key: cord-293139-uj1m3t79
authors: Hua, Jinxi; Zhang, Yuanxun; de Foy, Benjamin; Mei, Xiaodong; Shang, Jing; Feng, Chuan
title: Competing PM2.5 and NO2 holiday effects in the Beijing area vary locally due to differences in residential coal burning and traffic patterns
date: 2020-08-11
journal: Sci Total Environ
DOI: 10.1016/j.scitotenv.2020.141575
sha: 
doc_id: 293139
cord_uid: uj1m3t79

Abstract The holiday effect is a useful tool to estimate the impact on air pollution due to changes in human activities. In this study, we assessed the variations in concentrations of fine particulate matter (PM2.5) and nitrogen dioxide (NO2) during the holidays in the heating season from 2014 to 2018 based on daily surface air quality monitoring measurements in Beijing. A Generalized Additive Model (GAM) is used to analyze pollutant concentrations for 34 sites by comprehensively accounting for annual, monthly, and weekly cycles as well as the nonlinear impacts of meteorological factors. A Saturday effect was found in the downtown area, with about 4% decrease in PM2.5 and 3% decrease in NO2 relative to weekdays. On Sundays, the PM2.5 concentrations increased by about 5% whereas there were no clear changes for NO2. In contrast to the small effect of the weekend, there was a strong holiday effect throughout the region with average increases of about 22% in PM2.5 and average reductions of about 11% in NO2 concentrations. There was a clear geographical pattern in the strength of the holiday effect. In rural areas the increase in PM2.5 is related to the proportion of coal and biomass consumption for household heating. In the suburban areas between the Fifth Ring Road and Sixth Ring Road there were larger reductions in NO2 than downtown which might be due to decreased traffic as many people return to their hometown for the holidays. This study provides insights into the pattern of changes in air pollution due to human activities. By quantifying the changes, it also provides insights for improvements in air quality due to control policies implemented in Beijing during the heating season.

particles (PM 2.5 ) and nitrogen dioxide (NO 2 ), there have been conflicting results. (Cui et al., 2020) found concentrations of thirteen elements in PM 2.5 were higher on weekends than on weekdays. (Beirle et al., 2003; Chen et al., 2015; Liu et al., 2016) did not find a clear weekly pattern in the Beijing area. These studies did not consider the effect of meteorology on pollutant concentrations.

The regional transport plays an important role in the air pollution of Beijing: a clear spatial pattern was observed with decreasing concentrations from south to north (Li et al., 2015) . Because meteorology has a strong impact on air pollution in the region, it must be taken into account to prevent biases in the estimates of human-related air pollution changes (Sun et al., 2013; Sun et al., 2015; Wang et al., 2017a; .

In addition to meteorology, local emissions also affect air quality. The main anthropogenic sources of air pollution in Beijing during the heating season are coal-burning for heating and vehicle emissions due to rapid urbanization Xu et al., 2019) . A series of measures were implemented to improve air quality in the city, such as "Action Plan on Prevention and Control of Air Pollution" (Zhang et al., 2016) , "Joint Prevention and Control of Atmospheric Pollution" , "Coal to Electricity" and "Coal to Gas" projects (Shuxue et al., 2020) .

The actions have been found to effectively reduce the PM 2.5 and NO 2 concentrations (Barrington-Leigh et al., 2019; Cheng et al., 2019; de Foy et al., 2016b; Wang et al., 2017b; Zhang The holiday effect is similar to the weekend effect in that it is a measure of the difference in pollution levels between holidays and non-holidays (Tan et al., 2013) . In general, it shows higher concentrations during non-holidays and lower concentrations during holidays. The holiday effect is due to changes in human activities which are influenced by lifestyle, urbanization, energy consumption structure, and traditional culture (Forster and Solomon, 2003; Liu et al., 2019; Seidel and Birnbaum, 2015) . When there is no clear weekend effect, the holiday effect provides a useful tool to identify changes in air quality due to changes in human activities (Chen et al., 2019) .

