Geology topics

Jacob DeAngelo

Publications and source records attributed to Jacob DeAngelo.

At least 19 recordsLinked to original sources

USGS2026

Quantitative mineral resource assessment of lithium pegmatite deposits in the southern Appalachian orogen

The first quantitative mineral resource assessment for undiscovered lithium pegmatite deposits in the southern Appalachian region of the United States was conducted. Permissive tracts for lithium pegmatite deposits were delineated by integrating lithological, tectonic, geochemical, geophysical and mineral occurrence data. Lithium pegmatite prospectivity of the tracts was ranked with simplified mappable criteria, including proximity to Paleozoic felsic intrusions and major lithotectonic structures, stream sediment geochemical anomalies, and pegmatite occurrence data. The geospatial data and permissive tracts were used to estimate the number of undiscovered lithium pegmatite deposits. These estimates were integrated into probabilistic simulations along with a new global lithium pegmatite grade and tonnage dataset to quantify potential contained undiscovered lithium resources. An economic filter was applied to convert the probabilistic estimates of contained lithium into recoverable material. The identified lithium pegmatite resources for the Carolina Lithium and Kings Mountain deposits, North Carolina, contain 1589 thousand tons (kt) of Li 2 O. The median contained undiscovered resource for the southern Appalachian orogen was estimated to be 2240 kt Li 2 O. At 90% confidence, the region contains at least 130 kt Li 2 O, and 10,700 kt at 10% confidence. After applying economic filters, the median recoverable contained resource was 1430 kt Li 2 O, corresponding to approximately 201 years of current lithium imports for consumption in the United States. North and South Carolina are likely to contain most of these resources. Coarse data resolution and intra-state variations in the geological data contribute to uncertainty of undiscovered lithium pegmatite resources. Continued efforts to harmonize disparate geospatial datasets with updated or new information can improve the accuracy and precision of estimated undiscovered lithium pegmatite resources in the study area and at broader scales.

Alabama, Georgia, Maryland, North Carolina, South ↗

USGS2026

Conventional hydrothermal power-producing systems of the Great Basin, USA

As part of the update to the electric-grade conventional hydrothermal assessment of the Great Basin, USA, Monte Carlo analyses of identified resources within explored regions will be performed to make estimates of discovered resources and associated uncertainty. Analyses use conditional statistics where estimates are conditioned upon a hydrothermal favorability map, allowing for the likelihood that more resources exist in regions of higher hydrothermal favorability. For these analyses, a dataset of identified hydrothermal systems is compiled, and the new compilation is described herein. Recognizing that a single hydrothermal system may be developed with multiple power plants, and that the hydrothermal upflow zone may be several kilometers across with many measurements characterizing a single hydrothermal system, a procedure was developed and employed to create clusters of points (power plants, measurements, etc.) that are associated with a single system, and a new central point was defined as the best estimator of the center of the hydrothermal system. Hydrothermal systems were uniquely identified by grouping electric-grade hydrothermal measurements and operating power plants within a distance of 10 km. Groups that are >10 km apart are assumed to be different electric-grade hydrothermal systems. While 10 km was used as the threshold, most systems were significantly further apart, and most points within groups were typically within 5 km of each other. A well measurement was considered an electric-grade measurement of a hydrothermal system if it had two properties: a measured temperature of >85 °C and evidence of hydrothermal convection. Other points that were added to the dataset are locations of operating powerplants or locations that have been classified as an electric-grade hydrothermal resource by either the U.S Geological Survey (USGS) or the Great Basin Center for Geothermal Energy. After all points are assigned to systems, new points were computed with the goal of identifying the center of the throat of the hydrothermal upflow zone. If operating powerplants exist for a system, then the arithmetic average of all power plant locations is used. Otherwise, if USGS made an estimate, that location is used. In the absence of both powerplants or USGS estimates, the arithmetic average of all electric-grade measurement locations is used. An example is shown of how these newly compiled locations might be ranked for uncertainty analyses, where higher confidence is assumed if measured temperature is higher and there are many supporting measurements indicating an electric-grade resource. In summary, 28 systems have operating power plants, an additional 78 systems are known identified electric-grade hydrothermal resources, and 99 new systems were identified as probable electric-grade systems with varying levels of confidence. These 205 locations are shown as a function of a recent hydrothermal favorability map, conceptually illustrating the conditional statistics that can be used to make estimates of the undiscovered resources of the Great Basin. An accompanying data release provides summaries of developed capacity by system and USGS estimates of likely total capacity and associated uncertainty.

