Model configuration, parameterization, and evaluation
ELM simulations were conducted for a temperate low marsh at the Plum
Island Ecosystems Long Term Ecological Research site (PIE-LTER). The
PIE-LTER low marsh is located in the Plum Island Sound in northern
Massachusetts (42.7345, -70.8382) and has a humid continental climate.
The site has a tidal range of 2.5-3 m and the low marsh elevation is
approximately 1 m. Spartina alterniflora is the dominant
vegetation. The site is instrumented during the growing season with an
eddy covariance tower, water level sensors, and a conductivity sensor.
Hydrological parameters for ELM (marsh level relative to tidal channel,
mean water level height) were parameterized for the low marsh using
observed water level data from PIE-LTER (Giblin 2019, Giblin 2022).
Salinity optimum and tolerance (µ and ʎ ) were
parameterized by starting with literature values for optimal salinity
for S. alterniflora (Vasquez et al. 2006, Maricle and Lee, 2006)
and tolerance values used to model other brackish and salt marsh species
with a Gaussian function (Li et al. 2021). Parameters were then adjusted
until modeled gross primary production daily values (GPP) showed good
agreement with measured GPP at the low marsh eddy covariance tower
(Giblin and Forbrich 2022a, b, c), evaluated using root mean square
error (RMSE) values. The plant functional type was set as a C4 grass and
we used PFT parameters (including leaf traits, photosynthetic capacity,
and growth allocation) optimized in a previous modeling study forS. patens in Chesapeake Bay (O’Meara et al., 2021).
We examined how observed GPP rates correlated with observed water level
and salinity concentrations. To control for phenological patterns in GPP
rates, we first obtained a growing season curve by averaging the noon
GPP measured in 2018, 2019, and 2020 on each day followed by calculating
a 60-day rolling average to smooth the three-year average. This growing
season curve was then removed from the noon measurement of each year’s
dataset (i.e. 2018 noon GPP – seasonal average noon GPP). Pearson’s
correlation was used to determine the correlation of detrended GPP with
water level and salinity.
We drove the model with meteorological data collected at PIE-LTER
(Giblin 2019, Giblin 2020, Giblin 2021), salinity concentrations
measured at xx (Giblin and Forbrich 2022a, b, c) and estimated water
level using tide constituents available from NOAA Tides and Currents
(https://tidesandcurrents.noaa.gov/). Data was gapfilled using
linear interpolation. We assumed one salinity value for the entire soil
column because ELM does not currently represent salt transport through
soil layers. Model simulations were spun up using 100 years of
accelerated decomposition followed by 100 years of regular spinup (Koven
et al. 2013, Thornton et al. 2005). In default simulations, plant
responses to salinity and inundation were inactive, representing the
previous model vegetation function under site-specific meteorological,
hydrological, and salinity conditions. The second set of simulations
used the same meteorological, salinity, and water level forcing while
including vegetation response functions to salinity and flooding. We
evaluated model performance by comparing GPP from the default model
simulations, improved model simulations, and measured GPP from the field
site. A third set of scenarios were conducted for systematic changes in
salinity by adding or subtracting a constant value (-5, +5, +10 ppt)
from the 2018 salinity measurements used to force the improved model.
Finally, the improved model was used to conduct scenarios for systematic
changes in water level by changing the mean tide water level parameter
(-10, +10, +20, +50 cm).