Nearest neighbor method based on a coupled similarity indicator and its application in annual runoff prediction

The nearest neighbor method (NNM) based on hydrological similarity is a data-driven prediction method, the use of which allows for avoiding making certain assumptions about the dependence and probability distribution forms of objects of study. One of the key factors that affect the quality of the predicted results using this method is the selection of the measure of the similarity in eigenvectors. In order to give consideration to the information about "direction" and "quantity" while measuring similarity, this paper tries to couple Cosine distance and Euclidean distance to the technical indicator for similarity after thoroughly analyzing their differences and similarities. The nearest neighbor method based on coupled similarity indictor was used for the annual runoff prediction of Yichang hydrological station and Tangnaihai hydrological station, and the annual runoff data for different hydrological characteristics were used for checking. The results show that the nearest neighbor method based on the coupled indicator can appropriately predict annual runoff. It is an effective, feasible method.