The paper presents a new method of digital terrain model (DTM) estimation based on modified moving average interpolation. There are many methods that can be employed in DTM creation, such as kriging, inverse distance weighting, nearest neighbour and moving average. The moving average method is not as precise as the others; hence, it is not commonly comprised in scientific work. Considering the high accuracy, the relatively low time costs, and the huge amount of measurement data collected by multibeam echosounder, however, the moving average method is definitely one of the most promising approaches. In this study, several variants of this method are analysed. An optimization of the moving average method is proposed based on a new module of selecting neighbouring points during the interpolation process-the "growing radius" approach. Tests experiments performed on various multibeam echosounder datasets demonstrate the high potential of this modified moving average method for improved DTM generation.