DOI QR코드

DOI QR Code

Estimation of HCHO Column Using a Multiple Regression Method with OMI and MODIS Data

  • Hong, Hyunkee (Environmental Satellite Center, National Institute of Environmental Research: Department of Spatial Information Engineering, Pukyong National University) ;
  • Yang, Jiwon (Department of Spatial Information Engineering, Pukyong National University) ;
  • Kang, Hyeongwoo (Department of Spatial Information Engineering, Pukyong National University) ;
  • Kim, Daewon (Department of Spatial Information Engineering, Pukyong National University) ;
  • Lee, Hanlim (Department of Spatial Information Engineering, Pukyong National University)
  • Received : 2019.07.04
  • Accepted : 2019.07.31
  • Published : 2019.08.31

Abstract

We have estimated the vertical column density (VCD) of formaldehyde (HCHO) on a global scale using a multiple linear regression method (MRM) with Ozone Monitoring Instrument (OMI) and Moderate-Resolution Imaging Spectroradiometer (MODIS) data. HCHO VCDs were estimated in regions of biogenic, pyrogenic, and anthropogenic emissions using independent variables, including $NO_2$ VCD, land surface temperature (LST), an enhanced vegetation index (EVI), and the mean fire radiative power (MFRP), which are strongly correlated with HCHO. To evaluate the HCHO estimates obtained using the MRM, we compared estimates of HCHO VCD data measured by OMI ($HCHO_{OMI}$) with those estimated by multiple linear regression equations (MRE) ($HCHO_{MRE}$). Good MRM performances were found, having the average statistical values (R = 0.91, slope = 1.03, mean bias = $-0.12{\times}10^{15}molecules\;cm^{-2}$, percent difference = 11.27%) between $HCHO_{MRE}$ and $HCHO_{OMI}$ in our study regions where high HCHO levels are present. Our results demonstrate that the MRM can be a useful tool for estimating atmospheric HCHO levels.

Keywords

1. Introduction

Formaldehyde (HCHO), one of most abundant aldehyde species, is produced from photo-oxidation of volatile organic compounds (VOCs). HCHO plays an important role in the oxidation of VOCs, most of which are converted eventually to HCHO (Vrekoussis et al., 2010), which isthen released from biogenic, anthropogenic, and pyrogenic sources (De Smedt et al., 2010). Background levels of HCHO are generally determined by the amount of CH4, which exists in the troposphere at uniform concentrations on account ofitsstability (Heikes et al., 2001; Singh et al., 2001; Shim et al., 2005). Approximately 1600 Tg of HCHO is produced each year from the oxidation of CH4 globally (Stavrakou et al., 2009). In addition to VOC formation pathways, HCHO is also directly released into the atmosphere from biomass burning (Lee et al., 1997; Holzinger et al., 1999; Yokelson et al., 1999), fossil fuel combustion (Anderson et al., 1996; Geiger et al., 2002), and vegetation (Seco et al., 2007). However, such primary emissions of HCHO account for less than 1% of the global total. The presence of HCHO induces changes in tropospheric ozone levels, and the compound also plays an important role as a sink of the hydroxyl radical (OH) (Chance et al., 2000;Marais et al., 2014).Moreover, Formaldehyde has adverse effects on health, causing headaches and respiratory diseases; HCHO is also a known carcinogen (Xu et al., 2007).

To date,several environmentalsatellite sensors,such asthe Global Ozone Monitoring Experiment (GOME) sensor on board the European Remote Sensing-2 (ERS-2) satellite (e.g., Thomas et al., 1998; Chance et al., 2000; Wittrock et al., 2006), the SCanning Imaging Absorption SpectroMeter for Atmospheric CHartographY(SCIAMACHY) on board the Environment Satellite (ENVISAT)(De Smedt et al., 2008), the Ozone Monitoring Instrument (OMI) on board the Aura satellite (Gonzalez Abad et al., 2015) and the AtmosphericChemistryExperiment-FourierTransform Spectrometer (ACE-FTS) on board Science Satellite (SCISAT-1)(Dufour et al., 2009), have beenmonitoring the spatiotemporal distribution of atmospheric HCHO and its emission at regional and globalscalessince the mid-1990s. In addition, many studies have examined the distribution and origin of atmospheric HCHO and its precursors using the data obtained by such sensors. Abbot et al. (2003) compared HCHO vertical column densities (VCDs) obtained by the GOME sensor and VCDs simulated by the Goddard Earth Observing System-Chem Model (GEOS-Chem) over North America. Martin et al. (2004) validated HCHO values obtained by the GOME sensor in comparisons with data obtained by in situ measurements over the Southeastern United States.Asimilarinvestigation was also carried out over Asia. Witte et al. (2010) reported on characteristics of the HCHO/NO2 ratio in China using OMI observations. Diurnal variations in HCHO were revealed from Multi Axis-Differential Optical AbsorptionSpectroscopy (MAX-DOAS)measurements during the 2006 Pearl River Delta Regions Campaign (PRIDE-PRD2006) in China (Li et al., 2013). De Smedt et al. (2010) investigated temporal trends in the tropospheric HCHO column in Asia using SCIAMACHY data. Similar studies have also been conducted to understand long-term trends in HCHO distributions on a global scale (e.g., De Smedt et al., 2010; Vrekoussis et al., 2010). Several studies (e.g., Shim et al., 2005; Marais et al., 2014) have examined the concentration distributions of HCHO using satellitebased observation data. Moreover, efforts have been made to estimate HCHO concentrations and its emission utilizing chemical transport models, such as GEOS-Chem, Community Multi-scale Air Quality Model(CMAQ), the Model of Emissions of Gases and Aerosol from Nature (MEGAN).

