Spatial interpolation of mobile positioning data for population statistics

Mobile positioning is recognised to be one of the most promising new sources of data for the production of fast and cost-effective statistics regarding population and mobility. Considerable interest has been shown by government institutions in their search for a way to use mobile positioning data to produce official statistics, although to date there are only few examples of successful projects. Apart from data access and sampling, the main challenges relate to the spatial interpolation of mobile positioning data and extrapolation of recorded data to the level of the entire population. This area of work has to date received relatively little attention in the academic discussion. In the current study, we compare five different methods of spatial interpolation of mobile positioning data. The best methods of describing population distribution and size in comparison with Census data are the adaptive Morton grid and the Random forest model (R-2 > 0.9), while the more widely used point-in-polygon and areal-weighted methods produce results that are far less satisfactory (R-2 = 0.42; R-2 = 0.35). Careful selection of spatial interpolation methods is therefore of the utmost importance for producing reliable population statistics from mobile positioning data.