A relatively simple machine learning technique such as KNNs, could detect a complex pattern with a negative predictive value of 74%. But most importantly, this pattern is driven by the data, not by a prior geometry. If I were going for a parachute jump, I would prefer KNNs rather than ranges or ellipsoids!
In case that one diagnoses imbalance (or if the model predicts that the parachutist will fall in the sea), it would be useful to know how far the ionome is from the closest healthy island (or the closest path to swim towards the coast). I can compute a health index using the Euclidean distance from the closest healthy point (figure \ref{653216}).