Global projections of ecosystem responses to increasing climatic and anthropogenic pressures are needed to inform adaptation planning. However, data of appropriate spatiotemporal resolution are often not available to parameterize complex environmental processes at the global scale. Modeling approaches that can project the probability of ecosystem persistence when parameter uncertainty is high may offer a way forward. In particular, the conservation of coastal ecosystems with complex dynamics, like mangrove forests, may benefit from knowing where their future persistence is highly probable or, alternatively, cannot be reliably estimated without additional data of appropriate resolution. Here, we simulated network models to make probabilistic projections of the direction of net change in mangrove ecosystems worldwide under the SSP5-8.5 climate emissions scenario by the years 2040–2060. Seaward net loss was the most probable outcome in 77% [37%–78%; 95% confidence interval (CI)] of mangrove forest units, while 30% [15%–59%; CI] were projected to experience landward net gain or stability. In more than 50% of forest units, projections were ambiguous and therefore unreliable, with a near equal probability of net loss or gain. Quantitative models parameterized with locally accurate data could resolve uncertainty in the future persistence of mangroves in places with unreliable probabilistic projections. Projections made under conservation scenarios also showed that, with action to manage or restore, the number of mangrove forest units likely to experience net gain or stability in the future could nearly double. Our approach to simulating ecosystem responses to climatic and anthropogenic pressures provides a clear indication of how certain (or uncertain) ecosystem persistence is and thus can inform conservation planning.