Arizona, California, Idaho, Nevada, Oregon, Utah↗

USGS2025

Enhanced geothermal systems electric-resource assessment for the Great Basin, southwestern United States

The U.S. Geological Survey recently (2025) completed a provisional assessment of the geothermal-electric resources associated with high-temperature, low-permeability rock formations of the Great Basin, Southwestern United States. If sufficient technological advances to commercialize enhanced geothermal systems occur, then a current best provisional estimate for electric-power generation capacity of 135 gigawatts electric are available from the upper 6 kilometers of the Earth’s crust. This estimate is a potential substantial increase of the installed geothermal electricity-generating capacity from <1 to 10 percent of current total U.S. power production capacity.

California, Idaho, Nevada, Oregon, Utah↗

USGS2025

Favorability mapping for hydrothermal power resource assessments of the Great Basin, USA

The U.S. Geological Survey (USGS) is updating the 2008 assessment of conventional hydrothermal resources for the Great Basin in the western United States. As part of this work, the workflow for hydrothermal resource favorability maps is being modified to integrate modern data-driven machine learning (ML) methods. Improvements include: [1] using new and refined evidence layers (features); [2] using an order of magnitude more training sites (labeled examples); [3] utilizing simple but non-linear supervised ML algorithms; [4] representing positive training sites (wells with measured heat flow) with their ordinal value proportional to the magnitude of convective upflow (i.e., low, high, or very high convective signals instead of past strategies using positive-negative labels); [5] supplementing training sites with additional sites with low convective signals to represent diverse under-sampled areas where hydrothermal systems are unlikely to exist; [6] comparing with competing approaches; and [7] utilizing Monte Carlo cross-validation to estimate and evaluate prediction uncertainty. For the new favorability map, over half of the power-producing systems (i.e., 15 of 28) are predicted in the 99th percentile of most favorable locations (i.e., the highest 1 % of favorability, corresponding to 1 % of the map area), exceeding the performance of past models that have explicitly used power plants as training sites. Previous favorability maps predicted approximately half of the power-producing hydrothermal systems above the 80th percentile (i.e., 20 % of the map area). For the new favorability map, 93 % of power-producing systems (i.e., 26 of 28) are above the 80th percentile. The power-producing systems for which the new model does not perform well are either comparatively small, low-temperature systems or systems also not predicted well by prior modeling approaches, suggesting that these few systems are unusual when compared with most power-producing systems. Focusing research on these known, seemingly different systems may yield new insights and subsequent discovery of new prospects.

California, Idaho, Nevada, Oregon, Utah↗

USGS2025

Preventing overfitting when using tree-based methods for mapping hydrothermal favorability

Ensemble tree-based algorithms are robust tools for estimating sparsely distributed resources with non-linear dependencies (e.g., hydrothermal systems). These algorithms naturally accommodate the threshold conditions necessary to enable and support hydrothermal systems (e.g., having sufficient heat and permeability) and are simpler than many other non-linear machine learning strategies (e.g., artificial neural networks), which is an advantage when working with few labeled examples from which to learn. In previous work, we used eXtreme Gradient Boosting (XGBoost) to produce regional prediction and uncertainty maps of hydrothermal favorability; however, recent studies suggest that, even when properly applied, XGBoost has some risk of overfitting when there are few labeled examples from which to learn. To evaluate overfitting when constructing hydrothermal favorability maps with tree-based methods, we compare XGBoost with Extremely Randomized Trees (ExtraTrees), another ensemble tree-based algorithm that has the potential to underfit when using few labeled examples. We hold all other modeling parameters constant, resulting in two contrasting favorability maps of conventional geothermal resources for the Great Basin. Our results indicate that ExtraTrees demonstrably reduces overfitting compared with XGBoost. After considering overall performance, we conclude that ExtraTrees provides a more suitable modeling approach than XGBoost for the purposes of conventional hydrothermal resource assessments.

