X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 29 Apr 2015 07:10 AM by  anon
spectral resampling to sensor resolution
 1 Replies
Sort:
You are not authorized to post a reply.
Author Messages

anon



New Member


Posts:4
New Member


--
29 Apr 2015 07:10 AM
    I have a question regarding the options for spectral resampling in ENVI and I would like to better understand when ENVI does in the background. For an image with defined sensor type and defined wavelengths per band, I can perform a spectral resampling to a 'pre defined filter function'. These filter functions can be found in the envi install directory and seem to be identical to the sensor specific spectral response functions. What happens exactly computationally when I perform spectral resampling of e.g. a Landsat 7 ETM+ image using the Spot4 spectral response function? The result is an image with only the corresponding number of channels. Obviously the images only provide a single value for each band and nothing is known how the reflectance is distributed over the wavelength that is covered by a given band... I suspect that the overlap between both spectral response functions is used as a filter - but I would appreciate if someone could provide some additional details on what happens here. How could I recreate this process computationally in IDL (..without the Envi doit options)? Thanks, Patrick

    MariM



    Veteran Member


    Posts:2396
    Veteran Member


    --
    29 Apr 2015 08:04 AM
    The way ENVI does spectral resampling depends on the input information you have to give it. If you have both band center information and also FWHM data for each band, ENVI will use those to fit a Gaussian model to each band (to model the true band pass), and then that Gaussian model is integrated to find the area under the curve. If you use the filter function (spectral response function), then ENVI will use that data. The curve/response is divided by the calculated area to get the final value for the output band. This makes the band pass have unit area beneath its curve and the mean of the bands stays invariant. ENVI uses interpolation (Interpol) to accomplish this. Additionally, ENVI will apply 'weights' to the Gaussian curve when down-sampling (going to fewer bands).
    You are not authorized to post a reply.