The multiple line arregression method (MRM)is one of the statistical methodsfor estimating concentrations of trace gas and aerosols. The MRM was used with Moderate-Resolution Imaging Spectroradiometer (MODIS) aerosol optical thickness (AOT) data to estimate the concentration of particulate matter less than 2.5 μm in size (PM2.5) (Gupta et al., 2009). Kim et al. (2012) calculated surface concentrations of primary organic carbon and secondary organic carbon using the MRM with in situ measurement data, and Abdul-Wahab et al. (2005) used the MRM to estimate ozone concentrations. The MRM based on a simple least square fitting procedure generally provides reliable estimations, with the variance depending on the characteristics of the independent and dependent variable data used in in the multiple linear regression equation (MRE). Choi et al. (2015) estimated HCHO column in major cities over Asia using MRM with satellite data. However, to date, no study has attempted to estimate HCHO levels using the MRM on a global scale.

In this present study, we have, for the first time, estimated the HCHO VCD on a global scale using the MRM with OMI and MODIS data obtained in regions with high HCHO levels. Our aim was also to assessthe high reliability of MRE-derived HCHO results, by comparing HCHO column data estimated by MREs (HCHOMRE) with those measured by OMI(HCHOOMI).

2. Method and data

1) Selection of data and variables for MRM

The MRM was used to estimate the spatial distribution of HCHO VCD using MREs consisting of a dependent and multiple independent variables. Regression coefficients, which determine the nature of the relationship, can be solved by the least squares fitting method (Timm, 2002). In the present study, HCHOVCD wasthe dependent variable. Since HCHO originatesfrom primary sources and secondary formation related to biogenic, pyrogenic and anthropogenic sources, we selected independent variable candidates that represent each source type. Factorsthat are known to influence biogenic HCHO emissionsinclude the leaf area index (LAI), land surface temperature (LST), leaf age, and photosynthetic active radiation (PAR) (Geron et al., 2000; Palmer et al., 2006). The LST influences the growth rates and productivity of plants, while the EVIreflects canopy cover, land cover, and the LAI(Liu et al., 1995; Gao et al., 1996; Matsushita et al., 2007). Therefore, in this present study, the enhanced vegetation index (EVI) and LST were selected as the independent variable candidates influencing biogenic HCHO levels. LAI, however, is eventually excluded fromthe independent variable candidates due to its high value of variation inflation factor (VIF) against EVI. The mean fire radiative power (MFRP) and NO2 VCD was selected as the independent variable candidates influencing pyrogenic HCHO levels. According to Barkley et al. (2009), biomass burning is a significant source of HCHO which implies that the stronger biomass burning can lead to themore HCHO produced. Thus MFRPwasselected independent variable candidate to reflect HCHO in biomass burning areas. NO2 was also selected because biomass burning is amajorsource of NO2 (Andreae and Merlet, 2001). NO2 was also selected asthe independent variable candidate influencing anthropogenic HCHO levels since NO2 is a known trace gas closely related to fossil fuel combustion and population density (Lamsal et al., 2013). High levels of nitrogen oxides (NOx) can lead to active oxidation ofisoprene which is one ofthe precursor ofthe HCHO in physicochemical point of view (Trainer et al., 1987). Furthermore, the accuracy of OMI NO2 have relatively reliable even forthe poormeasurement conditions(e.g., high solar zenith angle and viewing zenith angle)since it uses visible wavelengthsintensity isstrongerthan UV wavelengths,so that the MRM method may be helpful to compensate the operational OMI HCHO retrievals. Although CO and CO2 levels are also related to fossil fuel combustion, they were not used in the study on account ofthe large discrepancy between theirlifetimes and that of HCHO.

The OMI and MODIS data were used as sources of data forthe independent and dependent variablesin the MRE. HCHO VCD, which is the dependent variable, is obtained from OMI observations(HCHOOMI)forthe period from January 2005 to December 2008; the dates after December 2008 were constrained on account of a degradation problem with the OMI charge-coupled device (CCD)(http://disc.sci.gsfc.nasa.gov/Aura/dataholdings/OMI/index.shtml#info). Data for the other independent variables were also obtained for the same period as that of the HCHOOMI data. The HCHOOMI data were obtained from the OMI Formaldehyde Level2G Global binned data (OMHCHOG, v003).The HCHOOMI used in this present study are those with cloud free (cloud fraction < 0.3) and those flagged as “0” which means good quality level. NO2 VCDs were obtained from the OMI Level-3 Global Gridded Total and Tropospheric NO2 VCD Data Product (NO2OMI) (OMNO2d).The MODIS instrument on board theTerra and Aqua satellites, launched in 2000 and 2002, respectively, was in operation throughout the study period, providing daily products including LST, EVI, MFRP, etc.(Justice et al., 2002).The LSTwas obtained from the MODIS LST(LSTMODIS) dataset(MOD11C3) and the EVI was obtained from the MODIS EVI (EVIMODIS) dataset (MOD13C2). The MFRP was obtained from the MODIS MFRP(MFRPMODIS) dataset (MYD14CHM).

