This paper investigates the performance of 10 Regional Climate Models (RCMs) hindcasts from the Coordinated Regional Climate Downscaling Experiments (CORDEX) over Central Africa, covering the period 1998–2008 and performed over a common model grid spacing 0.44° ( 50 km). Multiple observational data sets are used to evaluate model performances over four targeted subregions. Throughout the work, a measure of observational uncertainty is made and we discuss whether or not the models are found within or outside the range of observational uncertainty. Results indicate that RCMs generally capture rainfall and temperature basic features, though important biases exist and vary for models and seasons. Dry (wet) biases are common features over the Congo basin (northern and southern part of the domain). In terms of precipitation and temperature in both seasonal and annual scale, most RCMs along with their ensemble mean generally fall in the range of observational uncertainty. Furthermore, most RCMs show a good spread of grid points where the added value of RCMs is found although the added value in temperature is not as great as with precipitation. UC‐WRF is among models adding less value on ERAINT and this could explain why whatever the time scale of variability, UC‐WRF outputs are generally out from the observational uncertainty. The multimodel ensemble mean is generally found within observational range when most models are there as well. This highlights the fact that the ensemble mean, built from the equal treatment of RCMs, does not generally outperform individual RCMs realization as it is reported in several previous studies.

Spatially explicit quantification on design storms is essential for flood risk assessment and planning. Due to the limited temporal data availability from weather radar data, design storms are usually estimated on the basis of rainfall records of a few precipitation stations only that have a substantially long time coverage. To achieve a regional picture, these station-based estimates are spatially interpolated, incorporating a large source of uncertainty due to the typical low station density, in particular for short event durations. In this study we present a method to estimate spatially explicit design storms with a return period of up to 100 years on the basis of statistically extended weather radar precipitation estimates, based on the ideas of regional frequency analyses and subsequent bias correction. Associated uncertainties are quantified using an ensemble-sampling approach and event-based bootstrapping. With the resulting dataset, we compile spatially explicit design storms for various return periods and event durations for the federal state of Baden Württemberg, Germany. We compare our findings with two reference datasets based on interpolated station estimates. We find that the transition in the spatial patterns of the design storms from a rather random (short-duration events, 15 min) to a more structured, orographically influenced pattern (long-duration events, 24 h) seems to be much more realistic in the weather-radar-based product. However, the absolute magnitude of the design storms, although bias-corrected, is still generally lower in the weather radar product, which should be addressed in future studies in more detail.

Short-term flood and inundation forecasts, especially for spatially limited, convective heavy rainfall events, are challenging due to the very short lead time of heavy rainfall predictions. They are subject to different uncertainties. Due to the short forecasting period, data-driven machine learning methods have been developed to predict the inundated areas in almost real time. The time-consuming numerical, hydrodynamic simulation of flooding depths and flow velocities is replaced by a surrogate model.We use a neural network as surrogate model, which is trained with a large ensemble of hydrodynamically simulated water levels and flow velocities for a specific catchment using an ensemble of spatial and temporal surface runoff. The surface runoff is calculated by the hydrological model RoGeR at a 5 m resolution and spatial and temporal varying heavy rainfall events as input. The neural networks are able to predict spatially resolved maximum water depths, maximum discharges and maximum flow velocities for an event in the specific catchment area of 25km² in significantly less than one second.The fast prediction time allows to consider uncertainties in the forecast. Uncertainties in flood forecasts typically result mainly from uncertain precipitation input (or forecast), initial conditions (e.g. soil moisture), lack of data of relevant parameters or their limited transferability in space. Although the model error of the hydrological and hydraulic model itself, such as assumptions about geometry and parameters or underlying flow equations, is an important source of uncertainty, we address in this presentation only the uncertainties of input and initial conditions.To generate ensemble calculations that represent the probability distribution of predicted flood height and velocity, we create a large ensemble of input variables by statistically varying the initial soil moisture as well as the location and intensity of the precipitation fields. Since the trained neural network topology is another source of uncertainty, differently trained networks with different network topologies are also considered in the ensembles.Using the ensembles, we can specify prediction intervals for spatially resolved maximum water levels, maximum discharge and flow velocities, which result from the uncertainty of the model input, model parameters and the inaccuracy of the surrogate model. Using the ensemble approach, we discuss the impact of the different sources of uncertainty on the predictions.As an example, it is found that the statistical variation of the meteorological input data has a greater influence on the prediction interval width than the statistical variation of the initial soil moisture. In general, it can be concluded that sufficient variability in the training data needs to be covered to make reasonable uncertainty predictions.