With recent widespread usage of state-of-the-art technology (e.g., various mobile devices), environmental sensing is getting popular. The sensors used for sensing are small and due to the mobility they become more error-prone, which results in data corruption or loss from sensor. Therefore, cleaning of the sensed data is of high importance to recover the lost or corrupted data. In this paper, we propose a novel data cleaning mechanism to ensure better accuracy in environmental sensing applications. Based on the sensed data and the context relationship of each sensor, we update the credibility (or alternatively reliability) of the sensed data. We consider mobility pattern of the mobile sensor nodes while selecting the candidate sensor nodes for data stream cleaning. Through simulations, we evaluate the performance of our proposed approach. We compare our proposed sensor data stream cleaning approach with Influence Mean Cleaning (IMC) (a recent algorithm in data stream cleaning) and Mean-based cleaning. Simulation results show up to 24% reduction in root mean square error (RMSE) over IMC and up to 30% over Mean-based cleaning.