Journal Search Engine
Search Advanced Search Adode Reader(link)
Download PDF Export Citaion korean bibliography PMC previewer
ISSN : 1598-5504(Print)
ISSN : 2383-8272(Online)
Journal of Agriculture & Life Science Vol.53 No.6 pp.1-11

Correlation between Urban Forest and Satellite-borne Imagery-based Ambient Particulate Matter across Seoul, South Korea

Sang-Hoon Lee1*, Jin-Cheol Park2
1Graduate School of Urban Studies, Hanyang University, Seoul, 04763, Korea
2Hanyang University, Seoul, 04763, Korea
Corresponding author: Sang-Hoon Lee Tel: +82-2-2220-0278 Fax: +82-2-2220-1214 E-mail:
September 30, 2019 ; November 5, 2019 ; November 13, 2019


In South Korea, Particulate matter (PM) has become one of the major threats to public health and safety across the country. Urban forests have been considered as a possible contributor to mitigate the air pollutants in urban areas. However, there is lack of research on investigating the role of urban forests on mitigating PM. This study investigated on the relationship between urban forests and PM concentrations in Seoul, South Korea, by using urban forest data, PM measurements, satellite imagery, and meteorological data. The correlation between the size of urban forests and PM measurements within three concentric buffers of 1 km, 500 m, and 300 m in radius were analyzed. Overall PM10 and PM2.5 concentrations varied significantly with different seasons during the 2-year study period. Overall PM10 and PM2.5 concentrations tended to be reduced as the urban forest size increased. This tendency became less noticeable as smaller urban forest patches were predominant over larger patches in the buffers. Season-specific models were developed by using 30m-resolution satellite imageries of Landsat 8 and meteorological parameters for estimating PM concentrations. No noticeable correlations were found between the modeled PM concentrations and the Urban forest size showing the ualves of Pearson’s coefficient r of 0.08 to 0.23 for PM10 and -0.16 to 0.04 for PM2.5. In this study, the variations in PM measurements with the presence of high urban forests within buffers were investigated. Overall PM10 and PM2.5 concentrations were lower along the domains encompassing higher urban forests in elevation.



    There has been a growing global concern on efforts for mitigating ambient air pollution mainly due to its negative impacts on the increased risks to public health and safety (IPCC, 2014). Particulate matter (PM) such as PM10 and PM2.5 has been recently highlighted as one of the major air pollutants being increasingly found in many regions (WHO, 2016). Meanwhile, the possible mitigation efforts have been still perceived as challenging as rapid urban growth is expected in many small- to medium sized cities and megacities over the world (IPCC, 2014).

    With regard to the concerns in urban air quality, several studies have suggested that air pollution can possibly be mitigated with the presence of urban forests, which can absorb or capture fine air pollutants (Cavanagh et al., 2009;Dzierżanowski et al., 2011;Nowak et al., 2018). However, questions are still remaining on the effective way of reducing PM due the complexity of PM mitigation mechanisms that can be affected by a variety of environmental factors in proximity of urban forests such as tree species and density, ventilation-friendly forest structure, and meteorological factors (Cavanagh et al., 2009;Salmond et al., 2013;Suder & Szymanowski, 2014;Nowak et al., 2018). Accordingly, the individual size of urban forest that is the assemblage of trees could be considered as one of the environmental factors.

    This study was aimed at investigating positive effects of urban forests on reducing ambient air pollutants along urban areas. This study examined the spatiotemporal changes in PM10 and PM2.5 concentrations compared to different sizes of urban forests, by analyzing the correlation between PM10 and PM2.5 measurements and the sizes of the urban forests in Seoul in the use of the satellite-based models for estimating PM concentrations.

