This article was downloaded by: [UTSA Libraries] On: 08 September 2014, At: 05:57 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
International Journal of Remote Sensing Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/tres20
Evaluation of atmospheric correction models and Landsat surface reflectance product in an urban coastal environment a
a
Majid Nazeer , Janet E. Nichol & Ying-Kit Yung
b
a
Department of Land Surveying and Geo-Informatics, The Hong Kong Polytechnic University, Kowloon, Hong Kong b
The Hong Kong Environmental Protection Department, Water Policy and Science Group, Wan Chai, Hong Kong Published online: 29 Aug 2014.
To cite this article: Majid Nazeer, Janet E. Nichol & Ying-Kit Yung (2014): Evaluation of atmospheric correction models and Landsat surface reflectance product in an urban coastal environment, International Journal of Remote Sensing, DOI: 10.1080/01431161.2014.951742 To link to this article: http://dx.doi.org/10.1080/01431161.2014.951742
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
Conditions of access and use can be found at http://www.tandfonline.com/page/termsand-conditions
International Journal of Remote Sensing, 2014 http://dx.doi.org/10.1080/01431161.2014.951742
Evaluation of atmospheric correction models and Landsat surface reflectance product in an urban coastal environment Majid Nazeera, Janet E. Nichola*, and Ying-Kit Yungb a Department of Land Surveying and Geo-Informatics, The Hong Kong Polytechnic University, Kowloon, Hong Kong; bThe Hong Kong Environmental Protection Department, Water Policy and Science Group, Wan Chai, Hong Kong
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
(Received 21 March 2014; accepted 2 July 2014) Precise atmospheric correction is important for applications where small differences in surface reflectance (SR) are significant, such as biomass estimation, crop phenology, and retrieval of water quality parameters. It also enables direct comparison between different image dates and different sensors. As a precursor to monitoring different parameters of water quality around the coastline of Hong Kong using medium-resolution sensors Landsat TM/ETM, and HJ-1A/B, this study evaluated the performance of five atmospheric correction methods. The estimated SR of the first four reflective bands of Landsat 7 ETM+ and of the identical bands of the HJ-1A/B satellites was compared with in situ multispectral radiometer (MSR) SR measurements over sand, artificial turf, grass, and water surfaces for the five atmospheric correction methods – second simulation of the satellite signal in the solar spectrum (6S), fast line-of-sight atmospheric analysis of spectral hypercubes (FLAASH), atmospheric correction (ATCOR), dark object subtraction (DOS), and the empirical line method (ELM). Among the five methods, 6S was observed to be consistently more precise for SR estimation, with significantly less difference from the in-situ-measured SR, especially over lower reflective water surfaces. Of the two image-based methods, DOS performed well over the darker surfaces of water and artificial turf, although still inferior to 6S, while ELM worked well for grass sites as compared to the DOS and equalled the good performance of 6S over the high reflective sand surfaces. The study also evaluated the new standard Landsat SR product Landsat ecosystem disturbance adaptive processing system (LEDAPS) using the in situ measured SR data for the three land surface types – sand, artificial turf, and grass. For the highly and moderately reflecting bright sand and artificial turf, LEDAPS performed poorly, while for the darker grass site it performed better, although still inferior to 6S and ELM methods. This is probably due to the variable aerosol types and atmospheric conditions of Hong Kong, as LEDAPS was mainly compiled with reference to larger continental landmass areas.
1. Introduction The spectral radiance measured by satellite sensors is affected by absorption and scattering by atmospheric particles. Therefore, to obtain constant and accurate ground signatures, it is necessary to remove atmospheric artefacts using an efficient and reliable atmospheric correction method. Atmospheric correction is important when using ratio transformations, e.g. the normalized difference vegetation index (NDVI) (Song et al. 2001) as well as applications dependent on subtle differences in the surface reflectance (SR) such as crop phenology, or the retrieval of water quality parameters (WQPs). *Corresponding author: Email:
[email protected] © 2014 Taylor & Francis
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
2
M. Nazeer et al.
Numerous methods for atmospheric correction of satellite images have been presented (Song et al. 2001; Chavez 1988; Furby and Campbell 2001; Yang and Lo 2000). These fall into two types, namely physical methods including low-resolution atmospheric transmission (LOWTRAN), moderate-resolution atmospheric transmission (MODTRAN), and second simulation of the satellite signal in the solar spectrum (6S) and image-based methods including dark object subtraction (DOS) and the empirical line method (ELM). Physical methods use a radiative transfer model (RTM) to estimate SR, while image-based methods obtain relevant parameters from the image. In this study, three physical methods, FLAASH, ATCOR, and 6S, and two image-based methods, DOS and ELM, are evaluated for Landsat ETM+ and HJ-1A/B charge coupled device (CCD) images using in situ SR data. The Landsat series of satellites has been monitoring the Earth for more than four decades. In order to increase the comparability of Landsat data over space and time as well as to compare with other sensors, a SR product was developed by NASA Goddard Space Flight Center (GSFC) and the University of Maryland in 2006 (Masek et al. 2006) for Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) data sets (Feng et al. 2013). The Landsat SR products are generated from specialized software called the Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS). The method uses the 6S RTM to generate SR for a Landsat scene using a global data set of water vapour (WV), ozone, geopotential height, and digital elevation (USGS 2013), and aerosol optical thickness is extracted directly from the Landsat image (Vermote and Saleous 2007). A number of studies have examined the efficiency of the LEDAPS SR product by comparison with independent data sources (Masek et al. 2006; Ju et al. 2012; Ouaidrari and Vermote 1999; Vermote, Saleous, et al. 1997; Vermote, Tanre, et al. 1997). Most recently, Maiersperger et al. (2013) used field spectrometer data to evaluate the quality of LEDAPS SR at three sites including one grassland area at South Dakota State University and two spectrally brighter arid sites. For the grassland site, a total of 11 Landsat-5 (L5) and Landsat-7 (L7) scenes were matched to coincident spectrometer data from 2003 to 2010. For grassland, the differences between satellite and ground measurements for any particular day were in the range of 0–3.6% reflectance for each band. For the spectrally brighter sites, five scenes were matched with coincident spectrometer data, and the differences between LEDAPS SR and spectrometer measurements were greater than for grassland, within 6% reflectance for all match-ups except one which increased to 12%. The better results for the grassland sites suggest the dependence of LEDAPS SR on the presence of adequate dense dark vegetation cover near the area of interest (AOI). Maiersperger et al. (2013) also evaluated the temporal stability of LEDAPS SR for two of the CEOS (Committee on Earth Observation Satellites) pseudo-invariant calibration sites, Algodones Dunes and Libya-4. For Algodones Dunes, 83 L5 TM scenes from 2002 to 2010, and for Libya-4, 57 L5 TM scenes from 2000 to 2010, were compared with ground measurements collected on 27 February 2011. The overall LEDAPS SR uncertainty was reported to be within 3% (for bands 1–4 and 7) and 4% for band 5. Thus, although LEDAPS has already been validated (Ju et al. 2012; Maiersperger et al. 2013), this was over large continental landmasses which are far from the sea or large waterbodies. In order to test the accuracy of the operational LEDAPS SR product over the varied atmospheric conditions and subtropical coastal region of Hong Kong, it was compared with local in situ multispectral radiometer (MSR) SR data.
International Journal of Remote Sensing
3
2. Study area and data used
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
Hong Kong is a special administrative region of Mainland China, located at latitude/ longitude 22° 15′ N/114° 10′ E, on the eastern shore of the Pearl River estuary, having an area of 1095 km2. It shares a border to the north with Guangdong province of China and is surrounded by the South China Sea on the east, west, and south.
