key: cord-0997817-54lr29is authors: Wu, Qingru; Tang, Yi; Wang, Long; Wang, Shuxiao; Han, Deming; Ouyang, Daiwei; Jiang, Yueqi; Xu, Peng; Xue, Zhigang; Hu, Jingnan title: Impact of emission reductions and meteorology changes on atmospheric mercury concentrations during the COVID-19 lockdown date: 2020-09-11 journal: Sci Total Environ DOI: 10.1016/j.scitotenv.2020.142323 sha: f04deea0e067a4b59828f51b1fc8f2cc8a3607f7 doc_id: 997817 cord_uid: 54lr29is Controlling anthropogenic mercury emissions is an ongoing effort and the effect of atmospheric mercury mitigation is expected to be impacted by accelerating climate change. The lockdown measures to restrict the spread of Coronavirus Disease 2019 (COVID-19) and the following unfavorable meteorology in Beijing provided a natural experiment to examine how air mercury responds to strict control measures when the climate becomes humid and warm. Based on a high-time resolution emission inventory and generalized additive model, we found that air mercury concentration responded almost linearly to the changes in mercury emissions when excluding the impact of other factors. Existing pollution control and additional lockdown measures reduced mercury emissions by 16.7 and 12.5 kg/d during lockdown, respectively, which correspondingly reduced the concentrations of atmospheric mercury by 0.10 and 0.07 ng/m3. Emission reductions from cement clinker production contributed to the largest decrease in atmospheric mercury, implying potential mitigation effects in this sector since it is currently the number one emitter in China. However, changes in meteorology raised atmospheric mercury by 0.41 ng/m3. The increases in relative humidity (9.5%) and temperature (1.2 °C) significantly offset the effect of emission reduction by 0.17 and 0.09 ng/m3, respectively, which highlights the challenge of air mercury control in humid and warm weather and the significance of understanding mercury behavior in the atmosphere and at atmospheric interfaces, especially the impact from relative humidity. Mercury (Hg) pollution has caused global concern due to its long-range transmission, bio-toxicity, and neurotoxicity Gianga and Selina, 2016; Horowitz et al., 2017; Liu et al., 2019a) . To protect human beings and the environment from Hg emissions and releases, the Minamata Convention on Mercury (abbreviated as Convention) was signed in 2013 and has been taken into force since 2017 (UNEP, 2013) . Parties are required to reduce anthropogenic emissions and evaluate the effectiveness of emission control. Almost simultaneously, China took aggressive air pollution control measures during 2013-2017 to mitigate domestic haze pollution by launching the Air Pollution and Prevention Action Plan (SC, 2013) . A series of measures have been adopted, such as the ultra-low emission retrofit in coal-fired power plants (CFPP), eliminating backward production capacity, enhancing traditional air pollution control measures in industrial sources, replacing residential coal with electricity and natural gas, and off-peak production in winter. These strict policies effectively led to the reduction of atmospheric Hg emissions as a co-benefit Tian et al., 2015; Wen et al., 2020; Wu et al., 2018a; Wu et al., 2017; Wu et al., 2018b) . Atmospheric Hg emissions were reduced by approximately 22% from 2013-2017 (Liu et al., 2019b) . Correspondingly, the decrease of annual atmospheric Hg concentrations (TM) were observed at both urban sites and rural sites during 2013-2017 (Tang et al., 2018; Yin et al., 2018) . Afterwards, China released the Three-year Action Plan for Clean Air to further the effectiveness of air pollution control (SC, 2018) . In addition, to control the spread of the 2019 Novel Coronavirus , China issued a series of lockdown measures such as conditional industrial resumption, which indirectly led to restrictions in certain industrial activities. Such an emission scenario with strict pollution control and weakened industry demand is expected to occur in the coming ten years when China enters the J o u r n a l P r e -p r o o f Journal Pre-proof post-industrial era (Wu et al., 2018a) . Therefore, we are interested to learn to what extent atmospheric Hg emissions can be reduced under the joint measures of existing pollution control and the additional COVID-19 lockdown during this special period. Coincidentally, meteorology data has indicated that the relative humidity in East China during the lockdown period was the highest across the past ten years, and the average temperature increased by approximately 1.6℃ compared to the same period in 2019. The appearance of this kind of humid and warm weather is a silhouette of future climate change (IPCC, 2014) . Thus, the tragic outbreak of COVID-19 provides a natural experiment to find how atmospheric Hg (TM) responds to anthropogenic Hg emission reductions when the climate becomes humid and warm. In this study, we firstly compiled an Hg emission inventory with an hourly resolution to find the emission reductions due to anthropogenic control measures. Then, we established the relationship between TM concentrations and emissions, as well as meteorological fields, by applying a generalized additive model. Finally, by using the scenario analysis method, we predicted the response of TM to the change in anthropogenic measures and various meteorological parameters. By conducting this study, we aim to elucidate how policy and science can be prepared to better control TM pollution and implement