A numerical hydrodynamic simulation of the Texas-Louisiana continental shelf circulation is compared with hydrographic and moored observations of temperature, salinity, and current velocity. The spatial and temporal structure of the model error suggests that the model is able to reproduce the observed broad-scale features and seasonal patterns faithfully. However, there are energetic features that have small spatial and temporal scales, 30-50 km and 5-12 days, that are not reproduced in the model. These small-scale features are not well resolved concurrently in both space and time by most modern hydrographic and moored measurements, and thus appear as observational noise when comparing observations with numerical simulations. We demonstrate that these features are reproduced statistically within the model in terms of mean and variance, but are not reproduced exactly. Thus, these features create a substantial noise floor that constrain assessments of model skill to large spatial and temporal scales.