Data-driven site characterization (DDSC) aids geotechnical engineering by inferring and mapping soil parameters of the subsurface domain. In practice, the limited availability of site investigation data may hinder the performance of traditional machine learning methods and implies significant uncertainty in the predictions, which is typically not quantified. In this study, a framework for Bayesian site characterization (BaySiC) is applied on a benchmark example. Adopting Bayesian statistics enables the framework to deal with small training data sets and allows for coherent quantification of uncertainty, which is valuable to engineering practice for assessing the reliability and the determining characteristic values. BaySiC uses site investigation data to infer statistical estimators of cone penetration test (CPT) parameters and their dependence, as well as to learn spatial correlations. Consecutively, it generates a three-dimensional (3D) map of the subsurface by predicting the CPT parameter values and classifying the material type over the soil domain. For the benchmark example, the study formulated two models within the BaySiC framework and demonstrated their conduct in several cases of varying complexity. Eventually, the performance of the models was evaluated and compared in both deterministic and probabilistic terms. One of the models proved highly effective in predicting the material type at new locations of the subsurface domain, whereas the other provided accurate mapping of the CPT parameters even in complex stratigraphic cases. Also, investigating and comparing the results of the models led to insights regarding the effectiveness of their formulation. Moreover, the paper used hypothesis testing as a means of assessing the predictive power of the model independently from the validation data set. Stemming from the benchmark example, the paper draws conclusions that are meaningful to geotechnical engineering and decision-making.@en

Streamflow predictions remain a challenge for poorly gauged and ungauged catchments. Recent research has shown that deep learning methods based on Long Short-Term Memory (LSTM) cells outperform process-based hydrological models for rainfall-runoff modeling, opening new possibilities for prediction in ungauged basins (PUB). These studies usually feature local datasets for model development, while predictions in ungauged basins at a global scale require training on global datasets. In this study, we develop LSTM models for over 500 catchments from the CAMELS-US data base using global ERA5 meteorological forcing and global catchment characteristics retrieved with the HydroMT tool. Comparison against an LSTM trained with local datasets shows that, while the latter generally yields superior performances due to the higher spatial resolution meteorological forcing (overall median daily NSE 0.54 vs. 0.71), training with ERA5 results in higher NSE in most catchments of Western and North-Western US (median daily NSE of 0.83 vs. 0.78). No significant changes in performance occur when substituting local with global data sources for deriving the catchment characteristics. These results encourage further research to develop LSTM models for worldwide predictions of streamflow in ungauged basins using available global datasets. Promising directions include training the models with streamflow data from different regions of the world and with higher quality meteorological forcing.@en

Dike infrastructure is of vital importance for the safety against flooding. The standard methodologies for the assessment of dike safety for macrostability are based on limit equilibrium methods, which result in a safety factor against shear failure. The more advanced alternative consists of using finite element method to compute the safety factor against shear failure. However, these approaches do not take into account the capacity of the dike to retain water, but are only concerned with the mechanical equilibrium of the dike's initial composition. With the recent advancements in the modelling of large deformations within geotechnical engineering, e.g. by means of the material point method, the post failure behaviour of the dike can be predicted. The material point method is a mesh-free method that has been developed to address the problem of large deformation on a continuum level. The material point method offers the possibility to redefine the concept of factor of safety against shear failure. The initial shear failure of a dike does not necessarily lead to the loss of the dike's capability to retain water. In reality, after the initial shear failure the mass of soil will move and reach a new equilibrium position. This paper, after a brief description of the material point method, presents the analysis of a progressive dike failure, where the post failure behaviour is examined and a proposal is made to redefine the concept of factor of safety.@en

The Dutch design codes for the dikes with retaining walls rely on Finite Element Analysis (FEM) in combination with partial safety factors. However, this can lead to conservative designs. For this reason, in this study, a reliability analysis is carried out with FEM calculations aiming to demonstrate the feasibility of reliability analysis for a dike with an anchored sheet pile wall modelled in the 2D FEM, Plaxis. Sensitivity and reliability analyses were carried out and enabled by coupling the uncertainty package, OpenTURNS and Plaxis. The most relevant (ultimate) limit states concern the anchor, the sheet pile wall, the soil body failure (global instability) and finally the system. The case was used to investigate the applicability of the First Order Reliability Method (FORM) and Directional Sampling (DS) to analysing these limit states. The final goal is to estimate the probability of failure and identify the most important soil properties that affect the behaviour of each component and the system as a whole. The results of this research can be used to assess and optimize the current design procedure for dikes with retaining walls.@en