Spatial Accuracy Evaluation for Mobile Phone Location Data With Consideration of Geographical Context

In recent years, mobile phone location (MPL) data have been widely used to determine the spatial trajectories of users. Although this massive amount of MPL data can provide insight into human movement, definite conclusions cannot be drawn because of positioning bias: the locations of MPL data are usually not the phone users' actual locations. In recent years, the spatial accuracy of MPL data has been increasingly evaluated. Such efforts have led to many insights regarding the quality and applicability of MPL data. Despite these achievements, to the best of our knowledge, no studies have quantitatively assessed the spatial accuracy of MPL data by considering geographical influencing factors. In this study, we built a linear evaluation model based on geographical weighted regression (GWR) and a nonlinear evaluation model based on a random forest (RF) to quantify the relationship between geographical factors and the positioning bias of MPL data. Nanjing city in China is used as the test case. The results show that both the GWR model and RF model have good stability. However, the RF model's overall prediction performance is much better than that of the GWR model. The RF model can estimate the spatial accuracy of the MPL data within narrow margins of error. The importance ranking of geographical variables shows that the population density, elevation and building density are the three most important factors, while the normalised difference water index (NDWI) and distance to the nearest cell tower (DNCT) are the least important variables. The RF model constructed in this study can be used to evaluate the spatial accuracy of MPL data and simulate the spatial distribution of the positioning bias of the MPL data covering the study area.