Predicting mussel population density and age structure: the relationship between model complexity and predictive power

We use a model to predict the age structure and density of an intertidal mu ssel Mytilus edulis L, population on the 10 primary mussel beds of the Exe estuary, England. We investigate the relationship between the number of par ameters in the model and its accuracy in describing the observed density-ag e structure. The full detail version (382 parameters) assumes that recruitm ent rates are bed-specific and mortality rates are age-class, bed- and seas on-specific, and describes the observed age structure and density very accu rately; the predicted estuary-wide density in March is 581 m(-2) compared w ith an observed density of 578 m(-2) during 1977 to 1983. The simplest vers ion (24 parameters) describes the age structure and density almost as well (predicted estuary-wide density of 568 m(-2)) and assumes that: (1) density -dependence is absent in the 3rd-winter and older mussels, and density-inde pendent mortality in these age classes is the same on all beds, but increas es with age; (2) in younger mussels, density-dependent mortality operates a bove a threshold mussel density which is determined by bed exposure and has a strength which is the same on all beds, but which varies with age class; (3) recruitment rate is dependent on the density of adults on a bed and th e bed's substrate softness. The model is tested by comparing its prediction s, based on data collected during 1976 to 1983, with observed mussel densit ies during 1992 to 1997; a stable September mussel population of 524 adults m(-2) is predicted and, in accord with this, the observed density of adult s in September changed relatively Little between 1976 and 1983 (541 m(-2)) and between 1992 and 1997 (450 m(-2)). We discuss why the number of paramet ers in the model can be reduced so greatly with very little reduction in th e accuracy of predictions, and whether a similar approach could be used to model other shellfish populations.