“…We limit our discussion mostly to homogeneous HMMs, which means that the parameters, which describe state transition properties and noise distribution, are constant in time. Inhomogeneous HMMs, see, for instance, (Diehn et al 2019), are rarely used, as they are computationally more challenging and theoretical guarantees for parameter estimates are much harder to prove. As JULES detects short events, but finds many small events, which are most likely false positives, at areas of high conductance and high var-iance (see, for instance, the idealization of the observations around 0.36 nS in the middle panel).…”