Negative thresholds

In some cases, noisy pixels can produce S-Curves where the fit outputs negative values. This renders the dynamic scale of the threshold map useless:

I guess the problem should be treated by its root and negative values from the S-Cruve fit should be caught?