Skip to main content
NCBI home page
NASA Author Manuscripts logoLink to NASA Author Manuscripts
. Author manuscript; available in PMC: 2020 Feb 4.
Published in final edited form as: Remote Sens (Basel). 2019 Jan 10;11(2):10.3390/rs11020122. doi: 10.3390/rs11020122

Evaluation of Landsat-8 and Sentinel-2A aerosol optical depth retrievals across Chinese cities and implications for medium spatial resolution urban aerosol monitoring

Zhongbin Li a, David P Roy b, Hankui K Zhang a, Eric F Vermote c, Haiyan Huang a
PMCID: PMC6999735  NIHMSID: NIHMS1527447  PMID: 32021701

Abstract

In urban environments, aerosol distributions may change rapidly due to building and transport infrastructure and human population density variations. The recent availability of medium resolution Landsat-8 and Sentinel-2 satellite data provide the opportunity for aerosol optical depth (AOD) estimation at higher spatial resolution than provided by other satellites. A year of 30 m Landsat-8 and 10 m Sentinel-2A AOD data retrieved using the Land Surface Reflectance Code (LaSRC) were compared with coincident ground-based Aerosol Robotic Network (AERONET) Version 3 AOD data for 20 Chinese cities. Stringent selection criteria were used to select contemporaneous data – only satellite and AERONET data acquired within 10 minutes were considered. The average satellite retrieved AOD over a 1470 m × 1470 m window centered on each AERONET site was derived to capture fine scale urban AOD variations. AERONET Level 1.5 (cloud-screened) and Level 2.0 (cloud-screened and also quality assured) data were considered. For the 20 urban AERONET sites in 2016 there were 106 (Level 1.5) and 67 (Level 2.0) Landsat-8 AERONET AOD contemporaneous data pairs, and 118 (Level 1.5) and 89 (Level 2.0) Sentinel-2A AOD data pairs. The greatest AOD values (>1.5) occurred in Beijing, suggesting that the Chinese capital was one of the most polluted cities in China in 2016. The LaSRC Landsat-8 and Sentinel-2A AOD retrievals agreed well with the AERONET AOD data (linear regression slopes > 0.96; coefficient of determination r2 > 0.90; root mean square deviation < 0.175) and demonstrate that the LaSRC is an effective and applicable medium resolution AOD retrieval algorithm over urban environments. The Sentinel-2A AOD retrievals had better accuracy than the Landsat-8 AOD retrievals, which is consistent with previously published research. The implications of the research and the potential for urban aerosol monitoring by combining the freely available Landsat-8 and Sentinel-2 satellite data are discussed.

1.0. Introduction

In much of Asia, rapid economic growth has resulted in increased air pollution (Guo et al. 2014; Guan et al. 2014; Rohde and Muller 2015; Peng et al. 2016). In particular, the increased occurrence of urban haze, characterized by high concentrations of fine particulate matter with diameter less than 2.5 μm (PM2.5), has received growing attention due to human health and quality of life concerns (Gao et al. 2017; Lu et al. 2018). In situ air quality measurements provide high temporal resolution observations but can be difficult to extrapolate spatially. With wide spatial and repetitive coverage, satellite based remote sensing provides an effective way to quantify and map aerosol properties and in particular the aerosol optical depth (AOD), i.e., the amount of incoming solar radiation that is scattered and absorbed by aerosols at a given wavelength (Kaufman et al. 2002). Satellite based AOD products have been developed and used to estimate the concentration of fine particulate matter for epidemiological and air quality research (Wang and Christopher 2003; van Donkelaar et al. 2010; Kloog et al. 2014; Ma et al. 2016). However, satellite AOD products are defined at relatively coarse spatial resolution. For example, AOD products have been developed using data from Total Ozone Mapping Spectrometer (TOMS) at 1° resolution (Torres et al., 2002), Polarization and Directionality of Earth Reflectances (POLDER) at 1/6° resolution (Leroy et al. 1997), Multi-angle Imaging SpectroRadiometer (MISR) at 17.6 km resolution (Diner et al. 1999), MEdium Resolution Imaging Spectrometer (MERIS) at 10 km resolution (Mei et al. 2017), Visible Infrared Imaging Radiometer Suite (VIIRS) at 6 km resolution (Jackson et al. 2013), Himawari-8 Geostationary at 0.05° resolution (Kikuchi et al. 2018), and Moderate Resolution Imaging Spectroradiometer (MODIS) at 10 km, 3 km (Levy et al. 2013; Gupta et al. 2018) and 1 km resolution (Lyapustin et al. 2018). Coarse resolution products are less suitable for studying aerosols in urban environments where aerosol distributions may change at finer spatial scales due to factors including spatial variations in building and transport infrastructure and human population densities (Chudnovsky et al. 2013; Zhang and Cao, 2015; He et al. 2016).

The recently launched Landsat-8 and Sentinel-2 satellites can be used to derive AOD data at 30 m and 10 m resolution respectively and so have the potential for urban aerosol monitoring. The goals of this research are to study the accuracy of AOD retrievals from these satellites and provide insights into their suitability for medium spatial resolution urban aerosol monitoring. A year of Landsat-8 and Sentinel-2A 550 nm AOD data, retrieved using the recently published Land Surface Reflectance Code (LaSRC), were compared with coincident ground-based Aerosol Robotic Network (AERONET) AOD data at 20 Chinese cities. For fair comparison, the 500 nm AERONET AOD measurements were converted to 550 nm equivalent values using a standard Ångström parameter interpolation method. The LaSRC average AOD retrievals over 1470 m × 1470 m (i.e., 49 × 49 30 m Landsat-8 pixels and 147 × 147 10 m Sentinel-2A pixels) centered on the AERONET site locations were compared with the temporally closest AERONET AOD measured within 10 minutes of the satellite overpass. Differences between the LaSRC satellite AOD retrievals and AERONET AOD measurements were quantified statistically. The paper concludes with a discussion of the implications of this study for satellite based urban aerosol monitoring at moderate spatial resolution.

