Time Series Analysis of Climate and Air Pollution Factors Associated with Atmospheric Nitrogen Dioxide Concentration in Japan

Abstract

Nitrogen dioxide (NO₂) is an air pollutant discharged from combustion of human activities. Nitrous acid (HONO), measured as NO₂, is thought to impact respiratory function more than NO₂. HONO and NO₂ have an equilibrium relationship, and their reaction is affected by climate conditions. This study was conducted to discuss the extent of HONO contained in NO₂, depending on the level of urbanization. Whether climate conditions that promote HONO production enhanced the level of NO₂ measured was investigated using time series analysis. Climate and outdoor air pollution data measured in April 2009–March 2017 in urban (Tokyo, Osaka, and Aichi) and rural (Yamanashi) areas in Japan were used for the analysis. Air temperature had a trend of negative associations with NO₂, which might indicate the decomposition of HONO in the equilibrium between HONO and NO₂. The associations of relative humidity with NO₂ did not have consistent trends by prefecture: humidity only in Yamanashi was positively associated with NO₂. In high relative humidity conditions, the equilibrium goes towards HONO production, which was observed in Yamanashi, suggesting the proportion of HONO in NO₂ might be low/high in urban/rural areas.

1. Introduction

Nitrogen dioxide (NO₂) is an air pollutant discharged from combustion of human activities such as driving cars, and numerous epidemiological studies have suggested a relationship between NO₂ and impaired respiratory function or asthma symptoms [1,2,3,4].

NO₂ is monitored at air pollution monitoring stations (APMSs) in Japan [5], and previous studies have implied that climate factors affect NO₂ levels [1,6,7]. Since climate and air pollutant time series data demonstrate seasonality and autocorrelation effects, those effects should be adjusted to find associations among the factors. A large number of studies investigated the effect of NO₂ on human health status such as morbidity and mortality of diseases using time series analysis (e.g., [8,9,10]). In those studies, climate and other pollutant factors were dealt as confounders/covariates of NO₂ to control (i.e., an outcome, exposure, and confounders/covariates were the number of cases, NO₂, and climate/other pollutant factors, respectively), indicating they are associated with NO₂ concentration. However, studies that directly estimate the relationships between climate/co-pollutant factors and NO₂ using time series analysis are limited.

In the environment, nitrous acid (HONO) is in equilibrium with NO₂, nitric oxide (NO) and H₂O (Equation (1)). It is reported that this is the main chemical reaction for HONO production [11,12]:

$$2\mathrm{HONO}\rightleftarrows \mathrm{NO}+{\mathrm{NO}}_2+{\mathrm{H}}_2\mathrm{O}$$

(1)

The direction of this reaction depends on climate conditions: high air temperature promotes NO₂ production since the forward reaction is endothermic, and high humidity promotes HONO production since an abundance of H₂O promotes the reverse reaction [7,13], in which the effects on HONO and NO₂ production are also reversed. The other characteristics differ between HONO and NO₂ with climate conditions are water solubility and decomposition by sunlight. HONO is much more water-soluble [12] and is more easily decomposed by sunlight [7] than NO₂. Moreover, it has been found that conventional assays of NO₂ such as the Saltzmann reagent method and NOx analyzer measure HONO as NO₂ [14]. A few studies have suggested that HONO impacts respiratory function more than either NO₂ or sulfur dioxide alone [15], and that the main sources of indoor HONO are indoor combustion and outdoor NO₂ [6]. The World Health Organization [16] reported that it is unclear to what extent the health effects observed in epidemiological studies are attributable to NO₂ itself, or rather to primary and secondary combustion-related products (i.e., organic carbon and HONO) with which it is typically correlated. However, it is argued that the available scientific literature has not accumulated sufficient evidence to justify revising the existing WHO air quality guidelines for annual NO₂ concentrations. To clarify the effects of NO₂ and HONO, monitoring the concentrations of environment HONO at APMSs is desired; however, these concentrations are not currently measured.

In the equilibrium reaction (Equation (1)), HONO production is promoted by a temperature decline and humidity rise, and vice versa for NO₂. Thus, the associations of climate factors with NO₂ might indicate the effects of these factors on either NO₂ or HONO. This study was conducted to investigate the relationships between climate factors and NO₂ concentration measured at APMSs using time series analyses. Then, from the main results, the level of HONO (measured as a part of NO₂) in NO₂ concentration was discussed.

