Probabilistic Multivariate Forecasting of Hydrological Variables

Forough Allahyaripour, Mohammad Azmi, Shahab Araghinejad, and Reza Aasemi

Keywords

Multivariate Forecasting, Geostatistics, K-NN, ANN

Abstract

Rainfall and streamflow forecasts are needed in planning water resource systems. This paper presents a method of multivariate forecasting with the ability of modeling streamflow and rainfall of a basin mutually in a probabilistic manner. The proposed model benefits from geostatistical analysis in virtual fields to characterize the stochastic characteristics of forecast variables by producing conditional distribution of the predicted values for different hydro-climatic conditions. Semivariogram and crossvariogram functions can show the structure of correlation between dependent and independent variables. The distance parameter in those functions is known as distance between predictors the proposed method of this study has shown great ability in modeling and forecasting nonlinear hydrologic events in a real case study. The model was applied to forecast seasonal rainfall and streamflow in the Zayandeh-rud Basin, in Iran. The proposed method results are compared with k-nearest neighbor (K-NN) and artificial neural networks (ANNs) models. The results show acceptable advantages of proposed model in forecasting of hydro-climatic predicted variables.

Important Links:

Go Back