2.0. Data

2.1. AERONET data

The AERONET is a network of globally distributed ground-based sun and sky scanning radiometers that provide near-continuous daytime measurements of spectral solar irradiance, spectral aerosol optical thickness, water vapor, and inversion aerosol products (Holben et al. 1998; Dubovik et al. 2002). The AERONET data include spectral AOD in the visible to near infrared (typically from 340 nm or 440 nm, to 1020 nm or 1640 nm, depending on the site radiometer), Ångström parameters at different wavelengths, and column water vapor (g/cm2). The most recent Version 3.0 data (Giles et al. 2018) were obtained from the AERONET web site (http://aeronet.gsfc.nasa.gov/). The data are categorized into different quality levels. In this study the higher quality Level 1.5 (cloud-screened) and Level 2.0 (cloud-screened and also quality assured) data were used.

All the Version 3.0 AERONET data over mainland China, Hong Kong, and Taiwan, available in 2016, were considered. Two AERONET sites were removed as they were not in or close to urban areas. The majority of the remaining 26 sites had both Level 1.5 and Level 2.0 data and were in mainland China (Figure 1). Site differences in the annual amount of AERONET data and the processing level (Table 1) reflect cloud conditions, and AERONET site operational differences.

Figure 1.

Figure 1

Open in a new tab

Locations of 26 urban AERONET sites with Level 1.5 data (open circles, 11 sites) and with both Level 1.5 and Level 2.0 data (small solid circles, 15 sites) in 2016. The six sites shown in red were not used because they had no contemporaneous AERONET and satellite data after application of the stringent filtering criteria described in Section 3.2.

Table 1.

The 26 urban AERONET site (Figure 1) short names, geographic locations, and 2016 AERONET data availability. The coordinates of the Landsat-8 images (path/row) and Sentinel-2A tiles (tile ID) and 2016 data availability at each AERONET site are also summarized.

AERONET site (latitude, longitude) Number of days in 2016 with AERONET Level 1.5 (Level 2.0) data Landsat-8 Path/row Number of Landsat-8 images over the AERONET site in 2016 Sentinel-2A tile ID Number of Sentinel-2A images over the AERONET site in 2016
AOE_Baotou (40.852°N, 109.629°E) 173 (0) 127/32
128/31
128/32		 66 49TCF 56
Beijing (39.977°N, 116.381°E) 243 (243) 123/32 22 50TMK 59
Beijing-CAMS (39.933°N, 116.317°E) 299 (299) 123/32 22 50TMK 59
Beijing_PKU (39.992°N, 116.310°E) 147 (0) 123/32 22 50TMK 59
Beijing_RADI (40.005°N, 116.379°E) 224 (0) 123/32 22 50TMK 59
Hong_Kong_PolyU (22.303°N, 114.180°E) 53 (53) 121/45
122/45
122/44		 45 50QKK
49QHE		 122
Hong_Kong_Sheung (22.483°N, 114.117°E) 52 (52) 121/45
122/45
122/44		 49 50QKK
49QHE		 122
QOMS_CAS (28.365°N, 86.948°E) 198 (198) 140/40
140/41		 35 45RVM 21
SONET_Harbin (45.705°N, 126.614°E) 33 (0) 117/28
117/29
118/28		 46 52TCR
51TYL		 54
SONET_Hefei (31.905°N, 117.162°E) 23 (0) 121/38 17 50SNA 29
SONET_Nanjing (32.115°N, 118.957°E) 11 (0) 120/38 17 50SPA 33
SONET_Shanghai (31.284°N, 121.481°E) 16 (0) 118/38 17 51RUQ 62
SONET_Xingtai (37.182°N, 114.360°E) 22 (0) 124/34 20 50SKG
49SGB		 74
SONET_Zhoushan (29.994°N, 122.188°E) 13 (0) 118/39
117/39		 26 51RVP 23
Taihu (31.421°N, 120.215°E) 153 (129) 119/38 14 51RTQ
50RQV		 77
XiangHe (39.754°N, 116.962°E) 289 (289) 123/32
122/32
122/33		 60 50TMK 59
XuZhou-CUMT (34.217°N, 117.142°E) 258 (0) 121/36
121/37
122/36		 51 50SNC 29
Alishan (23.508°N, 120.813°E) 14 (14) 117/44
118/43
118/44		 58 51QTG
50QRM		 38
Bamboo (25.187°N, 121.535°E) 11 (11) 117/42
117/43		 27 51RUH 14
Cape_Fuguei Station (25.297°N, 121.538°E) 16 (0) 117/42
117/43		 27 51RUJ
51RUH		 28
Chen-Kung_Univ (23.000°N, 120.217°E) 137 (80) 118/44 20 51QTF
50QRL		 26
Chiayi (23.496°N, 120.496°E) 144 (144) 118/44
117/44
118/43		 39 51QTG
50QRM		 38
Douliu (23.712°N, 120.545°E) 19 (19) 117/44
118/43
118/44		 58 51QTG
50QRM		 38
EPA-NCU (24.968°N, 121.185°E) 201 (201) 117/43
118/42
118/43		 47 51RUH 14
Lulin (23.469°N, 120.874°E) 190 (190) 117/44 19 51QTG
51QTF
50QRM		 51
Taipei_CWB (25.030°N, 121.500°E) 188 (188) 117/43 16 51RUH 14
Open in a new tab

2.2. Landsat-8 and Sentinel-2A data

Landsat-8 was launched in 2013 and carries the optical wavelength Operational Land Imager (OLI) and the thermal wavelength Thermal Infrared Sensor (TIRS) (Roy et al. 2014a). The OLI is a 12 bit multispectral sensor providing eight reflective wavelength 30 m observations from 435 nm to 2294 nm (Loveland and Irons 2016). The OLI data are provided in approximately 185 × 180 km images defined in Worldwide Reference System (WRS) path/row coordinates in the Universal Transverse Mercator (UTM) projection. Landsat-8 has a 16-day revisit cycle and so nominally there are 22 or 23 OLI images over each path/row location per year. Only Landsat-8 Collection 1 Tier 1 data were used in this study as they are the highest quality data and are radiometrically calibrated and orthorectified (Dwyer et al. 2018). The image-to-image registration accuracy of the Landsat-8 Collection 1 Tier 1 scenes is ≤ 12 m radial root mean square error (Storey et al. 2014).

