Wind speed sensors of wind turbines are prone to suffer from performance degeneration or even drastic failures due to their inherent issues or environmental influence, which directly affect the quality of measurements and the safe operation of wind turbines. To overcome the limitations of inaccuracy and the partial missing of measurements by physical sensors, this work presents a digital twin-driven sensing methodology that enables to augment the physical sensor platform into virtual sensor arrays, thus, identifying the faulty sensors and accommodating them with appropriate estimated data. The proposed method is built upon a series of estimators, verifiers, setters, and selectors and focuses on fault identification, data verification, and reconstruction. The estimators, corresponding to virtual sensors of all physical wind speed sensors, are developed to detect the fault sensors and reconstruct the normal behaviors by using a spatiotemporal network based on multiturbine correlations model. Then, the verifiers are built to check whether the detection results by estimators are accurate by establishing spatiotemporal correlation constraints. In addition, the setters are built to adjust the estimated values and improve the safe operation of turbines based on the estimated values. The selectors are used to confirm the input data source for the turbine control system according to the identification results of both estimators and verifiers. Finally, a comprehensive statistical analysis on five baseline models and three wind speed datasets is conducted. The root mean square error and uncertainty of the proposed method is 0.45 and ±0.009, which demonstrate that the DTSense is capable of conducting the virtual sensor arrays for physical sensing platform effectively and improving the reliability of sensors in engineering applications.