Abstract. The Ross Sea is a region characterized by high primary productivity in comparison to other Antarctic coastal regions, and its productivity is marked by considerable variability both spatially (1–50 km) and temporally (days to weeks). This variability presents a challenge for inferring phytoplankton dynamics from observations that are limited in time or space, which is often the case due to logistical limitations of sampling. To better understand the spatiotemporal variability in Ross Sea phytoplankton dynamics and to determine how restricted sampling may skew dynamical interpretations, high-resolution bio-optical glider measurements were assimilated into a one-dimensional biogeochemical model adapted for the Ross Sea. The assimilation of data from the entire glider track using the micro-genetic and local search algorithms in the Marine Model Optimization Testbed improves the model–data fit by  ∼ 50 %, generating rates of integrated primary production of 104 g C m⁻² yr⁻¹ and export at 200 m of 27 g C m⁻² yr⁻¹. Assimilating glider data from three different latitudinal bands and three different longitudinal bands results in minimal changes to the simulations, improves the model–data fit with respect to unassimilated data by  ∼ 35 %, and confirms that analyzing these glider observations as a time series via a one-dimensional model is reasonable on these scales. Whereas assimilating the full glider data set produces well-constrained simulations, assimilating subsampled glider data at a frequency consistent with cruise-based sampling results in a wide range of primary production and export estimates. These estimates depend strongly on the timing of the assimilated observations, due to the presence of high mesoscale variability in this region. Assimilating surface glider data subsampled at a frequency consistent with available satellite-derived data results in 40 % lower carbon export, primarily resulting from optimized rates generating more slowly sinking diatoms. This analysis highlights the need for the strategic consideration of the impacts of data frequency, duration, and coverage when combining observations with biogeochemical modeling in regions with strong mesoscale variability.