Development of the hydrological rainfall-runoff model based on artificial neural network in small catchments

U nekim je situacijama štetne pojave, koje su uzrokovane djelovanjem vode, moguće umanjiti ili spriječiti posljedice istih uz pomoć sustava za rano uzbunjivanje pravovremenom obznanom obavijesti o mogućnosti nastanka štetne pojave. Motivacija za izradu ove disertacije temelji se na istraživanju mogućnosti predviđanja štetnih pojava uzrokovanih vodom na malim slivovima u cilju implementacije sustava za rano uzbunjivanje. Istraživanja unutar rada obuhvaćaju uspostavu kontinuiranog mjerenja meteoroloških i hidroloških podataka na istražnom području sliva Slani potok (Vinodolska dolina) koji je povijesno prepoznato hazardno područje, primjenu umjetnih neuronskih mreža pri razvoju hidrološkog modela predviđanja otjecanja, određivanje načina validacije i evalvacije modela te razvoj metodologije implementacije hidrološkog modela predviđanja otjecanja na malim slivovima. Shodno provedenom istraživanju, a u cilju dokazivanja postavljenih hipoteza razvijen je hidrološki model predviđanja otjecanja s malih slivova temeljen na umjetnoj neuronskoj mreži. Prikupljeni podaci korišteni su za treniranje, validaciju te evalvaciju mogućnosti predviđanja hidrološkog modela otjecanja. Model je potom validiran i evalviran vizualnim i numeričkim mjerama kvalitete prilikom čega su utvrđene dostatne mogućnosti predviđanja modela za potrebe implementacije sustava ranog uzbunjivanja. Temeljem razvijenog modela utvrđena je detaljna metodologija za implementaciju hidrološkog modela otjecanja na malim slivovima temeljenog na umjetnoj neuronskoj mreži.Occasionally, consequences caused by water induced events can somewhat be reduced or even prevented with the help of an early warning system whose aim is timely notification of local population on potentially upcoming hazardous event. The motivation for this thesis arises from the need to explore the possibilities to foresee such water caused events on small catchments with an aim to mitigate its consequences by implementing an early warning system. Research and analysis shown in this theses encompasses the establishment of continuous meteorological and hydrological data monitoring, on research area Slani Potok (Vinodol Valley) historically known as potentially hazardous area, the application of the artificial neural network as a means for the development of hydrological rainfall-runoff model, defining the methods for model validation and evaluation, as well as the development of the methodology for the hydrological rainfall-runoff model implementation on small catchments. Upon on this research a hydrological rainfall-runoff model for small catchments was developed based on artificial neural network. Gathered data was used for training, validation and evaluation of model`s accuracy and precision in rainfall-runoff prediction. The model was validated and evaluated using visual and numerical quality measures according to which needed accuracy in model prediction was determined for the implementation of early warning system. Based on this model a detailed methodology for the implementation of rainfall-runoff model on small catchments developed on artificial neural network was established