Environ Monit Assess DOI 10.1007/s10661-013-3588-2
Meteorological influence on predicting surface SO2 concentration from satellite remote sensing in Shanghai, China Dan Xue & Jingyuan Yin
Received: 9 July 2013 / Accepted: 12 December 2013 # Springer Science+Business Media Dordrecht 2013
Abstract In this study, we explored the potential applications of the Ozone Monitoring Instrument (OMI) satellite sensor in air pollution research. The OMI planetary boundary layer sulfur dioxide (SO2_PBL) column density and daily average surface SO2 concentration of Shanghai from 2004 to 2012 were analyzed. After several consecutive years of increase, the surface SO2 concentration finally declined in 2007. It was higher in winter than in other seasons. The coefficient between daily average surface SO2 concentration and SO2_PBL was only 0.316. But SO2_PBL was found to be a highly significant predictor of the surface SO2 concentration using the simple regression model. Five meteorological factors were considered in this study, among them, temperature, dew point, relative humidity, and wind speed were negatively correlated with surface SO2 concentration, while pressure was positively correlated. Furthermore, it was found that dew point was a more effective predictor than temperature. When these meteorological factors were used in multiple regression, the determination coefficient reached 0.379. The relationship of the surface SO2 concentration and meteorological factors was seasonally dependent. In summer and autumn, the regression model performed better than in spring and winter. The surface SO2 concentration predicting method proposed in this study can be easily D. Xue (*) : J. Yin School of Computer Engineering and Science, Shanghai University, Shanghai 200444, China e-mail: [email protected]
adapted for other regions, especially most useful for those having no operational air pollution forecasting services or having sparse ground monitoring networks. Keywords OMI . Sulfur dioxide . Shanghai . Meteorological influence . Stepwise regression
Introduction In recent decades, vigorous economic growth, expanding of anthropogenic activity, and urbanization resulted in significant deterioration of air quality in China. Sulfur dioxide (SO2) is considered one of the indicators of air quality. It is formed primarily from the combustion of sulfur-containing fuels. Owing to photochemical or catalytic reactions in the atmosphere, SO2 is partially converted into SO3 or sulfuric acid (H2SO4) aerosols and also its salt particles. The sulfate aerosols can exert influence on weather and climate (Stier et al. 2007), cause visibility impairments (Hand and Malm 2007), and pose a hazard to public health (He et al. 2002; Hu et al. 2010; Schlesinger and Cassee 2003). A number of studies have been conducted to investigate the sulfurous pollution in China. Surface observations of SO2 were made in and near Beijing (Li et al. 2007; Sun et al. 2009), Yangtze River Delta (YRD) (Costabile et al. 2006), Pearl River Delta (PRD) (Zhang et al. 2008), and rural areas (Meng et al. 2010). Although ground-based measurements are generally considered to be accurate, they are representative for only relatively small areas around point stations.
