USGS Science⌕ Search

SEARCH · USGS Science

Results for “Communications Earth and Environment”

Search indexed USGS publications on groundwater, aquifers, geologic maps, mineral resources and earthquakes. Explore source records by subject and place.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

1,662 records · Page 67Linked to original sources
USGS2026

Shallow differentiation of primitive arc magmas at the Jurassic Emigrant Gap mafic complex, Sierra Nevada, California

The Emigrant Gap composite pluton exposes ultramafic to silicic intrusive rocks that preserve the chemical evolution of primitive mafic arc magmas and their open-system interactions in the upper crust during mid-Jurassic growth of the Sierra Nevada batholith (California). We present field and petrographic observations and mineral and whole-rock chemistry of the ~35-km 2 ultramafic to dioritic Emigrant Gap mafic complex and an adjacent penecontemporaneous ~90-km 2 granodiorite that together make up the composite pluton. In the Emigrant Gap mafic complex, four roughly central masses of dunite, wehrlite, and olivine clinopyroxenite are surrounded by weakly layered gabbronorite and non-layered diorite. The ultramafic rocks are cumulates formed from near-liquidus minerals of primitive arc magmas that accumulated in steep feeder zones with substantial modification by melt–mush reaction as primitive liquids repeatedly transited the mush-filled conduits. The dominant gabbronoritic rocks are the variably accumulative products of more advanced crystallization–differentiation of arc tholeiitic basalts and basaltic andesites. The adjacent granodiorite intrusion originated separately and preserves field and geochemical evidence for assimilation of metasedimentary rocks. Open-system hybridization between the gabbronoritic mushes and the granodioritic magma produced an intervening body of two-pyroxene diorite. We infer that the ultramafic rocks and gabbronorite of the Emigrant Gap mafic complex crystallized from near-primitive arc basaltic to basaltic andesitic magmas at ~0.15–0.3 GPa, with estimated f O 2 of ≥FMQ +1 and dissolved H 2 O concentrations of only ~0.5–2 wt %. Notably, the Emigrant Gap composite pluton is distinct from other Mesozoic plutons in the Sierra Nevada batholith because of (1) its abundance of mafic and ultramafic rocks that crystallized from relatively primitive mafic melts and (2) the low inferred H 2 O concentrations of its parental magmas, indicated by a near absence of igneous amphibole and by the intermediate rather than calcic compositions of plagioclase. A Jurassic regional extension event probably accounts for the formation of relatively dry primitive arc magmas, as well as for their ascending to the upper crust.

California↗
USGS2025

Evaluating Laramide orogenesis via flexural basin response in the San Juan basin, New Mexico and Colorado

A challenge in interpreting the location, timing, and magnitude of ancient orogenic events is that ongoing uplift and erosion in the hinterlands often destroys much of the primary record of these events. However, basin-thickness patterns in the sedimentary record can provide complimentary evidence of uplift via flexural effects. Here, we deploy well-log correlation, isochores, basin modeling, flexural modeling, and subcrop mapping to evaluate the Late Cretaceous to Paleogene basin response to Laramide tectonism in the San Juan basin. A wedge of upper Campanian to Maastrichtian sedimentary rock thickens from 200 to 800 meters from southeast to northwest in the basin. This pattern can be successfully simulated via flexural modeling if we infer early Laramide uplift along the northwest basin flank that produced a 0.8 km high topographic load. The Laramide unconformity bounds the top of this Upper Cretaceous sedimentary wedge and truncates progressively older strata to the east, further supporting a westward tilt of the basin. The onset of Campanian Laramide flexure may have also contributed to the profound transgression from the upper Menefee Formation to the Lewis Shale. The Paleocene isochore map displays an approximately symmetrical pattern, with thickening towards the center of the basin. This suggests the possibility of competing flexural loads. The base Eocene structure indicates an asymmetric deep on the northeast flank of the basin, providing flexural evidence of contemporaneous uplift/loading of the Nacimiento uplift and Archuleta arch; this has been modeled as ~2.1 km load height. Both Cretaceous and Paleocene sedimentary wedges are narrow, suggesting low flexural rigidity; modeled effective elastic thicknesses (EET) are 20-30 km, comparable to estimates of modern EET for the region.

