The recent rapid development of data collection tools has led to an explosion in data sources. As centralized storage of big data sources becomes increasingly difficult, decentralized storage solutions are gaining more and more attention. Many traditional data mining techniques have become outdated and are no longer suitable for solving large, multidimensional data problems. This paper presents a collaborative possibilistic fuzzy c-means clustering algorithm (CPFCM) on distributed computer systems using graphics processing units (GPUs) for landcover classification. By combining collaborative clustering and the possibility of fuzzy c-means clustering, it is possible to create a collaborative clustering model on a distributed computer system. This proposal aims to solve the clustering problem with decentralized data. Furthermore, to improve the computational efficiency of the remote sensing image landcover classification problem, the proposed algorithm is also designed to run on local GPUs of the computers in the distributed system. Experiments on two datasets downloaded from the UCI Machine Learning Repository and satellite image data show that the proposed method gives significantly better results than others. This result shows that the collaborative possibilistic fuzzy c-means clustering algorithm can help solve the remote and distributed clustering problem and improve computational efficiency using local GPUs.