Among the different types of floods, flash floods can be particularly dangerous, notably due to specific challenges forecasters and risk managers face when dealing with this type of event: rapid rising of river water levels, short lead time to activate flood alert and social response, difficulties in forecasting localised and intense precipitation events, as well as the lack of quantitative data to improve the understanding and modelling of flash floods. The setup of flood alert systems for nowcasting usually focuses on capturing in advance a signal of an upcoming extreme event to generate short-range (0-6h ahead) forecasts and produce flash flood guidance and flood alert maps. This is usually based on the extrapolation in time of observed weather radar images or the use of high resolution numerical weather model outputs (or a combination of both). In this study, we discuss the use of ensemble approaches to quantify uncertainties in flash flood nowcasting systems and provide probabilistic information for flood alert at very short lead times. We present a review of existing approaches and recent developments, highlighting the potential ways the HEPEX community can contribute to tackle challenges specifically related to flash flood nowcasting. The presentation is illustrated with recent work carried out in France for the ensemble nowcasting of flash floods at gauged and ungauged catchments.