Abstract: The middle Yangtze River, Poyang Lake, and its five tributaries form a complex "river-lake" system, and understanding the time-series variation characteristics of the lake outflow and its driving factors is key to regulation and improvement of the river-lake relationship. This paper first selects 12 potential drivers of the outflow and illustrates the multi-collinearity among these drivers by combining the correlation coefficient evaluation criterion method and principal component analysis. Then, we construct a mathematical model of the lake outflow based on the random forest regression method, and use it to quantify the relative importance and effects of different potential drivers on the outflow. The results show that 1) no multi-collinearity has been observed among six driving factors－outflows from the Ganjiang, Fuhe, Raohe and Xiushui tributaries, lake stage at Xingzi, and outflow from the Three Gorges reservoir, which can fully characterize the multiple driving effects of the tributaries, lake water stage, and Yangtze mainstream. 2) We reveal that the top three effects come from lake stage, Yangtze mainstream discharge, and Ganjiang River outflow; the influence of the main driving factors is complicated and nonlinear. And relationships among all these six factors are also complicated and their influences are different. This study would deepen our understanding of the relationship underlying this river-lake system and be useful for river-lake regulation.

Key words: Poyang Lake, outflow, driving factors, multi-collinearity, random forest regression