Presentation #102.256 in the session Poster Session.
A planet formation model can only be tested against observations if the initial properties of the protoplanetary disk are constrained. The most robust data is however obtained for evolved Class II objects with ages around 2 Myr. Here, we present a Bayesian retrieval of initial conditions and unkown parameters like the strength of viscous turbulence based on ≈100 observed and well characterized disks in young clusters on the ALMA continuum millimeter flux and accretion luminosity plane. As forward model, we use a neural net trained on 100,000 realizations of a viscous disk model including modern external and internal photoevaporation prescriptions, a two-population dust and pebble evolution code, and a planetesimal formation module. This allows us to model the observable emission instead of relying on typical assumptions like a constant temperature and opacity, fixed cavity size, or one percent dust to gas ratio. Within this framework, we find preliminary posterior distributions shown in the supplementary data figure. First results of posterior distributions favor massive disks, small initial disk sizes, viscous α values around 10-4, low fragmentation velocities, and weak external FUV. The Bayesian framework is well suited to produce distributions of parameters which can be used in future planet formation studies.