We consider a decision maker who is responsible for issuing flood warnings for the population. The population is uncertain about the credibility of the warnings and adjusts its beliefs following false alerts or missed events. We show that low credibility leads the decision maker to issue warnings for lower probabilities of flooding. In practice, those probabilities are provided by hydrological forecasts. We therefore use our model to compare welfare under alternative real-world hydrological forecasts. We find that when forecasts include non-realistic extreme scenarios, the economy may remain stuck in a state characterized by many false alerts and poor credibility.