If your data was output to integers, then you must have specified the 'Scale output ot integers' option in the tool. This allows you to also specify a scale factor. The scale factor used here is the scale factor you should use in FLAASH. Since your data is ranging from 0-65397, I am guessing the scale factor must have been around 10000. You should only need a single scale factor for your bands for WV-2.

Thanks MariM for your reply. I think what I did was leave the default scale factor that was populated in the WV2 radiance calibration tool when I converted the raw data to radiance. The default value was set to "1000". So given that "1000" was used for the radiance conversion, I should now use "1000" for the scale factor in FLAASH, correct? Im guessing the important thing to note here is that whatever scale factor is used in the radiance conversion should be the same number to use in FLAASH, correct? Thanks for your help.

1000 is the default scale factor used in the WV-2 radiance calibration tool. However, you stated your output values from this are ranging up to 65000 which would be very unusual. When I calibrate an 11-bit WV-2 dataset (scaled by the default of 1000) the resulting radiance values hit maximums of around 15-16k: Basic Stats Min Max Mean Stdev Band 1 0 15420 2749.485174 754.241751 Band 2 0 15076 2061.628699 708.290648 Band 3 0 16302 2439.960502 1184.812554 Band 4 0 9160 733.507164 470.409779 Band 5 0 2350 284.642742 213.030039 Band 6 0 2335 200.275865 141.154531 Band 7 0 2611 483.519437 377.490965 Band 8 0 4651 322.171884 260.378120 So I am not sure about your scale factor. Is your original data unsigned integer as well?

The original raw data is: BSQ, unsigned integer, Host (Intel) with the following stats: Basic Stats Min Max Mean Stdev Band 1 154 2047 359.734425 47.509326 Band 2 42 2047 233.401740 54.037175 Band 3 35 2047 325.443517 97.285755 Band 4 1 2047 356.826447 147.028451 Band 5 1 2047 190.430847 95.754647 Band 6 1 2047 477.903502 128.446759 Band 7 1 2047 416.777984 143.497888 Band 8 1 2047 645.531369 214.825358 I reprojected the raw data with ArcGIS and the data is now: BIP, unsigned integer, Host (Intel). After checking the data values they are the same as the original data. Then calibrated the reprojected data to radiance to get a file in: BSQ, unsigned integer, Host (Intel) with the following stats: Basic Stats Min Max Mean Stdev Band 1 0 40229 6258.267299 2416.108147 Band 2 0 65397 6783.284861 2956.800282 Band 3 0 44326 6233.401688 2992.731380 Band 4 0 37277 5743.762234 3258.262081 Band 5 0 65359 5427.280122 3500.901520 Band 6 0 31581 6524.641586 2996.467973 Band 7 0 42447 7656.896297 3930.425147 Band 8 0 18584 5191.191779 2619.618109 Then I converted that data to BIL for use in FLAASH, checked the stats and data values did not change from those above. Do you think there is an issue somewhere? As a test I ran the WV2 calibration tool to convert the raw to radiance again and made sure to specify the default “1000” scale factor and I got the same resulting data values as above.

I find it strange that all your band minimums are not 0 since all of the WV-2 data I have seen has some background pixels which are set to 0. All the WV-2 datasets I have seen have all the band minimums set to 0 and the maximums set to 2047 for 11-bit data. Curious if you would get different results if you first calibrate the original TIFF or NITF data in ENVI prior to reprojection. You could of course reproject in ENVI. Are you using ENVI 4.8?

Yes I am using 4.8. I originally was planning on using ENVI for reprojection but when I attempted to do this the resulting output was not satisfactory. Converting from WGS84 latlong to NAD83UTM10N using nearest neighbor with zero edge and maintain extent of image when pixel size change and a combination of methods such as triangulation, RST, and rigorous the output cut off the bottom portion of the scene where zero padding should have begun for the new position of the image. The zero padding on the top portion of the image was fine but the bottom didn’t work. Instead of trying to troubleshoot I just ran it in ArcGIS. Maybe I overlooked something simple there and could have fixed it. Ill try calibrating the raw data now without reproj.

