Alluvial rivers are the most important agents of sediment transport in continental basins, whose fluvial deposits enclose information related to the time when rivers were active. In order to extract the most information from fluvial deposits in the sedimentary record, it is imperative to quantify the natural variability of channel patterns at the global scale, explore what controls may influence their development, and investigate whether channel pattern information is preserved in the alluvial plains in order to develop tools for recognizing them in the sedimentary record. By surveying 361 reaches of modern alluvial rivers with available water discharge data at a global scale, we present a quantitative channel pattern classification based on sinuosity and channel count index applicable to the recognition in the rock record. A continuum of channel patterns ranging from high-sinuosity single channel to lowsinuosity multichannels is documented, along with the proportion of depositional elements in their alluvial plains and their conditions of occurrence. Preserved barforms in the alluvial plains of these rivers are used to infer and quantify paleoflow directions at the channel-belt scale and result in ranges of paleocurrent circular variance that may lead to channel pattern identification in the rock record. Data from this work indicate that the recognition of channel patterns may be used to predict paleogeographic features such as the scale of drainage basin area and discharge, slope, and annual discharge regimes.

Abstract. Data on grain sizes of pebbles in gravel-bed rivers are of key importance for the understanding of river systems. To gather these data efficiently, low-cost UAV (uncrewed aerial vehicle) platforms have been used to collect images along rivers. Several methods to extract pebble size data from such UAV imagery have been proposed. Yet, despite the availability of information on the precision and accuracy of UAV surveys as well as knowledge of errors from image-based grain size measurements, open questions on how uncertainties influence the resulting grain size distributions still persist. Here we present the results of three close-range UAV surveys conducted along Swiss gravel-bed rivers with a consumer-grade UAV. We measure grain sizes on these images by segmenting grains, and we assess the dependency of the results and their uncertainties on the photogrammetric models. We employ a combined bootstrapping and Monte Carlo (MC) modeling approach to model percentile uncertainties while including uncertainty quantities from the photogrammetric model. Our results show that uncertainty in the grain size dataset is controlled by counting statistics, the selected processed image format, and the way the images are segmented. Therefore, our results highlight that grain size data are more precise and accurate, and largely independent of the quality of the photogrammetric model, if the data are extracted from single, undistorted nadir images in opposition to orthophoto mosaics. In addition, they reveal that environmental conditions (e.g., exposure to light), which control the quality of the photogrammetric model, also influence the detection of grains during image segmentation, which can lead to a higher uncertainty in the grain size dataset. Generally, these results indicate that even relatively imprecise and inaccurate UAV imagery can yield acceptable grain size data, under the conditions that the photogrammetric alignment was successful and that suitable image formats were selected (preferentially single, undistorted nadir images).

No abstract

No abstract

Abstract. Data on grain sizes of pebbles in gravel-bed rivers are of key importance for the understanding of river systems. To gather these data efficiently, low-cost UAV (unmanned aerial vehicle) platforms have been used to collect images along rivers. Several methods to extract pebble size data from such UAV imagery have been proposed. Yet, despite the availability of information on the precision and accuracy of UAV surveys, a systematic analysis of the uncertainties that might be introduced into the resulting grain size distributions is still missing. Here we present the results of three close-range UAV surveys conducted along Swiss gravel-bed rivers with a consumer-grade UAV. We measure grain sizes on these images by segmenting grains, and we assess the dependency of the results and their uncertainties on the photogrammetric models. We employ a combined bootstrapping and Monte Carlo (MC) modelling approach to model percentile uncertainties while including uncertainty quantities from the photogrammetric model. Our results show that uncertainty in the grain size dataset is controlled by counting statistics, the selected orthoimage format, and the way the images are segmented. Therefore, our results highlight that grain size data are more precise and accurate, and largely independent on the quality of the photogrammetric model, if the data is extracted from single, undistorted orthoimages. In addition, they reveal that environmental conditions (e.g., exposure to light), which control the quality of the photogrammetric model, also influence the detection of grains during image segmentation, which can lead to a higher uncertainty in the grain size dataset. Generally, these results indicate that even relative imprecise and not accurate UAV imagery can yield acceptable grain size data, under the conditions that the photogrammetric alignment was successful and that suitable image formats were selected (preferentially single orthoimages).

