Browsing by Person "Witte, Irene"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Publication
    Multi-objective and multi-variate global sensitivity analysis of the soil-crop model XN-CERES in Southwest Germany
    (2021) Witte, Irene; Streck, Thilo
    Soil-crop models enjoy ever-greater popularity as tools to assess the im- pact of environmental changes or management strategies on agricultural production. Soil-crop models are designed to coherently simulate the crop, nitrogen (N) and water dynamics of agricultural fields. However, soil-crop models depend on a vast number of uncertain model inputs, i.e., initial conditions and parameters. To assess the uncertainty in the simulation results (UCSR) and how they can be apportioned among the model inputs of the XN-CERES soil-crop model, an uncertainty and global sensitivity analysis (GSA) was conducted. We applied two different GSA methods, moment-independent and variance-based methods in the sense of the Factor Prioritization and the Factor Fixing setting. The former identifies the key drivers of uncertainty, i.e., which model input, if fixed to its true value, would lead to the greatest reduction of the UCSR. The latter identifies the model inputs that cannot be fixed at any value within their value range without affecting the UCSR. In total we calculated six sensitivity indices (SIs). The overall objective was to assess the cross-sub-model impact of parameters and the overall determinability of the XN-CERES applied on a deep loess soil profile in Southwest Germany. Therefore, we selected 39 parameters and 16 target variables (TGVs) to be included in the GSA. Furthermore, we assessed a weekly time series of the parameter sensitivities. The sub-models were crop, water, nitrogen and flux. In addition, we also compared moment-independent (MI) and variance-based (VB) GSA methods for their suitability for the two settings. The results show that the parameters of the TGVs of the four groups cannot be considered independently. Each group is impacted by the parameters of the other groups. Crop parameters are most important, followed by the Mualem van Genuchten (MvG) parameters. The nitrate (NO3-) content and the matric potential are the two TGVs that are most affected by the inter- action of parameters, especially crop and MvG parameters. However, the model output of these two TGVs is highly skewed and leptokrutic. Therefore, the variance is an unsuitable representation of the UCSR, and the reliability of the variance-based sensitivity indices SIVB is curtailed. Nitrogen group parameters play an overall minor role for the uncertainty of the whole XN-CERES, but nitrification rates can be calibrated on ammonium (NH4+) measurements. Considering the initial conditions shows the high importance of the initial NO3-; content. If it could be fixed, the uncertainty of crop groups’ TGVs, the matric potential and the N content in the soil could be reduced. Hence, multi-year predictions of yield suffer from uncertainty due to the simulated NO3-; content. Temporally resolved parameter show the big dependence between the crop’s development stage and the other 15 TGVs becomes visible. High temporally resolved measurements of the development stage are important to univocally estimate the crop parameters and reduce the uncertainty in the vegetative and generative biomass. Furthermore, potential periods of water and N-limiting situations are assessed, which is helpful for deriving management strategies. In addition, it become clear that measurement campaigns should be conducted at the simulation start and during the vegetation period to have enough information to calibrate the XN-CERES. Regarding the performance of the different GSA methods and the different SIs, we conclude that the sensitivity measure relying on the Kolmogorov-Smirnov metric (betaks) is most stable. It converges quickly and has no issues with highly skewed and leptokrutic model output distributions. The assessments of the first-effect index and the betaks provide information on the additivity of the model and parameters that cannot be fixed without impacting the simulation results. In summary, we could only identify three parameters that have no direct impact on any TGV at any time and are hence not determinable from any measurements of the TGVs considered. Furthermore, we can conclude that the groups’ parameters should not be calibrated independently because they always affect the uncertainty of the selected TGV directly or via interacting. However, no TGV is suitable to calibrate all parameters. Hence, the calibration of the XN-CERES requires measurements of TGVs from each group, even if the modeler is only interested in one specific TGV, e.g., yield. The GSA should be repeated in a drier climate or with restricted rooting depth. The convergence of the values for the Sobol indices remains an issue. Even larger sample sizes, another convergence criteria or graphical inspection cannot alleviate the issue. However, we can conclude that the sub-models of the XN-CERES cannot be considered in- dependently and that the model does what it is designed for: coherently simulating the crop, N and water dynamics with their interactions.