Conference Paper↗

USGS2024

Geothermal play fairway analysis, part 1: Example from the Snake River Plain, Idaho

The Snake River Plain (SRP) volcanic province overlies the track of the Yellowstone hotspot, a thermal anomaly that extends deep into the mantle. Most of the area is underlain by a basaltic volcanic province that overlies a mid-crustal intrusive complex, which in turn provides the long-term heat flux needed to sustain geothermal systems. Previous studies have identified several known geothermal resource areas within the SRP. For the geothermal study presented herein, our goals were to: (1) adapt the methodology of Play Fairway Analysis (PFA) for geothermal exploration to create a formal basis for its application to geothermal systems, (2) assemble relevant data for the SRP from publicly available and private sources, and (3) build a geothermal PFA model for the SRP and identify the most promising plays, using GIS-based software tools that are standard in the petroleum industry.

Idaho↗

USGS2024

Geothermal Play Fairway Analysis, Part 2: GIS methodology

Play Fairway Analysis (PFA) in geothermal exploration originates from a systematic methodology developed within the petroleum industry and is based on a geologic, geophysical, and hydrologic framework of identified geothermal systems. We tailored this methodology to study the geothermal resource potential of the Snake River Plain and surrounding region, but it can be adapted to other geothermal resource settings. We adapted the PFA approach to geothermal resource exploration by cataloging the critical elements controlling exploitable hydrothermal systems, establishing risk matrices that evaluate these elements in terms of both probability of success and level of knowledge, and building a code-based ‘processing model’ to process results.

Geothermics↗

USGS2024

Updated three-dimensional temperature maps for the Great Basin, USA

As part of the periodic update of the geothermal energy assessments for the USA (e.g., last update by Williams and others, 2008), a new three-dimensional temperature map has been constructed for the Great Basin, USA. Williams and DeAngelo (2011) identified uncertainty in estimates of conductive heat flow near land surface as the largest contributor to uncertainty in previously published temperature maps. The new temperature maps incorporate new conductive heat flow estimates developed by DeAngelo and others (2023). Predicted temperatures at depth are compared with representative measurements (for conductively dominated conditions), showing good agreement under relatively simple uniform conditions. Inputs included radiogenic heat production for all layers of 1.89 μW/m 3 , effective bulk thermal conductivity of 2.7 W/m/°C for all rocks underlying sedimentary basins, and a previously published (Williams and DeAngelo, 2011) empirically driven estimate of increasing thermal conductivity with depth in sedimentary sequences. The resulting three-dimensional temperature model is published in a USGS data release associated with this manuscript (Burns and others, 2023).

Arizona, California, Idaho, Nevada, Oregon, Utah↗

USGS2024

Predicting hydrothermal reservoir depth from chemical geothermometers using a three-dimensional temperature model in the Great Basin, USA

Recent work in the Great Basin region of the western United States has made it possible to predict the depth of hydrothermal reservoirs (i.e., the depth at which heat is accumulated prior to ascent via hydrothermal upflow) identified through geochemistry and to contextualize the spatial patterns of these reservoir depths. Chemical geothermometers use the chemical and mineral constituents of hydrothermal fluids to predict the temperature at which fluids equilibrated with the host rocks at depth. Assuming that most of the Great Basin is dominated by conductive conditions until a vertically connected hydrothermal flow path is created (e.g., by faulting), geothermometers reflect the chemical and thermal conditions at the depth interval that the fluid has conductively equilibrated over a long period before a vertical conduit allows convective upflow. By pairing geothermometer temperature estimates with our recent three-dimensional temperature model of conductive heat flow in the Great Basin, we estimate the corresponding reservoir depths and construct a map of circulation depths. The predicted depths from geothermometers have spatial patterns across the Great Basin that relate to patterns seen in other geologic and geophysical data. Deeper springs generally occur disproportionately in areas with higher strain rates and in basins. We posit that current elevated strain rates reflect patterns of historic deformation where ongoing tectonic activity maintains permeable pathways to deeper reservoirs, some of which are estimated to exceed 6 km depth. Basins, as expected, contain a disproportionate number of these deep systems, because the underlying aquifers are closer to the surface in basins, thus requiring less water pressure to reach the surface than in mountain ranges. Most springs estimated to have their source in a deep reservoir occur at places known to host a hydrothermal system; these refined depth estimates of the source reservoir can help to better constrain the source depth for many known hydrothermal systems across the Great Basin.

