Abstract: In recent years, water level in Poyang Lake has changed significantly, threatening the ecological security of wetlands. First, this paper uses the Mann-Kendall method to analyze water level changes in this lake in the period of 1990 to 2016.Then, we construct a BP neural network model for simulation of the lake water level, and quantify through scenario analysis the temporal and spatial characteristics of the contribution made to the lake level change by the flow changes in the Yangtze mainstream, lake inflows from the Poyang’s sub-basins, and the topographic change. In time evolution, flow variations in the Yangtze mainstream is the main driving factor of the decreases in the average monthly water level at all the hydrological stations from July through September, and topography is the main driving factor of the decreases in the Poyang’s water level in dry season. Spatially, in the period of July to November, the pull down effects of the Yangtze mainstream flow on the monthly average water levels at different stations rank in a decreasing order as Xingzi station, Duchang station, Tangyin station, and Kangshan station; the pull down effects of topography rank as Duchang station, Xingzi station, Tangyin station, and Kangshan station. Changes in topography are the major cause of the maximum water level variations that occur at the Duchang station.

Key words: water level change, BP neural network model, contribution rate, spatiotemporal characteristics, Poyang Lake