the Convention. We applied our China Atmospheric Mercury Emission (CAME) model to compile the anthropogenic emission inventory (Hui et al., 2017; Liu et al., 2018; Liu et al., 2019b; Wu et al., 2016; Wu et al., 2018b; Zhang et al., 2015) . This model covers almost all emission sectors (29 sectors) in the Mercury Assessment Report. The inventory contains monthly emissions and approximately 13 types of sources are point sources (covering more than 70% of emissions in 2017). This model estimates Hg emissions based on the J o u r n a l P r e -p r o o f Journal Pre-proof mass-balance method and considers Hg input and removal in the studied sectors. The database for this model included Hg content of raw materials/fuel, coal quality, activity level, types of air pollution control device (APCD) combinations, and Hg removal and speciation for different types of APCDs. In this study, we applied this model to compile the high-time resolution Hg emission inventory. We focused on the emissions in the Beijing-Tianjin-Hebei (BTH) region because this area is the main Hg emission region in China. It is also the source area of Hg emissions for our observation site based on the Potential Source Contribution Factors result (Tang et al., 2019) . The studied period was from Jan 1, 2020 to Mar 10, 2020. China activated the First-Level Public Health Emergency Response in late January when strict lockdown measures were implemented, and economic activities gradually resumed after mid-February. Thus, we divided the studied period into three sub-periods based on the execution times of the COVID-19 lockdown measures. We denoted these three sub-periods as PRE_LOCK (Jan 1 -Jan 19), LOCK (Jan 20 -Feb 15), and AFTER_LOCK (Feb 16 -Mar 10), respectively. Moreover, COVID-19 broke during the Chinese New Year and there were measures to control air pollutants. To separate the impact of existing measures and additional lockdown measures, we complied an additional emission inventory (em_BAU) for the studied period. We first compiled the 2019 emissions (em_2019) and derived the 2015-2017 emissions from the CAME model. Then, we assumed that the emission amount of each sector followed their emission trend from 2015-2019. This trend continued to 2020 and we obtained the em_BAU. We followed most of the datasets in the CAME model, but we assumed that all new released capacity in 2019 and 2020 had applied advanced APCDs considering the strict pollution control requirement. We also updated the activity level based on a field and questionnaire investigation, literature review, and expert consultant. The detailed data sources are listed in Table S1 . We obtained daily clinker yields for cement clinker J o u r n a l P r e -p r o o f Journal Pre-proof production (CEM) by dividing the monthly production by the daily operation rate of production lines. We adopted a similar method for the iron and steel production sector (ISP). Daily coal consumption data of coal-fired power plants (CFPP) were from the Coal Electricity Group. Daily residential coal combustion (RCC) was based on the field and questionnaire investigation. We generated the coal consumption data for the heating supply sector (HS) based on the change of heating areas. Biomass combustion (BIO) was estimated based on the number of fire points. The change in coal consumption for the coking sector (COK) followed the change in crude steel production, considering that more than 80% of coke was produced in steel plants. Daily mobile oil combustion was adjusted based on oil combustion and the change in transportation data. For the other sectors, we simply calculated the emissions based on the change in activity level and then distributed them to a daily resolution. We finally improved the time resolution of the inventory to a 1-h resolution based on the hourly distribution coefficient (HDC) (Supporting information S1 and Fig.S1 ). We conducted TM observations in Tsinghua University, Haidian district of Beijing, China (Fig.S2 ). The sampling site (40.00 N, 116.33 E) was located at the top of the Environmental Engineering Laboratory Building. Only the campus and residential areas fell within a 1 km radius, and no industrial point sources were within a 30 km radius. It is a representative site for the atmospheric environment in urban Beijing, where many atmospheric observation experiments have been conducted over the years (Hu et al., 2020; Tang et al., 2019) . We used atmospheric Hg monitoring system (Tekran TM 2537B/1130/1135)to monitor speciated Hg in the atmosphere, which has been widely used for atmospheric Hg observations around the world. Previous studies have described the principle of this instruments in detail, as well as routine maintenance for it We used a generalized additive model (GAM) to establish the relationship between TM and various variables, which have been used to simulate the changing trends of air pollutants (Gong et al., 2018; Gong et al., 2017) . The GAM produces a simple and explicit formulation of response-predictor relationships in a neural network model (Aldrin and Haff, 2005) . The equation is as follows: where i is the observation of the i th hour, g is the link function, and μ i is the expectation of the dependent variable, which specifies the relationship between the linear formulation on the right side of equation (1) Step 1 Dataset Preparation We prepared dataset_2020 for the studied period ( (Polissar et al., 1999) . Step 2 Variable Selection According to Woods' study (Wood, 