<p class="Correspondence"><span lang="EN-GB">The size of coarse sedimentary particles in fluvial systems is key for quantifying sedimentation and transport conditions in both active and ancient fluvial systems. In particular, the grain size of the bed load in gravel-bed rivers allows inferring information on sediment entrainment or deposition mechanisms, and on the hydraulic conditions controlling them. However, collecting data on such coarse-grained sedimentary particles traditionally involved time-intensive and costly fieldwork, leading to the development of image-based techniques for grain size estimation over the last two decades. Nevertheless, despite much progress and the recent deployment of deep learning methods that were trained on large datasets (i.e., > 100 000 manually annotated grains; Lang et al., 2021; Chen et al., 2022), image-based grain size data is limited to single percentile values, often due to a systematic bias and/or a low accuracy (e.g., Chardon et al., 2020; Mair et al., 2022). Specifically, the core challenge for most existing methods is the accurate segmentation, i.e., the identification and delineation of individual grains, across distinctly different types of data. </span></p> <p class="Correspondence"><span lang="EN-GB">Here we present a new approach designated to improve the segmentation in images, which is based on the capability of transfer learning of deep learning models. Such a strategy allows us to re-train existing models for new tasks that are similar to their original purpose. In particular, we use the python-based and open-source tool cellpose (Stringer et al., 2021), which is a state-of-the-art machine-learning model based on neural networks and designed to segment cells in biomedical images. We retrained such a cellpose model on several image datasets of fluvial gravel. The rationale for our approach is based on an inferred geometric similarity between cell nuclei and rock pebbles. Our re-trained models outperform existing methods designed for the segmentation of fluvial pebbles in all datasets, despite an order of magnitude smaller number of training data than currently used in machine learning models. Furthermore, our results show that models trained on specialized datasets for a specific sediment setting yield significantly better results than models trained on larger and more diverse datasets. Fortunately, the model’s flexibility, accessibility, and ability for easy and fast training (Pachitariu and Stringer, 2022) enable the training of task- or image-type-specific models. To facilitate the segmentation power of such models, we built an open-source software tool, ImageGrains. This tool allows for easy use of the models we trained, or of other custom models, as well as streamlined grain size and shape measurements. This allows for fast and nearly automated measurements of large numbers of coarse sedimentary particles with high precision and across vastly different image settings. </span></p> <p class="Correspondence"><strong><span lang="EN-GB">References</span></strong></p> <p class="Correspondence">Chardon, V., et al., 2022: River Res. Appl., 38, 358–367, https://doi.org/10.1002/rra.3910.</p> <p class="Correspondence">Chen, X., et al., 2022: Earth Surf. Dyn., 10, 349–366, https://doi.org/10.5194/esurf-10-349-2022.</p> <p class="Correspondence">Lang, N., et al. 2021: Hydrol. Earth Syst. Sci., 25, 2567–2597, https://doi.org/10.5194/hess-25-2567-2021.</p> <p class="Correspondence">Mair, D., et al. 2022: Earth Surf. Dyn., 10, 953–973, https://doi.org/10.5194/esurf-10-953-2022.</p> <p class="Correspondence">Pachitariu, M. and Stringer, C. 2022: Nat. Methods, 19, 1634–1641, https://doi.org/10.1038/s41592-022-01663-4.</p> <p class="Correspondence">Stringer, C., et al. 2021: Nat. Methods, 18, 100–106, https://doi.org/10.1038/s41592-020-01018-x.</p>

Abstract. Climate changes have been considered an essential factor controlling the shaping of the recent alluvial landscapes in central Amazonia, with implications for explaining the biogeographic patterns in the region. This landscape is characterized by wide floodplains and various terrace levels at different elevations. A set of older terraces with ages between 50 and >200 ka occupy the higher portions of central Amazonia, whereas multiple terraces next to floodplains occur at lower elevations and display ages of a few thousand years. These lower terraces, referred to as middle–lower terraces, reveal what can be perceived as a stochastic pattern both in space and time. Despite the widespread occurrence of these geomorphic features, no process-oriented analysis has been conducted to explain their formation. Here, we develop a landscape evolution model referred to as SPASE (Sedimentary Processes and Alluvial Systems Evolution) to explicitly account for fluvial erosion and deposition in combination with lateral channel migration to explore the controls on terrace development. The model results show that the higher terraces were deposited under the condition of a higher base level for the basins upstream of the confluence between the Solimões and Negro rivers. The subsequent decrease in the base level initiated a phase of gradual incision, thereby resulting in the current fluvial configuration. The model also predicts that high-frequency climate changes resulted in the construction of middle–lower terraces at various elevations which, however, are all situated at lower elevation than the higher terrace levels. Our model shows that dry-to-wet shifts in climate, in relation to the modern situation, yield a landscape architecture where middle–lower terrace levels are better preserved than wet-to-dry changes in climate, again if the current situation is considered as reference. Finally, our results show that fast and widespread landscape changes possibly occurred in response to high-frequency climate changes in central Amazonia, at least since the Late Pleistocene, with great implications for the distribution and connectivity of different biotic environments in the region. Because of this short timescale of response to external perturbations, we suggest that the streams in central Amazonia possibly also respond in rapid and sensitive ways to human perturbations.

The size of sedimentary particles in gravel-bed rivers allows for inferring information on sediment entrainment or deposition mechanisms and on the hydraulic conditions controlling them. However, collecting data on these coarse-grained sediments is costly and time-consuming, both in the field and remotely from images. Therefore, recent attention has turned to machine learning models to improve such measurements. Despite their success, current methods need large quantities of data and yield results limited to a few percentile values of grain size datasets, often affected by systematic bias and low accuracy. In most cases, the root of these limitations is the challenge of accurately segmenting grains. Here we present a new approach to improve the segmentation of individual grains based on the capacity of transfer learning in convolutional neural networks. Specifically, we re-train a state-of-the-art model for cell segmentation in biomedical images to find and segment coarse-grained particles in images of fluvial sediments. Our results show that our re-trained models outperform existing methods so that the performance in segmentation tasks can be directly transferred to images of fluvial sediments. With our approach, these results are achievable with only 10-20% of the data that have been previously needed for training other machine learning models. Moreover, we find that primarily traits in our data control the segmentation performance, enabling data-driven approaches to improve future segmentation models. Additionally, comparing our automatically measured grains with the results retrieved from various image and field-based surveys confirms that these improvements in segmentation are directly leading to more precise and more accurate grain size data across different image settings. Finally, we release a software package, the trained models, and the used data. The goal is to offer a tool to efficiently segment and measure grains in images of sediments in an automated way, which can be adapted to different settings.