2.1. Satellite data The L7 satellite was launched in April 1999 with the ETM+ sensor on board. The sensor has a spatial resolution of 30 m for the six reflective bands, 60 m for the thermal band, and 15 m for the panchromatic band. All ETM+ acquisitions after 31 May 2003 have an anomaly caused by the failure of the scan line corrector (SLC). The malfunction of the SLC mirror assembly resulted in the loss of approximately 22% of the normal scene area, while the middle of each scene (approximately 22 km wide) contains very little data loss (Chander et al. 2009). The HJ-1A/B satellites were launched by China in September 2008 and together carry four CCD cameras on board, i.e. two for HJ-1A (CCD1 and CCD2) and two for HJ-1B (CCD1 and CCD2) (Zhang et al. 2013). The spectral range and spatial resolution of the four CCD cameras is similar to the first four bands of Landsat TM/ETM+ (Table 1), while the main advantage of HJ-1A/B over Landsat is the shortening of the revisit time from 16 days to 48 hours or less. The study used four HJ-1A/B satellite images, two L7 ETM+ images, and two atmospherically corrected ETM+ SR product (LEDAPS) images. The HJ-1A/B images were obtained from the China Centre for Resources Satellite Data and Application (CRESDA), the ETM+ images were obtained from the United States Geological Survey (USGS) Earth Explorer website (http://earthexplorer.usgs.gov/), and the atmospherically corrected ETM+ SR product was obtained from the USGS On-Demand Landsat SR facility (landsat.usgs.gov/CDR_LSR.php). Table 1. Band
Band designations of Landsat ETM+, HJ-1A/B CCD cameras, and MSR-16R radiometer. WL (µm)
Landsat 7 ETM+ B1 – Blue 0.450–0.515 B2 – Green 0.525–0.605 B3 – Red 0.630–0.690 B4 – Near infrared 0.750–0.900 HJ-1A/B CCD cameras B1 – Blue 0.430–0.520 B2 – Green 0.520–0.600 B3 – Red 0.630–0.690 B4 – Near infrared 0.760–0.900 Cropscan multispectral radiometer (MSR-16R) B2 – Blue 0.457–0.525 B5 – Green 0.525–0.597 B7 – Red 0.633–0.691 B9 – Near infrared 0.760–0.906 Note: WL, wavelength; CWL, central wavelength.
CWL (µm)
Resolution (m)
0.483 0.565 0.660 0.825
30 30 30 30
0.475 0.560 0.660 0.830
30 30 30 30
0.491 0.561 0.662 0.833
— — — —
4
M. Nazeer et al.
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
2.2. Field spectral data Field spectral data, representing SR, were collected using a Cropscan MSR-16R MSR (Cropscan Inc., Rochester, NY, USA) for wavelengths similar to the wavelengths of the first four bands (B1, B2, B3, and B4) of the ETM+ and HJ-1A/B CCD cameras (Table 1). Previous studies (Wu and Bauer 2012; Wong, Nichol, and Lee 2011; Nichol, Wong, and Chan 2008) have also used the Cropscan MSR-16R for the retrieval of SR in conjunction with QuickBird, Terra Moderate Resolution Imaging Spectroradiometer (MODIS), and CHRIS/PROBA imagery. The SR data were collected from four types of surfaces, namely sand, artificial turf of sports pitches, grass, and over water, between 10:00 am and 2:00 pm under cloud-free conditions. The spectral data were collected (Figure 1) over sand from three beach sites, from three sports pitches of artificial turf, over grass in three urban park sites, and over water from seven sampling locations in the Tolo Harbour region (northeast of Hong Kong). Figure 2 shows the MSR sampling locations and their names. Multiple spectra (at least 30 from each sampling site) were collected from each field site, and an averaged value was used for final comparison with satellite estimated SR. The standard deviation of measurements for each date, each sampling site, and each band was generally low, which shows that the collected spectra were bunched together.
2.3. Pixel selection criteria for validation The validation of satellite-measured SR with direct ground-measured SR provides the best method to evaluate different atmospheric correction methods. Thus, the comparison needs sufficiently large homogenous surfaces, but in an urban and densely built city, such as Hong Kong, it is difficult to find large areas. Therefore, for each sampling site, pure pixels were selected, and the reflectance contribution from the neighbouring pixels, i.e. adjacency effect, was compensated for in the physical atmospheric correction methods. For the beach sites, care was also taken to avoid the influence of the neighbouring water pixels from the seaward side and vegetation/built-up area pixels from the landward side. The number of pixels selected for each sampling location was based on its area. For two large beach sites (RBB and SOB), an averaged value of nine pixels (three adjacent pixels in groups of three at the east, centre, and west of the beaches) was used, while for one small beach site (BWBB), the average value of three pixels was used. Due to the small areas of artificial turf sports pitches, a single pixel value was used. For grass areas, an averaged value of two adjacent pixels was used for each site, and over water surface, an averaged value from a window of 3 × 3 pixels was used. For L7 ETM+, the image data were extracted only for those in situ MSR sampling locations which were not affected by ETM+ scan line errors. The reflectance contribution from neighbouring pixels (adjacency effect) was incorporated into the physical atmospheric correction methods (6S, FS, and AC). The 6S accounts for adjacency effects based on view and azimuth angles of the sensor (Vermote, Tanre, et al. 1997) and FS and AC account for adjacency effects by accounting for the SR of the pixel under observation compared with its surroundings. In the case of turf areas, although a single pixel (30 m × 30 m) was considered, the actual area of each sampled turf site was considerably larger than a single pixel, e.g. 100 m × 60 m for KPHG, 100 m × 60 m for BSPG, and 60 m × 35 m for SKMPG, which means that the SR contribution from any immediately adjacent area was the same as that from the sampled pixel. Each sample was located using a Garmin eTrex H GPS, with positional accuracy of less than 5 m.
International Journal of Remote Sensing
5
4.0 SL1 SL2 SL3 SL4 SL5 SL6 SL7
3.5
2.5 2.0 1.5 1.0 0.5 (a) 0.0 1
3
2
4
Band Number 50
KPHG BSPG SKMPG RBB SOB BWBB THTRG-S THTRG-N AAF
40 Surface Reflectance (%)
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
Surface Reflectance (%)
3.0
30
20
10
(b) 0 1
2
3
4
Band Number
Figure 1. Spectral signatures of different features: (a) spectral signatures of different water sampling locations for 17 January 2013 and (b) averaged spectral signatures from sand, grass, and artificial turf areas for 7 December 2013 except KPHG (6 November 2013).
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
6
M. Nazeer et al.
Figure 2. The MSR sampling locations on a Landsat image. The gold-coloured box represents sand areas, green-coloured box represents the artificial turf and grass sites, and blue-coloured box represents the sampling locations over water [Shek O Beach (SOB), Lat: 22.22909782, Long: 114.25096563; Repulse Bay Beach (RBB), Lat: 22.23695841, Long: 114.19589620; Big Wave Bay Beach (BWBB), Lat: 22.246531, Lon: 114.247032; Artificial Turf and Grass Sites: Boundary Street Play Ground (BSPG), Lat: 22.32652465, Lon: 114.17058144; King’s Park Hockey Ground (KPHG), Lat: 22.30743958, Lon: 114.17742377; Shek Kip Mei Play Ground (SKMPG), Lat: 22.33548323, Lon: 114.16932907; Tai Hang Tung Recreation Ground-North (THTRG-N), Lat: 22.32867778, Lon: 114.17121111; Tai Hang Tung Recreation Ground-South (THTRG-S), Lat: 22.32733079, Lon: 114.17089067; Aberdeen Athletic Field (AAF), Lat: 22.24961444, Lon: 114.17175397; and Sampling Locations Over Water: SL1, Lat: 22.412153, Long: 114.218057; SL2, Lat: 22.447288, Long: 114.203479; SL3, Lat: 22.432922, Long: 114.219518; SL4, Lat: 22.456713, Long: 114.224295; SL5, Lat: 22.443590, Long: 114.241809; SL6, Lat: 22.448390, Long: 114.267782; SL7, Lat: 22.473857, Long: 114.299733].
