HOME

This study was conducted on 76 gers and 40 dwellings in ger district, Ulaanbaatar, with 36 gers measured in Jan., 2016, 40 gers measured in Jan, 2018 and 40 dwellings measured in Jan, 2017. In each year, researchers measured temperature, relative humidity, and PM_2.5 number concentration of residence for 4 days during daytime (11:00 - 18:00). Temperature and relative humidity were measured by Onset HOBO Datalogger UX100-003 (Onset Computer Corporation, USA), and the number concentration of PM_2.5 was measured by utilizing a Dylos DC1799 (Dylos Corporation, USA). The instruments were located at a minimum distance of 0.5 m from the floor. The measurement interval of all instruments was one minute. The obtained number concentrations of PM_2.5 were converted to mass concentrations through the following equation (1).

During the measurement, the researchers observed the activities of the inhabitants and made notations in an observation log. The start and end times of activities such as cooking, fuel addition and fuel amount, indoor smoking, candle usage, food and beverage consumption, residents’ exiting their residences, cleaning and ventilation were investigated through observation log. The amount of added fuel was examined quantitatively using a scale.

We used public outdoor PM_2.5 concentration data of ger district (Figure1). Measuring station was located in Nisekh, sub-district of Khaan-Uul district in Ulaanbaatar (Figure 1). These data were accessed from the OpenAQ Platform (openaq.org) and originated from Mongolia National Agency of Meteorology and Environmental Monitoring (Accessed 3 March, 2018).

Figure 1. 
Indoor-outdoor PM_2.5 distribution of two residential types

Error bars indicate one standard deviation from sample average. Regression equation was C_in = 1.07C_out + 32.78 (r = 0.63; 95% C.I 0.47-0.75, p<0.001) for ger and C_in = 0.86C_out + 0.52 (r = 0.72; 95% C.I 0.52-0.84, p<0.001) for dwellings.

The average indoor PM_2.5 concentrations were calculated as the arithmetic mean of the measurement time after excluding adjustment time of the device. The average outdoor PM_2.5 concentrations were calculated using the arithmetic mean of PM_2.5 from 09:00 to 18:00.

The I/O ratio, representing the strength of indoor PM_2.5 concentration in that household, was calculated by dividing average indoor PM_2.5 concentration by average outdoor PM_2.5 concentration of the day. Based on I/O ratio, we excluded data that were more than three standard deviations from the mean. One dwelling was excluded by the criterion. Data with instrument malfunction during observation was removed. The final number of samples were 76 for the gers and 38 for the dwellings.

Simple linear regression was applied to obtain correlation of indoor-outdoor PM_2.5 concentrations for the two residential types. The regression equation was as follows: C_out = aC_in+ b, where C_in and C_out represent average indoor and outdoor PM_2.5 concentrations, respectively.

Multiple linear regression was used to identify determinants of daytime average indoor PM_2.5 concentrations in ger. Gers with missing observation were removed from original data. The final number of samples used in multiple linear regression was 70. The predictors consist of the following three categories outdoor PM_2.5 concentrations, occupants’ indoor activities and characteristics of gers. The activity factors include cooking frequency, fuel usage, and smoking (indoor smoking more than once during observation time). And separation of cooking room and stove type (traditional, improved) were classified into the characteristics of gers. To make the PM_2.5 data normally distributed, we performed a square root transformation for indoor PM_2.5 concentration and outdoor PM_2.5 concentration. After transformation, PM_2.5 data met normality assumption for the reliability of statistical analysis.

The peak analysis was conducted for residents’ activity factor (fuel usage, cooking and smoking) to determine effect to the real-time indoor PM_2.5 concentration in ger. Total number of gers used in peak analysis was 68 (8 gers were excluded for missing observation log and measuring time error).

We defined the peak as the case where the difference between the lowest value and the highest value within 30 minutes of active time was 35 or more. All statistical analysis was performed using R software, version 3.4.1 (R Core Development Team, 2017).

Table 1 shows descriptive statistics of temperature, humidity and indoor PM_2.5 concentrations by observation dates in gers and dwellings. On average, temperature and indoor PM_2.5 concentration were higher in ger, but relative humidity was higher in the dwelling.

