Read about geothermal machine learning, woody plants, and mangroves in this Bulletin.
by USGS-PSU Partnership
December 12, 2023
Share
Announcements
Andrew Fountain on NPR!
Listen to Professor Andrew Fountain talk about disappearing glaciers in the American West with NPR's Ayesha Rascoe.
Partner Publications
In their recent publication, "Cursed? Why One Does Not Simply Add New Data Sets to Supervised Geothermal Machine Learning Models," Stanley P. Mordensky (GMEG), Erick R. Burns (USGS), John J. Lipor, and Jacob DeAngelo explore whether or not the addition of new but partially correlated feature data in machine learning (ML) pipelines will improve resulting models of hydrothermal favorability. They concluded that producing optimal ML models for geothermal resource assessments is dependent on feature engineering and selection, as well as dimensionality reduction. This is because in higher-dimensional feature spaces, it is more difficult to discern the relationships between features and the presence or absence of hydrothermal systems.
Jennifer Morse (PSU) contributed to a paper, “Woody Plant–Soil Relationships in Interstitial Spaces Have Implications for Future Forests Within and Beyond Urban Areas,” that discusses how unmanaged interstitial areas can support the growth of woody plant communities. Researchers quantified woody vegetation at the residential-wildland interface and in exurban reference natural areas in six metropolitan regions across the continental USA. In consideration of anthropogenic differences between reference and interstitial sites, they also analyzed soil N and C cycling processes. The authors discovered differences in woody plant community composition between interstitial and reference sites in most metropolitan regions.
Pascal D. Caraccioli (GMEG), Stanley P. Mordensky (GMEG), Erick R. Burns (USGS), and the Nevada Machine Learning research team applied machine learning techniques using an artificial neural network (ANN) to identify suitable sites for geothermal exploration. The research team noted 83 known hydrothermal systems as positive and needed negative sites for the ANN to develop a model that could discern the two classes. They selected 62 negative sites that did not contain hydrothermal resources. After taking into account the sparsity of hydrothermal systems and location types without hydrothermal systems, the research team developed improved models for predicting hydrothermal resource favorability. Check out their paper, “Don’t Let Negatives Hold You Back: Accounting for Underlying Physics and Natural Distributions of Hydrothermal Systems When Selecting Negative Training Sites Leads to Better Machine Learning Predictions” to learn more!
In “Non-native Rhizophora mangle as Sinks for Coastal contamination on Molokai, Hawai'i,” Elise Granek (PSU) and other researchers compared microplastic and pesticide contamination in coastal sites at areas with non-native mangroves and with open coastline. They collected sediment, porewater, and mangrove plant tissues to quantify microplastic and pesticide concentrations and detected six pesticides, most commonly bifenthrin. This insecticide had higher concentrations in mangrove roots than in sediment, suggesting that roots can accumulate pesticides and act as a sink for organic contaminants.