Objectives Digital elevation model (DEM) is the foundation of many geoscientific applications. Currently there are limited studies with the accuracy evaluation of various open DEM products. We aim to analyze the accuracy of five open DEM products (including ASTER, AW3D30, MERIT, SRTM, and TanDEM-X 90 m).

Methods The five products were compared to airborne LiDAR derived reference DEM and digital surface model (DSM) respectively. Statistical measures including mean error (ME), root mean square error (RMSE), as well as linear error at 90% probability (LE90) were used.

Results The results demonstrate significant differences in the accuracy among 5 open DEM products. None of them meet the designed accuracy in this mountain forest study area. When used as DEM, all 5 DEM products have severe overestimation of the terrain elevation. Among them, the MERIT (ME is 2.82 m,LE90 is 12.19 m ) and SRTM (ME is 4.74 m,LE90 is 11.75 m) has much higher accuracy than the other 3 products. TanDEM-X 90 m DEM has the worst accuracy (ME is 10.39 m,LE90 is 22.83 m). When used as DSM, all 5 DEM products underestimated the elevation of surface objects. Among them, the TanDEM-X 90 m (ME is －3.69 m,LE90 is 16.99 m ) and AW3D30 (ME is －5.34 m,LE90 is 16.81 m ) have much higher accuracy than the other 3 products. The spatial distribution of elevation errors are influenced by the type of land cover. Errors in forested areas are higher than in residential areas. Compared to slope, the height of ground objects has more impact on the elevation errors.

Conclusions When used as DEM, the MERIT and SRTM products have higher accuracy, while when used as DSM, the TanDEM-X 90 m and AW3D30 have higher accuracy.