Sensing in building operations is essential to realize intelligent buildings and digital twin-enabled operations. However, buildings have a limited sensing environment with sensor absences, faulty sensors, low redundancy, and less dense deployment because of the inherent building characteristics (e.g., massive, heterogeneous, and long-term). It impedes the advanced building operations. To tackle these issues, this study proposes a novel sensing method for in-situ observation virtual sensors (OVS) in building operations. The proposed OVS is intended to predict the unmeasured variables in real time, without the target sensors. To do so, the OVS is modeled in-situ and then calibrated indirectly without the target observation (Y) so that the virtual sensing can be effectively applied in the building sector, where it is very difficult to establish a laboratory environment having Y for the OVS modeling. The OVS can be developed and operated in the building digital twins, thus extending the physical sensing coverages for digital twin-enabled intelligent operations. For real application, the proposed OVS was developed to observe the return water temperature of a real district heating system. The OVS was experimentally validated with the real measurement to discuss the in-situ OVS model and calibration performance. The OVS could be successfully modeled without using the target observation (Y), and indirect calibration improved the initial OVS performance by 32 %. The OVS demonstrates a significant performance with a root mean square error of 0.55 °C.