A critical overview of model estimates of net primary productivity for the Australian continent

Net primary production links the biosphere and the climate system through the global cycling of carbon, water and nutrients. Accurate quantification of net primary productivity (NPP) is therefore critical in understanding the response of the world’s ecosystems to global climate change, and how changes in ecosystems might themselves feed back to the climate system. Twelve model estimates of long-term annual NPP for the Australian continent were reviewed. These models varied considerably in the approaches adopted and the inputs required. The model estimates ranged 5-fold, from 0.67 to 3.31 Gt C y–1. Within-continent variation was similarly large, with most of the discrepancies occurring in the arid zone of Australia, which comprises most of the continent. It is also within this zone that empirical NPP data are most lacking. Comparison with a recent global-scale analysis of six dynamic global vegetation models showed a similar level of variability in continental total NPP, 0.38 to 2.85 Gt C y–1, and similar within-continent spatial variability. As a first tentative step towards model validation the twelve NPP estimates were compared with existing field measurements, although the ability to reach definitive conclusions was limited by insufficient data, and incompatibilities between the field-based observations and the model predictions. It was concluded that the current NPP-modelling capability falls short of the accuracy required for effective application in understanding the terrestrial biospheric implications of global atmospheric / climatic change. Potential methods that could be used in future work for improving modelled estimates of Australian continental NPP and their validation are discussed. These include increasing the spatial coverage of empirical NPP estimates within arid ecosystems, the use of existing high quality site data for more detailed model exploration, and a formal model inter-comparison using uniform driver datasets to investigate more intensively differences in model behaviour and assumptions.