3. Methodology Atmospheric correction of the Landsat ETM+ and HJ-1A/B CCD images was carried out using three physical based methods – 6S, FLAASH (FS), and ATCOR (AC) – and two image-based methods – DOS and ELM. Three different aerosol models namely urban (UB), maritime (MT), and continental (CT) were tested with the physical based models. The effect of off-nadir viewing for HJ-1A/B imagery was considered for physical atmospheric correction methods by considering the sensor zenith and azimuth angles.
International Journal of Remote Sensing
7
3.1. Atmospheric correction using 6S The 6S is an advanced radiative transfer code designed to simulate the reflectance of solar radiation by a coupled atmosphere–surface system for a wide range of atmospheric, spectral, and geometric conditions (Kotchenova et al. 2006). It calculates the atmospheric correction coefficients xa, xb, and xc for each band separately based on input data (Tables 2 and 3), which indicate the most likely atmospheric conditions during image acquisition. These atmospheric correction coefficients are applied to the top of atmosphere (TOA) radiance (Lsatλ) to obtain SR for each band. Using the 6S model, the SR (ρ) free from atmospheric effects was calculated using Equation (1).
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
ρ¼
Y ; 1:0 þ ðxc Y Þ
(1)
where Y ¼ ðxa Lsatλ Þ xb . xa is the inverse of transmittance, xb is the scattering term of the atmosphere, and xc is the reflectance of the atmosphere for isotropic light. Information on solar zenith angle (θs), solar azimuth angle (φs), sensor zenith angle (θo), sensor azimuth angle (φo), and image acquisition date and time were extracted from the image metadata file, while values of WV and aerosol optical depth (AOD) were retrieved from MODIS Terra Daily Level-3 (1° × 1°) global atmospheric product (MOD08_D3.051) (Sriwongsitanon, Surakit, and Thianpopirug 2011) from http://gdata1.sci.gsfc.nasa.gov/ daac-bin/G3/gui.cgi?instance_id=MODIS_DAILY_L3. The local equatorial crossing time of MODIS Terra is 10:30 am (Gkikas et al. 2013); therefore, the Landsat ETM+ and Table 2.
6S model input parameters for Landsat ETM+ images.
Parameters Sensor Image time (GMT hd) Scene centre longitude (dd) Scene centre latitude (dd) Water vapour (g cm–2) Ozone (cm-atm) Aerosol model Aerosol optical depth at 550 nm
Table 3.
20 October 2013
7 December 2013
ETM+ 2.80 113.58 23.11 4.5 0.26 UB/MT/CT 0.54
ETM+ 2.81 113.58 23.11 2.5 0.25 UB/MT/CT 0.66
6S model input parameters for HJ-1A/B CCD images.
Parameters Sensor Image time (GMT hd) θs φs θo φo Water vapour (g cm–2) Ozone (cm-atm) Aerosol model Aerosol optical depth at 550 nm
17 January 2013
3 December 2013
6 December 2013
7 December 2013
HJ-1B CCD2 2.42 52.25 323.87 14.35 96.86 3.0 0.25 UB/MT/CT 0.66
HJ-1A CCD1 2.28 52.25 323.87 14.35 96.86 2.0 0.25 UB/MT/CT 0.38
HJ-1A CCD2 1.92 57.50 320.23 26.90 96.36 1.5 0.25 UB/MT/CT 0.42
HJ-1A CCD2 2.33 55.99 323.46 31.50 −79.88 2.5 0.25 UB/MT/CT 0.66
8
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
Table 4.
M. Nazeer et al. ATCOR and FLAASH input parameters for Landsat ETM+ and HJ-1A/B CCD images.
Parameters
20 October 7 December 17 January 3 December 6 December 7 December 2013 2013 2013 2013 2013 2013
Sensor
ETM+
ETM+
Sensor altitude (km) Pixel size (m) θs θo φo Atmospheric model Aerosol model Visibility (km)
705 30.0 39.08 — — Tropical
705
HJ-1B CCD2 650
HJ-1A CCD1 650
HJ-1A CCD2 650
HJ-1A CCD2 650
30.0 50.55 — — Tropical
30.0 52.25 14.35 96.86 Tropical
30.0 52.25 14.35 96.86 Tropical
30.0 57.50 26.90 96.36 Tropical
30.0 55.99 31.50 −79.88 Tropical
UB/MT/CT 13.0
UB/MT/CT 12.0
UB/MT/CT UB/MT/CT 13.0 12.0
UB/MT/CT UB/MT/CT 8.0 10.0
HJ-1A/B images used in this study were within 35 min of the MODIS Terra overpass. Columnar ozone (O3) data were retrieved from the OMI Daily Level-3 (0.25° × 0.25°) global gridded product from http://gdata1.sci.gsfc.nasa.gov/daac-bin/G3/gui.cgi? instance_id=omi. The expected uncertainty for the vertical OMI O3 density is 1.3–2.2% (Veefkind et al. 2006), for the MODIS WV it is 5–10% (Gao and Kaufman 2003), while for the MODIS Level-3 AOD the expected error (EE) can be determined using Equation (2) (Ruiz-Arias et al. 2013). EEAODðL3Þ ¼ 0:29 τ 2 þ 0:06 τ þ 0:06 ;
(2)
where τ is the true AOD value.
3.2. Atmospheric correction using ATCOR ATCOR is an add-on module of ERDAS IMAGINE (Richter 1996), which uses MODTRAN 4 RTM to create look-up tables for different atmospheric input parameters (Table 4). The visibility values for the image acquisition date and time were obtained from the Hong Kong Observatory (HKO), and as Hong Kong falls in the tropical climate zone, a tropical atmospheric model was selected. The calibration files for Landsat ETM+, available in ATCOR 2, were updated for each date according to the band-specific gain and bias values, while the calibration files for HJ-1A/B CCDs were generated using the band-specific gain and bias values retrieved from the image metadata file.
3.3 Atmospheric correction using FLAASH FLAASH is an atmospheric correction module in the Envi software that corrects wavelengths in the visible through near-infrared and shortwave infrared regions using MODTRAN 4 (Yuan and Niu 2008) and several input parameters (Table 4). The spectral response function (SRF) files for Landsat ETM+ bands were available in FLAASH, while for HJ-1 the SRF files were built using the Spectral Library Builder in Envi.
International Journal of Remote Sensing
9
3.4. Atmospheric correction using the dark object subtraction (DOS) method DOS is an image-based atmospheric correction method that assumes that some objects were under complete shadow during image acquisition, and these must have zero reflectance. In fact, atmospheric scattering and absorption make the imaging system record a non-zero digital number (DN) value for these dark objects. This constant non-zero DN value, DN haze, can be observed from the image histogram and is subtracted from the whole band. A relative scattering model is also considered to incorporate the atmospheric scattering effect. Equation (3) is the general equation used for SR estimation using the DOS method (Chavez 1996).
