Conceptual urban water metabolism (CUWM) models provide a holistic view of the efficiency of urban water systems. These models can be linked with multi-agent reinforcement learning (MARL) models to mimic stakeholders' reactions to various strategies. However, the outcomes derived from CUWM-MARL models come with uncertainty. As a result, this paper introduces a decision support system (DSS) that is aware of these uncertainties and uses this information to select robust management strategies based on the output of CUWM-MARL models. First, future scenarios are generated by exploring all combinations of critical values of deeply uncertain variables and values of uncertain variables sampled from an appropriate multivariate copula distribution. Then, a computationally efficient surrogate model is developed to alleviate the computational load of the MARL sub-model. The surrogate model is run through all future scenarios to calculate the system's key performance indicators (KPIs) for each management strategy. Urban water managers can use these KPIs and social choice methods to find consensus management strategies. The proposed methodology has been implemented in the western part of the Tehran metropolitan area, Iran. This study evaluates 26 management strategies consisting of various projects, including reducing water distribution network leakage, implementing urban demand control projects, and utilizing semi-centralized or decentralized wastewater treatment plants. The strategy chosen by the uncertainty-aware DSS enhances the total utility and fairness indices by 125% and 4%, respectively. Moreover, it effectively improves groundwater quality, reduces energy consumption and greenhouse gas emissions, and enhances water supply reliability, ultimately contributing to farmers' job security.