2013) , it is best to use the p-values from the larger model rather than those calculated through model comparisons to assess whether or not a variable should be left in the model. Thus, we included all parameters of the dataset in the model and deleted the insignificant ones based on the significance (F-statistics) and p-value of each variable. We used Penalized Cubic Regression Splines for the smoothing function of each parameter to balance between not fitting the observed data well and overfitting the data by choosing an J o u r n a l P r e -p r o o f effective number of degrees of freedom. The restricted maximum likelihood approach was used to conduct the overall smoothness selection criteria. Finally, we obtained 9 independent parameters for the model (Table S2) . Step 3 Model Evaluation The model evaluation results were shown in Table S3, S4 and Fig.S4 . We used the 10-fold cross validation method to evaluate the accuracy of the GAM models (Fushiki, 2009 ). The strong correlation, with a determination coefficient (R 2 ) of approximately 97%, demonstrated the accuracy of GAM ( Table S3 ). The parameters in the GAM model explained 63.2% of the variance in TM concentrations and the adjusted R 2 reached 0.620 (Table S3) . Thus, the independent variables could interpret the TM concentrations well. All the selected independent variables significantly correlated with TM concentrations ( Table S4 ). The QQ-plot results showed that the GAM produced good results around the average concentration and had larger uncertainties in larger and lower values (Fig.S4a) . The residuals vs predictors (Fig.S4b) and histogram of the residuals (Fig.S4c) showed an almost unbiased simulation of TGM. The sharp decrease in the autocorrelation of residuals reflected almost no lag days during the simulation (Fig.S4d) . The model evaluation process validated the underlying assumptions of the homogeneity, normality, and independence of the GAM, and the results ensured the quality of the model. We used the scenario analysis method to predict the impact of emission reductions and meteorology changes on air Hg concentrations during the LOCK sub-period of 2020. The emissions and meteorology during the PRE_LOCK period were used for comparison. The fitted TM concentrations during the LOCK sub-period, which was predicted based on dataset_2020, was regarded as the BASE scenario. Then, we replaced the actual emissions during the LOCK sub-period with em_BAU (em_covid scenario), and the difference in TM between this scenario and the BASE was regarded as the impact of the J o u r n a l P r e -p r o o f COVID-19 lockdown. We normalized the emissions during the LOCK sub-period by using the hour-mean emissions from PRE-LOCK (em_ctrl scenario). The TM change between em_ctrl and BASE reflected the impact of existing pollution control measures. To set the other scenarios, the basic principle was that we normalized the studied parameter during the LOCK sub-period by using the hour-mean value from the PRE_LOCK sub-period and kept the others unchanged as the BASE scenario. Then, we ran the predict.gam command to generate TM concentrations. The difference in TM between the BASE and objected scenario represented the impact of the corresponding parameter. Atmospheric Hg emissions in the BTH region were approximately 7.7 t during the studied period. Dominant sources presented various trends, especially CEM, RCC, CFPP, and ISP (Fig.1a) . CEM emitted the largest amount of Hg during PRE_LOCK, accounting for 28% of total Hg emissions, and then sharply reduced at the beginning of the LOCK sub-period. Such a decline almost returned to the PRE_LOCK level at the end of the LOCK sub-period. Contrary to the emissions from CEM, Hg emissions from RCC increased from 10.6 kg/d during the PRE_LOCK sub-period to 13.9 kg/d during the LOCK sub-period, increasing by approximately 31%. Then, the emissions gradually decreased to 6.1 kg/d in the AFTER_LOCK sub-period with the increase of temperature. As for CFPP, 1081 kg of Hg emissions were released during the studied period. The daily mean value of Hg emissions during AFTER_LOCK was still 8.1 kg/d (37%) lower than that during PRE_LOCK. For ISP, daily Hg emissions during LOCK and AFTER_LOCK decreased by 7% and 13%, respectively. As for diurnal variation trends, they were almost the same, with two peaks around 11:00 and 18:00 and two troughs around 4:00 and 15:00, respectively (Fig.1b) . Compared with the sectoral HDC curve (Fig.S2) J o u r n a l P r e -p r o o f We built the GAM model for TM following the procedure in Section 2.3. The time series of both the observed and fitted TM concentrations generally showed good agreement (Fig.S6) . The fitted TM concentrations were approximately 2.15±0.53 ng/m 3 during the PRE_LOCK sub-period and 2.39±0.63 ng/m 3 during the LOCK sub-period. Based on the scenario analysis, we predicted the contributions of different impacting factors. As shown in Fig.3 The influence of detailed parameters varied. TM concentrations increased almost linearly with the raise of anthropogenic emissions (Fig.4a) . Of all the sources, the emission reductions from CEM was the most significant and contributed to the decrease of TM concentrations by 0.11 ng/m 3 across the studied period (Table 1) . Emission reductions from CEM were mainly caused by the effect of existing air pollution control meaasures, such as off-peak production. As for ISP and CFPP, TM concentrations were reduced by 0.01 ng/m 3 and 0.05 ng/m 3 , respectively, due