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
ρ¼
π d 2 ðLsatλ Lhazeλ Þ ; ðEsunλ cos θS Þ
(3)
where Lhazeλ is the path radiance for band λ (W m–2 sr–1 μm–1), Esunλ is the Exo-atmospheric solar irradiance for band λ (W m–2 μm–1), and d is the Earth–Sun distance (astronomical units). Generally, it is assumed that dark objects comprise approximately 1% of the whole image radiance (Mahiny and Turner 2007). Therefore, the 1% radiance of a dark object can be calculated using Equation (4). L1% ¼
0:01Esunλ cos θ : π d2
(4)
Thus, L1% needs to be subtracted from the path radiance (Lhazeλ), and finally the SR is calculated using Equation (5). ρ¼
π d 2 ðLsatλ ðLhazeλ L1%λ ÞÞ : Esunλ cos θ
(5)
3.5. Atmospheric correction using the empirical line method (ELM) The ELM uses field-measured SR from a series of invariant-in-time calibration targets to estimate SR for each band (Smith and Milton 1999). Although ELM can be performed using just two very distinct calibration targets, the use of more targets allows the parameters of the relationship between at-sensor radiance and at-surface reflectance to be estimated with greater confidence (Karpouzli and Malthus 2003). Using the ELM, a regression equation is developed for each band of an image to perform the atmospheric correction. In this study, eight calibration sites were used to develop a regression equation between the in situ MSR SR (collected coincident with the image acquisition time) and the at-sensor radiance of an HJ-1A CCD2 image of 7 December 2013. For ETM+, seven calibration sites were used to construct a regression equation. Because of the ETM+ scan line error, some MSR sampling locations were missing if only one image was considered. Therefore the at-sensor radiance values from the seven locations were extracted from two ETM+ images, 20 October 2013 image (for at-satellite radiance extraction of sites KPHG, RBB, SOB, and BWBB) and 7 December 2013 image (for at-satellite radiance extraction of sites BSPG and THTRG-S). The regression equations are given in Figure 3.
M. Nazeer et al.
MSR Surface Reflectance (%)
60
(a)
Band 1 Y = –81.30 + 1.296 X R = 0.96, RMSE = 2.3
50 40 30 20 10
60 MSR Surface Reflectance (%)
10
40 30 20 10 0
0 0
15
30
45
60
75
0 15 30 45 60 75 90 105 At-sensor Radiance (W m–2 sr–1 µm–1)
90 105
At-sensor Radiance (W m–2 sr–1 µm–1)
(c)
Band 3 Y = –40.02 + 1.264 X R = 0.97, RMSE = 3.2
50 40 30 20 10
60 MSR Surface Reflectance (%)
MSR Surface Reflectance (%)
60
0
(d)
50 40 30 Band 4 Y = –23.56 + 1.263 X R = 0.84, RMSE = 4.6
20 10 0
0 15 30 45 60 75 90 105 At-sensor Radiance (W m–2 sr–1 µm–1)
(e)
Band 1 Y = –35.45 + 0.650 X R = 0.93, RMSE = 3.0
50 40 30 20 10
15
30
45
60
75
90 105
At-sensor Radiance (W m–2 sr–1 µm–1) 60 MSR Surface Reflectance (%)
MSR Surface Reflectance (%)
60
Band 2 Y = –17.89 + 0.561 X R = 0.93, RMSE = 3.5
50
(f)
40 30 20 10
0 0 15 30 45 60 75 90 105 At-sensor Radiance (W m–2 sr–1 µm–1)
0 15 30 45 60 75 90 105 At-sensor Radiance (W m–2 sr–1 µm–1) 60 50 40 30 20 10 0 0
15
30
45
60
60
(g)
Band 3 Y = –18.05 + 0.690 X R = 0.97, RMSE = 3.5
MSR Surface Reflectance (%)
MSR Surface Reflectance (%)
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
(b)
Band 2 Y = –42.08 + 1.016 X R = 0.95, RMSE = 3.0
50
75 –2
90 105 –1
–1
At-sensor Radiance (W m sr µm )
50
Band 4 Y = –10.33 + 0.769 X R = 0.94, RMSE = 3.8
(h)
40 30 20 10 0 0 15 30 45 60 75 90 105 At-sensor Radiance (W m–2 sr–1 µm–1)
Figure 3. Regression lines and prediction equations for HJ-1A/B CCDs bands 1 to 4 (panels (a), (b), (c), and (d)) and for Landsat ETM+ bands 1 to 4 (panels (e), (f), (g), and (h)).
International Journal of Remote Sensing
11
4. Results The performance of each atmospheric correction method was assessed by comparing the SR difference between the in situ MSR-measured and satellite-estimated SR (i.e. MSR SR – estimated SR). In Figures 4–8, the x-axis represents different atmospheric correction methods and the y-axis represents the percentage difference between the in-situmeasured SR and the satellite-estimated SR. The colour of each bar represents a specific band. Negative difference values show that a specific method has overestimated the SR, while a positive value shows that the method has underestimated the SR.
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
4.1. Sand sites Among the different methods, overall, 6S performed best (Figure 4), showing results much closer to the in situ SR compared with FLAASH and ATCOR. Over the four dates, the maximum SR difference (absolute) for 6S was 10.8%, 7.4%, 11.0%, and 14.6% for B1, B2, B3, and B4, respectively. For the three physical based atmospheric correction methods, the urban aerosol model appeared best for such highly reflective sand sites. Of the two image-based methods, ELM showed comparable results to 6S, with maximum SR differences from the in situ SR of 9%, 7.8%, 10%, and 14% for B1, B2, B3, and B4, respectively. LEDAPS showed large differences from the in-situ-measured SR, underestimating SR in all the four bands. For one date (20 October 2013), the differences between LEDAPS and in-situ-measured SR were 12.8%, 15.4%, 18%, and 12.6%, while for the same date, the 6S urban aerosol model showed maximum SR differences of 4.6%, 7.4%, 10.6%, and 4.1% for B1, B2, B3, and B4, respectively. The LEDAPS differences are also higher than the SR difference reported by Maiersperger et al. (2013) for spectrally brighter sites.
4.2. Artificial turf sites The 6S and FLAASH with maritime aerosol model were the best performing models over moderately reflecting artificial turf areas (Figure 5). Over the three dates, the maximum SR difference was 8.9%, 10.6%, 6.4%, and 7.8% for B1, B2, B3, and B4, respectively. While for one date (7 December 2013), LEDAPS performed poorly compared to 6S and FLAASH. Among the image-based methods, DOS was better overall than the ELM and performed even better than 6S for bands 3 and 4.
4.3. Grass sites The 6S with continental aerosol model was the overall best method for the grass sites (Figure 6) except for two sites (THTRG-S and AAF) on one date (7 December 2013 HJ1A image) where band 4 showed a difference of 10%. For these, the 6S urban model gave better results. For the grass area, LEDAPS performed better, compared to its performance on other sites with SR differences of 2.4%, 1.3%, 0%, and 2.7% for B1, B2, B3, and B4, respectively, and within the error range reported by Maiersperger et al. (2013) for grass sites.
