Global forecasts of ecosystem responses to increasing climatic and anthropogenic pressures are needed to inform adaptation planning. However, data of appropriate spatio-temporal resolution are not typically available to parameterise complex processes at the global scale. Forecast uncertainty associated with ‘data-process’ scale incongruities must be quantified and effectively communicated to avoid over-confident decision-making. Here, we used network models to make probabilistic forecasts of the direction of change in mangrove extent globally under the SSP5-8.5 climate emissions scenario by 2040-2060. We forecast that seaward net loss is the most likely outcome in 77% [±37-78%; 95% confidence] of mangrove forest units, while 30% [±15-59%] will experience landward net gain or stability. Parameter uncertainty limited our capacity to make reliable forecasts everywhere, highlighting where current understanding and global datasets are deficient. However, with action to manage or restore, the number of mangrove forest units likely to experience net gain or stability could nearly double.