(Madhavi Latha and Highwood, 2006) found coarse particulate matter (PM 10 ) concentrations are lower during the Christmas holidays than non-holidays in the United Kingdom, which are mainly due to reductions of local traffic emissions. (Chen et al., 2019; Tan et al., 2009) investigated the difference of six air pollutants in Taipei between the Spring Festival and non-Spring Festival, finding significant reductions during the Spring Festival for nitrogen oxides (NO x ), carbon monoxide (CO), volatile organic compounds or non-methane hydrocarbon (NMHC), sulfur dioxide (SO 2 ), and PM 10 , while O 3 concentrations increased due to a reduction in the titration effect. (Tan et al., 2013) reported a distinct spatial holiday effect associated with the degree of urbanization in Taipei and found that the holiday effects of NO x , CO and NMHC become greater when the population, and hence the number of motor vehicles increases. was available from the Beijing Municipal Environmental Monitoring Center (http://www. bjmemc.com.cn/). These sites are designed to reflect urban background conditions, regional transmission, traffic pollution, and urban air quality, covering most of the spatial range and multiple land use types (Ji et al., 2019; Sun et al., 2019) . For each site, daily averages were calculated from the valid data points when data was available for more than 10 hours per day. The location of the sites and the number of observations as shown in Table S1 . PM 2.5 and NO 2 were available for more than 95% of the days for most sites during the study period (756 days in total).

Beijing includes 16 administrative regions ( Fig. 1) , with a total population of 21.54 million in 2018 (Beijing Municipal Bureau of Statistics) and an area of 16,410km 2 (Dong et al., 2018) . Rapid urbanization caused a drastic expansion from core districts (Dongcheng and Xicheng). The area within the Fifth Ring Road is considered the most densely populated area with around half of the inhabitants and only 4% of the surface area of Beijing (Xin , Dong, et al., 2018 . Due to relatively short commuting time and cheap housing, the area between the Fifth Ring Road and the Sixth Ring Road accounts for 27% of the population of Beijing City (Xin Zhang, 2019).

Outside the Sixth Ring Road, the population density is relatively sparse, and rural residents account for a large proportion, with approximately 86% of land area only 24% of the population (Xin Zhang, 2019 , Dong, et al., 2018 .

Different population distribution and development patterns will lead to variations in air pollution due to variations in human activities. To explore the spatial variation of the holiday effect, the sites were clustered by geographical locations based on different population and development patterns (Fig. 1) . The sites within the Fifth Ring Road were defined as downtown, the sites between the Fifth Ring Road and the Sixth Ring Road were defined as suburban, the sites outside the Sixth J o u r n a l P r e -p r o o f Journal Pre-proof

The holiday periods referred to New Year"s Day (January 1), the Spring Festival (Lunar New

Year), and the Lantern Festival based on the national legal holiday arrangements issued by the State Council. The Spring Festival is the most important holiday involving family reunions, and it will bring travel peaks of people returning to their hometowns before the festival and then back to work afterwards. For the purposes of this study, the Festival holiday period is defined as starting two days before the eve of the Spring Festival and finishing four days after the Spring Festival.

This excludes the influence of the travel peaks which occur before and after that .

The specific dates are listed in Table S2 . Overall, the holiday periods last around ten days each

year.

In this study, the non-parametric Mann-Whitney U test, which does not require the data to be normally distributed (Chen et al., 2019) , was used to compare the pollutant concentrations and meteorological observations during the holidays and non-holidays, weekends and weekdays.

The predictors in our GAM model include time vectors to represent inter-annual, monthly, and weekday variations, as well as meteorological variables (boundary layer height, east-west wind component, south-north wind component, relative humidity, air temperature, dew point temperature, and surface pressure). The time factors were defined as linear terms, and the meteorological variables were defined as smooth terms. The equation is as follows: S (·) is the P-spline smoothing function that optimizes the fitting and controls the smoothness through a penalty term. S 1 (BLH) are the smoothers that characterize the non-linear influence of boundary layer height on the measurements. S 2 (U, V) are interaction terms that denote the influence of horizontal ventilation due to zonal and meridional wind speeds. " " represent the meteorological variables selected in the candidate list based on their contribution to the model R 2 . The candidate list included relative humidity, air temperature, dew point temperature, and pressure. To avoid the collinearity problem, the test process input only one variable to GAM at a time. The one leading to the greatest increase in correlation coefficient was included. All meteorological variables were scaled linearly in order to approximate the normal distribution of zero mean to reduce the effects of extreme observations.

