Aggregation error: Research objectives and relevant model structure☆

Abstract A model that simulated the daily production of 100 theoretical algal populations was used to define a sampling universe for the construction and evaluation of aggregated models of phytoplankton community structure. Three model communities consisting of ten populations each were constructed by aggregating parameters from subsets of the original 100 populations. Populations were aggregated according to different criteria: (1) correlations between their individual annual production and total annual production of all 100 populations, (2) correlations between individual population late summer production and total algal production during this period, and (3) similarities among model parameters that determined individual population growth rates. A fourth model community was constructed by classifying all 100 populations into four groups according to their individual seasonal patterns of growth. Cross comparisons of model performance with total community production values demonstrated that aggregates based on correlation with annual production reproduced total annual production within 6% of the actual value. The community aggregate derived from seasonal patterns of individual population growth was the poorest overall predictor, underestimating total annual production by 24% and late summer production by 4%. The model community derived from late summer production underestimated total annual production by 18%, but predicted summer production under simulated increased temperatures and nutrient availability within 1% of the actual value. These aggregation exercises demonstrated the importance of identifying community structures in relation to the specific ecological question of interest and indicated the potential for inaccuracy when a model structure derived for a particular objective is more broadly applied.