Environ Monit Assess
Surface SO2 concentration is subject to change depending on the local topography, source emission, and surrounding meteorological conditions. It has been correlated with the combination of the various meteorological factors statistically in several studies over the past decade (Olcese and Toselli 1997; Ilten and Selici 2008). Cuhadaroglu and Demirci (1997) performed a study to show the influence of some meteorological factors on air pollution in Trabzon City in Turkey. They obtained correlations for SO2 and particle concentrations between meteorological factors. Their results indicated that there is a moderate and weak level of relation between the SO2 level and the meteorological factors in Trabzon City. In the study presented by Bridgman et al. (2002), the relationship of SO2 concentrations to six major meteorological parameters has been investigated. Results showed that SO2 concentrations were strongly related to colder temperature, higher relative humidity, and lower wind speed. For prediction of SO2 and smoke concentrations of Kayseri, Turkey, multiple regression equations including meteorological parameters and previous day's pollutant concentrations have been used by Kartal and Özer (1998). The changes of air quality in Erzurum, Turkey and the correlation of SO2 and total suspended particle (TSP) pollution in Erzurum City with meteorological parameters such as wind speed, temperature, atmospheric pressure, precipitation, and relative humidity were researched by the Turalıoğlu et al. (2005). Akpinar et al. (2008) investigated the relationship between SO2 and meteorological factors during the winter season for the city of Elazığ, Turkey, where a moderate and weak level of relation was found. Ilten and Selici (2008) reported that TSP and SO2 concentrations were highly correlated with meteorological variables in Balikesir, Turkey. Banerjee et al. (2011) evaluated the statistical models correlating air pollutants TSP, SO2, and NO2 and meteorological variables at Pantnagar, India. Source emission is another important factor that influences surface SO2 concentration, but getting its statistics is difficult. However, a number of satellite sensors launched in the last 15–20 years, such as GOME, SCIAMACHY, and the Ozone Monitoring Instrument (OMI), offer powerful tools for studying SO2 in the atmosphere. SO2 is a highly active pollutant gas, which mainly exists in the vicinity of emission sources and the lower atmosphere, so SO2 column density of the boundary layer is an indicator of SO2 pollution in the lower atmosphere. Through analyzing the SO 2 column
density, ground SO2 emission can be detected from space by using spectroscopic measurements in the ultraviolet. For example, Li et al. (2010) observed dramatic reductions of SO2 emissions by OMI in 2008 over several areas in northern China due to the strict emission control measurements of the 2008 Beijing Olympic Games. But none of these researches have been devoted to explore the potential applications of satellite data in estimating surface SO2 pollution. The variability of air pollutant concentrations in a location has different characteristics based on the prevalent meteorological conditions (Olcese and Toselli 1997). Moreover, as meteorological factors differ significantly under varying geographical conditions, it is therefore essential to study the impact of meteorology on the variation of ground level pollutant concentrations of megacity Shanghai. In this study, we explored the potential applications of combining satellite data and meteorological factors in estimating surface SO2 concentration. Air quality and meteorological and satellite data were analyzed for the city of Shanghai, China. Besides examining the relationship between the surface SO2 concentration and satellite-retrieved SO2 data, we also demonstrated a method that utilized satelliteretrieved SO2 data and meteorological data in the forecasting of surface SO2 pollution. We envision that this method not only can be easily adapted for other regions, but also will be more useful for those having limited coverage from ground monitors. It should also be noted that the objective of this study is not to establish an operational forecasting system, but rather to investigate the strengths and limitations of satellite and meteorological data in air pollution applications.
Materials and methods Surface SO2 concentration Daily values of the air pollution index (API) of SO2 in Shanghai were collected from the Shanghai Environmental Protection Bureau (http://www.envir.gov.cn/ airnews/). The API data have been extensively used in studies on air pollution in China. It calculates the 24-h average concentrations of pollutant from noon of the present day to noon of the previous day. The measurement methods, validity, and issues of the dataset have been reviewed by Qu et al. (2010). For a given city, a network consisting of multiple monitoring stations
Environ Monit Assess
measures designated pollutants. These monitoring stations are located at different zones (traffic, residential, commercial, and industrial) and the suburban background. The daily average concentration across the entire network for each pollutant is converted to the subAPI of that pollutant, based on the air quality classification table (Table 1) and linear interpolation. The formula used for converting sub-API to concentration is
18 °C. The city has a frost-free period lasting up to 230 days a year. The average annual rainfall is 1,200 mm (about 50 in.). The daily average meteorological data of Shanghai used in this study, including temperature (T), dew point (DP), relative humidity (RH), pressure (P), and wind speed (WS), were downloaded from the weather underground (http:// www.wunderground.com).
