Adaptation to the multi-scale impacts of climate change in water resources systems is challenged by substantial uncertainty in future hydrologic projections (Asadieh & Krakauer, 2017;Dottori et al., 2018;Wilby & Dessai, 2010), as well as land use and water demand (Clarke et al., 2018). Under these conditions, dynamic planning provides a basis for responding to new observations as they occur, aiming to prevent both over-and under-investment (De Neufville & Scholtes, 2011;Walker et al., 2013). The implementation of dynamic decisions requires a control policy mapping observed indicator variables to actions, which can be optimized to determine the sequence, timing, and/or threshold values on which actions are conditioned (Herman et al., 2020). Several difficult aspects of the infrastructure planning problem, such as discrete actions, implementation delays, and path dependency, may prevent analytical approaches to optimal control. However, numerical approaches have been widely used, including stochastic dynamic programming (