We are conducting research and development on a deep learning methods for discovering and predicting spatial and temporal association patterns in heterogeneous sensing data related to the urban environment and applying them to the short-term prediction and behavioral support of the effects of abnormal weather and air pollution on social life (environmental quality), such as traffic and health. By devising a prediction model inconsideration of the spatial and temporal continuity and locality of the association pattern, both high versatility and prediction accuracy are achieved. We are also working on the development of technology to optimize predictions by linking user-collected data.

This technology has been implemented in the xData platform on NICT’s Integrated Testbed, and is being applied to support early warning and alert issuance for air pollution, smart services for environment-aware mobility and healthcare, etc. We are also engaged in user-participatory development of environmental quality prediction models and smart services. In addition, we are conducting collaborative research both in Japan and overseas to reuse, share, and transfer environmental quality prediction models.