    Materials and Methods

    1 Study area and PM measurements

    This study focused on Seoul, the most populated metropolis in South Korea (approximately 600 km2) (KOSIS, 2017). In this study, the land use is largely divided into densely distributed urban surfaces, urban forests and open green spaces (Fig. 1). The city of Seoul consists of 25 administrative districts, of which the ambient air qualities including PM10 and PM2.5 are monitored by the metropolitan air quality network. The network provides the calibrated datasets of the measurements which include hourly mean concentrations of PM10 and PM2.5 in the city since 2014 (Fig. 1) (AirKorea, 2019). In order to identify the most recent trend of the spatiotemporal changes in ambient air quality across the study area, the observation data of PM10 and PM2.5 concentrations measured from August 2017 to July 2019 were collected (

    2 Urban forest data

    In order to identify the relationship between urban forest size and PM measurements along the areas close by the forests, the spatial information of the urban forests including sizes and locations was collected from Seoul metropolitan government (2019). The sizes of the urban forests neighboring the monitoring stations were computed and classified into two groups by size. Each domain was defined by given three distances in radius: 1 km, 500 m, and 300 m; the domain was called ‘buffer’ in this study.

    3 Satellite data

    There are many areas without the monitoring station, which might possibly lead public to have misperception in surrounding air quality. To cover such areas, remotely sensed data from Landsat 8 OLI (Landsat 8) were used for interpreting regional atmospheric information. Standard Landsat 8 Level 1 product (LC08_L1XX) retrieved from August 2017 to July 2019 was obtained from USGS online archive ( The acquisition time of images available for the study area was approximately 02:00 UTC (11:00 in local time). Based on multispectral band data (at band 2, 3, 4, 5, 6, and 7), radiometric correction, simple dark object subtraction, and atmospheric correction were conducted to obtain the estimated Aerosol Optical Depth (AOD) values over the study area based on the algorithms used in earlier aerosol studies (Sobrino et al., 2004;Nadzri et al., 2010;Saleh & Hasan, 2014). The software of ENVI version 5.1 (Exelis Visual Information Solutions, 2013) and ArcMap version 10.2 (ESRI, 2013) were used for pre-processing of the images.

    4 Meteorological data

    In order to examine the meteorological influences on the PM concentration, wind speed (m/s), wind direction (0~360º), temperature (ºC), and relative humidity (%) from August 2017 to July 2019 were collected from 25 automatic weather stations (AWS) located within each district in Seoul. The observation data were obtained from the online archive provided by Korea Meteorological Administration (KMA, 2019).

    5 Statistical analysis

    To examine the effect of urban forest size on PM concentrations, PM measurements were compared to the total sizes of individual urban forests located within a radius of 1 km, 500 m, and 300 m from the each centers of the monitoring stations. The sizes of urban forests neighboring each monitoring station were calculated by using the software ArcMap 10.2 (ESRI, 2013). Hourly mean PM10 and PM2.5 concentrations measured at each monitoring station during the period from 2017 to 2018 were computed into quarterly mean PM10 and PM2.5 concentrations to examine seasonal changes in PM concentrations and to be compared to the forest size. In this study, the season was designated as the period from January to March, April to June, July to September, and October to December.

    To develop multiple linear regression models, Landsat 8-derived AOD, hourly mean PM measurements, and hourly mean measurements of meteorological parameters were used. Among the images collected for the study period, a season-specific (spring, summer, and fall) images were selected based on the acquisition time to develop single-day models for estimating PM concentrations. Among 25 monitoring stations as sample sites, 20 sites were selected as training data for the numerical model and the other five samples were selected as validation data for the cross-validation of the model. The model-derived PM10 and PM2.5 estimations were compared to PM10 and PM2.5 measurements. Correlation between the model-derived PM estimations and PM measurements was analyzed on the same spots as the monitoring station sites within the PM estimated map. All modeling was performed by using the software SPSS Statistics version 25.0 (IBM Inc., 2017).

    In the rasterized map of PM estimations, correlation between PM estimations and urban forest size were also statistically analyzed on the spots of the monitoring sites. The PM estimations were computed from the mean raster values within each buffer by using ‘raster calculator’ in ArcMap version 10.2 (ESRI, 2013).


