Modelling carbon and water cycles in a beech forest Part I : model description and uncertainty analysis on modelled NEE

A forest ecosystem model (CASTANEA) is developed with the aim to bridge the gap between soil–vegetation–atmosphere (SVAT) and growth models. A physiologically multi-layer process-based model is built, completed with a carbon allocation model and coupled with a soil model. CASTANEA describes canopy photosynthesis and transpiration, maintenance and growth respiration, seasonal development, partitioning of assimilates to leaves, stems, branches, coarse and fine roots, evapotranspiration, soil heterotrophic respiration, water and carbon balances of the soil. Net primary productivity (NPP) is calculated as the difference between gross photosynthesis and plant respiration. The net ecosystem exchange (NEE) between soil-plant system and atmosphere is calculated as the difference between gross photosynthesis and total respiration (soil + plants). The meteorological driving variables are global radiation, rainfall, wind speed, air humidity and temperature (either half-hourly or daily values). A complete description of the model parameterization is given for an eddy flux station in a beech stand (Hesse, France). A parametric sensitivity analysis is carried out to get a classification of the model parameters according to their effect on the NEE. To determine the key input parameters, a +10% or −10% bias is applied on each of the 150 parameters in order to estimate the effect on simulated NEE. Finally 17 parameters, linked to photosynthesis, vegetative respiration and soil water balance, appear to have a significant effect (more than 2.5%) on the NEE. An uncertainty analysis is then presented to evaluate the error on the annual and daily NEE outputs caused by uncertainties in these input parameters. Uncertainties on these parameters are estimated using data collected in situ. These uncertainties are used to create a set of 17,000 simulations, where the values of the 17 key parameters are randomly selected using gaussian random distributions. A mean uncertainty of 30% on the annual NEE is obtained. This uncertainty on the simulated daily NEE does not totally explain the discrepancies with the daily NEE measured by the eddy covariance technique (EC). Errors on daily measurements by EC technique and uncertainty on the modelling of several processes may partly explain the discrepancy between simulations and measurements