1: \begin{abstract}

2: A new efficient ensemble prediction strategy is developed for a general

3: turbulent model framework with emphasis on the nonlinear interactions

4: between large and small scale variables. The high computational cost

5: in running large ensemble simulations of high dimensional equations

6: is effectively avoided by adopting a random batch decomposition of

7: the wide spectrum of the fluctuation states which is a characteristic

8: feature of the multiscale turbulent systems. The time update of each

9: ensemble sample is then only subject to a small portion of the small-scale fluctuation

10: modes in one batch, while the true model dynamics with multiscale

11: coupling is respected by frequent random resampling of the batches

12: at each time updating step. We investigate both theoretical and numerical

13: properties of the proposed method. First, the convergence of statistical

14: errors in the random batch model approximation is shown rigorously

15: independent of the sample size and full dimension of the system. Then,

16: the forecast skill of the computational algorithm is tested on two

17: representative models of turbulent flows exhibiting many key statistical

18: phenomena with direct link to realistic turbulent systems. The random

19: batch method displays robust performance in capturing a series of

20: crucial statistical features of general interests including highly

21: non-Gaussian fat-tailed probability distributions and intermittent

22: bursts of instability, while requires a much lower computational cost

23: than the direct ensemble approach. The efficient random batch method

24: also facilitates the development of new strategies in uncertainty

25: quantification and data assimilation for a wide variety of complex

26: turbulent systems in science and engineering.

27: \end{abstract}

28: