Updated Landsat 8 Reflectance Tutorial w/ Imagery
GIS AG MAPS Image-Based DOS Surface Reflectance Methods & TOA
Landsat surface reflectance is available for free from the USGS, though it can be beneficial to also be able to convert to surface reflectance independently; for example, surface reflectance may not available for a particular scene or it may be needed as soon as possible. GIS Ag Maps offers simple and fast dark-object subtraction (DOS; Chavez ) methods - you do need software that will enable you to view a raster attribute table. Results published in Remote Sensing of Environment based on using GIS Ag Maps DOS SR for NDVI can be viewed here.
Data Download for Surface Reflectance (SR) Tutorial Includes: 1) Landsat 8 DN imagery for tutorial, 2) corresponding USGS algorithm surface reflectance imagery for comparison (in native integer raster format; divide by 10,000 for surface reflectance units), and 3) a composite RGB surface reflectance image (based on tutorial values). See below for more information.
Tutorial Data Location (orange); Red Band Scatter (.032949)
July 30th, 2014; East Central Illinois, USA (orange extent; 23.7 km x 15.0 km)
Landsat surface reflectance is available for free from the USGS but converting to surface reflectance independently can be useful if, for example, surface reflectance is not available for a particular scene (for whatever reason) or if reflectance is needed as soon as possible. You need GIS software for the data. Free Quantum GIS (opens in a new tab to a page on this website to access and download Quantum GIS) can be used; or you can use other software, such as ArcGIS. See the Landsat 8 Surface Reflectance Results page for results information.
This tutorial is based on the original tutorial but has new imagery and a specific recommended method (GIS Ag Maps Preferred Method2 Surface Reflectance [PM2 SR]). Important to know: Image-based atmospheric correction for Landsat 8 imagery is much different than it is for other Landsat imagery.
The download above includes Landsat 8 digital number (DN) imagery for bands 2, 3, 4, 5, and 6 (blue, green, red, NIR, and SWIR, respectively), corresponding Landsat 8 USGS Surface Reflectance imagery for the same bands, and a composite bands image produced by PM2 SR imagery (click on the composite bands image with an Identify Tool and reflectance for the visible bands can be seen). The LandsatLook image is not included. The imagery is from the fourth area from the top on the Landsat 8 Surface Reflectance Results page.
* GIS Ag Maps DOS Preferred Method Surface Reflectance Steps (DOS SR & DOS2 SR) *
(See table below tutorial steps for converted surface reflectance values after completing tutorial)
GIS Ag Maps Surface Reflectance Methods are outlined in the steps below and use DOS with relative scatter based on band 4 (red). One of the two methods deducts relative scatter from the visible bands and NIR band, while the other just deducts relative scatter from the visible bands and uses TOA reflectance for band 5 (NIR) because solely using TOA for the NIR band retrieves closer values to those derived by the USGS Landsat 8 Surface Reflectance Algorithm (based on research here as is shown in the Landsat 8 Surface Reflectance Results page). Also, of of the methods uses a slightly modified band 3 (Green) relative scatter value to produce values closer to the USGS values. This is all explained below and through the links. Keep in mind, this is a very simple and straightforward process after it is completed a couple times. The relative scatter table can be accessed here: Landsat 8 Continuous Relative Scatter Lookup Table (opens in new window).
Steps to calculate the Preferred Method surface Landsat 8 with the Continuous Relative Scatter Lookup Table are as follows:
1) Apply the following equation to the entire raster for each band: ([DN x .00002] - 0.1).
2) Divide the values from Step 1 by the cosine of the solar zenith to calculate Top of the Atmosphere (TOA) reflectance. The solar zenith = (90 - solar elevation); the scene center solar elevation is given in the .MTL file can be used (or a more local value can be used if available). The cosine of the solar zenith angle for the data in the tutorial is 0.878335597. The COST Method (Chavez, 1996) where the square of the cosine of the solar zenith is applied in the denominator, while very important for previous Landsat imagery, should not be applied for Landsat 8 - the orginial DOS method (Chavez, 1988), where the cosine is not squared, should be applied. TOA is not surface reflectance but should be used for bands 6 & 7 (SWIR) because there is virtually no scatter. USGS Landsat 8 Algorithm band 5 surface reflectance is virtually the same as TOA reflectance, though there are small amount included in the Relative Scatter Lookup table.
