Now, the problem that we immediately face is that we don’t know this error. With lower tolerance values we would likely increase the false positive rate, i.e., the probability that we choose a knickpoint that is due to an artefact. More specifically, I would argue that one should choose tolerance values that are higher than the maximum expected error between the measured and the true river profile. I wrote that “the value of tol should reflect uncertainties that are inherent in longitudinal river profile data”.
So, what value of ‘tol’ should we choose? At the same time, we know that river profile data often has some erroneous values and we wouldn’t want to pick knickpoints that are merely artefacts of the data. The ‘tol’-parameter determines how many knickpoints the finder will detect and we might not want to miss a knickpoint. However, I’m still confused about how to translate this sentence into code, or in another word, how can I validate my choice of tol by some criteria based on codes? You said that “the value of tol should reflect uncertainties that are inherent in longitudinal river profile data”. Kai Deng from the GFZ Potsdam commented on my previous post on knickpointfinder and asked the question about the parameter ‘tol’: Which tolerance should I choose when using knickpointfinder?
