Model complexity and performance: How far can we simplify?

Handling model complexity and reliability is a key area of research today. While complex models containing sufficient detail have become possible due to increased computing power, they often lead to too much uncertainty. On the other hand, very simple models often crudely oversimplify the real ecosystem and can not be used for management purposes. Starting from a complex and validated 1D pelagic ecosystem model of the Ligurian Sea (NW Mediterranean Sea), we derived simplified aggregated models in which either the unbalanced algal growth, the functional group diversity or the explicit description of the microbial loop was sacrificed. To overcome the problem of data availability with adequate spatial and temporal resolution, the outputs of the complex model are used as the baseline of perfect knowledge to calibrate the simplified models. Objective criteria of model performance were used to compare the simplified models’ results to the complex model output and to the available data at the DYFAMED station in the central Ligurian Sea. We show that even the simplest (NPZD) model is able to represent the global ecosystem features described by the complex model (e.g. primary and secondary productions, particulate organic matter export flux, etc.). However, a certain degree of sophistication in the formulation of some biogeochemical processes is required to produce realistic behaviors (e.g. the phytoplankton competition, the potential carbon or nitrogen limitation of the zooplankton ingestion, the model trophic closure, etc.). In general, a 9 state-variable model that has the functional group diversity removed, but which retains the bacterial loop and the unbalanced algal growth, performs best.