Equity-aware multi-objective optimization of nature-based solutions for flood susceptibility reduction

Nature-based solutions (NBS) are increasingly promoted for flood mitigation, yet how to allocate limited resources while balancing aggregate effectiveness and social equity remains poorly understood. In particular, it is unclear whether directing greater mitigation benefits toward socially vulnerable communities necessarily requires substantial losses in overall flood-susceptibility reduction. We develop an equity-aware multi-objective optimization framework for allocating tree planting and wetland restoration across census block groups under fixed budget, suitability, and intervention-capacity constraints. The framework explicitly represents the trade-off between aggregate flood-susceptibility reduction and vulnerability-weighted mitigation benefit, while high-priority locations are identified using a composite flood risk metric integrating flood susceptibility, population exposure, and social vulnerability. We apply the framework to Dorchester and Somerset Counties, Maryland, and evaluate exact optimization alongside greedy, random, and reinforcement-learning benchmarks. Under a USD 20 million budget, moderate vulnerability weighting produced substantial gains in vulnerability-weighted mitigation benefit with little loss in aggregate performance. In Dorchester, aggregate susceptibility reduction changed from 33.47% under the pure-efficiency solution to 33.21% under moderate vulnerability weighting, while vulnerability-weighted reduction increased from 40.48% to 42.06%. In Somerset, aggregate reduction changed from 31.97% to 31.62%, while vulnerability-weighted reduction increased from 34.58% to 36.37%; population-weighted reduction also increased slightly from 34.91% to 35.38%. The same trade-off persisted when expenditure and intervention mix were held constant and across alternative budgets, census-block-group inclusion thresholds, intervention assumptions, and parameter settings. Greedy allocation closely reproduced the exact optimum, whereas reinforcement learning remained substantially suboptimal, indicating that classical optimization is better suited to the deterministic allocation problem considered here. Overall, the results show that meaningful improvements in the distribution of NBS mitigation benefits toward socially vulnerable communities can be achieved at only a modest cost in aggregate flood-susceptibility reduction. The framework provides a transparent approach for evaluating equity–efficiency trade-offs in spatial flood-mitigation planning.