To evaluate the impacts of time correlated error models on the estimates of soil moisture profiles with the Ensemble Square Root Filter (EnSRF), an observing system simulation experiment was set up, in which the near-surface soil moisture observations were assimilated into NOAH Land Surface Model (LSM). The experiments used a combination technique of EnSRF and state augmentation, in which both the model states and model errors were updated at the same time. The results showed that EnSRF could not successfully estimate the soil moisture profiles, if a large model bias was directly ignored in the process of data assimilation. On the contrary, the soil moisture estimates had a large improvement, if the model errors were corrected. Furthermore, the sensitive tests showed that the value of model bias,the magnitudes of observational errors and the time intervals would all influence the estimates in the different way.