The ocean surface boundary layer (OSBL) tends to be vertically well-mixed, but it can be horizontally inhomogeneous. For example, buoyancy fronts can develop as steep horizontal gradients of temperature, salinity and density in the OSBL. Misalignment of these steep horizontal buoyancy gradients with either the horizontal gradients of surface elevation or bathymetry is known to generate circulation which can deflect ocean currents which transport fluid properties, as well as pollution and debris. In turn, the generation of circulation itself can break up these fronts, thereby cascading horizontally circulating structures to smaller scales.Taking advantage of the vertically well-mixed property of the OSBL and working in the stochastic Euler-Poincaré variational framework introduced in [HL19], we derive the thermal rotating shallow water (TRSW) equations with stochastic advection by Lie transport (SALT), as a theoretical foundation for uncertainty quantification and data assimilation in computational models of the effects of submesoscale oceanic circulation using data-driven stochastic parametrisations, as in [CCH + 18, CCH + 19]. The key feature of SALT for geophysical fluid dynamics (GFD) is that SALT respects the Kelvin circulation theorem, which is the essence of cyclogenesis. Asymptotic expansion in the three small parameters present in the TRSW model in the neighbourhood of thermal quasi-geostrophic balance among the flow velocity and the gradients of free surface elevation and buoyancy leads first to the deterministic thermal versions of the Lagrangian 1 (TL1) equations and then to the thermal quasi-geostrophic (TQG) theory. We illustrate the instabilities of TQG which cascade circulation to smaller, typically unresolvable scales. We derive the stochastic version of this hierarchy of models TRSW/TL1/TQG in the framework of the stochastic Euler-Poincaré variational principle. Finally, we indicate the next steps in applying these results for uncertainty quantification and data assimilation of the cascading cyclogenetic effects of buoyancy fronts using the data-driven stochastic parametrisation algorithm based on SALT at these three levels of description.