关键词: 模糊粗糙集, 碳通量, 属性约简, 神经网络, 预测

Abstract: Finding the main factors of affecting carbon flux is the key of knowing the law of the carbon cycle.As ecological factors that closely related to carbon flux have many uncertainties and lots of redundant information，so that attribute reduction becomes difficulty.A fuzzy rough set theory has advantages of dealing effectively with complex issues and dealing with continuous attributes，and is used in reduction of carbon flux data attributes.Experimental results show that using attribute having been reduced by fuzzy rough set to establish the carbon flux prediction model based on BP network has higher prediction accuracy，and this method can increase in reduction efficiency substantially，while keeping the amount of information effectively.

Key words: fuzzy rough set, carbon flux, attribute reduction, neural networks, prediction