Colorado, New Mexico↗
USGS2025

No evidence for an active margin-spanning megasplay fault at the Cascadia Subduction Zone

It has been previously proposed that a megasplay fault within the Cascadia accretionary wedge, spanning from offshore Vancouver Island to Oregon, has the potential to slip during a future Cascadia subduction zone earthquake. This hypothetical fault has major implications for tsunami size and arrival times and is included in disaster-planning scenarios currently in use in the region. This hypothesis is evaluated in this study using CASIE21 deep-penetrating and U.S. Geological Survey high-resolution seismic reflection profiles. We map changes in wedge structural style and seismic character to identify the inner-outer wedge transition zone where a megasplay fault has been previously hypothesized to exist and evaluate evidence for active faulting within this zone. Our results indicate that there is not an active, through-going megasplay fault in Cascadia, but instead, the structure and activity of faulting at the inner-outer wedge transition zone is highly variable and segmented along strike, consistent with the segmentation of other physical and mechanical properties in Cascadia. Wedge sedimentation, plate dip, and subducting topography are proposed to play a major role in controlling megasplay fault development and evolution. Incorporating updated megasplay fault location, geometry, and activity into modeling of Cascadia earthquakes and tsunamis could help better constrain associated hazards.

British Columbia, Oregon, Washington↗
USGS2026

Long-term monotonic trends in water budget components in the contiguous United States: Insights from two hydrologic models

Characterizing changes to water availability for domestic, industrial, agricultural, and other uses is essential to support water management. To better quantify these changes, the U.S. Geological Survey and National Science Foundation National Center for Atmospheric Research produced two hydrologic models simulating water budget components from 1980 to 2021 over the contiguous United States (CONUS). Both hydrologic models were driven by a common atmospheric forcing dataset and aggregated to common spatial and temporal scales, which enables a novel evaluation of congruency between the models. We present annual and seasonal trends in six water budget components (precipitation, evapotranspiration, streamflow, groundwater recharge, soil saturation, and snow water equivalent) based on the Mann–Kendall test for monotonic trend and Theil-Sen slope estimate for the water year 1983–2021 period for ~86,000 catchments in CONUS. Additional components and metrics from our analysis pipeline are available in an associated published dataset, which contains more than 46 million trend results. The water budget trends showed broad agreement with prior observational and modeling studies that indicate increasing trends in the northeast and decreasing trends in southwestern CONUS. We found the seasonal variability in water budget trends was greatest in the southern, central, and northwest CONUS. These findings support integrated trend assessments when coupled with trends in water quality and use.

Contiguous United States↗
USGS1986

Regionalization of low-flow characteristics of Tennessee streams

Procedures for estimating 3-day 2-year, 3-day 10-year, 3-day 20-year, and 7-day 10-year low flows at ungaged stream sites in Tennessee are based on surface geology and drainage area size. One set of equations applies to west Tennessee streams, and another set applies to central and east Tennessee streams. The equations do not apply to streams where flow is significantly altered by activities of man. Standard errors of estimate of equations for west Tennessee are 24 to 32% and for central and east Tennessee 31 to 35%. Streamflow recession indexes, in days/log cycle, are used to account for effects of geology of the drainage basin on low flow of streams. The indexes in Tennessee range from 32 days/log cycle for clay and shale to 350 days/log cycle for gravel and sand, indicating different aquifer characteristics of the geologic units that sustain streamflows during periods of no surface runoff. Streamflow recession rate depends primarily on transmissivity and storage characteristics of the aquifers, and the average distance from stream channels to basin divides. Geology and drainage basin size are the most significant variables affecting low flow in Tennessee streams according to regression analyses. (Author 's abstract)

Tennessee↗
USGS2024

Seismic response comparison of a historical masonry church subject to real and simulated ground motions

In recent years, advanced numerical models and high-performance computing have facilitated the utilization of ground motion time series in the assessment of the non-linear dynamic behavior of historic masonry structures. Since recorded accelerograms can be sparse for specific analysis conditions, stochastic ground motion simulations have become a viable alternative to overcome this limitation. This study simulates the recorded acceleration time series of the Central Italy 2016 earthquake event at the closest station to the town of Macerata using a site-based stochastic approach. The simulated motions are seismologically evaluated using a goodness-of-fit method in terms of various intensity measures. The simulated records, in conjunction with real records, are used to study the non-linear dynamic behavior of San Filippo Neri church located in Macerata. The church of San Filippo represents an important example of Baroque religious architecture in central Italy, which was damaged and closed off to the public after the 2016 earthquake events. The construction was investigated with a vast diagnostic campaign which included on-site testing and dynamic identification tests. The collected data is used to calibrate the dynamic response of a three-dimensional finite element model of the church. The model is finally used to compare the non-linear seismic responses under real and simulated ground motions with the site recorded damage. The results of structural responses demonstrate a strong agreement between the real and simulated records, providing evidence to support the validation of the site-based stochastic simulation.