% Difference estimated versus in situ SR
% Difference estimated versus in situ SR
% Difference estimated versus in situ SR
–15
–10
–5
0
5
10
15
20
–20
–15
–10
–5
0
5
10
15
20
–15
–10
–5
0
5
10
15
20
–5
0
5
10
15
20
B2
B3
B4
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
B1
(RBB)
DOS
(RBB)
DOS
(RBB)
DOS
(RBB)
6S UB 6S MT 6S CT FS UB FS MT FS CTAC UBAC MTAC CT ELM DOSLEDAPS
–15
–10
–5
0
5
10
15
20
25
30
–20
–15
–10
–5
0
5
10
15
20
25
–20
–15
–10
–5
0
5
10
15
20
25
–5
0
5
10
15
20
25
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
(SOB)
DOS
(SOB)
DOS
(SOB)
DOS
(SOB)
6S UB6S MT 6S CT FS UBFS MT FS CTAC UBAC MTAC CT ELM DOSLEDAPS
–5
0
5
10
15
20
25
30
–20
–15
–10
–5
0
5
10
15
20
25
30
35
–25
–20
–15
–10
–5
0
5
10
15
20
25
–5
0
5
10
15
20
25
DOS
DOS
DOS
(BWBB)
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
(BWBB)
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
(BWBB)
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
(BWBB)
6S UB 6S MT6S CT FS UBFS MT FS CTAC UBAC MTAC CT ELM DOSLEDAPS
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
Figure 4. Comparison of in situ MSR SR and the estimated SR for sand sites. From top, the first row shows the results for the 20 October 2013 image of ETM+, the second row shows the results for the 3 December 2013 image of HJ-1A CCD1, the third row shows the results for the 6 December 2013 image of HJ-1A CCD2, and the fourth row shows the results from the 7 December 13 image of HJ-1A CCD2, for three sandy beach areas: Repulse Bay Beach (RBB), Shek O Beach (SOB), and Big Wave Bay Beach (BWBB).
% Difference estimated versus in situ SR
% Difference estimated versus in situ SR % Difference estimated versus in situ SR % Difference estimated versus in situ SR % Difference estimated versus in situ SR
% Difference estimated versus in situ SR % Difference estimated versus in situ SR % Difference estimated versus in situ SR % Difference estimated versus in situ SR
12 M. Nazeer et al.
B2
B3
B4
(BSPG)
0
–5
–10
6S UB 6S MT 6SCT FS UB FS MT FS CT ACUB AC MT ACCT ELM DOS 10 5
(BSPG)
0 –5 –10 –15 –20 –25 –30
6S UB 6S MT 6S CT FS UB FS MT FS CT ACUB AC MT ACCT ELM DOS
(BSPG) 5
0
–5
–10
–15
% Difference estimated versus in situ SR
B1
% Difference estimated versus in situ SR
6S UB6S MT 6SCT FS UB FS MT FS CTACUBACMTAC CT ELM DOSLEDAPS 5
% Difference estimated versus in situ SR
% Difference estimated versus in situ SR % Difference estimated versus in situ SR
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
% Difference estimated versus in situ SR
International Journal of Remote Sensing
13
6S UB6S MT 6SCT FS UB FS MT FS CTACUBACMTAC CT ELM DOSLEDAPS 15
(SKMPG) 10
5
0
–5
6SUB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT ACCT ELM DOS 15
(SKMPG)
10 5 0 –5 –10 –15 –20
6SUB 6S MT 6S CT FS UB FS MT FS CT ACUB AC MT ACCT ELM DOS 15
(SKMPG)
10 5 0 –5 –10
Figure 5. Comparison of in situ MSR SR and the estimated SR for two artificial turf areas. From top, the first row shows the results for the 7 December 2013 image of ETM+, the second row shows the results for the 6 December 2013 image of HJ-1A CCD2, and the third row shows the results from the 7 December 2013 image of HJ-1A CCD2.
4.4. Water sites Over water, the 6S model was significantly better than the other methods when using the continental or maritime aerosol model for all the sampling sites (Figure 7). The maximum differences were 1.2%, 2.6%, 1.4%, and 2.3% for B1, B2, B3, and B4, respectively. ATCOR gave negative or zero SR for B1 to B3 and showed no differences between the bands or between the sampling sites for all three aerosol models, suggesting insensitivity of ATCOR to scene and spectral parameters. Of the two image-based methods, DOS outperformed ELM in all bands, while B1 and B3 of ELM greatly underestimated the SR. Overall, the 6S physical method slightly outperformed DOS and greatly outperformed all other methods.
4.5. Sensitivity analysis As 6S outperformed all the other methods, it was further subjected to a sensitivity analysis, to analyse the effect of sensor-specific uncertainties in AOD (Equation (2)) –
M. Nazeer et al.
B1
B2
B3
% Difference estimated versus in situ SR
% Difference estimated versus in situ SR
(a)
B4
0
–5
–10
6S UB 6S MT 6S CT FSUB FSMT FS CT AC UB AC MT ACCT ELM
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
6SUB 6S MT 6SCT FS UB FSMT FS CT AC UB AC MT AC CT ELM
6SUB 6S MT 6SCT FSUB FSMT FSCT ACUB ACMT ACCT ELM DOSLEDAPS 5
10
(c) 5
0
–5
–10
–15
DOS
20 15
(b)
10 5 0 –5 –10 –15
DOS
6S UB 6S MT 6S CT FS UB FSMT FS CT ACUB ACMT AC CT ELM
% Difference estimated versus in situ SR
% Difference estimated versus in situ SR
14
DOS
35 30
(d)
25 20 15 10 5 0 –5 –10 –15
Figure 6. Comparison of in situ MSR SR and the estimated SR for three grass areas. Panel (a) shows the results for the 7 December 2013 image of ETM+ from a dry grass area of THTRG-S, panels (b), (c), and (d) show the results for the 7 December 2013 image of HJ-1A CCD2.
WV (5–10%) and O3 concentration (1.3–2.2%). The analysis was undertaken by randomly defining 20 AOIs, each consisting of a 3 × 3 pixel window over water surfaces for the 17 January 2013 image. The value of AOD was increased/decreased by 2–5% and also according to the EE (about 23%) (calculated using Equation (2)), WV was increased/ decreased by 5–10%, and O3 was increased/decreased by 2–5%. The descriptive statistics (Table 5) used are the mean, difference (Diff) (between the initial and new estimated SR value), standard deviation (StDev), minimum (Min), and maximum (Max). The Sr#1 ‘Initial’ in Table 5 represents the statistics of the 17 January 2013 image when the value of AOD was 0.66, WV was 3.0 (g cm–2), and O3 was 0.25 (cm-atm) (Table 3). In Table 5, Sr#2–7 represent the new AOD values when it was varied from its initial value to ±2%, ±5%, and the EE, Sr#8–11 represent the new WV values when it was varied from its initial value to ±5% and ±10%, and Sr#12–15 represent the new O3 values when it was varied from its initial value to ±2% and ±5%, respectively. The sensitivity analysis in 6S showed that the 6S accuracy was not greatly sensitive to the increase/decrease in AOD from ±2 to ±5% but was sensitive when its value was increased/decreased according to the EE (about 23%) and when the difference in SR was varied from −2 to 3% of SR for B1–B4. It was found that the 6S SR was not sensitive to the variation in WV (±5 to ±10%) and O3 (±2 to ±5%) as there was no change at all in SR in any band (except for band 2 [1%] when the O3 value was increased by 5%). Therefore, it can be concluded that when using the maritime aerosol model, the 6S SR is sensitive when the AOD value is increased/decreased by 23% or above from its initial value, while it is insensitive for variations in WV and O3. As both satellite systems (Landsat ETM+ and HJ-1A/B) have same spatial (30 m) and spectral resolution (8 bit), it is assumed that any differences in results are not due to sensor differences.
