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

• Martine Mathieu: Spatial Predictive Dispersion Modeling of Particulate Matter (PM_2.5) Concentrations as Surrogates for Exposure in Colfax, Louisiana, USA (advised by Jennifer Richmond-Bryant, FER)

Abstract: In Colfax, LA, several residents have experienced adverse health effects in an area surrounding an open burn/open detonation hazardous waste thermal treatment facility that accepts waste including soils from Superfund sites, fireworks, military ordnances, and munitions, resulting in emission of fine particulate matter (PM). To predict the spatial distribution of PM_2.5 concentrations as surrogates for exposures and help Colfax residents understand their health risks from exposure to the facility emissions, we used AERMOD to model dispersion in Colfax.  Our preliminary results showed that PM_2.5 dispersion occurred over two communities adjacent to the facility: Colfax and The Rock. The average concentrations exceeded the EPA’s annual ambient PM_2.5 standard of 12 mg/m³.

• Ian McGregor: The Importance of Trade-Offs Between Detection Time and Accuracy for Multi-Source Deforestation Monitoring (advised by Josh Gray, FER)

Abstract: Detecting deforestation as quickly and accurately as possible is important for near real-time monitoring (NRTm) of forests, especially via remote sensing. Recently, multi-source approaches have improved our detection capability, but some studies have hinted at the lack of an optimal solution. To explore this accuracy-latency trade-off and determine if it’s limited to optical sensors, we developed a novel approach combining data from Landsat-8, Sentinel-2, and Sentinel-1 in northern Myanmar. We found three main results: 1) we quantified the full spectrum of trade-offs; 2) including Sentinel-1 did not improve results; and 3) our approach returned fast detections with high accuracy.

• Annie Paulukonis: Agricultural Field Delineations for Use in Large-Scale Ecological Risk Assessments (co-advised by Helena Mitasova (MEAS) and Tom Purucker (US EPA))

Abstract: The U.S. EPA uses spatially explicit agricultural information from the USDA Cropland Data Layer (CDL) to quantify potential pesticide exposures of non-target species, but the CDL lacks well-defined field boundaries necessary for higher tier modeling efforts. This work describes a methodology that uses unique pixel histories and basic geostatistical approaches to define field boundaries for use in ecological risk assessment efforts. The approach captures field edges while closely matching both original CDL crop patterns and reported crop metrics for areas of high proportional agriculture use and can be altered to reflect trends in county-level field size/use for delineation.

• Felipe Sanchez: Spatiotemporal Relative Risk Distribution of Porcine Reproductive and Respiratory Syndrome Virus in the United States (advised by Gustavo Machado, Population Health and Pathobiology, CVM)

Abstract: Identifying areas and farms at greater risk for porcine reproductive and respiratory syndrome virus (PRRSV) transmission is pivotal for targeted surveillance and disease control strategies. In this study, we used an adaptive kernel density approach to define the spatial distance at which the risk of PRRSV transmission between farms is high, and describe how risk is distributed across different farm types. We then used a Bayesian spatiotemporal model to identify variables associated with PRRSV outbreaks. Information gathered from this study addresses an important gap in the understanding related to the spatial range of PRRSV local transmission, and can be used to calibrate future PRRSV transmission models.