Parsimonious modelling of nutrient fluxes for a terrestrial ecosystem on Svalbard

Abstract

MBL-MEL, a simple model of ecosystem biogeochemistry, is amended and applied to plant and soil C, 14N and 15N data for the summers of 2001–2003 from Brandalpynten, a maritime high Arctic site on Svalbard following the application of 15N (99 atom%) as 15NO3-N at or 15NH4-N at concentrations of 1 or 5 kg N ha−1. Variants of this Parent model are also developed to incorporate: temperature dependence into equations describing nutrient fluxes (Temp model); cryptogams (Cryp model); both features combined (CrypTemp model). Goodness-of-fit (GOF) statistics suggest that the addition of temperature-dependence improves the utility of models with and without cryptogams: the residual weighted sums of squares per data point were 5.69, 3.91, 4.31 and 3.93 for the Parent, Temp, Cryp and CrypTemp models respectively. The application of model selection criteria confirm that the addition of temperature-dependence also improves model generalisability. Across all models, the principal discrepancies between observation and prediction are associated with the inorganic soil 15N pool. We conclude that models in which fluxes are described using simple equations that can be augmented to include additional complexity, are an ideal starting point from which the relationship between GOF and model generalisability can be assessed.