Abstract. The dynamics of biogeochemical models are determined by the mathematical
equations used to describe the main biological processes. Earlier studies
have shown that small changes in the model formulation may lead to major
changes in system dynamics, a property known as structural sensitivity. We
assessed the impact of structural sensitivity in a biogeochemical model of
intermediate complexity by modelling the chlorophyll and dissolved inorganic
nitrogen (DIN) concentrations. The model is run at five different
oceanographic stations spanning three different regimes: oligotrophic,
coastal, and the abyssal plain, over a 10-year timescale to observe the
effect in different regions. A 1-D Model of Ecosystem Dynamics, nutrient
Utilisation, Sequestration, and Acidification (MEDUSA) ensemble was used with
each ensemble member having a combination of tuned function parameterizations
that describe some of the key biogeochemical processes, namely nutrient
uptake, zooplankton grazing, and plankton mortalities. The impact is
quantified using phytoplankton phenology (initiation, bloom time, peak
height, duration, and termination of phytoplankton blooms) and statistical
measures such as RMSE (root-mean-squared error), mean, and range for
chlorophyll and nutrients. The spread of the ensemble as a measure of
uncertainty is assessed against observations using the normalized RMSE ratio
(NRR). We found that even small perturbations in model structure can produce
large ensemble spreads. The range of 10-year mean surface chlorophyll
concentration in the ensemble is between 0.14 and 3.69 mg m−3 at
coastal stations, 0.43 and 1.11 mg m−3 on the abyssal plain, and 0.004
and 0.16 mg m−3 at the oligotrophic stations. Changing both
phytoplankton and zooplankton mortalities and the grazing functions has the
largest impact on chlorophyll concentrations. The in situ measurements of
bloom timings, duration, and terminations lie mostly within the ensemble
range. The RMSEs between in situ observations and the ensemble mean and
median are mostly reduced compared to the default model output. The NRRs for
monthly variability suggest that the ensemble spread is generally narrow (NRR
1.21–1.39 for DIN and 1.19–1.39 for chlorophyll profiles, 1.07–1.40 for
surface chlorophyll, and 1.01–1.40 for depth-integrated chlorophyll). Among the five stations, the most reliable ensembles are
obtained for the oligotrophic station ALOHA (for the surface and integrated
chlorophyll and bloom peak height), for coastal station L4 (for inter-annual
mean), and for the abyssal plain station PAP (for bloom peak height). Overall
our study provides a novel way to generate a realistic ensemble of a
biogeochemical model by perturbing the model equations and parameterizations,
which will be helpful for the probabilistic predictions.
Trophic modeling of the Northern Humboldt Current Ecosystem, Part II: Elucidating ecosystem dynamics from 1995 to 2004 with a focus on the impact of ENSO
