Due to the hardware and network conditions, data collected in Wireless Sensor Networks usually suffer from loss and corruption. Most existing research works mainly consider the reconstruction of missing data without data corruption. However, the inevitable data corruption poses a great challenge to guarantee the recovery accuracy. To address this problem, this letter proposes a data recovery method based on tensor singular value decomposition. Data collected by the spatial distributed sensor nodes in each time slot is arranged in matrix form instead of vector to further exploit the spatial correlation of the data. Therefore, data collected in consecutive time slots can form a three-way tensor. To avoid the influence of corruption on recovery accuracy, a Tensor Robust Principal Component Analysis model is developed to decompose the raw data tensor into a low-rank normal data tensor and a sparse error tensor. The recovery accuracy is further improved by incorporating total variation constraint. Computer experiments corroborate that the proposed method significantly outperforms the existing method in the recovery accuracy.