Macerata↗
USGS2026

Software to support remote sensing of river discharge based on critical flow theory

Water resource management requires accurate observations of streamflow but standard field methods for measuring river discharge ( Q ) are costly and can be hazardous for equipment and personnel. Remote sensing has become a viable alternative, but many image-based techniques require field data for calibration and depth and velocity can seldom be mapped with a single sensor. A new approach based on critical flow theory, in contrast, allows both of these attributes to be inferred from readily available image data. This technique only pertains to sites with standing waves, called undular hydraulic jumps (UHJs), but a recent investigation demonstrated its potential to provide accurate discharge estimates. This paper introduces software designed to facilitate Inferring Q from UHJs Identified in River Images (InQUIRI). The package includes modules for retrieving data from image servers, making the measurements of wavelength and width required to calculate discharge, inferring a representative wavelength from a profile digitized along a wave train, combining multiple estimates to obtain an ensemble median discharge, and assessing accuracy via comparison to gage records from the U.S. Geological Survey. By making these steps easier to implement, InQUIRI enables users to apply the workflow to a variety of UHJ-containing images. Accumulating more case studies, some successful and others less so, would help constrain the range of applicability of the critical flow approach and foster development of refined guidelines for selecting and measuring waves. The software described herein could play an important role in promoting informed use of this new technique for non-contact streamflow measurement.

Arizona, Colorado, New Mexico, Utah↗
USGS2024

Mapping the resistivity structure of Walker Ridge 313 in the Gulf of Mexico using the marine CSEM method

A marine controlled source electromagnetic (CSEM) campaign was carried out in the Gulf of Mexico to further develop marine electromagnetic techniques in order to aid the detection and mapping of gas hydrate deposits. Marine CSEM methods are used to obtain an electrical resistivity structure of the subsurface which can indicate the type of substance filling the pore space, such as gas hydrates which are more resistive. Results from the Walker Ridge 313 study (WR 313) are presented in this paper and compared with the Gulf of Mexico Gas Hydrate Joint Industry Project II (JIP2) logging while drilling (LWD) results and available seismic data. The hydrate, known to exist within sheeted sand deposits, is mapped as a resistive region in the two dimensional (2D) CSEM inversion models. This is consistent with the JIP2 LWD resistivity results. CSEM inversions that use seismic horizons provide more realistic results compared to the unconstrained inversions by providing sharp boundaries and architectural control on the location of the resistive and conductive regions in the CSEM model. The seismic horizons include: 1) the base of the gas hydrate stability zone (BGHSZ), 2) the top of salt, and 3) the top and bottom of a fine grained marine mud interval with near vertical hydrate filled fractures, to constrain the CSEM inversion model. The top of salt provides improved location for brines, water saturated salt, and resistive salt. Inversions of the CSEM data map the occurrence of a ‘halo’ of conductive brines above salt. The use of the BGHSZ as a constraint on the inversion helps distinguish between free gas and gas hydrate as well as gas hydrate and water saturated sediments.

Louisiana↗
USGS2025

Variable partitioning of lithium in rhyolitic melt during decompression and ascent