We used the coefficient of variation (c_v) as an indicator of between-home variability. In case of temperature, between-home variability was relatively small (c_v=0.18). It indicated that there was not much difference in temperature between gers. The relative humidity were moderately variable between gers (c_v=0.33). However, the variability of indoor PM_2.5 was considerable (c_v=0.59). The between-home variability of temperature and indoor PM_2.5 concentration in dwelling showed similar to ger (c_v=0.16, 0.62, respectively). The variability of relative humidity was slightly smaller than that of ger (c_v=0.26), but it was not statistically significant.

The average indoor PM_2.5 concentration in the ger was higher than dwelling (p<0.05). The proportion of households where average indoor PM_2.5 concentration is high (> 300 μg/m³) differed between the two residential types. In the case of ger, 15.8 % (n=12) of households had a PM_2.5 level above 300 μg/m³, compared to 7.9 % (n=3) for dwellings. 6.6 % (n=5) of gers exhibited extremely high indoor PM_2.5 levels over 400 μg/m³, whereas no applicable data were observed in dwellings.

Table 2 shows the I/O ratios of gers and dwellings. The I/O ratio of ger was significantly higher than that of dwellings (p<0.001). It suggested that indoor PM_2.5 was still larger than dwellings after accounting for influence of outdoor air. In the case of the I/O ratio over 1 households, ger was 64 % (n=49) and dwelling was 42 % (n=16). For households with I/O ratio over 2, ger was 17 % (n=13) and there was no dwelling.

As shown in Figure 1, correlation of indooroutdoor PM_2.5 concentration was considerable (r = 0.63, 0.72 for ger and dwelling).

Despite the short observation period, daily variation of average outdoor PM_2.5 concentration was large so that data was relevant to infer indoor-outdoor PM_2.5 correlation (p<0.001 for both). Regression coefficients were not significantly different between two residential types (i.e. effect modification of outdoor PM_2.5 concentration was not confirmed).

In regression models, mean absolute error calculated as average absolute distance from expected value was larger in gers (p<0.05). It resulted in a smaller correlation coefficient of ger than that of dwelling.

Table 3 shows the multiple linear regression result on average indoor PM_2.5 concentration in ger. Under this model, outdoor PM_2.5 concentration was the most influential factor for average indoor PM_2.5 concentration (p<0.001).

The average indoor PM_2.5 concentration was higher in the house where smoking was observed more than once during the observation period (p<0.05). The separated cooking room, cooking frequency, stove type, fuel amount and fuel usage frequency were not significantly associated with average indoor PM_2.5 concentration.

Three activity factors - fuel usage, cooking and indoor smoking affected real-time PM_2.5 concentration causing peaks of indoor PM_2.5 concentration (Table 4). More than half of the fuel usage raised the peaks of more than 35 μg/m³ within 30 minutes before and after fuel usage. In case of the cooking, the average increase was larger than fuel usage, but the peak occurrence rate was smaller than that of fuel usage. As for indoor smoking, it showed that most smoking caused a strong peak within a short time. The average increase of indoor smoking was 407.7 μg/m³, which was significantly the largest factor among the three activity factors.

Figure 2 shows examples of real-time changes of indoor PM_2.5 concentration in gers according to three activity factors. Each activity frequently caused large increases of PM_2.5 concentration in a short time. After peak occurred, PM_2.5 concentration gradually decreased and returned to the original level.

Figure 2. 
Influences of indoor activities to real-time indoor PM_2.5 concentrations

The overall indoor environment of ger was not suitable for occupant’s well-being. The relative humidity was considerably deviated from the indoor air quality standard. Previous studies suggested that atmospheric humidity of at least 30 - 40 % had to be maintained to prevent drying of the nasal mucosa membrane [16]. However, small proportion of households met this standard (22 %). The low indoor humidity was known to affect eyes and skin irritation [17, 18]. According to our survey, substantial proportion of residents had been suffering eye irritation and skin problems (42 %, 17 %, respectively). It is necessary to maintain relative humidity at an adequate level for reducing related adverse health risk of residents.

