Quantile Model Regression of Nameplate Capacities: P50 and P90 Megawatt Hydrothermal Projection

This paper provides the most data-driven estimates of hidden hydrothermal technical capacity of the Great Basin. Geothermal has recently been included in large scale renewable energy generation models, with same resolution as wind and solar. Previous estimates for hydrothermal did not include any measurement of permeability nor uncertainty. The Great Basin area provides the most data rich location to identify positive and negative hydrothermal sites and data observables (features), which have been used in XGBoost regression to make hydrothermal capacity predictions. This work is novel in it uses 37 nameplate capacities (megawatts) as the positive labels and 248 negative (0 megawatt) locations. The independent variables or features are 48 geophysical and geologic attributes. Three XGBoost predictions were input in the reV (Renewable Energy Potential) model. These models were chosen because of their fits to the 37 geothermal plants and negative locations and reasonable prediction ranges and totals within the INGENIOUS area. We present a method for subsampling the negative sites to bring the labels into balance that utilizes the geologic domain knowledge to represent the negatives proportionally. Overall, the distributions of hydrothermal technical capacity and site levelized cost of energy (LCOE) are much tighter and lower than previous estimates for the Great Basin, which only used CONUS scaled temperature estimates. Percentile (50th and 90th) models provide bookends for these metrics.