An ensemble smoother is used to estimate the initial conditions and mortality rates for a spatially explicit model of Alexandrium fundyense. The data assimilation procedure is effective at reducing model-data misfit in this strong constraint problem formulation. The skill of this estimation procedure is assessed through cross-validation. The estimation is carried out with three different representations of circulation: no flow, climatology, and a data assimilative hindcast. Although the misfit to the assimilated data is lowest with no flow, the skill of the biological hindcast is best with the hindcast and climatological velocity fields. Mortality estimates fall within the range of observed values, but the inferred spatial structure is not testable with existing data.