Drought-driven salinity intrusion drives fast decline and slow recovery of methane fluxes in a brackish tidal wetland

Abstract

Methane fluxes in brackish tidal wetlands are challenging to predict because common controls interact with temporally varying salinity. We measured ecosystem-scale methane fluxes in a brackish marsh in Massachusetts during two hydrologically distinct years during which methane fluxes collapsed and then recovered. The wetland experienced exceptionally high salinity levels during a drought in 2022 and consistently moderate levels in the subsequent year. Soil salinity did not return to the initiicensal low values which was likely due to limited infiltration of fresh surface water in 2023. Methane fluxes averaged 0.120 (Formula presented) (Formula presented) mol  (Formula presented) (Formula presented)   (Formula presented) (Formula presented) before they were reduced to effectively zero when salinity increased rapidly. Fluxes recovered to 67% of original levels in August 2023. To understand the timescale and drivers of this ecosystem response as well as their interactions, we developed neural network models to predict methane fluxes for each year. We derived functional relationships by systematically varying each driver with the remaining drivers set constant. We found porewater specific conductivity, air temperature and to lesser degree gross primary production to be the dominant drivers of methane fluxes. Our modeling determined strong interactions between specific conductivity and temperature controlling methane fluxes. We identified a threshold of about 15 mS  (Formula presented) (Formula presented) (8.7 psu) above which modeled methane fluxes decreased substantially, especially at higher temperatures. In 2023, the variation in measured specific conductivity was low and the neural network less predictive. At that time, our porewater measurements indicated variability of sulfate concentrations not captured by specific conductivity observations. Porewater methane concentrations were consistently detectable even during periods of flux suppression, indicating a role of methane oxidation in the prolonged flux suppression. Our findings demonstrate the value of applying machine learning approaches to flux analysis in dynamic wetland systems and suggest that drought-induced salinization can alter methane cycling in brackish tidal wetlands for prolonged periods of time.