Sentinel-2A was launched in June 2015 and carries the optical wavelength multi-spectral instrument (MSI). The MSI is a 12 bit multispectral sensor with four 10 m, six 20 m and three 60 m reflective bands observations from 433 nm to 2280 nm (Drusch et al. 2012). Sentinel-2A has a 10-day revisit cycle and the data are provided in fixed 109 × 109 km spatially overlapping tiles in the UTM projection (Roy et al. 2016). The image-to-image registration accuracy of the data was 16 m before June 15th 2016 and was 4 m after this date (Yan et al. 2018) when the European Space Agency updated the processing software (Gascon et al. 2017).

All of the Landsat-8 OLI and Sentinel-2A MSI data available in 2016 over the 26 urban AERONET sites were obtained. Top of atmosphere (TOA) reflectance OLI images were obtained from the United States Geological Survey portal (https://earthexplorer.usgs.gov/) and TOA Sentinel-2A L1C tile products were obtained from the European Space Agency portal (https://scihub.copernicus.eu/dhus). This provided between 14 and 66 (mean = 20) Landsat-8 OLI images and between 14 and 122 (mean = 30) Sentinel-2A tile acquisitions at each AERONET site (Table 1). More Sentinel-2A acquisitions were available because of the higher 10-day satellite revisit cycle.

3.0. Methods

3.1. Landsat-8 and Sentinel-2A aerosol optical depth (AOD) retrieval

Reliable surface monitoring with optical wavelength remotely sensed data requires atmospheric correction to minimize the scattering and absorbing effects of atmospheric gases and aerosols. Radiative transfer algorithms and atmospheric characterization data are used for automated large-area atmospheric correction (Vermote et al. 2002; Roy et al. 2014b). The correction of aerosol effects is particularly challenging because aerosols are highly variable in space and time (Kaufman et al. 1997; Dubovik et al. 2002). In this study, the LaSRC (Land Surface Reflectance Code) V3.5.5 atmospheric correction algorithm (Vermote et al. 2016) was used. The LaSRC algorithm was developed for atmospheric correction of Landsat-8 imagery (Vermote et al. 2016) and has been adapted for Sentinel-2A application (Zhang et al. 2018; Doxani et al. 2018). It is based on the 6SV radiative transfer code (Kotchenova and Vermote 2007). The AOD retrieval takes advantage of the short wavelength red and blue bands on the Landsat-8 and Sentinel-2 sensors. Two reflectance ratios, red to blue and red to ultra-blue, and the difference between them are used to invert the AOD using a global ratio data set derived from MODIS and MISR data, and expressed as a function of a mid-infrared vegetation index (Vermote et al. 2016). A fixed urban-clean aerosol model (Dubovik et al. 2002) is assumed. Despite using a fixed aerosol model, the Ångström parameter (related to the dependence of AOD with wavelength) is not fixed (Vermote et al. 2016). This has a similar effect as allowing a dynamic aerosol model with varying aerosol size distribution and refractive index (Kaufman et al. 1997; Torres et al. 2002) and is needed for aerosol retrieval over regions with different aerosol types. The LaSRC AOD is defined at 550 nm for each 10 m (Sentinel-2A) and 30 m (Landsat-8) pixel. The LaSRC algorithm also generates a per-pixel cloud mask using the red-to-blue surface reflectance ratios (Vermote et al. 2016). The LaSRC 10 m (Sentinel-2A) and 30 m (Landsat-8) cloud masks were used in this study to remove cloud contaminated pixels.

3.2. Comparison of contemporaneous AERONET and satellite retrieved AOD

The ground-based AERONET 500 nm AOD measurements were compared with the LaSRC satellite retrieved AOD 550 nm data. First, the AERONET 500 nm AOD measurements were converted to 550 nm equivalent values using the standard Ångström parameter interpolation method (King et al. 1999) as:

τ(550) = τ(500)(500/550)α [1]

where τ(550) is the interpolated AERONET 550 nm AOD data, τ(500) is the AERONET 500 nm AOD measurement, and α is the AERONET 440–675 nm Ångström parameter. Then the following temporal and spatial data selection criteria were applied to select contemporaneous satellite and AERONET AOD data.

At each AERONET site the closest AERONET AOD 550 nm value within 10 minutes (before or after) of the satellite overpass time was selected, and the mean 550 nm LaSRC AOD over a 1470 m × 1470 m image window centered on the AERONET site location was derived. If more than half of the satellite AOD retrievals in the 1470 m × 1470 m window were cloudy or missing, the data were discarded. A 10 minute period was used because the median closest temporal difference between the satellite overpass times and the AERONET measurements for the 2016 data was 2.9 minutes (Sentinel-2A) and 4.1 minutes (Landsat-8), and the median second closest temporal difference was 7.7 minutes (Sentinel-2A) and 9.5 minutes (Landsat-8). In China, the annual urban mean wind speed is 2.4 m/s (Guo et al. 2011) and so in 10 minutes aerosols could be blown 1440 m which is smaller than the 1470 m window dimension. Using too large a window size will increase the likelihood of averaging aerosols from different sources, which is a concern in urban environments where aerosols may vary spatially quite rapidly.

4.0. Results

4.1. Contemporaneous data availability and example annual satellite and AERONET AOD data comparison

Table 2 summarizes the number of contemporaneous satellite and AERONET AOD data sets in 2016. Due to the stringent selection criteria (Section 3.2), six of the 26 AERONET sites had no remaining contemporaneous data. There were generally more contemporaneous Sentinel-2A than Landsat-8 data due to the greater Sentinel-2A temporal data availability. There were more Level 1.5 data than Level 2.0 AERONET data due to the additional quality assurance used to generate the Level 2.0 data (Giles et al. 2018). On average, among the 20 sites in 2016, there were 5.9 Level 1.5 and 4.45 Level 2.0 AERONET AOD and Sentinel-2A contemporaneous data pairs, and on average for Landsat-8, there were 5.3 Level 1.5 and 3.35 Level 2.0 AERONET AOD data pairs.

Table 2.

Summary of the 26 urban AERONET sites (Figure 1), the number of days with 2016 AERONET data, and the number of contemporaneous LaSRC satellite AOD and AERONET AOD data pairs.

