Modelling of ammonia volatilisation in fertilised and flooded rice systems

In flooded rice systems that are broadcast with urea, significant amounts of nitrogen (N) may be lost to the atmosphere in the form of ammonia (NH3). Many models with different complexities with regards to describing the process of NH3 volatilisation and the overall N dynamics in the systems are available. However, given the differences in local conditions, both too simple and too complex models may not be able to predict NH3 volatilisation correctly or may lead to large prediction uncertainties. Therefore, the main objective of this thesis is to provide a framework to determine an appropriate process-based model with corresponding uncertainty characteristics for estimating NH3 volatilisation in fertilised and flooded rice systems. As a first step in the selection of a model for a specific application, an overview on the modelling concepts and the performance of 14 models developed to simulate N dynamics in flooded soil systems is given. Next, in order to understand differences in modelling concepts for a specific process, co-validation was conducted at single process level: urea hydrolysis, NH3 volatilisation, and floodwater pH. Then, a new process-based model for estimating NH3 volatilisation in fertilised and flooded rice systems, which is of a complexity appropriate for scarce soil N data, is presented and evaluated with field observations. For the flooded rice systems in the Philippines, conceptualisation of the two-step urea hydrolysis, partitioning between ammonium and NH3, and a time-varying rate coefficient of NH3 volatilisation in the proposed model improved the prediction of the net NH3 loss. Subsequently, a set-membership parameter estimation approach with soft-error-bounds was used to characterise the uncertainty in the parameter estimates in the proposed model. The set-membership approach is appropriate for poor quality data sets as it allows simultaneous consideration of the different sources of uncertainty affecting the model prediction, such as uncertainty in the model structure, parameters, and observations. Findings of this study can be used as criteria for stakeholders to make an informed selection of models, to modify the existing models for a specific purpose, and to interpret model-output responses critically.