Plant community composition influences carbon, water, and energy fluxes at regional to global scales. Vegetation demographic models (VDMs) allow investigation of the effects of changing climate and disturbance regimes on vegetation composition and fluxes. Such investigation requires that the models can accurately resolve these feedbacks to simulate realistic composition. Vegetation in VDMs is composed of plant functional types (PFTs), which are specified according to plant traits. Defining PFTs is challenging due to large variability in trait observations within and between plant types and a lack of understanding of model sensitivity to these traits. Here we present an approach for developing PFT parameterizations that are connected to the underlying ecological processes determining forest composition in the mixed-conifer forest of the Sierra Nevada of California, USA. We constrain multiple relative trait values between PFTs, as opposed to randomly sampling within the range of observations. An ensemble of PFT parameterizations are then filtered based on emergent forest properties meeting observation-based ecological criteria under alternate disturbance scenarios. A small ensemble of alternate PFT parameterizations is identified that produces plausible forest composition and demonstrates variability in response to disturbance frequency and regional environmental variation. Retaining multiple PFT parameterizations allows us to quantify the uncertainty in forest responses due to variability in trait observations. Vegetation composition is a key emergent outcome from VDMs and our methodology provides a foundation for robust PFT parameterization across ecosystems.