    By computing the size of urban forests, total 12 sample sites were selected from the entire rankings. The 12 sample sites were divided into two groups: one group with six sites containing the largest forests within given buffers (L1-L6), and the other with six sites of the smallest forests (S1-S6). The distribution of urban forests and the ranks by the forest size are summarized (Fig. 2; Table 1). In L1 to L6, largesized forest masses were dominant and tended to cut across the given buffers whereas small-sized and fragmented forest patches were distributed sporadically nearby the monitoring stations (Fig. 2). In S1 to S6, while small-sized and fragmented forest patches were dominant, no large-sized forest mass cut across the given buffers (Fig. 2). As the buffer distances decreased, small-sized fragments of forest patches became more dominant in the urban areas followed by the reduced coverages of the large forests (Fig. 2). While the overall PM10 concentrations were slightly lowered with the increased urban forest sizes within 1 km and 500 m buffers, there was no noticeable trend followed by different forest sizes within 300 m buffers (Fig. 3). The overall decrease in PM2.5 concentrations with the increased forest sizes was more noticeable (Fig. 3). This trend was found only in the domains within 1 km buffers; however, no noticeable trend was found in the domains within 500 m and 300 m buffers. These trends might be derived from the locations of the monitoring sites which, in fact, were designed to measure the air quality close by human neighborhoods.

    The relationships between the sizes of urban forests and PM10 and 2.5 concentrations in the time series and by buffer are plotted in Fig. 3. Regardless of the radius of buffer, the mean PM10 concentrations showed a significant decrease at the third quarter and then a significant increase at the fourth quarter of the period. There was no significant difference in PM10 concentrations between two groups. The mean PM2.5 concentrations generally showed a significant decrease at the second and third quarter, and then a significant increase at the fourth quarter except for the concentration between the third and fourth quarter in the group of small urban forest which increase was not significant, whereas there was no significant difference in PM2.5 concentrations between two groups (Fig. 3).

    The regression models derived from Landsat 8 data and meteorological data, the model factors, and the model reliability are summarized in Table 2. The PM10 models indicated r2 value of 0.57 in spring and 0.54 in fall. Marginally strong correlations were found between the modelled and measured PM10 concentrations, showing r value of 0.71 in spring and 0.73 in fall (Table 2). PM2.5 models presented Person’s correlation coefficient of 0.54 in spring and 0.71 in fall. Correlations between the modelled and measured PM2.5 concentrations, indicating the coefficient of 0.52 in spring and 0.33 in fall (Table 2).

    In the selection of the season-specific Landsat-derived models, winter was excluded to avoid the resting season of vegetation. Based on the correlations between the estimated PM concentrations and the sizes of the urban forests, the best fit lines are plotted in Fig. 4 and 5. The correlations between the estimated PM10 and PM2.5 concentrations showed r values ranging from 0.08 to 0.23 for PM10 and from -0.16 to 0.04 for PM2.5. There was no noticeable seasonal trend based on the observation days (Fig. 4, 5). The smaller pixel size of the imagery might not necessarily indicate the more accurate interpretation of the atmospheric parameters since the fine spatial resolution may cause it to be more susceptible to radiometric noise and contamination on the bright urban surfaces, resulting in biased image products (Munchak et al., 2013;Remer et al., 2013). This result suggested that the Landsat 8-derived model requires further improvement in interpreting the atmospheric information in detail.


    In this study, expecting that PM concentrations would be lower in proximity of larger forest masses, the correlations between seasonal PM10 and PM2.5 concentrations and urban forest sizes were analyzed. The result showed the tendency that both the PM10 and PM2.5 concentrations were lowered with the presence of larger urban forest masses (Fig. 2, 3). With regard to the PM mitigation effect of urban forests, it was found in this study that there might be a threshold of urban forest in area that could be effective for reducing PM concentrations. A similar question was suggested in Cavanagh et al.’s study (2009), presenting reduced PM concentrations with increased buffer distance from the inner forest patches to the outer forest edge. Since the investigation was designed to examine the gradients of PM concentrations measured within each of the forest patches. However, the question about the effective threshold size of forests still remains unanswered.

