Background With the vegetation restorating, vegetation in the hilly-gully region of the Loess Plateau has been gradually improving. However, due to the shelter of vegetation, the Digital Elevation Model (DEM) cannot well describe the real undulation of ground surface. But the vegetation in the hilly-gully region of the Loess Plateau are quite different between the growing season and the non-growing season, thus this study mainly analyzed the influence of vegetation seasonal changes on the production of high-precision DEMs.

Methods This study took Qiaogou, Suide county, Shaanxi province as the typical hilly-gully region of the Loess Plateau. Three different sites were selected and surveyed by 3D laser scanning. Based on the point cloud obtained by 3D laser scanning in September and December 2018, after vegetation filtering and interpolating (Kriging or triangular irregular network, TIN), the difference of DEMs obtained by two surveys was compared and analyzed.

Results 1) Vegetation seasonal change presented effects on point cloud surveyed by 3D laser scanner. The mean elevation of point cloud in Sept. was higher than that in Dec. by 0.22, 0.47 and 0.07 m in site A, B and C, respectively. That result was mainly from the vegetation seasonal changes. 2) Slope-based filtering increased the accuracy of DEM. The mean elevation of point cloud in Sept. after filtering was still higher than that in Dec. by 0.15 m, indicating the effects of vegetation seasonal change cannot be eliminated completely by slope-based filtering. 3) Compared with the Kriging, the DEM generated by the TIN was in high accuracy. The gully geomorphological parameters, e.g., gully length and gully depth obtained by Kriging method in Dec. were both often larger than those by TIN. This might result from that Kriging can roughly remove tall vegetation during the vegetation growing season and made the gully edges smooth during the non-growing season. The DEMs interpolated by TIN were closer to the real geomorphology than that by Kriging.

Conclusions Seasonal change of vegetation shows certain effect on DEMs generation. Most of the vegetation can be removed by point cloud vegetation filtering algorithms, and the DEMs generated are closer to the real geomorphology. Regarding different DEMs interpolated methods, the accuracy of both Kriging and TIN can meet the needs of production and research but the accuracy of Kriging is less than that by TIN. The accuracy of DEMs can be effectively improved by choosing the non-growing season for point cloud survey and applying the appropriate vegetation filtering algorithms and interpolation.