In this study, two criteria were applied in the selection of areasfor estimating the HCHO VCD using the MRM; a) an absolute correlation coefficient (|R|) between HCHOOMI and at least two of the independent variable candidates greater than 0.4; b) a monthly maximum value of HCHOOMI within the top 30% of the global monthly average (i.e., > 2 × 1016 molecules cm-2 ), as these areas are high in HCHO and are likely to show seasonal HCHO cycles.

The pyrogenic regions, where biomass burning is a dominant source of HCHO, and where correlations between HCHOOMI and both NO2OMI and MFRPMODIS are high, include the Amazon (Barkley et al., 2008), SouthBorneo Island,IndochinaPeninsula, andSouthern Mexico. The regions, where oxidation of VOCs (such as isoprene) provides the dominant source of HCHO, include the Southeast United States, the correlation between HCHOOMI and both LSTMODIS and EVIMODIS is high, whereas no significant correlation is observed between HCHOOMI andMFRPMODIS.The regions, where high anthropogenic emissions of HCHO associated with fossil fuel combustion and population density is high, include Beijing, Nanjing, Seoul, NIIT, and New York.In areas of high anthropogenic emissions, annual mean NO2OMI levels are greaterthan 1 × 1016 molecules cm-2 and high negative correlations are observed between HCHOOMI and NO2OMI; however, no correlation is observed between HCHOOMI and MFRPMODIS in these areas.

2) Study areas

To classify the regionsinto biogenic, pyrogenic, and anthropogenic source types, we investigated global distributions of the correlation coefficient (R) obtained from linear regressions between monthly means of HCHOOMI and MFRPMODIS, EVIMODIS, LSTMODIS, and NO2OMI,from2005 to 2008 (Fig. 1(a)-(d),respectively). Correlation coefficients in regions where satellite data are missing, or where data were observed at large solar zenith angles (SZAs), were excluded from Fig. 1. The negative correlations observed in high-latitude (~60°) regions can be attributed to large uncertainties in HCHOOMI and NO2OMI values related to the large SZA (Fig. 1(d)). Fig. 1(a) shows positive correlations between HCHOOMI and MFRPMODIS in high-biomass regions (0.4-0.8), such as the Amazon [20°S-0°N, 50°W-70°W], Indochina [12°N-17°N, 100°E-108°E], and south Borneo Island [15°S-3°N, 110°E-120°E]. Fig. 1(b)shows negative correlations betweenHCHOOMI and EVIMODIS (R = -0.6) in Africa and the Amazon, where biomass burning is a dominant source of HCHO. The average correlation between HCHOOMI and EVIMODIS in most temperate forests and grasslands in the Southeast United States [30°N-42°N, 80°W100°W] and East Asia [25°N-42°N, 112°E-140°E] is ~0.65. The spatial distribution of correlations between HCHOOMI and LSTMODIS values (Fig. 1(c)) is similar to that between HCHOOMI and EVIMODIS (Fig. 1(b)); however, a positive correlation is observed in theAfrica and Amazon regions (Fig. 1(c)). Fig. 1(d) shows that correlations between HCHOOMI and NO2OMI of -0.55 to -0.85 occur in the Southeastern United States, the industrialized area in the eastern coastal region of China, the Northern Italy Industrial Triangle (NIIT; Turin, Milan, and Genoa), and in northeast Asian megacities, such as Beijing and Seoul.

OGCSBN_2019_v35n4_503_f0001.png 이미지

Fig. 1. Global distribution of the correlation coefficients (R) of linear regressions between HCHOOMI and (a) MFRPMODIS, (b) EVIMODIS, (c) LSTMODIS, and (d) NO2OMI , from 2005 to 2008.

3) Multiple regression method

The MRE can be defined as following equation:

\(Y = a_0 + a_1 X_1 +a_2 X_2 \cdots + a_n X_n + \varepsilon\)        (1)

Here, Y is dependent variable (HCHOMRE), a0 is constant coefficient, X1, X2, …, Xn are the independent variables(NO2 OMI,LSTMODIS,EVIMOIDS, andMFRPMODIS), a1, a2, …, an are the regression coefficient, and ε is the difference between observations (HCHOOMI) and estimated values(HCHOMRE).The regression coefficients can be estimated by the leastsquare fitting (Equation 2).