After considering the impact of meteorological parameters, a net contribution to the time cycle can be obtained from the regression coefficient. In this way, temporal profiles are closer to changes caused by emissions than meteorological factors. The temporal indicators are input for each year, each month, and each day of the week, and therefore do not have a unique solution. As described J o u r n a l P r e -p r o o f Journal Pre-proof in (de Foy, 2018), a weighting factor of one was used on the penalty term for the regression coefficients which solves the dummy variable trap problem while also forcing  to have the smallest possible values.

The GAM results can be interpreted using Equation (2):

Where p corresponds to the percentage change in the concentration during the time intervals relative to long-term averages.

Block-bootstrapping (de Foy and Schauer, 2015; Requia et al., 2019) with seven-day chunks was used to estimate the uncertainty of the linear terms. The model was obtained 100 times using a randomly resampled dataset each time. When resampling the dataset, the data points to be included were selected at random with replacement so that each dataset was of the same size as the original. The standard deviation of the temporal coefficients was obtained from the 100 model simulations. The 95% confidence interval of the nonlinear terms in Equation (1) and 20% lower NO 2 concentrations ( Table 1) . The mean (± SD) PM 2.5 concentrations are 96±82μg/m 3 and 81±80μg/m 3 in holidays and non-holidays, respectively. The mean (± SD) NO 2 concentrations are 44±29μg/m 3 and 55±31μg/m 3 during holidays and non-holidays, respectively.

Daily means of PM 2.5 and NO 2 between the holidays and non-holidays, weekends and weekdays periods were found to be statistically different with p<0.01 using the Mann-Whitney U test (Table   1) . Meteorological variables were also statistically different, which means that it was important to consider these factors in the model in order to properly quantify the human-related air pollution changes. During the holidays, the mean BLH is 47m lower than non-holidays and the air temperature is around 3℃ lower. The main reason for the difference is that the holiday periods fall mainly within January and February which are colder than March. The same weather parameters from ERA5 and ISD show a consistent bias: BLH, D2M, T2M, RH, SP, and U are lower during the holidays and V is higher.

The average standard deviations of PM 2.5 scaling factors are around 4.5% and 9.9% in non-holidays and holidays, respectively, and they are stable from site to site ( Table S3 ). The relatively high uncertainty of holiday effects is due to the fact that there are fewer data points for the holiday periods than for the non-holidays. For NO 2 , the uncertainty of non-holidays is around 1.5% and for holidays it is around 2.6%. The uncertainty for PM 2.5 is higher than NO 2 mainly because PM 2.5 concentrations are affected by complex factors such as various emission sources, air mass transport, chemical transformation, day-to-day carry over, and complex interaction effect J o u r n a l P r e -p r o o f between synoptic conditions and PM 2.5 (Wang, et al., 2017 , Khuzestani, et al., 2018 .

The performance of the model fit was assessed by calculating the regression coefficient (r 2 ) and

the Root Mean Square Error (RMSE) using the 100 bootstrap runs ( Fig. S1 and Fig. S2) .

Overall, the GAM estimates of the holiday effects were found to be robust for each site with respect to the selection of the optimal set of meteorological variables.

The list of variables selected as input to the GAM for each site is shown in Table S1 . Daily average boundary layer height and ISD winds were included in most cases, and the average relative humidity was the optimal choice at the majority of sites during the sensitivity tests. We will take the site YZ as an example to discuss the impact of meteorology on pollutant concentrations. The site is located in the plain area and near a subway station, which means that its topographic features are similar to most of the other sites and it is strongly impacted by human activities. The increase in the mixing layer height led to the strong diffusion of pollutants (Miao et al., 2019) , leading to approximately a 65% decrease in PM 2.5 with one standard deviation increase in the boundary layer height at YZ (Fig. S3) . High relative humidity promotes the hygroscopic growth of particulate matter (Cheng et al., 2015) , with about a 65% increase in PM 2.5 with one standard deviation increase in the relative humidity. Beijing is surrounded by mountains to the west, north, and northeast; and the southeast is a plain that slopes slowly towards the Bohai Sea.

Because of the topographical features, the air quality at most sites is influenced by southerly winds.

Frequent northerly winds during the heating periods also transport air pollution from downtown to J o u r n a l P r e -p r o o f Journal Pre-proof YZ site which is a suburban site to the southeast. For NO 2 , an increase in one standard deviation of the boundary layer height is associated with a 22% decrease at YZ (Fig. S4) . Higher relative humidity is associated with higher NO 2 concentrations. Except for the influence from southerly winds, local emissions were found to have high contributions to NO 2 levels because the YZ site is very busy as it is near a subway station.

