Search

Soil erosion by water is a main form of land degradation worldwide. The problem has been addressed, among others, in the United Nations Sustainability Goals. However, for mitigation of erosion consequences and adequate management of affected areas, reliable information on the magnitude and spatial patterns of erosion is needed. Although such need is often addressed by erosion modelling, precise erosion monitoring is necessary for the calibration and validation of erosion models and to study erosion patterns in landscapes. Conventional methods for quantification of rill erosion are based on labour-intensive field measurements. In contrast, remote sensing techniques promise fast, non-invasive, systematic and larger-scale surveying. Thus, the main objective of this study was to develop and evaluate automated and transferable methodologies for mapping the spatial extent of erosion rills from a single acquisition of remote sensing data. Data collected by an uncrewed aerial vehicle was used to deliver a highly detailed digital elevation model (DEM) of the analysed area. Rills were classified by two methods with different settings. One approach was based on a series of decision rules applied on DEM-derived geomorphological terrain attributes. The second approach utilized the random forest machine learning algorithm. The methods were tested on three agricultural fields representing different erosion patterns and vegetation covers. Our study showed that the proposed methods can ensure recognition of rills with accuracies between 80 and 90% depending on rill characteristics. In some cases, however, the methods were sensitive to very small rill incisions and to similar geometry of rills to other features. Additionally, their performance was influenced by the vegetation structure and cover. Besides these challenges, the introduced approach was capable of mapping rills fully automatically at the field scale and can, therefore, support a fast and flexible assessment of erosion magnitudes.

Environmental observations are crucial for understanding the state of the environment. However, current observation networks are limited in spatial and temporal resolution due to high costs. For many applications, data acquisition with a higher resolution would be desirable. Recently, Internet of Things (IoT) -enabled low-cost sensor systems offer a solution to this problem. While low-cost sensors may have lower quality than sensors in official measuring networks, they can still provide valuable data. This study describes the requirements for such a low-cost sensor system, presents two implementations, and evaluates the quality of the factory calibration for a widely used low-cost precipitation sensor. Here, twenty sensors have been tested for an 8-month period against three reference instruments at the meteorological site of the TU Dresden. Further, the factory calibration of 66 rain gauges has been evaluated in the lab. Results show that the used sensor falls short for the desired out-of-the-box use case. Nevertheless, it could be shown that the accuracy could be improved by further calibration.

Northern rivers are responding to global warming by changes in seasonal discharges, sediment transport rates and morphology. Very limited amount of studies about bedload transport rates have been carried out in the winter-season. Traditional methods of measuring bedload transport are limited by their proneness to user error, small spatial scales and uncertainties related to the equipment itself. Computer vision-based particle image velocimetry (PIV) and particle tracking velocimetry (PTV) methods have been successfully applied to measurements of water surface velocities and preliminary results show that they can be applied to underwater sediment transport velocity measurements. The aims of this study are to 1) to enhance the bedload calculations, by comparing traditional mechanical methods and computer vision-based particle image velocimetry methods applied to underwater video data sets. Additionally, topography created from underwater imagery is used to scale and georeference the results with very good precision, and 2) understand the seasonal variation in bedload transport amounts based on both mechanical and image velocimetry methods.The study is based on field data, measured at sub-arctic Pulmanki river, located in northern Finland. The data has been gathered in 2021 autumn and winter, 2022 spring, 2022 autumn, to cover different possible sediment transport conditions, from low flow ice-covered to high flow open channel periods. The preliminary results are presented. They show that the method is promising in enhancing the understanding of sediment transport processes and the seasonal transported amounts.

Between June and August 2022, the European Forest Fire Information System (EFFIS) reported more fires in Europe than in any other recent summer season. This is particularly true for Central Europe, where the largest forest fire in recent Czech history occurred in the German-Czech border region. With global warming and resulting longer dry periods, the length and severity of wildfire seasons in central Europe will likely increase. Therefore, easy to implement and cost-effective methods to assess wildfire damage and regeneration of the ecosystems are getting increasingly important. In this study we evaluated how different datasets obtained by uncrewed aerial system (UAS) can be incorporated with datasets obtained from the ground to describe the fire affected landscape. Thereby, multi-spectral 3D point clouds were derived from low-cost UAV laser scanning and using structure from motion (SfM) photogrammetry applied to RGB and multi-spectral imagery. The aerial datasets were combined with ground-based terrestrial and mobile laser scanning. The datasets were acquired in several surveys following the forest fire event in the German part of the National park Saxonian/Bohemian Switzerland.Initial results show the potential of UAS-based sensing for efficient mapping of a burned area with a high resolution (600-1000 pts/m²). The combination of point clouds from UAS-based laser scanning and photogrammetry enables a detailed representation of the burned forest with different levels of fire damage (e.g., in still present canopy) when compared to the single datasets. The UAS based laser scanning data reveals higher noise compared to the SfM-based point clouds. However, the accuracy is still sufficient to improve the quality of orthomosaics in densely vegetated areas. In a next step, further investigations on data accuracy are conducted and automated point cloud fusion algorithms based on classified points are considered.

