After a brief overview of our work on transition to turbulence in complex three-dimensional flows, the core of this seminar will focus on the problem of identifying interpretable and physically consistent reduced-order models from data. Over the past few years, the sparse identification of nonlinear dynamics (SINDy) has sparked a renewed interest in the identification of continuous-time nonlinear dynamical systems from a limited set of available measurements. Since its introduction in 2015, SINDy has proven to be an extremely versatile framework with numerous variations proposed, e.g. vanilla SINDy, SINDy with control, MANDy, etc. In this talk, we thus aim to give the audience a general overview of the capabilities offered by the SINDy framework. For that purpose, the chaotic thermal convection in annular thermosyphon will be used as example. We will most notably focus on how one can incorporate prior knowledge about the system (e.g. invariants, symmetries, etc) in the identification step.