Choosing the Optimal Global Digital Elevation Model for Stream Network Delineation: Beyond Vertical Accuracy.

Satellite‐derived global digital elevation models (DEMs) are essential for providing the topographic information needed in a wide range of hydrological applications. However, their use is limited by spatial resolution and vertical bias due to sensor limitations in observing bare terrain. Significant efforts have been made to improve the resolution of global DEMs (e.g., TanDEM‐X) and create bare‐earth DEMs (e.g., FABDEM, MERIT, CEDTM). We evaluated the vertical accuracy of bare‐earth and global DEMs in Central European mountains and submontane regions, and assessed how DEM resolution, vegetation offset removal, land cover, and terrain slope affect stream network delineation. Using lidar‐derived DTM and national stream networks as references, we found that: (a) bare‐earth DEMs outperform global DEMs across all land cover types. RMSEs increased with increasing slope for all DEMs in non‐forest areas. In forests, however, the negative effect of the slope was outweighed by the vegetation offset even for bare‐earth DTMs; (b) the accuracy of derived stream networks was affected by terrain slope and land cover more than by the vertical accuracy of DEMs. Stream network delineation performed poorly in non‐forest areas and relatively well in forests. Increasing slope improved the streams delineation performance; (c) using DEMs with higher resolution (e.g., 12 m TanDEM‐X) improved stream network delineation, but increasing resolution also increased the need for effective vegetation bias removal. Our results indicate that vertical accuracy alone does not reflect how well DEMs perform in stream network delineation. This underscores the need to include stream network performance in DEM quality rankings. Plain Language Summary: We evaluated the accuracy of different types of digital elevation models (DEMs) and derived stream networks in Central European mountains. Our focus was on satellite‐derived global DEMs, including bare‐earth DEMs (i.e., DEMs with vegetation and building offsets removed). Using lidar DTM and national stream networks as references, we found that bare‐earth DEMs consistently outperformed other DEMs across all evaluated land cover types. We observed that slope and land cover have contrasting effects on the accuracy of DEMs and delineated stream networks. As slope increases, DEMs accuracy decreases, while for delineated stream networks, the opposite trend is observed. Where land cover is concerned, DEMs' vertical accuracies (i.e., how well they represent the bare‐earth terrain) are lowest in forests, whereas the accuracy of stream network is lowest in non‐forest areas. Furthermore, we demonstrated that slope and land cover type considerably affect the accuracy of stream network delineation, more so than the vertical accuracy of the DEMs. Additionally, we found that using DEMs with higher resolution can improve stream network delineation, but increasing resolution also increases the need for effective vegetation bias removal. Therefore, removing vegetation and buildings offset from Tandem‐X DEM at a 12 m resolution would represent a major next step forward. Key Points: Differences between digital elevation models (DEMs) (i.e., their absolute vertical accuracy) have a lower effect on stream delineation than terrain slope and landcoverThe use of higher‐resolution DEMs to derive stream networks increases the importance of removing vegetation and building biasVertical accuracy alone does not reflect DEMs' performance in stream delineation, emphasizing the need to include this in quality evaluation [ABSTRACT FROM AUTHOR]