Time-lapse photogrammetry has been proven to be a valuable tool to support the understanding of earth surface processes since it can be used to create 3D models with an unprecedented temporal resolution. Overlapping images are captured simultaneously and structure from motion (SfM) photogrammetry is used to reconstruct 3D point clouds from these images automatically.We performed one rainfall simulation on an erosion plot in the field covering about 16 square meters and having a slope of 9 degrees and another experiment in the laboratory with a plot of about 4 square meters and a slope of 20 degrees. Rainfall intensities were similar and high in both simulations to ensure rill formation. During the experiment, we measured soil surface changes with a time series of 3D point clouds derived via SfM photogrammetry. We also estimated runoff flow velocities with a tracer and observed the spatial pattern of runoff velocities with particle tracking velocimetry (PTV) applied to videos captured during the field experiment. At the plot outlet, we also measured runoff and sediment yield. These datasets were used as validation data for the soil erosion model.Soil erosion was simulated with the physically-based model RillGrow and SMODERP, which conceptualizes the formation of erosion rills as a self-organizing dynamic system. We ran several thousand erosion simulations using a Monte Carlo approach. The aim was first to assess the sensitivity of the input parameters of the model, and secondly to automatically find the best fitting set of input parameter values for the given field site conditions and rainfall intensity. We compare the simulation results to global values, such as the sediment yield and runoff, and to local changes, measured by the photogrammetric 4D data.Our study combines established as well as newly developed data recording and processing methods to create a spatio-temporal high-resolution dataset. This is used to test a soil erosion model with the aim of enhancing understanding of rill erosion processes.This research was supported by the projects DFG EL926/3-1, SFZP 085320/2022, SS05010180, and SGS OHK1-086/23/11143.

The measuring of flood events is associated with many challenges. Among them is the determination of flow velocities for the derivation of discharge. Most of the applied methods for velocity determination the disadvantage that they work in direct contact with water. This often makes measuring under critical flow conditions dangerous. Optical measurement methods have a great advantage because they can work remotely, i.e., without water contact.For representative discharge measurements, flow velocity measurements over the entire width of the river cross-section are required. This is a major challenge in the application of PTV, because visible particles must be present across the entire cross-section, which is not always the case. The potential measurement gaps in the surface velocity distribution have a negative effect on the quality of the discharge determination. Because optical measurement methods are relatively new in hydrology, there is not yet a standardised procedure with which the discharge can be determined. The "OptiQ" method presented here is an approach for determining discharge using PTV. This method is based on the continuity equation, which is dependent on two variables, the flow area and the mean flow velocity. The challenge here is to determine the depth-averaged flow velocity, because PTV is used to determine the surface velocity. To get the depth-averaged flow velocities, the PTV results are averaged over a transect and converted using a velocity coefficient. The arithmetic mean, the velocity area method (DIN EN ISO 748:2008-02) and the moving average are considered as averaging methods. A statistical approach was chosen for closing measurement gaps that occurred in the velocity distribution. In this approach, the measurement results with similar discharge conditions in the entire time series, i.e. PTV results for the same water levels, are statistically analysed, filtered and summarised in a lookup table. The gaps in the measurements due to missing particles are filled with the data from the lookup table.For the data collection, three camera gauges were installed at regular gauging stations of the Saxon State Agency for Environmental and Agricultural Monitoring (BfUL). The camera gauges recorded short video sequences at regular time intervals, which were used to determine the velocity distributions using the FlowVelo tool (Eltner et al., 2020). This resulted in three time series covering a period of 10-15 months. For the validation of the optical discharge time series, the regular water level and discharge measurements of the BfUL are used. The application of "OptiQ" shows a significant adjustment of the optically determined discharge data to the reference measurement at all three gauging stations. While acceptable results were determined with the arithmetic mean only at higher discharge, the results with the velocity area method and the moving average are similarly good at all discharges. At the gauging station in Elbersdorf, the average difference from the reference value could be reduced from 29% to 15% with "OptiQ". In the next step, it is planned to further develop the statistical model "OptiQ" by using Deep Learning.

