Formation process of water bloom was complicated, time-varied and uncertain. So far water bloom prediction of urban lake was still difficult. An integrated modeling approachby using interior-outer-set, rough sets reduction algorithm and RBF neural network model was proposed for early-warning of water bloom risk. Interior-outer-set model was employed to define the threshold of chlorophyll a for predictingwater bloom risk, and a method was put forward for calculating the risk probability of water bloom.Rough sets reduction algorithm was used to identify the keydriving factors ofwater bloom. An early-warning model of water bloom risk was developed based on RBF neural network model. Feasibility of themodeling approach was proved though the application in Qingjing Lake. The results indicated thatthe threshold value of chlorophyll a was 70.98μg/L; water bloom risk was divided into five grades based on the risk probability of water bloom; fourwater quality indexes including water temperature, dissolved oxygen, permanganate index and total dissolved solids were identified as the indicators of water bloom. Result of model validation showed that the RBF neural network model's accurate rate exceeded 85%, and could be applied to early-warning of water bloom risk in Qingjing Lake.