Machine Learning Predictions of Vertical Accretion in the Mississippi River Deltaic Plain

Deltaic landscapes consist of vast wetland systems that rely on sedimentation to maintain their elevation and ecological communities against relative sea-level rise. In the Mississippi River Deltaic plain, rising relative sea level and anthropogenic activities are causing land loss that will continue unless vertical accretion of sediment on the wetland surface is enough to fill the accommodation space. Even though the fate of the Mississippi Deltaic plain is tied directly to vertical accretion, there is not yet a clear understanding of the system-wide controls on this process. Here, we investigate vertical accretion in coastal Louisiana using a data set of 266 stations from the Coastwide Reference Monitoring System (CRMS). Using linear regression models, we analyze vertical accretion in freshwater-intermediate, brackish, and saline marsh communities. Integrating results from these models into a Gaussian Process regression model, we predict controls on vertical accretion rates across the deltaic plain. Consistent with previous studies, our results suggest that tidal amplitude and flood depth are critical controls on vertical accretion. These effects are additive and marshes with high tidal amplitudes and flood depths experience the most vertical accretion. Interestingly, the normalized difference vegetation index is found to be important for predicting vertical accretion, but not because of an increase in biomass production, but because it records unique marsh communities and flooding regimes. This study emphasizes the importance of incorporating marsh specific information into predictive models for the vertical accretion of coastal wetlands and that better predictions of wetland accretion probably require denser observational data. The Mississippi River Deltaic plain (MRDP) is a threatened landscape as the Gulf of Mexico encroaches inland due to relative sea-level rise. To prevent further coastal land loss, enough sediment must be deposited onto the wetland surface to offset the relative sea-level rise. Even though the fate of the MRDP is directly tied to sedimentation, we still do not have a system-wide understanding of what controls this process. We used 266 stations that recorded numerous environmental variables from the Coastwide Reference Monitoring System to investigate the controls on sedimentation. Using a machine learning framework, we find that tidal amplitude and flood depth have additive effects that positively affect sedimentation rates. Interestingly, the normalized difference vegetation index is found to be important for predicting sedimentation, but not due to an increase in biomass production, but because it records unique wetland communities and flooding regimes. This study emphasizes the importance of building predictive models for sedimentation that consider specific information about different wetland types. We use a machine learning framework to understand how vertical accretion varies by marsh community and across the Mississippi River Deltaic plain We found that large tidal amplitude and flood depth positively affect vertical accretion Normalized difference vegetation index is also important in our model because it distinguishes marsh communities in the delta