Ecological communities can vary greatly in species composition. Often this variation is discontinuous, in that abrupt changes in composition occur over small distances in space or short periods of time. A wide range of hypotheses from different subfields of ecology have been proposed to explain these patterns. I suggest a framework to quantitatively evaluate these hypotheses with observational data by characterizing 1) how community composition varies across sites in space, 2) how community composition varies through time, and 3) the possible drivers of this variation. I applied this approach to understand the community composition of producers in temporary and semipermanent wetlands in Michigan, USA. I identified several distinct community states which were variously dominated by particular plant functional groups (submerged, floating or emergent plants) or had no plants throughout a season. Evaluating possible hypotheses to explain this variation, I found that similar communities were not necessarily clustered near each other, suggesting that dispersal was not limited for these plants. Some sites exhibited a great deal of change in plant composition among years, shifting between two community states, but there was relatively little change at sites within a year. Moreover, these shifts did not occur in a particular order to suggest directional change or repeating cycles. Community composition was associated with several environmental variables such as pH, light and depth, and multivariate analyses suggested that species had complex, nonlinear responses to these possible drivers. Alternative stable states and interactions among multiple nonlinear drivers best explained the patterns observed in these wetlands. By formalizing initial data collection in other systems with the framework suggested here, we may gain insight into the causes of alternative community states beyond wetlands and the role of climate change and other anthropogenic forces in precipitating transitions between states.