USGS Science⌕ Search

SEARCH · USGS Science

Results for “Mining, Metallurgy and Exploration”

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.

1,691 records · Page 69Linked to original sources
USGS2025

Mineralization processes at Escanaba Trough

Escanaba Trough off northern California is the southernmost and only sediment covered segment of the Gorda Ridge. Hydrothermal mineralization was first discovered in Escanaba Trough in 1985, and subsequent investigations included the delineation of the NESCA (northern Escanaba) and SESCA (southern Escanaba) sections of Escanaba Trough, as well as drilling by Ocean Drilling Program (ODP) Leg 169 in 1996. Preliminary research conducted in the 1980s distinguished two classes of sulfide mineral assemblages within the Escanaba Trough: one dominated by pyrrhotite (Fe 1-x S) and one containing ‘mixed sulfides’ including chalcopyrite (CuFeS 2 ) and sphalerite (ZnS). In 2022, we conducted a research expedition to the Escanaba Trough to more thoroughly characterize mineralization processes and consider preservation of precipitated minerals throughout hydrothermally active and inactive sections of Escanaba Trough, with the specific objective of characterizing the host minerals, and processes of enrichment for specific elements which are currently listed as critical elements and can also be toxic under certain conditions (Sb, As, Ba, Co, Ga, Ge, Te). The following will include initial results from that work.

California↗
USGS2026

Transit time modeling framework for predicting freshwater salinization in urban catchments

The salinity of inland freshwaters is rising globally, particularly in urban watersheds where winter road deicers are widely applied. Attributing stream salinity dynamics to specific sources and transport pathways remains challenging due to episodic salt inputs, engineered drainage, and strong coupling between hydrology and subsurface storage. We present a modeling framework that couples climate-driven deicer build-up and wash-off with transient transit time distribution theory to simulate salt transport through drainage, interflow, and groundwater pathways. Applied to an urban watershed in Northern Virginia (USA), the model reproduces ten years of high-frequency stream salinity measurements across daily-to-decadal timescales. The calibrated model implies an average deicer application of 206 tonnes Cl −1 yr −1 , or roughly one 20 kg bag of rock salt person −1 yr −1 when normalized by the 20,000 people living in the watershed. In winter months, higher infiltration routes a large fraction of snowmelt and deicers into shallow subsurface pathways, enhancing vadose-zone and interflow contributions to stream salinity. Limited subsurface storage capacity and seasonal hydrologic turnover flush excess chloride from the vadose zone and groundwater during subsequent summer storms. By linking climate-driven deicer inputs, hydrologic connectivity, and stream water age, the framework provides a transferable basis for diagnosing and managing freshwater salinization in urban watersheds.

Virginia↗
USGS2024

Potential corrosivity of untreated groundwater in Louisiana

Corrosive groundwater can cause lead, copper, and other metals to leach from pipes and plumbing fixtures in water distribution systems. Metals, if ingested, could lead to serious health implications to the nearly 2.9 million people in Louisiana who obtain their drinking water from groundwater sources. Four indices—the Langelier Saturation Index (LSI), Ryznar Stability Index (RSI), Puckorius Scaling Index (PSI), and the Potential to Promote Galvanic Corrosion (PPGC)—in addition to an analysis which normalized the results from the existing indices, the Combined Index (CI), were used to assess the corrosivity of groundwater in Louisiana and identify areas within eight major aquifers and aquifer systems with moderate to high corrosivity potential. The purpose of this study is to provide State and local governments, public water system managers, and the nearly 500,000 private well owners in Louisiana with information needed to manage drinking-water supplies and mitigate potential health risks related to leaching of metals from water pipes and fixtures. The average scores of untreated groundwater samples from approximately 375 wells by index are as follows: LSI, −1.28; RSI, 9.78; PSI, 9.34; and CI, 4.14. The PPGC does not produce a numerical score, but the total percentage of class counts can be used to assign a classification; overall, samples in Louisiana were classified as significant concern. The percentages of groundwater samples from wells classified as potentially corrosive, by index, are as follows: LSI, 53 percent; RSI, 94 percent; PSI, 81 percent; PPGC, 98 percent; and CI, 81 percent. The percentages of samples classified as indeterminate, by index, are as follows: LSI, 46 percent; RSI, 5 percent; PSI, 12 percent; PPGC, 0 percent; and CI, 18 percent.

