(Massachusetts Institute of Technology and Woods Hole Oceanographic Institution, 2008-02)
I build on the deterministic phytoplankton growth model of Sosik et al. by introducing
process error, which simulates real variation in population growth and inaccuracies
in the structure of the matrix model. Adding a stochastic component allows me to
use maximum likelihood methods of parameter estimation.
I lay out the method used to calculate parameter estimates, confidence intervals,
and estimated population growth rates, then use a simplified three-stage model to
test the efficacy of this method with simulated observations. I repeat similar tests
with the full model based on Sosik et al., then test this model with a set of data from
a laboratory culture whose population growth rate was independently determined.
In general, the parameter estimates I obtain for simulated data are better the
lower the levels of stochasticity. Despite large confidence intervals around some model
parameter estimates, the estimated population growth rates have relatively small
confidence intervals. The parameter estimates I obtained for the laboratory data fell
in a region of the parameter space that in general contains parameter sets that are
difficult to estimate, although the estimated population growth rate was close to the
independently determined value.