<p>Studies have demonstrated that the geodetic global navigation satellite system (GNSS) stations enable sea level measurement using GNSS interferometric reflectometry (GNSS-IR). This technique benefits from a rich wealth of signals, allowing for the sea level retrieval using signal-to-noise ratio (SNR), pseudorange combination and phase combination of multi-constellation GNSS observations. Furthermore, these observations will provide more retrievals contributing, and limited attempt has yet been made by researchers to integrate them to improve accuracy. A new multi-type observation combination method is developed to enhance GNSS-IR sea level retrieval, considering unequal precision retrievals of different observation types and acceleration of sea level variation. Three GNSS sites were tested, of which SC02 and AT01 stations located in the USA, and MAYG station on the east coast of Africa. The results show that the initial retrievals only reach decimeter-level accuracy, and there are differences in retrieval performance of the three stations for three observation types. However, the multi-type and multi-GNSS sea level retrievals using the new method greatly improve the accuracy by more than 70% compared with the initial retrievals. The accuracy of multi-type retrievals, with ability to achieve centimeter level, is better than that of multi-GNSS retrievals from only single observation type. The weight changes in the iterative process of Helmert variance component estimation shows that SNR, pseudorange combination and phase combination are unequal precision observations. There are great differences in retrievals among different stations.</p>