PhD Researcher: Michelle Viswanathan
Supervisors: Thilo Streck (University of Hohenheim), Tobias Weber (University of Hohenheim), Juliane Mai (University of Waterloo)
The fate and behaviour of agrochemicals at the landscape scale are not well understood. Soil-crop-atmosphere processes control water and element cycles, thus influencing the fluxes of these agrochemicals. Models of these complex processes have been developed with the aim of improving our understanding of the system. With the increasing threat of climate change, these models are also used to make predictions so as to assess the impact on food security and the water cycle, and develop suitable adaptation strategies. However, robust predictions and the identification of model inadequacies are challenging with these models when uncertainties are not appropriately accounted for.
A Bayesian approach can tackle this limitation by providing a range of probable model outcomes by quantifying input, parameter and model uncertainties, while enabling the use of prior information or expert knowledge (1). Multi-model ensembles that capture the uncertainty in our understanding of underlying processes (2) can also be incorporated in a Bayesian framework. Bayesian Model Averaging (BMA) has been applied on an ensemble of soil-crop-atmosphere system models, where the reliability of predicting evapotranspiration and soil water drainage was evaluated (3). This study aims to extend the application of BMA by assessing its effectiveness in the prediction of crop yield, water and nitrate fluxes.
As measured data set for this study, field measurements carried out from 2009 to 2018 at six sites in the Swabian Alb and Kraichgau (Southern Germany) within the collaborative research project Regional Climate Change (DFG Research Unit: FOR 1695) will be used. The dataset includes meteorological conditions, soil moisture and mineral nitrogen, crop development stages, biomass, carbon and nitrogen content, and field management information. The soil-crop-atmosphere system processes will be modelled using the agro-ecosystem program package Expert-N. Prediction quality will be assessed through cross-validation.
The observable differences in plant carbon and nitrogen content, biomass, phenology and yield across the sites and years are hypothesised to be caused by environmental factors like differences in prevailing weather conditions, soil characteristics etc. Prediction quality will be assessed for dependencies on environmental conditions, prior information and model capabilities in suitably capturing crop development stages.
- Bayesian model averaging: A systematic review and conceptual classification. Fragoso T. M., Neto F. L. 2015, Statistical Science
- Multimodel ensembles of wheat growth: many models are better than one. Martre P., Wallach D., Asseng S. et al. 2015, Global Change Biology, pp. 21, 911-925
- Bayesian model averaging to explore the worth of data for soil-plant model selection and prediction. Woehling T., Schoeniger A., Gayler S., Nowak W. 2015, Water Resources Research