Louisiana↗
USGS2025

Revisiting an enigma on California's north coast: The Mw6.5 Fickle Hill earthquake of 21 December 1954

Many earthquakes occur along the North Coast of California in the vicinity of the Mendocino Triple Junction (MTJ), where the Pacific, Gorda, and North American (NA) plates meet, and on the adjacent plate boundaries. The MTJ marks the nexus of the Mendocino and San Andreas faults with the Cascadia subduction zone (CSZ). Historically, most large earthquakes around the MTJ have been within the offshore Gorda plate and its subducted portion beneath the NA plate. North of the MTJ, active faults mapped in the NA plate are part of the CSZ fold‐and‐thrust belt. Although some events have been detected in the NA plate, no large historic events have been associated with mapped surface faults. The 21 December 1954 M w 6.5 earthquake in Humboldt County is one possible exception. Using published data from catalogs and articles, unpublished data from Berkeley’s archives, and S ‐ P times interpreted from two U.S. Coast and Geodetic Survey (USCGS) accelerometers, we determine a probability cloud for the earthquake’s hypocenter using NonLinLoc. The highest probability location lies beneath Fickle Hill just east of the city of Arcata, California, at 40.87° N, 124.03° W, and ∼11 km depth. Using P ‐wave polarities from Berkeley stations and the digitized waveforms from the accelerometers, we find that the focal mechanism most consistent with the data indicates thrust movement with strike, dip, and rake of 350°, 10°, and 90°, respectively, at a depth of 14 km. Given the depth uncertainties of both this event and the megathrust, this implies that the earthquake most likely took place on the subduction interface rather than on the mapped faults in the Mad River fault zone that trend 322° and dip to the northeast. The revisited intensity in the epicentral region also supports a location beneath Fickle Hill to the east of the city of Arcata, California.

California, Idaho, Nevada, Oregon, Washington↗
USGS2024

On-orbit calibration and performance of the EMIT imaging spectrometer

The Earth surface Mineral dust source InvesTigation (EMIT) is a remote visible to shortwave infrared (VSWIR) imaging spectrometer that has been operating onboard the International Space Station since July 2022. This article describes EMIT's on-orbit spectroradiometric calibration and validation. Accurate spectroscopy is vital to achieve consistent mapping results with orbital imaging spectrometers. EMIT takes a unique approach to this challenge, with just six optical elements, no shutter, and no onboard calibration systems. Its simple design focuses on uniformity and stability to enable vicarious spectroradiometric calibration. Our experiments demonstrate that this approach is successful, approaching the fidelity of manual field spectroscopy in some cases, and enabling new and more accurate products across diverse Earth science disciplines. EMIT achieves several notable firsts for an instrument of its class. It demonstrates successful on-orbit adjustments of Focal Plane Array (FPA) alignment with sub-micron precision. It offers spectral uniformity better than 98%. Optical artifacts in the measurement channels are at least three orders of magnitude below the primary solar-reflected surface signals. Its noise performance enables percent-level discrimination in the depths of mineral absorption features. In these aspects, EMIT satisfies the stringent performance needs for the next generation of VSWIR imaging spectrometers to observe the Earth's ecosystems, geology , and water resources.

Remote Sensing of Environment↗
USGS2024

Recent applications of the USGS National Crustal Model for Seismic Hazard Studies

The U.S. Geological Survey is developing the National Crustal Model (NCM) for seismic hazard studies to facilitate modeling site, path, and source components of seismic hazard across the conterminous United States. The NCM is composed of a 1km grid of geophysical profiles, extending from the Earth’s surface into the upper mantle. It is constructed from a threedimensional (3D) geologic framework and geophysical rules that use (1) a petrologic and mineral physics database; (2) a 3D temperature model; and (3) a calibrated rock type- and age-dependent porosity model. Parameters needed to estimate site response for existing ground motion models (GMMs), including the time-averaged velocity in the upper 30 meters (VS30), the depths to 1.0 and 2.5 km/s shear-wave velocity (Z1.0 and Z2.5), and sediment thickness, can be computed from the NCM. As GMMs continue to improve in the future, other metrics could also be extracted or derived from the NCM, such as fundamental period, site attenuation (ko), a fully frequency-dependent site response function, or 3D geophysical volumes for wavefield simulations. Application of the NCM may also benefit other aspects of seismic hazard analysis, including better accounting for path-dependent attenuation and geometric spreading, more accurate estimation of earthquake source properties such as hypocentral location and stress drop, and calculation of crustal strength profiles that inform estimates of the base of seismicity.

