Abstract: Based on waterfowl birds' survey data, such as Osprey, Black winged kite, Common buzzard, Oriental monty Yao, Red billed blue magpie and other 13 species of waterbirds, we select the 15 districts of it in Maipo-Deep Bay wetland of Hong Kong, January 2009 as a case. Then, collect the high resolution remote sensing data, climate and terrain data of this area at the same period. Finally, use GIS and RS technology to obtain nine kinds of impact factors, what is closely related to the birds. The Genetic Algorithm for Rule-Set Prediction (GARP) niche model is used to simulate the space distribution of a random sampling points of wetland-dependant birds. The result was not ideal for the waterfowl data space is not continuous, and we do not consider the waterfowl non-Presence points in the GARP model. So we introduce the BP neural network model to improve the simulation results of waterfowl. The results showed that the actual distribution of GARP model simulation waterfowl improved by BP algorithm has achieved good effect. AUC value from ROC curve was used to check and measure the training sample points, the AUC value of 0.905, has a great increase compared with the last result of 0.762. The simulation results demonstrate the characteristics of the distribution of wetland-dependent birds, intertidal zone and large areas of the pond area is its main habitat, mangrove, marsh, followed by habitat, due to human activities frequently interfered with the habitat of waterfowl, the waterfowl distribution less or no in deep bay, urban land used and urban green space area. The relationship between water birds and geographical environment was revealed very good.