, temperature, and humidity were entered into the model as independent variables to analyze the qualitative and quantitative relationships between different dependent variables and indoor bioaerosol concentration, and to establish other linear regression models and Poisson regression models.

Bioaerosols and particulate matter

Linear regression analysis

Linear regression analysis is a method of studying quantitative relationships between variables that are interdependent.⁶¹ Linear regression uses the F-test for statistical test, which is used to compare whether the average between two or more groups is significantly different, to test whether the regression model is meaningful. If the model passes the F-test (p < .05), it indicates that the model is meaningful, and at least one independent variable will affect dependent variable. On the contrary, if the model does not pass the F-test (p > .05), it means that the model construction is meaningless, and independent variable will not affect dependent variable. In F (n₁, n₂), n₁ is the molecular degree of freedom, and n₂ is the denominator degree of freedom. Usually, they have no practical significance, and only the p-value needs to be checked for analysis. The p-value is the value corresponding to the T-test. When p < .01, it can be considered that the model is significant at α = 0.01 level, or with a confidence level of 99%. VIF is the variance inflation factor used to determine collinearity problems. When the value of VIF is greater than 5, it indicates the existence of collinearity problems. D-W is the value corresponding to the Durbin Watson test and is a test method for autocorrelation. If the D-W value is around 2 (between 1.7 and 2.3), it indicates that there is no autocorrelation and that the model is well constructed. R² is a coefficient of determination and also a model fitting indicator, used to reflect how much proportion of the fluctuation of the dependent variable can be described by the fluctuation of the independent variable. Adjusted R² is the adjusted coefficient of determination and also the model fitting index. Adjusted R² is more accurate than R² when there are many independent variables.

PM₁, PM_2.5, PM₄, PM₁₀, and TPM were independent variables, while indoor bioaerosol concentration was dependent variable for linear regression analysis. The results showed that the model did not pass the F-test when F-test was conducted on the model, meaning that it is not practical to conduct linear regression analysis on indoor bioaerosol concentration separately for different particulate matters. Therefore, it is not possible to specifically analyse the influence of a single independent variable on the dependent variable. Therefore, the particulate matter (PM₁, PM_2.5, PM₄, PM₁₀, and TPM are analysed as independent and dependent variables, and the results are shown in Table 4.

Table 4.

Results of linear regression analysis (n = 36).

	Unstandardized coefficients		Standardization coefficient	t	p	VIF	R²	Adjustment R²	F
	B	Standard error	Beta	t	p	VIF	R²	Adjustment R²	F
Constant	419.350	89.636	—	4.678	0.000**	—	0.555	0.481	F (5.30) = 7.493, p = .000
PM₁	19780.313	122896.521	0.392	0.161	0.873	399.451
PM_2.5	46962.076	256474.817	2.943	0.183	0.856	17425.196
PM₄	−210679.939	173479.651	-22.018	−1.214	0.234	22175.685
PM₁₀	156209.070	43741.562	19.761	3.571	0.001**	2065.727
TPM	−8769.556	4720.047	-1.155	−1.858	0.073	26.066

D-W: 1.712.

** p < .01.

As shown in Table 4, the model passed the F test (F = 7.493, p = .000<0.05), meaning that at least one PM₁, PM_2.5, PM₄, PM₁₀, and TPM would affect the indoor bioaerosol concentration. And the model equation is indoor biological aerosol concentration = 419.350 + 19780.313 × PM₁ + 46962.076 × PM_2.5 − 210679.939 × PM₄ + 156209.070 × PM₁₀ − 8769.556 × TPM. And the R² value of the model is 0.555, meaning that PM₁, PM_2.5, PM₄, PM₁₀, and TPM could explain 55.5% of the change in indoor bioaerosol concentration. In addition, PM₁₀ will have a very significant positive impact on the indoor bioaerosol concentration, but PM₁, PM_2.5, PM_4, and TPM will not have an impact on the indoor bioaerosol concentration. Therefore, the closely related independent variables should be checked, and the closely related independent variables should be removed before re-analysis.

Therefore, the PM₁₀ index is removed, PM_2.5, PM₄, PM₁, and TPM are taken as independent variables and indoor bioaerosol concentration is taken as dependent variable for linear regression analysis again. The results are shown in Table 5. The R² value of the model is 0.366, meaning that PM_2.5, PM₄, PM₁, and TPM could explain the 36.6% change of indoor bioaerosol concentration, indicating that the accuracy of the model has not been improved after removing PM₁₀. Therefore, the results in Table 4 are used for comparative analysis.

Table 5.

Results of linear regression analysis (n = 36).

	Unstandardized coefficients		Standardization coefficient	t	p	VIF	R²	Adjustment R²	F
	B	Standard error	Beta	t	p	VIF	R²	Adjustment R²	F
Constant	485.648	102.984	—	4.716	0.000**	—	0.366	0.285	F (4.31) = 4.480, p = .006
PM_2.5	−658272.051	192185.219	−41.248	−3.425	0.002**	7094.468
PM₄	330851.108	98959.237	34.576	3.343	0.002**	5232.200
PM₁	318376.407	105774.465	6.304	3.010	0.005**	214.554
TPM	3374.994	3843.974	0.444	0.878	0.387	12.535

D-W: 1.243.