\(\sum {^m _{j=1} \varepsilon _j ^2} = \sum {^m _{j=1}(y_j -\widehat y_j )^2}\)        (2)

Where yj is observed value with m numbers of data points. By minimizing the sum of ε2, regression coefficients can be derived. Among the four independent variable candidates described in Section 2.1, to select optimal independent variables used in the MREs, two criteria were applied: variation inflation factor (VIF) and p-value. First, we examined theVIF that quantifies themulticollinearity of an independent variable candidate with regard to other independent variable candidates. TheVIF ofthe j-th independent variable is expressed as:

\(VIF(x_j) ={1 \over 1-R_j^2}\)        (3)

Where Rj2 is the R-squared value for the regression of xj against the other covariates(a regression that does not involve the dependent variable j).TheVIF indicates how much xj is correlated with the variables. A candidate for independent variable with a very high VIFcan be considered redundant and should be removed fromtheMRE.The candidatesforindependent variables, which do not satisfy the criterion VIF > 10 (Kutner et al., 2004), were excluded from the independent variables. We also used p-value to select independent variables. The significance level is set to 0.05 (5%) (Sellke et al., 2001). Among the independent variable candidatesthatsatisfy the VIF criterion, those that also satisfy the p-value less than 0.05 (p-value < 0.05) are selected asfinal independent variablesin the MRE.The independent variables and regression coefficients determined by least square fitting for each area are shown in Table 1.

Table 1. Coefficients of the multiple regression equations (MRE) used to predict HCHO (HCHOMRE), and coefficient of determination (R2) for the linear regressions between HCHOOMI and HCHOMRE, for different regions

OGCSBN_2019_v35n4_503_t0001.png 이미지

4) Monthly characteristics of the variables used in MRE

Temporal tendencies and amplitudes between the dependent variable (HCHOOMI) and the independent variable candidates (NO2OMI, LSTMODIS, EVIMODIS, and MFRPMODIS) are important for obtaining optimal regression coefficients for the MRE. To understand how temporal trends and amplitudes HCHOOMI and the other independent variable candidates depend on biogenic, pyrogenic, and anthropogenic source areas, we investigated the temporal characteristics ofHCHOOMI and the other independent variable candidates that are strongly correlated with HCHOOMI in different geographic areas (see Fig. 2, 3 and 4)

OGCSBN_2019_v35n4_503_f0002.png 이미지

Fig. 2. Time series of HCHOOMI, LSTMODIS, NO2OMI, and MFRPMODIS (see text for an explanation of these variables) in different pyrogenic source regions of HCHO: (a) Amazon, (b) south Borneo, (c) south Mexico, (d) Indochina.

OGCSBN_2019_v35n4_503_f0003.png 이미지

Fig. 3. Time series of HCHOOMI, NO2OMI, EVIMODIS, and LSTMODIS (see text for an explanation of these variables) in biogenic source regions of HCHO: southeast United States.

OGCSBN_2019_v35n4_503_f0004.png 이미지

Fig. 4. Time series of HCHOOMI, LSTMODIS, EVIMODIS, and NO2OMI (see text for an explanation of these variables) in different anthropogenic source regions of HCHO: (a) Seoul, (b) Beijing, (c) Nanjing, (d) Atlanta, (e) New York, and (f) NIIT.

(1) Biomass burning (pyrogenic) regions

Fig. 2 showsthe time series of HCHOOMI, LSTMODIS, NO2OMI, and MFRPMODIS from 2005 to 2008 in the pyrogenic regions: (a) Amazon, (b) south Borneo, (c) south Mexico, and (d) Indochina. In general, the temporal cycles and amplitudes of the variables are in good agreement, which implies, for example, that both HCHOOMI and NO2OMI are emitted from periodic biomass burning eventsin savanna and tropicalforests, as has also been reported in previous studies (Barkley et al., 2008; Marais et al., 2012). Monthly average MFRPMODIS values in the Amazon in September 2008 were 30% lowerthan those in 2005 and 2006 (Fig. 2(a)). Similarly, monthly averages of HCHOOMI (2.7 × 1016 molecules cm-2 ) and NO2 OMI (5.0 × 1015 molecules cm-2) in September 2008 were 80% of those in 2005 and 70% ofthose in 2006,respectively.The MFRPMODIS values observed during the biomass burning period from July to October in south Borneo were 30% higherin 2006 than in 2005, 2007, and 2008 (Fig. 2(b)). During the biomass burning period in 2006, monthly average of HCHOOMI was 4.0 × 1016 molecules cm-2; this value is also approximately 60% higher than those in 2005, 2007, and 2008. Monthly average of NO2OMI was 5.7 × 1015 molecules cm-2 ; this value is also approximately 45% higher than those in 2005, 2007, and 2008. Temporal characteristics of HCHOOMI in central Africa were different from those in other pyrogenic regions.In centralAfrica,seasonal cycles of NO2OMI and MFRPMODIS were similar to those in other pyrogenic areas; however, the trend of decreased HCHOOMI in January corresponding tomaximumlevels of MFRPMODIS is unique to this region. The causes of this unique trend in HCHOOMI levels in January have not been so far examined, and this subject requires further investigations.