conterminous United States↗
USGS2026

Surface variable‐based machine learning for scalable arsenic prediction in undersampled areas

In the United States, private wells are not federally regulated, and many households do not test for Arsenic (As). Chronic exposure is linked with multiple health outcomes, and risk can change sharply over short distances and with well depth. Coarse maps or sparse sampling often miss exceedances. Most existing models operate at ∼1 km resolution and use groundwater chemistry or detailed geologic logs, which limits their use in undersampled areas where improved guidance is most needed. We overcome these limitations by developing a machine learning model for Minnesota, USA, that predicts As exposure risk using only surficial variables from remote sensing and global data sets. Variables related to surface water hydrology and geomorphology are selected based on mechanistic links that control redox conditions and As mobilization. Local training was essential, and surficial geology variables that are more sensitive to local conditions were needed to maximize model accuracy. The resulting complete model was sufficiently sensitive to generate accurate and detailed risk maps and depth profiles of As concentrations above the 10 μg/L maximum contaminant level. Accuracy depended on local training data density. We identified a training data density of 0.07 wells/km 2 as a practical target for stable county-level performance. Maps of exceedance probabilities highlight priority areas for testing that are particularly important in rural communities that have received less sampling. These results support public health action by guiding where to install wells and where to test them, how much new sampling is needed, and where treatment outreach is most urgent.

Minnesota↗
USGS2025

Earthquake recurrence estimates for northern Caribbean faults from combinatorial optimization

We use combinatorial optimization to find the optimal spatial distribution of random samples of earthquakes (≥6.5) that minimize the misfit in target slip rates for all faults in the northeast Caribbean, and we derive magnitude-frequency relationships with uncertainties for these faults. Slip rates for many faults are derived from GPS block models, not direct measurements, because of their underwater locations. Predicted recurrence rates for eastern Hispaniola and Puerto Rico Trench faults are 220-450 yr for M7 and 3-5 kyr for M8, with maximum feasible magnitude of M8.2. The most frequent earthquakes with magnitudes ≥7.0 are predicted on the large upper plate strike-slip faults, Enriquillo (EF) and Septentrional Fault, commensurate with the historical record. Calais et al. (2023) suggested that shortening in western Hispaniola is accommodated on the offshore Jérémie and onshore Malpasse faults north and south of EF, instead of on terrestrial faults in western Hispaniola and EF. Because of our system-modeling approach, such a configuration predicts less frequent earthquakes on EF and on western Hispaniola and Muertos convergent zones. Recurrence times of a few 100s yr for M6.7 earthquakes is predicted on the submerged faults in Mona Passage, and infrequent M>7 earthquakes are predicted on the Virgin Islands faults.

Lesser Antilles↗
USGS2024

Geomorphic map of the Umatilla River corridor, Oregon

This map portrays the distribution of landforms along the Umatilla River in northeastern Oregon and covers a corridor 127 kilometers long from the confluence of the Umatilla River with the Columbia River upstream to Meacham Creek. The map encompasses the valley bottom and extends about 1 kilometer up the adjoining hillslopes. Map data are intended to support water quality and fisheries enhancement efforts pursuant to the First Foods, a resource-management approach that focuses on traditionally gathered foods including water, fish, big game, roots, and berries and calls attention to the reciprocity between people and the foods upon which humans depend. The Umatilla River drains about 6,300 square kilometers on the northwest slope of the Blue Mountains in northeast Oregon. Most of the drainage basin is underlain by Miocene basalt flows of the Columbia River Basalt Group. Younger, weakly lithified, late Miocene and early Pliocene gravel deposits of local origin (for example, McKay Formation) are mapped in a few places. Upland surfaces are mantled with windborne silt (loess) correlative with deposits elsewhere known as the Palouse Formation. Surfaces below an elevation of about 340 meters were inundated repeatedly by large Pleistocene glacial outburst floods, most emanating from glacial Lake Missoula in western Montana. In backflooded areas such as the lower Umatilla River valley, Missoula floods deposited extensive slack-water silt. Areas mapped as open water, active channel and tie channel, flood basin, valley bottom, and modified land constitute the geomorphic floodplain: the area subject to occasional inundation by the Umatilla River. Deposits and landforms within the floodplain are inset into Missoula flood deposits and hence postdate the 20–15-kilo-annum Missoula floods. Some floodplain deposits are no more than a few centuries old, as indicated by substantial erosion and deposition during the Umatilla River flood of February 2020, the largest since systematic measurements began in October 1903. Deposits and landforms of the floodplain are transient features within the longer-term incision of the Umatilla River into mid-Miocene flood basalts and younger gravel of the McKay Formation.

