Maybe I should add for others, that I am aware of one free alternative, which is not enough for me unfortunately. GRASS module i.landsat.toar, which primarily converts DN values to TOA radiance or reflectance, has support for OLI sensor (in current stable GRASS version) and has an option to use several DOS-based methods of atmospheric correction. Some details of the correction can be found in the manual:
I still have to thoroughly test it, but the main reason I think it would probably not work well for me (multitemporal water quality parameters modeling on regional scale) is, that here in Czech Republic there is typically quite some spatial variability in atmospheric properties, and all the methods of this GRASS module are spatially invariable
Just to keep informed those still interested: I have developed a GRASS GIS script to do relative, spatially-variable atmospheric normalization based on regression between a reference and corrected image. If absolutely atmospherically corrected image is used as the reference, it is possible to get ground reflectance as result. Now in the phase of testing and preparing an article.
The open source ARCSI software developed at Aberystwyth university is capable of atmospherically correcting Landsat 8 data and other sensors to top of atmosphere and surface reflectance. Some details on usage are available in the following post:
Generally, you will need AOT and Aerosol from the nearest Aeronet station (aeronet.gsfc.nasa.gov/) to make those program work. Keep in mind that aerosols are fluctuent, and that even on a single scene it can vary quite a lot. So if your nearest Aeronet station is too far away, you will have to look for another solution for those parameters. Some meteo models produce aerols maps.
For the TIR band 10, the objective is to obtain the surface temperature; To do this, you must suppress the effect of the atmosphere int the thermal region. This is well described in this short conference paper of Julia Barsi: http://srv2.lemig.umontreal.ca/donnees/geo6333/Barsi_IGARSS03.pdf.
Thanks, Michel for the detailed answer. As for GRASS 6S based module, I would be very grateful for advice, if you have experience with it, or an article to point me to. I was trying to use it, it works somehow, but the results are always a quantized reflectance (i.e. not reflectance in 0-1 region, and it is not even some simple multiplied reflectance like LEDAPS product). There seems to be no clue anywhere, how is the reflectance quantized, or rather how to compute absolute reflectance from the pixel values in the result of the atmospheric correction.
For those still interested I have decided to publish preliminary version of the spatially-variable normalization script I mentioned in the comment from April 29. The testing and article preparation takes much longer than I had anticipated.
Edit: Since the post I uploaded several new releases, see https://www.researchgate.net/profile/Tomas_Brunclik/publications?pubType=dataset and find the latest one.
For data of Landsat 8, must apply the first step (convert DN to radiance) and the second step convert to athmospheric correction (convert to reflection)????
i saw some videos in youtube directly apply the second step ( convert to reflection), is that true ??
Malik, the i.landsat.toar has an option to do simple DOS (Dark Object Subtraction method, several variants of them in fact) image-based atmospheric correction. In this case it is one step process using just that command. Or you can use i.landsat.toar just to convert to TOA (top of atmosphere, or at-sensor) radiance or TOA reflectance, and then use i.atcorr to atmospherically correct the result in second step. i.atcorr is using the 6S algorithm. I have heard, that in GRASS 7 the problem of i.atcorr I described few posts above is adressed, but had no time yet to test it.
Landsat CDR contains more high-level products (surface reflectance, vegetation indices), and some of these may be obtained from several ordering interfaces. On EarthExplorer there is now available provisional surtface reflectance product. See
http://landsat.usgs.gov/CDR_ECV.php
http://landsat.usgs.gov/CDR_LSR.php
As for license and use constraints, it is mentioned in several places (for L7 example http://www.landsat.org/aboutcontent.html#price) that the Landsat data are now public domain, although I am not currently able to find this explicitly stated in the USGS official web pages. All the descriptions how to use and download the data and lack of permitted or not permitted uses specification may suggest as much, however. There are no use constraints as far as I know.
I just came across a few pages that may be interesting regarding the Ladsat data license. On NASA websites, there are remarks suggesting Landsat data are public domain, for example: http://landsat.gsfc.nasa.gov/?p=6111. But public domain is not "viral" license like some opensource licenses, which means anyone can legally get the data from a public domain source and then republish (after modification or unmodified) under any other license, even commercial or very restrictive one. For example Geoscience Australia is distributing Landsat data under CC: https://creativecommons.org/tag/landsat (not so restrictive, but still not public domain).
For data on USGS websites, which should include atmospherically corrected SR products in CDR archives or EarthExplorer, there should aply these guidelines: http://www.usgs.gov/laws/info_policies.html, where it is stated:
"USGS-authored or produced data and information are considered to be in the U.S. public domain ...
When using information from USGS information products, publications, or Web sites, we ask that proper credit be given. ...."
Public domain means no constraints at all. But to be on safe side, you can always ask regarding certain product.
Most imortantly it features orthogonal regression, which should work better than ordinary least squares for the type of data used. But I have to warn Windows users, the script does not work for me in WinGRASS. It may work for you, but most probably will not, as it seems to be caused by a problem in the Windows version of GRASS. Linux and Mac users should be fine.