(2) Biogenic regions

Fig. 3 shows time series of HCHOOMI, NO2OMI, EVIMODIS, and LSTMODIS in biogenic regions: the Southeastern United States. Levels of HCHOOMI, LSTMODIS, and EVIMODIS tend to increase in summer and are strongly correlated with one another. In the southeast United States, HCHO levels are reported to originate from biogenic sources, such as deciduous forests, woodland regeneration, CH4 oxidation, and industrial emissions (Shim et al., 2005; Millet et al., 2008).Annual cycles of HCHOOMI are similar to those of both LSTMODIS and EVIMODIS, which implies that biogenic sources are dominant in this area.

(3) Anthropogenic regions

Fig.4showstimeseriesofHCHOOMI,NO2OMI,EVIMODIS, and LSTMODIS in anthropogenic regions: Seoul,Beijing, Nanjing, Atlanta, New York, and NIIT. Fig. 5 shows correlations between HCHOOMI and NO2OMI in the representative anthropogenic areas of Seoul, NewYork, and NIIT from 2005 to 2008. Levels of HCHOOMI are strongly correlated with EVIMODIS and LSTMODIS in the representative anthropogenic areas, asshown in Fig. 4. The similarity of temporal cycles of EVIMODIS and LSTMODIS can be attributed to the locationsin temperate climate regionsin the Northern Hemisphere. However, the similarity of seasonal cycles of HCHOOMI with those of EVIMODIS does not necessarily mean that HCHO is affected by EVI, but that the two are both related to increased secondary formation of HCHO fromanthropogenicVOCs under conditions of enhanced solar radiation. The negative correlation between HCHOOMI and NO2OMI in biogenic areasismuch smaller than that in anthropogenic areas. The correlation between HCHOOMI and NO2OMI is-0.66 in the southeast United States, and is -0.53 and -0.37 in biogenic regions such as the states of Alabama and Georgia, respectively. However, high negative correlations between HCHOOMI and NO2OMI occurin anthropogenic areas in NorthAmerica such as New York andAtlanta (-0.77 and -0.84,respectively).Similarly,strong negative correlations between HCHOOMI and NO2OMI occur in Seoul and NIIT(-0.76 and -0.79,respectively)(Fig. 5). Levels of HCHOOMI and NO2OMI are strongly correlated in anthropogenic areas, as HCHO and NO2 exhibit both opposite seasonal cycles and also large-amplitude variations.

OGCSBN_2019_v35n4_503_f0005.png 이미지

Fig. 5. Correlation between HCHOOMI and HCHONO2 at (a) Seoul, (b) New York, and (c) NIIT between January 2005 and December 2008. Black lines are regression lines calculated using two measurement datasets.

3. Results

1) Determination of the MREs

Monthly average values of NO2OMI, LSTMODIS, EVIMODIS, andMFRPMODIS were used asthe independent variable candidates in the MRE, and that of HCHOOMI was used as the only dependent variable. All monthly datasetsfrom 2005 through 2008 (N = 48), which were used to determine MREs.Regression coefficients were determined using a linear least squares fitting method (Timm, 2002). The NO2OMI variable was categorized into two types depending on the dominant NO2 source type in each area: NO2_BB was used to represent the NO2 in pyrogenic-dominant regions, such as the Amazon, south Borneo, south Mexico, and Indochina, whereas NO2_FF was used to represent the NO2 in anthropogenic 0.84, indicating that HCHOMRE is in good agreement with HCHOOMI.In pyrogenic-dominantregionssuch as Amazon, SouthBorneo, South Mexico, and Indochina, the best estimates of HCHO were obtained when NO2_BB, LSTMODIS were used as independent variables in the MRE; note that MFRPMODIS was excluded due to a high p-value from the final MREs in several pyrogenic regions (see Table 1). The R2 between HCHOMRE and HCHOOMI represents 0.69. In biogenicdominant regions, the southeast United States, the best HCHO estimate was achieved when EVIMODIS were used asindependent variables(R2 = 0.84)(seeTable 1). In anthropogenic-dominant regions, the best estimates of HCHO were obtained when NO2_FF, EVIMODIS, and LSTMODIS were used as independent variables in the MRE. The R2 between HCHOMRE and HCHOOMI is 0.62. 

Fig. 6(a) shows the slopes and correlation coefficient obtained fromthe linearregressions between HCHOOMI and HCHOMRE. The correlation coefficient (R) and the slopes are found to be up to 0.97 and 0.94,respectively, showing very good agreements between HCHOOMI and HCHOMRE. Fig. 6(b) presents root mean square error (RMSE) and percent difference obtained from the linear regressions between HCHOOMI and HCHOMRE. The RMSE and percent difference are found to be up to 2.5 and 25%, respectively, showing good performances of MRM. The mean bias values are found to be either close to orslightly smaller than zero in Fig. 6(c), implying negligible bias of HCHOMRE against HCHOOMI.

OGCSBN_2019_v35n4_503_f0006.png 이미지

Fig. 6. Slopes and correlation coefficient (a), root mean square error (RMSE) and percent difference (b), and mean bias error and mean absolute error (c) obtained from the linear regressions between HCHOOMI and HCHOMRE.