C ¼ C low þ ðI−I low Þ= I high −I low C high −C low
OMI SO2 data
ð1Þ where C is the concentration and I is the API value. Ihigh and Ilow, the two values most approaching to value I in Table 1, Chigh and Clow represent the concentrations corresponding to Ihigh and Ilow, respectively. Study area and meteorological data Located in the delta region of the Yangtze River on China's eastern coast, Shanghai is one of the major economic centers of China (Fig. 1). It has a total area of 6,304 km2 and is divided into 17 districts. In 2011, 23.5 million people were living in this densely populated city. With a century-long history of industrialization, Shanghai is currently one of the fastest growing economies in the world. Its per capita GDP for the year 2011 came to $12,784. With a pleasant northern subtropical maritime monsoon climate, Shanghai weather has four distinct seasons, generous sunshine, and abundant rainfall. Its spring and autumn are relatively short compared with summer and winter. The average annual temperature is Table 1 Breakpoints for converting air pollutant concentrations to API values Air pollution index (API)
Pollutant concentration(μg m−3) SO2
NO2
Air pollution level
PM10
50
50
80
50
Excellent
100
150
120
150
Good
200
800
280
350
Slight polluted
300
1,600
565
420
Lightly polluted
400
2,100
750
500
Moderately polluted
500
2,620
940
600
Heavily polluted
SO2_PBL is measured by OMI, launched in 2004, onboard NASA's EOS Aura satellite. OMI is a DutchFinnish nadir-viewing hyperspectral imager that provides daily global coverage at a high spatial resolution (13 km×24 km at nadir) capable of mapping pollution products on urban scales (Witte et al. 2009). The OMI sensor PBL SO2 column amounts from measurements of backscattered solar UV (BUV) radiation in the wavelength range of 311–315 nm using the band residual difference (BRD) algorithm (Krotkov et al. 2008). The retrieved SO2 slant column density (SCD) (i.e., the effective total column along the mean path of BUV photons) is converted to the total SO2 vertical column density (VCD) in Dobson units (1 DU=2.69×1016 molecules/cm2) using a constant air mass factor (AMF) 0.36, Total SO2 VCD ¼ SCD=AMF
ð2Þ
For this study, we used the SO2_PBL data from the OMSO2 L2G product (http://disc.sci.gsfc.nasa.gov/ Aura/data-holdings/OMI/omso2g_v003.shtml). Daily retrievals, allocated to grid cells of 0.125°×0.125°, were first filtered to remove data with a large solar zenith angle (>70°), or relatively high cloud fraction (OMI retrieved radiative cloud fraction >0.3), or possible contamination due to the OMI row anomaly (Witte et al. 2009) and then averaged to 0.25°×0.25° resolution. The daily OMI Level2 SO2_PBL data with a large temporal scale from October 2004 to September 2012 of Shanghai (112°E–115.5°E, 21.5°N–24°N) was used in this study. Models for estimating surface SO2 concentration To find a better approach for inferring surface SO2 concentration from remote sensing, here we evaluated a simple linear regression model and a multiple linear
Environ Monit Assess Fig. 1 The location of Shanghai City
regression model for the city of Shanghai. The simple linear model takes the form of ½SO2 ¼ a0 ðSO2 PBLÞ þ b0
ð3Þ
The dependent variable on the left-hand side, [SO2], is the daily average surface SO2 concentration measured at the monitoring network. The independent variable SO2_PBL on the right-hand side denotes the PBL SO2 column density derived by OMI. a0 and b0 are regression coefficients determined through simple linear regression. So far, the response of surface SO2 concentration to meteorological condition changes at the ground level has not been well addressed in the city of Shanghai. Therefore, in the multiple linear regression model, meteorological parameters were introduced as additional predictor variables, and surface SO2 concentration was estimated using
