ABSTRACT

The objective of this project is to develop a framework to achieve real-time smoke transport prediction and air quality forecasting. Wildfire smoke can transport very fast and pose significant health hazards to communities. State-of-the-art smoke forecasting models typically have infrequent updates and provide predictions with a coarse spatial resolution due to spatiotemporal resolution limitations of input data and the tremendous computational power required to simulate atmospheric conditions. This project will develop real-time smoke transport and air quality prediction methodologies with better spatial resolution for improving the scalability and efficiency of the underlying data processing system to enable timely air quality alerts. While this project is applied towards smoke transport and air quality prediction, this work can be generalized to solve many other big data problems that require such design. The principal investigators will use the materials and topics from this project to enhance education by creating new big data analytics related courses and developing a Big Data Minor program at the University of Nevada, Reno. The project will also provide opportunities to engage more students from underrepresented groups and impact the education of several students, via K-12 outreach and mentoring undergraduate and graduate students.

The intellectual merit of this research is in establishing a novel big data driven air quality prediction for wildfire smoke to provide timely and effective health alerts. The planned new prediction methodology will integrate the novel Gaussian Markov Random Field based real-time spatiotemporal prediction with statistical-based long-term spatiotemporal prediction. To tackle the challenge of missing high-resolution data, a data fusion methodology is planned to integrate fine-grained image data collected from camera networks with air pollution monitoring data to increase data resolution. A Deep Neural Network based smoke density detection process will extract air quality information from camera image data. The planned novel signature time-series based prediction methodology will open opportunities to process larger amounts of spatiotemporal data using limited resources. By identifying critical data based on spatiotemporal properties, the project will develop a communication framework that enables efficient camera data transfer. Efficient parallel and distributed data processing is utterly important to support processing large scale data in real time. The planned decomposition-based parallelization methodology and a performance model driven scheduling framework will enable efficient dynamic computing resource management, which is key to the success of this project.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Please report errors in award information by writing to: awardsearch@nsf.gov.