Target vs. reference band regression

You can only perform regression analysis on linked spectra. A target spectrum can only be linked to a single reference spectrum (and vice versa). When a target and reference spectrum are linked, the regression statistics (gain, bias, and so forth) are calculated.

Linking target and reference spectra

In either the Target or Reference tables, select a spectrum. A list of available spectra appears, in which you select the corresponding spectrum. When spectra are linked, the plot is updated, and the linked spectrum inherits the color characteristics of the defining spectrum. When you link target to reference, the target spectrum is the defining spectrum and the reference spectrum inherits the qualities (color) of the target spectrum. When you link reference to target, the reference spectrum becomes the defining spectrum and the target spectrum inherits the qualities of the reference spectrum. When two spectra are linked, the regression statistics gain and bias are calculated. To calculate the regression slope, intercept, correlation coefficient (r) and coefficient of determination (r*2), you must select two or more pairs of linked spectra.

Interpreting the regression statistics

The Target table displays the regression per signature across all bands and shows the gain, bias, and R2 statistics. The regression statistics are applied with the parameters defined by the header and the ATCOR parameters.

The gain is calculated with the header gain applied.
A good match will approach a value of 1.
The bias is calculated with the header bias applied.
A good match will approach a value of 0.
The R2, the determination coefficient, should converge to a value of 1.

Target versus reference band regression provides per-band regression across the selected spectra. This is per band between spectra.

The active gain is defined in the header for the selected band and used in the current computation.
The active bias is defined in the header for the selected band and used in the current computation.
The derived gain is the result of the computed regression for the selected band and is equal to 1 / active gain.
The derived bias is the result of the computed regression for the selected band and is equal to the active bias less the regression bias.
The regression slope for the selected band should approach a value of 1 for a good match.
The regression intercept for the selected band should approach a value of 0 for a good match.
The correlation coefficient (r) for the selected band should converge to a value of 1 for a good match.

Applying derived calibration coefficients

The derived calibration coefficients (Gain and Bias) will update the values stored in the file and channel metadata for the target file Gain and Bias for all bands.

Attention: Applying derived calibration coefficients requires advanced knowledge of working with data that has incorrect gain and bias values in the header file.

Goodness of "fit" of selected spectra

When linked spectra are selected in the Target or Reference table, they have an inherent quality of fit. The computed value represents the regression across all bands for all of the selected spectra. If no spectra are selected in either of the tables, all linked spectra are used in the computation. Nonlinked spectra are excluded.