This study evaluates the sensitivity of a multiscale ensemble assimilation system to different configurations of satellite soil moisture observations, namely the retrieval accuracy, spatial availability, and revisit time. We perform horizontally coupled assimilation experiments where pixels are updated not only by observations at the same location but also all in the study domain. Carrying out sensitivity studies within a multiscale assimilation system is a significant advancement over previous studies that used a 1-D assimilation framework where all horizontal grids are uncoupled. Twin experiments are performed with synthetic soil moisture retrievals. The hydrologic modeling system is forced with satellite estimated rainfall, and the assimilation performance is evaluated against model simulations using in-situ measured rainfall. The study shows that the assimilation performance is most sensitive to the spatial availability of soil moisture observations, then to revisit time and least sensitive to retrieval accuracy. The horizontally coupled assimilation system performs reasonably well even with large observation errors, and it is less sensitive to retrieval accuracy than the uncoupled system, as reported by previous studies. This suggests that more information may be extracted from satellite soil moisture observations using multiscale assimilation systems resulting in a potentially higher value of such satellite products.