How to define the spatial distance ε for EDR and LCSS similarity algorithms?

Let’s assume a dataset with human trajectories like Geolife. I want to measure the similarity of two random trajectories from this dataset using the EDR or/and LCSS algorithm(s).

I wonder, which factors to consider in order to define the value of spatial distance threshold ε used by EDR and LCSS?

I suppose that if I use a different dataset, like Hurricanes or Animals movements, this threshold should be different as the movement pattern is different. The speed of a hurricane is different than the speed of a walking human.

Also, the sampling rate is something to take into consideration?