** p < .01.

Poisson regression analysis

PM₁, PM_2.5, PM₄, PM₁₀, and TPM were taken as independent variables and indoor bioaerosol concentration was taken as dependent variables for Poisson regression analysis. The results showed that all models except PM₁₀ could pass the F-test, meaning that different particulate matter could be used to conduct Poisson regression analysis on indoor bioaerosol concentration separately. However, the R² of Poisson regression analysis with PM₁, PM_2.5, PM₄ and TPM as independent variables is 0.002, 0.003, 0.002, and 0.004, respectively, interpreted poorly. Therefore, the particles (PM₁, PM_2.5, PM₄, PM₁₀, and TPM) are all analysed as independent variables.

First, the overall effectiveness of the model is tested and analysed by the likelihood ratio test, and the results are shown in Table 6. The original assumption of the model test here is whether the model quality is the same when the independent variables (PM₁, PM_2.5, PM₄, PM₁₀, and TPM) are added. The p-value is less than 0.05 this time, meaning that the original assumption is rejected, that is, the independent variables put in this model construction are effective, and this model construction is meaningful.

Table 6.

Likelihood ratio test results of poisson regression model.

Model	−2 times log likelihood	Chi square	df	p
Intercept	6058.988
Final model	3164.873	2894.114	5	0.000

Therefore, five items of PM₁, PM_2.5, PM₄, PM₁₀, and TPM are independent variables, while indoor bioaerosol concentration is the dependent variable for Poisson regression analysis. The results are shown in Table 7. The McFadden R² of the model is 0.478, meaning that PM₁, PM_2.5, PM₄, PM₁₀, and TPM can explain 47.8% of the change in indoor bioaerosol concentration, slightly lower than the multiple linear regression can explain 55.5% of the change. The model equation is:

\log (u) = 6.219 + 7.586 \times P M_{1} + 78.395 \times P M_{2.5} - 251.923 \times P M_{4} + 178.307 \times P M_{10} - 10.525 \times T P M

(4)

Where: u represents the expected mean.

Table 7.

Summary of poisson regression analysis results (n = 36).

Item	Regression coefficient	Standard error	z	p
PM₁	7.586	17.225	0.440	0.660
PM_2.5	78.395	35.983	2.179	0.029
PM₄	−251.923	24.302	−10.366	0.000
PM₁₀	178.307	5.971	29.863	0.000
TPM	−10.525	0.699	−15.065	0.000
Intercept	6.219	0.013	485.933	0.000

McFadden R²: 0.478.

PM_2.5 and PM₁₀ will have a significant positive impact on the indoor bioaerosol concentration, and PM₄ and TPM will have a significant negative impact on the indoor bioaerosol concentration. However, PM₁ does not affect the indoor bioaerosol concentration.

Bioaerosol and CO₂ concentration, air exchange rate

Linear regression analysis

Take CO₂ concentration and ACH as independent variables and indoor bioaerosol concentration as dependent variables for linear regression analysis. The results are shown in Table 8. However, the model did not pass the F test (F = 3.013, p=.063>0.05), meaning that CO₂ concentration and ACH do not impact indoor bioaerosol concentration, so linear regression analysis cannot be performed.

Table 8.

Results of linear regression analysis (n = 36).

	Unstandardized coefficients		Standardization coefficient	t	p	VIF	R²	Adjustment R²	F
	B	Standard error	Beta	t	p	VIF	R²	Adjustment R²	F
Constant	−714.221	635.472	—	−1.124	0.269	—	0.154	0.103	F (2.33) = 3.013, p = .063
CO₂	1.689	0.878	0.308	1.924	0.063	1.001
ACH	216.336	136.942	0.253	1.580	0.124	1.001

D-W: 0.662.

** p < .01.

Poisson regression analysis

Take CO₂ concentration and ACH as independent variables and indoor bioaerosol concentration as dependent variables for Poisson regression analysis.

First, the overall effectiveness of the model is tested and analysed by likelihood ratio, and the results are shown in Table 9. The results show that the p-value is less than 0.05, meaning that the original hypothesis is rejected, that is, the independent variables put in this model construction are effective, and this model construction is meaningful.

Table 9.

Likelihood ratio test results of Poisson regression model.

Model	−2 times log likelihood	Chi square	df	p
Intercept	6058.988
Final model	5157.010	901.978	2	0.000

The results are shown in Table 10. McFadden R² of the model is 0.149, meaning that CO₂ and ACH can explain the 14.9% change of indoor bioaerosol concentration. And the model equation is:

\log (u) = 4.684 + 0.002 \times C O_{2} + 0.276 \times A C H

(5)

And CO₂ and ACH have a significant positive impact on the indoor bioaerosol concentration.

Table 10.

Summary of poisson regression analysis results (n = 36).

Item	Regression coefficient	Standard error	z
CO₂	0.002	0.000	23.682
ACH	0.276	0.014	19.541
Intercept	4.684	0.069	67.660

McFadden R²: 0.149.