The partitioning behavior of Li in magmatic systems is increasingly being investigated due to the economic importance of Li in the transition to sustainable energy resources (e.g., batteries). However, at upper crustal pressures, it remains uncertain whether Li preferentially partitions into the vapor or liquid (brine) phase or remains in the silicate melt. This complicates our ability to determine where Li resides—silicate melt, minerals, or fluid phase—upon eruption, a crucial factor for understanding its postdepositional movement and concentration into a brine or volcano-sedimentary deposit. Here, we present a novel investigation into the behavior of Li within natural evolved melts during continuous magma decompression and ascent using melt embayments (open melt inclusions). Mineral-hosted melt embayments preserve records of the evolving composition of the exterior melt, including degassing pathways and ascent timescales, when paired with appropriate diffusion coefficients. Lithium concentration profiles were measured in quartz-hosted melt embayments from the rapidly quenched eruptive phases of five rhyolitic, caldera-forming eruptions to investigate the behavior of Li during magma decompression and ascent, where vapor partitioning and ascent dynamics were previously established by investigating H 2 O and CO 2 profiles. We find that in four systems, embayments contain lower interior Li concentrations than the coerupted melt inclusions; the fifth system contains the same Li concentrations in embayments and melt inclusions. However, many of these embayments contain gradients, with 84 % preserving Li enrichment near the melt-bubble interface, as compared to their interior concentration. We interpret these characteristics to represent two distinct stages of Li partitioning during magma decompression and ascent, in contrast to existing literature that proposes only one type of partitioning behavior. The first stage is interpreted as melt depletion of Li, likely driven by partitioning into an exsolved supercritical fluid phase, supported by the strong correlation between the extent of Li depletion and Cl concentration in the melt, as well as the decompression rate. This behavior then fundamentally shifts, where Li reenriches in the melt, postulated to be driven by the unmixing of the supercritical fluid phase at shallow pressures. For the one system that did not develop Li gradients through decompression, we attribute this to the lower values of Na and Cl in the melt, potentially inhibiting the partitioning of Li into a fluid phase. Importantly, the behavior of Li during decompression is not consistent within or between volcanic centers, highlighting the need for systematic experimental investigation in variable composition melts at pressures relevant to conduit dynamics. This knowledge would improve our ability to model Li profiles to understand magma decompression, and predict where Li resides (e.g., stored in volcanic glass, gas, or crystals) upon eruption prior to any later extraction.

Economic Geology↗
USGS2026

Estimation of parental abundance using hierarchical Bayesian modeling with data augmentation

Pedigree-based estimation methods leverage the fact that each offspring in a cohort is genotypically “marked” by its parents and represent a recent and promising toolset for estimating population dynamics. This includes pedigree accumulation estimators that model the “accumulation” of inferred unique parents within a given cohort to estimate parental abundance. Unlike close-kin mark-recapture approaches, which rely on intercohort comparisons, pedigree accumulation modeling can be completed solely using intracohort samples. This is particularly advantageous for semelparous species, where intercohort pairs are impossible and adult life stages can be difficult to sample without affecting their likelihood of successfully reproducing. Previous work has evaluated a range of estimators for such datasets, concluding that the non-parametric Chao estimator provides the most accurate and precise estimates for feasible levels of sampling effort. We used simulated data to evaluate an alternative estimator based on hierarchical modeling and data augmentation in a Bayesian framework. Results indicate that estimates from the hierarchical Bayesian estimator had comparable accuracy and better precision than both the previously tested Chao1 estimator and the improved iChao formulation across a range of sample sizes and sex ratios. Furthermore, the Bayesian estimator was far more robust to simulated errors in pedigree reconstruction, especially the presence of false negatives. Hierarchical Bayesian pedigree accumulation models can also provide additional insight into underlying reproductive ecology through their use of an explicit observation process, allowing for the incorporation or estimation of species- and population-specific reproductive dynamics. More broadly, the parametric nature of these models offers opportunities to efficiently pool information among datasets as well as to propagate uncertainty within more complex models.

Ecology and Evolution↗
USGS2024

Changes in soil erosion caused by wildfire: A conceptual biogeographic model

Soil erosion rates after wildfire are strongly controlled by intrinsic properties such as topography, weather, climate, soil, and vegetation. These landscape and hydroclimatic properties are important in determining post-fire erosion rates; however, their influence on post-fire erosion and their interaction with the intensity of a wildfire remains uncertain. A key limitation in resolving this uncertainty is the lack of conceptual models and frameworks for organising data related to the geomorphic sensitivity of landscapes to wildfire. Our aim is to develop a framework for consolidating understanding of post-fire erosion in the context of hydroclimatic conditions which contribute to system states, for example soil and vegetation properties, and wildfire regime. The framework is developed around a simple conceptual model where the change in erosion due to wildfire is a product of change in runoff generation and sediment supply, which is strongly related to landscape net primary productivity (NPP). We hypothesised that geomorphic sensitivity to wildfire should vary as a unimodal humped relationship across a gradient of NPP, peaking at an intermediate level. To develop this framework and to test the hypothesis, we first review intrinsic soil and vegetation properties related to the supply and transport of sediment from burned and unburned hillslopes. Net primary productivity is systematically related to these intrinsic properties because it integrates many processes involved in soil and vegetation development. Empirical data indicate a trend in the change in surface runoff generation with NPP after wildfire, peaking at an NPP of approximately 15 Mg C ha −1 y −1 . A simple model of fuel availability and soil heating are correlated with a similar “humped” trend in sediment supply. These results are consistent with our conceptual model, which indicates that sediment supply and runoff contribute towards a distinct peak in wildfire effects on erosion at an intermediate level of NPP. We propose that landscapes of intermediate NPP typically have the highest quantity of fuel available to burn, which cause large changes to the soil surface properties. Landscapes at intermediate NPP also tend to produce intrinsic soil and vegetation properties that promote erosion after wildfire. The interplay between these short and long-term landscape characteristics is strongest at intermediate levels of NPP. Our proposed biogeographic model of geomorphic sensitivity to wildfire was supported by erosion data from burned hillslope and zero-order catchments studies from a range fire-prone landscapes in Australia and North America. Our proposed conceptual model will help identify areas most vulnerable to post-fire erosion changes.

