The grass layer of African savannas can be described as heterogeneous mosaics of lawn and bunch grass communities. The transformation of bunch grass states to lawn grass states is expected when bunch grasslands are heavily utilised by herbivores allowing lawn grass invasion to persist under frequent grazing. The dynamic nature of these transformations suggests lawn and bunch grasslands can exist as alternative ecosystem states however, empirical evidence has so far been lacking. In this study, we focus on how invasion of lawn grasses in bunch dominant grass communities may change with different grazing frequencies, we used an experimental approach to create artificial ecosystems in mesocosms to test if different clipping treatments coupled with invasion by lawn grasses can shift bunch grass communities into lawn dominant communities. Our experiment investigated (1) under which clipping frequencies are lawn grasses able to invade bunch grass systems successfully. The initial species composition was dominated by 80% bunch grass with 20% lawn grass, we manipulated different clipping treatments ranging from frequent to infrequent clipping. (2) To understand how the average growth rate of lawn and bunch grass species differ with clipping treatments, a biomass model was developed within a Bayesian state-space modeling framework. We used biomass data from the experiment and a Markov chain Monte Carlo method to estimate model parameters. We found that under frequent clipping lawn grasses have higher growth rates and are able to invade bunch-dominated systems and under infrequent clipping lawn grasses are unable to invade. These results suggest frequent clipping can maintain grass communities in a lawn-dominant state while infrequent clipping removes lawn grasses from the system. Our analysis provides evidence to the theory that lawn grass species persist under intense frequent grazing and that two alternative grass states can exist under different clipping treatments.