B1
B2
B3
B4
5
0
–5
–10
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM DOS 10
(SL3)
5
0
–5
–10
% Difference estimated versus in situ SR
10
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
DOS (SL1)
(SL5)
5
0
–5
–10
% Difference estimated versus in situ SR % Difference estimated versus in situ SR % Difference estimated versus in situ SR
% Difference estimated versus in situ SR % Difference estimated versus in situ SR
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM 10
6S UB 6S MT 6S CT FS UB FSMT FS CT ACUB ACMT AC CT ELM DOS
% Difference estimated versus in situ SR
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
International Journal of Remote Sensing
10
15 DOS
(SL2)
5
0
–5
–10
6S UB 6S MT 6S CT FS UB FS MT FS CT AC UB AC MT AC CT ELM
DOS
10 (SL4) 5
0
–5
–10
6S UB 6S MT 6S CT FS UB FSMT FS CT ACUB ACMT AC CT ELM DOS (SL6) 10
5
0
–5
–10
6S UB 6S MT 6S CT FSUB FSMT FS CT AC UB ACMT AC CT ELM DOS (SL7) 10
5
0
–5
–10
Figure 7. Comparison of in situ MSR SR and the estimated SR for water sites for the 17 January 2013 image of HJ-1B CCD2.
5. Discussion 5.1. Physical atmospheric correction methods Overall, the bright and moderately brighter surfaces of sand and artificial turf observed significantly lower accuracy for all the bands and methods used. Overall, 6S was the most accurate for all the selected surfaces and was highly accurate over dark surfaces of water and grass, greatly outperforming all other methods. It was also found that the aerosol models used are heavy drivers in SR estimation. The contribution of the aerosol models varies with respect to the surface type, according to the surface reflectivity. For example, for the highly reflective sand surfaces, the urban aerosol model gave the best estimate. Whereas for moderately reflective surfaces such as artificial turf and lower reflective surfaces such as grass, the maritime and continental aerosol models gave the best SR
Initial AOD + 0.02 AOD AOD – 0.02 AOD AOD + 0.05 AOD AOD – 0.05 AOD AOD + EE AOD AOD – EE AOD WV + 0.05 WV WV – 0.05 WV WV + 0.10 WV WV – 0.10 WV O3 + 0.02 O3 O3 – 0.02 O3 O3 + 0.05 O3 O3 – 0.05 O3
1 2
12 13 14 15
11
10
9
8
7
6
5
4
3
Variable
Sr#
0
0 0 0 0
0.05 0.05 0.05 0.05
0
0
0.02 0.02 0.02 0.02
0.02
0.02
0.02
0.02
0.02
−3
0
0.02
0.02
−1
+3
0.02
0.02
0.02 0.02
0.03 0.03 0.03 0.03
0.09 0.09 0.09 0.09
0.03 0.09
0.03 0.09
0.03 0.09
0.03 0.09
0.06 0.11
0.00 0.06
0.04 0.09
0.03 0.09
0.04 0.09
0.03 0.09 0.03 0.09
0.06 0.06 0.07 0.06
0.06
0.07
0.06
0.06
0.08
0.04
0.07
0.06
0.07
0.06 0.06
StDev Min Max Mean
0
0
– 0
Diff (%)
0.05
0.05
0.05
0.05
0.08
0.02
0.06
0.05
0.05
0.05 0.05
Mean
Band 1
0 0 −1 0
0
0
0
0
−2
+2
−1
0.02 0.02 0.02 0.02
0.02
0.02
0.02
0.02
0.02
0.03
0.02
0.02
0.02
−1 0
0.02 0.02
0.04 0.04 0.04 0.04
0.11 0.11 0.12 0.11
0.04 0.11
0.04 0.12
0.04 0.11
0.04 0.11
0.06 0.13
0.02 0.09
0.05 0.12
0.04 0.11
0.04 0.12
0.04 0.11 0.04 0.11
0.04 0.04 0.04 0.04
0.04
0.04
0.04
0.04
0.06
0.02
0.04
0.03
0.04
0.04 0.04
StDev Min Max Mean
– 0
Diff (%)
Band 2
0 0 0 0
0
0
0
0
−2
+2
0.02 0.02 0.02 0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
−1 0
0.02
0.02 0.02
0.02 0.02 0.02 0.02
0.09 0.09 0.09 0.09
0.02 0.09
0.02 0.09
0.02 0.09
0.02 0.09
0.04 0.11
0.00 0.07
0.02 0.09
0.01 0.09
0.02 0.09
0.02 0.09 0.02 0.09
0.02 0.02 0.02 0.02
0.02
0.02
0.02
0.02
0.04
0.00
0.02
0.02
0.02
0.02 0.02
StDev Min Max Mean
0
– 0
Diff (%)
Band 3
0 0 0 0
0
0
0
0
−2
+2
0
0
0
– 0
Diff (%)
0.02 0.02 0.02 0.02
0.02
0.02
0.02
0.02
0.02
0.01
0.02
0.02
0.02
0.02 0.02
0.00 0.00 0.00 0.00
0.07 0.07 0.07 0.07
0.00 0.07
0.00 0.07
0.00 0.07
0.00 0.07
0.02 0.08
0.00 0.05
0.00 0.07
0.00 0.07
0.00 0.07
0.00 0.07 0.00 0.07
StDev Min Max
Band 4
Table 5. Descriptive statistics for the 6S (maritime aerosol model) sensitivity with aerosol optical depth (AOD), water vapour (WV), and ozone (O3) for the 17 January 2013 image.
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
16 M. Nazeer et al.
International Journal of Remote Sensing
% Difference estimated versus in situ SR
B4 AC
(a)
20 October 2013 (ETM+)
25
3 December 2013 (HJ-1A) 6 December 2013 (HJ-1A) 7 December 2013 (HJ-1A)
20 15 10 5 0 FS
AC
10
5
0
25
Surface Type: Grass Aerosol Model: Continental 7 December 2013 (HJ-1A) 7 December 2013 (ETM+)
10
FS
AC
DOS Surface Type: Artificial Turf Aerosol Model: Maritime 6 December 2013 (HJ-1A) 7 December 2013 (HJ-1A) 7 December 2013 (ETM+)
(b) 20 15 10 5 0 6S
ELM
15
(c)
6S
ELM Surface Type: Sand Aerosol Model: Urban
6S
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
B3 FS
% Difference estimated versus in situ SR
% Difference estimated versus in situ SR
30
B2
% Difference estimated versus in situ SR
B1 6S
17
(d)
FS
AC
DOS Surface Type: Water Aerosol Model: Maritime 17 January 2013 (HJ-1B)
5
0
Figure 8. Overall comparison of in situ MSR SR and estimated SR using different models for all the dates and sites. SR comparison for sand areas (a), artificial turf areas (b), grass areas (c), and water sampling locations (d).
estimate. In the case of low reflective water surfaces, the good performance of the maritime aerosol model is probably due to its sensitivity to coarse maritime particles composed of sea salt. A summary of the best atmospheric correction methods and aerosol models (discussed in Section 4) based on the analysis of all the dates and sites is presented in Figure 8. For all the dates, 6S gave the lowest difference from the in situ data, over all surface types, and was especially effective over water. The rationale for the testing of different atmospheric correction methods and aerosol models over different surfaces specifically over water was to select the most appropriate atmospheric correction method and an aerosol model for use over water surfaces. As this study is a precursor to the retrieval of WQPs, accurate SR estimation will enable us to estimate the WQPs with lower atmospheric error. Although the SR results over water have been validated for only one date, it is still useful compared to selecting an atmospheric correction method and aerosol model randomly, or by using results from other study areas. For example, some studies (Sriwongsitanon, Surakit, and Thianpopirug 2011; Duan et al. 2007) have estimated WQPs using Landsat data but have not validated the atmospheric correction method or aerosol model, while others have used an image-based method for retrieval of WQPs (Kabbara et al. 2008; Alparslan et al. 2007; Vincent et al. 2004).