Bioaerosol and temperature

Linear regression analysis

The linear regression analysis was conducted with temperature as the independent variable and indoor bioaerosol concentration as the dependent variable. The results are shown in Table 11. However, the model did not pass the F test (F = 0.229, p = .635>0.05), meaning that temperature does not have a linear impact on indoor bioaerosol concentration, so linear regression analysis cannot be conducted.

Table 11.

Results of linear regression analysis (n = 36).

	Unstandardized coefficients		Standardization coefficient	t	p	VIF	R ²	Adjustment R²	F
	B	Standard error	Beta	t	p	VIF	R ²	Adjustment R²	F
Constant	1547.231	1651.700	—	0.937	0.355	—	0.007	−0.023	F (1.34) = 0.229, p = .635
T	−29.560	61.784	−0.082	−0.478	0.635	1.000	0.007	−0.023	F (1.34) = 0.229, p = .635

McFadden R²: 0.149.

D-W: 0.609.

** p < .01.

Poisson regression analysis

Poisson regression analysis was conducted with temperature as an independent variable and indoor bioaerosol concentration as a dependent variable.

First, the overall effectiveness of the model is tested and analysed by likelihood ratio, and the results are shown in Table 12. The results show that the p-value is less than 0.05, meaning that the original hypothesis is rejected, that is, the independent variables put in this model construction are effective, and this model construction is meaningful, and Poisson regression analysis can be conducted.

Table 12.

Likelihood ratio test results of poisson regression model.

Model	−2 times log likelihood	Chi square	df	p
Intercept	6058.988
Final model	6019.109	39.879	1	0.000

Therefore, Poisson regression analysis was conducted with temperature as an independent variable and indoor bioaerosol concentration as a dependent variable. The results are shown in Table 13. McFadden R² of the model is 0.007, meaning that the temperature could explain the 0.7% change of indoor bioaerosol concentration. And the model equation is:

\log (u) = 7.689 - 0.040 \times T

(6)

Where u represents the expected mean.

Table 13.

Summary of poisson regression analysis results (n = 36).

Item	Regression coefficient	Standard error	z	p
T	−0.040	0.006	−6.280	0.000
Intercept	7.689	0.169	45.601	0.000

McFadden R²: 0.007.

And temperature will have a significant negative impact on the indoor bioaerosol concentration.

Bioaerosol and relative humidity

Linear regression analysis

The linear regression analysis was conducted to analyse the humidity as an independent variable and the indoor bioaerosol concentration as a dependent variable. The results are shown in Table 14. However, the model did not pass the F test (F = 0.280, p = .600>0.05), meaning that RH does not impact indoor bioaerosol concentration, so linear regression analysis cannot be conducted.

Table 14.

Results of linear regression analysis (n = 36).

	Unstandardized coefficients		Standardization coefficient	t	p	VIF	R ²	Adjustment R²	F
	B	Standard error	Beta	t	p	VIF	R ²	Adjustment R²	F
Constant	380.463	715.455	—	0.532	0.598	—	0.008	−0.021	F (1,34)=0.280, p=.600
RH	6.463	12.220	0.090	0.529	0.600	1.000	0.008	−0.021	F (1,34)=0.280, p=.600

D-W: 0.589.

** p < .01.

Poisson regression analysis

The Poisson regression analysis was conducted to analyse the humidity as an independent variable and the indoor bioaerosol concentration as a dependent variable.

First, the overall effectiveness of the model is tested and analysed by likelihood ratio, and the results are shown in Table 15. The results show that the p-value is less than 0.05, indicating that the independent variable humidity rejected is effective. This model construction is meaningful, and Poisson regression analysis can be carried out.

Table 15.

Likelihood ratio test results of poisson regression model.

Model	−2 times log likelihood	Chi square	df	p
Intercept	6058.988
Final model	6011.085	47.903	1	0.000

The results are shown in Table 16. McFadden R² of the model is 0.008, meaning that humidity can explain the 0.8% change of indoor bioaerosol concentration. The model equation is:

\log (u) = 6.135 + 0.008 \times R H

(7)

Where: u represents the expected mean.

Table 16.

Summary of Poisson regression analysis results (n=36).

Item	Regression coefficient	Standard error	z	p
RH	0.008	0.001	6.938	0.000
Intercept	6.135	0.072	85.529	0.000

McFadden R²: 0.008.

The final specific analysis shows that the regression coefficient value of RH is 0.008, and presents a significant level of 0.01 (z = 6.938, p = .000<0.01), which means that RH will have a significant positive impact on the indoor bioaerosol concentration.

Bioaerosol and all environmental parameters

Linear regression analysis

The linear regression analysis was carried out by taking particles with different particle sizes (PM₁, PM_2.5, PM₄, PM_10, and total particles), CO₂ concentration, air exchange rate, temperature, and humidity as independent variables, and indoor bioaerosol concentration as dependent variables.

As shown in Table 17, the model passed the F test (F = 4.360, p = .002<0.05), meaning that at least one of the particulate matter (PM₁, PM_2.5, PM₄, PM_10, and TPM), CO₂ concentration, air exchange rate, temperature, and humidity will affect the indoor bioaerosol concentration and linear regression analysis can be conducted. The model equation is:

Table 17.

