Object-Oriented Bayesian Networks for Holistic Understanding of Groundwater: An Application to Coastal Groundwater Management

It is challenging to model complex groundwater systems if in addition to the physical processes also socio-economic and environmental aspects are considered. In this contribution we propose a new approach for integrated groundwater resources management decision support, utilizing Object-Oriented Bayesian Networks (OOBN) and simulation-based optimization. The proposed methodology incorporates all the available evidence and conflicting objectives to evaluate implications of alternative actions in the decision-making process under uncertainty. An OOBN provides a framework within which the contributions of stakeholders can be taken into account allowing for a range of different factors and their probabilistic relationship to be considered. The best non-dominated Pareto front solutions of the simulation-based optimization are selected and the corresponding water allocation policies are then used to train the BNs. The developed BNs are validated using a new efficient sensitivity analysis method which is based on Relevance Reasoning (RR). The proposed methodology is applied to the south Al-Batinah region in Oman which is affected by saltwater intrusion into a costal aquifer due to excessive groundwater extraction for irrigation agriculture. This in turn has led to a conflict among various water actors as well as uncertainty about the consequences of water management interventions. The complex water system in the study area is represented at a farm scale by individual BNs. These individual BNs are linked to produce an OOBN, which is then used to simulate the water system at aquifer level. The results of our investigation demonstrate that the OOBN developed is a powerful tool to analyze the entire water system simultaneously, characterizes complex uncertainty information as probabilities, and suggests best decision pathways under uncertainty evaluating different management options. In addition, our practice of using sensitivity analysis to validate the Bayesian networks for groundwater management indicates that the technique is useful in modeling Bayesian networks.