Analysis of vegetation modelling uncertainties due to soil moisture stress during droughts

Many model uncertainties results from parameter tuning to compensate for errors in model outputs. A number of studies have focused on the analysis of uncertainties in modelled gross primary production (GPP), particularly with regard to the representation of soil moisture stress. GPP is often overestimated by models during dry periods in water-limited regions, and this bias increases during drought events. Soil moisture stress functions are widely applied to correct this. However, soil moisture stress is not always the direct constraining factor on GPP, and the functions adopted by models do not correspond to accepted mechanisms. We have used eco-evolutionary optimality principles, via the so-called P model, to estimate carbon uptake at sites where leaf area index (LAI) was routinely measured. We used observational networks (including FLUXNET) and Fraction of Absorbed Photosynthetically Active Radiation (fAPAR) data from satellites. By comparing modelled and observed GPP we determined whether there is a significant difference between model performance during the dry and wet seasons, or between energy- and water-limited sites. We found that the soil moisture stress function used in one version of the P models essentially compensates for uncertainties in fAPAR data from satellites, especially in grasslands and other areas subject to seasonal drought. This situation is problematic, since soil moisture is a driver or modulator of other ecosystem processes, including soil evaporation and runoff generation. A possible way forward involves implementing phenological components dependent on soil and atmospheric conditions. The new challenge this poses is to apply eco-evolutionary optimality principles to model the seasonal time course of LAI, which is often poorly simulated by complex ecosystem models.