Remote sensing images play an important role in the development of species distribution model. However, low spatial resolution ecological niche derived from remote sensing data and lack of fine-scale presence data requires alternative approaches. With the application of various data acquisition devices, a large number of animal movement data can be considered as the presence data. In this study, we use DBSCAN method to cluster the movement data and each cluster represents a stopover. Then, we split the remote sensing image into 16×16 patches and divide them into positive and negative samples on the basis of clustering result. In addition, a multi-convolutional neural network model is proposed for the training and prediction of the potential distribution of wild geese surrounding Qinghai Lake. We evaluate the proposed system using a real GPS dataset collected on 29 birds over three years. The experiments show that the proposed method outperforms the GLCM method in terms of overall accuracy, F1 score, and AUC. The proposed method also can obtain a better result in a potential distribution prediction experiment.