AERONET site (latitude, longitude) Number of days in 2016 with AERONET Level 1.5 (Level 2.0) data Number of contemporaneous Landsat-8 and AERONET Level 1.5 (Level 2.0) AOD data pairs Number of contemporaneous Sentinel-2A and AERONET Level 1.5 (Level 2.0) AOD data pairs
AOE_Baotou (40.852°N, 109.629°E) 173 (0) 13 (0) 9 (0)
Beijing (39.977°N, 116.381°E) 243 (243) 11 (11) 22 (22)
Beijing-CAMS (39.933°N, 116.317°E) 299 (299) 12 (12) 29 (29)
Beijing_PKU (39.992°N, 116.310°E) 147 (0) 3 (0) 11 (0)
Beijing_RADI (40.005°N, 116.379°E) 224 (0) 4 (0) 3 (0)
Hong_Kong_PolyU (22.303°N, 114.180°E) 53 (53) 0 (0) 1 (1)
Hong_Kong_Sheung (22.483°N, 114.117°E) 52 (52) 3 (3) 0 (0)
QOMS_CAS (28.365°N, 86.948°E) 198 (198) 2 (2) 0 (0)
SONET_Harbin (45.705°N, 126.614°E) 33 (0) 2 (0) 0 (0)
SONET_Hefei (31.905°N, 117.162°E) 23 (0) 0 (0) 0 (0)
SONET_Nanjing (32.115°N, 118.957°E) 11 (0) 0 (0) 0 (0)
SONET_Shanghai (31.284°N, 121.481°E) 16 (0) 0 (0) 1 (0)
SONET_Xingtai (37.182°N, 114.360°E) 22 (0) 0 (0) 0 (0)
SONET_Zhoushan (29.994°N, 122.188°E) 13 (0) 2 (0) 0 (0)
Taihu (31.421°N, 120.215°E) 153 (129) 2 (2) 5 (5)
XiangHe (39.754°N, 116.962°E) 289 (289) 18 (18) 23 (23)
XuZhou-CUMT (34.217°N, 117.142°E) 258 (0) 15 (0) 4 (0)
Alishan (23.508°N, 120.813°E) 14 (14) 0 (0) 0 (0)
Bamboo (25.187°N, 121.535°E) 11 (11) 0 (0) 0 (0)
Cape_Fuguei_Station (25.297°N, 121.538°E) 16 (0) 0 (0) 0 (0)
Chen-Kung_Univ (23.000°N, 120.217°E) 137 (80) 2 (2) 1 (0)
Chiayi (23.496°N, 120.496°E) 144 (144) 5 (5) 2 (2)
Douliu (23.712°N, 120.545°E) 19 (19) 0 (0) 1 (1)
EPA-NCU (24.968°N, 121.185°E) 201 (201) 3 (3) 2 (2)
Lulin (23.469°N, 120.874°E) 190 (190) 4 (4) 0 (0)
Taipei_CWB (25.030°N, 121.500°E) 188 (188) 5 (5) 4 (4)
Open in a new tab

The Beijing-CAMS and XiangHe AERONET sites had the greatest number of contemporaneous satellite and AERONET data pairs. Figure 2 illustrates the AERONET AOD (black dots) and the contemporaneous Landsat-8 (blue circles) and Sentinel-2A (red circles) AOD for the Beijing-CAMS site. The vertical lines in Figure 2 are drawn to connect the satellite and AERONET AOD retrievals (by definition they occurred within 10 minutes) and illustrate that the Sentinel-2A and Landsat-8 sensors usually overpass on different days. On day 221 (August 8th) both Landsat-8 and Sentinel-2A overpassed the AERONET site on the same day, and the satellite and contemporaneous AERONET AOD data were less than 0.2 different. Specifically, the Landsat-8 AOD was 0.295 (sensed at 02:53:32.9 UTC) compared to the temporally closest AERONET AOD of 0.193 (02:51:11 UTC), and the Sentinel-2A AOD was 0.337 (02:59:14.1 UTC) compared to the temporally closest AERONET AOD of 0.143 (03:00:57 UTC).

Figure 2.

Figure 2

Open in a new tab

Beijing-CAMS site 2016 results showing the AERONET Level 2.0 AOD (41 small black dots), the Sentinel-2A (29 red filled circles) and the Landsat-8 (12 blue filled circles) AOD values. The vertical lines connect the satellite AOD retrievals with the contemporaneous AERONET AOD measurements (always within 10 minutes of each other). Satellite AOD values are mean non-cloudy values derived from a 1470 m × 1470 m window over the AERONET site. All AOD values are at 550 nm.

The biggest Beijing-CAMS site satellite AERONET AOD difference occurred on day 181 (June 29th) when the AERONET AOD was 1.117 (03:06:58 UTC) and the Sentinel-2A AOD was 1.948 (03:06:27.6 UTC). However, visual inspection of the day 181 Sentinel-2A 1470 m × 1470 m window revealed that this was due to residual cloud contamination (nearly one third of the Sentinel-2A pixels were cloud contaminated but only about half were labeled by the LaSRC algorithm as cloudy). Cloud contamination is a significant error source for satellite AOD retrieval and usually results in inflated AOD retrieval (Kaufman et al. 2005; Zhao et al. 2013). Examination of the other days with contemporaneous data in 2016 indicated no other cloud detection omission errors. In summary, for the illustrated year of Beijing-CAMS site data, discarding the day 181 cloud contaminated data, the root mean square deviation (RMSD) between the AERONET AOD and the satellite AOD was 0.183 for Landsat-8 (n=12) and 0.102 for Sentinel-2A (n=28). The correlation between the satellite AOD and the AERONET AOD for the year was 0.967 for Landsat-8 and 0.962 for Sentinel-2A.

4.2. Satellite and AERONET AOD comparison

Figure 3 shows scatterplots of the contemporaneous Landsat-8 and AERONET Level 1.5 (left) and Level 2.0 (right) AOD data over the urban sites for 2016. Three of the Landsat-8 overpasses had cloud detection omission errors and resulted in inflated Landsat-8 AOD values (filled circles); they occurred on April 8 (day 99) at the Taipei-CWB site, June 12 (day 164) at the QOMS_CAS site, and December 4 (day 339) at the Taipei-CWB site. The open circles show the remaining AOD pairs (n=103 for Level 1.5 and n=64 for Level 2.0) that were used for statistics analysis. The results are similar considering the Level 1.5 and Level 2.0 AERONET data. The ordinary least squares (OLS) linear regression lines (black) show a high level agreement (r2 > 0.9; regression slopes close to unity and > 0.96). The root mean square deviation (RMSD) between the Landsat-8 AOD and the AERONET AOD was 0.161 (Level 1.5) and 0.171 (Level 2). The Level 2 AOD provided higher RMSD likely because the errors in the satellite AOD retrievals are larger than the errors in the AERONET AOD measurements; any improvement of the quality assured Level 2.0 over the Level 1.5 AERONET AOD is marginal compared to the satellite AOD retrieval errors.