Arizona, California, Idaho, Nevada, Oregon, Utah↗

USGS2024

Separating signals in elevation data improves supervised machine learning predictions for hydrothermal favorability

A recent study identified topography (land surface elevation above sea level) as an important input dataset (feature) for predicting the location of hydrothermal systems in the Great Basin in Nevada. Yet, topography is generally a result of more than one geological process and may consequently contain multiple distinct signals. For example, the geologic evolution of the Great Basin has produced both crustal thickening (i.e., regional-scale trends in elevation) and thinning via Basin and Range extensional faulting (i.e., valley-scale topographic relief). We postulate that these geologic processes may affect the occurrence of hydrothermal systems differently. Therefore, we separate the regional trend from the valley-scale signal in the Great Basin, and then use them separately to evaluate the importance of each as predictors for hydrothermal favorability. Our prior work applying supervised machine learning (ML) using the data from the Nevada Machine Learning Project demonstrated that employing a training strategy that randomly selects negative training sites produces better performing models for predicting hydrothermal favorability than a training strategy that uses expert-selected negatives. The models created using both training strategies exhibited a west-east geographic trend in the predictions for the favorability of hydrothermal resources. These models generally predicted higher favorability in western Nevada and lower favorability in eastern Nevada. This west-east trend in predicted favorability correlates with elevation across the Great Basin, which trends higher from west to east. By separating the original elevation feature into distinct features for elevation trend (i.e., regional-scale topography) and detrended elevation (i.e., valley-scale or local relative topography), we find that models using the separated topographic signals consistently outperform competing models that use the original elevation feature. Although western Nevada still exhibits higher favorability than eastern Nevada, using separated signals for regional elevation and local structure reduces the west-east prediction trend in the region and emphasizes structures associated with hydrothermal upflow. This work emphasizes how carefully engineering features to represent geological conditions relevant to hydrothermal systems allows ML algorithms to detect important patterns for predicting hydrothermal resource favorability and leads to better model performance.

Conference Paper↗

USGS2023

When less is more: How increasing the complexity of machine learning strategies for geothermal energy assessments may not lead toward better estimates

Previous moderate- and high-temperature geothermal resource assessments of the western United States utilized data-driven methods and expert decisions to estimate resource favorability. Although expert decisions can add confidence to the modeling process by ensuring reasonable models are employed, expert decisions also introduce human and, thereby, model bias. This bias can present a source of error that reduces the predictive performance of the models and confidence in the resulting resource estimates. Our study aims to develop robust data-driven methods with the goals of reducing bias and improving predictive ability. We present and compare nine favorability maps for geothermal resources in the western United States using data from the U.S. Geological Survey's 2008 geothermal resource assessment. Two favorability maps are created using the expert decision-dependent methods from the 2008 assessment ( i.e., weight-of-evidence and logistic regression). With the same data, we then create six different favorability maps using logistic regression (without underlying expert decisions), XGBoost, and support-vector machines paired with two training strategies. The training strategies are customized to address the inherent challenges of applying machine learning to the geothermal training data, which have no negative examples and severe class imbalance. We also create another favorability map using an artificial neural network. We demonstrate that modern machine learning approaches can improve upon systems built with expert decisions. We also find that XGBoost, a non-linear algorithm, produces greater agreement with the 2008 results than linear logistic regression without expert decisions, because the expert decisions in the 2008 assessment rendered the otherwise linear approaches non-linear despite the fact that the 2008 assessment used only linear methods. The F1 scores for all approaches appear low (F1 score < 0.10), do not improve with increasing model complexity, and, therefore, indicate the fundamental limitations of the input features ( i.e., training data). Until improved feature data are incorporated into the assessment process, simple non-linear algorithms ( e.g., XGBoost) perform equally well or better than more complex methods ( e.g., artificial neural networks) and remain easier to interpret.

