Multi-purpose reservoirs are critical infrastructure for the management of water resources, and reservoir operators must often balance multiple competing water demands. Climate change is an important threat to the resiliency of water resources in these reservoirs, as it can limit the efficacy of historical operations strategies as well as water demands and availability. The Conowingo Reservoir, a large multi-purpose reservoir located in Pennsylvania and Maryland along the Susquehanna River, has been managed actively for decades, yet current operating policies fail to adequately account for internal system variability and emerging threats to regional water resource reliability. Havre de Grace, Maryland, a small community at the mouth of the Susquehanna River, sources drinking water from the river, and during periods of low riverine freshwater flow, saline waters from Chesapeake Bay can intrude upstream, degrading public water supply and corroding infrastructure. To address this emerging problem, we first develop a statistical model for saltwater intrusion at the Havre de Grace drinking water intake using historical discharge, tide, and wind data from 2007 – 2024. Salinity responses to wind and tidal forcing under low flow conditions are rare and nonlinear, highlighting the need for targeted reservoir releases prior to and during salinity events. Second, we incorporate this model into a simulation-optimization framework for the Conowingo, training adaptive, state-aware, and dynamic release policies that release water from the reservoir downstream to flush out intruding saltwater while continuing to satisfy pre-existing regional water demands. Updating the current operations strategies will expand the scope of water resource management to an emerging compound stressor of water reliability, providing finer control over rare intrusion events that currently, and will continue to, threaten public health and infrastructure.