Spatially Representative Online Big Data Sampling for Smart Cities

The diversity of sensing options that IoT offers imposed requirements to evolve stream processing engines so to cope with highly heterogeneous and fast-pace data streams challenging their computing capacities. Location intelligence applications aim at exploiting those geo-referenced data in generating visualizations and dashboards that provide deep insights for assisting decision making in smart cities and urban planning. As data arriving are mostly geo-referenced and the rate is fluctuating in pace and skewness, computations upon streams should depend on approximation by applying methods such as sampling. Representativeness in sampling designs remains the pivotal concern in the literature. In spatial data streams contexts, it loosely means selecting proportional counts of spatial tuples from each group of tuples that belong to the same real geometry (i.e., geographically residing in the same proximity) within each streaming time window. This is challenging in streaming settings because spatial data is parametrized, loosing hence it is real geometries, which requires costly geometric operations to project them back to maps. To close this void, we have designed SpatialSPE in a previous work and incorporated an efficient fine-grained spatial online sampling method (SAOS) transparently within its layers. In this paper, we extend SAOS (the novel method is termed ex-SAOS) by new features that allow efficient online spatial sampling on a coarser level, which is a requirement in smart city scenarios. Our results show that ex-SAOS is efficient and effectively extends SAOS for more general smart city and urban computing scenarios.