Geothermics↗

USGS2023

Exploratory analysis of machine learning techniques in the Nevada geothermal play fairway analysis

Play fairway analysis (PFA) is commonly used to generate geothermal potential maps and guide exploration studies, with a particular focus on locating and characterizing blind geothermal systems. This study evaluates the application of machine learning techniques to PFA in the Great Basin region of Nevada. Following the evaluation of various techniques, we identified two approaches to PFA that produced promising results, 1) supervised Bayesian probabilistic neural networks to generate geothermal potential maps with confidence intervals, and 2) unsupervised principal component analysis paired with k-means clustering to generate both cluster maps to help identify spatial patterns, as well as new combined feature inputs. We applied these techniques to perform a comparative analysis between two principal sets of geological and geophysical features related to permeability and heat and a set of positive (known geothermal resources) and negative training sites (known drill sites with unsuitable geothermal conditions). We found that these methods constrain previously unrecognized feature controls on geothermal favorability, many of which are spatially organized within the extent of cluster groups and the major structural-hydrologic domains of the study area. Furthermore, we utilized exploratory unsupervised modeling to highlight spatial relationships between input data and predictive output results of our supervised modeling. Finally, we demonstrate how our models compare to the previous Nevada PFA and how the rapid insights these machine learning techniques offer may support future assessments of both known and undiscovered blind geothermal systems in the Great Basin region of Nevada and beyond.

Nevada↗

USGS2023

New maps of conductive heat flow in the Great Basin, USA: Separating conductive and convective influences

Geothermal well data from Southern Methodist University and the U.S. Geological Survey (USGS) were used to create maps of estimated background conductive heat flow across the Great Basin region of the western United States. These heat flow maps were generated as part of the USGS hydrothermal and Enhanced Geothermal Systems resource assessment process, and the creation process seeks to remove the influence of hydrothermal convection from the predictions of the background conductive heat flow. The heat flow maps were constructed using a custom-developed iterative process using weighted regression, in which convectively influenced outliers were de-emphasized by assigning lower weights to measurements with heat flow values further from the estimated local trend (e.g., local convective influence). The local linear weighted regression algorithm is two-dimensional locally estimated scatterplot smoothing where smoothness was controlled by varying the number of nearby wells used for each local interpolation. Three maps resulting from conductive heat flow models are detailed in this paper, highlighting the influence of measurement confidence. The three maps use either: measurements from all wells with equal weight (no confidence weights), or one of two different published categorization methods to de-emphasize low-quality measurements; one categorization method graded thermal gradient quality, the other categorization method graded thermal conductivity quality. Each map is an estimate of background conductive heat flow as a function of reported data quality, and a point coverage is also provided for all wells in the compiled dataset. The point coverage includes an important new attribute for geothermal wells: the residual, which can be interpreted as the departure of a well from the estimated background heat flow conditions, and the value of the residual may be useful in identifying the influence of fluids (hydrothermal or groundwater) on conductive heat flow. Of the three maps presented, the map that de-emphasized the impact of wells with low-quality thermal gradient measurements appears to perform best because it did not incorporate many of the wells in the Snake River Plain that do not penetrate the aquifer and are therefore very unlikely to reflect true conductive conditions.

Great Basin↗

USGS2023

Detrending Great Basin elevation to identify structural patterns for identifying geothermal favorability

