Der er observeret en intensivering i antallet af regnbetingede oversvømmelser forårsaget af kraftig nedbør, hvilket har store økonomiske konsekvenser for såvel berørte borgere som kommuner og forsikringsselskaber. Ved at varsle imod kommende oversvømmelser, får borgere og beredskab muligheden for at foretage afværgende foranstaltninger, inden uheldet er ude. For at være i stand til at varsle en by imod en kommende regnbetinget oversvømmelse, kræver det en model, der beregner oversvømmelsesudbredelse og vandniveau i realtid.Kernen i varslingssystemet er realtidsmodellen, som udarbejdes med baggrund i surrogat modellering, hvor der opbygges et katalog af responser via simulering af historisk observeret nedbør, igennem en MIKE FLOOD model. Responserne i kataloget afslører, hvorledes ovsersvømmelsesgivende regnhændelser er korreleret til oversvømmelsesudbredelse samt vandniveauet over terræn. Denne viden anvendes i realtidsmodelleringen, som bygger på metoderne; logistiske funktioner og neurale netværk.Derudover undersøges potentialet for at anvende kunstigt konstrueret nedbør til udarbejdelsen af en realtidsmodel.Det kan konkluderes, at det er muligt at anvende neurale netværk til modellering af oversvømmelsesudbredelse samt vandniveau over terræn med høj præcision, i realtid. Realtidsmodellen med baggrund i logistiske funktioner har ligeledes vist sig anvendelig til at modellere oversvømmelsesudbredelse. Dette skyldes, at middelintensiteten over forskellige regnvarigheder af en regnhændelse, i høj grad er korreleret til oversvømmelsesudbredelsen.\\Desuden konkluderes det, at CDS-regn kan anvendes til opbygning af en realtidsmodel, der med tilfredsstillende præcision modellerer responsen i systemet ud fra en regnhændelse.Intensification of the number of rainfall floods caused by heavy rainfall has been observed, which has major economic consequences for both affected citizens and municipalities. By notifying citizens and preparedness before a servere flooding, they have the opportunity to take precautionary measures, and thereby reducing the consequences. It has been chosen to investigate the potential of a warning system against rain-flooding using a real-time model that calculates the response of the system in real time, without compromising the complexity of the model. It has been chosen to construct the real-time model based on surrogate modeling, whereby a catalog of responses is produced. The catalog of responses is produced by simulating the MIKE FLOOD model with historical rain series as input. The catalog of responses is analyzed by using two methods; logistic functions and neural network and subsequently examine the methods in relation to describing the response in the system in real time. In addition, the potential for using artificially constructed precipitation for the production of the real-time model is examined. It can be concluded that a real-time model can be produced which calculates whether future rainfall events result in flooding in real time, by both methods, where the neural network calculates the response with a greater certainty. In addition, it can be concluded that artificially constructed precipitation in the form of CDS-rain can be used to build a real-time model that calculates the response with satisfying certainty.