The 95% confidence interval range is very narrow for the meteorological parameters ( Fig. S3 and 

For the day of the week profiles, the PM 2.5 scaling factors vary from site to site, but on average there is not much variation from Monday to Sunday (Fig. 3, Table 2 Figure S8 ) suggesting that the weekend effects are robust with respect to spatial variability. The weekend effect in PM 2.5 can be clearly seen even though PM 2.5 is influenced by carry over on the scale of 3 to 5 days. While this shows that the weekend effect is strong enough to be identified over other factors, future analysis would be required to estimate the impacts of carry over, for example using aerosol chemical speciation. The probability distribution of percentage changes by type of day shows that the differences between the days are statistically significant. As an example, the distributions of PM 2.5 factors at WSXG are clearly different for Saturdays, Sundays and Holidays, and the distributions of NO 2 factors are clearly different for Saturdays and Holidays ( Figure S9 ). This result updates the previous research on the weekend effect in Beijing. (Beirle et al., 2003; Chen et al., 2015; Liu et al., 2016) which showed that there was no obvious weekend effect for PM 2.5 or NO 2 in Beijing, mainly because the previous study mostly did not consider the impact of meteorological parameters and of spatial patterns. This implies that the GAM analysis provides valuable information with respect to temporal variation and meteorological influence, and that a network of widely distributed monitoring stations can provide more subtle information on the weekend effect. suggest that there are multiple factors involved in the change of human-activities, as will be discussed in Section 3.4.

From downtown to suburban, to rural areas, there is a clear change in the holiday effects in the direction of a larger increase of PM 2.5 and a smaller decrease of NO 2 (Fig. 4) . For the most part, sites in close geographical proximity behave similarly, although there are a few outliers. DL, DGC, and TZ exhibit different characteristics from the neighboring sites, probably because they are affected by local land use effects. For example, DGC is geographically close to PG, but DGC is surrounded by farmland and forest while PG is located in a residential area (Fig. S10) , which suggests that the different holiday effects are due to local variations in land use type. DL was purposely designed to monitor urban background air quality surrounded by mountainous areas (Chen et al., 2015; Li et al., 2015) . These sites are excluded in the analysis that follows.

For downtown, the changes are relatively consistent, with an increase in PM 2.5 concentrations ranging from 20.4% to 27.1% and a reduction in NO 2 ranging from 8.6% to 12.5% ( Table 2) . For most sites in the suburban region, the PM 2.5 changes are close to downtown sites with 19.5~25.1%

increases, and the changes in NO 2 are stronger than downtown with 12.0~15.7% decreases.

The effects of holidays in rural areas are clearly distinct. For the PM 2.5 holiday effect, the southwest and northeast rural areas demonstrate the largest increases, being on average 30.6%

higher than weekdays. The sites in the southeast and northwest areas show the lowest increases J o u r n a l P r e -p r o o f Journal Pre-proof with 12.2% changes on average. This pattern is similar to the feature of PM 2.5 emissions from household heating in (Cai et al., 2018) , who developed a village-based emission inventory of household combustion based on the investigation of all villages in Beijing. For NO 2 , the sites in the southwest and northeast rural areas have the lowest decrease (7.3% on average) while the sites in the southeast and northwest areas have the highest decrease (14.6% on average).

The PM 2.5 holiday effect consisted of increased concentrations ranging from 2% to 39% depending on the site. In contrast, the NO 2 holiday effect consisted of decreased concentrations ranging from 3% to 18% depending on the site. That PM 2.5 has a stronger holiday effect than NO 2 implies that there are greater differences in human activities for PM 2.5 emission sources.

Rural areas to the northeast of Beijing experienced larger increases in PM 2.5 concentrations. These are mainly concentrated in Miyun, Huairou, and Pinggu, which are districts in northeast Beijing and are more heavily forested, and where coal and biomass burning is frequently used as a heating fuel (Cai et al., 2018) . found that coal-fired boilers have been mostly eliminated from urban areas but remain in rural areas and especially in the districts of Miyun, Huairou, and PingGu in 2014.

