Land surface models struggle to accurately and precisely predict evapotranspiration (ET) and streamflow. We hypothesize this is partially due to poor representation of the spatial variability of relevant land surface model parameters. For example, land surface models parameterize the slope of the Medlyn stomatal conductance equation (g1) by plant functional type (PFT), but g1 varies widely within PFTs. We developed a parameterization scheme based on trait-environment relationships that predicts grid cell-specific g1 within the Catchment-CN4.5 land surface model as a function of local mean precipitation and canopy height. Although the functional form of this relationship was prescribed, its parameters were updated within Catchment-CN4.5 using particle swarm optimization constrained by observed ET and streamflow. We compared this model to a version of Catchment-CN4.5 that optimized PFT-based g1 values under the same optimization function and to one that used default parameters. We found that trait-environment-based g1 had significant within-PFT variability in forests. Furthermore, the trait-environment-based Catchment-CN4.5 outperformed the default Catchment-CN4.5 for mean absolute error by 11.8% for ET and 20.1% for streamflow. It also outperformed the PFT-based optimization of Catchment-CN4.5 for streamflow, decreasing mean absolute error by as much as 80% in some basins and by 5.4% across all basins. The PFT-based optimization, however, performed better for ET predictions by 1.2%. When extrapolating beyond where the trait-environment relationship was fit, the trait-environment-based model outperformed the default model by 7.8% and was outperformed by the PFT-based optimization by 2.8%. Our work demonstrates that trait-environment-based parameterization can improve land surface model performance.