Results of linear regression analysis (n = 36).

	Unstandardized coefficients		Standardization coefficient	t	p	VIF	R²	Adjustment R²	F
	B	Standard error	Beta	t	p	VIF	R²	Adjustment R²	F
Constant	−1000.288	1846.110	—	−0.542	0.593	—	0.601	0.464	F (9.26) = 4.360, p = .002
PM₁	64481.318	136393.400	1.277	0.473	0.640	475.783
PM_2.5	−45532.923	277870.248	−2.853	−0.164	0.871	19779.289
PM₄	−158650.425	187398.373	−16.580	−0.847	0.405	25023.609
PM₁₀	153808.610	48739.519	19.457	3.156	0.004**	2480.191
TPM	−10381.176	5169.215	−1.367	−2.008	0.055	30.232
T	29.265	61.700	0.081	0.474	0.639	1.901
RH	0.702	10.558	0.010	0.066	0.947	1.421
CO₂	1.040	0.777	0.190	1.338	0.192	1.311
ACH	−50.879	134.190	-0.059	−0.379	0.708	1.607

D-W: 1.814.

** p < .01.

Indoor bioaerosol concentration= −1000.288 + 64481.318 × PM₁ − 45532.923 × PM_2.5 − 158650.425 × PM₄ + 153808.610 × PM₁₀ − 10381.176 × TM + 29.265 × T + 0.702 × RH + 1.040 × CO₂ − 50.879 × ACH.

And the model R² value is 0.601, which means that these independent variables could explain 60.1% of the change in indoor bioaerosol concentration. And PM₁₀ will have a significant positive impact on indoor bioaerosol concentration. However, PM₁, PM_2.5, PM₄, TPM, T, RH, CO₂, and ACH would not affect the bioaerosol concentration.

Therefore, linear regression may not be able to effectively use particles with different particle sizes, CO₂ concentration, ACH, T, and RH to explain indoor bioaerosol concentration.

Poisson regression analysis

The Poisson regression analysis was carried out by taking different particle sizes (PM₁, PM_2.5, PM₄, PM₁₀, and TPM), CO₂ concentration, ACH, T and RH as independent variables, and indoor bioaerosol concentration as dependent variables.

First, the overall effectiveness of the model is tested and analysed by likelihood ratio, and the results are shown in Table 18. The results show that the p-value is less than 0.05, indicating that the independent variable humidity rejected is effective. This model construction is meaningful, and Poisson regression analysis can be carried out.

Table 18.

Likelihood ratio test results of poisson regression model.

Model	−2 times log likelihood	Chi square	df	p
Intercept	6058.988
Final model	2800.998	3257.990	9	0.000

As shown in Table 19, the McFadden R² of the model is 0.538, meaning that PM₁, PM_2.5, PM₄, PM₁₀, TPM, T, RH, CO₂, and ACH can explain 53.8% of the changes in the concentration of biological aerosols. The model equation is:

\log (u) = 5.183 + 98.407 \times P M_{1} - 97.234 \times P M_{2.5} - 153.453 \times P M_{4} + 171.800 \times P M_{10} - 12.300 \times T P M + 0.012 \times T - 0.006 \times R H + 0.002 \times C O_{2} - 0.077 \times A C H

(8)

Table 19.

Summary of Poisson regression analysis results (n=36).

Item	Regression coefficient	Standard error	z	p
PM₁	98.407	18.786	5.238	0.000
PM_2.5	−97.234	37.910	−2.565	0.010
PM₄	−153.453	25.426	−6.035	0.000
PM₁₀	171.800	6.509	26.395	0.000
TPM	−12.300	0.751	−16.375	0.000
T	0.012	0.009	1.301	0.193
RH	−0.006	0.002	−3.998	0.000
CO₂	0.002	0.000	15.565	0.000
ACH	−0.077	0.019	−4.006	0.000
Intercept	5.183	0.260	19.966	0.000

PM₁, PM₁₀, and CO₂ will have a significant positive impact on indoor bioaerosol concentration, and PM_2.5, PM₄, TPM, RH, and ACH will have a significant negative impact on indoor bioaerosol concentration. However, temperature does not affect the indoor bioaerosol concentration.

Analysis and establishment of prediction model for bioaerosol concentration

The R² results of linear regression and Poisson regression between bioaerosols and independent variables are shown in Table 20, and the following results are obtained:

1. A single indicator may not be able to use linear regression to explain the bioaerosol concentration. Although most single indicators could use Poisson regression to explain the bioaerosol concentration, compared with multiple indicators, the effect of a single indicator to explain the bioaerosol concentration is not significant.

2. The correlation between particulate matter (especially PM₁₀) and bioaerosol concentration is the best, consistent with the Aileen’s research.⁶²

3. The more indicators describing the concentration of bioaerosols, the higher the model’s explanatory power.

4. Poisson regression could basically explain the concentration of biological aerosols. When T and RH are regarded as single independent variables in relation to the concentration of biological aerosols, Poisson regression and linear regression have the same degree of interpretation, and linear regression cannot perform single-factor analysis on each particle separately.