We compared the strength of the holiday effect with household energy consumption by town in most districts of Beijing in 2017 provided by (Cai et al., 2018) . For each rural measurement site, we calculated the average coal and biomass consumption in neighboring districts. The strength of the holiday effect in PM 2.5 in rural areas was found to have a positive correlation with coal and biomass consumption (Fig. 5) . The northeast and southwest rural areas had a higher proportion of coal consumption for household heating than the northwest and southeast rural areas, and had a J o u r n a l P r e -p r o o f Journal Pre-proof correspondingly larger increase in PM 2.5 concentrations during the holidays. The spatial variation of the holiday effect suggests that indoor household heating activities could be a possible cause for the increases in PM 2.5 in rural areas. The results suggest that the promotion of the "coal to gas" project ( Barrington-Leigh et al., 2019; Zhao et al., 2020b) will play an important role in improving air quality in Beijing in the next few years.

Household heating is also a significant source of NO 2 emissions (Luo et al., 2019) , such that the NO 2 emitted by coal combustion probably offset the reduction of NO 2 due to reduced traffic. The area with more coal-fired activities have a smaller reduction in NO 2 than the areas with lower coal-fired activities. Therefore, the net NO 2 holiday effect in rural areas is a result of two competing effects. For the area within the Sixth Ring Road, although suburban areas and downtown show similar PM 2.5 increases, the suburban area exhibited an extra 2.7% reduction in NO 2 relative to the downtown area. The difference is probably because the suburban area is home to a large number of commuters thanks to relatively cheap housing costs and short commute times, and consequently has a greater fraction of people traveling during the holidays (Zhao et al., 2020a) .

Changes in travel patterns are particularly strong during the Spring Festival when a large number of people return to their hometown leading to reduced transportation emissions during the Spring Festival itself. To distinguish between the effect of different holidays it will be necessary to have a longer time series of measurements.

In this study, we used GAM to estimate the holiday effects of PM 2.5 and NO 2 in Beijing by concentrations increased by about 5% whereas there were no clear changes for NO 2 . Although there is uncertainty due to meteorological factors such as temperature, wind speed and direction, and boundary layer height, the weekend variations are consistent at all downtown sites and robust with respect to temporal variability. The holiday effect was found to be much stronger than the weekend effects, and had opposite signs for PM 2.5 and NO 2 . There were increases in PM 2.5 ranging from 2% to 30% depending on the site. In contrast, NO 2 decreased from 3% to 18% depending on the site.

Furthermore, a clear spatial pattern was found in the strength of the holiday effect. In the rural areas, the strength of the PM 2.5 increases were associated with the extent of coal and biomass consumption for household heating in districts surrounding the measurement sites. It is worth noting that the promotion of renewable energy can therefore be expected to improve air quality in rural hotspots as well as in the greater Beijing area. In addition, the suburban area where more people travelled during the holidays experienced greater reductions in NO 2 than the downtown area. The spatial variation in the holiday effect at different sites reflects two distinct ways that human activities impact air quality: increased residential heating tended to increase both PM 2.5 and NO 2 , whereas reduced traffic emissions leads to lower NO 2 .

This study investigated the holiday effect in Beijing, providing evidence for the influence of human activities on air quality on short time scales. Studies of the holiday effects as well as other natural experiments such as the impacts of the Olympic games (Liu et al., 2012) , APEC Blue (Gao et al., 2017; Sun et al., 2016) , and new studies emerging on the impact of the COVID-19

lockdown (Bauwens et al., 2020; Chauhan and Singh, 2020; Wang and Su, 2020) will provide valuable information for the formulation of policies for both holiday and non-holiday periods. 

Pollutant emissions from residential combustion and reduction strategies estimated via a village-based emission inventory in Beijing

Decline in PM2.5 concentrations over major cities around the world associated with COVID-19

Impacts of holiday characteristics and number of vacation days on "holiday effect

Diurnal, weekly and monthly spatial variations of air pollutants and air quality of Beijing

Effects of meteorology and emission reduction measures on air pollution in Beijing during heating seasons

Humidity plays an important role in the PM2.5 pollution in Beijing

In situ continuous observation of hourly elements in PM2.5 in urban beijing, China: Occurrence levels, temporal variation, potential source regions and health risks

City-level variations in NOx emissions derived from hourly monitoring data in Chicago

