To improve the soil moisture initial conditions for numerical weather prediction models the potential of assimilating both screen-level parameters (2m-temperature T2m, 2m-relative humidity RH2m) and 1.4 GHz TB is investigated. A soil moisture analysis system based on the Extended Kalman Filter theory is applied to the single column version of the ECMWF numerical weather prediction model for 130 summer days during the MUREX field experiment. A comparison with independent observations shows that the prediction of sensible heat flux, root zone and especially near-surface soil moisture benefits from the assimilation of all three observation types (T2m, RH2m, TB).