5. In multifactor analysis, multiple linear regression has a slightly higher interpretation of indoor bioaerosol concentration than Poisson regression, consistent with Tseng's research.⁴⁹ However, only PM₁₀ will have a very significant positive impact on indoor bioaerosol concentration in multiple linear regression. At the same time, PM₁, PM_2.5, PM₄, and TPM indicators do not play their due role, and Poisson regression effectively uses all indicators. Hence Poisson regression is more suitable than linear regression to explain indoor bioaerosol concentration.

Table 20.

R² results of linear regression and poisson regression.

Independent variable	Linear regression (R²)	Poisson Regression(R²)
PM₁	—	0.002
PM_2.5	—	0.003
P_M4	—	0.002
PM₁₀	—	—
TMP	—	0.004
T	—	0.007
RH	—	0.008
CO₂ and ACH	—	0.149
PM₁、PM_2.5、PM₄ 、PM₁₀, and TMP	0.555	0.478
All indicators	0.601	0.538

According to the results of Poisson regression in 3.2.5, except for temperature, other indicators would impact indoor bioaerosol concentration. The possible reason is that Shenzhen is located in a hot summer and warm winter area. Most of the time, the weather is hot and the indoor temperature is relatively warm and stable. And as shown in Table 2, during the measurement, the indoor temperature is 25∼28°C, and the temperature difference is small. Therefore, the temperature index is removed for re-analysis. The likelihood ratio test and analysis of the overall effectiveness of the model can be used for Poisson regression analysis. The results of the Poisson regression analysis are shown in Table 21.

Table 21.

Summary of poisson regression analysis results (n = 36).

Item	Regression coefficient	Standard error	z	p
PM₁	102.513	18.499	5.541	0.000
PM_2.5	−102.482	37.641	−2.723	0.006
PM₄	−148.105	25.054	−5.912	0.000
PM₁₀	169.038	6.144	27.513	0.000
TMP	−11.992	0.712	−16.844	0.000
RH	−0.007	0.001	−4.576	0.000
CO₂	0.002	0.000	16.905	0.000
ACH	−0.074	0.019	−3.863	0.000
Intercept	5.485	0.115	47.502	0.000

McFadden R²: 0.537.

As shown in Table 21, PM₁, PM₁₀, and CO₂ will have a significant positive impact on indoor bioaerosol concentration, and PM_2.5, PM₄, TPM, RH, and ACH will have a significant negative effect on indoor bioaerosol concentration. And the model McFadden R² is 0.537, meaning PM₁, PM_2.5, PM₄, PM₁₀, TPM, RH, CO₂, and ACH can explain 53.7% of the change of biological aerosol concentration, so the fitting effect is significant.

According to the regression coefficient, the indoor bioaerosol concentration is linked to the particle diameter. The regression coefficient of particles is significantly greater than that of other parameters, and the regression coefficient is gradually increasing with the increase of particle radius. This shows that particles may have a greater impact on indoor bioaerosol than other indicators, and with the increase of particle radius, the impact on the concentration of bioaerosol gradually increases, of which PM₁₀ has the largest impact on indoor bioaerosol concentration. The possible reason is that the particle size of PM₁₀ is larger than PM_2.5 and PM₁, and it has a larger surface area and mass, so bacteria are more likely to attach to PM₁₀.¹³

However, the influence of oversize particles on bioaerosol concentration may be reduced. When the particle concentration is greater than 10 μm, the effect on indoor bacterial concentration decreases with the increase in particle size⁶³ Therefore, PM₁₀ may be the indicator that has the greatest impact on indoor bioaerosol concentration and may be used as an indirect indicator of indoor bioaerosol concentration.

The fitting equation of the final indoor bioaerosol concentration is equation (9).

\log (u) = 5.485 + 102.513 \times P M_{1} - 102.482 \times P M_{2.5} - 148.105 \times P M_{4} + 169.038 \times P M_{10} - 11.992 \times T P M - 0.007 \times R H + 0.002 \times C O_{2} - 0.074 \times A C H

(9)

The original data is substituted into the validation and corrected according to equation (3). The corrected indoor bioaerosol concentration prediction equation of the office building is equation (10).

U = e^{u} - 130

(10)

As shown in Figure 6, compared with the original data, the average error rate of the modified analogue data is −0.02% and the absolute error is 181 CFU/m³, and the median is similar. Therefore, the revised simulation prediction data is accurate and reliable. However, the absolute error of the seventh and ninth groups of data (measured at 9:00 and 11:00 in the 2^nd F conference room) is −567.59 CFU/m³ and −518.57 CFU/m³. The reason may be that on the day of measurement, workers entered the air conditioning for maintenance at 9:00 and 11:00 respectively. Although researchers promptly stopped it, it still brought in external microorganisms, particulate matter, etc., damaging the original indoor environment. Similarly, the measurement error of the 31st group of data (measured at 9:00 am in the 1^st F restroom) is 676.25 CFU/m³. The office building's working time is 9:00, and the time for people to go to the bathroom is relatively concentrated. The measurement results show (see Table 2) that the particle concentration in the 1^st F toilet is significantly higher than that in the 2^nd F toilet (the researchers ensure that no one goes to the bathroom on the day of measurement except the researchers), which may lead to higher predicted values of bioaerosol concentration. The most likely cause of these three groups of abnormal data is the influence of human factors.²² Therefore, this reflects the influence of human behaviour on indoor bioaerosol concentration from the side, and human behaviour can be included in the prediction model, which may further improve the accuracy and effectiveness of the prediction model.

