High-resolution observations of evolved stars have uncovered a plethora of intricate structures and substructures in their outflows. The origin of the complexity of these structures probably lies in the interplay between several effects, such as convective motions at the stellar surface, or one or more companions stirring up the outflow. As a result, sophisticated forward models, that can incorporate these effects, are required to model the observed intricacies. However, despite much progress on these forward models, their complexity and the sheer size of the relevant parameter space make it difficult to model observations of a specific object. This, in turn, makes it difficult to interpret these observations, to learn exactly what is going on, and use this understanding to inform the forward modelling.
In this talk, we aim to alleviate this emerging tension between modelling and observations, caused by the intricacies that are observed and thus have to be modelled. We present the first steps that have been taken to build a novel methodology, based on deprojection or inverse radiative transfer, in which spectral line images (e.g. obtained with ALMA) can be converted and combined into a probabilistic 3D model for a specific object. The resulting models are probabilistic in the sense that they allow one to sample from a probability distribution of potential 3D models corresponding to the observations of a particular object. With this probabilistic approach, one can account for the ambiguity that is inherent to this process, as some model parameters might not be constrained by the observations. Furthermore, it allows one to account for the uncertainties in the observations. As a first example, we demonstrate this methodology for a set of ALMA observations of the intricate circumstellar environment of the evolved star R Aquilae.