In this talk, we will provide a high-level explanation of why stochasticity is a necessary tool in computational turbulence modelling. While the discussion focuses on fluid dynamics, the underlying concepts belong to a wider mathematical framework suited for dynamical systems where one can only observe and simulate a projection of the full phase space. This framework is particularly relevant for systems where “coarse” dynamics take place in a significantly lower-dimensional space than the “true” dynamics. The wide range of scales of motion in turbulent flow provides a major challenge in the prediction of fluid-dynamical processes. Feasible simulation strategies require solving computational problems with reduced complexity, for example by filtering out the smaller scales of motion and/or numerically solving the governing equations on a coarse grid. However, these necessary simplifications induce systematic errors and uncertainty in flow prediction. In this presentation, we explain how errors and the loss of information inherent in complexity reduction motivate a probabilistic modelling approach. We will explain the need for i) (data-driven) stochastic closure models to account for discretisation errors and unresolved scales, and ii) data assimilation methods to steer predictions toward observations.