to emission reductions in these two sectors. Compared to these three sectors, the increased energy requirement from residential coal combustion increased TM concentrations by 0.02 ng/m 3 during the LOCK sub-period. Surface meteorology increased TM concentrations by 0.27 ng/m 3 during the LOCK sub-period, mainly due to changes in relative humidity and temperature (Table 1) . According to Fig.4b , TM concentration increased with the raise of temperature, possibly because the increase of temperature would promote the Hg emissions from natural surface (Qin et al., 2019; (Horowitz et al., 2017) . According to an empirical rule that a 10 ℃ temperature increase doubles the rates for chemical reactions near room temperature (Kissinger, 1957) . This rule has been proved to be applicable to the reduction of oxidized Hg in boreal soil (Pannu et al., 2014) . We observed a similar relationship between TM and relative humidity (Fig.4c) . The raise in relative humidity might enhance the liquid phase reduction process that converts reactive Hg to GEM (Horowitz et al., 2017; Huang et al., 2019; Saiz-Lopez et al., 2018) . During the LOCK sub-period, hourly surface temperature and relative humidity increased by 1.2℃ and 9.5%, respectively, which correspondingly raised the TM by 0 Moreover, our study identified the importance of relative humidity on TM concentrations during the LOCK sub-period. This could be because liquid aerosol enhanced Hg photoreduction, based on a previous isotope study (Huang et al., 2019) . This process is supposed to change global Hg cycling (Horowitz et al., 2017) , but details on this are still unclear. A model study assumed that Hg liquid reduction occurs when relative humidity is higher than 35% and that the reaction rate is independent of relative humidity (Horowitz et al., 2017) . However, the spline of relative humidity indicated a positive correlation between TM and relative humidity. This means that if the relative humidity affects the liquid reduction of Hg by changing the phase state of particles, then the assumption about the liquid phase (when relative humidity is higher than 35%) requires reconsideration. Otherwise, relative humidity may directly or indirectly impact the reaction rate. To summarize, we still need to understand the physicochemical and biogeochemical transformation of Hg in the atmosphere and at atmospheric interfaces to evaluate the effect of future climate change on Hg behavior, especially the impact from relative humidity. Table1 The impact ☒ √The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. ☒The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: J o u r n a l P r e -p r o o f Generalised additive modelling of air pollution, traffic volume and meteorology Trans-provincial health impacts of atmospheric mercury emissions in China Observations of atmospheric mercury in China: a critical review Estimation of prediction error by using K-fold cross-validation Ozone in China: Spatial Distribution and Leading Meteorological Factors Controlling O-3 in 16 Chinese Cities A new mechanism for atmospheric mercury redox chemistry: implications for the global mercury budget Chemical characteristics and sources of water-soluble organic aerosol in southwest suburb of Beijing Diel variation in mercury stable isotope ratios records photoreduction of PM2.5-bound mercury Climate change 2014: synthesis report. Contribution of working groups I, II and III to the fifth assessment report of the intergovernmental panel on climate change Atmospheric mercury in the Lake Michigan basin: Influence of the Chicago/Gary urban area Source attribution for mercury deposition with an updated atmospheric mercury emission inventory in the Pearl River Delta Region A highly resolved mercury emission inventory of Chinese coal-fired power plants Measure-specific effectiveness of air pollution control on China's atmospheric mercury concentration and deposition during Current and future levels of mercury atmospheric pollution on a global scale Quantifying the effects of soil temperature, moisture and sterilization on elemental mercury formation in boreal soils The aerosol at Barrow, Alaska: long-term trends and source locations Characteristics of atmospheric mercury in a suburban area of east China: sources, formation mechanisms, and regional transport Quantitative assessment of atmospheric emissions of toxic heavy metals from anthropogenic sources in China: historical trend, spatial distribution, uncertainties, and control policies Impact of ultra-low emission technology retrofit on the mercury emissions and cross-media transfer in coal-fired power plants On p-values for smooth components of an extended generalized additive model Mitigation options of atmospheric Hg emissions in China Updated atmospheric speciated mercury emissions from iron and steel production in China during Temporal trend and spatial distribution of speciated atmospheric mercury emissions in China during 1978-2014 Emission-limit-oriented strategy to control atmospheric mercury emissions in coal-fired power plants towards the implementation of Minamata Convention Multi-year monitoring of atmospheric total gaseous mercury at a remote high-altitude site (Nam Co, 4730 m a.s.l.) in the inland Tibetan Plateau region Updated emission inventories for speciated atmospheric mercury from anthropogenic sources in We would like to thank Dian Ding for helping us extract the WRF data. The authors declare that they have no conflicts of interest.