2. Materials and Methods

2.1. Study Area

Tokyo, Osaka, Aichi, and Yamanashi Prefectures were chosen as study areas. Tokyo, Osaka and Aichi were chosen as urban study sites. These prefectures include the three largest cities in Japan, the special wards of Tokyo, Osaka City, and Nagoya City, and are highly populated (first, third, fourth highest populations out of 47 prefectures in Japan in 2015, respectively) [17]. Yamanashi was selected as a rural study area and has a lower population (the seventh lowest out of 47 prefectures) and is located around the mid-point between Tokyo and Aichi.

2.2. Data Collection

Time series data of outdoor air pollution and climate factors between April 2009 and March 2017 were used in this study. From each prefecture studied, the roadside APMS (RAPMS) that had the highest monthly NO₂ emission in September 2009 and the closest ambient air pollution monitoring station (AAPMS) from each RAPMS was selected, and hourly outdoor air pollutant concentration data were collected from the monitoring stations [5]. The air pollutants consisted of NO₂ (ppb), nitric oxide (NO, ppb), and ozone (O₃, ppb). If data were missing for three consecutive hours, the missing data were kept as missing, otherwise they were imputed by linear interpolation. Each daily datum was calculated as the average of the hourly data of that day unless they contained missing data. Then, all daily missing data were imputed further by linear interpolation. The numbers of missing hourly and daily data by the compounds and APMSs are shown in Table S1. The concentration of active O₃ (AO) was defined as the difference in concentrations between potential ozone (PO) and NO₂ at the RAPMS. PO was calculated as the sum of NO₂ and O₃ at the AAPMSs. AO was used as the alternative of O₃ at the RAPMS for the analysis below because the concentration of O₃ is negligibly low at roadside due to high consumption by reactions with NO and others. Daily climate data were also collected from the meteorological station [18] in each prefecture that was the closest to the air pollution monitoring stations. The climate data included the mean air temperature (°C), relative humidity (%), mean wind speed (m/s), total duration of sunshine (hour), and mean global solar radiation (MJ/m²). The average monthly concentrations of outdoor pollutants and the average or total monthly values of climate factors were calculated from the data collected. All the data used were publicly available, and no ethical approval was required.

2.3. Statistical Analysis

2.3.1. Descriptive Analysis

Monthly NO₂ concentration dynamics by monitoring stations and regions were plotted to assess NO₂ levels for the type of monitoring stations and for the regions.

2.3.2. Time Series Prediction Model for NO₂ (SARIMAX Model)

The time series data were divided into training (April 2009–March 2016) and testing (April 2016–March 2017) data to allow for cross-validation. With the training data, time series models for monthly NO₂ concentrations were built using seasonal autoregressive integrated moving average (SARIMA) and SARIMA with exogeneous variable (SARIMAX) models [19]. These models were built to assess the relationships between NO₂ and the climate factors/outdoor air pollutants after removing the effects of autocorrelation and seasonality which are commonly contained in time series data. The logarithm of monthly average NO₂ concentration was used as a response variable. One of the climate factors/air pollutants was added as an exogeneous variable for each model. In total, 64 models were built for the four regions, two monitoring stations, and eight exogeneous variables (including non-exogeneous (SARIMA) models). A SARIMA model is an extension from an ARIMA model to which the seasonal terms are added. An ARIMA model consists of non-seasonal autoregressive and moving average terms and were transformed to stationary from non-stationary series by differencing. The relationships between NO₂ and exogeneous variables can be evaluated by adjusting the coefficient of the exogeneous variables given the autocorrelation terms are included in the models. For the non-seasonal terms, the number of previous time series used for autoregression and moving averaging, and the number of differencing to reach stationary series were described as p, d, and q, while those for seasonal terms were P, D, and Q, respectively.