Obtaining real-time water level estimations is crucial for effective monitoring and response during emergencies caused by heavy rainfall and rapid flooding. Typically, this type of monitoring can be a difficult task, requiring river reach preparations and specialized equipment. Moreover, in extreme flood events, standard observation methods may become ineffective. This is why the possibility of developing low-cost, automatic monitoring systems represents a significant advancement in our ability to monitor river courses and allow emergency teams to respond appropriately.Image-based methods for water level estimation facilitate the development of a low-cost river monitoring strategy in a quick and remote approach. These techniques are faster and more convenient regarding the setup than traditional water stage monitoring methods, allowing us to efficiently monitor the river from different locations with a cost-effective approach. By increasing the density of the observation network, we can improve flood warning and management.The approach presented involves placing cameras in secure locations to capture images of the river, for which we have previously modelled the terrain in 3D using Structure from Motion (SfM) algorithms supported by GNSS data. With the images obtained every 15 minutes, we perform a Convolutional Neural Network (CNN) segmentation based on artificial intelligence algorithms that allow us to automatically extract the contours of the water surface area. In this study, two different neural network approaches are presented to segment water in the images.Using a photogrammetric strategy, we reproject the water line extracted by the AI on the 3D model of the scene. This reprojection is also supported by the use of a keypoint detection neural network that allows us to accurately identify the ground control points (GCPs) observed in the images captured by the surveillance camera. This approach allows us to automatically assign to each image the real coordinates of the GCPs and subsequently estimate the camera pose.This AI segmentation and automatic reprojection into the 3D model has allowed us to generate a robust centimetre-accurate workflow, capable of estimating the water level in near-real time for daylight conditions. In addition, the automatic detection of the GCP has permitted to obtain automatic water level measurements over a longer period of time (one year). This approach represents the basis for obtaining other river monitoring parameters, such as velocity or discharge, which allow a better understanding of river floods and represent key steps for the development of early warning systems for flood events.

In recent years, photogrammetric models have gained widespread use in geosciences due to their ability to reproduce natural surfaces. These models offer a cost-effective and user-friendly alternative to other systems, such as LiDAR, for creating 3D point clouds. On the other hand, rockfalls pose a significant risk to society, as they are the most common natural hazard in mountainous areas and can occur with great speed, resulting in high levels of danger. The aim of this communication is to show results on the development of new algorithms and time-lapse photogrammetric systems for automatic rockfall monitoring (Blanch, 2022).To acquire the data, a photogrammetric system consisting of different photographic modules and a data transmission module has been developed. This system uses conventional cameras (24Mpx-48Mpx) powered by solar panels and it is controlled by a Raspberry Pi. The system captures time-lapse images, can be programmed, configured flexibly, and it can send images remotely for near real-time processing. The system has been installed at two sites with rockfall activity. One in the Puigcercós cliff, located in the Origens UNESCO Gobal Geopark (Spain), and the other in the Tajo de San Pedro cliff located in the Alhambra de Granada - UNESCO World Heritage Site (Spain).Data processing comprises two main steps. The first step involves the automatic photogrammetric process using SfM-MVS algorithms. Thereby, the MEMI workflow is applied to improve the level of detection in the change-detection comparison (Blanch et al., 2021). Afterwards, a workflow based on M3C2 (Lague et al., 2013) comparison and DBSCAN clustering is applied to identify possible rockfall clusters. The resulting clusters are processed via a machine learning approach to automatically discriminate the true rockfall events from the candidate clusters . To perform this task, various metric parameters, i.e. features, of the candidate clusters are calculated, and a Random Forest machine learning model is used to perform the classification.The presented approach facilitates the automated monitoring of rockfalls in near-real time, while improving the detection threshold in the 3D change-detection models, resulting in a more detailed characterisation of active zones and defining the framework that allows for automated 4D rockfall monitoring in high temporal frequency.Blanch, X., 2022. Developing Advanced Photogrammetric Methods for Automated Rockfall Monitoring. Doctoral dissertation. http://hdl.handle.net/10803/675397Blanch, X.; Eltner, A.; Guinau, M.; Abellan, A., 2021. Multi-Epoch and Multi-Imagery (MEMI) Photogrammetric Workflow for Enhanced Change Detection Using Time-Lapse Cameras. Remote Sens. , 13, 1460. https://doi.org/10.3390/rs13081460Lague, D., Brodu, N., Leroux, J., 2013. Accurate 3D comparison of complex topography with terrestrial laser scanner: Application to the Rangitikei canyon (N-Z). ISPRS Journal of Photogrammetry and Remote Sensing 82, 10–26. https://doi.org/10.1016/j.isprsjprs.2013.04.009

This study assesses the suitability to use RGB and thermal infrared imagery acquired from an UAV to measure surface flow velocities of rivers. The reach of a medium-scale river in Hungary is investigated. Image sequences with a frame rate of 2 Hz were captured with two sensors, a RGB and an uncooled thermal camera, at a flying height that ensures the visibility of both shores. The interior geometry of both cameras were calibrated with an in-house designed target field. The image sequences were automatically co-registered to account for UAV movements during the image acquisition. The TIR data was processed to keep loss-free image information solely in the water area and to enhance the signal to noise ratio. Image velocimetry with PIV applied to the TIR data and PTV applied to the RGB data was utilised to retrieve surface flow velocities. Comparison between RGB and TIR data reveal an average deviation of about 0.01 m/s. Future studies are needed to evaluate the transferability to other non-regulated river reaches.

