Multipath delay is a significant error in GNSS precise positioning and affected by the station surroundings. This error cannot be effectively eliminated by observation difference techniques and must be carefully calibrated. Although the spatiotemporal correlation feature of multipath plays an important role in static multipath mitigation, this feature lacks research and needs to be thoroughly investigated. Therefore, the local filtering (LF) approach based on least square collocation (LSC) and moving average (MA) methods is proposed and investigated. The space correlation feature is characterized by covariance function or correlation function, and the residuals with small correlations are neglected while only that with relative larger correlations are used in multipath modelling and calibration. The experiments with short baseline MAT1_MATE and KERG_KRGG for both GPS and Galileo multipath calibration show that, the LF approach based on LSC, simple MA and weighted MA methods achieve better performance in multipath mitigation over short and long-time spans compared with conventional grid approach. For example, the GPS double difference (DD) residual variance reduction rates of baseline MAT1_MATE for the calibration of the next day with LF approach based on the three methods reach73.2%, 72.1% and 70.2%, respectively, which are obviously larger than 63.7% of the grid approach. The improvement reaches maximum 16.4% in the case of baseline KERG_KRGG for Galileo and the average value is around 11.5% i.e. the relatively numbers of the two values reach 29.9% and 20.2%, respectively.