excluded from the input into the models. A generally used measure of the goodness of fit of a linear model is R2, sometimes called the coefficient of determination. The coefficient of determination is that proportion of the total variability in the dependent variable that is accounted for by the regression equation. A value of R2 =1 indicates that the fitted equation accounts for all the variability of the values of the dependent variables in the sample data. At the other extreme, R2 =0 indicates that the regression equation explains none of the variability. The model performance was assessed by a comprehensive package of model accuracy prediction methods as suggested by Banerjee et al. (2011). A brief description of the applied performance measures is presented below. a. Root mean square error (RMSE): N
½SO2 ¼ a0 ðSO2 PBLÞ þ b0 þ a1 Par1 … þ an Parn
Mean square error MSE ¼ N
ð4Þ in which Par1…n represented meteorological parameters, including temperature (T), dew point (DP), relative humidity (RH), pressure (P), and wind speed (WS), and a1…n were regression coefficients for these meteorological factors. Pursuing the experiment, it was assumed that the dependent variables follow the normal distribution, homoscedasticity, i.e., the data have equal variance, and the differences between actual and theoretical values of dependent variables were independent (İçağa and Sabah 2009). We determined the regression coefficients and the intercept b0 via stepwise regression. Data from the period October 2004 to September 2011 were used to construct the two models. To verify the validity of the two regression models, they were applied to the period October 2011 to September 2012, being the data period
−1
2
∑ ½Oi −Pi i¼1
RMSE ¼
pffiffiffiffiffiffiffiffiffiffi MSE
b. Mean absolute error (MAE) MAE ¼ N −1
N X
jOi −Pi j
i¼1
c. Correlation (r): The formula used for the determination of the correlation of coefficient (r) has the form: " r¼
# " n #1=2 n n 2 X X . X 2 Oi −O Pi −P Oi −O Pi −P i¼1
i¼1
i¼1
where N is the number of data points, Oi is the observed surface SO2 concentrations, Pi is the model predicted
Environ Monit Assess
is the average value of observed concentration, O is the average value of predicted concentrations, and P concentration. All statistical analyses were conducted using the SPSS17.0 statistical software program. Model parameter estimates were presented with three decimal places or at least one significant digit.
Results and discussion Temporal variations of SO2_PBL and surface SO2 concentration Figure 2 shows the interannual variations of SO2_PBL column densities and surface SO2 concentration from 2005 to 2011 in Shanghai. Overall, the variation trends of both SO2_PBL and surface SO2 concentration were downward in the past 7 years. Surface SO2 concentration was the highest in 2005 with mean daily value of 60.870 μg m−3. It reduced to 50.826 μg m−3 in 2006, but increased in 2007 with the concentration of 55.407 μg m−3. This change was the result of the rapid increase of annual electricity consumption and the desulfurization program of Shanghai City. Coal is the main energy source in China and its combustion is the main cause of the atmospheric SO2 emissions in the country. The main coal consumption in Shanghai comes from coal-fired power plants. Shanghai started desulfurizing its coal-fired plants in 2005. After several consecutive years of increase, SO2 concentration finally declined in 2007. It has been reduced by half in the past 5 years and reached 29.181 μg m−3 in 2011. This showed that the desulfurization program for coal-fired plants had an
impact on overall SO2 emissions against rising electricity generation. By June 2009, Shanghai installed flue gas desulfurization devices for all the 10-GW capacities of coal-fired stations. Meanwhile, 695 MW of small and inefficient coal-fired plants in total had been shut down. Now with all the coal-fired power stations desulfurized, there are good reasons to believe that SO2 emissions will keep decreasing. However, the SO2_PBL column amount has different characteristics with the surface SO2 concentration. The highest column appeared in 2005 with a concentration of 1.294 DU, while the lowest column appeared in 2009 with a concentration of 0.973 DU. In the years 2010 and 2011, the concentrations