Does maximization of net carbon profit enable the prediction of vegetation behaviour in savanna sites along a precipitation gradient?

Abstract

Most terrestrial biosphere models (TBMs) rely on more or less detailed information about the properties of the local vegetation. In contrast, optimality-based models require much less information about the local vegetation as they are designed to predict vegetation properties based on general principles related to natural selection and physiological limits. Although such models are not expected to reproduce current vegetation behaviour as closely as models that use local information, they promise to predict the behaviour of natural vegetation under future conditions, including the effects of physiological plasticity and shifts of species composition, which are difficult to capture by extrapolation of past observations. A previous model intercomparison using conventional TBMs revealed a range of deficiencies in reproducing water and carbon fluxes for savanna sites along a precipitation gradient of the North Australian Tropical Transect (Whitley et al., 2016). Here, we examine the ability of an optimality-based model (the Vegetation Optimality Model, VOM) to predict vegetation behaviour for the same savanna sites. The VOM optimizes key vegetation properties such as foliage cover, rooting depth and water use parameters in order to maximize the net carbon profit (NCP), defined as the difference between total carbon taken up by photosynthesis minus the carbon invested in construction and maintenance of plant organs. Despite a reduced need for input data, the VOM performed similarly to or better than the conventional TBMs in terms of reproducing the seasonal amplitude and mean annual fluxes recorded by flux towers at the different sites. It had a relative error of 0.08 for the seasonal amplitude in ET and was among the three best models tested with the smallest relative error in the seasonal amplitude of gross primary productivity (GPP). Nevertheless, the VOM displayed some persistent deviations from observations, especially for GPP, namely an underestimation of dry season evapotranspiration at the wettest site, suggesting that the hydrological assumptions (free drainage) have a strong influence on the results. Furthermore, our study exposes a persistent overprediction of vegetation cover and carbon uptake during the wet seasons by the VOM. Our analysis revealed several areas for improvement in the VOM and the applied optimality theory, including a better representation of the hydrological settings as well as the costs and benefits related to plant water transport and light capture by the canopy. The results of this study imply that vegetation optimality is a promising approach to explain vegetation dynamics and the resulting fluxes. It provides a way to derive vegetation properties independently of observations and allows for a more insightful evaluation of model shortcomings as no calibration or site-specific information is required.