Figure 3.

Figure 3

Open in a new tab

Scatterplots of the Landsat-8 AOD against the contemporaneous AERONET Level 1.5 (left) and Level 2.0 (right) AOD data over the urban AERONET sites for 2016. The three filled circles are outliers due to Landsat-8 cloud detection omission errors and are not used in statistics analysis. The solid lines show ordinary least square regression lines. The dotted lines are 1:1 lines superimposed for reference.

Figure 4 shows scatterplots of the contemporaneous Sentinel-2A and AERONET AOD data over the urban sites for 2016. The open circles show the valid AOD pairs (n=116 for Level 1.5 and n=87 for Level 2.0) used for the statistical analysis. Two outliers (filled circles) were due to cloud detection omission errors in Sentinel-2A images acquired on June 29 (day 181) at the Beijing-CAMS site (also evident in Figure 2), and on August 9 (day 222) at the Taipei_CWB site. They are not considered in the statistical analysis. The Sentinel-2A AOD retrievals show high agreement (OLS regression slopes close to unity and > 0.96; r2 > 0.920; RMSD < 0.110) with both the AERONET Level 1.5 and Level 2.0 AOD data. Similar to the Landsat-8 results (Figure 3), there are no significant differences between the Level 1.5 (left) and Level 2.0 (right) results.

Figure 4.

Figure 4

Open in a new tab

Scatterplots of the Sentinel-2A AOD against the contemporaneous AERONET Level 1.5 (left) and Level 2.0 (right) AOD data over the urban AERONET sites for 2016. The three filled circles are outliers due to Sentinel-2A cloud detection omission errors and are not used in statistics analysis. The solid lines show ordinary least square regression lines. The dotted lines are 1:1 lines superimposed for reference.

Notably, more than 13% of the AERONET Level 2.0 AOD values were > 1.0 (indicating very hazy urban conditions). For five overpass dates the AERONET AOD values were > 1.5; all occurred in Beijing (at AERONET sites: Beijing, Beijing-CAMS, and XiangHe) in the months of March, June, July, August, and September. The greatest AERONET AOD value with contemporaneous non-cloudy Landsat-8 satellite data was 1.940 and occurred at the Beijing AERONET site on July 7th (day 189), and the greatest AERONET AOD value with contemporaneous non-cloudy Satellite-2A satellite data was 2.094 and occurred at the Beijing_PKU site on August 11th (day 224). On these two days the difference between the satellite and AERONET AOD values was <0.12. This, and the good agreements illustrated in Figures 3 and 4, indicate the ability of the LaSRC satellite AOD retrieval algorithm to work over a wide range of urban AOD values.

5.0. Conclusion and discussion

Stringent selection criteria were used to select contemporaneous medium resolution satellite and AERONET data located in, or close to, major urban conurbations, across China. In 2016 there were 26 urban AERONET sites, and after applying the spatial and temporal selection criteria 20 sites remained with an annual total of 106 (Level 1.5) and 67 (Level 2.0) Landsat-8 AERONET AOD contemporaneous data pairs, and 118 (Level 1.5) and 89 (Level 2.0) Sentinel-2A AOD data pairs. The resulting AERONET data encompassed a range of AOD from clear days (minimum = 0.008) to very hazy conditions (maximum = 2.094). The greatest AOD values (>1.5) occurred in Beijing, suggesting that the Chinese capital was one of the most polluted cities in China in 2016.

The Landsat-8 and Sentinel-2A AOD retrievals agreed well with the AERONET AOD measurements (OLS linear regression slopes > 0.96, r2 > 0.90, and RMSD < 0.175) with no significant pattern of over- or under-estimation of the AOD. Although the AERONET Level 2.0 AOD data are quality assured, there were no pronounced difference between the results considering the Level 1.5 and considering the Level 2.0 AOD data. The Sentinel-2A AOD retrievals had slightly better agreement (RMSD < 0.11) with the contemporaneous AERONET AOD measurements than the Landsat-8 retrievals (RMSD < 0.175). Recently, the LaSRC AOD algorithm was assessed, with seven other AOD algorithms, considering a year of Landsat-8 and seven months of Sentinel-2A overpasses at 19 globally distributed urban and non-urban AERONET sites using a ±15 minute and 9 km window dimension (Doxani et al. 2018). The Sentinel-2A LaSRC AOD performed the best among the different AOD algorithms (0.097 RMSD, n=48), and, as in this study, the Landsat-8 LaSRC AOD was less accurate (0.387 RMSD, n=119). The reasons for this sensor difference are unknown but are likely related to the higher spatial resolution of the Sentinel-2A compared to Landsat-8 and to sensor spectral band differences.

The selection criteria used in this study were purposefully stringent to capture fine scale Chinese urban aerosol variations. The closest AERONET AOD value within 10 minutes of each satellite overpass was selected, and the mean LaSRC AOD over a 1470 m × 1470 m image window centered on the AERONET site location was derived. A ±10 minute period was used to reflect the approximately 3 minute frequency of most AERONET data (Giles et al. 2018) and because the median closest and second closest temporal differences between the satellite overpass times and the AERONET measurements for the 2016 data were less than 10 minutes. A 1470 m × 1470 m image was used, as it corresponds to an integer multiple of the Landsat-8 30 m and the Sentinel-2A 10 m bands, and because in 10 minutes aerosols would not be transported completely across the window at the reported annual urban 2.4 m/s mean wind speed (Guo et al. 2011). We undertook a sensitivity analyses, considering 2970 m × 2970 m and 5970 m × 5970 m window dimensions for the same ±10 minute period, and found only small differences in the reported results (differences only in the second decimal place of the reported RMSD values).

