The statistical distributions of particle velocities in turbulent flows of superfluid helium display non-classical tails at small enough scales. The shape of these tails is affected by uncertainties originating from the use of numerical differentiation for particle velocity computation.
We compare here the performance of a multi-point velocity estimator based on Gaussian blur kernels with that of a simple linear estimator. We find that the particle velocities obtained by the former are less noisy but, at the same time, the estimator parametrization may depend on the studied physical problem.