Studies which have attempted to assimilate remotely-sensed soil moisture (SM) into land-surface models have mainly focused on the application of retrievals from active and passive microwave (MW) sensors. However, SM retrievals from thermal (TIR) sensors have been shown to add unique information especially in areas where MW sensors are not able to provide accurate SM retrievals (i.e. moderate to dense vegetation). In this study, the TIR-based product is provided by a soil moisture methodology associated with surface flux estimates from the Atmosphere Land Exchange Inverse (ALEXI) model, while the MW-based product is provided by Vrijie Universiteit Amsterdam (VUA)-NASA surface SM product based on Land Surface Parameter (LPRM) model. Here a series of data assimilation experiments using an ensemble Kalman filter (EnKF) are proposed to quantify the impact of assimilating TIR and MW SM retrievals in isolation and within a dual assimilation framework: (a) no assimilation, (b) only ALEXI SM, (c) only LPRM SM and (d) ALEXI and LPRM SM. The relative skill of each assimilation configuration is assessed through a data-denial experimental design, where the LSM is forced with an inferior precipitation dataset. The ability of each assimilation configuration to correct for precipitation errors is quantified through the comparison of SM predictions with a LSM simulation which is forcing with a high-quality precipitation dataset. Finally, the entire assimilation framework is repeated using a new technique based on the estimation of triple collocation error as an estimate of retrieval error in the EnKF. These results are compared to the first series of assimilation experiments which employ a constant, ad-hoc specification of model and retrieval error in the EnKF to assess the impact of a more sophisticated representation of error.