Understanding ecosystem stability and functioning is a long-standing goal in theoretical ecology, with one of the main tools being dynamical modelling of species abundances. With the help of dynamical population models limits to stability and regions of various ecosystem dynamics have been extensively mapped in terms of diversity (number of species), types of interactions, interaction strengths, varying interaction networks (for example plant-pollinator, food-web) and varying structures of these networks. Although it is apparent that ecosystems reside in and are affected by a spatial environment, local differences (spatial heterogeneity) is often excluded from studies mapping stability boundaries under the assumption of an average and equal amount of interaction for all individuals of a species. Here we show that extending the classic dynamical Generalised-Lotka-Volterra model into a connected space the boundaries of stability change. When viewing the ecosystem as a spatially heterogeneous whole, limits previously marking the end of stability can now be crossed without any remarkable change in species abundances and without loss of stability. Thus limits previously thought to mark catastrophic transitions are not critical due to the possibility of spatial heterogeneity within the system. In addition, we show that too much spatial fragmentation of ecosystem habitats acts destabilising and leads back to the stability boundaries found in spatially homogeneous ecosystems with average interactions. Thus, we conclude that spatially heterogeneous but connected systems are the most robust. In terms of ecosystem management, the risk of collapse or irreversible changes is lower in spatially heterogeneous systems, which real ecosystems are, and we should expect local changes in populations well in advance of system collapse. Although, too much fragmentation of an ecosystem's available space can lead to a less robust system with higher risk of extinctions and collapse.