4. Conclusions

To estimate HCHO levelsin pyrogenic regions,such as the Amazon, south Borneo, south Mexico, and Indochina, the independent variables NO2_BB, LSTMODIS, and MFRPMODIS were used in MREs. In biogenic regions such as the Southeastern United States, LSTMODIS, and EVIMODIS were used as the independent variables in the MRE. In anthropogenic regions, such as Seoul, Beijing, Nanjing, New York, Atlanta, and NIIT, NO2_FF, LSTMODIS, and EVIMODIS were used as the independent variables. The HCHOMRE values were in good agreement with HCHOOMI values in most regions.

Acknowledgements

This work was supported by a Research Grant of Pukyong National University (2017 year).

References

  1. Abbot, D. S., P. I. Palmer, R. V. Martin, K. V. Chance, D. J. Jacob, and A. Guenther, 2003. Seasonal and interannual variability of North American isoprene emissions as determined by formaldehyde column measurements from space, Geophysical Research Letters, 30(17).
  2. Abdul-Wahab, S. A., C. S. Bakheit, and S. M. Al-Alawi, 2005. Principal component and multiple regression analysis in modelling of ground-level ozone and factors affecting its concentrations, Environmental Modelling & Software, 20(10): 1263-1271. https://doi.org/10.1016/j.envsoft.2004.09.001
  3. Anderson, L. G., J. A. Lanning, R. Barrell, J. Miyagishima, R. H. Jones, and P. Wolfe, 1996. Sources and sinks of formaldehyde and acetaldehyde: An analysis of Denver's ambient concentration data, Atmospheric Environment, 30(12): 2113-2123. https://doi.org/10.1016/1352-2310(95)00175-1
  4. Andreae, M. O., and P. Merlet, 2001. Emission of trace gases and aerosols from biomass burning, Global Biogeochemical Cycles, 15(4): 955-966. https://doi.org/10.1029/2000GB001382
  5. Barkley, M. P., P. I. Palmer, U. Kuhn, J. Kesselmeier, K. Chance, T. P. Kurosu, R. V. Martin, D. Helmig, and A. Guenther, 2008. Net ecosystem fluxes of isoprene over tropical South America inferred from Global Ozone Monitoring Experiment (GOME) observations of HCHO columns, Journal of Geophysical Research: Atmospheres, 113(D20).
  6. Bey, I., D. J. Jacob, R. M. Yantosca, J. A. Logan, B. D. Field, A. M. Fiore, Q. Li, H. Y. Liu, L. J. Mickley, and M. G. Schultz, 2001. Global modeling of tropospheric chemistry with assimilated meteorology: Model description and evaluation, Journal of Geophysical Research: Atmospheres, 106(D19): 23073-23095. https://doi.org/10.1029/2001JD000807
  7. Boeke, N. L., J. D. Marshall, S. Alvarez, K. V. Chance, A. Fried, T. P. Kurosu, B. Rappengluck, D. Richter, J. Walega, and P. Weibring, 2011. Formaldehyde columns from the Ozone Monitoring Instrument: Urban versus background levels and evaluation using aircraft data and a global model, Journal of Geophysical Research: Atmospheres, 116(D5).
  8. Chance, K., P. I. Palmer, R. J. D. Spurr, R. V. Martin, T. P. Kurosu, and D. J. Jacob, 2000. Satellite observations of formaldehyde over North America from GOME, Geophysical Research Letters, 27(21): 3461-3464. https://doi.org/10.1029/2000GL011857
  9. Choi, W., H. Hong, J. Park, and H. Lee, 2015. First-time estimation of HCHO column in major cities over Asia using multiple regression with satellite data, Korean Journal of Remote Sensing, 31(6): 523-530 (in Korean with English abstract). https://doi.org/10.7780/kjrs.2015.31.6.3
  10. De Smedt, I., J. F. Muller, T. Stavrakou, R. Van Der A, H. Eskes, and M. Van Roozendael, 2008. Twelve years of global observations of formaldehyde in the troposphere using GOME and SCIAMACHY sensors, Atmospheric Chemistry and Physics, 8: 4947-4963. https://doi.org/10.5194/acp-8-4947-2008
  11. De Smedt, I., T. Stavrakou, J. F. Muller, R. Van Der A, and M. Van Roozendael, 2010. Trend detection in satellite observations of formaldehyde tropospheric columns, Geophysical Research Letters, 37(18).
  12. Dufour, G., S. Szopa, M. P. Barkley, C. D. Boone, A. Perrin, P. I. Palmer, and P. F. Bernath, 2009. Global upper-tropospheric formaldehyde: seasonal cycles observed by the ACE-FTS satellite instrument, Atmospheric Chemistry and Physics, 9(12): 3893-3910. https://doi.org/10.5194/acp-9-3893-2009
  13. Fu, T. M., D. J. Jacob, P. I. Palmer, K. Chance, Y. X. Wang, B. Barletta, D. R. Blake, J. C. Stanton, and M. J. Pilling, 2007. Space-based formaldehyde measurements as constraints on volatile organic compound emissions in east and south Asia and implications for ozone, Journal of Geophysical Research: Atmospheres, 112(D6).