of SO2_PBL column density were a little more than 1 DU. The monthly variations of both SO2_PBL column density and surface SO2 concentration in Shanghai have been plotted in Fig. 3 from October 2004 to September 2012 with their characteristic seasonal variations. As illustrated from the observed data, higher surface SO2 concentrations are experienced from the month of November and onward, which are particularly high in winters (December to February) in Shanghai. The reason is during wintertime the need for electricity is highest. Thus, SO2 pollution is higher in winter than in other seasons since anthropogenic emissions are expected to be higher because of heating. One-way analysis of variance (ANOVA) results show that SO2_PBL column density is significantly different between seasons. Among them, the SO2_PBL column density of winter is significantly different with the other three seasons (p
Meteorological influence on predicting surface SO2 concentration from satellite remote sensing in Shanghai, China Dan Xue & Jingyuan Yin
Received: 9 July 2013 / Accepted: 12 December 2013 # Springer Science+Business Media Dordrecht 2013
Abstract In this study, we explored the potential applications of the Ozone Monitoring Instrument (OMI) satellite sensor in air pollution research. The OMI planetary boundary layer sulfur dioxide (SO2_PBL) column density and daily average surface SO2 concentration of Shanghai from 2004 to 2012 were analyzed. After several consecutive years of increase, the surface SO2 concentration finally declined in 2007. It was higher in winter than in other seasons. The coefficient between daily average surface SO2 concentration and SO2_PBL was only 0.316. But SO2_PBL was found to be a highly significant predictor of the surface SO2 concentration using the simple regression model. Five meteorological factors were considered in this study, among them, temperature, dew point, relative humidity, and wind speed were negatively correlated with surface SO2 concentration, while pressure was positively correlated. Furthermore, it was found that dew point was a more effective predictor than temperature. When these meteorological factors were used in multiple regression, the determination coefficient reached 0.379. The relationship of the surface SO2 concentration and meteorological factors was seasonally dependent. In summer and autumn, the regression model performed better than in spring and winter. The surface SO2 concentration predicting method proposed in this study can be easily D. Xue (*) : J. Yin School of Computer Engineering and Science, Shanghai University, Shanghai 200444, China e-mail: [email protected]
adapted for other regions, especially most useful for those having no operational air pollution forecasting services or having sparse ground monitoring networks. Keywords OMI . Sulfur dioxide . Shanghai . Meteorological influence . Stepwise regression
Introduction In recent decades, vigorous economic growth, expanding of anthropogenic activity, and urbanization resulted in significant deterioration of air quality in China. Sulfur dioxide (SO2) is considered one of the indicators of air quality. It is formed primarily from the combustion of sulfur-containing fuels. Owing to photochemical or catalytic reactions in the atmosphere, SO2 is partially converted into SO3 or sulfuric acid (H2SO4) aerosols and also its salt particles. The sulfate aerosols can exert influence on weather and climate (Stier et al. 2007), cause visibility impairments (Hand and Malm 2007), and pose a hazard to public health (He et al. 2002; Hu et al. 2010; Schlesinger and Cassee 2003). A number of studies have been conducted to investigate the sulfurous pollution in China. Surface observations of SO2 were made in and near Beijing (Li et al. 2007; Sun et al. 2009), Yangtze River Delta (YRD) (Costabile et al. 2006), Pearl River Delta (PRD) (Zhang et al. 2008), and rural areas (Meng et al. 2010). Although ground-based measurements are generally considered to be accurate, they are representative for only relatively small areas around point stations.