Historical aerial photographs captured since the early 1900s and spy satellite photographs from the 1960s onwards have long been used for military, civil, and research purposes in natural sciences. These historical photographs have the unequalled potential for documenting and quantifying past environmental changes caused by anthropogenic and natural factors.The increasing availability of historical photographs as digitized/scanned images, together with the advances in digital photogrammetry, have heightened the interest in these data in the scientific community for reconstructing long-term surface evolution from local to regional scale.However, despite the available volume of historical images, their full potential is not yet widely exploited. Currently, there is a lack of knowledge of the types of information that can be derived, their availability over the globe, and their applications in geoscience. There are no standardized photogrammetric workflows to automatically generate 3D (three-dimensional) products, in the form of point clouds and digital elevation models from stereo images (i.e. images capturing the same scenery from at least two positions), as well as 2D products like orthophotos. Furthermore, influences on the quality and the accuracy of the products are not fully understood as they vary according to the image quality (e.g. photograph damage or scanning properties), the availability of calibration information (e.g. focal length or fiducial marks), and data acquisition (e.g. flying height or image overlap).We reviewed many articles published in peer reviewed journals from 2010 to 2021 that explore the potential of historical images, covering both photogrammetric reconstruction techniques (methodological papers) and the interpretation of 2D and 3D changes in the past (application papers) in different geoscience disciplines such as geomorphology, cryosphere, volcanology, bio-geosciences, geology and archaeology. We present an overview of these published studies and a summary of available image archives. In addition, we compare the main methods used to process historical aerial and satellite images, highlighting new approaches. Finally, we provide our advice on image processing and accuracy assessment.

The importance of optical measurement methods in hydrology is increasing in the last years. In contrast to conventional gauging techniques, they can be applied remotely, making the measurement safe for humans and equipment, even under difficult measurement conditions. One important hydrological parameter to measure is discharge. Deriving discharge with remote sensing can be done by applying particle tracking velocimetry (PTV) in combination with the velocity area method (VAM). VAM is a standardized and established method in hydrology. For reliable discharge results with the VAM, surface flow velocity measurements and thus trackable particles in the case of PTV usage are required across the entire width of the river cross section, which is not always the case in natural observation conditions. To fill these data gaps several statistical methods were investigated that incorporate information provided at different measurement times but with similar discharge conditions.In this study, data were collected over longer time periods with different cameras at a gauging station of a medium scale river in Saxony, Germany. Stationary cameras recorded short videos, which are used to estimate the velocity distribution at the water surface using PTV incorporated in the FlowVelo tool (Eltner, 2020), and afterwards, to estimate the discharge using VAM. The obtained discharge time series from different cameras and camera positions were used to analyse the performance of different gap filling approaches. The results were compared to discharge and water level measurements of the official gauging station maintained by the federal measuring agency. They show, that the adjustment to the data of the reference measurements increases significantly by application of the gap filling methods. Next steps are to enhance the presented methods by using targeted data filtering and deep learning.Keywords: velocity area method, particle tracking velocimetry, camera based discharge estimation

The handling of natural disasters, especially heavy rainfall and corresponding floods, requires special demands on emergency services. The need to obtain a quick, efficient and real-time estimation of the water level is critical for monitoring a flood event. This is a challenging task and usually requires specially prepared river sections. In addition, in heavy flood events, some classical observation methods may be compromised.With the technological advances derived from image-based observation methods and segmentation algorithms based on neural networks (NN), it is possible to generate real-time, low-cost monitoring systems. This new approach makes it possible to densify the observation network, improving flood warning and management. In addition, images can be obtained by remotely positioned cameras, preventing data loss during a major event.The workflow we have developed for real-time monitoring consists of the integration of 3 different techniques. The first step consists of a topographic survey using Structure from Motion (SfM) strategies. In this stage, images of the area of interest are obtained using both terrestrial cameras and UAV images. The survey is completed by obtaining ground control point coordinates with multi-band GNSS equipment. The result is a 3D SfM model georeferenced to centimetre accuracy that allows us to reconstruct not only the river environment but also the riverbed.The second step consists of segmenting the images obtained with a surveillance camera installed ad hoc to monitor the river. This segmentation is achieved with the use of convolutional neural networks (CNN). The aim is to automatically segment the time-lapse images obtained every 15 minutes. We have carried out this research by testing different CNN to choose the most suitable structure for river segmentation, adapted to each study area and at each time of the day (day and night).The third step is based on the integration between the automatically segmented images and the 3D model acquired. The CNN-segmented river boundary is projected into the 3D SfM model to obtain a metric result of the water level based on the point of the 3D model closest to the image ray.The possibility of automating the segmentation and reprojection in the 3D model will allow the generation of a robust centimetre-accurate workflow, capable of estimating the water level in near real time both day and night. This strategy represents the basis for a better understanding of river flooding and for the development of early warning systems.

