Mitigation of drought and flood rely on objectives that are often a combination of several flow thresholds to be respected for different locations downstream the reservoirs. In this context, multi-objective optimisation techniques quickly show their limits due to the curse of dimensionality (Bellman, 1957). To tackle this issue, we propose an approach in which we first evaluate the risk of non-achievement of each objective independently for a given climatology and a given state of the system. Then, we derive management rules by prioritising the riskiest objectives in the daily decision making.This approach is applied on the Seine catchment (located in the North of France), which is equipped with a system of four large reservoirs to protect against floods and water shortages multiple locations downstream including the Paris region.First, catchment naturalized flows are modelled at a daily time step with a semi-distributed GR4J model based on the R package airGRiwrm (Dorchies et al., 2022) and forced by 11 GCM/RCM scenarios for both RCP4.5 and RCP8.5 between 1950 and 2100.Then, these flows are used to assess the risk of non-achievement of each objective taking into account the current reservoir volume, the day of the year and a selection of climate scenarios and periods. This assessment is derived from the statistical distribution of the minimum (resp. maximum) volume required in the reservoirs for a given drought (resp. flood) objective calculated by a single objective dynamic programming optimisation. The result of this assessment is available to the public through an interactive Shiny interface (http://irmara.g-eau.fr) that allows to experiment management scenarios in real time.Finally, management rules are derived by prioritising the riskiest objectives and balancing proactive and reactive decisions taking into account a hedging policy. The performance of this management is compared to the current management of the reservoirs over the historical and future periods.This approach has the advantage of providing a decision based on a risk assessment and prioritisation process that allows the manager to justify the decision and paves the way for an operational application.ReferencesBellman, R. Dynamic programming. Princeton, N.J.: Princeton University Press, 1957.Dorchies, David, Olivier Delaigue, et Guillaume Thirel. « airGRiwrm: Modeling of Integrated Water Resources Management based on airGR. R Package version 0.6.1 ». Portail Data INRAE, 7 mars 2022. https://doi.org/10.15454/3CVD1I.