Since communication is often constrainted and the computational resources are limited in wireless sensor networks, it is more important for local sensors to send in compressed data. In this paper, a nonlinear coordinate compression rule is constructed based on arctangent function. Beneficial from the nonlinear feature of arctangent function, near the centroid the compression ratio is low and apart from the centroid the compression ratio becomes higher and higher. The proposed algorithm is more suitable for the signal that has a useful high frequency near centroid. And the proposed algorithm has the following features: the sampling interval is not even; the compression can be done before sampling, which is similar to a compression sensing; it has low computation amount, is simple and easy to implement in a real system.