Multi-model ensemble simulations to assess the impact of climate change on agro-ecosystems

Abstract

Global climate change will alter the water, nitrogen and carbon cycles of agroecosystems. To predict future agricultural production under climate change, numerical soil-crop models are used. These soil-crop models can represent the complex and coupled processes of agroecosystems in a deterministic manner for a given environment. The projections made by soilcrop models suffer from two kinds of uncertainty: (1) epistemic uncertainty and (2) parameter uncertainty. Additionally, it is assumed that the parameterization is applicable to other environments. Therefore, this study has two major aims. The first aim is to quantify the above-mentioned uncertainties simultaneously by combining two methods: multi-model ensemble modeling and Bayesian statistics. The multi-model ensemble allows to quantify epistemic uncertainty by comparing individual model outputs. This has been demonstrated in many studies. Bayesian methods are common to assess credible parameter intervals for highly nonlinear process models. The second aim of this study is to provide a framework for assessing the robustness of the parametrization of soil-crop models. Therefore, a preliminary numerical study was conducted to test different calibration schemes and to investigate parameters sensitivities in dependence of the environment. The soil-crop modelling software ExpertN 3.0 will be used to set up a multi-model ensemble with eight soilcrop models. The model output will be analyzed by comparison with data from two sites, five soil types and two crops gathered by the DFG Research Unit 1695 since 2010. To achieve the second aim a global sensitivity analyses was conducted to rank the input factors for each soil-crop model. The result of the global sensitivity analyses will clarify the impact of model input on model output in regard to environment, model combinations, and extent. Additionally, different calibration schemes will be tested to identify the method yielding the most robust parametrization. We used a Latin Hypercube sampling scheme. In total, the whole study requires 1,000,000 CPU hours. We expect that the results will enable us to develop a generally applicable and feasible strategy of how soil-crop models have to be set up to produce reliable predictions of agroecosystem behavior under climate change.