Oregon↗
USGS2025

Risk implications of Poisson assumptions and declustering inferred from a fully time-dependent earthquake forecast

We use the Third Uniform California Earthquake Rupture Forecast Epidemic Type Aftershock Sequence model, which is fully time-dependent in terms of including spatiotemporal clustering, to evaluate the effects of the Poisson assumption and declustering algorithms on statewide loss exceedance curves. The model is simulation based, meaning it produces synthetic catalogs that exhibit realistic behavior with respect to aftershocks and multi-fault earthquakes. A Poisson version of the model was constructed by randomizing event times, and the influence of two declustering algorithms was examined as well. We demonstrate that the probability of one-or-more loss exceedances (occurrence exceedance probability) is greater for the Poisson model because it has fewer seismically quiet time windows. The discrepancy between dollar loss estimates with a given exceedance probability is up to a factor of 32% but varies depending on the loss threshold (the x-axis value) and the forecast duration (we examined a range between 24 h and 50 years, with the discrepancy for the latter being negligible). We discuss how the one-or-more loss exceedance metric is questionable because it ignores all but the maximum loss experienced in each timeframe. An alternative metric based on total aggregate loss in each time window (aggregate exceedance probability) was therefore also examined, for which the Poisson model again implies higher risk at intermediate losses but lower risk at higher losses (because large, triggered events now contribute to total aggregate losses for the fully time-dependent model). We also argue that declustering is not a scientifically justifiable way to deal with full time dependence, in agreement with a chorus from other recent studies. It is difficult to draw generally applicable conclusions from our study, in part because application specific details will likely be important, but our results highlight how full time dependence can be reckoned with once authoritative forecast models are made available.

California↗
USGS2025

Structural controls on splay fault rupture dynamics during Cascadia megathrust earthquakes

Great subduction earthquakes ( M w ≥ 8.0) can generate devastating tsunamis by rapidly displacing the seafloor and overlying water column. These potentially tsunamigenic seafloor offsets result from coseismic fault slip and deformation beneath or within the accretionary wedge. The mechanics of these shallow rupture phenomena and their dependence on subduction zone properties remain unresolved, partly due to the sparsity of offshore observations of shallow megathrust earthquake deformation. Here, we analyze how offshore structure influences shallow rupture mechanics and slip partitioning using 3D dynamic earthquake simulations of the Cascadia subduction zone (CSZ) megathrust with and without variably dipping seaward- or landward-vergent splay faults in the wedge that sole into the megathrust. Resulting tradeoffs between splay and megathrust slip reveal structural controls on rupture partitioning, with greater splay slip leading to less shallow megathrust slip updip. Gently dipping and seaward-vergent splays host more slip than those with steeper, landward-vergent splays. To isolate the underlying mechanisms, we compare models with Andersonian and plunging principal stresses. Results suggest distinct static and dynamic processes control the dip- and vergence-dependence of splay rupture: static (mis)alignment relative to far-field tectonic loading favors slip on more optimally oriented, shallowly dipping splay faults. In contrast, dynamic stress interactions of an updip-propagating megathrust rupture front with the free surface and potential branch faults favor forward branching onto seaward-vergent splays and inhibit backward branching onto landward-vergent splays. Resulting seafloor displacements suggest splay fault structure may influence coseismic tsunami source processes, highlighting the importance of dynamically viable rupture scenarios in subduction hazard assessments.

Cascadia subduction zone↗
USGS2026

Simulation of groundwater flow to evaluate hydrogeologic controls on a PFAS plume, Coakley Landfill Superfund Site, Rockingham County, New Hampshire

