Practical and optimal reduction of watershed loads under deep uncertainty requires sufficient search alternatives and direct evaluation of robustness. These requirements contribute to the understanding of the tradeoff between cost and robustness; while they are not well addressed in previous studies. This study thereby (a) uses preconditioning technique in Evolutionary Algorithm to reduce unnecessary search space, which enables a sufficient search; and (b) derives Robustness Index (RI) as a second-tier optimization objective function to achieve refined solutions (solved by GA) that address both robustness and optimality. Uncertainty-based Refined Risk Explicit Linear Interval Programming is used to generate alternatives (solved by Controlled elitist NSGA-II). The robustness calculation error is also quantified. Proposed approach is applied to Lake Dianchi, China. Results demonstrate obvious improvement in robustness after conducting sufficient search and negative robustness-optimality trade-offs, and provides a detailed characteristic of robustness that can serve as references for decision-making.