The loss of coastal marshes is a topic of great concern, because these habitats provide tangible ecosystem services and are at risk from sea-level rise and human activities. In recent years, significant effort has gone into understanding and modeling the relationships between the biological and physical factors that contribute to marsh stability. Simulation-based process models suggest that marsh stability is the product of a complex feedback between sediment supply, flooding regime and vegetation response, resulting in elevation gains sufficient to match the combination of relative sea-level rise and losses from erosion. However, there have been few direct, empirical tests of these models, because long-term datasets that have captured sufficient numbers of marsh loss events in the context of a rigorous monitoring program are rare. We use a multi-year data set collected by the Coastwide Reference Monitoring System (CRMS) that includes transitions of monitored vegetation plots to open water to build and test a predictive model of near-term marsh vulnerability. We found that despite the conclusions of previous process models, elevation change had no ability to predict the transition of vegetated marsh to open water. However, we found that the processes that drive elevation change were significant predictors of transitions. Specifically, vegetation cover in prior year, land area in the surrounding 1 km2 (an estimate of marsh fragmentation), and the interaction of tidal amplitude and position in tidal frame were all significant factors predicting marsh loss. This suggests that 1) elevation change is likely better a predictor of marsh loss at time scales longer than we consider in this study and 2) the significant predictive factors affect marsh vulnerability through pathways other than elevation change, such as resistance to erosion. In addition, we found that, while sensitivity of marsh vulnerability to the predictive factors varied spatially across coastal Louisiana, vegetation cover in prior year was the best single predictor of subsequent loss in most sites followed by changes in percent land and tidal amplitude. The model’s predicted land loss rates correlated well with land loss rates derived from satellite data, although agreement was spatially variable. These results indicate 1) monitoring the loss of small scale vegetation plots can inform patterns of land loss at larger scales 2) the drivers of land loss vary spatially across coastal Louisiana, and 3) relatively simple models have potential as highly informative tools for bioassessment, directing future research, and management planning.