The light reflected from the Earth’s surface (surface reflectance) is important for monitoring environmental resources – such as agricultural production and mining activities – over time.
We need to make accurate comparisons of imagery acquired at different times, seasons and geographic locations. However, inconsistencies can arise due to variations in atmospheric conditions, sun position, sensor view angle, surface slope and surface aspect. These need to be reduced or removed to ensure the data is consistent and can be compared over time.
What this product offers
This product has been corrected to account for variations caused by atmospheric properties, sun position and sensor view angle at time of image capture.
These corrections have been applied to all satellite imagery in the Sentinel-2 archive. This is undertaken to allow comparison of imagery acquired at different times, in different seasons and in different geographic locations.
These products also indicate where the imagery has been affected by cloud or cloud shadow, contains missing data or has been affected in other ways. The Surface Reflectance products are useful as a fundamental starting point for any further analysis, and underpin all other optical derived Digital Earth Australia products.
This product eliminates pre-processing requirements for a wide range of land and coastal monitoring applications and renders more accurate results from analyses, particularly those utilising time series data.
Such applications include various forms of change detection, including:
- monitoring of urban growth, coastal habitats, mining activities, and agricultural production
- compliance surveys
- scientific research emergency management
The standardised SR data products deliver calibrated optical surface reflectance data across land and coastal fringes. SR is a medium resolution (10/20/60 m) grid based on the Sentinel 2 MSI archive and presents surface reflectance data in 10, 20 and 60m pixels.
Radiance measurements from EO sensors do not directly quantify the surface reflectance of the Earth. Such measurements are modified by variations in atmospheric properties, sun position, sensor view angle, surface slope and surface aspect. To obtain consistent and comparable measures of Earth surface reflectance from EO, these variations need to be reduced or removed from the radiance measurements (Li et al., 2010). This is especially important when comparing imagery acquired in different seasons and geographic regions.
The SR product is created using a physics-based, coupled Bidirectional Reflectance Distribution Function (BRDF) and atmospheric correction model that can be applied to both flat and inclined surfaces (Li et al., 2012). The resulting surface reflectance values are comparable both within individual images and between images acquired at different times and/or with different sensors.
No data values and pixels that were determined as not view-able in NBART products are expressed as -999.
Accuracy and limitations
Atmospheric correction accuracy is dependent on the quality of aerosol data available to determine the atmospheric profile at time of image acquisition.
BRDF correction is based on low resolution imagery (MODIS) which is assumed to be relevant to medium resolution imagery such as Sentinel-2 MSI. BRDF correction is applied to each whole Sentinel 2 MSI tiles and does not account for changes in land cover. It also excludes effects due to topographic shading and local BRDF. This algorithm is dependent on the availability of relevant MODIS BRDF data.
Topographic shading correction algorithm is designed for medium resolution imagery and assumes that hill slopes are resolved by the sensor system (Li et al., 2012). The algorithm assumes that: a. BRDF effect for inclined surfaces is modelled by the surface slope and does not account for land cover orientation relative to gravity (as occurs for some broad leaf vegetation with vertical leaf orientation).
Atmospheric and BRDF Correction
The algorithm was validated using Landsat data. As detailed in Li et al. (2010), the atmospheric and BRDF correction algorithm was validated in three parts:
1. Validate combined atmospheric and surface BRDF correction using field reflectance measurements at two very different sites, Gwydir, NSW, and Lake Frome, SA - correlation (measured as r) between corrected image values and field data was >0.95.
2. Validate surface BRDF correction using data from image overlap areas of adjacent paths acquired one week apart in northeast Queensland - normalised reflectance factor was very close in corrected images, but not in original images, and difference in reflectance factor values between corrected and uncorrected images can be >0.05.
3. Cross-validate Landsat TM data for accuracy of spectral reflectance using the MODIS reflectance product for Lake Frome correlation (measured as r2) between corrected Landsat TM image values and MODIS reflectance product was 0.93-0.97 in all bands except Landsat TM band 5, which was 0.90.
The results clearly show that the algorithm can remove most of the BRDF effect without empirical adjustment and that cross-calibration between the Landsat ETM+ and MODIS sensors is achievable.
The algorithm was validated using Landsat data. As detailed in Li et al. (2012), two high relief areas in southeast Australia (Australian Alps in northeast Victoria and the Blue Mountains in NSW) were used to test the algorithm against eight Landsat images with varying solar angles (seasons), and terrain types. Validation included four parts.
1. Visual assessment showed that the algorithm removed much of the topographic effect and detected deep shadows in all eight images.
2. An indirect validation based on the change in correlation between the data and terrain slope showed that the correlation coefficient between the surface reflectance factor and the cosine of the incident (sun) angle reduced dramatically after the topographic correction algorithm was applied. The correlation coefficient typically reduced from 0.80-0.70 to 0.05 in areas of significant relief.
3. Validated using land cover classification. The corrected surface reflectance can provide suitable input data for multi-temporal land cover classification in areas of high relief based on spectral signatures and spectral albedo, while the products based only on BRDF and atmospheric correction cannot.
4. Compared with two empirical methods based on the C-correction were used as well as the established sun-canopy-sensor SCS-method to provide benchmarks. The proposed method was found to achieve the same (better) measures of shade reduction without empirical regression.
Li, F., Jupp, D. L. B., Reddy, S., Lymburner, L., Mueller, N., Tan, P., & Islam, A. (2010). An evaluation of the use of atmospheric and brdf correction to standardize landsat data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 3(3), 257–270. https://doi.org/10.1109/JSTARS.2010.2042281
Li, F., Jupp, D. L. B., Thankappan, M., Lymburner, L., Mueller, N., Lewis, A., & Held, A. (2012). A physics-based atmospheric and BRDF correction for Landsat data over mountainous terrain. Remote Sensing of Environment, 124, 756–770. https://doi.org/10.1016/j.rse.2012.06.018
- Level 1 Satellite Imagery
- Systematic Terrain Correction Imagery
- Ephemeris Data
- Maximum Satellite View Angle
- Earth rotational angular velocity
- Earth-sun Distance
- Spectral Filter
- Solar Irradiance
- MODIS BRDF Shape Function
- BRDF Database
- SRTM DSM/DEM data
- Aerosol Optical Depth
- Monthly Ozone Imagery
- Total Water Vapour
- CO2 Concentration
- MODTRAN Seasons
- Extract metadata from data sources
- Calculate sun and sensor angles per pixel (Vincenty, 1975; Edberg and Oliver, 2013)
- Determine values for six base atmospheric parameters across each image scene
- Derive normalised surface reflectance for sun angle of 45°
OwnerCommonwealth of Australia (Geoscience Australia)
LicenseCC BY Attribution 4.0 International License
© Commonwealth of Australia (Geoscience Australia) 2015. Creative Commons Attribution 4.0 International License.