Davide, I highly recommend against using QUAC for Landsat imagery- there aren't enough spectral bands, and you can get inconsistent results from neighboring rows in the same path on a different date. I suggest using FLAASH. Surface reflectance needs to be calculated from radiance, not TOA reflectance- both QUAC and FLAASH will produce poor results if you run them using the latter. Guy

I think Quac is fine to use with multispectral imagery and it is created by the same developers of FLAASH. It is estimated to be within +/-15% of the physical model of FLAASH and does not require the user know all of the parameters or units of their dataset. You can actually use DN, radiance or TOA reflectance in Quac. My guess is that your input data has values of 0 in some of the bands. You will need to create a mask for all bands that covers all 0s (or negative values) in all bands. You can use the Build mask tool in ENVI for this. Then this mask should be used within Quac.

We noticed some issues with QUAC and Landsat with a pair of adjacent rows from the same path and acquisition date, specifically that the reflectance values did not match up well, as opposed to with FLAASH where they agreed across the row borders. I think that QUAC requires at least 12 different materials and does best with hyperspectral data, whereas FLAASH is better for multispectral datasets.

That is good information to know. Both FLAASH and Quac work best with hyperspectral (more information) but in most cases Quac should work well - especially if you don't know the proper parameters for FLAASH. FLAASH can actually provide worse results in cases where bad information is given. Quac does not work well with scenes that have little variability in materials - as you mentioned. It requires a minimum of 10 'endmembers' or unique materials in the scene.

Thank everybody for your useful information. I have aset of several Landsat scenes at different years. I want to select the best ones and replace pixels with clouds with data from other years and then mosaick to obtain a bigger scene, no matter if the years differ, the months is the same. Gathering the data for FLAASH input is very difficult in my case because of the different years and because I work on a remote area in the mountains of China... I tried to mask the values: I select 0 as ignore value in the header of the input scene and then run QUAC again, setting the ignore value as mask, both with DN and reflectance. Results: DN works also without ignore value and the values after QUAC differ very slightly. Reflectance now works better, band 5 has a very bad result and some bands have negative values (probably coming from the mask. Note that after calibrating the original landsat scene to reflectance values, I can't see the values in the stats... only ********** but I can see regular values in the image display. I performed band math (bx>0) and then ignore value 0 before running QUAC. So the atmospherically corrected values differ when I input DN or reflectance, but I guess they should be the same... I don't know what's worng with my reflectances... for the moment the best results are still using DN with mask (or even without...) Kind regards Davide

FLAASH really isn't that complicated and is better suited for multispectral sensors like Landsat TM. Yes, it's extra work compared to QUAC, but the parameters are pretty easy to determine. Here is what I would do if I were you: 1. Read the MTL file into ENVI. 2. Run Landsat calibration to radiance. 3. Convert the radiance image from BSQ to BIL, flagging "convert in place" to conserve disk space. 4. From the MTL file, determine the scene extents and center. 5. In Google Earth, glean information on elevation, and use a mean value for your target. 6. In FLAASH, set the scaling factor to 10 for all input bands. Input the various sensor and view parameters. Acquisition times, etc. can be found in the MTL file. I would assume a rural aerosol value unless you have better information. Do not use water retrieval as Landsats 1-7 lack the appropriate bands. The model atmosphere is dependent upon the surface temperature. Also, there is no need to tinker with the advanced settings. From there you just need to run it.

Thank you for the tips :) Settings for FLAASH are quite difficult for my area, elevation range is very variate (mountainous area in Yunnan, China). I'm trying to use it and my first results show a high number of negative values, especially in the shorter wavelenghts. Concerning my progress with QUAC, I started again from the beginning and finally got the best results with reflectances, edges of the scenes cut and 0 values ignored. But I still can't understand why after QUAC values from images differ when usinf reflectances or DN as input... Best regards Davide

Have you tried using LEDAPS (http://ledaps.nascom.nasa.gov/tools/t...)? You'll need a Linux machine (can be a virtual machine running under another OS), but it's a 6S-based atmospheric correction tool specifically designed for Landsat. Compiling it and running it can be a little buggy, but once it works, it does a good job. You will, however, need to update the ancillary files regularly for it to work with new imagery.