Coffield, Shane | Engineering Papers

DOE OSTI2021

Machine learning and artificial intelligence for wildfire prediction

Wildfire ignition, intensity, and spread rates are tightly linked with water cycle extremes. The science of wildfire prediction has traditionally encompassed the use of physical and empirical models to quantify the direction and speed of fire spread, plume injection and fire-aerosol impacts on atmospheric composition, predictions of fire season severity on subseasonal-to-seasonal (S2S) time scales, and assessment of the spatial and temporal patterns of fire risk across landscapes. Together with expanding observation networks, machine learning and artificial intelligence (AI) have the potential to revolutionize the application of such models for fire science, saving lives, protecting critical infrastructure, and providing more accurate estimates of wildfire-climate feedbacks.

54 ENVIRONMENTAL SCIENCES↗

NASA NTRS2016

Evaluation of Different MODIS AOD Retrieval Algorithms for PM (sub 2.5) Estimation in the Western, Midwestern and Southeastern United States with Implications for Public Health

This study was part of the research activities of the Center for Applied Atmospheric Research and Education (CAARE) funded by the NASA MUREP (Minority University Research and Education Project) Institutional Research Opportunity (MIRO) Program. Satellite measurements of Aerosol Optical Depth (AOD) have been shown to be correlated with ground measurements of fine particulate matter less than 2.5 microns PM (sub 2.5), which in turn has been linked to respiratory and heart diseases. The strength of the correlation between AOD and PM (sub 2.5) varies for different AOD retrieval algorithms and geographic regions. We evaluated several Moderate Resolution Imaging Spectrometer (MODIS) AOD products from different satellites (Aqua vs. Terra), retrieval algorithms (Dark Target versus Deep Blue), Collections (5.1 versus 6) and spatial resolutions (10-kilometers versus 3-kilometers) for cities in the Western, Midwestern and Southeastern U.S. We developed and validated PM (sub 2.5) prediction models using remotely-sensed AOD data, which were improved by incorporating meteorological variables (temperature, relative humidity, precipitation, wind speed, and wind direction) from the North American Land Data Assimilation System Phase 2 (NLDAS-2). Adding these meteorological data significantly improved the predictive power of all the PM (sub 2.5) models, especially in the Western U.S. Temperature, relative humidity and wind speed were the most significant meteorological variables throughout the year in the Western U.S. Wind speed was the most significant meteorological variable for the cold season while temperature was the most significant variable for the warm season in the Midwestern and Southeastern U.S. Our study re-establishes the connection between PM (sub 2.5) and public health concerns including respiratory and cardiovascular diseases (asthma, high blood pressure, coronary heart disease, heart attack, and stroke). Using PM (sub 2.5) data and health data from the Centers for Disease Control and Prevention (CDC)'s Behavioral Risk Factor Surveillance System (BRFSS), our statistical analysis showed that heart attack and stroke occurrences had the strongest correlations with PM (sub 2.5).

Environmental Public Health↗