Parallel computing for Fast Spatiotemporal Weighted Regression

The Spatiotemporal Weighted Regression (STWR) model is an extension of the Geographically (GWR) for exploring heterogeneity spatiotemporal processes. A key feature STWR that it utilizes data points observed at previous time stages to make better fit and prediction latest stage. Because temporal bandwidths a few other parameters need be optimized in STWR, calibration computationally intensive. In particular, when amount large, becomes heavily time-consuming. For example, with 10,000 10 stages, takes about 2307 s single-core PC process STWR. Both distance weighted matrix are memory intensive, which may easily cause insufficiency as increases. To improve efficiency computing, we developed parallel computing method by employing Message Passing Interface (MPI). cache MPI processing approach was proposed routine. Also, splitting strategy designed address problem insufficiency. We named overall design Fast (F-STWR). experiment, tested F-STWR High-Performance Computing (HPC) environment total number 204,611 observations 19 years. results show can significantly STWR's capability large-scale data.