Per- and polyfluoroalkyl substances (PFAS), including perfluorooctanoic acid (PFOA) and perfluorooctanesulfonic acid (PFOS), have been detected at combined concentrations above 2,000 nanograms per liter (ng/L) at groundwater seep locations near the Coakley Landfill Superfund site, in North Hampton, New Hampshire. The landfill was active from 1972 to 1985. An impermeable cap was placed on the landfill in 1998. The adjacent area to the Coakley Landfill has many water supply wells, and transport of PFAS compounds to the wells is a concern. Fracture anisotropy in the underlying bedrock aquifer complicates the understanding of PFAS transport because groundwater preferentially travels along fractures that may not align with the prevailing groundwater flow direction. In 2018, the U.S Environmental Protection Agency and the U.S. Geological Survey began an investigation of the groundwater flow from the Coakley Landfill site. This report describes the modification of a numerical groundwater-flow model for the local area around the Coakley Landfill and summarizes findings of the investigation. In addition, this report includes a brief description of PFOA and PFOS occurrence, a discussion of model construction, evaluation of model performance through calibration, and discussion of simulation results for two periods (before and after capping). Limitations are also discussed. Results show that simulated groundwater flow moves from the Coakley Landfill to the west and north. Advective transport modeling using particle tracking shows that groundwater from the landfill discharges primarily to streams to the west and north, and a small amount is transported to distal wells. Dilution of contaminants through advection and dispersion likely plays a role in whether PFAS compounds from the landfill will be detected above laboratory reporting levels at distal wells.

EarthArXiv↗
USGS2025

A framework for understanding the effects of subsurface agricultural drainage on downstream flows

Understanding controls on streamflow volume and magnitude is important to water resource management applications, such as critical water and transportation structure design and floodplain mapping. Changes in land use and agricultural practices, such as subsurface agricultural drainage, may be contributing to changes in streamflow characteristics. Subsurface agricultural drainage, also known as tile drainage, is the practice of installing drains in the subsurface of agricultural fields to improve productivity. Because of the complex interactions between subsurface drainage systems, precipitation, local soil conditions, and land management practices, it is difficult to determine how subsurface agricultural drainage affects downstream flow. Previously developed subsurface agricultural drainage conceptual models under dry, saturated, and winter conditions are summarized, and current literature on the effects of subsurface agricultural drainage on downstream flows, focusing on peak flow, non-event flow, and total flow to develop frameworks for discussing these systems is compiled. The effects that subsurface drainage has on hydrologic systems are expected to vary by site and are seasonally based on system design, soil type, moisture conditions, precipitation characteristics, and land conditions. Subsurface drainage can affect the magnitude of peak flow by converting surface runoff from a storm event to subsurface runoff. By increasing hydrologic connectivity of a catchment, subsurface drainage can increase non-event flow or the flow between two storm events, typically dependent on lateral flow through the subsurface and groundwater. Theoretically, by diverting water from groundwater recharge or by reducing water available for evapotranspiration, subsurface drainage may increase the total volume of flow. Precipitation changes may increase infiltration, excess overland flow, and flood risk regardless of the presence or absence of subsurface drainage.

Illinois, Iowa, Michigan, Minnesota, Missouri, Mon↗
USGS2025

Assessment of density pattern retention of generalized data for 1:100,000-scale United States topographic maps

Cartographic generalization reduces the complexity of geographic data to produce legible, smaller-scale displays that retain essential information and logical geographic patterns. Generalization is a vital process in topographic map production. An important challenge in this process is managing and evaluating consistency across scale in the density and spatial distribution of map features such as buildings, roads, streams, water bodies, and elevation contours. Density patterns in these features reflect underlying physiographic conditions, which include factors such as bedrock geology, tectonics, climate, and landforms. Assessments of an acceptable level of change in feature density patterns are critical to ensuring the readability, usability, and accuracy of generalized maps and data. Preserving realistic density patterns across mapping scales also supports sustainable development goals in cartography, by helping to prioritize and communicate the relative reliability of geospatial data at specific scales.

Conference Paper↗
USGS2025

Bayesian mapping of regionally grouped, sparse, univariate earth science data

Some earth science data are naturally grouped by region, and it is often desirable to map these data by region. However, if there are only a few samples within each region, then the map should be smoothed in an appropriate way to mitigate the problems that arise from having only a few samples. A smoothing algorithm based on a Bayesian hierarchical model is developed and presented in this report. This algorithm has several features that make it especially suitable for mapping earth science data: it can account for measurements that are censored, it can process multiple datasets with different measurement errors and different censoring thresholds, and it can calculate the uncertainty in any statistic that is mapped. The algorithm is demonstrated by mapping gold concentrations that are measured in streambed sediments in the Taylor Mountains quadrangle in southwestern Alaska.