    In terms of the overall differences in PM concentrations with different particle sizes (Fig. 3), the result in this study falls into the findings by Dzierżanowski et al. (2011). The amount of the particles captured on tree leaf surfaces varied with the size of the particles; significantly smaller quantity of PM was found in the form of fine particles (0.2-2.5μm) than that of coarser particles (2.5-10μm). A variety of conditions across the urban environment should be examined with increased number of the monitoring sites for better understanding of the relationship between PM and urban forests.

    Such complexity of examining the PM mitigation effects of urban forests was suggested in Nowak et al.’s study (2018). There were some positive estimates of air purification due to the aerosol capturing ability of individual urban forests, while the result indicated a necessity in further investigations on possible effects of urban forests for PM mitigation in the collaboration with other environmental factors. Several studies suggested that urban forests do not necessarily depend only on absorbing or intercepting air pollutants by tree leaves. Rather, a greater reduce in air pollutant concentrations were expected when other factors supported the urban forests such as the healthiness of vegetation, optimal forest structure and density of trees for better ventilation (Salmond et al., 2013;Nowak et al., 2018). This suggestion also falls into the findings from Suder & Szymanowski’s study (2014). The presence, location and elevation of hills and mountains have been used as factors that affect the distribution of urban ventilation corridors, which can aerodynamically contribute to the dispersion of air mass stagnated within urban areas.

    Based on these findings, the capability of topographical factors was tested in this study to understand the characteristics of the individual urban forests and the subsequent changes in PM concentrations. It was considered that mountainous forest could provide urban ventilation due to the elevation gradient. The monitoring sites were ranked based on the maximum elevation within a 1 km buffer, and were categorized into two groups: one for high urban forests of five, and the other for low forests; some sites that are too close to each other were excluded in the analysis (Fig. 6). Overall quarterly mean PM concentrations were slightly higher in the group of low urban forests during the study period, whereas there was no significant difference in PM concentrations between two groups. There was the tendency that both PM10 and PM2.5 concentrations were lower across the sites with high urban forest in consistent manner (Fig. 7).

    This study suggested that analyzing the relationship between urban forest and PM concentration could be improved synergistically in the use of more favorable environmental factors, particularly dealing with detailed characteristics of urban forest. It is expected that this approach could be a useful indicator for creating and coordinating the urban forest, and for planning urban forest management strategies on the basis of scientific outcomes.


    This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (Ministry of Science and ICT) (No. NRF-2017R1C1B5017787).



    Study area in this study and the locations of ambient air quality monitoring stations and urban forests (shaded) (AirKorea, 2019;Seoul metropolitan government, 2019).


    Distribution of the urban forests located within a buffer of 1 km, 500 m, and 300 m in radius from each center of the monitoring stations located in Seoul; the domains that contain large forests (L1~L6) and small forests (S1~S6) in proximity are named in the order from the largest to smallest.


    Quarterly mean PM10 (a) and PM2.5 (b) concentrations and standard errors measured at each group of the domains (Small and Large) (n=6) during the observation period of 2017-2018; the groups were named based on the urban forests size within a buffer of 1 km, 500 m and 300 m in radius from each center of the domains.


    Correlation between the estimated mean PM10 concentrations and the sizes of the urban forest contained within a radius of 1 km, 500 m, and 300 m from the center of each monitoring station (n=22).


    Correlation between the estimated mean PM2.5 concentrations and the sizes of the urban forest contained within a radius of 1 km, 500 m, and 300 m from the center of each monitoring station (n=22).


    Site name and the maximum elevation (m) (AGL) within a 1 km radius from the center of each monitoring stations; the ranks are sorted based on the elevation of the largest to the smallest (data from NSDIP, 2019).