Environ Monit Assess
Surface SO2 concentration is subject to change depending on the local topography, source emission, and surrounding meteorological conditions. It has been correlated with the combination of the various meteorological factors statistically in several studies over the past decade (Olcese and Toselli 1997; Ilten and Selici 2008). Cuhadaroglu and Demirci (1997) performed a study to show the influence of some meteorological factors on air pollution in Trabzon City in Turkey. They obtained correlations for SO2 and particle concentrations between meteorological factors. Their results indicated that there is a moderate and weak level of relation between the SO2 level and the meteorological factors in Trabzon City. In the study presented by Bridgman et al. (2002), the relationship of SO2 concentrations to six major meteorological parameters has been investigated. Results showed that SO2 concentrations were strongly related to colder temperature, higher relative humidity, and lower wind speed. For prediction of SO2 and smoke concentrations of Kayseri, Turkey, multiple regression equations including meteorological parameters and previous day's pollutant concentrations have been used by Kartal and Özer (1998). The changes of air quality in Erzurum, Turkey and the correlation of SO2 and total suspended particle (TSP) pollution in Erzurum City with meteorological parameters such as wind speed, temperature, atmospheric pressure, precipitation, and relative humidity were researched by the Turalıoğlu et al. (2005). Akpinar et al. (2008) investigated the relationship between SO2 and meteorological factors during the winter season for the city of Elazığ, Turkey, where a moderate and weak level of relation was found. Ilten and Selici (2008) reported that TSP and SO2 concentrations were highly correlated with meteorological variables in Balikesir, Turkey. Banerjee et al. (2011) evaluated the statistical models correlating air pollutants TSP, SO2, and NO2 and meteorological variables at Pantnagar, India. Source emission is another important factor that influences surface SO2 concentration, but getting its statistics is difficult. However, a number of satellite sensors launched in the last 15–20 years, such as GOME, SCIAMACHY, and the Ozone Monitoring Instrument (OMI), offer powerful tools for studying SO2 in the atmosphere. SO2 is a highly active pollutant gas, which mainly exists in the vicinity of emission sources and the lower atmosphere, so SO2 column density of the boundary layer is an indicator of SO2 pollution in the lower atmosphere. Through analyzing the SO 2 column
density, ground SO2 emission can be detected from space by using spectroscopic measurements in the ultraviolet. For example, Li et al. (2010) observed dramatic reductions of SO2 emissions by OMI in 2008 over several areas in northern China due to the strict emission control measurements of the 2008 Beijing Olympic Games. But none of these researches have been devoted to explore the potential applications of satellite data in estimating surface SO2 pollution. The variability of air pollutant concentrations in a location has different characteristics based on the prevalent meteorological conditions (Olcese and Toselli 1997). Moreover, as meteorological factors differ significantly under varying geographical conditions, it is therefore essential to study the impact of meteorology on the variation of ground level pollutant concentrations of megacity Shanghai. In this study, we explored the potential applications of combining satellite data and meteorological factors in estimating surface SO2 concentration. Air quality and meteorological and satellite data were analyzed for the city of Shanghai, China. Besides examining the relationship between the surface SO2 concentration and satellite-retrieved SO2 data, we also demonstrated a method that utilized satelliteretrieved SO2 data and meteorological data in the forecasting of surface SO2 pollution. We envision that this method not only can be easily adapted for other regions, but also will be more useful for those having limited coverage from ground monitors. It should also be noted that the objective of this study is not to establish an operational forecasting system, but rather to investigate the strengths and limitations of satellite and meteorological data in air pollution applications.
Materials and methods Surface SO2 concentration Daily values of the air pollution index (API) of SO2 in Shanghai were collected from the Shanghai Environmental Protection Bureau (http://www.envir.gov.cn/ airnews/). The API data have been extensively used in studies on air pollution in China. It calculates the 24-h average concentrations of pollutant from noon of the present day to noon of the previous day. The measurement methods, validity, and issues of the dataset have been reviewed by Qu et al. (2010). For a given city, a network consisting of multiple monitoring stations
Environ Monit Assess
measures designated pollutants. These monitoring stations are located at different zones (traffic, residential, commercial, and industrial) and the suburban background. The daily average concentration across the entire network for each pollutant is converted to the subAPI of that pollutant, based on the air quality classification table (Table 1) and linear interpolation. The formula used for converting sub-API to concentration is
18 °C. The city has a frost-free period lasting up to 230 days a year. The average annual rainfall is 1,200 mm (about 50 in.). The daily average meteorological data of Shanghai used in this study, including temperature (T), dew point (DP), relative humidity (RH), pressure (P), and wind speed (WS), were downloaded from the weather underground (http:// www.wunderground.com).
