Extracting the regional common‐mode component of GPS station position time series from dense continuous network

We develop a spatial filtering method to remove random noise and extract the spatially correlated transients (i.e., common‐mode component (CMC)) that deviate from zero mean over the span of detrended position time series of a continuous Global Positioning System (CGPS) network. The technique utilizes a weighting scheme that incorporates two factors—distances between neighboring sites and their correlations of long‐term residual position time series. We use a grid search algorithm to find the optimal thresholds for deriving the CMC that minimizes the root‐mean‐square (RMS) of the filtered residual position time series. Comparing to the principal component analysis technique, our method achieves better (>13% on average) reduction of residual position scatters for the CGPS stations in western North America, eliminating regional transients of all spatial scales. It also has advantages in data manipulation: less intervention and applicable to a dense network of any spatial extent. Our method can also be used to detect CMC irrespective of its origins (i.e., tectonic or nontectonic), if such signals are of particular interests for further study. By varying the filtering distance range, the long‐range CMC related to atmospheric disturbance can be filtered out, uncovering CMC associated with transient tectonic deformation. A correlation‐based clustering algorithm is adopted to identify stations cluster that share the common regional transient characteristics. Key Points A correlation‐weighted spatial filtering is developed for CGPS data analysis The filter achieves significant common‐mode signal reduction The filter can extract any signals common to a subset of a CGPS network