DESC0024738

In this proposal, we are going to build a data exploratory and visualization tool to help analyze large-scale of geospatial data from various sources in different spatial resolutions and formats and facilitate users' understanding of the environment. The tool will provide the capabilities to preprocess and combine the data with different resolutions and from multiple modalities and harmonize them to a uniform resolution level that allows for neighborhood (ideally building) level information visualization. The backend machine learning model will produce fine-scale predictions given the sparsely distributed observations of particulate matter and heat levels, together with relevant environmental factors (e.g., weather and built environments). The system will mainly handle two challenges. Firstly, data collection and management from an array of diverse sources in real time is no small feat. The second challenge is the inherent limitations of certain data sources as the data might only be available at limited locations. HoodLab will leverage edge computing techniques to optimize the data transfer costs in the network and quantify uncertainties to the extent possible. HoodLab will also build a cloud-based system for big spatial data management using PostgreSQL/PostGIS and Sedona. HoodLab will develop a machine learning model to predicts fine-scale air quality and urban heat islands given the multi-modal datasets. HoodLab will develop a web-application interface to support efficient query and dynamic visualizations by selecting products, region of interest and time frames to help understand fine-scale predictions and other environmental products. We also discuss in the proposal the anticipated benefits using our existing technologies and the proposed computational methods in environmental, social, economic, and Department of Energy (DOE). We have the detailed work plan towards the technical objectives, and the facility support from the Technology Holding LLC. and the University of Minnesota.