  14. Geiger, H., J. Kleffmann, and P. Wiesen, 2002. Smog chamber studies on the influence of diesel exhaust on photosmog formation, Atmospheric Environment, 36(11): 1737-1747. https://doi.org/10.1016/S1352-2310(02)00175-9
  15. Geron, C., A. Guenther, T. Sharkey, and R. R. Arnts, 2000. Temporal variability in basal isoprene emission factor, Tree Physiology, 20(12): 799-805. https://doi.org/10.1093/treephys/20.12.799
  16. Gonzalez Abad, G., X. Liu, K. Chance, H. Wang, T. P. Kurosu, and R. Suleiman, 2015. Updated Smithsonian Astrophysical Observatory Ozone Monitoring Instrument (SAO OMI) formaldehyde retrieval, Atmospheric Measurement Techniques, 8(1): 19-32. https://doi.org/10.5194/amt-8-19-2015
  17. Gupta, P. and S. A. Christopher, 2009. Particulate matter air quality assessment using integrated surface, satellite and meteorological products: Multiple regression approach, Journal of Geophysical Research: Atmospheres, 114(D14).
  18. Heikes, B., J. Snow, P. Egli, D. O'Sullivan, J. Crawford, J. Olson, G. Chen, D. Davis, N. Blake, and D. Blake, 2001. Formaldehyde over the central Pacific during PEM-Tropics B, Journal of Geophysical Research: Atmospheres, 106(D23): 32717-32731. https://doi.org/10.1029/2001JD900012
  19. Holzinger, R., C. Warneke, A. Hansel, A. Jordan, W. Lindinger, D. H. Scharffe, G. Schade, and P. J. Crutzen, 1999. Biomass burning as a source of formaldehyde, acetaldehyde, methanol, acetone, acetonitrile, and hydrogen cyanide, Geophysical Research Letters, 26(8): 1161-1164. https://doi.org/10.1029/1999GL900156
  20. Jeong, J. I. and R. J. Park, 2013. Effects of the meteorological variability on regional air quality in East Asia, Atmospheric Environment, 69: 46-55. https://doi.org/10.1016/j.atmosenv.2012.11.061
  21. Justice, C., J. Townshend, E. Vermote, E. Masuoka, R. Wolfe, N. Saleous, D. Roy, and J. Morisette, 2002. An overview of MODIS Land data processing and product status, Remote Sensing of Environment, 83(1-2): 3-15. https://doi.org/10.1016/S0034-4257(02)00084-6
  22. Kim, W., H. Lee, J. Kim, U. Jeong, and J. Kweon, 2012. Estimation of seasonal diurnal variations in primary and secondary organic carbon concentrations in the urban atmosphere: EC tracer and multiple regression approaches, Atmospheric Environment, 56: 101-108. https://doi.org/10.1016/j.atmosenv.2012.03.076
  23. Lee, M., B. G. Heikes, D. J. Jacob, G. Sachse, and B. Anderson, 1997. Hydrogen peroxide, organic hydroperoxide, and formaldehyde as primary pollutants from biomass burning, Journal of Geophysical Research: Atmospheres, 102(D1): 1301-1309. https://doi.org/10.1029/96JD01709
  24. Li, X., T. Brauers, A. Hofzumahaus, K. Lu, Y. Li, M. Shao, T. Wagner, and A. Wahner, 2013. MAX-DOAS measurements of $NO_2$, HCHO and CHOCHO at a rural site in Southern China, Atmospheric Chemistry and Physics, 13(4): 2133-2151. https://doi.org/10.5194/acp-13-2133-2013
  25. Marais, E. A., D. J. Jacob, T. Kurosu, K. Chance, J. Murphy, C. Reeves, G. Mills, S. Casadio, D. Millet, and M. P. Barkley, 2012. Isoprene emissions in Africa inferred from OMI observations of formaldehyde columns, Atmospheric Chemistry and Physics, 12(14): 6219-6235. https://doi.org/10.5194/acp-12-6219-2012
  26. Marais, E. A., D. Jacob, A. Guenther, K. Chance, T. Kurosu, J. Murphy, C. Reeves, and H. Pye, 2014. Improved model of isoprene emissions in Africa using OMI satellite observations of formaldehyde: implications for oxidants and particulate matter, Atmospheric Chemistry and Physics Discussions, 14(5): 6951-6979. https://doi.org/10.5194/acpd-14-6951-2014
  27. Martin, R., D. Parrish, T. Ryerson, D. Nicks, K. Chance, T. Kurosu, D. Jacob, E. Sturges, A. Fried, and B. Wert, 2004. Evaluation of GOME satellite measurements of tropospheric $NO_2$ and HCHO using regional data from aircraft campaigns in the southeastern United States, Journal of Geophysical Research: Atmospheres, 109(D24).
  28. Matsushita, B., W. Yang, J. Chen, Y. Onda, and G. Qiu, 2007. Sensitivity of the enhanced vegetation index (EVI) and normalized difference vegetation index (NDVI) to topographic effects: a case study in high-density cypress forest, Sensors, 7(11): 2636-2651. https://doi.org/10.3390/s7112636
