The most commonly used tools in applied hydrology are essentially data-driven, which poses a problem for several drainage basins in the world that are ungauged or poorly gauged. These basins are characterized by the absence of discharge measurements or by incomplete data series, with several gaps and a lack of information content for hydrological applications. In flood forecasting, one of the main challenges when dealing with ungauged basins concerns the need of data for, on one hand, the calibration of model parameters and, on the other hand, the updating of initial conditions at the onset of the forecasts. Information has to be retrieved from other hydrologically similar gauged catchments, which usually correspond to neighbouring catchments within a homogeneous region, and transposed to the ungauged site. The uncertainty associated with such procedure is considerable and strategies to reduce the predictive uncertainty need to be put in place to produce reliable forecasts that can be useful to decision-makers. The objective of this study is to investigate different strategies to reduce the predictive uncertainty in ungauged catchments, with the help of the geographical neighbourhood. Different scenarios are considered, combining both the transfer of model parameters from neighbour catchments ("donors") for rainfall-runoff simulation, and the transfer of observed discharges for updating the initial conditions at the time of forecast. In search of an optimal use of the information coming from neighbours, the assumption that using different donors to infer model parameters and to derive initial conditions for updating will improve forecast performance is also tested. A leave-one-out cross validation approach is applied for 211 catchments in France, considering each catchment as ungauged at a time. Daily forecasts are issued by a lumped soil-moisture accounting type rainfall-runoff model, using observed rainfalls as input, as well as ensemble precipitation forecasts from the French meteorological centre (Météo-France) available for a 4-year period. Flow forecasts are evaluated using mathematical skill scores and the performances of the different tests are compared. The results show that the main contribution of neighbouring gauged catchments is in the transfer of model parameters. The transfer of specific discharges for updating requires a more meticulous search for donors. The added value of having at least local data available at the time of the forecast to perform the real-time updating is illustrated. Key issues on alternative strategies to apply when no local updating is possible are discussed.