A method to statistically characterize turbulent data with physically motivated parameters, illustrated on a centroid velocity map

TL;DR

This paper demonstrates how a physically motivated turbulence model can be used to statistically characterize and compare 2D and 3D turbulent data, using an automated parameter fitting approach on centroid velocity maps.

Contribution

It introduces a method to fit a turbulence model to observational data using MCMC, enabling statistical characterization and comparison of turbulent fields.

Findings

The model can reproduce the statistics of centroid velocity maps.

Parameter variations significantly affect the visual and statistical properties.

The approach is computationally efficient and suitable for systematic exploration.

Abstract

We investigate the potential of a recently proposed model for 3D compressible MHD turbulence (Chevillard et al. 2010; Durrive et al. 2021) to be used as a tool to characterize statistically 2D and 3D turbulent data. This model is parametrized by a dozen of free (intuitive, physically motivated) parameters, which control the statistics of the fields (density, velocity and magnetic fields). The present study is a proof of concept study: (i) we restrict ourselves to the incompressible hydrodynamical part of the model, (ii) we consider as data centroid velocity maps, and (iii) we let only three of the free parameters vary (namely the correlation length, the Hurst parameter and the intermittency parameter). Within this framework, we demonstrate that, given a centroid velocity map, we can find in an automated manner (i.e. by a Markov Chain Monte Carlo analysis) values of the parameters such that the model resembles the given map, i.e. which reproduces its statistics fairly well. Hence, thanks to this procedure, one may characterize statistically, and thus compare, various turbulent data. In other words, we show how this model may be used as a metric to compare observational or simulated data sets. In addition, because this model is numerically particularly fast (nearly 500 times faster than the numerical simulation we use to generate our reference data) it may be used as a surrogate model. Finally, by this process we also initiate the first systematic exploration of the parameter space of this model. Doing so, we show how the parameters impact the visual and the statistical properties of centroid velocity maps, and exhibit the correlations between the various parameters, providing new insight into the model.