Time series analysis was conducted using the Box-Jenkins approach [20]. The Box-Jenkins approach contains model stabilization, identification, model validation, forecasting and cross-validation. For model stabilization and identification, the Hyndman-Khandakar algorithm [21] for automatic ARIMA modelling was used, which is available in the “forecast” package in R as the “auto.arima” function, to select the best model based on the Akaike Information Criterion (AIC, the lowest AIC models were selected as the best models). The Ljung-Box tests were then used to test the randomness of the residuals for the model validation. If the model failed to pass the Ljung-Box test, differenced data were used to select the best fit model based on AIC [19] using the “auto.arima” function. Finally, using the testing data, a forecasting for the preceding 12 months was conducted. The model fitting and forecasting were evaluated using the root mean squared errors (RMSEs). All the analyses were performed using Statistical Software R version 4.0.0 [22].

3. Results

3.1. Descriptive Analysis (NO₂ Dynamics in April 2009–March 2017)

Among the four study areas, Tokyo had the highest NO₂ concentration both at AAPMS and RAPMS, and Yamanashi has the lowest (

Table 1. Mean (standard deviation) NO₂ concentration (ppb) by month between April 2009 and March 2017 at AAPMSs and RAPMSs in Tokyo, Aichi, Osaka, and Yamanashi Prefectures, Japan.

	AAPMS				RAPMS
Month	Tokyo	Aichi	Osaka	Yamanashi	Tokyo	Aichi	Osaka	Yamanashi
January	26.5 (2.0)	22.2 (3.5)	22.9 (2.9)	17.8 (1.6)	34.5 (3.1)	34.3 (2.2)	28.0 (3.5)	18.5 (3.0)
February	25.2 (4.1)	23.0 (3.6)	24.6 (8.6)	16.3 (3.0)	37.9 (6.2)	36.7 (4.2)	28.6 (4.7)	17.8 (3.2)
March	23.1 (2.4)	21.9 (2.9)	22.5 (3.0)	12.3 (2.7)	38.8 (3.7)	36.2 (3.7)	30.9 (3.5)	14.4 (2.5)
April	21.1 (2.7)	21.0 (2.3)	20.8 (2.7)	10.6 (1.6)	41.5 (5.8)	33.6 (3.0)	32.1 (3.5)	13.3 (3.0)
May	18.8 (2.2)	18.4 (2.3)	18.7 (1.7)	8.7 (1.6)	39.4 (5.4)	29.5 (3.5)	29.5 (3.7)	11.6 (3.3)
June	19.6 (3.6)	19.0 (2.5)	17.4 (2.9)	8.7 (2.2)	42.4 (5.0)	29.0 (2.8)	28.1 (2.9)	11.2 (3.1)
July	18.2 (1.1)	16.7 (2.0)	16.1 (2.1)	8.1 (1.5)	37.2 (3.4)	24.7 (2.3)	21.7 (1.8)	9.5 (2.0)
August	15.2 (2.3)	14.4 (2.0)	13.8 (2.2)	7.7 (1.4)	32.7 (6.3)	21.8 (3.7)	21.1 (2.6)	8.4 (2.0)
September	18.2 (3.6)	18.0 (2.8)	14.3 (1.6)	8.8 (1.7)	37.6 (6.5)	28.3 (4.0)	26.6 (2.9)	9.9 (1.7)
October	22.5 (4.9)	22.2 (3.5)	17.6 (3.8)	12.1 (2.4)	38.6 (4.8)	34.8 (4.3)	30.5 (4.2)	12.7 (2.1)
November	27.8 (3.0)	24.4 (2.1)	23.3 (3.3)	16.9 (2.5)	38.2 (5.0)	37.2 (2.8)	32.5 (3.4)	17.7 (2.7)
December	28.9 (1.8)	23.1 (2.2)	22.8 (2.5)	19.5 (1.9)	36.8 (2.7)	34.2 (1.7)	29.0 (2.9)	20.1 (3.0)

NO₂, nitrogen dioxide; AAPMSs, ambient air pollution monitoring stations; RAPMSs, roadside air pollution monitoring stations.

and Figure 1). The dynamics of NO₂ had annual seasonality showing the highest/lowest concentrations in summer (August)/winter (basically November–February), respectively; however, the seasonality at RAPMS was less clear than that at AAPMS (Table 1).