Changes in ozone photochemical regime in Fresno, California from 1994 to 2018 deduced from changes in the weekend effect

Impacts of control strategies, the Great Recession and weekday variations on NO 2 columns above North American cities

Satellite NO2 retrievals suggest China has exceeded its NOx reduction goals from the twelfth Five-Year Plan

Origin of high particle number concentrations reaching the St. Louis, Midwest Supersite

Changes in speciated PM2.5 California, due to NOx concentrations in Fresno, reductions and variations in diurnal emission profiles by day of week

Observations of a

Distinguishing the roles of meteorology, emission control measures, regional transport, and co-benefits of reduced aerosol feedbacks in

Characteristics and sources of aerosol pollution at a polluted rural site southwest in Beijing

Spatial distribution differences in PM2.5 concentration between heating and non-heating seasons in Beijing

Spatial-Temporal Analysis on Spring Festival Travel Rush in China Based on Multisource Big Data

Diurnal, seasonal, and spatial variation of PM 2.5 in Beijing

Temporal Patterns in Fine Particulate Matter Time Series in Beijing: A Calendar View

Source Apportionment of Fine-Particle, Water-Soluble Organic Nitrogen and Its Association with the Inflammatory Potential of

A statistical model to evaluate the effectiveness of PM2.5 emissions control during the Beijing 2008 Olympic Games

Source apportionment of black carbon during winter in Beijing

Origins of aerosol nitrate in Beijing during late winter through spring

Studies on particulate matter (PM10) and its precursors over urban environment of Reading

Modeling the Effect of Weekday-Weekend Differences in Motor Vehicle Emissions on Photochemical Air Pollution in Central California

Interaction Between Planetary Boundary Layer and PM2.5 Pollution in Megacities in China: a Review

Day-of-week patterns of particulate matter and its chemical components at selected sites in California

in the United States: A trend analysis over the last 30years

Effects of Independence Day fireworks on atmospheric concentrations of fine particulate matter in the United States

Coal-to-electricity" project is ongoing in north China

Analysis of PM2.5 pollution episodes in Beijing from 2014 to 2017: Classification, interannual variations and associations with meteorological features

The impact of relative humidity on aerosol composition and evolution processes during wintertime in Beijing

APEC Blue": secondary aerosol reductions from emission controls in Beijing

Long-term real-time measurements of aerosol particle composition in Beijing, China: seasonal variations, meteorological effects, and source analysis

Impact of urbanization on the air pollution "holiday effect" in Taiwan

Air pollution "holiday effect" resulting from the Chinese New Year

A joint prevention and control mechanism for air pollution in the Beijing-Tianjin-Hebei region in china based on long-term and massive data mining of pollutant concentration

A preliminary assessment of the impact of COVID-19 on environment -A case study of China

Impacts of regional transport on black carbon in Huairou

Effectiveness of temporary control measures for lowering PM 2.5 pollution in Beijing and the implications

Relative impact of emissions controls and meteorology on air pollution mitigation associated with the Asia-Pacific Economic

Results of investigation on population distribution of Beijing's various ring roads. Favorite Big Beijing

Nitrate dominates the chemical composition of PM2.5 during haze event in Beijing

Trends of multiple air pollutants emissions from residential coal combustion in Beijing and its implication on improving air quality for control measures

A Study on the model for heating influence on PM2.5 Emission in Beijing China

Air pollution and control action in Beijing

Mass human migration and Beijing"s urban heat island during the Chinese New Year holiday

Historic and future trends of vehicle emissions in Beijing, 1998-2020: A policy assessment for the most stringent vehicle emission control program in China

Long-term trends in NO2 columns related to economic developments and air quality policies from 1997 to 2016 in China

Long commutes and transport inequity in China"s growing megacity: New evidence from Beijing using mobile phone data

Effect of the "coal to gas" project on atmospheric NOX during the heating period at a suburban site between Beijing and Tianjin

Pollution levels, composition characteristics and sources of atmospheric PM2.5 in a rural area of the North China Plain during winter

Human activities and urban air pollution in Chinese mega city: An insight of ozone weekend effect in Beijing

Village energy survey reveals missing rural raw coal in northern China: Significance in science and policy

This work was supported by the National Natural Science Foundation of China (NSFC, No. 

J o u r n a l P r e -p r o o f