The result of calibrating the raw data in its original projection is the following: BSQ, unsigned integer, host (intel) with stats: Basic Stats Min Max Mean Stdev Band 1 3026 40229 7069.698920 933.676617 Band 2 59 65397 7666.422980 1774.679377 Band 3 758 44326 7047.123870 2106.612524 Band 4 18 37277 6497.991427 2677.455103 Band 5 32 65359 6143.296022 3088.729015 Band 6 15 31581 7373.023980 1981.657390 Band 7 21 42447 8642.465576 2975.626438 Band 8 9 18584 5860.488802 1950.302430

Ok then I will assume the values are good and the scale factor for FLAASH should be 1000. They look a little higher than what I have seen before but perhaps it is the characteristics of your data that make the pixel values higher. I would go ahead and try the FLAASH process on the dataset. Remember, the output from FLAASH is also scaled integers which are scaled by 10,000.

Ok thanks for your help. I know this is sightly off topic but since I am new to FLAASH, for worldview 2 data I am unsure what bands to set the KT upper and lower channels to for aerosol retrieval. FLAASH wont run unless they are specified. I would of thought that if they were not defined by the user that the defaults for the sensor would be used. The sensor info for FLAASH has been set to WV2. Do you have any advice on this or can refer me to material? Thank you for your time.

Unfortunately you cannot do an aerosol retrieval with WV data because it has no bands in the 2100nm spectral range, which is required. So in this case you will need to turn the aerosol retrieval to 'None' and use an initial visibility estimate (the default is usually fine). You will also need to set the water retrieval to 'No' because there is not enough spectral resolution in WV to perform this correction. Your image would need a spectral resolution of 15nm or better. The description of these parameters and their requirements can be found in the ENVI help under ENVI Modules ->Atmospheric Correction.

Thank you MariM for your time with this issue and quick responses, I appreciate it. FLAASH is able to run now.

Hi sablanchard! You need to be careful using ENVI when convert DN to radiance for WV-02 data (at least for ENVI 4.7 or previous). ENVI calibration utilities is not for WorldView-2, but for WorldView. To convert DN to radiance for using in FLAASH module (results in microW cm-2 nm-2 sr-1) , you need use band math: (float(b1)*abscalfactor_b1/effectiveBandwidht_b1)/10 Abscalfactor and effectiveBandwidht for each band are avalaible in IMD file, with the image. Then, when you use the FLAASH, you must use factor scale: "1". I hope have been usefull to you! Good luck! Laura

Hi, thanks for the previous post I have only one question. Why divide by 10 the equation? Thanks and good luck in everything Rene

Hi Rene, sorry for the delay in my response! When you apply the equation float(bx)*absCalFactorBx/effective bandwidthBx the result will be in W m-2 sr-1 microm-1, but units required by flaash are: microW cm-2 nm-1 sr-1. For this reason you need divide by 10. Kind regards! Laura

mlzoffoli wrote: Hi sablanchard! You need to be careful using ENVI when convert DN to radiance for WV-02 data (at least for ENVI 4.7 or previous). ENVI calibration utilities is not for WorldView-2, but for WorldView. To convert DN to radiance for using in FLAASH module (results in microW cm-2 nm-2 sr-1) , you need use band math: (float(b1)*abscalfactor_b1/effectiveBandwidht_b1)/10 Abscalfactor and effectiveBandwidht for each band are avalaible in IMD file, with the image. Then, when you use the FLAASH, you must use factor scale: "1". I hope have been usefull to you! Good luck! Laura Sorry to dig up a two year-old thread, but I have a very similar question. I want to use FLAASH in ENVI 4.8 to convert my WorldView-2 images to TOA reflectance. However, I received images in product level LV3D, which are dynamically range adjusted and according to the DigitalGlobe imagery guide do not have a absolute calibration factor, but there is an effective bandwidth. Does this mean that I cannot use FLAASH or is there a work around (plus the stupid question being if this conversion is even necessary)? Thanks.

I am afraid DRA images are not appropriate for calibration. It really works best with 11-bit data that have not been adjusted. However, you could use Quac, which uses a different method for estimating atmospheric effects but is typically within +/-15% of the physical model used in FLAASH. It is important to mask out background pixel when using Quac so be sure to include a mask file for best results.

Thank for you the response. I will give QUAC a go.