You can use DN in Quac but I tend to always use radiance so that I can compare the results to FLAASH. When using Quac, it is important to use a mask of any background values or values of 0 or less. If you did not use a mask on these values, then Quac may return strange results. Negative reflectance values in water is not uncommon in either FLAASH or Quac because water tends to return very low radiance in visible and NIR so the pixels do not model well.

I got your point. I'm actually using the coefficient (gains and offsets) for the ETM+ sensor published in Chandler et al. 2009 B1 0.778740 -6.98 B2 0.798819 -7.2 B3 0.621654 -5.62 B4 0.639764 -5.74 B5 0.12622 -1.13 B7 0.043898 -0.39 If I'm correct in band 4 for a DN=5 the radiance value is 5 * 0.6398 *-5.7400 = -2.5412 W/(m2 sr μm). This value is negative: this meas very low radiance. I guess that even in FLAASH this conversion will produce negative values. According to you, negative values should be "replaced" with zero radiance: however my need is to use such low radiance pixels for my study. Working with DN would avoid reaching negative values and exploit the positive DN value. I have two questions 1) Do you think what I have summarized (working with raw DNs as input of QUAC) still represent a valid approach, when comparing results with field data, even if comparison with FLAASH is not possible? 2) Why only using radiance as input to QUAC can provide an output which is comparable to the results of FLAASH? Thanks in advance Salvatore Chander, G., Markham, B. L., & Helder, D. L. (2009). Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. Remote Sensing of Environment, In Press, Corrected Proof.