    Quarterly mean PM concentrations and standard errors measured at each group of the sites (High and Low) during the observation period of 2017-2018; the groups were named based on the site elevation (n=5).


    Domain names and urban forest sizes (ha) located within a buffer of 1 km, 500 m, and 300 m in radius from each center of the monitoring stations located in Seoul; the ranks are sorted by the forest size from the largest to smallest

    Season-specific single-day models for estimating PM concentrations based on AOD derived from high-resolution sensors (n=25)


    1. AirKorea.2019. Latest hourly ambient air quality data archive. (2019.8.6.).
    2. Cavanagh JE , Zawar-Reza P and Wilson G. 2009. Spatial attenuation of ambient particulate matter air pollution within an urbanised native forest patch. Urban Forestry and Urban Greening, 8 (1): 21-30.
    3. Dzierżanowski K , Popek R , Gawrońska H , Sæbø A and Gawroński S. 2011. Deposition of particulate matter of different size fractions on leaf surfaces and in waxes of urban forest species. International Journal of Phytoremediation, 13 (10): 1037-1046,
    4. ESRI.2013. ArcGIS Desktop: Release 10.2. Redlands, CA: Environmental Systems Research Institute.
    5. IPCC (Intergovernmental Panel on Climate Change).2014. Climate change 2014: Mitigation of climate change. Contribution of working group III to the fifth assessment report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA.
    6. KMA (Korean Meteorological Administration).2019. Meteorological dataset archive. Accessed in
    7. KOSIS (Korean Statistical Information Service).2018. Administrative division statistics. Accessed in (2019.9.12).
    8. Munchak LA , Levy RC , Mattoo S , Remer LA , Holben BN , Schafer JS , Hostetler CA and Ferrare RA. 2013. MODIS 3 km aerosol product: applications over land in an urban/suburban region, Atmospheric Measurement Techniques, 6: 1747-1759, (2013).
    9. Nadzri O , Mat MZ , Lim HS. 2010. Estimating particulate matter concentration over arid region using satellite remote sensing: A case study in Makkah, Saudi Arabia. Modern Applied Science. 4.
    10. Nowak DJ , Hirabayashi S , Doyle M , McGovern M and Pasher J. 2018. Air pollution removal by urban forests in Canada and its effect on air quality and human health. Urban Forestry and Urban Greening, 29: 40-48, ISSN 1618-8667,
    11. NSDIP (National Spatial Data Infrastructure Portal).2019. Korea National Geographic Information Institute data archive. Accessed in (2019.9.9.).
    12. Remer LA , Mattoo S , Levy RC and Munchak LA. 2013. MODIS 3 km aerosol product: Algorithm and global perspective. Atmospheric Measurement Techniques Discussions 6: 69-112.
    13. Saleh SAH, Hasan G.2014 Estimation of PM10 concentration using ground measurements and Landsat 8 OLI satellite image. Journal of Geophysics & Remote Sensing 3: 120.
    14. Salmond JA , Williams DE , Laing G , Kingham S , Dirks K , Longley I and Henshaw GS. 2013. The influence of vegetation on the horizontal and vertical distribution of pollutants in a street canyon. Science of the Total Environment. 443: 287-298.
    15. Seoul Metropolitan Government.2019. Seoul city green space geospatial information (datum: ITRF2000). Accessed in (2019.8.22).
    16. Sobrino J , Jiménez-Muñoz JC and Paolini L. 2004. Land surface temperature retrieval from LANDSAT TM 5. Remote Sensing of Environment. Elsevier, 90: 434-440.
    17. Suder A and Szymanowski M. 2014. Determination of ventilation channels in urban area: A case study of Wrocław (Poland). Pure and Applied Geophysics. Vol. 171, no. 6,
    18. WHO (World Health Organization).2016. Ambient air pollution: A global assessment of exposure and burden of disease. World Health Organization. Accessed in (2019.9.10.).
    오늘하루 팝업창 안보기 닫기