Ocean forecasts include observational data that describe the surface, such as sea surface height/temperature, or profiles of temperature and salinity in only the upper part of the water. This leaves the deep ocean initially uninfluenced by inclusion of observations for these forecasts. We use two ocean forecasts that each contain 32 sub-forecasts, each with slightly changed initial ocean states, to analyze if differences in the structure of the deep ocean help determine the outcome of sub-forecast performance. The time-period chosen includes both a challenging event to predict, as well as deep ocean observations, allowing for a direct comparison between sub-forecast output and collected data. We then use an analysis method to help determine key differences in the size and timing of the deep ocean structures found in the models that do, or do not, perform well.