Conclusions

In this study, the indoor air quality of six typical spaces in an office building in Shenzhen, China, was measured. The influential factors of indoor bioaerosol concentration were studied using linear regression analysis and Poisson regression analysis methods. The results show that the Poisson regression has better prediction of the bioaerosol concentration. PM₁, PM_10, and CO₂ will have a significant positive impact on the indoor bioaerosol concentration, i.e., mean bioaerosol concentration increases with increasing PM₁, PM_10, and CO₂. Conversely, PM_2.5, PM₄, TPM, RH, and ACH are associated with reduced indoor bioaerosol concentrations. In addition, PM₁₀ may be the most influential indicator of indoor bioaerosol concentration. A simple, fast, and low-cost online monitoring method for indoor microorganisms was developed and a prediction model for indoor bioaerosol concentration in an office building in Shenzhen was established.

Whilst results are promising, it is prudent to outline the limitations of the work. Firstly, the model only applies indoor factors as input parameters. The relatively few input features indeed simplify model training and enhance repeatability, however, other characteristics that may affect indoor bioaerosol concentration, such as outdoor environment, seasonal and meteorological conditions,^64,65 human occupancy levels^22,66,67 or behaviour activities,^22,68,69 are not taken into account as model inputs. These characteristics thus should be included in the model to further improve the prediction accuracy. Then, since the sampling points in this study are all located in the office buildings in Shenzhen, the empirical equation in this study is only applicable to the office buildings in Shenzhen but can be used as the research basis and basic model of prediction models for other similar climate regions and indoor spaces. It is likely that model training based on specific field monitoring data may be necessary. Thirdly, although the cost of environmental sensors is affordable, there are still inherent uncertainties through the measurement and calibration of these instruments, which may reduce the predictive accuracy of the model.⁷⁰ Finally, the study used only indoor bioaerosol concentrations, rather than specific viruses. As a result, further research is needed to understand the classification and composition of indoor air microorganisms, to determine whether there are pathogenic fungi species in the research site, which will allow a more detailed assessment of the exposure risk of staff in different office buildings.

Accepting the potential limitations to the study, this study does establish a relationship between the indoor measurable parameters of the office building and bioaerosol concentration and demonstrates that the concentrations of indoor bioaerosols can be determined and predicted based on observed parameters (including particulate matter, CO₂ concentration, ACH, T, and RH) using the real-time environmental monitoring sensors in a real office building. Because the prevention of epidemic spread in the indoor environment is key, there is an urgent need for a simple, fast, low-cost, and efficient technology that can accurately monitor the concentration of bioaerosols in the air, so that bioaerosol monitoring and forecasting systems can be widely used in the indoor environment. This prediction will enable remedial measures to be evaluated and implemented (such as increasing fresh air supply) to maintain indoor air quality and timely predict the risk of indoor epidemic transmission.

Footnotes

Acknowledgements

Thanks to Shenzhen Lemniscare Medical Technology Co., Ltd., China, for their support in instrument and equipment, and thanks to the Research Institute of Tsinghua University in Shenzhen, Shenzhen Lihe Technology Co., Ltd. for their support in the microbiology laboratory and measurement site. The authors are grateful to Mr. Yezhao Cai of Tsinghua Innovation Centre in Zhuhai for his English proof reading.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research is financially supported by Practical Research on Low Carbon and Healthy Building System in Fujian and Guangdong for China Construction Fangcheng Investment& Development Group Co., Ltd. (CSCECFC-2022-KJZX--06), Research on Intelligent Systems of Healthy Residential Buildings for Smart Building Research Centre of China Real Estate Association, Industrial Transfer, Environmental Regulation and Industrial Ecological Development: A Case Study of Anhui Province.(SK2020A0572). Key Research Project of Humanities and Social Sciences in Universities in Anhui Province: Research on Green Finance Development from the Perspective of Regional Economic Development and Carbon Tax Mechanism Constraints--Taking Anhui Province as an Example (SK2020A0570) and Research and Verification of Epidemic Prevention System of Residential Buildings (Shenzhen science and technology plan project in engineering construction field in 2021).

ORCID iD

Xiaoqiang Gong

References

Awada

Becerik-Gerber

Hoque

, et al. Ten questions concerning occupant health in buildings during normal operations and extreme events including the COVID-19 pandemic. Build Environ 2021; 188: 107480.

Gong

Liu

, et al. Development of a healthy assessment system for residential building epidemic prevention. Build Environ 2021; 202: 108038.

Cincinelli

Martellini

. Indoor air quality and health. Int J Environ Res Publ Health 2017; 14(11): 1286.

Atarodi

Karimyan

Gupta

, et al. Evaluation of indoor air quality and its symptoms in office building – a case study of Mashhad, Iran. Data Brief 2018; 20: 74–79.