3.2. Climate and Air Pollution Factors Associated with Nitrogen Dioxide (NO₂) Concentration at Ambient Air Pollution Monitoring Station (AAPMS)

Overall, air temperature, wind speed, solar radiation, and O₃ at AAPMS had a trend of negative associations with NO₂, while NO had positive associations (

Table 2. Climate and air pollution factors associated with NO₂ concentration in the atmosphere from SARIMAX models at AAPMSs/RAPMSs in Tokyo, Aichi, Osaka, and Yamanashi Prefectures, Japan.

Model 1 (SARIMA/ SARIMAX)	AAPMS	RAPMS
	Coefficient 2 (95%CI)
Tokyo
SARIMA	NA	NA
Temperature	−0.022 (−0.029, −0.015)	−0.001 (−0.005, 0.003)
Humidity	0.004 (−0.001, 0.008)	0.002 (−0.001, 0.004)
Wind speed	−0.181 (−0.242, −0.120)	−0.297 (−0.344, −0.251)
Solar radiation	−0.021 (−0.033, −0.009)	−0.002 (−0.01, 0.005)
Sunshine duration	−0.018 (−0.037, 0.001)	−0.021 (−0.039, −0.003)
NO	0.024 (0.016, 0.031)	0.006 (0.005, 0.008)
O3/AO 3	−0.003 (−0.009, 0.002)	−0.003 (−0.012, 0.001)
Aichi
SARIMA	NA	NA
Temperature	−0.015 (−0.020, −0.009)	−0.016 (−0.022, −0.009)
Humidity	−0.003 (−0.010, 0.004)	−0.005 (−0.010, 0.000)
Wind speed	−0.270 (−0.343, −0.196)	−0.169 (−0.238, −0.100)
Solar radiation	−0.019 (−0.031, −0.007)	−0.002 (−0.014, 0.011)
Sunshine duration	−0.013 (−0.034, 0.008)	−0.002 (−0.020, 0.016)
NO	0.020 (0.013, 0.028)	0.009 (0.007, 0.012)
O3/AO 3	−0.001 (−0.008, 0.005)	0.001 (−0.005, 0.007)
Osaka
SARIMA	NA	NA
Temperature	−0.022 (−0.026, −0.018)	−0.011 (−0.013, −0.008)
Humidity	−0.010 (−0.019, −0.001)	−0.006 (−0.012, −0.001)
Wind speed	−0.306 (−0.413, −0.199)	−0.257 (−0.317, −0.198)
Solar radiation	0.002 (−0.025, 0.029)	0.01 (−0.006, 0.026)
Sunshine duration	0.061 (0.031, 0.091)	0.025 (0.003, 0.047)
NO	0.033 (0.027, 0.039)	0.015 (0.012, 0.018)
O3/AO 3	−0.002 (−0.009, 0.005)	−0.006 (−0.011, −0.001)
Yamanashi
SARIMA	NA	NA
Temperature	−0.012 (−0.034, 0.010)	−0.031 (−0.038, −0.024)
Humidity	0.010 (0.006, 0.013)	0.005 (0.001, 0.010)
Wind speed	−0.198 (−0.251, −0.144)	−0.120 (−0.208, −0.032)
Solar radiation	−0.016 (−0.029, −0.004)	−0.005 (−0.022, 0.012)
Sunshine duration	−0.025 (−0.043, −0.006)	0.012 (−0.008, 0.032)
NO	0.034 (0.021, 0.047)	0.038 (0.025, 0.050)
O3/AO 3	−0.007 (−0.012, −0.002)	−0.009 (−0.015, −0.005)

Notes: The first 7 years [April 2009–March 2016, 84-point (12 months times 7 years) time series data] were used as training data and the last 1 year [April 2016–March 2017, 12-point (12 months times 1 year) time series data] were as training data for each model. AAPMS, ambient air pollution monitoring stations; RAPMS, roadside air pollution monitoring stations; SARIMA, seasonal autoregressive integrated moving average; SARIMAX, seasonal autoregressive integrated moving average with exogeneous variables; CI, confidence interval; NO₂, nitrogen dioxide; NO, nitric oxide; O₃, ozone. ¹ This column shows the types of models. The exogeneous variables are shown for SARIMAX models. An exogeneous variable was included in each SARIMAX model. ² Coefficients in log scale for exogeneous variable. ³ O₃/AO were used for the analyses at AAPMSs/RAPMSs, respectively.

