Abstract: The use of biophysical models in agroecology has increased in the last few decades for two main reasons: the need to formalize empirical knowledge and the need to disseminate model-based decision support for decision makers (such as farmers, advisors, and policy makers). The first has encouraged the development and use of mathematical models to enhance the efficiency of field research through extrapolation beyond the limits of site, season, and management. The second reflects the increasing need (by scientists, managers, and the public) for simulation experimentation to explore options and consequences, for example, future resource use efficiency (i.e., management in sustainable intensification), impacts of and adaptation to climate change, understanding market and policy responses to shocks initiated at a biophysical level under increasing demand, and limited supply capacity. Production concerns thus dominate most model applications, but there is a notable growing emphasis on environmental, economic, and policy dimensions. Identifying effective methods of assessing model quality and performance has become a challenging but vital imperative, considering the variety of factors influencing model outputs. Understanding the requirements of stakeholders, in respect of model use, logically implies the need for their inclusion in model evaluation methods. We reviewed the use of metrics of model evaluation, with a particular emphasis on the involvement of stakeholders to expand horizons beyond conventional structured, numeric analyses. Two major topics are discussed: (1) the importance of deliberative processes for model evaluation, and (2) the role computer-aided techniques may play to integrate deliberative processes into the evaluation of agroecological models. We point out that (i) the evaluation of agroecological models can be improved through stakeholder follow-up, which is a key for the acceptability of model realizations in practice, (ii) model credibility depends not only on the outcomes of well-structured, numerically based evaluation, but also on less tangible factors that may need to be addressed using complementary deliberative processes, (iii) comprehensive evaluation of simulation models can be achieved by integrating the expectations of stakeholders via a weighting system of preferences and perception, (iv) questionnaire-based surveys can help understand the challenges posed by the deliberative process, and (v) a benefit can be obtained if model evaluation is conceived in a decisional perspective and evaluation techniques are developed at the same pace with which the models themselves are created and improved. Scientific knowledge hubs are also recognized as critical pillars to advance good modeling practice in relation to model evaluation (including access to dedicated software tools), an activity which is frequently neglected in the context of time-limited framework programs.

Abstract: Eddy covariance data from four European grassland sites are used to probabilistically invert the CARAIB (CARbon Assimilation In the Biosphere) dynamic vegetation model (DVM) with 10 unknown parameters, using the DREAM((ZS)) (DiffeRential Evolution Adaptive Metropolis) Markov chain Monte Carlo (MCMC) sampler. We focus on comparing model inversions, considering both homoscedastic and heteroscedastic eddy covariance residual errors, with variances either fixed a priori or jointly inferred together with the model parameters. Agreements between measured and simulated data during calibration are comparable with previous studies, with root mean square errors (RMSEs) of simulated daily gross primary productivity (GPP), ecosystem respiration (RECO) and evapotranspiration (ET) ranging from 1.73 to 2.19, 1.04 to 1.56 g C m(-2) day(-1) and 0.50 to 1.28 mm day(-1), respectively. For the calibration period, using a homoscedastic eddy covariance residual error model resulted in a better agreement between measured and modelled data than using a heteroscedastic residual error model. However, a model validation experiment showed that CARAIB models calibrated considering heteroscedastic residual errors perform better. Posterior parameter distributions derived from using a heteroscedastic model of the residuals thus appear to be more robust. This is the case even though the classical linear heteroscedastic error model assumed herein did not fully remove heteroscedasticity of the GPP residuals. Despite the fact that the calibrated model is generally capable of fitting the data within measurement errors, systematic bias in the model simulations are observed. These are likely due to model inadequacies such as shortcomings in the photosynthesis modelling. Besides the residual error treatment, differences between model parameter posterior distributions among the four grassland sites are also investigated. It is shown that the marginal distributions of the specific leaf area and characteristic mortality time parameters can be explained by site-specific ecophysiological characteristics.