C ¼ C low þ ðI−I low Þ= I high −I low C high −C low
OMI SO2 data
ð1Þ where C is the concentration and I is the API value. Ihigh and Ilow, the two values most approaching to value I in Table 1, Chigh and Clow represent the concentrations corresponding to Ihigh and Ilow, respectively. Study area and meteorological data Located in the delta region of the Yangtze River on China's eastern coast, Shanghai is one of the major economic centers of China (Fig. 1). It has a total area of 6,304 km2 and is divided into 17 districts. In 2011, 23.5 million people were living in this densely populated city. With a century-long history of industrialization, Shanghai is currently one of the fastest growing economies in the world. Its per capita GDP for the year 2011 came to $12,784. With a pleasant northern subtropical maritime monsoon climate, Shanghai weather has four distinct seasons, generous sunshine, and abundant rainfall. Its spring and autumn are relatively short compared with summer and winter. The average annual temperature is Table 1 Breakpoints for converting air pollutant concentrations to API values Air pollution index (API)
Pollutant concentration(μg m−3) SO2
NO2
Air pollution level
PM10
50
50
80
50
Excellent
100
150
120
150
Good
200
800
280
350
Slight polluted
300
1,600
565
420
Lightly polluted
400
2,100
750
500
Moderately polluted
500
2,620
940
600
Heavily polluted
SO2_PBL is measured by OMI, launched in 2004, onboard NASA's EOS Aura satellite. OMI is a DutchFinnish nadir-viewing hyperspectral imager that provides daily global coverage at a high spatial resolution (13 km×24 km at nadir) capable of mapping pollution products on urban scales (Witte et al. 2009). The OMI sensor PBL SO2 column amounts from measurements of backscattered solar UV (BUV) radiation in the wavelength range of 311–315 nm using the band residual difference (BRD) algorithm (Krotkov et al. 2008). The retrieved SO2 slant column density (SCD) (i.e., the effective total column along the mean path of BUV photons) is converted to the total SO2 vertical column density (VCD) in Dobson units (1 DU=2.69×1016 molecules/cm2) using a constant air mass factor (AMF) 0.36, Total SO2 VCD ¼ SCD=AMF
ð2Þ
For this study, we used the SO2_PBL data from the OMSO2 L2G product (http://disc.sci.gsfc.nasa.gov/ Aura/data-holdings/OMI/omso2g_v003.shtml). Daily retrievals, allocated to grid cells of 0.125°×0.125°, were first filtered to remove data with a large solar zenith angle (>70°), or relatively high cloud fraction (OMI retrieved radiative cloud fraction >0.3), or possible contamination due to the OMI row anomaly (Witte et al. 2009) and then averaged to 0.25°×0.25° resolution. The daily OMI Level2 SO2_PBL data with a large temporal scale from October 2004 to September 2012 of Shanghai (112°E–115.5°E, 21.5°N–24°N) was used in this study. Models for estimating surface SO2 concentration To find a better approach for inferring surface SO2 concentration from remote sensing, here we evaluated a simple linear regression model and a multiple linear
Environ Monit Assess Fig. 1 The location of Shanghai City
regression model for the city of Shanghai. The simple linear model takes the form of ½SO2 ¼ a0 ðSO2 PBLÞ þ b0
ð3Þ
The dependent variable on the left-hand side, [SO2], is the daily average surface SO2 concentration measured at the monitoring network. The independent variable SO2_PBL on the right-hand side denotes the PBL SO2 column density derived by OMI. a0 and b0 are regression coefficients determined through simple linear regression. So far, the response of surface SO2 concentration to meteorological condition changes at the ground level has not been well addressed in the city of Shanghai. Therefore, in the multiple linear regression model, meteorological parameters were introduced as additional predictor variables, and surface SO2 concentration was estimated using
excluded from the input into the models. A generally used measure of the goodness of fit of a linear model is R2, sometimes called the coefficient of determination. The coefficient of determination is that proportion of the total variability in the dependent variable that is accounted for by the regression equation. A value of R2 =1 indicates that the fitted equation accounts for all the variability of the values of the dependent variables in the sample data. At the other extreme, R2 =0 indicates that the regression equation explains none of the variability. The model performance was assessed by a comprehensive package of model accuracy prediction methods as suggested by Banerjee et al. (2011). A brief description of the applied performance measures is presented below. a. Root mean square error (RMSE): N