5.2. Image-based atmospheric correction methods The ELM performed equally well as the physical methods for the sand and grass areas, while DOS performed well over water and showed greater differences than the physical methods. The good performance of DOS is probably because it is based on the selection of dark pixels similar to water in the image. DOS also gave slightly better results than ELM for the artificial turf areas. The results suggest that among the image-based methods,
18
M. Nazeer et al.
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
DOS is a good choice for SR estimation of dark surfaces such as water as well as moderately reflective surfaces, while ELM is good for SR estimation of highly reflective surfaces such as sand. The results also suggest that the ELM needs to be trained using more surface types to obtain good SR estimates over a variety of surfaces. In this study, the ELM performed well and better than DOS for grass and sand sites but underestimated SR over water, which suggests some uncertainty in training of the ELM for dark water surfaces, as the line in the ELM method represents both dark and bright targets. It can be expected that if the line construction had been restricted to dark areas, including a wider range of water types, ELM would have performed better over water. 5.3. Evaluation of LEDAPS As the aerosol model in LEDAPS is not flexible (Ju et al. 2012), i.e. it uses a continental aerosol model globally for the SR estimation, the results were compared to the direct runs of 6S using different aerosol models. LEDAPS greatly underestimated SR for the high reflective surfaces and showed moderate but variable performance for artificial turf. Although LEDAPS performed better over grass than over other surfaces, it was still significantly less accurate than the 6S continental model over the grass site. The LEDAPS validation results showed that the product accuracy was not within the expected uncertainty (3% for bands 1 to 4, defined by Maiersperger et al. 2013) for sand and artificial turf sites, while for grass the LEDAPS SR difference from the in situ measured SR was within the expected uncertainty. The discrepancies in LEDAPS SR estimation might also be due to the use of relatively coarser resolution (2.5° × 2.5°) WV data from National Centers for Environmental Prediction (NCEP) (Masek et al. 2006). Some limitations in validation of LEDAPS with the coincident ground measurements include the difficulty of obtaining coincident satellite and in situ data due to cloud-covered images and scan line errors of ETM+ coincident with the in situ sampling data. 6. Conclusions In this study, five different atmospheric correction methods, including three physical methods based on RTMs (6S, FLAASH, and ATCOR) and two image-based methods (DOS and ELM), were evaluated over sand, artificial turf, grass, and water using in situ MSR SR measurements. Of the physical methods, three different aerosol models – urban, maritime, and continental – were also tested. The in situ MSR SR data coincident with the two Landsat ETM+ images and four HJ-1A/B CCD images were used to select the best atmospheric correction method as well as the most suitable aerosol model. Overall, the 6S model was closest to the ground spectral data over all surfaces. The 6S was also tested for its sensitivity to AOD, WV, and ozone, considering the expected uncertainties of the respective sensors. By increasing/decreasing these variables from ±2% to ±10%, it was found that when using a maritime aerosol model, the 6S SR is sensitive when the AOD value is increased/decreased from its initial value, while it is insensitive to variations in WV and O3. Although the in situ data for the validation of LEDAPS were limited, overall LEDAPS showed greater differences from the in-situ-measured SR than the overall best methods 6S and DOS, for all the surface types (sand, artificial turf, and grass), and is clearly not suitable for the variable aerosol types, coastal urban area such as Hong Kong. As the validation of LEDAPS from previous studies suggests, this product may be more suitable for large continental landmass areas. The LEDAPS SR product uses a continental aerosol
International Journal of Remote Sensing
19
model globally for SR estimation, while Hong Kong is a densely populated urban area and has high, as well as mixed, aerosol loadings from both anthropogenic and marine sources. Therefore, 6S, which has the option to input different aerosol models such as urban and maritime, is better suited for SR estimation in the urban coastal environment. It is suggested that the LEDAPS SR product may be more applicable to urban coastal regions if a selection of aerosol models were available to users.
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
Acknowledgements This work would not have been possible without The Hong Kong Environmental Protection Department’s help in collecting the MSR data over water, the US Geological Survey for providing Landsat images, and the CRESDA for providing the HJ1-A/B images. The Landsat SR product is courtesy of the US Geological Survey Earth Resources Observation and Science Center. We acknowledge the mission scientists and principal investigators who provided the MODIS and OMI data products used in this research effort. Special thanks to Man Ho Lee for helping in MSR field data collection, Xiang Zhao (Beijing Normal University, Beijing, China) for providing the HJ-1A/B CCD bands SRF files, and Muhammad Bilal, Sawaid Abbas, Syed Muhammad Irteza (The Hong Kong Polytechnic University) and Muhammad Imran Shahzad (COMSATS Institute of Information Technology, Islamabad Pakistan) for their valuable suggestions. We thank the anonymous reviewers for their valuable suggestions and comments that have improved the quality of the article.
Funding This work was supported by The Hong Kong Polytechnic University Grant 1-ZVCY.
References Alparslan, E., C. Aydöner, V. Tufekci, and H. Tüfekci. 2007. “Water Quality Assessment at Omerli Dam Using Remote Sensing Techniques.” Environmental Monitoring and Assessment 135 (1– 3): 391–398. Chander, G., B. L. Markham, and D. L. Helder. 2009. “Summary of Current Radiometric Calibration Coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI Sensors.” Remote Sensing of Environment 113 (5): 893–903. Chavez, P. S. 1988. “An Improved Dark-Object Subtraction Technique for Atmospheric Scattering Correction of Multispectral Data.” Remote Sensing of Environment 24 (3): 459–479. doi:10.1016/0034-4257(88)90019-3. Chavez, P. S. 1996. “Image-Based Atmospheric Corrections-Revisited and Improved.” Photogrammetric Engineering and Remote Sensing 62 (9): 1025–1036. Duan, H., Y. Zhang, B. Zhang, K. Song, and Z. Wang. 2007. “Assessment of Chlorophyll-a Concentration and Trophic State for Lake Chagan Using Landsat TM and Field Spectral Data.” Environmental Monitoring and Assessment 129 (1–3): 295–308. doi:10.1007/s10661006-9362-y. Feng, M., J. O. Sexton, C. Huang, J. G. Masek, E. F. Vermote, F. Gao, R. Narasimhan, S. Channan, R. E. Wolfe, and J. R. Townshend. 2013. “Global Surface Reflectance Products from Landsat: Assessment Using Coincident MODIS Observations.” Remote Sensing of Environment 134: 276–293. doi:10.1016/j.rse.2013.02.031. Furby, S. L., and N. A. Campbell. 2001. “Calibrating Images from Different Dates to ‘like-Value’ Digital Counts.” Remote Sensing of Environment 77 (2): 186–196. doi:10.1016/S0034-4257(01) 00205-X. Gao, B.-C., and Y. J. Kaufman. 2003. “Water Vapor Retrievals Using Moderate Resolution Imaging Spectroradiometer (MODIS) near-Infrared Channels.” Journal of Geophysical Research 108 (D13): 4389. doi:10.1029/2002JD003023. Gkikas, A., N. Hatzianastassiou, N. Mihalopoulos, V. Katsoulis, S. Kazadzis, J. Pey, X. Querol, and O. Torres. 2013. “The Regime of Intense Desert Dust Episodes in the Mediterranean Based on