Airborne Laser Scanning (ALS) is a widely used method in Earth science, Agriculture or Forestry. This method could provide high resolution and accurate spatial data for the better understanding of surface structures, moreover, based on the laser pulses, it can even show important features of the ground below dense vegetation. However, these ALS surveys requires specially designed aircrafts, pilots and operators, detailed flight planning, which leads to an expensive way of data analysis. The application of laser scanners for Unmanned Aerial Systems (UAS) has started in the last few years. These sensor payloads provide less weight and size and decreased accuracy compared to the traditional ALS surveys but still serve as more reliable mapping technology contrary to the photogrammetric methods in many cases. However, several new UAS laser scanners are being developed but their accuracy conditions and applicability for agricultural monitoring must be studied in many ways.In our study we applied the novel Zenmuse L1 LiDAR sensor mounted on a DJI Matrice M300 RTK UAS. We surveyed a ~50 ha area of corn field near Berettyóújfalu, Hungary in the summer of 2021. Our aim was to reveal the applicability of UAS laser scanning for the precise ground surface reconstruction. In this period, the corn was under irrigated condition, therefore, extensive weed patches were observed between the paths. The laser scanner ground filtering data was compared to a photogrammetry-based aerial survey that we have carried out at the beginning of the vegetation cycle at the same parcel. Our results showed both the potentials and limitations of this sensor for precision agriculture. The laser beams produced significant amount of noise between the paths that had to be cleaned to extract the ground surface below the corn canopy. Based on our data processing methods we were able to delineate similar drainage networks within the parcel that was also processed from the initial aerial survey. However, the UAS LiDAR gained the most accurate surface reconstruction at the more clear grassland patches around the parcel. L. Bertalan was supported by the INKP2022-13 grant of the Hungarian Academy of Sciences. This research was funded by the Thematic Excellence Programme (TKP2020-NKA-04) of the Ministry for Innovation and Technology in Hungary. This research was also influenced by the COST Action CA16219 “HARMONIOUS - Harmonization of UAS techniques for agricultural and natural ecosystems monitoring”.

High-resolution information on the processes and rates of soil erosion, transport, and deposition, offer important knowledge for soil erosion modelling, and the protection and sustainable management of soil. It helps improve the cross-scale understanding on aspects as aggregate breakdown, rill erosion, swelling and shrinking effects, and rill-network evolution. As a non-invasive, high-resolution, and cost as well as time-efficient method, Structure from Motion (SfM) presents a valuable tool to calculate soil loss, depict soil surface change detection, and offer high-resolution information on soil and soil erosion processes. Even though SfM shows in general higher erosion rates, due to the influence of non-erosive processes, the technique is altogether in good agreement with the sampling data at the outlet. We monitor soil erosion on multiple erosional plots and with spatial and temporal high-resolution photogrammetry to assess its feasibility over time.For this purpose, we conduct 12 rainfall simulations on a three times one metre plot, on different sides, with different vegetation cover, tillage, and initial soil conditions. Seven to ten synchronized time-lapse cameras are set up around the plot, taking pictures every 10-60 seconds. The data thus obtained allow change detection assessment via digital elevation models of difference at least once per minute. The elevation change by SfM is validated via bulk density measurements, and sampling at the plot’s outlet assessing runoff, and sediment concentration at minute intervals. During an overflow experiment, we measure flow velocity via video using particle tracer and manually via colour tracer, gaining spatial and temporal distribution information on the flow velocity. Using low-cost sensors, we furthermore monitor the progress of the soil moisture and temperature during the whole rainfall simulation.We present sampled and photogrammetric results based on a dozen rainfall simulations at the micro-scale with a very high temporal and spatial resolution. This gives an insight into spatial distribution and development of soil erosion processes on a sub-minute resolution. We compare these data to gain knowledge on the feasibility of temporal and spatial high-resolution SfM soil erosion assessment and their usability for the validation and calibration of process-based soil erosion models.

Rain driven flash floods have severe impacts on society and landscape functions. The July 2021 flood in the Eifel region, west Germany, was one drastic example of such impact. While media and scientists rightfully highlighted the meteorological and hydrological aspects of this flood, it was the concurrent reorganisation of important landscape conditions and the debris carried by the fast flowing water that made this flood so devastating and unpredictable.Here, we take a process-based impact perspective and systematically ask, which were the specific roles of non-hydraulic but geomorphic dynamics that implemented the damage, caused flood non-linearities and amplified the landscape deterioration. We combine insights from field mapping campaigns during, right after and within the relaxation phase of the flood with high resolution geophysical and LiDAR surveys to discuss the role of hillslopes, vegetation, fluvial sediment mobilisation and the legacy of anthropogenic landscape reorganisation. We conclude that some of these elements emerged as the flood event evolved, causing either transient effects or persistent landscape features, thus modifying the response of the landscape to future events, also to less intense precipitation events.Our findings not only support more tailored recovery efforts for the flood affected Eifel catchments, but should also inform landscape development trajectories and potentially crucial factors in other Central European regions.

