Long-term assessment of terrain changes and calculation of erosion rates in an alpine catchment based on SfM-MVS processing of historical aerial images. How camera information and processing strategy affect quantitative analysis

The long-term assessment of terrain changes and calculation of erosion rates in alpine catchments based on Structure from Motion and Multi-View-Stereo (SfM-MVS) analysis of historical aerial images (HAI) has attracted increasing interest in geomorphology in recent years. Processing HAI with SfM-MVS offers tremendous potential for reconstructing surfaces in highly dynamic environments and determining long-term landscape changes over several decades. This study focuses on the effects of image compression and processing strategy on the accuracy of the respective digital terrain models (DTMs) computed from two HAI sets of the Kaunertal valley of different quality and resolution. The DTMs calculated with lens calibration information yielded the highest accuracies (standard deviation = 1.12 m). The differences due to image compression are negligible when focal length and fiducial marks are included in the processing. Not so in the automatic calibration mode, where DTMs calculated from TIFF images are more accurate (StD between 1.40 and 6.61m) than those calculated from JPEG images (standard deviations between 1.56m and 7.45m). The relatively high reprojection and errors of ground control points indicate a stronger influence of systematic errors in these models. All DTMs show a deviation from the normal distribution, but with different strengths. The auto-calibrated data showed surface deformations (doming) in the peripheral parts of the survey area and artefacts distributed throughout the survey area, most pronounced in JPEGbased models. Change detection analysis on a LIA lateral moraine showed, that the mean annual sediment balances are negative throughout the entire observation period and the annual erosion volumes and rates are almost stable (7 to 10 cm). Quantitative analysis revealed highest negative changes in areas in close proximity to the 1953 lateral glacier boundary (–35 m). In the upper part of the slope linear erosion led to the formation of deep and very steep gullies, which serve as transport routes for material mobilized from higher elevations. Location and shape indicate that the deposits occur primarily in areas of larger depressions (due to dead ice) and in the lower slope areas. The case study shows that the fluvial erosional activity in deeply incised gullies remains unchanged over the entire 65-year study period. Overall, our study shows that the inclusion of fiducials and focal length in SfM processing increases the accuracy of the respective DTM, while the image type has a smaller impact. The results presented here demonstrate the capabilities, but also the limitations, of HAI for assessing the long-term geomorphological evolution of alpine catchments over periods relevant to climate change impact assessment.