  29. Michael H. K., C. J. Nachtsheim, J. Neter, and W. Li, 2005. Applied Linear Regression Models (Vol. 5), McGraw-Hill/Irwin, New York, NY, USA.
  30. Lamsal, L., R. Martin, D. D. Parrish, and N. A. Krotkov, 2013. Scaling relationship for $NO_2$ pollution and urban population size: A satellite perspective, Environmental Science & Technology, 47(14): 7855-7861. https://doi.org/10.1021/es400744g
  31. Millet, D. B., D. J. Jacob, K. F. Boersma, T. M. Fu, T. P. Kurosu, K. Chance, C. L. Heald, and A. Guenther, 2008. Spatial distribution of isoprene emissions from North America derived from formaldehyde column measurements by the OMI satellite sensor, Journal of Geophysical Research: Atmospheres, 113(D2).
  32. Palmer, P. I., D. S. Abbot, T. M. Fu, D. J. Jacob, K. Chance, T. P. Kurosu, A. Guenther, C. Wiedinmyer, J. C. Stanton, and M. J. Pilling, 2006. Quantifying the seasonal and interannual variability of North American isoprene emissions using satellite observations of the formaldehyde column, Journal of Geophysical Research: Atmospheres, 111(D12).
  33. Park, R. J., D. J. Jacob, B. D. Field, R. M. Yantosca, and M. Chin, 2004. Natural and transboundary pollution influences on sulfate-nitrate-ammonium aerosols in the United States: Implications for policy, Journal of Geophysical Research: Atmospheres, 109(D15).
  34. Qing Liu, H. and A. Huete, 1995. A feedback based modification of the NDVI to minimize canopy background and atmospheric noise, IEEE Transactions on Geoscience and Remote Sensing, 33(2): 457-465. https://doi.org/10.1109/TGRS.1995.8746027
  35. Seco, R., J. Penuelas, and I. Filella, 2007. Short-chain oxygenated VOCs: Emission and uptake by plants and atmospheric sources, sinks and concentrations, Atmospheric Environment, 41(12): 2477-2499. https://doi.org/10.1016/j.atmosenv.2006.11.029
  36. Sellke, T., M. J. Bayarri, and J. O. Berger, 2001. Calibration of ${\rho}$ values for testing precise null hypotheses, The American Statistician, 55(1): 62-71. https://doi.org/10.1198/000313001300339950
  37. Shim, C., Y. Wang, Y. Choi, P. I. Palmer, D. S. Abbot, and K. Chance, 2005. Constraining global isoprene emissions with Global Ozone Monitoring Experiment (GOME) formaldehyde column measurements, Journal of Geophysical Research: Atmospheres, 110(D24).
  38. Singh, H., L. Salas, R. Chatfield, E. Czech, A. Fried, J. Walega, M. Evans, B. Field, D. Jacob, and D. Blake, 2004. Analysis of the atmospheric distribution, sources, and sinks of oxygenated volatile organic chemicals based on measurements over the Pacific during TRACE-P, Journal of Geophysical Research: Atmospheres, 109(D15).
  39. Stavrakou, T., J.-F. Muller, I. D. Smedt, M. V. Roozendael, G. van der Werf, L. Giglio, and A. Guenther, 2009. Evaluating the performance of pyrogenic and biogenic emission inventories against one decade of space-based formaldehyde columns, Atmospheric Chemistry and Physics, 9(3): 1037-1060. https://doi.org/10.5194/acp-9-1037-2009
  40. Timm, N. H., 2002. Applied Multivariate Analysis: Springer Texts in Statistics, Springer-Verlag New York, NY, USA.
  41. Vrekoussis, M., F. Wittrock, A. Richter, and J. Burrows, 2010. GOME-2 observations of oxygenated VOCs: what can we learn from the ratio glyoxal to formaldehyde on a global scale?, Atmospheric Chemistry and Physics, 10(21): 10145-10160. https://doi.org/10.5194/acp-10-10145-2010
  42. Witte, J. C., B. N. Duncan, A. R. Douglass, T. P. Kurosu, K. Chance, and C. Retscher, 2011. The unique OMI HCHO/$NO_2$ feature during the 2008 Beijing Olympics: Implications for ozone production sensitivity, Atmospheric Environment, 45(18): 3103-3111. https://doi.org/10.1016/j.atmosenv.2011.03.015
  43. Xu, J., X. Jia, X. Lou, G. Xi, J. Han, and Q. Gao, 2007. Selective detection of HCHO gas using mixed oxides of ZnO/ZnSnO3, Sensors and Actuators B: Chemical, 120(2): 694-699. https://doi.org/10.1016/j.snb.2006.03.033
  44. Yokelson, R. J., J. G. Goode, D. E. Ward, R. A. Susott, R. E. Babbitt, D. D. Wade, I. Bertschi, D. W. Griffith, and W. M. Hao, 1999. Emissions of formaldehyde, acetic acid, methanol, and other trace gases from biomass fires in North Carolina measured by airborne Fourier transform infrared spectroscopy, Journal of Geophysical Research: Atmospheres, 104(D23): 30109-30125. https://doi.org/10.1029/1999JD900817