Climate change, accompanied by intensified extreme weather events, results in changes in intensity, frequency and magnitude of soil erosion. These unclear future developments make adaption and improvement of soil erosion modelling approaches all the more important. Hypothesizing that models cannot keep up with the data, this review gives an overview of 44 process based soil erosion models, their strengths and weaknesses and discusses their potential for further development with respect to new and improved soil and soil erosion assessment techniques. We found valuable tools in areas, as remote sensing, tracing or machine learning, to gain temporal and spatial distributed high resolution parameterization and process descriptions which could lead to a more holistic modelling approach. Most process based models are so far not capable to implement cross-scale erosional processes or profit from the available resolution on a temporal and spatial scale. We conclude that models need further development regarding their process understanding, adaptability in respect to scale as well as their parameterization and calibration. The challenge is the development of models which are able to simulate soil erosion processes as close to reality as possible, as user-friendly as possible and as complex as it needs to be.

Deep learning (DL) methods, integrated with imagery obtained by remote sensors, are considered a novel source of information in the field of hydrometry. Their results can be a support for standard gauging systems. In this research, three different model definitions based on the SegNet architecture were developed: training a new model, using a pre-trained model, and performing transfer learning. The main goal of this approach is the performance assessment of convolutional neural network (CNN) generalization to segment images containing different water bodies automatically. Diverse sensors were used to obtain RGB images from different areas of the world. The effectiveness of the CNN was estimated using pixel accuracy and IoU metrics. Training a new model and using transfer learning revealed similar high accuracies that were at least twice as accurate compared to the pre-trained model. However, the transfer learned model is preferred due to significantly lower training expenses.

While the availability and affordability of unmanned aerial systems (UASs) has led to the rapid development of remote sensing applications in hydrology and hydrometry, uncertainties related to such measurements are still to be quantified and mitigated. Physical instability of the UAS platform inevitably induces motion in the acquired videos and can have a significant impact on the accuracy of camera-based measurements such as velocimetry. A common practice in the data preprocessing stages is the compensation of platform-induced motion by means of digital image stabilisation (DIS) methods, which use the visual information from the captured videos – in the form of physically static features – to first estimate and then to compensate such motion. Most existing stabilisation approaches rely either on in-house built tools based on different algorithms, or on general-purpose commercial software. Intercomparison of different stabilisation tools for UAS remote sensing purposes that could serve as a basis for a selection of a particular tool in given conditions has not been found in the available literature. In this paper we have attempted to summarise and describe several freely available DIS tools applicable to UAS velocimetry purposes. A total of seven tools – six aimed specifically at velocimetry and one general purpose software – were investigated in terms of their (1) stabilisation accuracy in various conditions, (2) robustness, (3) computational complexity, and (4) user experience, using three case study videos with different flight and ground conditions. In attempt to adequately quantify the accuracy of the stabilisation using different tools, we have also presented a comparison metric based on root-mean-squared differences (RMSD) of interframe pixel intensities for selected static features. The most apparent differences between the investigated tools have been found with regards to the method for identifying and selecting static features in videos – manual selection of features or automatic. State-of-the-art methods which rely on automatic selection of features require fewer user-provided parameters and are able to select a significantly higher number of potentially static features (by several orders of magnitude) when compared to the methods which require manual identification of such features. This allows the former to achieve a higher stabilisation accuracy, but manual feature selection methods have demonstrated lower computational complexity and better robustness in complex field conditions. While this paper does not intend to identify the optimal stabilisation tool for UAS-based velocimetry purposes, it does aim to shed a light on implementational details which can help engineers and researchers choose the tool suitable for their needs and specific field conditions. Additionally, the RMSD comparison metric presented in this paper can also be used in order to measure the velocity estimation uncertainty induced by UAS motion.

Soil erosion as a major environmental challenge, plays a central role in land degradation. Accurate erosion rates assessment and information on erosion, deposition and on occurring processes are important to support soil protection and recovery strategies.Due to the complexity, variability and discontinuity of erosional processes, model approaches to predict soil erosion are non-transferable to different temporal and spatial scales. Present process-based models are only valid for the particular observation scale which they were parameterized and validated for. In reality processes occur (e.g. spontaneous rill initiation) which are only to some extent reproducible, resulting in an incomplete process description. While model parameterization in the past was limited by the availability and resolution of data, constant development of data assessment technologies help overcome these confines. Time and cost in collecting data decreases, computing power is constantly expended and both the temporal and spatial resolution offer new possibilities on new scales.Addressing the issue ‘data overhaul models’ we present a unique experimental setup, including flow velocity, erosion and deposition measurements at nested temporal and spatial scales, acquired using high resolution photogrammetric data (RGB and thermal) and structure from motion techniques. At the micro plot scale (3 m2), we perform rainfall simulations, monitored with up to eleven cameras. Using time lapse intervals of 10-20 seconds processes of pool formation and aggregate breakdown are observed. At the hillslope scale (60 m2), we installed a permanent setup – three rigs at three slope positions at four meter height, each equipped with five synchronized RGB cameras, a RGB video-camera and a low cost thermal camera. To capture changes in soil surface during rainfall events, time lapse images are triggered by a low-cost rain gauge. Soil surface changes at the small catchment scale (4 ha) are measured by taking UAV-images before and after rainfall events. These observations are used as parameterization, calibration and validation for modelled soil surface changes and erosion fluxes, using Erosion3D and FullSWOF.The continuous development and improvement of soil erosion assessment techniques leads to spatially and temporally highly resolved information on different scales. Eventually the adjustment of the erosion models can enable a cross-scale description and validation of scale-dependent processes, offering new perspectives on both interconnectivity of sediment transport and the relationship between event frequency and magnitude.