Alaska↗
USGS2024

Geologic input databases for the 2025 Puerto Rico – U.S. Virgin Islands National Seismic Hazard Model update: Crustal faults component

The last National Seismic Hazard Model (NSHM) for Puerto Rico and the U.S. Virgin Islands (PRVI) was published in 2003. In advance of the 2025 PRVI NSHM update, we created three geologic input databases to summarize new onshore and offshore fault source information in the northern Caribbean region between 62°–70° W and 16°–21° N. These databases, of fault sections, fault‐zone polygons, and geologic estimates of fault activity (fault‐slip rate and earthquake recurrence intervals) at specific sites, document updates to fault parameters used in prior seismic hazard models in PRVI. Fault sources were reviewed from published studies since 2003, which document substantial changes to the understanding of fault location, geometry, or activity. New fault section sources were added for features that meet the criteria of (1) length ≥7 km, (2) unequivocal evidence of recurrent tectonic Quaternary activity, and (3) documentation that is publicly available in a peer‐reviewed source. In addition, we revised several broad areal sources, such as the Mona and Anegada extensional zones. The 2003 model included three fault sections and two fault‐zone polygons (areal sources). These databases include 35 fault sections, 6 fault‐zone polygons, and 51 earthquake geology sites. To characterize fault activity rates, slip‐rate bins were assigned based on landscape expression and paleoseismic trench observations for faults without published slip‐rate sites. Additional fault sources were evaluated but not included in these databases due to a lack of published information about fault location, geometry, or recurrent Quaternary activity. The PRVI NSHM 2025 geologic input databases describe crustal faulting; the geometries and coupling of Puerto Rico subduction zone and Muertos Trough models are considered in a separate database. Updates to the fault sections, fault‐zone polygons, and earthquake geology databases can help inform the location and recurrence rate of damaging earthquakes in the PRVI NSHM implementation.

Puerto Rico, U.S. Virgin Islands↗
USGS2026

Cnidarian–algal partnerships structure bacterial communities during strobilation in Cassiopea xamachana

Cnidarian–algal (Symbiodiniaceae) symbioses rely on complex interactions among the cnidarian host, algal symbionts, and associated bacterial communities. In the upside-down jellyfish Cassiopea xamachana , the polyp-to-medusa transition (strobilation) requires the establishment of symbiosis with Symbiodiniaceae algal partners, yet bacterial community dynamics during this developmental process remain unknown. Here, we experimentally induced symbiosis in aposymbiotic polyps using four algal treatments: xenic Symbiodinium microadriaticum (native symbiont), xenic Breviolum minutum , antibiotic-treated B. minutum , and a photosynthetically impaired B. minutum mutant. We combined 16S rRNA gene sequencing with measurements of photosynthetic efficiency, asexual budding, and algal surface N-glycan profiles to characterize holobiont assembly during symbiosis onset and strobilation. Algal treatment structured bacterial communities in both algal cultures and polyp tissues. Our analyses identified a set of amplicon sequence variants that consistently distinguished strobilating polyps from non-strobilating aposymbiotic and mutant polyps, in addition to potential bacterial biomarkers associated with successful metamorphosis. Strobilation was associated with the enrichment of bacterial communities putatively involved in sulfur and nitrogen cycling, whereas non-strobilating aposymbiotic and mutant polyps were characterized by opportunistic bacteria and increased community variability. Together, these results reveal coordinated changes in algal physiology, surface glycan profiles, and bacterial community structure associated with successful strobilation in C. xamachana and support a model in which tripartite host–alga–bacteria interactions influence cnidarian life stage transitions.

ISME Communications↗
USGS2024

Classification of lakebed geologic substrate in autonomously collected benthic imagery using machine learning

Mapping benthic habitats with bathymetric, acoustic, and spectral data requires georeferenced ground-truth information about habitat types and characteristics. New technologies like autonomous underwater vehicles (AUVs) collect tens of thousands of images per mission making image-based ground truthing particularly attractive. Two types of machine learning (ML) models, random forest (RF) and deep neural network (DNN), were tested to determine whether ML models could serve as an accurate substitute for manual classification of AUV images for substrate type interpretation. RF models were trained to predict substrate class as a function of texture, edge, and intensity metrics (i.e., features) calculated for each image. Models were tested using a manually classified image dataset with 9-, 6-, and 2-class schemes based on the Coastal and Marine Ecological Classification Standard (CMECS). Results suggest that both RF and DNN models achieve comparable accuracies, with the 9-class models being least accurate (~73–78%) and the 2-class models being the most accurate (~95–96%). However, the DNN models were more efficient to train and apply because they did not require feature estimation before training or classification. Integrating ML models into benthic habitat mapping process can improve our ability to efficiently and accurately ground-truth large areas of benthic habitat using AUV or similar images.

Michigan, Wisconsin↗