9 ene 2022

Scale and linear / Scale and quadratic

Scale and linear and Scale and quadratic are two of the scatter corrections that we can find in Win ISI 4, and they are used to remove as much as possible the multiplicative scatter we found in the NIR spectra. Two are the packages that have a detrend function in R, so depending on the case of the multiplicative effect (linear or quadratic, we can use the pracma or prospectr packages.

The detrend function of the Prospectr package is equivalent to the math treatment in WinISI 4 called "scale and quadratic” and combine the SNV treatment with the remove of a quadratic trendline in the spectra. In the case of the combination of SNV and Detrend from the Pracma package we obtain the Win ISI math treatment "Scale and Linear" and in this case we remove the linear trend line due to the scatter. We can try both to check which one gives the better performance.

I show the spectra of both treatments applied to the soil LUCAS database spectra filtered for the Spanish soils:




Samples are coloured according the soil type (cropland, grassland, woodland,....). The jump in the detector change it is common due that we treat individually the math treatment for the NIR and VIS ranges (trend are different in VIS and NIR regions).



1 comentario:

  1. Really interesting posts here as always...
    one comment: the jump around the 1100 nm due to detector change can be corrected before any preprocessing using splice correction (e.g. using the prospectr::spliceCorrection() function in R). After this correction, the jump will disappear from the plots fo the preprocessed spectra

    ResponderEliminar