Unmanned Aerial Vehicles (UAV) have become a commonly used measurement tool in geomorphology due to their affordable cost, flexibility, and ease of use. They are regularly used in fluvial geomorphology, among other fields, because the high spatiotemporal resolution of UAV data makes it possible to assess the continuum rather than relying on single samples.In this study, UAV data are used to hydro-morphologically describe three different river reaches of lengths between 150 and 1000 m. Specifically, the surface flow velocity and bathymetry of the rivers were reconstructed. The flow velocities were calculated using the Particle Tracking Velocimetry (PTV) method applied to UAV video sequences. In addition, UAV-based imagery was acquired to perform 3D reconstruction above and below the water surface using SfM (Structure from Motion) photogrammetry, taking into account refraction effects as well as frame processing to increase the visibility of underwater features. Reference data for flow velocities were generated at selected positions using current meters as well as ADCP (Acoustic Doppler Current Profiler) readings. The image-based calculated bathymetry was compared with RTK-GNSS sampling depth measurements and also ADCP data.The developed workflow enables rapid and regular measurement of hydrological and morphological data of river channels. This ultimately enables multi-temporal assessment and significantly improves hydro-morphodynamic modelling, in particular their calibration.

While the availability and affordability of unmanned aerial systems (UASs) has led to the rapid development of remote sensing applications in hydrology and hydrometry, uncertainties related to such measurements must be quantified and mitigated. The physical instability of the UAS platform inevitably induces motion in the acquired videos and can have a significant impact on the accuracy of camera-based measurements, such as velocimetry. A common practice in data preprocessing is compensation of platform-induced motion by means of digital image stabilisation (DIS) methods, which use the visual information from the captured videos – in the form of static features – to first estimate and then compensate for such motion. Most existing stabilisation approaches rely either on customised tools developed in-house, based on different algorithms, or on general purpose commercial software. Intercomparison of different stabilisation tools for UAS remote sensing purposes that could serve as a basis for selecting a particular tool in given conditions has not been found in the literature. In this paper, we have attempted to summarise and describe several freely available DIS tools applicable to UAS velocimetry. A total of seven tools – six aimed specifically at velocimetry and one general purpose software – were investigated in terms of their (1) stabilisation accuracy in various conditions, (2) robustness, (3) computational complexity, and (4) user experience, using three case study videos with different flight and ground conditions. In an attempt to adequately quantify the accuracy of the stabilisation using different tools, we have also presented a comparison metric based on root mean squared differences (RMSDs) of inter-frame pixel intensities for selected static features. The most apparent differences between the investigated tools have been found with regards to the method for identifying static features in videos, i.e. manual selection of features or automatic. State-of-the-art methods which rely on automatic selection of features require fewer user-provided parameters and are able to select a significantly higher number of potentially static features (by several orders of magnitude) when compared to the methods which require manual identification of such features. This allows the former to achieve a higher stabilisation accuracy, but manual feature selection methods have demonstrated lower computational complexity and better robustness in complex field conditions. While this paper does not intend to identify the optimal stabilisation tool for UAS-based velocimetry purposes, it does aim to shed light on details of implementation, which can help engineers and researchers choose the tool suitable for their needs and specific field conditions. Additionally, the RMSD comparison metric presented in this paper can be used in order to measure the velocity estimation uncertainty induced by UAS motion.

As key-components of the urban-drainage system, storm-drains and manholes are essential to the hydrological modeling of urban basins. Accurately mapping of these objects can help to improve the storm-drain systems for the prevention and mitigation of urban floods. Novel Deep Learning (DL) methods have been proposed to aid the mapping of these urban features. The main aim of this paper is to evaluate the state-of-the-art object detection method RetinaNet to identify storm-drain and manhole in urban areas in street-level RGB images. The experimental assessment was performed using 297 mobile mapping images captured in 2019 in the streets in six regions in Campo Grande city, located in Mato Grosso do Sul state, Brazil. Two configurations of training, validation, and test images were considered. ResNet-50 and ResNet-101 were adopted in the experimental assessment as the two distinct feature extractor networks (i.e., backbones) for the RetinaNet method. The results were compared with the Faster R-CNN method. The results showed a higher detection accuracy when using RetinaNet with ResNet-50. In conclusion, the assessed DL method is adequate to detect storm-drain and manhole from mobile mapping RGB images, outperforming the Faster R-CNN method. The labeled dataset used in this study is available for future research.