½SO2 ¼ a0 ðSO2 PBLÞ þ b0 þ a1 Par1 … þ an Parn
Mean square error MSE ¼ N
ð4Þ in which Par1…n represented meteorological parameters, including temperature (T), dew point (DP), relative humidity (RH), pressure (P), and wind speed (WS), and a1…n were regression coefficients for these meteorological factors. Pursuing the experiment, it was assumed that the dependent variables follow the normal distribution, homoscedasticity, i.e., the data have equal variance, and the differences between actual and theoretical values of dependent variables were independent (İçağa and Sabah 2009). We determined the regression coefficients and the intercept b0 via stepwise regression. Data from the period October 2004 to September 2011 were used to construct the two models. To verify the validity of the two regression models, they were applied to the period October 2011 to September 2012, being the data period
−1
2
∑ ½Oi −Pi i¼1
RMSE ¼
pffiffiffiffiffiffiffiffiffiffi MSE
b. Mean absolute error (MAE) MAE ¼ N −1
N X
jOi −Pi j
i¼1
c. Correlation (r): The formula used for the determination of the correlation of coefficient (r) has the form: " r¼
# " n #1=2 n n 2 X X . X 2 Oi −O Pi −P Oi −O Pi −P i¼1
i¼1
i¼1
where N is the number of data points, Oi is the observed surface SO2 concentrations, Pi is the model predicted
Environ Monit Assess
is the average value of observed concentration, O is the average value of predicted concentrations, and P concentration. All statistical analyses were conducted using the SPSS17.0 statistical software program. Model parameter estimates were presented with three decimal places or at least one significant digit.
Results and discussion Temporal variations of SO2_PBL and surface SO2 concentration Figure 2 shows the interannual variations of SO2_PBL column densities and surface SO2 concentration from 2005 to 2011 in Shanghai. Overall, the variation trends of both SO2_PBL and surface SO2 concentration were downward in the past 7 years. Surface SO2 concentration was the highest in 2005 with mean daily value of 60.870 μg m−3. It reduced to 50.826 μg m−3 in 2006, but increased in 2007 with the concentration of 55.407 μg m−3. This change was the result of the rapid increase of annual electricity consumption and the desulfurization program of Shanghai City. Coal is the main energy source in China and its combustion is the main cause of the atmospheric SO2 emissions in the country. The main coal consumption in Shanghai comes from coal-fired power plants. Shanghai started desulfurizing its coal-fired plants in 2005. After several consecutive years of increase, SO2 concentration finally declined in 2007. It has been reduced by half in the past 5 years and reached 29.181 μg m−3 in 2011. This showed that the desulfurization program for coal-fired plants had an
impact on overall SO2 emissions against rising electricity generation. By June 2009, Shanghai installed flue gas desulfurization devices for all the 10-GW capacities of coal-fired stations. Meanwhile, 695 MW of small and inefficient coal-fired plants in total had been shut down. Now with all the coal-fired power stations desulfurized, there are good reasons to believe that SO2 emissions will keep decreasing. However, the SO2_PBL column amount has different characteristics with the surface SO2 concentration. The highest column appeared in 2005 with a concentration of 1.294 DU, while the lowest column appeared in 2009 with a concentration of 0.973 DU. In the years 2010 and 2011, the concentrations of SO2_PBL column density were a little more than 1 DU. The monthly variations of both SO2_PBL column density and surface SO2 concentration in Shanghai have been plotted in Fig. 3 from October 2004 to September 2012 with their characteristic seasonal variations. As illustrated from the observed data, higher surface SO2 concentrations are experienced from the month of November and onward, which are particularly high in winters (December to February) in Shanghai. The reason is during wintertime the need for electricity is highest. Thus, SO2 pollution is higher in winter than in other seasons since anthropogenic emissions are expected to be higher because of heating. One-way analysis of variance (ANOVA) results show that SO2_PBL column density is significantly different between seasons. Among them, the SO2_PBL column density of winter is significantly different with the other three seasons (p
