Parameter sensitivity study of the NEMURO lower trophic level marine ecosystem model

Abstract

We investigated the parameter sensitivity of the NEMURO ecosystem model calibrated to field at two typical Stations (A7 and P) in the subarctic North Pacific. The NEMURO model follows various forms of nitrogen and silicon, and the daily biomasses of two phytoplankton and three zooplankton groups for multiple years. Previously calibrated versions to data at two stations provided the basis for comparing parameter sensitivities under different environmental and biological conditions. Four sensitivity analysis experiments were performed involving 72 parameters: 3 experiments used Monte Carlo methods and 1 experiment used a 1-parameter-at-a-time approach. Normalized sensitivities and correlation were used as sensitivity measures for comparison among the four experiments. Monte Carlo and one-parameter-at-a-time methods that used relatively small variations in parameter values (less than or equal to ±10%) yielded very similar rankings of the top five parameters. Parameter rankings from the Monte Carlo analysis that used relatively large variation in parameters (−50% to +100%) differed slightly from the rankings obtained with the small variations. Additional examination of the ±10% results showed that parameters deemed important differed among prognostic variables and differed between Stations A7 and P. For example, annual small phytoplankton biomass was most sensitive to the maximum grazing rate of small zooplankton at both stations, while large phytoplankton biomass was most sensitive to its own maximum photosynthetic rate parameter. While at Station A7 large phytoplankton was sensitive to zooplankton parameters, at Station P it was more sensitive to phytoplankton parameters. A few of certain phytoplankton parameters were consistently important in all sensitivity experiments. Despite sometimes complex relationships between prognostic variables and parameters, our analyses showed that the NEMURO model was generally well-behaved and was robust to parameter variation and to the method used for the sensitivity analysis. Extensions of our analysis could involve computing the sensitivity measure over time through the year, and the use other types of prognostic variables than biomass such as the timing of the phytoplankton bloom. Sensitivity analyses, such as those performed here, are useful tools for applying the NEMURO to other locations and for helping to interpret and diagnose existing NEMURO applications.