Predicting and forecasting air quality at a fine spatiotemporal scale is not only essential for studying the impact of air pollutant on health conditions but also critical for making informed decisions. For example, accurately forecasting the air quality at a fine spatial resolution in a city can help school officials make their advanced prevention plan based on their locations. (School on the east side of the town might need to cancel the afternoon physical education classes due to the poor air quality but not other schools.) Existing work on air quality modeling typically relies on area-specific, expert-selected data features and fail to model the complex spatial and temporal relationships between the air quality data generated from a sensor network. In this talk, I will present our latest approach for forecasting the short-term (next 24 hours) PM 2.5 concentrations using a deep learning model. The model learns the spatial relationships between air quality sensors by first mining publicly available geographic data to determine how the built environment affects air quality and then performing a diffusion convolutional process on the sensor network. Next, the model learns the temporal dependencies of the air quality readings by leveraging the sequence-to-sequence encoder-decoder architecture. We have evaluated our model on two real-world air quality datasets (Beijing and Los Angeles) and showed consistent improvement over the state-of-the-art deep learning approaches.
Yao-Yi Chiang is an Associate Professor (Research) in Spatial Sciences, the Director of the Spatial Computing Laboratory at the Spatial Sciences Institute, and the Associate Director of the Integrated Media Systems Center at the University of Southern California. He received his Ph.D. degree in Computer Science from the University of Southern California; his Bachelor degree in Information Management from the National Taiwan University. Dr. Chiang's general area of research is artificial intelligence and data informatics, with a focus on information integration and spatial data analytics. He develops computer algorithms and intelligent systems that discover, collect, fuse, and analyze data from heterogeneous sources to solve real-world problems. Dr. Chiang is also an expert on digital map processing and geospatial information system (GIS). Prior to USC, Dr. Chiang worked as a research scientist for Geosemble Technologies and Fetch Technologies in California. Geosemble Technologies was founded based on a patent on geospatial data fusion techniques, and he was a co-inventor. Geosemble Technologies was acquired by TerraGo, and Fetch Technologies was acquired by Connotate, both in 2012.