Manipulation to prediction: integrating flood experiments and AI to understand coastal forest mortality

Rising sea levels and intensifying storms are driving increased flooding and salinization of coastal forests, yet the mechanistic pathways linking belowground disturbances to forest mortality remain poorly constrained. We designed an ecosystem-scale flood manipulation experiment in a coastal forest to disentangle the roles of inundation and salinity in initiating the hypothesized “tree mortality spiral”. Our experimental plots are outfitted with an extensive array of sensors to complement high resolution sampling campaigns, allowing us to observe immediate and lagged responses to flooding. Experimental flooding drove rapid, consistent shifts in soil biogeochemistry indicative of oxygen stress and altered carbon cycling, followed by a lagged response in aboveground vegetation. The temporal disconnect between belowground process thresholds and observable forest impacts demonstrates how manipulative experiments can benchmark the early stages of transitions in the coastal Critical Zone. 

Building on our field-based findings and substantial AI-ready datasets produced over multiple years of flooding experiments, we are developing a coupled modeling framework that leverages both AI-based and process-based models to predict forest responses under future flooding regimes. Through this integrated approach, we aim to understand how disturbance intensity, duration, and legacy effects propagate across time and space to control coastal forest resilience. The combination of controlled large-scale ecosystem manipulation and data-driven predictive modeling provides a framework for bridging disciplines and scales—linking soil biogeochemistry, ecohydrology, and vegetation dynamics—to improve projections of coastal forest mortality and its consequences for coastal Critical Zone carbon cycling.