Steps 3, 4, 5 & 6 Describe Methods to Establish and Deduct Scatter to Derive Surface Reflectance
3) There are two recommended methods for deducting scatter (haze) from visible bands - either deduct scatter based on individual band values or apply relative scatter based on one band (the red band is recommended here). In any case, establish the scatter (dark object) DN by determining the Lowest Valid Value. If you calculate individual band scatter, make sure there is a high power correlation (or state the correlation) between band center wavelength (in micrometers, blue=0.480; green=0.560; red=0.655; NIR=0.865 [see the Noteworthy Information section below about NIR scatter]) and band Lowest Valid Value reflectance; producing a power line is a simple function that can be done in Excel (or other software) where you insert a scatter chart and add the power line and correlation value (a power exponent should be from about -2 to -4). If applying relative scatter, use band 4 (red) as the starting scatter value for the REVISED Relative Scatter Lookup Table. The lowest valid band 4 value for the tutorial data is 6447. (The Lowest Valid Value is a from an entire scene; the tutorial data is part of a scene. The Lowest Valid Value of 6447 happens to be from an area of the scene outside the extent of the tutorial data.)
4) Apply the following equation to the scatter DN (the Lowest Valid Value from the red band) from Step 3: ([DN x .00002] - 0.1).
5) Divide the scatter value from Step 4 by the cosine of the solar zenith (0.878335597) to convert to reflectance; the reflectance value of the band 4 starting scatter for the tutorial data is 0.032949 (hazier than normal day).
6) If calculating Lowest Valid Value scatter for each individual band, simply subtract that amount from TOA (Step 2) to calculate surface reflectance. If applying relative scatter, acquire relative scatter for the visible bands from the Landsat 8 Continuous Relative Scatter Lookup Table and deduct them from the visible band TOA reflectance from Step 2 to calculate surface reflectance. There are three scatter deduction methods described in the following paragraph (two of which are included in the tutorial) - at this point in time, DOS SR then DOS1 SR are the recommended methods (there is currently not enough background research to recommend the the DOS2 SR method, though information is included in the L8 SR Results page).
Noteworthy Information: We have assessed many images and compared individual band Lowest Valid Value scatter values (relative to red starting band scatter) in the the Relative Scatter Lookup Table - we have found there are inconsistencies, of course, between calculated individual band scatter and relative values. Check back here for values listed comparing the most recent individual band Lowest Valid Value and relative scatter.
For this tutorial, the scatter values relative to the starting band 4 value of 0.032949 are used and are as follows: 0.08121 (Blue), 0.04961 (Green), 0.04149 (Green2; explained below), and 0.01895 (NIR). (NIR2 is not calculated in this tutorial but has recently been recommended based on research here and is described in the previous paragraph). Use the original band 4 value of 0.032949 instead of the slightly different band 4 lookup value 0.03290. Green2 is a relative green scatter value modified slightly from the original scatter table value (though both are listed in the Relative Scatter Table, which was developed well before Landsat 8 surface reflectance was made available by the USGS). Green2 derives DOS2 SR (formerly named PM2 SR) values, and is based on the research shown and described in the Landsat 8 SR Results page that shows how the modified green values retrieve surface reflectance overall closer to USGS Landsat 8 algorithm green values. (Green2 is included for information purposes and is not recommend as there has not been enough background research completed.) DOS2 SR also uses Top of Atmosphere (TOA) reflectance for Band 5 (NIR) because USGS Landsat 8 Algorithm band 5 surface reflectance values are nearly identical to TOA. As the DOS2 SR method focuses on retrieving values closer to the USGS Landsat 8 algorithm reflectance values, the DOS SR (formerly named PM1 SR) method focuses on the power relationships between scatter amounts and band center wavelength the should exist (power relationships between scatter and band center wavelength in the Relative Scatter Lookup Table decrease more with DOS2 SR as scatter becomes higher). (DOS1 SR is shown in the Landsat 8 SR Results page and is the same as DOS SR except that TOA is used for NIR surface reflectance.) Choose the method you prefer based on these main points, but be consistent with the method used. These methods are now based on the Revised Lookup Table.
PLEASE READ THE FOLLOWING REGARDING ONE-PERCENT DARK OBJECT REFLECTANCE, WHICH IS NOT APPLIED FOR LANDSAT 8.
One-Percent Dark Object Reflectance Background
One-percent surface reflectance is the process of deducting .01 from the established scatter (haze) in order to acquire one-percent reflectance for the low value. This has been shown to improve accuracy of retrieved surface reflectance when applied to Landsat 5 and 7. Based on the dark object selection method applied here for Landsat 8 (Lowest Valid Value Method), analysis shows that it is better to bypass this step (not apply one-percent reflectance) and simply deduct scatter (haze) from the different bands based on the Relative Scatter Table. This is likely due to the fact that the Lowest Valid Value scatter amount is based on a different histogram method than the one used for previous Landsats (which relied on finding the low-end abrupt increase in the histogram) - the change in the histogram methods is mostly due to the fact that the Landsat 8 has greater bit depth than previous Landsat satellites.
Tutorial Reflectance Units
(DN values should be converted to the reflectance units below; USGS raster reflectance values are in the native downloaded 4-integer format [divide by 10,000 for reflectance units])
|Landsat 8||USGS SR||DOS SR||DOS2 SR||TOA|
|Landsat 8||Stan. Dev.||Stan. Dev.||Stan. Dev.||Stan. Dev.|
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.