The indoor PM_2.5 concentrations of ger need instant improvement. Indoor PM_2.5 concentrations of gers was extremely high on average. It was much worse than the WHO Air Quality Guideline of indoor PM_2.5 concentration (25 μg/m³). In some gers, average PM_2.5 concentration exceeded the industrial air quality standard. The threshold of eight hour time weighted average exposures to PM_2.5 was 300 μg/m³ [19]. This threshold was exceeded in some gers (15.8 %).

In the most gers, the temperature was acceptable. The average temperature in gers met the winter temperature standard of 19.2 - 27.8 ºC of the American Society of Heating, Refrigerating and Air-conditioning Engineers [20].

There was a remarkable difference in PM_2.5 level for each residential type: almost as factor of one and half for I/O ratio (Table 2). The I/O ratio was commonly used to indicate the strength of indoor generated pollutants [21]. High indoor PM_2.5 level in ger is presumably caused by influence of indoor generated pollutants and infiltration of outdoor pollution.

It could be inferred that the characteristics of house may be related to high indoor PM_2.5 level in ger. Because indoor activities of dwelling was not particularly different with ger [12]. The effect of indoor generated pollutants on PM_2.5 level according to the characteristics of house remain to be determined.

Controlling indoor particulate matter concentration will be necessary to improve overall health outcomes of Ulaanbaatar. More than half families in Ulaanbaatar live in ger or dwelling other than apartment [11]. Our results suggest a large number of population of Ulaanbaatar live in an indoor environment that is similar to, or even worse than outdoor environment.

Determining factors of PM_2.5 was a crucial issue for Mongolian government to establish policy for improving indoor PM_2.5 level. Our results provided an understanding of factors for indoor PM_2.5 concentrations. As expected, the outdoor PM_2.5 concentration was a significant factor of indoor PM_2.5 concentration of ger. We showed significant association of indoor-outdoor PM_2.5 correlation in ger district, Ulaanbaatar (Table 2). It implied that serious air pollution in Ulaanbaatar could directly affect an indoor air of residence in Ulaanbaatar, possibly causing an adverse health effects on residents.

Among indoor activities, smoking was the most influential factor of indoor PM_2.5 concentration. Indoor smoking was only significantly associated with average indoor PM_2.5 concentrations. Influence of second hand smoke exposure on human health is well documented [22]. It indicated that indoor smoking might have large impact on occupant’s health compared to the other indoor activities.

The impacts of cooking and fuel usage on average PM_2.5 concentrations were not determined. However, they frequently increased indoor PM_2.5 concentrations after actions (Table 3). It suggested that real-time analysis was more suitable to determine impact of these activities on indoor PM_2.5 concentrations.

All of three indoor activities considerably affected real-time indoor PM_2.5 concentrations raising peak after actions. Smoking had the largest peak occurrence rate and magnitude. Peaks of PM_2.5 were observed immediately after smoking for most of cases (Table 3).

Cooking also had a large impact on indoor PM_2.5 concentration. For cooking, peak occurrence rate was the smallest, but the peak magnitude was larger than fuel usage. It was known that certain type of cooking had a significant effect on indoor PM_2.5 concentrations. Frying food affected indoor PM_2.5 in the cooking space [23]. According to the observations in this study, only 24.3 % of ger had separate cooking space. Therefore, indoor PM_2.5 exposure from cooking might be considerable in ger.

Because 24 hours could not be observed, it was difficult to understand how the behavior factors of various residents affected indoor air quality. There were overlapping of various activities at the time when peak concentration occurred. In particular, it was difficult to grasp the influence of certain variables in situations where various activities overlapped in the time of peak occurrence. There might be observational biases from different observers. Outdoor PM_2.5 was approximated using publicly available data. Given that the PM_2.5 distribution in Ulaanbaatar could be spatially different, it would have been better to collect outdoor PM_2.5 concentration data near target households as much as possible to ascertain outdoor influence on indoor PM_2.5 concentration. However, we used publicly available data, because device could not directly measure outdoor PM_2.5 due to very low temperature of Ulaanbaatar.