Here, we have tackled the challenge of estimating water flow in areas without direct measurements, a crucial task in hydrology. We have applied deep learning techniques to a large sample of French catchments with various hydrological regimes. We have also compared our approach with traditional methods. We found that incorporating more data improves the accuracy of our deep learning predictions. Notably, our method outperforms traditional approaches in certain regimes, though not universally.

This article results from a master's research project which was part of a natural hazards programme developed by the French Ministry of Ecological Transition. The objective of this work was to investigate a possible way to improve the operational flash flood warning service by adding rainfall forecasts upstream of the forecasting chain. The results showed that the tested forecast product, which is new and experimental, has a real added value compared to other classical forecast products.

This paper proposes a methodological framework designed for event-based evaluation in the context of an intense flash-flood event. The evaluation adopts the point of view of end users, with a focus on the anticipation of exceedances of discharge thresholds. With a study of rainfall forecasts, a discharge evaluation and a detailed look at the forecast hydrographs, the evaluation framework should help in drawing robust conclusions about the usefulness of new rainfall ensemble forecasts.

Hydrologists have long dreamed of a tool that could adequately predict runoff in catchments. Data-driven long short-term memory (LSTM) models appear very promising to the hydrology community in this respect. Here, we have sought to benefit from traditional practices in hydrology to improve the effectiveness of LSTM models. We discovered that one LSTM parameter has a hydrologic interpretation and that there is a need to increase the data and to tune two parameters, thereby improving predictions.

This contribution presents the first evaluation of Variational Data Assimilation successfully applied over a large sample to the spatially distributed calibration of a newly taylored grid-based parsimonious model structure and corresponding adjoint. High performances are obtained in spatio-temporal validation and at flood time scales, especially for mediterranenan and oceanic catchments. Regional sensitivity analysis revealed the importance of the non conservative and production components.

We compared three hydrological models in a flash flood modelling framework. We first identified the sensitive parameters of each model, then compared their performances in terms of outlet discharge and soil moisture simulation. We found out that resulting from the differences in their complexities/process representation, performance depends on the aspect/measure used. The study then highlights and proposed some future investigations/modifications to improve the models.

In this note, a new method (RAT) is proposed to assess the robustness of hydrological models. The RAT method is particularly interesting because it does not require multiple calibrations (it is therefore applicable to uncalibrated models), and it can be used to determine whether a hydrological model may be safely used for climate change impact studies. Success at the robustness assessment test is a necessary (but not sufficient) condition of model robustness.

Efficient flood mapping methods are needed for large-scale, comprehensive identification of flash flood inundation hazards caused by small upstream rivers. An evaluation of three automated mapping approaches of increasing complexity, i.e., a digital terrain model (DTM) filling and two 1D–2D hydrodynamic approaches, is presented based on three major flash floods in southeastern France. The results illustrate some limits of the DTM filling method and the value of using a 2D hydrodynamic approach.

This study contributes to flash flood prediction using a hydrological model. The model describes the spatial properties of the watersheds with hundreds of unknown parameters. The Gardon d'Anduze watershed is chosen as the study benchmark. A sophisticated numerical algorithm and the downstream discharge measurements make the identification of the model parameters possible. Results provide better model predictions and relevant spatial variability of some parameters inside this watershed.