Using large window sizes and long temporal periods as selection criteria will increase the likelihood of aerosols from different sources being compared between the AERONET and satellite AOD data. This is a concern in urban environments where aerosols may vary spatially quite rapidly. For example, Chen et al. (2015) considered a year of PM2.5, PM10, and greenhouse gas concentration data from 35 Beijing municipal environmental monitoring stations and found significant spatial differences between urban, suburban, and traffic sites, and significant diurnal variations at individual sites. Similarly, Wang et al. (2015) documented urban PM2.5 concentrations in the morning and evening that were nearly twice those in the mid-afternoon in northeast China. Previously, researchers have compared coarser spatial resolution but near daily MODIS AOD retrievals with AERONET AOD using temporal selection criteria of ±30 minutes and spatial window side dimensions varying from 9 km to 50 km (Ichoku et al. 2002; Chu et al. 2002; Levy et al. 2013; Lyapustin et al. 2011; Nichol and Bilal, 2016; Gupta et al. 2018) and ±7.5 minutes with a 10 km spatial window dimension (Xie et al. 2011). An explicit study of spatial scaling effects, comparing MODIS AOD data with PM2.5 measurements in the Boston metropolitan area (Chudnovsky et al. 2013), found reduced correlation with increasing window size. Further research to undertake scaling analyses, including wind speed information, using the Landsat-8 and Sentinel-2 AOD data is recommended, subject to the availability of data from a dense ground-based atmospheric monitoring network.

The current state-of-the-practice for global coverage satellite AOD monitoring is based on near daily, but coarse spatial resolution, polar-orbiting data, such as from MODIS, that may be spatially too coarse for monitoring urban pollution (Li et al. 2005; Chudnovsky et al. 2013; Hu et al. 2014). The Landsat-8 and Sentinel-2A satellites provide a higher spatial resolution (10 m to 30 m) global coverage AOD monitoring capability but with lower temporal resolution. For example, in this study there were an average of 20 Landsat-8 and 30 Sentinel-2A overpasses per year over the Chinese cities. Data from the recently launched (March 2017) Sentinel-2B were not available for the study period, but combined with Sentinel-2A and Landsat-8, will increase the temporal observation frequency to a global median average revisit interval of 2.9 days with different overpass times (Li and Roy, 2017). Given their high accuracy in AOD retrieval, future research to investigate the combined use of Landsat-8 and the two Sentinel-2 satellites for AOD monitoring is recommended.

Acknowledgments

This research was funded by the NASA Land Cover/Land Use Change Multi-Source Land Imaging Science Program Grant NNX15AK94G and by NASA Making Earth System Data Records for Use in Research Environments (MEaSUREs) program (cooperative agreement NNX13AJ24A). The European Space Agency and Copernicus program management and staff, and the USGS Landsat program management and staff, are thanked for the free provision of the Sentinel-2A and Landsat-8 data respectively.

