Landsat 8 DOS (Chavez, 1988) vs. COST (Chavez, 1996) Soybean Field
Visible & NIR Surface Reflectance Comparison
Below, Landsat 8 NIR imagery is shown on left (lighter shades are higher reflectance, scaled from ± 2 standard deviations from the mean). The Landsat 8 image date is 8/23/13; the corresponding photograph date (on the right) is 8/26/13. Surface reflectance (SR) is calculated with the Landsat 8 DOS Method as detailed in the Landsat 8 Imagery & Reflectance Tutorial page and used the Revised Relative Scatter Lookup Table (which is somewhat different than the current recommended Relative Scatter Table). The Landsat imagery was the most recent imagery released by the USGS at the time of this analysis. The COST method uses the square of the cosine of the solar zenith in the denominator when converting to SR, which increases SR for all solar elevations less than 90°.
IMPORTANTLY FOR THE COMPARISON HERE, NIR SR WOULD BE VIRTUALLY THE SAME NO MATTER WHAT SCATTER TABLE IS USED AND IT IS NEARLY THE SAME AS TOP OF ATMOSPHERE (TOA) REFLECTANCE IN ALL ENVIRONMENTAL CONDITIONS, AS EVEN IN HAZY CONDITIONS THERE WILL ONLY BE ABOUT .01 (1%) OF SCATTER TO DEDUCT.
For the soybean field above with the arrow, DOS NIR SR median and mode are both .60, realistic for this stage (earlier in August, SR for visible bands may have been slightly lower and surface reflectance for NIR might have been slightly higher); while, COST NIR reflectance median is .73 and mode is .74, both too high (n = 497 pixels; for scale, the image is 2.5 miles wide; north is ↑). See graph below for soybean reflectance information.
Retrieved SR values shown in the chart below represent well-documented, proper relationships for a soybean field at this stage (median values are shown on graph; mean values are nearly identical). The soybean field with reflectance values calculated has an arrow on it which approximates the aspect of the photo of the field. For scale, the NIR image is 2.5 miles wide (north is ↑). Though the COST method (Chavez, 1996; uses the square of the cosine of the solar zenith in the denominator to increase surface reflectance) has been shown to be more the more accurate and correct method to use to convert Landsat 5 & 7 imagery to surface reflectance, based on the NIR surface reflectance values below, the DOS method should be used for Landsat 8 atmospheric correction. (See below graphic for more details.)
Surface reflectance for above field for visible and NIR bands for reprocessed Landsat 8 imagery (most recently released imagery for area by USGS at the time of this analysis). On the y-axis is (Preferred Method) surface reflectance units (where, .05 = 5%, .50 = 50%, etc.). On x-axis is Landsat 8 band center wavelength, from left to right, blue, red, green and NIR.
SR
Landsat 8 band center wavelength, from left to right: band 2 (blue); band 3 (green); band 4 (red); band 5 (NIR)
There was no NIR scatter deduction necessary for this particular image, as the Revised Relative Scatter Table has no relative scatter deduction based on the red band when the red band value was less than .01 (though the current Recommended Relative Scatter Table has a NIR deduction corresponding to all starting red band scatter values; it was a very clear day). Nevertheless, if there were a corresponding relative scatter deduction listed, the value would be very minor and insignificant. As a result, in this particular case, DOS and TOA reflectance are the same.
The COST atmospheric correction method has been shown to be more accurate than DOS to derive reflectance for Landsat 5 (Chavez, 1996) and Landsat 7 has nearly identical bandwidths to Landsat 5; however, the COST method (which adds a factor in the denominator that increases reflectance) has not been validated with Landsat 8 data. Landsat 8 has a much different sensor and radiometric resolution than Landsat 5 (also, the NIR band has been refined to decrease atmospheric absorption). The COST method results in NIR reflectance that is too high for soybeans while the DOS method results in realistic reflectance (see below). The COST and DOS methods are described more in the Atmospheric Correction Guide and Landsat 8 Atmospheric Correction pages.
Pixels used for field for reflectance (n=497; north is ↑)
Stayed a minimum of 15 meters (half-pixel width or height) away from non-field areas (e.g. boundaries, channels, non-crop vegetation; tree stand in northwest seemed slightly more than 15 meters based on zooming in to background imagery) to help account for Landsat horizontal accuracy error.
(OGRIP, 2013)
Plot from Gitelson (2004) that shows soybean NIR reflectance
Landsat 8 Conversion to TOA Reflectance Equation (USGS, 2013)
OLI band data can also be converted to TOA planetary reflectance using reflectance rescaling coefficients provided in the product metadata file (MTL file). The following equation is used to convert DN values to TOA reflectance for OLI data as follows:
ρλ' = MρQcal + Aρ
where:
ρλ' = TOA planetary reflectance, without correction for solar angle. Note that ρλ' does not contain a correction for the sun angle.
Mρ = Band-specific multiplicative rescaling factor from the metadata (REFLECTANCE_MULT_BAND_x, where x is the band number)
Aρ = Band-specific additive rescaling factor from the metadata (REFLECTANCE_ADD_BAND_x, where x is the band number)
Qcal = Quantized and calibrated standard product pixel values (DN)
TOA reflectance with a correction for the sun angle is then:
where:
ρλ = TOA planetary reflectance
θSE = Local sun elevation angle. The scene center sun elevation angle in degrees is provided in the metadata (SUN_ELEVATION).
θSZ = Local solar zenith angle; θSZ = 90° - θSE
For more accurate reflectance calculations, per pixel solar angles could be used instead of the scene center solar angle, but per pixel solar zenith angles are not currently provided with the Landsat 8 products.
References
Chavez, P.S., Jr. 1996. Image-based atmospheric corrections–revisited and improved. Photogrammetric Engineering and Remote Sensing 62(9): pp. 1025 - 1036.
Chavez, P.S., Jr. 1988. An improved dark-object subtraction technique for atmospheric scattering correction of multispectral data. Remote Sensing of Environment 24: pp.459-479.
Gitelson, A.A. 2004. Wide dynamic Range vegetation index for remote quantification of biophysical characteristics of vegetation. Journal of Plant Physiology 161: pp. 165 –173.
OGRIP. 2013. Ohio Geographically Referenced Information Program. Website last updated in 2013. Imagery not from 2013. Downloaded imagery from viewer at: http://gis4.oit.ohio.gov/osiptiledownloads/osip2.aspx
USGS. 2013b. Using the USGS Landsat 8 Product. Cited at: http://landsat7.usgs.gov/Landsat8_Using_Product.php