Geomorphology↗
USGS1983

Base flow and ground water in upper Sweetwater Valley, Tennessee

The upper Sweetwater Valley area has a flow system with complex interaction between surface and ground water. A water budget study indicated that during dry years approximately three-fourths of the annual flow to Sweetwater Creek may be derived from ground-water sources. Hydrograph analysis showed seasonal variation of recharge to the ground-water flow system. Streamflow records were analyzed to estimate the frequency of low flow of Sweetwater Creek at river mile 16.7, and indicated the lowest average flow for 1 day in 20 years to be about 5.1 cubic feet per second. Two periods of base-flow measurements of Sweetwater Creek identified channel reaches with significant gains and losses of streamflow. Base flow measurements also showed interbasin transfer of water among sub-basins of the valley. Major flow surpluses were associated with areas in which the majority of flow originated at a spring. Topographically low areas adjacent to the main stem of Sweetwater Creek generally have surplus flow. Topographically higher areas generally have deficient surface outflow unless significant spring flow occurs in the basin. Ground-water recharge occurs by water draining into sinkholes, faults, and fractures. Ground-water flow is regionally diffused across formation strikes from the topographically low areas unless the water is exposed to highly permeable formations or impervious formations. Ground water infiltrates the highly permeable formations and flows along strike. Ground water encountering impervious formations may discharge at small springs at the contact, or may reroute along the contact if the upgradient rock is sufficiently permeable or has well developed secondary porosity. Groundwater discharges to streams at innumerable seeps and at a few large springs. Areas of ground-water flow up-gradient of large springs are hypothesized as likely areas of significant ground-water reservoirs.

Tennessee↗
USGS2024

High resolution identification and quantification of diffuse deep groundwater discharge in mountain rivers using continuous boat-mounted helium measurements

Discharge of deeply sourced groundwater to streams is difficult to locate and quantify, particularly where both discrete and diffuse discharge points exist, but diffuse discharge is one of the primary controls on solute budgets in mountainous watersheds. The noble gas helium is a unique identifier of deep groundwater discharge because groundwater with long residence times is commonly enriched in helium. In this study, a portable mass spectrometer was used to measure longitudinal variation in dissolved helium concentrations in two mountainous rivers at high spatial resolution not feasible with traditional sampling techniques. Helium profiles were then simulated using a mass-balance model to quantify longitudinal variation in groundwater discharge to the receiving rivers. Results indicate helium concentrations were enriched by multiple orders of magnitude above atmospheric equilibrium in both rivers and that this persisted for up to 18 km below observed pulse inputs in the Colorado River. Helium mass-balance models match observed longitudinal patterns with the exception of sharp initial increases in helium observed in the rivers. Increased longitudinal groundwater discharge rates correspond to mapped geologic structures in both watersheds that likely transport deep geothermal water. Models show variable sensitivity to spatial assignment of input variables representing the groundwater source, illustrating the importance of collecting data from discrete groundwater discharges where possible. The methodology shows promise for field experiments designed to assess air–water exchange rates and to quantify total groundwater discharge from a combination of discrete and diffuse sources.

Colorado, Utah↗
USGS2026

King Rails in southeastern Arkansas are majority resident

King Rails ( Rallus elegans ), a declining species of secretive marsh bird, have a mixed migratory strategy, with some individuals migrating and some being year-round residents. Determining the distribution of migratory versus resident King Rails is critical to guiding habitat management and assessing risks to populations. We examined the migratory status of adult King Rails at the Choctaw West Wildlife Management Area in southeastern Arkansas, the only known site in Arkansas with multiple breeding pairs. Of the 23 adult King Rails we tagged in 2022 and 2023, two birds migrated away from the local area, one in each year. One migrated 145 km, and the other 450 km. Our findings suggest that the line between the migratory and resident range of Arkansas King Rails could be north of our site.