References

  1. Che H, Zhang XY, Xia X, Goloub P, Holben B, Zhao H, Wang Y, Zhang XC, Wang H, Blarel L, Damiri B, 2015. Ground-based aerosol climatology of China: aerosol optical depths from the China Aerosol Remote Sensing Network (CARSNET) 2002–2013. Atmospheric Chemistry and Physics, 15(13), 7619–7652. [Google Scholar]
  2. Chen W, Tang H, Zhao H, 2015. Diurnal, weekly and monthly spatial variations of air pollutants and air quality of Beijing. Atmospheric Environment, 119, 21–34. [Google Scholar]
  3. Chu DA, Kaufman YJ, Ichoku C, Remer LA, Tanré D, Holben BN, 2002. Validation of MODIS aerosol optical depth retrieval over land. Geophysical research letters, 29(12), MOD2–1. [Google Scholar]
  4. Chudnovsky AA, Kostinski A, Lyapustin A, Koutrakis P, 2013. Spatial scales of pollution from variable resolution satellite imaging. Environmental Pollution, 172, 131–138. [DOI] [PubMed] [Google Scholar]
  5. Diner DJ, Abdou W, Ackerman T, Crean K, Gordon H, Kahn R, Martonchik J, McMuldroch S, Paradise S, Pinty B, Verstrete M, 1999. MISR Level 2 Aerosol Retrieval Algorithm Theoretical Basis, JPL D11400. [Google Scholar]
  6. Doxani G, Vermote E, Roger JC, Gascon F, Adriaensen S, Frantz D, Hagolle O, Hollstein A, Kirches G, Li F, Louis J, Mangin A, Pahlevan N, Pflug B, Vanhellemont Q, 2018. Atmospheric correction inter-comparison exercise. Remote Sensing, 10(2), 352. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Drusch M, Del Bello U, Carlier S, Colin O, Fernandez V, Gascon F, Hoersch B, Isola C, Laberinti P, Martimort P, Meygret A, 2012. Sentinel-2: ESA’s optical high-resolution mission for GMES operational services. Remote sensing of Environment, 120, 25–36. [Google Scholar]
  8. Dubovik O, Holben B, Eck TF, Smirnov A, Kaufman YJ, King MD, Tanré D, Slutsker I, 2002. Variability of absorption and optical properties of key aerosol types observed in worldwide locations. Journal of the atmospheric sciences, 59(3), 590–608. [Google Scholar]
  9. Dwyer JL, Roy DP, Sauer B, Jenkerson CB, Zhang HK, Lymburner L, 2018, Analysis Ready Data: Enabling analysis of the Landsat archive, Remote Sensing, 10(9), 1363. [Google Scholar]
  10. Gascon F, Bouzinac C, Thépaut O, Jung M, Francesconi B, Louis J, Lonjou V, Lafrance B, Massera S, Gaudel-Vacaresse A, Languille F, Alhammoud B, Viallefont F, Pflug B, Bieniarz J, Clerc S, Pessiot L, Trémas T, Cadau E, De Bonis R, Isola C, Martimort P, Fernandez V, 2017. Copernicus Sentinel-2A calibration and products validation status. Remote Sensing, 9(6), 584. [Google Scholar]
  11. Giles DM, Sinyuk A, Sorokin M, Schafer JS, Smirnov A, Slutsker I, Eck TF, Holben BN, Lewis J, Campbell J, Welton EJ, Korkin S, Lyapustin A, 2018. Advancements in the Aerosol Robotic Network AERONET) Version 3 Database - Automatic Quality Assurance of Sun Photometer Aerosol Optical Depth, Atmospheric Measurement Techniques, in preparation. [Google Scholar]
  12. Guan D, Su X, Zhang Q, Peters GP, Liu Z, Lei Y, He K, 2014. The socioeconomic drivers of China’s primary PM2. 5 emissions. Environmental Research Letters, 9(2), 024010. [Google Scholar]
  13. Guo H, Xu M, Hu Q, 2011. Changes in near-surface wind speed in China: 1969–2005. International Journal of Climatology, 31(3), 349–358. [Google Scholar]
  14. Guo S, Hu M, Zamora ML, Peng J, Shang D, Zheng J, Du Z, Wu Z, Shao M, Zeng L, Molina MJ, 2014. Elucidating severe urban haze formation in China. Proceedings of the National Academy of Sciences, 111(49), 17373–17378. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Gupta P, Remer LA, Levy RC, Mattoo S, 2018. Validation of MODIS 3 km land aerosol optical depth from NASA’s EOS Terra and Aqua missions. Atmospheric Measurement Techniques, 11(5), 3145. [Google Scholar]
  16. He J, Wu L, Mao H, Liu H, Jing B, Yu Y, Ren P, Feng C, Liu X, 2016. Development of a vehicle emission inventory with high temporal–spatial resolution based on NRT traffic data and its impact on air pollution in Beijing–Part 2: Impact of vehicle emission on urban air quality. Atmospheric Chemistry and Physics, 16(5), 3171–3184. [Google Scholar]
  17. Hu X, Waller LA, Lyapustin A, Wang Y, Al-Hamdan MZ, Crosson WL, Estes MG Jr, Estes SM, Quattrochi DA, Puttaswamy SJ, Liu Y, 2014. Estimating ground-level PM2. 5 concentrations in the Southeastern United States using MAIAC AOD retrievals and a two-stage model. Remote Sensing of Environment, 140, 220–232. [Google Scholar]
  18. Holben BN, Eck TF, Slutsker I, Tanre D, Buis JP, Setzer A, Vermote E, Reagan JA, Kaufman YJ, Nakajima T, Lavenu F, 1998. AERONET—A federated instrument network and data archive for aerosol characterization. Remote sensing of environment, 66(1), 1–16. [Google Scholar]
  19. Ichoku C, Chu DA, Mattoo S, Kaufman YJ, Remer LA, Tanré D, Slutsker I and Holben BN, 2002. A spatio-temporal approach for global validation and analysis of MODIS aerosol products. Geophysical Research Letters, 29(12), MOD1–1. [Google Scholar]
  20. Jackson JM, Liu H, Laszlo I, Kondragunta S, Remer LA, Huang J, Huang HC, 2013. Suomi-NPP VIIRS aerosol algorithms and data products. Journal of Geophysical Research: Atmospheres, 118(22), 12–673. [Google Scholar]
  21. Kaufman YJ, Remer LA, Tanré D, Li RR, Kleidman R, Mattoo S, Levy RC, Eck TF, Holben BN, Ichoku C, Martins JV, 2005. A critical examination of the residual cloud contamination and diurnal sampling effects on MODIS estimates of aerosol over ocean. IEEE Transactions on Geoscience and Remote Sensing, 43(12), 2886–2897. [Google Scholar]
  22. Kaufman YJ, Tanré D, Boucher O, 2002. A satellite view of aerosols in the climate system. Nature, 419(6903), 215. [DOI] [PubMed] [Google Scholar]
  23. Kaufman YJ, Tanré D, Remer LA, Vermote EF, Chu A, Holben BN, 1997. Operational remote sensing of tropospheric aerosol over land from EOS moderate resolution imaging spectroradiometer. Journal of Geophysical Research: Atmospheres, 102(D14), 17051–17067. [Google Scholar]
  24. Kikuchi M, Murakami H, Suzuki K, Nagao TM, Higurashi A, 2018. Improved Hourly Estimates of Aerosol Optical Thickness Using Spatiotemporal Variability Derived From Himawari-8 Geostationary Satellite. IEEE Transactions on Geoscience and Remote Sensing, 56(6), 3442–3455. [Google Scholar]
  25. King MD, Kaufman YJ, Tanré D, Nakajima T, 1999. Remote sensing of tropospheric aerosols from space: Past, present, and future. Bulletin of the American Meteorological society, 80(11), 2229–2260. [Google Scholar]
  26. Kloog I, Chudnovsky AA, Just AC, Nordio F, Koutrakis P, Coull BA, Lyapustin A, Wang Y, Schwartz J, 2014. A new hybrid spatio-temporal model for estimating daily multi-year PM2.5 concentrations across northeastern USA using high resolution aerosol optical depth data. Atmospheric Environment, 95, 581–590. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Kotchenova SY, Vermote EF, 2007. Validation of a vector version of the 6S radiative transfer code for atmospheric correction of satellite data. Part II. Homogeneous Lambertian and anisotropic surfaces. Applied optics, 46(20), 4455–4464. [DOI] [PubMed] [Google Scholar]
