The mesoscale organization and structure of convective clouds is thought to be rooted in the thermodynamic properties of the atmosphere and in the turbulent to mesoscale dynamics of the flow. Such structure may contribute to the transition between shallow and deep convection. The thermodynamic state of the boundary layer is forced by the amount of surface fluxes from below. Conversely, landscape patterns and land-cover heterogeneity may equally give rise to focused regions for deep convection triggering, in particular when patch sizes exceed 10 km. Since the convective boundary layer has a mediating function between the surface and deep storm clouds, the connection between surface and upper atmosphere is not straightforward. It is generally believed to involve local erosion of the capping inversion layer, the build-up of a moist energy supply, gradual humidification of the lower-free troposphere that reduces dry air entrainment into burgeoning deeper clouds, and thermal mesoscale circulations that can generate moisture convergence and locally forced ascent. To what extent microscale realistic surface heterogeneity and an interactive surface response matter to shallow and deep convection and its organization remains an open question. In this dissertation, we describe the coupling of a physiology-based vegetation model (HYBRID) and of a sea surface flux algorithm (COARE) to the cloud-resolving Active Tracer High-resolution Atmospheric Model (ATHAM). We investigate the full diurnal cycle of convection based on the example of the Hector storm over Tiwi Islands, notably the well-characterized event on 30th November 2005. The model performs well in terms of timing and cloud dynamics in comparison to a range of available observations. Also, ATHAM-HYBRID seems to do well in terms of flux partitioning. Whilst awaiting more thorough flux validation, we remain confident that the interactive surface response of both HYBRID and COARE is suited for the purpose of simulating convective-scale processes. We find the storm system evolution in 3D simulations to be robust with respect to differences in surface configuration and initialization. Within our 3D sensitivity runs, we could not identify a strong dependence on either realistic surface heterogeneity in the island landscape or on the interactive surface response. We conclude that in our case study at least, atmospheric (turbulent) dynamics likely dominate over surface heterogeneity effects, provided that the bulk magnitude of the surface energy fluxes, and their partitioning into sensible and latent heat (Bowen ratio), remain unaltered. This is consistent with 2D sensitivity studies, where we find model grid-spacing and momentum diffusion, governing the dynamics, to have an important influence on the overall evolution of deep convection. Fine grid-spacing is necessary, as the median width of updraught cores mostly does not exceed 1000 m. We associate this influence with the dry air entrainment rate in the wake of rising parcels, and with how resolution and diffusion act on coherent structures in the flow. In 2D sensitivity studies with differences in realistic heterogeneities of surface properties, we find little evidence for a clear deterministic influence of these properties on the transition between shallow and deep convection, in spite of largely different storm evolutions across the various runs. In these runs, we tentatively ascribe triggering to stochastic features in the flow, without discarding the relevance of convergence lines produced by mesoscale density currents, such as the sea breeze and cold pool storm outflows.