Geospatial Analytics Ph.D. students will share current research in a series of short talks.

Speakers:

> Christina Perella – Climate as Push and Pull: Forecasting Movement of US Populations (co-advised by Dr. Adam Terando and Dr. Jelena Vukomanovic)

Abstract: Human choices and preferences shape and impact landscapes. Historically, migration has been driven in large part by push and pull factors such as economic opportunity, political oppression, or social networks. Recently, with rising incomes and fewer barriers to relocation, pull factors increasingly include amenities that enhance quality of life, such as outdoor recreation and milder summers and winters. To understand how this trend might continue into the future, we recreate a widely-cited econometric model documenting the relationship between climate variables and population growth in the US and project future population growth under different climate scenarios.

> Rebecca Composto – Best of Both Worlds: Comparing Satellite- and Process-Based Methods to Map Urban Flooding (advised by Dr. Mirela Tulbure)

Abstract: Flooding causes many types of harm from economic losses and damages to disrupting daily life. Flood maps help decision-makers recover from and prepare for future events. Satellite-based and process-based flood models are two effective approaches for mapping floods; however, they are rarely tested in urban areas or compared. To address these gaps, we produced a flood extent using satellite imagery and a flood model for Hurricane Ida (2021) and compared the results.

> Owen Smith – Accelerating Land Surface Phenology Estimation with Computationally Efficient Bayesian LSP Modeling (advised by Dr. Josh Gray)

Abstract: The Bayesian Land Surface Phenology (BLSP) model is a hierarchical Bayesian model which enables the creation of long-term phenology time series from sparse data plus uncertainty quantification through Markov Chain Monte Carlo (MCMC) sampling. However, MCMC methods are computationally intensive, making pixel-wise processing at high resolution and over large spatial regions challenging. I show a reformulation of the BLSP inference problem from an algorithmic perspective along with memory and CPU optimizations to facilitate the computational feasibility of the BLSP approach.

> Randi Butler – Assessing Climate- and Weather-Driven Impacts to Crops of the U.S. National Crop Yields and Losses: Which Data Source is Best? (advised by Dr. Natalie Nelson)

Abstract: The USDA National Agricultural Statistics Survey (NASS) is the premier data source for agricultural production statistics in the U.S, but data quality is compromised by reliance on voluntarily self-reported data. Meanwhile, the USDA Risk Management Agency (RMA), which manages crop insurance for two-thirds of planted acres in the U.S. and mandates reporting, may provide higher quality data as compared to NASS and serve as a more robust alternative. A comparative analysis of NASS and RMA annual crop loss and yields data over 10 years, focusing on corn, cotton, soybean, and wheat, was applied to quantify differences between the two datasets.