CRII: III: Discovering Complex Mixture Patterns in Spatial Data to Advance Resilience of Communities

Our communities have continued to face unprecedented challenges in recent years, including food-shortages, the COVID-19 pandemic, and natural disasters exacerbated by climate change such as droughts, hurricanes, and fires. To improve the resilience of communities, a key challenge is to understand "where" existing weaknesses are in order to make timely policy intervention and changes in planning. The goal of this project is to investigate novel data science and computational techniques to identify such vulnerabilities using emerging spatial and spatiotemporal big data such as Earth observation data (e.g., satellite-based crop maps), urban points-of-interest, biodiversity databases, geo-tagged social media streams, etc. Specifically, this project will explore the detection of mixture patterns - a new pattern family that focuses on compositions of the types of data points in space and time - which is necessary in revealing many of the vulnerabilities. For example, regions with low crop diversification are subject to devastating impact of a single disease, and places with low economic diversification are vulnerable to changes in supply or demand for a single sector. If successful, the results will lead to improved resilience in various domains, including agriculture (e.g., lowering risks of food shortages), economy (e.g., reducing impact of disturbances such as COVID-19), ecosystems (e.g., biodiversity) and many more. More broadly, mixture patterns may also help improve other data science techniques. For example, in machine learning, mixture patterns of error types may be used to inform better architecture designs or training strategies. Proposed techniques will be disseminated via open-source packages on GitHub as well as incorporation with popular software/tools to enhance research infrastructure and promote reproducible research. This research will also facilitate development of new undergraduate courses at the University of Maryland and help engage students from underrepresented groups in STEM.

This project is expected to result in multiple data science and computing innovations. First, it will explore novel statistically-robust formulations of mixture patterns (e.g., new test statistics and point processes) to allow explicit control of the rate of spurious results, which is critical in real-world applications with limited resources and high societal impact. Second, it will design scalable computational frameworks to identify mixtures patterns with irregular shapes to capture real-world mixture processes with complex footprints. A unique challenge is that local- and pattern-level mixture signatures can be different, which violates the assumption of local-criteria-based search paradigms widely used in clustering-type of techniques. New algorithmic designs will be explored to bridge this gap. Finally, it will investigate new spatiotemporal formulations to capture the non-stationarity of mixture patterns across spatial scales and time-series. If successful, the results will expand data science knowledge with new pattern families, and have the potential to transform related domain science research by opening new lenses for data analytics.

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.