Model-data comparison
Several sets of salinity parameters (optimal salinity, µ, and salinity tolerance, λ, from Equation 1) were tested with the improved model to see which parameters best represented measured GPP for 2018. The simulation using (µ = 0, λ = 20) predicted the closest cumulative GPP to measurements (Fig. 3a). Modeled GPP rates with each set of parameters were compared to measured GPP rates using linear regression (Fig 3b). The simulation using (µ = 0, λ = 20) had the best agreement with data (RMSE = 5.65).
Simulations without a salinity response overestimated rates of GPP by as much as 3x (Fig. 4, 5). The inclusion of salinity and flooding functions simulated GPP values very close to measurements (Fig. 4, Fig. 5). At hourly time scales, the improved model shows decreased GPP rates during high tides (Fig. 4). The modeled water level based on NOAA tide constituents does not always represent high tide at the time it is measured in the low marsh, but the predicted water level patterns are similar to observed water levels at the site.
At annual time scales, the improved model represents the cumulative GPP within 20% of measurements (Fig. 5). The default model simulated annual productivity to be between 2700-3000 gC m-2, whereas the improved model simulated annual productivity between 700-1050 gC m-2. In general, the improved model tends to overestimate GPP during green-up and senescence and underestimate GPP during the middle of the growing season. The model overestimated 2018 and 2019 GPP by 10 and 19% respectively, but underestimated GPP by 21% in 2020, a year in which salinity was higher throughout the growing season, but especially during the first half.