, the result of model validations is available in Table S2).

Specifically, air temperature had significant negative associations with NO₂ in all the prefectures except Yamanashi. Wind speed had negative associations with NO₂ in all the prefectures. Solar radiation had negative associations with NO₂ in all the prefectures except Osaka. O₃ was negatively and significantly associated with NO₂ only in Yamanashi. NO had positive associations with NO₂ in all the prefectures. The associations of relative humidity and sunshine duration with NO₂ did not have consistent trends by prefecture (Table 2). Humidity in Osaka was negatively associated with NO₂ while that in Yamanashi was positively associated. Sunshine duration had a trend of negative association with NO₂ except in Osaka; Yamanashi had a significant negative association, while Osaka had a positive association. RMSEs for forecasting were close to those for model fitting (Table S2), and most of the testing data were within the 95% confidence interval (CI) of the forecasting (see Figure 2 as examples).

3.3. Climate and Air Pollution Factors Associated with Nitrogen Dioxide (NO₂) Concentration at Roadside Air Pollution Monitoring Station (RAPMS)

Air temperature, wind speed, NO, and AO (the alternative of O₃) at RAPMS had similar trends as those at AAPMS, and the first two were negatively associated with NO₂, while NO was positively associated (Table 2, the result of model validations is available in Table S3). Specifically, air temperature had significant negative associations with NO₂ in all the prefectures except Tokyo. Wind speed had negative associations with NO₂ in all the prefectures. NO had positive associations with NO₂ in all the prefectures. AO had significant negative associations with NO₂ in Osaka and Yamanashi. The associations of relative humidity with NO₂ did not have a consistent trend by prefecture (Table 2) as observed at AAPMS. The trend of relative humidity at RAPMS was the same as that at AAPMS: in Osaka it was negatively associated with NO₂ while in Yamanashi it was positively associated. Sunshine duration had a trend of positive association with NO₂ except in Osaka; significant associations were observed in Tokyo and Osaka at RAPMS. No significant relationship between solar radiation and NO₂ was observed at RAPMS.

RMSEs for forecasting were close to those for model fitting (Table S3), and most of the testing data were within the 95% CI of the forecasting (see Figure 3 as examples, showing SARIMA model fittings and forecastings).

4. Discussion

NO₂ air pollutant concentration shows seasonality, that is, low in summer and high in winter. The seasonality of NO₂ at RAPMS was not clear compared to that at AAPMS, especially in urban areas (Tokyo, Aichi, and Osaka, i.e., NO₂ levels at AAPMSs had regular annual patterns as shown in Figure 1 while those at RAPMSs in urban areas had irregular shape). Although the reasons for this are not clear, it might be affected by the emission of air pollutants from road traffic. According to the Ministry of Land, Infrastructure, Transport and Tourism, Japan [23], the average traffic volume of measuring points within 2 square kilometres of the centre of the RAPMSs in Tokyo, Aichi, and Osaka were 53,892, 51,894, and 43,004 cars per day in 2015, respectively, and the volumes were higher than in Yamanashi at 13,910 cars per day. NO₂ was high in the urban areas and low in the rural area. The traffic volume could be one of the factors that affects NO₂ concentrations in the atmosphere.

Air temperature, wind speed, and solar radiation at AAPMSs had a trend of negative association with NO₂, while NO had a positive association (Table 2). The effect of solar radiation and wind speed on HONO and NO₂ are consistent. HONO is decomposed by solar light [7], and the effect of decomposition by sunlight for NO₂ is much weaker [24]. Wind decreases the concentrations of NO₂ and HONO in the atmosphere [1,7]. When considering the relationships between climate factors and NO₂ concentration in the atmosphere, the equilibrium relationship (Equation (1)) can also be considered. Given a high air temperature, the reaction proceeds toward the right side (i.e., producing NO₂), while it goes to the left side (producing HONO) given high humidity [7]. Although the proportion of HONO out of NO₂ is a few percent, the change in HONO concentration must be larger than NO₂ from the equilibrium Equation (1) (i.e., one molecule of NO₂ produces two HONO molecules). The negative relationship between air temperature and the NO₂ concentration measured at AAPMS from this study suggests that the identified associations showed the same trends for HONO with climate factors, and that a certain proportion of HONO was in measured NO₂.

