[gabrielc]

If processing is not perfect (e.g. improper 1st data point scaling in the indirect dimension, which depends on the existence of a non-zero 1st order phase correction), the baseline (in other words the noise ) can have an offset from 0. This can be spotted for example when looking at 1D traces in the indirect dimension of the ROESY or NOESY spectrum: it looks like the average noise line for the most intense peaks (e.g. diagonal peaks or methyl peaks) lies above (or below) the true 0 line (i.e. the average noise line of the traces that don't contain strong peaks).

This results into a larger relative error in peak volumes compared to peak intensities. Say the average noise level is offset above (below) 0, it will add (subtract) a large quantity to the peak volume because it's close to the peak base, which is broad (basically the additional volume added/subtracted will be be approx. the noise average times the broad area of the peak base). The error propagated in the intensity is only the noise average. Again, the noise average is nonzero because the noise is artificially above or below the true 0 baseline.

Since this error is non-uniform (it applies only to cross peaks that align in F2 frequency to the strong peaks, or ultimately only to peaks with a baseline offset), it may decrease significantly the overall accuracy of the resulting NOE constraint set.