Bartlett

Kennedy

Brauer

, et al. Evaluation and a predictive model of airborne fungal concentrations in school classrooms. Ann Occup Hyg 2004; 48(6): 547–554. DOI: 10.1093/annhyg/meh051

Chao

Schwartz

Milton

, et al. Populations and determinants of airborne fungi in large office buildings. Environ Health Perspect 2002; 110(8): 777–782.

Goh

Obbard

Viswanathan

, et al. Airborne bacteria and fungal spores in the indoor environment a case study in Singapore. Acta Biotechnol 2000; 20(1): 67–73.

Law

AKY

Chau

Chan

GYS

. Characteristics of bioaerosol profile in office buildings in Hong Kong. Build Environ 2001; 36(4): 527–541.

Mui

Wong

Hui

. Risks of unsatisfactory airborne bacteria level in air-conditioned offices of subtropical climates. Build Environ 2008; 43(4): 475–479.

10.

Chan

Rao

, et al. Characteristics, determinants, and spatial variations of ambient fungal levels in the subtropical Taipei metropolis. Atmos Environ (1994). 2007, 41(12):2500-2509.

11.

Zhu

Phelan

Duan

, et al. Characterizations and relationships between outdoor and indoor bioaerosols in an office building. China Particuol 2003; 1(3): 119–123.

12.

Ming-Jian

Guo-Qing

Xin

. Review of key technologies for forecast of indoor microbial contamination levels. China Environ Sci 2018; 38(11): 4040–4049.

13.

Zhao

Chen

Miao

, et al. Sources and influence factors of bacterial aerosol in university classrooms. Build Sci 2020; 36(10): 20–26.

14.

Lopez

Dessi

Cabrera-Codony

, et al. CO₂ in indoor environments: from environmental and health risk to potential renewable carbon source. Sci Total Environ 2023; 856(Pt 2): 159088.

15.

González-Martín

Kraakman

NJR

Pérez

, et al. A state–of–the-art review on indoor air pollution and strategies for indoor air pollution control. Chemosphere 2021; 262: 128376.

16.

Kraakman

NJR

González-Martín

Pérez

, et al. Recent advances in biological systems for improving indoor air quality. Rev Environ Sci Biotechnol 2021; 20(2): 363–387.

17.

Kim

Kabir

Jahan

. Airborne bioaerosols and their impact on human health. J Environ Sci (China) 2018; 67: 23–35.

18.

Marcovecchio

Perrino

. Bioaerosol contribution to atmospheric particulate matter in indoor university environments. Sustainability 2021; 13(3): 1149.

19.

Pertegal

Lacasa

Cañizares

, et al. Understanding the influence of the bioaerosol source on the distribution of airborne bacteria in hospital indoor air. Environ Res 2023; 216: 114458.

20.

Duquenne

. On the identification of culturable microorganisms for the assessment of biodiversity in bioaerosols. Ann Work Expo Health 2018; 62(2): 139–146.

21.

Schafer

Weiss

Jackel

. Preliminary validation of a method combining cultivation and cloning-based approaches to monitor airborne bacteria. Ann Work Expo Health 2017; 61(6): 633–642.

22.

Lee

JYY

Miao

Chau

RLT

, et al. Artificial intelligence-based prediction of indoor bioaerosol concentrations from indoor air quality sensor data. Environ Int 2023; 174: 107900.

23.

Obayori

. Chapter 4 - indoor and outdoor aeromicrobiology. In: Developments in applied microbiology and biotechnology, aeromicrobiology. New York: Academic Press, 2023, pp. 59–88. DOI: 10.1016/B978-0-323-96122-6.00009-8

, et al. Dynamics of SARS-CoV-2 spreading under the influence of environmental factors and strategies to tackle the pandemic: a systematic review. Sustain Cities Soc 2022; 81: 103840.

, et al. The effects of air pollution, meteorological parameters, and climate change on COVID-19 comorbidity and health disparities: a systematic review. Environmental Chemistry and Ecotoxicology 2022; 4: 194–210.

26.

Aganovic

Cao

, et al. Estimating the impact of indoor relative humidity on SARS-CoV-2 airborne transmission risk using a new modification of the Wells-Riley model. Build Environ 2021; 205: 108278.

27.

Jones

Harrison

. The effects of meteorological factors on atmospheric bioaerosol concentrations—A review. Sci Total Environ 2004; 326(1): 151–180.

28.

Lin

Marr

. Humidity-Dependent decay of viruses, but not bacteria, in aerosols and droplets follows disinfection kinetics. Environ Sci Technol 2020; 54(2): 1024–1032.

, et al. Predicting the effects of environmental parameters on the spatio-temporal distribution of the droplets carrying coronavirus in public transport – A machine learning approach. Chem Eng J 2022; 430: 132761.

, et al. An overview of SARS-CoV-2 transmission and engineering strategies to mitigate risk. J Build Eng 2023; 73: 106737.

. A mathematical model for predicting the viability of airborne viruses. J Virol Methods 2010; 164(1): 88–95.

32.

