Urban simulations are governed by strongly coupled thermo-fluid processes involving radiation, turbulence, vegetation drag, and evapotranspiration, which renders parametric design approaches computationally prohibitive at city scale. While advanced simulation models for urban climate studies are already being used intensively to study the significant expected future urban climate changes, simulation-based, goal-oriented optimization studies are still in their infancy. We present a continuous adjoint optimization approach to determine the optimal spatial distribution of urban vegetation for minimizing human heat stress in an urban climate model. To this end, we embed a continuous adjoint solver into the open-source urban climate model PALM [1] to compute gradients of heat-stress functionals with respect to the urban plant canopy distribution. The continuous adjoint of the governing equations is augmented with adjoint enthalpy and mixing-ratio transport equations, vegetation-model counterparts, and surface fluxes derived from an adjoint Monin–Obukhov similarity theory. Moreover, dedicated regularization strategies are introduced to constrain the vegetation field towards realistic tree designs. The gradient-based optimization method is verified against finite-difference gradients. The strategy is subsequently assessed, with a focus on boundary conditions and control, as well as frozen versus coupled thermodynamics. Finally, the method is applied to a selected urban area to reduce heat-stress levels on urban dwellers at critical times of the diurnal cycle. [1] Maronga B. et al., Overview of the PALM model system 6.0, Geoscientific Model Development, 13(3), pp. 1335-1372, 2020.