Topography provides information about the structural controls of the Great Basin and therefore information that may be used to identify favorable structural settings for geothermal systems. The Nevada Machine Learning Project (NVML) tested the use of a digital elevation map (DEM) of topography as an input feature to predict geothermal system favorability. A recent study re-examines the NVML data, identifying the DEM as the most important feature, showing a broad uniform pattern of high-favorability in the lower-elevation west and low-favorability in the higher elevation east of their study area in north-central Nevada. This regional elevation trend conflicts with the geologic notion that local relative topography should be used to identify geologic structures associated with favorable structural settings for hydrothermal upflow. Specifically, local relative topography gives information about position in the mountains, in the valleys, or at the transitions between, aiding in identification of faults and fault intersections. As part of U.S. Geological Survey efforts to engineer features that are useful for predicting geothermal resources, we construct a detrended elevation map that emphasizes local relative topography and highlights features that geologists use for identifying geothermal systems (i.e., providing machine learning algorithms with features that may improve predictive skill by emphasizing the information used by geologists). Herein, we describe the removal of the regional trend in elevation to emphasize the basin-and-range scale structural features, creating detrended elevation maps. Regional elevation trends were estimated using a local linear regression and subtracted from the actual elevation using a 30-m DEM. In an effort to optimize the detrended surface, alternate versions were produced with different rates of smoothness resulting in three detrended elevation maps. The resulting elevation trend surfaces (a proxy for crustal thickness) are compared with conductive heat flow maps, and a general pattern was observed of a negative correlation between heat flow and regional elevation in many areas, indicating that thinner crust may be causing elevated heat flow in some areas and thicker crust may cause the observed heat flow lows. Because these detrended elevation maps emphasize geologic structure and relative displacement, these products may also be useful for other geologic research including mineral exploration, hydrologic research, and defining geologic provinces.

Geothermal Resources Council Transactions↗

USGS2023

Cursed? Why one does not simply add new data sets to supervised geothermal machine learning models

Recent advances in machine learning (ML) identifying areas favorable to hydrothermal systems indicate that the resolution of feature data remains a subject of necessary improvement before ML can reliably produce better models. Herein, we consider the value of adding new features or replacing other, low-value features with new input features in existing ML pipelines. Our previous work identified stress and seismicity as having less value than the other feature types (i.e., heat flow, distance to faults, and distance to magmatic activity) for the 2008 USGS hydrothermal energy assessment; hence, a fundamental question regards if the addition of new but partially correlated features will improve resulting models for hydrothermal favorability. Therefore, we add new maps for shear strain rate and dilation strain rate to fit logistic regression and XGBoost models, resulting in new 7-feature models that are compared to the old 5-feature models. Because these new features share a degree of correlation with the original relatively uninformative stress and seismicity features, we also consider replacement of the two lower-value features with the two new features, creating new 5-feature models. Adding the new features improves the predictive skill of the new 7-feature model over that of the old 5-feature model; albeit, that improvement is not statistically significant because the new features are correlated with the old features and, consequently, the new features do not present considerable new information. However, the new 5-feature XGBoost model has a statistically significant increase in predictive skill for known positives over the old 5-feature model at p = 0.06. This improved performance is due to the lower-dimensional feature space of the former than that of the latter. In higher-dimensional feature space, relationships between features and the presence or absence of hydrothermal systems are harder to discern (i.e., the 7-feature model likely suffers from the “curse of dimensionality”).

Geothermal Resources Council Transactions↗

USGS2023

Predicting large hydrothermal systems

We train five models using two machine learning (ML) regression algorithms (i.e., linear regression and XGBoost) to predict hydrothermal upflow in the Great Basin. Feature data are extracted from datasets supporting the INnovative Geothermal Exploration through Novel Investigations Of Undiscovered Systems project (INGENIOUS). The label data (the reported convective signals) are extracted from measured thermal gradients in wells by comparing the total estimated heat flow at the wells to the modeled background conductive heat flow. That is, the reported convective signal is the difference between the background conductive heat flow and the well heat flow. The reported convective signals contain outliers that may affect upflow prediction, so the influence of outliers is tested by constructing models for two cases: 1) using all the data (i.e., -91 to 11,105 mW/m2), and 2) truncating the range of labels to include only reported convective signals between -25 and 200 mW/m2. Because hydrothermal systems are sparse, models that predict high convective signal in smaller areas better match the natural frequency of hydrothermal systems. Early results demonstrate that XGBoost outperforms linear regression. For XGBoost using the truncated range of labels, half of the high reported signals are within < 3 % of the highest predictions. For XGBoost using the entire range of labels, half of the high reported signals are in < 13 % of the highest predictions. While this implies that the truncated regression is superior, the all-data model better predicts the locations of power-producing systems (i.e., the operating power plants are in a smaller fraction of the study area given by the highest predictions). Even though the models generally predict greater hydrothermal upflow for higher reported convective signals than for lower reported convective signals, both XGBoost models consistently underpredict the magnitude of higher signals. This behavior is attributed to low resolution/granularity of input features compared with the scale of a hydrothermal upflow zone (a few km or less across). Trouble estimating exact values while still reliably predicting high versus low convective signals suggests that a future strategy such as ranked ordinal regression (e.g., classifying into ordered bins for low, medium, high, and very high convective signal) might fit better models, since doing so reduces problems introduced by outliers while preserving the property of larger versus smaller signals.

