Uncertainty analysis of agroforestry models

Correspondance: wopke.vanderwerf@wur.nlInternational audienceModels for productivity and resource use in agroforestry systems are needed for land use explorations and support of management decisions by farmers and policy makers in European agriculture. Alternative or complementary avenues for building predictive models are available, ranging from parameter-sparse models, containing not more than some dozens of parameters and only few dynamic equations, to parameter-rich models with hundreds of parameters, based on detailed descriptions of ecophysiological processes and covering many pages of code. The strength of parameter-sparse models lies in their transparency and amenability to mathematical techniques for reduction of uncertainty and improvement of predictive capacity. The strength of parameter-rich models is their biological realism and suitability for studying intricate trade-offs in agroforestry systems at short temporal and detailed spatial scales. In both classes of models, problems are faced in parameterization and the containment of uncertainties in model predictions. Parameter-sparse models require long-term data sets for model calibration and evaluation of uncertainty. Such data sets are rare. Parameter-rich models contain many parameters that do not have a significant impact on long-term model predictions; moreover, they are cumbersome to develop and maintain, understand and document. This presentation will focus on how these problems are being dealt with in a European-wide research project aiming at evaluation of possibilities of agroforestry systems as a rediscovered component of sustainable land use in Europe