In the traditional geostatistical model, the effect of covariates are usually assumed to be constants. However, this assumption sometimes restricts the real application. In this work, we mainly focus on the spatial varying coefficient model (SVCM), which agrees that the effect of the covariates can vary as the observation locations. Compared with the conventional stationary model, a more general mean structure in the concerned model is considered. To estimate the related model information, the non-parametric approaches with the penalty scheme is developed. The proposed approach is verified through the simulation studies.