PROBABILISTIC MULTIVARIATE FORECASTING OF HYDROLOGICAL VARIABLES

摘要

Rainfall and streamflow forecasts are needed in planningrnwater resource systems. This paper presents a method ofrnmultivariate forecasting with the ability of modelingrnstreamflow and rainfall of a basin mutually in arnprobabilistic manner. The proposed model benefits fromrngeostatistical analysis in virtual fields to characterize thernstochastic characteristics of forecast variables byrnproducing conditional distribution of the predicted valuesrnfor different hydro-climatic conditions. Semivariogramrnand crossvariogram functions can show the structure ofrncorrelation between dependent and independent variables.rnThe distance parameter in those functions is known asrndistance between predictors the proposed method of thisrnstudy has shown great ability in modeling and forecastingrnnonlinear hydrologic events in a real case study. Thernmodel was applied to forecast seasonal rainfall andrnstreamflow in the Zayandeh-rud Basin, in Iran. Thernproposed method results are compared with k-nearestrnneighbor (K-NN) and artificial neural networks (ANNs)rnmodels. The results show acceptable advantages ofrnproposed model in forecasting of hydro-climatic predictedrnvariables.