Arkansas↗
USGS2026

Tringa flavipes (Lesser Yellowlegs) from separate breeding sites subdivides the Prairie Pothole Region in space and time during southbound migration

Some staging regions support multiple groups of the same migratory species, each of which may use the region differently. Characterizing the ways, in which separate groups use such regions can therefore help to identify vulnerabilities during this sensitive period of the annual cycle. The Prairie Pothole Region (PPR) is a massive wetland complex in the northern Great Plains of North America used by ∼11 million shorebirds during migration. The PPR has been heavily modified by agriculture and is experiencing varied effects of global climate change, threatening the health of the shorebirds that rely on it. Here, we used 6 seasons of southbound tracking data of Tringa flavipes (Lesser Yellowlegs)—a long-distance migratory shorebird species with an estimated population decline of 63% over the last 4 decades—from 9 sites across their breeding range to explore differences in migratory behavior within this important staging region. We found that 75% of tracked individuals used the region during southbound migration, and T. flavipes from different breeding sites detoured 110–875 km from their most direct migratory route to access the PPR. Individuals that arrived later stayed longer and made more stops within the region than those that arrived early. Individuals originating from different breeding sites also displayed spatial and temporal segregation within the region: T. flavipes from southwest and central Alaska relied heavily on the northwestern PPR, while those from Canada used the central and southeastern portions of the PPR. Finally, timing of use varied among groups, but the southeastern PPR became increasingly important over the course of the southbound migratory window, as other wetlands likely dried out. Our study highlights the portions of the PPR of critical importance to migrating T. flavipes and the diversity of ways, in which different groups from within the same species can use a single staging region.

Alaska, Manitoba, Northwest Territories, Ontario, ↗
USGS2025

Central Valley Hydrologic Model version 2 (CVHM2): Decision support tool for groundwater and land subsidence management

The San Joaquin Valley (SJV) of California is one of the world’s most productive agricultural regions. Reliance on groundwater has led to some of the greatest rates of human-induced land subsidence in the world in the 20th century, as well as more recently. The United States Geological Survey (USGS) has recently developed an integrated surface–subsurface hydrologic model, the Central Valley Hydrologic Model 2 (CVHM2), that represents the major components of the hydrologic system of California’s Central Valley. In this study, CVHM2 was applied as a decision support tool while simulating various management strategies to mitigate the land subsidence caused by the extraction of groundwater. CVHM2 was extended through to 2073 and applied to simulate management scenarios in terms of three primary drivers and their impact on subsidence along the Delta–Mendota Canal (DMC), a critical piece of infrastructure in the western SJV. The drivers considered were agricultural water demands, managed aquifer recharge (MAR), and changes in future climate. The results show that future subsidence is most sensitive to water demands, second most sensitive to future changes in climate, and relatively insensitive to MAR when it is applied as a surface application in the western SJV. However, we demonstrate via proof-of-concept scenarios that the MAR is capable of arresting subsidence when implemented via injection below the Corcoran Clay Member of the Tulare Formation instead of as a surface application. We also examine the uncertainty that is the result of climate variability and how to use the tool to identify the most appropriate strategies to constrain future subsidence to acceptable levels.

California↗
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↗
USGS2024

Uncertainty and spatial correlation in station measurements for mb magnitude estimation

The body‐wave magnitude (⁠⁠) is a long‐standing network‐averaged, amplitude‐based magnitude used to estimate the magnitude of seismic sources from teleseismic observations. The U.S. Geological Survey National Earthquake Information Center (NEIC) relies on in its global real‐time earthquake monitoring mission. Although waveform modeling‐based moment magnitudes are the modern standard to characterize earthquake size, is important because (1) in many cases, waveform modeling is not possible (e.g., low signal‐to‐noise events), (2) is applicable over a broad range of magnitudes, ∼M 4–7, and (3) there is a many decades‐long history of estimating magnitudes. We use the NEIC Preliminary Determination of Epicenters earthquake catalog to investigate the uncertainty in NEIC station measurements. We show that measurements are spatially correlated, which can bias event ⁠, and we describe an empirical relation between this spatial correlation and station‐to‐station distance. We further describe an approach to mitigate bias from the spatial correlation. Accounting for the spatial covariance of observations can change the event from −0.15 to 0.07 units (10th to 90th percentile) for smaller events (⁠⁠). These smaller events have the largest standard deviations ranging from 0.05 to 0.15 units (10th to 90th percentile).

The Seismic Record↗