We introduce a Python based software tool to measure surface flow velocities and to estimate discharge eventually. Minimum needed input are image sequences, some camera parameters and object space information to scale the image measurements. Reference information can be provided either indirectly via ground control point measurements or directly providing camera pose parameters. To improve the reliability and density of velocity measurements the area of interest has to be masked for image velocimetry. This can either be performed with a binary mask file or considering a 3D point cloud, for instance retrieved with Structure from Motion (SfM) photogrammetry, describing the region of interest. The tracking task can be done with particle image velocimetry (PIV) considering small interrogation regions or using particle tracking velocimetry (PTV) and thus detecting and tracking features at the water surface. To improve the robustness of the tracking results, filtering can be applied that implements statistical information about the flow direction, flow steadiness and average velocities.The FlowVeloTool has been tested with two different datasets; one at a gauging station and one at a natural river reach. Thereby, UAV and terrestrial data were acquired and processed. Velocities can be estimated with an accuracy of 0.01 m/s. If information about the river topography and bathymetry are available, as in our demonstration, discharge can be estimated with an error ranging from 5 to 31 % (Eltner et al. 2019). Besides these results we demonstrate further developments of the FlowVeloTool regarding filtering of tracking results, discharge estimation, and processing of time series. Furthermore, we illustrate that thermal data can be used, as well, with our tool to retrieve river surface velocities. Eltner, A., Sardemann, H., and Grundmann, J.: Flow velocity and discharge measurement in rivers using terrestrial and UAV imagery, Hydrol. Earth Syst. Sci. Discuss., https://doi.org/10.5194/hess-2019-289, 2019.

Soil erosion is one of the most prominent environmental problems of major interest to a vast field of research. Due to the complexity, variability and discontinuity of erosional processes, erosion model approaches are non-transferable to different spatial and temporal scales.The objective of our project is the across-scale modelling of soil erosion, using photogrammetric measurements and optimization methods as well as physical based model approaches. Present process-based models are only valid for the observation scale they are parametrized and validated for. In the observed reality phenomena therefore occur, which are not or only to some extent reproducible by complex model concepts (e.g. development of rills or concentrated runoff within driving lanes). We present the synergetic combination of a physically described model with highly redundant observations from photogrammetric data processing. This enables both the validation of the erosion model EROSION-3D as well as the optimization of its parameters and potentially advancement of the mathematical process description. The photogrammetric observations (RGB and thermal) offer the opportunity of a temporal and spatial differentiated process assessment (splash, sheet and rill erosion, as well as deposition and transport). To this purpose, the acquisition of the respective operating processes and contributing factors, will be nested defined at three different scales (micro plot, single slope and catchment scale) on two sites (loess soil and residual soil).Flexible cross-scale applicable photogrammetric methods, considering 3D reconstruction and flow measurement, combined with physical-based methods of soil erosion modelling shall enable a better and reliable understanding of soil erosion processes on various spatial and temporal observation scales. Consequently, the implementation of the adjusted model is aimed for to enable a cross-scale description and validation of scale-dependent processes (e.g. discrete consideration of thin sheet flow and rill genesis) to offer new perspectives on both interconnectivity of sediment transport and relationship between event frequency and magnitude.

The Sajó River in Hungary is a medium-sized sand-bed river along which intensive meander development and bank erosion occur. The process threatens agricultural lands and populated areas extensively. Therefore, preventive river management is needed.Main geomorphological features, processes and in-channel flow conditions have to be studied in detail in order to reveal main driving factors. Datasets with high spatio-temporal resolution are necessary to identify relevant parameters. However, so far data density at this river is sparse and gauging stations are distributed poorly.The aim of this study is the improvement of data availability to measure and model hydromorphodynamics of single reaches of the Sajó River. Therefore, multi-temporal field campaigns along selected sub-reaches are conducted with Unmanned Aerial Vehicles (UAV) and Unmanned Water Vehicles (UWV) to survey the topography, the river bed and flow conditions. The channel bathymetry is measured by a single-beam echo sounder mounted on a self-designed remotely controlled boat. The boat also integrates a Mobile Laser Scanner (MLS) to measure the river banks. Furthermore, a panorama camera system is installed to improve the pose estimation of the UWV functioning as a calibrated multi-sensor platform. UAV surveys were performed, using RGB and Thermal Infrared image sequences, to apply image velocimetry algorithms to characterize the river flow at selected cross-sections. ADCP measurements and Terrestrial Laser Scans (TLS) are used for accuracy assessment of the novel datasets.Eventually, data captured over a 2-years period will be implemented into hydrodynamic modeling of the studied sub-reaches to better understand seasonal variations in channel morphodynamics.----The project has been founded by the DAAD (57448822) and (Tempus Public Foundation & DAAD 307670). The research is also influenced by the HARMONIOUS COST Action (CA16219).