Liu

Luo

, et al. Prediction and control of aerosol transmission of SARS-CoV-2 in ventilated context: from source to receptor. Sustain Cities Soc 2022; 76: 103416.

33.

Park

Lee

, et al. The effects of indoor temperature and humidity on local transmission of COVID-19 and how it relates to global trends. PLoS One 2022; 17(8): e0271760.

. The transmission of SARS-CoV-2 is likely comodulated by temperature and by relative humidity. PLoS One 2021; 16(7): e0255212.

, et al. Human occupancy as a source of indoor airborne bacteria. PLoS One 2012; 7(4): e34867.

36.

Cai

. A novel CO₂-based demand-controlled ventilation strategy to limit the spread of COVID-19 in the indoor environment. Build Environ 2022; 219: 109232.

, et al. Safe CO₂ threshold limits for indoor long-range airborne transmission control of COVID-19. Build Environ 2023; 234: 109967.

38.

Peng

Jimenez

. Exhaled CO₂ as a COVID-19 infection risk proxy for different indoor environments and activities. Environ Sci Technol Lett 2021; 8(5): 392–397.

, et al. Assessment of indoor air quality and risk of COVID-19 infection in Spanish secondary school and university classrooms. Build Environ 2022; 226: 109717.

, et al. Adverse effects of exposure to fine particles and ultrafine particles in the environment on different organs of organisms. J Environ Sci (China) 2024; 135: 449–473.

, et al. Atmospheric particulate matter effects on SARS-CoV-2 infection and spreading dynamics: a spatio-temporal point process model. Environ Res 2022; 212: 113617.

, et al. Machine learning reveals that prolonged exposure to air pollution is associated with SARS-CoV-2 mortality and infectivity in Italy. Environ Pollut 2020; 267: 115471.

43.

Kim

Huang

, et al. Experimental and numerical analysis of indoor air quality affected by outdoor air particulate levels (PM_1.0, PM_2.5 and PM₁₀), room infiltration rate, and occupants’ behaviour. Sci Total Environ 2022; 851: 158026.

, et al. Biosensors for the detection of respiratory viruses: a review. Talanta Open 2020; 2: 100007.

, et al. Prediction model for airborne microorganisms using particle number concentration as surrogate markers in hospital environment. Int J Environ Res Public Health 2020; 17(19).

, et al. Environmental monitoring and disease prediction. Advancements and Technologies in Pig and Poultry Bacterial Disease Control 2021: 145–169.

, et al. Development of the internet of things sensory technology for ensuring proper indoor air quality in hospital facilities: taxonomy analysis, challenges, motivations, open issues and recommended solution. Measurement 2022; 192: 110920.

48.

Wang

Lei

, et al. Quantifying the dynamic characteristics of indoor air pollution using real-time sensors: current status and future implication. Environ Int 2023; 175: 107934.

, et al. Examining the feasibility of prediction models by monitoring data and management data for bioaerosols inside office buildings. Build Environ 2011; 46(12): 2578–2589.

50.

Zhang

Tian

, et al. On-site investigation of the concentration and size distribution characteristics of airborne fungi in a university library. Environ Pollut 2020; 261: 114138.

51.

State Administration for Market Regulation, Standardization Administration . State administration for market regulation and standardization administration. Standards for indoor air quality GB/T 18883-2022. Beijing: China Construction Industry Press, 2022.

, et al. Mapping urban aerosolized fungi: predicting spatial and temporal indoor concentrations. Hum Ecol Rev 2018; 24: 81–103.

53.

Municipality MBOS . Overview of Shenzhen’s climate and characteristics of four seasons, 2022, http://weather.sz.gov.cn/qixiangfuwu/qihoufuwu/qihouguanceyupinggu/qihougaikuang/

54.

Cehlin

Karimipanah

Larsson

, et al. Comparing thermal comfort and air quality performance of two active chilled beam systems in an open-plan office. J Build Eng 2019; 22: 56–65.

55.

Tang

. Study on ventilation rates and assessment of infection risks of COVID-19 in an outpatient building. J Build Eng 2021; 42: 103090.

56.

Zender Świercz

. Improvement of indoor air quality by way of using decentralised ventilation. J Build Eng 2020; 32: 101663.

57.

Yuzhu

Fenghua

. Mathematical treatment of error in the indirect measurement. Physical Experiment of College 2001; 02: 51–53.

58.

Nitish

Sahar

Elyse

, et al. Chapter 6 - qualitative and quantitative research methods, Substance use and addiction research. New York: Academic Press, 2023, pp. 65–69. DOI: 10.1016/B978-0-323-98814-8.00008-1

59.

. Evaluation of indoor air quality in residencies and office buildings. Taipei: National Science Council, 1998. NSC 87-EPA-P-002-013.

60.

Chiang

, et al. Changing microbial concentrations are associated with ventilation performance in Taiwan’s air-conditioned office buildings. Indoor Air 2005; 15(1): 19–26.

61.

Qiu

. Chapter 8 linear and nonlinear regression models, applied multivariate statistical analysis and related topics with R. Les Ulis: EDP Sciences 2021: 111–130. DOI: 10.1051/978-2-7598-2602-5.c009