To investigate relevant processes as well as to predict the possible impact of soil erosion, many soil erosion modelling tools have been developed. The most productive development of process-based models took place at the end of the 20th century. Since then, the methods available to observe and measure soil erosion features as well as methods to inter- and extrapolate such data have undergone rapid development, e.g., photogrammetry, light detection and ranging (LiDAR) and sediment tracing are now readily available methods, which can be applied by a broader community with lower effort. This review takes 13 process-based soil erosion models and different assessment techniques into account. It shows where and how such methods were already implemented in soil erosion modelling approaches. Several areas were found in which the models miss the capability to fully implement the information, which can be drawn from the now-available observation and data preparation methods. So far, most process-based models are not capable of implementing cross-scale erosional processes and can only in parts profit from the available resolution on a temporal and spatial scale. We conclude that the models’ process description, adaptability to scale, parameterization, and calibration need further development. The main challenge is to enhance the models, so they are able to simulate soil erosion processes as complex as they need to be. Thanks to the progress made in data acquisition techniques, achieving this aim is closer than ever, if models are able to reap the benefit.

The Shivalik foothills of northwestern India are very prone to soil erosion by water due to undulating slopes, highly erodible soils and high intensity rainstorm events during monsoon season. Physically based soil erosion modeling is seen as viable method for planning of measurements to reduce damages done by soil erosion. Nevertheless, parametrization of such models is a major challenge for large inaccessible areas. Several methods do exist for the estimation of the input parameters skin factor, surface roughness and resistance to erosion for the physically based soil erosion model EROSION-3D. Four rainfall experiments, each including dry and wet run, were conducted on different land use conditions on a research farm of the Regional Research Station Ballowal Saunkhri to test estimation methods. Modeling parameters were determined from these experiments. Parameter estimation by two methods for experimental conditions produced values in close range to experimentally determined values for resistance to erosion and surface roughness. Therefore, existing estimation methods are considered to be applicable for the conditions of the Shivalik mountains, except for skin factor. A first modeling with EROSION-3D using preliminary data of a small example catchment shows uncertainties resulting from range of determined and estimated soil parameters.

Soil processes are known to stretch over many scales – some processes, like erosion, are of particular interest due to their quick and complex characteristics with high impact. The analysis of soil erosion processes is challenging through heterogeneous field situations, involved spatio-temporal scales and by a reconfiguration of the system itself. Various experimental procedures and analytical methods were developed, which can analyze erosion processes. But because the procedures are driven by specific model assumptions which in effect also relate to a plethora of central state variables and parameters, the data of different groups are rarely compatible. Interoperability is hindered further through inhomogeneous data structures and a lack of metadata.Within the NFDI4Earth pilot SoilPulse (soilpulse.github.io) we are developing an interactive metadata generator which shall assist researchers to make their soil process related data sets reusable by humans and machines. Instead of forcing the user to adhere to a defined metadata standard, the tool semi-automatically and interactively builds a translation procedure i) to map various existing data structures to a common scheme and ii) to feedback valuable but missing information to be provided by the researcher. While treating a dataset the researcher is aided by visualizations of the data in relation to other datasets which are already made machine readable through SoilPulse, allowing to easily discover non-plausible data and errors within the dataset. Once treated the dataset can be queried along with other datasets through a common interface and can be linked to erosion models through an API. The PICO presentation demonstrates the functionality of the SoilPulse metadata generator prototype and invites attendees to apply it themselves on their data sets. As SoilPulse is in active development we highly appreciate comments, hints and impulses to further improve the tool!

Measuring runoff formation on soil surfaces by rainfall simulators predominantly provide lumped values without spatiotemporal information in regard to water dynamics (e.g., water ponding timing and connectivity). Understating spatial and temporal variations of water storage on soil surface is key to assess hydrological connectivity and runoff generation. Furthermore, it is very relevant for erosion studies. Computer vision and deep learning has presented state-of-the-art results in environmental sciences, for instance to segment water using cameras as gauges or performing flood mapping with remote sensing images. However, automatic mapping of water forming on soil surfaces due to rainfall is very challenging because the water area is considerably smaller and water ponds present complex shapes and similar color characteristics to the soil itself, which is a challenge for deep learning models. The aim of this study is to assess the potential of computer vision and deep learning to estimate water ponding timing, connectivity and runoff formation behavior during rainfall simulations, with emphasis on data imbalance and label uncertainty.We conducted rainfall simulations at three different soil erosion plots with different soil and tillage caracteristics. Runoff was measured at the plot outlet. We collected time lapse images from the plot surface. And ground control points for model scaling were measured with a total station. To train the deep learning models, we manually labeled a selected set of images from all the plot images to derive binary masks (i.e., water and background). We trained three different convolution neural networks (CNN) and further considered techniques that take class imbalance and label uncertainty into account. Eventually, we assess the performance of ensemble models. We applied the best model on the whole set of time lapse images and measured the water pixel area and pond connectivity in terms of connected components. Our findings suggest that considering class imbalance and label uncertainty is key to reach satisfactory segmentation performance, being more important than the model architecture. Furthermore, ensemble models result in better performance when compared to single models. By comparing the measured discharge and the water area derived from the best deep learning model, we can observe different characteristics of the runoff formation related to distinct ponding and intensity of ponding and connectivity. Our approach presents an innovative visual and automatic observation option to quantify the water pond formation and its spatial temporal development. It is a step towards a better understanding of the runoff generation.

Within the research project RUINS we assess the risk of inland floods of the Krummhörn region at the German North Sea coast. One third of this area lies below mean sea level, which demands to drain inland water during low tides by sluicing or otherwise by pumping. If, at any point in time, the drainage demand exceeds the drainage capacity, the available storage in polders and canals will be filled. Once this storage capacity is exceeded inland floods will occur.Previous risk assessment for such inland floods assumed a constant daily drainage capacity, which results from the installed pump capacity. We analysed process data provided by the operator of the drainage system (1. Entwässerungsverband Emden) at sub hourly resolution. The recorded water levels within the canal system showed that under current conditions the maximum areal drainage capacity is usually limited by the flow capacity within the canal network. The capacity of the pumps is dependend on the gradient from canal to North Sea water level.Under increased tidal water levels in the North Sea (e.g. storm flood situations) the pumping capacity can drop below the canal flow capacity. In consequence the areal drainage capacity is variable and can become much smaller than the constant daily drainage capacity assumed in previous studies. Due to the predicted increase in mean sea level with climate change the area might face an increased risk of inland floods despite a situation of insignificant changes in predicted rainfall patterns.Instead of costly infrastructural improvements, we propose a forecast system i) to optimise the drainage capacity in foresight of short term extreme situations and ii) to enable preparation for inland floodings. The proposed system includes the inherent uncertainty of the analysed processes and predicts the magnitude of upcoming inland floods. Currently, we use synthetic data as drivers, but these shall be exchanged by available weather and tide level predictions. The forecast system is realized as online accessible app, providing an easy usable and understandable access point for the operator and the interested public.