Abstract: The Heihe River is a great inland river in arid zone of the northwestern China. Its main stream, with a length of 821 km and a catchment area of 13 000 km², rises in the Qilian Mountain and flows through the Hexi Corridor region of Gansu Province, and enters the western part of Inner Mongolia Plateau. The surface water resource of the river come completely from the mountain region. The Yingluoxia hydrometric station, with a elevation of 1 700 m a.s.l. and a catchment area of 10 009 km², is the controlling station for measuring the mountain outlet runoff of the Heihe River. The Zhangye Basin in the middle course of the Heihe River is an important base for producing grain, where there are extremely developed irrigation agriculture. The western part of Inner Mongolia in the lower reaches of the river is a desert oasis, where the environment is degenerating severely. At present, the over exploitation and utilization of water resources in the Zhangye Basin endanger the existence of the desert oasis. The development of society and economy in the Heihe River basin like other arid regions are seriously limited by the quantity and quality of the water resources. Therefore, it's very important to study the runoff forecasting methods. Kalman filter method is a estimation method on mathematical statistics. The optimal estimator of physical parameters can be obtained through treating a series of observational data with errors to the reality, and the method enable the errors to become minimum. Based on Kalman filter method and with mean precipitation, air temperature, the annual mean values of sunspot quantity and the main 6 year cycle of annual mean discharge through the Yingluoxia Hydrometric Station as the controlling parameters, the forecasting for the annual runoff at the mountain outlet of the Heihe River was conducted.

Key words: Kalman Filter Method, Multivariate regression model, Runoff forecast Recurrence parameters