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
20
M. Nazeer et al.
Contemporary Satellite Observations and Ground Measurements.” Atmospheric Chemistry and Physics 13: 12135–12154. doi:10.5194/acp-13-12135-2013. Ju, J., D. P. Roy, E. F. Vermote, J. Masek, and V. Kovalskyy. 2012. “Continental-Scale Validation of MODIS-Based and LEDAPS Landsat ETM+ Atmospheric Correction Methods.” Remote Sensing of Environment 122: 175–184. doi:10.1016/j.rse.2011.12.025. Kabbara, N., J. Benkhelil, M. Awad, and V. Barale. 2008. “Monitoring Water Quality in the Coastal Area of Tripoli (Lebanon) Using High-Resolution Satellite Data.” ISPRS Journal of Photogrammetry and Remote Sensing 63 (5): 488–495. doi:10.1016/j.isprsjprs.2008.01.004. Karpouzli, E., and T. Malthus. 2003. “The Empirical Line Method for the Atmospheric Correction of IKONOS Imagery.” International Journal of Remote Sensing 24 (5): 1143–1150. doi:10.1080/0143116021000026779. Kotchenova, S. Y., E. F. Vermote, R. Matarrese, and F. J. Klemm, Jr. 2006. “Validation of a Vector Version of the 6S Radiative Transfer Code for Atmospheric Correction of Satellite Data. Part I: Path Radiance.” Applied Optics 45 (26): 6762–6774. doi:10.1364/AO.45.006762. Mahiny, A. S., and B. J. Turner. 2007. “A Comparison of Four Common Atmospheric Correction Methods.” Photogrammetric Engineering & Remote Sensing 73 (4): 361–368. doi:10.14358/ PERS.73.4.361. Maiersperger, T. K., P. L. Scaramuzza, L. Leigh, S. Shrestha, K. P. Gallo, C. B. Jenkerson, and J. L. Dwyer. 2013. “Characterizing LEDAPS Surface Reflectance Products by Comparisons with AERONET, Field Spectrometer, and MODIS Data.” Remote Sensing of Environment 136: 1–13. doi:10.1016/j.rse.2013.04.007. Masek, J. G., E. F. Vermote, N. E. Saleous, R. Wolfe, F. G. Hall, K. F. Huemmrich, F. Gao, J. Kutler, and T.-K. Lim. 2006. “A Landsat Surface Reflectance Dataset for North America, 1990–2000.” IEEE Geoscience and Remote Sensing Letters 3 (1): 68–72. doi:10.1109/LGRS.2005.857030. Nichol, J. E., M. S. Wong, and Y. Y. Chan. 2008. “Fine Resolution Air Quality Monitoring from a Small Satellite: CHRIS/PROBA.” Sensors 8 (12): 7581–7595. doi:10.3390/s8127581. Ouaidrari, H., and E. F. Vermote. 1999. “Operational Atmospheric Correction of Landsat TM Data.” Remote Sensing of Environment 70 (1): 4–15. doi:10.1016/S0034-4257(99)00054-1. Richter, R. 1996. “A Spatially Adaptive Fast Atmospheric Correction Algorithm.” International Journal of Remote Sensing 17 (6): 1201–1214. doi:10.1080/01431169608949077. Ruiz-Arias, J. A., J. Dudhia, C. A. Gueymard, and D. Pozo-Vázquez. 2013. “Assessment of the Level-3 MODIS Daily Aerosol Optical Depth in the Context of Surface Solar Radiation and Numerical Weather Modeling.” Atmospheric Chemistry and Physics 13 (2): 675–692. doi:10.5194/acp-13-675-2013. Smith, G. M., and E. J. Milton. 1999. “The Use of the Empirical Line Method to Calibrate Remotely Sensed Data to Reflectance.” International Journal of Remote Sensing 20 (13): 2653–2662. doi:10.1080/014311699211994. Song, C., C. E. Woodcock, K. C. Seto, M. P. Lenney, and S. A. Macomber. 2001. “Classification and Change Detection Using Landsat TM Data.” Remote Sensing of Environment 75 (2): 230– 244. doi:10.1016/S0034-4257(00)00169-3. Sriwongsitanon, N., K. Surakit, and S. Thianpopirug. 2011. “Influence of Atmospheric Correction and Number of Sampling Points on the Accuracy of Water Clarity Assessment Using Remote Sensing Application.” Journal of Hydrology 401 (3–4): 203–220. doi:10.1016/j. jhydrol.2011.02.023. USGS. 2013. Product Guide: Landsat Climate Data Record (CDR) Surface Reflectance. Department of Interior U.S. Geological Survey. Accessed May 21, 2013. http://landsat.usgs. gov/documents/cdr_sr_product_guide.pdf Veefkind, J. P., J. F. de Haan, E. J. Brinksma, M. Kroon, and P. F. Levelt. 2006. “Total Ozone from the Ozone Monitoring Instrument (OMI) Using the DOAS Technique.” IEEE Transactions on Geoscience and Remote Sensing 44 (5): 1239–1244. doi:10.1109/TGRS.2006.871204. Vermote, E. F., and N. Saleous. 2007. LEDAPS Surface Reflectance Product Description. Version 2.0. College Park: University of Maryland Department of Geography. Accessed April 10, 2013. http://ledaps.nascom.nasa.gov/docs/pdf/SR_productdescript_dec06.pdf Vermote, E. F., N. E. Saleous, C. O. Justice, Y. J. Kaufman, J. L. Privette, L. Remer, J. C. Roger, and D. Tanré. 1997. “Atmospheric Correction of Visible to Middle-Infrared EOS-MODIS Data over Land Surfaces: Background, Operational Algorithm and Validation.” Journal of Geophysical Research 102 (D14): 17131. doi:10.1029/97JD00201.
Downloaded by [UTSA Libraries] at 05:57 08 September 2014
International Journal of Remote Sensing
21
Vermote, E. F., D. Tanre, J. L. Deuze, M. Herman, and J.-J. Morcette. 1997. “Second Simulation of the Satellite Signal in the Solar Spectrum, 6S: An Overview.” IEEE Transactions on Geoscience and Remote Sensing 35 (3): 675–686. doi:10.1109/36.581987. Vincent, R. K., X. Qin, R. M. L. McKay, J. Miner, K. Czajkowski, J. Savino, and T. Bridgeman. 2004. “Phycocyanin Detection from LANDSAT TM Data for Mapping Cyanobacterial Blooms in Lake Erie.” Remote Sensing of Environment 89 (3): 381–392. doi:10.1016/j.rse.2003.10.014. Wong, M. S., J. E. Nichol, and K. H. Lee. 2011. “An Operational MODIS Aerosol Retrieval Algorithm at High Spatial Resolution, and Its Application over a Complex Urban Region.” Atmospheric Research 99 (3–4): 579–589. doi:10.1016/j.atmosres.2010.12.015. Wu, J., and M. E. Bauer. 2012. “Estimating Net Primary Production of Turfgrass in an UrbanSuburban Landscape with QuickBird Imagery.” Remote Sensing 4 (12): 849–866. doi:10.3390/ rs4040849. Yang, X., and C. P. Lo. 2000. “Relative Radiometric Normalization Performance for Change Detection from Multi-Date Satellite Images.” Photogrammetric Engineering and Remote Sensing 66: 967–980. Yuan, J., and Z. Niu. 2008. “Evaluation of Atmospheric Correction Using FLAASH.” In Paper Presented at International Workshop on Earth Observation and Remote Sensing Applications, Beijing, June 30–July 2, 1–6. IEEE. doi: 10.1109/EORSA.2008.4620341. Zhang, X., X. Zhao, G. Liu, Q. Kang, and D. Wu. 2013. “Radioactive Quality Evaluation and Cross Validation of Data from the HJ-1A/B Satellites' CCD Sensors.” Sensors 13 (7): 8564–8576. doi:10.3390/s130708564.