The associations of relative humidity and sunshine duration with NO₂ did not have consistent trends by prefecture. In the case of relative humidity, the relationship observed in Yamanashi seemed to be an effect of humidity on HONO in terms of the equilibrium relationship between HONO and NO₂ (Equation (1)), while that in Osaka was an effect on NO₂. Urban areas are more polluted by NO/NO₂ compared to rural areas due to higher amounts of combustion, and the level of HONO does not reach an equilibrium concentration. Because of this, it is suggested that the proportion of HONO in measured NO₂ was low/high in urban/rural areas, respectively, and the change in HONO concentration could be more sensitive in rural areas. This idea is consistent with a report saying the HONO/NO_x ratio at roadsides was low [25], while that indoors, which may correlate with the ambient HONO/NO_x concentrations, was high [6]. As described in (Equation (1)), HONO is produced from NO and NO₂ discharged from, for example, cars. This reaction takes time to reach the equilibrium status [26], and it indicates that rural areas are polluted by HONO not locally but diffusely.

The trends of association of climate factors/air pollutants with NO₂ at RAPMS were almost consistent with those at AAPMS although a significant relationship between solar radiation and NO₂ was not observed at RAPMS. The lower effects of solar radiation on NO₂ measured at RAPMSs suggested a higher proportion of NO and NO₂ concentration due to their higher emission, thus, a lower proportion of HONO could be considered at RAPMS than at AAPMS (i.e., the level of HONO had not reached the equilibrium concentration at RAPMSs) because the effect of decomposition by sunlight for NO₂ is much weaker than that for HONO [24]

Correlations between NO₂ and O₃ previously reported vary depending on the temporal averaging periods (i.e., from within-hourly to annual or longer-term correlations) and also depending on the studies [1]. One study using monthly averaging period, which is the same averaging period as the current study, showed similar results to this study (i.e., negative correlation between NO₂ and O₃). O₃ is consumed by reactions with NO producing NO₂ ($\mathrm{NO}+{\mathrm{O}}_3\to {\mathrm{NO}}_2+{\mathrm{O}}_2$), and its concentration can be negligibly low due to high consumption where NO concentration is high. The negative relationship between O₃ and NO₂ from this study can be explained by this reaction and the extent of O₃ concentrations, which was relatively low. The relationships between NO₂ and O₃ concentrations may have seasonal patterns, although the evidence is limited [1]; however, the present study did not deal with the effect modification of seasons, and further studies are required to assess the interaction effect between O₃ and seasons on NO₂ concentrations.

SARIMAX models can estimate coefficient of exogeneous variables by adjusting autocorrelation and seasonality effects. However, in this study only one exogeneous variable was used for each SARIMA model for the estimation. It should be noted that other confounders were not adjusted because underlying causal relationships between factors were not well established, and data used were limited to climate factors and air pollutants.

5. Conclusions

The relationships between climate factors and NO₂ concentrations measured in the atmosphere, the differences between the types of monitoring stations, and differences between urban and rural areas were assessed using SARIMA and SARIMAX models in this study. Overall, wind speed was negatively and NO was positively associated with NO₂ concentration. Air temperature had a negative association with NO₂, which might indicate the decomposition of HONO (that was in measured NO₂) in the equilibrium (Equation (1)). Humidity in Osaka was negatively associated with NO₂ (the reaction of NO₂) while that in Yamanashi was positively associated (the reaction of HONO), indicating the proportion of HONO in NO₂ might be low/high in urban/rural areas, respectively. The associations between solar radiation and NO₂ concentrations were only observed at the AAPMSs where the proportion of HONO in NO₂ measured might have been higher. The findings from this study may suggest that the equilibrium reaction of HONO had a large impact against the ambient NO₂ concentrations. To test the hypotheses that arose from this study (i.e., the level of HONO in NO₂ in different settings), monitoring of HONO concentration and further analyses of the relationship between HONO and climate/pollutant factors are required.