Geothermal Resources Council Transactions↗

USGS2023

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

Selecting negative training sites is an important challenge to resolve when utilizing machine learning (ML) for predicting hydrothermal resource favorability because ideal models would discriminate between hydrothermal systems (positives) and all types of locations without hydrothermal systems (negatives). The Nevada Machine Learning project (NVML) fit an artificial neural network to identify areas favorable for hydrothermal systems by selecting 62 negative sites where the research team had confidence that no hydrothermal resource exists. Herein, we compare the implications of the expert selection of negatives (i.e., the NVML strategy) with a random sample strategy, where it is assumed that areas outside the favorable structural ellipses defined by NVML are negative. Because hydrothermal systems are sparse, it is highly probable that, in the absence of a favorable geological structure, hydrothermal favorability is low. We compare three training strategies: 1) the positive and negative labeled examples from NVML; 2) the positive examples from NVML with randomly selected negatives in equal frequency as NVML; and 3) the positive examples from NVML with randomly selected negatives reflecting the expected natural distribution of hydrothermal systems relative to the total area. We apply these training strategies to the NVML feature data (input data) using two ML algorithms (XGBoost and logistic regression) to create six favorability maps for hydrothermal resources. When accounting for the expected natural distribution of hydrothermal systems, we find that XGBoost performs better than the NVML neural network and its negatives. Model validation was less reliable using F1 scores, a common performance metric, than comparing probability estimates at known positives, likely because of the extreme natural class imbalance and the lack of negatively labeled sites. This work demonstrates that expert selection of negatives for training in NVML likely imparted modeling bias. Accounting for the sparsity of hydrothermal systems and all the types of locations without hydrothermal systems allows us to create better models for predicting hydrothermal resource favorability.

Geothermal Resources Council Transactions↗

USGS2022

Predicting geothermal favorability in the western United States by using machine learning: Addressing challenges and developing solutions

Previous moderate- and high-temperature geothermal resource assessments of the western United States utilized weight-of-evidence and logistic regression methods to estimate resource favorability, but these analyses relied upon some expert decisions. While expert decisions can add confidence to aspects of the modeling process by ensuring only reasonable models are employed, expert decisions also introduce human bias into assessments. This bias presents a source of error that may affect the performance of the models and resulting resource estimates. Our study aims to reduce expert input through robust data-driven analyses and better-suited data science techniques, with the goals of saving time, reducing bias, and improving predictive ability. We present six favorability maps for geothermal resources in the western United States created using two strategies applied to three modern machine learning algorithms (logistic regression, support-vector machines, and XGBoost). To provide a direct comparison to previous assessments, we use the same input data as the 2008 U.S. Geological Survey (USGS) conventional moderate- to high-temperature geothermal resource assessment. The six new favorability maps required far less expert decision-making, but broadly agree with the previous assessment. Despite the fact that the 2008 assessment results employed linear methods, the non-linear machine learning algorithms (i.e., support-vector machines and XGBoost) produced greater agreement with the previous assessment than the linear machine learning algorithm (i.e., logistic regression). It is not surprising that geothermal systems depend on non-linear combinations of features, and we postulate that the expert decisions during the 2008 assessment accounted for system non-linearities. Substantial challenges to applying machine learning algorithms to predict geothermal resource favorability include severe class imbalance (i.e., there are very few known geothermal systems compared to the large area considered), and while there are known geothermal systems (i.e., positive labels), all other sites have an unknown status (i.e., they are unlabeled), instead of receiving a negative label (i.e., the known/proven absence of a geothermal resource). We address both challenges through a custom undersampling strategy that can be used with any algorithm and then evaluated using F1 scores.

western United States↗