Geothermal resource favorability: select features and predictions for the Nevada Machine Learning subregion of the United States Great Basin curated for DOI 10.1016/j.geothermics.2026.103833

These datasets were compiled to support the publication Caraccioli, P.D., Mordensky, S.P., DeAngelo, J., Burns, E.R., Lipor, J.J., 2026, How Selecting Representative Training Sites and Features Can Improve Machine Learning Models for Hydrothermal Resource Favorability. Geothermics, 103833, https://doi.org/10.1016/j.geothermics.2026.103833 that details how training strategies (e.g., the choice of negative training sites) and the selection of evidence layers (i.e., feature data) influences model performance. The datasets contained herein are the point data for the 13 input features from the Nevada Machine Learning (NVML) grid (n=1,695,126). For this grid, we provide the predictions from the NVML artificial neural network, the hydrothermal favorability maps presented by Caraccioli et al. (2025) from exploring the training strategies and feature sets therein as the median prediction percentile (P50) of 60 realizations, and the uncertainty maps as differences between the 95th and 5th and then 75th and 25th prediction percentiles from the 60 realizations. This grid includes positive training sites (sites with known hydrothermal systems; n=83), negative training sites (sites with a confirmed absence of a hydrothermal system; n=62). The remainder of the grids cells are unlabeled (i.e., grid data with no label regarding the presence or absence of a hydrothermal system). Corresponding geoTiffs files are located in the corresponding subdirectories for model predictions and model uncertainty. Note that the CSV files have too many examples to open in MS Excel. We suggest opening this file using Notepad or Notepad++. As of ArcGIS Pro 3.3.2, ArcGIS Pro could import and display the data in CSV files on a computer with at least 64 GB of RAM.