Prediction of Greenhouse Gas (GHG) Fluxes from Coastal Salt Marshes using Artificial Neural Network

Coastal salt marshes are among the most productive ecosystems on earth. Given the complex interactions between ambient environment and ecosystem biological exchanges, it is difficult to predict the salt marsh greenhouse gas (GHG) fluxes (CO2 and CH4) from their environmental drivers. In this study, we developed an artificial neural network (ANN) model to robustly predict the salt marsh GHG fluxes using a limited number of input variables (photosynthetically active radiation, soil temperature and porewater salinity). The ANN parameterization involved an optimized 3-layer feed forward Levenberg-Marquardt training algorithm. Four tidal salt marshes of Waquoit Bay, MA — incorporating a gradient in land-use, salinity and hydrology — were considered as the case study sites. The wetlands were dominated by native Spartina Alterniflora, and characterized by high salinity and frequent flooding. The developed ANN model showed a good performance (training R2 = 0.87 - 0.96; testing R2 = 0.84 - 0.88) in predicting the fluxes across the case study sites. The model can be used to estimate wetland GHG fluxes and potential carbon balance under different IPCC climate change and sea level rise scenarios. The model can also aid the development of GHG offset protocols to set monitoring guidelines for restoration of coastal salt marshes.