Mobile phone data are shedding a new light on population dynamics. Yet, spatial resolution of cellular-based statistics is scarcely documented, highly dependent on the network topology and very heterogeneous across territories. In this paper, we propose a general framework to evaluate the location estimation precision of cellular network events. With a quadtree algorithm, we build an adaptive spatial where the smallest the tiles, the highest the accuracy embedding precision within the dissemination grid. Our proposals are directly tested on producing a new present population statistics for metropolitan France using signalling data, network data and a highly precise residency location data.