Search

Rapid mapping of event landslides is crucial to identify the areas affected by damages as well as for effective disaster response. Traditionally, such maps are generated with visual interpretation of remote sensing imagery (manned/unmanned airborne systems or spaceborne sensors) and/or using pixel-based and object-based methods exploiting data-intensive machine learning algorithms. Recent works have explored the use of convolutional neural networks (CNN), a deep learning algorithm, for mapping landslides from remote sensing data. These methods follow a standard supervised learning workflow that involves training a model using a landslide inventory covering a relatively small area. The trained model is then used to predict landslides in the surrounding regions. Here, we propose a new strategy, i.e., a progressive CNN training relying on combined inventories to build a generalized model that can be applied directly to a new, unexplored area. We first prove the effectiveness of CNNs by training and validating on event landslides inventories in four regions after earthquakes and/or extreme meteorological events. Next, we use the trained CNNs to map landslides triggered by new events spread across different geographic regions. We found that CNNs trained on a combination of inventories have a better generalization performance, with a bias towards high precision and low recall scores. In our tests, the combined training model achieved the highest (Matthews correlation coefficient) MCC score of 0.69 when mapping landslides in new unseen regions. The mapping was done on images from different optical sensors, resampled to a spatial resolution of 6 m, 10 m, and 30 m. Despite a slightly reduced performance, the main advantage of combined training is to overcome the requirement of a local inventory for training a new deep learning model. This implementation can facilitate automated pipelines providing fast response for the generation of landslide maps in the post-disaster phase. In this study, the study areas were selected from seismically active zones with a high hydrological hazard distribution and vegetation coverage. Hence, future works should also include regions from less vegetated geographic locations.

Past exploration missions have revealed that the lunar topography is eroded through mass wasting processes such as rockfalls and other types of landslides, similar to Earth. We have analyzed an archive of more than 2 million high-resolution images using an AI and big data-driven approach and created the first global map of 136.610 lunar rockfall events. Using this map, we show that mass wasting is primarily driven by impacts and impact-induced fracture networks. We further identify a large number of currently unknown rockfall clusters, potentially revealing regions of recent seismic activity. Our observations show that the oldest, pre-Nectarian topography still hosts rockfalls, indicating that its erosion has been active throughout the late Copernican age and likely continues today. Our findings have important implications for the estimation of the Moon's erosional state and other airless bodies as well as for the understanding of the topographic evolution of planetary surfaces in general.

We use multitemporal analyses based on Synthetic Aperture Radar differential interferometry (DInSAR) to study the slope adjacent to the large Punatsangchhu-I hydropower plant, a concrete gravity dam under construction in Bhutan since 2009. Several slope failures affected the site since 2013, probably as a consequence of toe undercutting of a previously unrecognised active landslide. Our results indicate that downslope displacement, likely related to the natural instability, was already visible in 2007 on various sectors of the entire valley flank. Moreover, the area with active displacements impinging on the dam site has continuously increased in size since 2007 and into 2018, even though stabilization measures have been implemented since 2013. Stabilisation measures currently only focus on a small portion of the slope, however, the unstable area is larger than previously evaluated. Highly damaged rock is present across many areas of the entire valley flank, indicating that the volumes involved may be orders of magnitude higher than the area on which stabilisation efforts have been concentrated after the 2013 failure. The results highlight that satellite-based DInSAR could be systematically used to support decision making processes in the different phases of a complex hydropower project, from the feasibility study, to the dam site selection and construction phase.

High-resolution 3D geological models are crucial for underground development projects and corresponding numerical simulations with applications in e.g., tunneling, hydrocarbon exploration, geothermal exploitation and mining. Most geological models are based on sparse geological data sampled pointwise or along lines (e.g., boreholes), leading to oversimplified model geometries. In the framework of a hydraulic stimulation experiment in crystalline rock at the Grimsel Test Site, we collected geological data in 15 boreholes using a variety of methods to characterize a decameter-scale rock volume. The experiment aims to identify and understand relevant thermo-hydro-mechanical-seismic coupled rock mass responses during high-pressure fluid injections. Prior to fluid injections, we characterized the rock mass using geological, hydraulic and geophysical prospecting. The combination of methods allowed for compilation of a deterministic 3D geological analog that includes five shear zones, fracture density information and fracture locations. The model may serve as a decameter-scale analog of crystalline basement rocks, which are often targeted for enhanced geothermal systems. In this contribution, we summarize the geological data and the resulting geological interpretation.

Background: NMR chemical shift prediction plays an important role in various applications in computational biology. Among others, structure determination, structure optimization, and the scoring of docking results can profit from efficient and accurate chemical shift estimation from a three-dimensional model.A variety of NMR chemical shift prediction approaches have been presented in the past, but nearly all of these rely on laborious manual data set preparation and the training itself is not automatized, making retraining the model, e.g., if new data is made available, or testing new models a time-consuming manual chore., Results: In this work, we present the framework NightShift (NMR Shift Inference by General Hybrid Model Training), which enables automated data set generation as well as model training and evaluation of protein NMR chemical shift prediction.In addition to this main result - the NightShift framework itself - we describe the resulting, automatically generated, data set and, as a proof-of-concept, a random forest model called Spinster that was built using the pipeline., Conclusion: By demonstrating that the performance of the automatically generated predictors is at least en par with the state of the art, we conclude that automated data set and predictor generation is well-suited for the design of NMR chemical shift estimators.The framework can be downloaded from https://bitbucket.org/akdehof/nightshift. It requires the open source Biochemical Algorithms Library (BALL), and is available under the conditions of the GNU Lesser General Public License (LGPL). We additionally offer a browser-based user interface to our NightShift instance employing the Galaxy framework via https://ballaxy.bioinf.uni-sb.de/.