Photogrammetric techniques for across-scale soil erosion assessment: Developing methods to integrate multi-temporal high resolution topography data at field plots

Soil erosion is a complex geomorphological process with varying influences of different impacts at different spatio-temporal scales. To date, measurement of soil erosion is predominantly realisable at specific scales, thereby detecting separate processes, e.g. interrill erosion contrary to rill erosion. It is difficult to survey soil surface changes at larger areal coverage such as field scale with high spatial resolution. Either net changes at the system outlet or remaining traces after the erosional event are usually measured. Thus, either quasi-point measurements are extrapolated to the corresponding area without knowing the actual sediment source as well as sediment storage behaviour on the plot or erosion rates are estimated disrupting the area of investigation during the data acquisition impeding multi-temporal assessment. Furthermore, established methods of soil erosion detection and quantification are typically only reliable for large event magnitudes, very labour and time intense, or inflexible. To better observe soil erosion processes at field scale and under natural conditions, the development of a method is necessary, which identifies and quantifies sediment sources and sinks at the hillslope with high spatial resolution and captures single precipitation events as well as allows for longer observation periods. Therefore, an approach is introduced, which measures soil surface changes for multi-spatio-temporal scales without disturbing the area of interest. Recent advances regarding techniques to capture high resolution topography (HiRT) data led to several promising tools for soil erosion measurement with corresponding advantages but also disadvantages. The necessity exists to evaluate those methods because they have been rarely utilised in soil surface studies. On the one hand, there is terrestrial laser scanning (TLS), which comprises high error reliability and retrieves 3D information directly. And on the other hand, there is unmanned aerial vehicle (UAV) technology in combination with structure from motion (SfM) algorithms resulting in UAV photogrammetry, which is very flexible in the field and depicts a beneficial perspective. Evaluation of the TLS feasibility reveals that this method implies a systematic error that is distance-related and temporal constant for the investigated device and can be corrected transferring calibration values retrieved from an estimated lookup table. However, TLS still reaches its application limits quickly due to an unfavourable (almost horizontal) scanning view at the soil surface resulting in a fast decrease of point density and increase of noise with increasing distance from the device. UAV photogrammetry allows for a better perspective (birds-eye view) onto the area of interest, but possesses more complex error behaviour, especially in regard to the systematic error of a DEM dome, which depends on the method for 3D reconstruction from 2D images (i.e. options for additional implementation of observations) and on the image network configuration (i.e. parallel-axes and control point configuration). Therefore, a procedure is developed that enables flexible usage of different cameras and software tools without the need of additional information or specific camera orientations and yet avoiding this dome error. Furthermore, the accuracy potential of UAV photogrammetry describing rough soil surfaces is assessed because so far corresponding data is missing. Both HiRT methods are used for multi-temporal measurement of soil erosion processes resulting in surface changes of low magnitudes, i.e. rill and especially interrill erosion. Thus, a reference with high accuracy and stability is a requirement. A local reference system with sub-cm and at its best 1 mm accuracy is setup and confirmed by control surveys. TLS and UAV photogrammetry data registration with these targets ensures that errors due to referencing are of minimal impact. Analysis of the multi-temporal performance of both HiRT methods affirms TLS to be suitable for the detection of erosion forms of larger magnitudes because of a level of detection (LoD) of 1.5 cm. UAV photogrammetry enables the quantification of even lower magnitude changes (LoD of 1 cm) and a reliable observation of the change of surface roughness, which is important for runoff processes, at field plots due to high spatial resolution (1 cm²). Synergetic data fusion as a subsequent post-processing step is necessary to exploit the advantages of both HiRT methods and potentially further increase the LoD. The unprecedented high level of information entails the need for automatic geomorphic feature extraction due to the large amount of novel content. Therefore, a method is developed, which allows for accurate rill extraction and rill parameter calculation with high resolution enabling new perspectives onto rill erosion that has not been possible before due to labour and area access limits. Erosion volume and cross sections are calculated for each rill revealing a dominant rill deepening. Furthermore, rill shifting in dependence of the rill orientation towards the dominant wind direction is revealed. Two field plots are installed at erosion prone positions in the Mediterranean (1,000 m²) and in the European loess belt (600 m²) to ensure the detection of surface changes, permitting the evaluation of the feasibility, potential and limits of TLS and UAV photogrammetry in soil erosion studies. Observations are made regarding sediment connectivity at the hillslope scale. Both HiRT methods enable the identification of local sediment sources and sinks, but still exhibiting some degree of uncertainty due to the comparable high LoD in regard to laminar accumulation and interrill erosion processes. At both field sites wheel tracks and erosion rills increase hydrological and sedimentological connectivity. However, at the Mediterranean field plot especially dis-connectivity is obvious. At the European loess belt case study a triggering event could be captured, which led to high erosion rates due to high soil moisture contents and yet further erosion increase due to rill amplification after rill incision. Estimated soil erosion rates range between 2.6 tha-1 and 121.5 tha-1 for single precipitation events and illustrate a large variability due to very different site specifications, although both case studies are located in fragile landscapes. However, the susceptibility to soil erosion has different primary causes, i.e. torrential precipitation at the Mediterranean site and high soil erodibility at the European loess belt site. The future capability of the HiRT methods is their potential to be applicable at yet larger scales. Hence, investigations of the importance of gullys for sediment connectivity between hillslopes and channels are possible as well as the possible explanation of different erosion rates observed at hillslope and at catchment scales because local sediment sink and sources can be quantified. In addition, HiRT data can be a great tool for calibrating, validating and enhancing soil erosion models due to the unprecedented level of detail and the flexible multi-spatio-temporal application