  28. Leroy M, Deuzé JL, Bréon FM, Hautecoeur O, Herman M, Buriez JC, Tanré D, Bouffies S, Chazette P, Roujean JL, 1997. Retrieval of atmospheric properties and surface bidirectional reflectances over land from POLDER/ADEOS. Journal of Geophysical Research: Atmospheres, 102(D14), 17023–17037. [Google Scholar]
  29. Levy RC, Mattoo S, Munchak LA, Remer LA, Sayer AM, Patadia F, Hsu NC, 2013. The Collection 6 MODIS aerosol products over land and ocean. Atmospheric Measurement Techniques, 6(11), 2989. [Google Scholar]
  30. Li C, Lau A, Mao J, Chu D, 2005. Retrieval, validation, and application of the 1-km aerosol optical depth from MODIS measurements over Hong Kong, IEEE Transactions on Geoscience and Remote Sensing, 43, 2650–2658. [Google Scholar]
  31. Li J, Roy DP, 2017. A global analysis of sentinel-2A, sentinel-2B and Landsat-8 data revisit intervals and implications for terrestrial monitoring. Remote Sensing, 9(9), 902. [Google Scholar]
  32. Loveland TR, Irons JR, 2016. Landsat 8: The plans, the reality, and the legacy. Remote Sensing of Environment, 185, 1–6. [Google Scholar]
  33. Lu Y, Wang Y, Zuo J, Jiang H, Huang D, Rameezdeen R, 2018. Characteristics of public concern on haze in China and its relationship with air quality in urban areas. Science of the Total Environment, 637, 1597–1606. [DOI] [PubMed] [Google Scholar]
  34. Lyapustin A, Wang Y, Korkin S, Huang D, 2018. MODIS Collection 6 MAIAC algorithm. Atmospheric Measurement Techniques, 11(10). [Google Scholar]
  35. Lyapustin A, Wang Y, Laszlo I, Kahn R, Korkin S, Remer L, Levy R, Reid JS, 2011. Multiangle implementation of atmospheric correction (MAIAC): 2. Aerosol algorithm. Journal of Geophysical Research: Atmospheres, 116(D3). [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Ma Z, Hu X, Sayer AM, Levy R, Zhang Q, Xue Y, Tong S, Bi J, Huang L, Liu Y, 2015. Satellite-based spatiotemporal trends in PM2.5 concentrations: China, 2004–2013. Environmental health perspectives, 124(2), 184–192. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Mei L, Rozanov V, Vountas M, Burrows JP, Levy RC, Lotz W, 2017. Retrieval of aerosol optical properties using MERIS observations: Algorithm and some first results. Remote sensing of environment, 197, 125–140. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Nichol JE, Bilal M, 2016. Validation of MODIS 3 km resolution aerosol optical depth retrievals over Asia. Remote Sensing, 8(4), 328. [Google Scholar]
  39. Peng J, Chen S, Lü H, Liu Y, Wu J, 2016. Spatiotemporal patterns of remotely sensed PM2.5 concentration in China from 1999 to 2011. Remote Sensing of Environment, 174, 109–121. [Google Scholar]
  40. Rohde RA, Muller RA, 2015. Air pollution in China: mapping of concentrations and sources. PloS one, 10(8), e0135749. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Roy DP, Li J, Zhang HK, Yan L, 2016. Best practices for the reprojection and resampling of Sentinel-2 Multi Spectral Instrument Level 1C data. Remote Sensing Letters, 7(11), 1023–1032. [Google Scholar]
  42. Roy DP, Wulder MA, Loveland TR, Woodcock CE, Allen RG, Anderson MC, Helder D, Irons JR, Johnson DM, Kennedy R, Scambos TA, Schaaf CB, Schott JR, Sheng Y, Vermote EF, Belward AS, Bindschadler R, Cohen WB, Gao F, Hipple JD, Hostert P, Huntington J, Justice CO, Kilic A, Kovalskyy V, Lee ZP, Lymburner L, Masek JG, McCorkel J, Shuai Y, Trezza R, Vogelmann J, Wynne RH, Zhu Z, 2014a. Landsat-8: Science and product vision for terrestrial global change research. Remote sensing of Environment, 145, 154–172. [Google Scholar]
  43. Roy DP, Qin Y, Kovalskyy V, Vermote EF, Ju J, Egorov A, Hansen MC, Kommareddy I, Yan L, 2014b. Conterminous United States demonstration and characterization of MODIS-based Landsat ETM+ atmospheric correction. Remote Sensing of Environment, 140, 433–449. [Google Scholar]
  44. Storey J, Choate M, Lee K, 2014. Landsat 8 Operational Land Imager on-orbit geometric calibration and performance. Remote sensing, 6(11), 11127–11152. [Google Scholar]
  45. Torres O, Bhartia PK, Herman JR, Sinyuk A, Ginoux P, Holben B, 2002. A long-term record of aerosol optical depth from TOMS observations and comparison to AERONET measurements. Journal of the Atmospheric Sciences, 59(3), 398–413. [Google Scholar]
  46. van Donkelaar A, Martin RV, Brauer M, Kahn R, Levy R, Verduzco C, Villeneuve PJ, 2010. Global estimates of ambient fine particulate matter concentrations from satellite-based aerosol optical depth: development and application. Environmental health perspectives, 118(6), 847. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Vermote EF, El Saleous NZ, Justice CO, 2002. Atmospheric correction of MODIS data in the visible to middle infrared: first results. Remote Sensing of Environment, 83(1–2), 97–111. [Google Scholar]
  48. Vermote E, Justice C, Claverie M, Franch B, 2016. Preliminary analysis of the performance of the Landsat 8/OLI land surface reflectance product. Remote Sensing of Environment, 185, 46–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Wang J, Christopher SA, 2003. Intercomparison between satellite-derived aerosol optical thickness and PM2.5 mass: Implications for air quality studies. Geophysical research letters, 30(21). [Google Scholar]
  50. Wang YQ, Zhang XY, Sun JY, Zhang XC, Che HZ, Li Y, 2015. Spatial and temporal variations of the concentrations of PM 10, PM 2.5 and PM 1 in China. Atmospheric Chemistry and Physics, 15(23), 13585–13598. [Google Scholar]
  51. Xie Y, Zhang Y, Xiong X, Qu JJ, Che H, 2011. Validation of MODIS aerosol optical depth product over China using CARSNET measurements. Atmospheric Environment, 45(33), 5970–5978. [Google Scholar]
  52. Yan L, Roy DP, Li Z, Zhang HK, Huang H, 2018. Sentinel-2A multi-temporal misregistration characterization and an orbit-based sub-pixel registration methodology. Remote Sensing of Environment, 215, 495–506. [Google Scholar]
  53. Zhang HK, Roy DP, Yan L, Li Z, Huang H, Vermote E, Skakun S, Roger JC, 2018. Characterization of Sentinel-2A and Landsat-8 top of atmosphere, surface, and nadir BRDF adjusted reflectance and NDVI differences. Remote Sensing of Environment, 215, 482–494. [Google Scholar]
  54. Zhang YL, Cao F, 2015. Fine particulate matter (PM 2.5) in China at a city level. Scientific Reports, 5, 14884. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Zhao TXP, Chan PK, Heidinger AK, 2013. A global survey of the effect of cloud contamination on the aerosol optical thickness and its long-term trend derived from operational AVHRR satellite observations. Journal of Geophysical Research: Atmospheres, 118(7), 2849–2857. [Google Scholar]

RESOURCES