Abstract: Objectives Regional height reference frame is usually maintained by geodetic leveling network, however, leveling observation is characterized by heavy workload, high cost and low efficiency, which makes the height reference frame quite difficult to maintain, especially in the area of land subsidence. We present a method to dynamically maintain the regional height reference frame based on the combination of GNSS reference station and surface mass loading data aiming to reduce the workload and cost and improve the timeliness. Methods In this method, GNSS reference stations are selected as the core nodes of the regional height reference frame, GNSS observations and surface mass loading data (land water storage, sea level height and atmospheric pressure) are used to determine the normal height changes and then update the normal heights at these core stations. The normal height change is composed of change in geodetic height and height anomaly. The GNSS reference stations are taken as the core nodes of height control network, geodetic height changes of these stations are obtained by GNSS coordinate time series analysis using least squares method. Height anomaly changes at these stations are computed using the remove-restore method and local Green's function integration based on surface mass loading data. Results The method is validated using GNSS observations over 8 years at 5 reference stations, surface mass loading data and multi-session geodetic leveling data in the subsidence area of Beijing. The numerical results demonstrate that the accuracy of the normal height changes determined from the combination of GNSS data and surface mass loading data is better than 11 mm, with most stations achieving accuracy better than 5 mm. The maximum extrapolation error for one month of normal height changes is 13 mm. Conclusions By applying the height anomaly change corrections that derived from the surface mass loading data using the remove-compute-restore technique and the Green's function integral method, the accuracy of normal height changes can be significantly improved. By selecting the GNSS reference stations with longterm continuous observations and good data quality as the core stations, the dynamic maintenance of regional height reference frame can be realized.