USGS Science⌕ Search

SEARCH · USGS Science

Results for “Algorithms”

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.

At least 1,063 records · Page 59Linked to original sources
USGS2000

Landsat 7 on-orbit geometric calibration and performance

The Landsat 7 Image Assessment System was developed to characterize and calibrate the radiometric and geometric performance of the Landsat 7 Enhanced Thematic Mapper Pius (ETM+) instrument. Algorithms and software assess the geometric performance of the Landsat 7 spacecraft and ETM+ sensor system and perform geometric calibration by estimating sensor and spacecraft geometric parameters. Following the initial on-orbit calibration, performed during the Landsat 7 on- orbit initialization and verification period, all geometric performance goals were met. Geometric characterization and calibration activities will continue for the life of the Landsat 7 mission.

Conference Paper↗
USGS2002

Absolute irradiance of the Moon for on-orbit calibration

The recognized need for on-orbit calibration of remote sensing imaging instruments drives the ROLO project effort to characterize the Moon for use as an absolute radiance source. For over 5 years the ground-based ROLO telescopes have acquired spatially-resolved lunar images in 23 VNIR (Moon diameter ???500 pixels) and 9 SWIR (???250 pixels) passbands at phase angles within ??90 degrees. A numerical model for lunar irradiance has been developed which fits hundreds of ROLO images in each band, corrected for atmospheric extinction and calibrated to absolute radiance, then integrated to irradiance. The band-coupled extinction algorithm uses absorption spectra of several gases and aerosols derived from MODTRAN to fit time-dependent component abundances to nightly observations of standard stars. The absolute radiance scale is based upon independent telescopic measurements of the star Vega. The fitting process yields uncertainties in lunar relative irradiance over small ranges of phase angle and the full range of lunar libration well under 0.5%. A larger source of uncertainty enters in the absolute solar spectral irradiance, especially in the SWIR, where solar models disagree by up to 6%. Results of ROLO model direct comparisons to spacecraft observations demonstrate the ability of the technique to track sensor responsivity drifts to sub-percent precision. Intercomparisons among instruments provide key insights into both calibration issues and the absolute scale for lunar irradiance.

Conference Paper↗
USGS2002

Advances in directional borehole radar data analysis and visualization

The U.S. Geological Survey is developing a directional borehole radar (DBOR) tool for mapping fractures, lithologic changes, and underground utility and void detection. An important part of the development of the DBOR tool is data analysis and visualization, with the aim of making the software graphical user interface (GUI) intuitive and easy to use. The DBOR software system consists of a suite of signal and image processing routines written in Research Systems' Interactive Data Language (IDL). The software also serves as a front-end to many widely accepted Colorado School of Mines Center for Wave Phenomena (CWP) Seismic UNIX (SU) algorithms (Cohen and Stockwell, 2001). Although the SU collection runs natively in a UNIX environment, our system seamlessly emulates a UNIX session within a widely used PC operating system (MicroSoft Windows) using GNU tools (Noer, 1998). Examples are presented of laboratory data acquired with the prototype tool from two different experimental settings. The first experiment imaged plastic pipes in a macro-scale sand tank. The second experiment monitored the progress of an invasion front resulting from oil injection. Finally, challenges to further development and planned future work are discussed.

Conference Paper↗
USGS2003

Modeling the Radiance of the Moon for On-orbit Calibration

The RObotic Lunar Observatory (ROLO) project has developed radiometric models of the Moon for disk-integrated irradiance and spatially resolved radiance. Although the brightness of the Moon varies spatially and with complex dependencies upon illumination and viewing geometry, the surface photometric properties are extremely stable, and therefore potentially knowable to high accuracy. The ROLO project has acquired 5+ years of spatially resolved lunar images in 23 VNIR and 9 SWIR filter bands at phase angles up to 90??. These images are calibrated to exoatmospheric radiance using nightly stellar observations in a band-coupled extinction algorithm and a radiometric scale based upon observations of the star Vega. An effort is currently underway to establish an absolute scale with direct traceability to NIST radiometric standards. The ROLO radiance model performs linear fitting of the spatially resolved lunar image data on an individual pixel basis. The results are radiance images directly comparable to spacecraft observations of the Moon. Model-generated radiance images have been produced for the ASTER lunar view conducted on 14 April 2003. The radiance model is still experimental - simplified photometric functions have been used, and initial results show evidence of computational instabilities, particularly at the lunar poles. The ROLO lunar image dataset is unique and extensive and presents opportunities for development of novel approaches to lunar photometric modeling.

Conference Paper↗
USGS2004

Digital elevation extraction from multiple MTI data sets

The Digital Elevation Model (DEM) extraction process traditionally uses a stereo pair of aerial photographs that are sequentially captured using an airborne metric camera. Standard DEM extraction techniques have been naturally extended to utilize satellite imagery. However, the particular characteristics of satellite imaging can cause difficulties in the DEM extraction process. The ephemeris of the spacecraft during the collects, with respect to the ground test site, is the most important factor in the elevation extraction process. When the angle of separation between the stereo images is small, the extraction process typically produces measurements with low accuracy. A large angle of separation can cause an excessive number of erroneous points in the output DEM. There is also a possibility of having occluded areas in the images when drastic topographic variation is present, making it impossible to calculate elevation in the blind spots. The use of three or more images registered to the same ground area can potentially reduce these problems and improve the accuracy of the extracted DEM. The pointing capability of the Multispectral Thermal Imager (MTI) allows for multiple collects of the same area to be taken from different perspectives. This functionality of MTI makes it a good candidate for the implementation of DEM extraction using multiple images for improved accuracy. This paper describes a project to evaluate this capability and the algorithms used to extract DEMs from multi-look MTI imagery.

Conference Paper↗
USGS2004

Littoral Assessment of Mine Burial Signatures (LAMBS) buried land mine/background spectral signature analyses

The objective of the Office of Naval Research (ONR) Rapid Overt Reconnaissance (ROR) program and the Airborne Littoral Reconnaissance Technologies (ALRT) project's LAMBS effort is to determine if electro-optical spectral discriminants exist that are useful for the detection of land mines in littoral regions. Statistically significant buried mine overburden and background signature data were collected over a wide spectral range (0.35 to 14 ??m) to identify robust spectral features that might serve as discriminants for new airborne sensor concepts. LAMBS has expanded previously collected databases to littoral areas - primarily dry and wet sandy soils - where tidal, surf, and wind conditions can severely modify spectral signatures. At AeroSense 2003, we reported completion of three buried mine collections at an inland bay, Atlantic and Gulf of Mexico beach sites.1 We now report LAMBS spectral database analyses results using metrics which characterize the detection performance of general types of spectral detection algorithms. These metrics include mean contrast, spectral signal-to-clutter, covariance, information content, and spectral matched filter analyses. Detection performance of the buried land mines was analyzed with regard to burial age, background type, and environmental conditions. These analyses considered features observed due to particle size differences, surface roughness, surface moisture, and compositional differences.

Conference Paper↗
USGS2006

Using self-organizing maps to determine observation threshold limit predictions in highly variant data

A significant data quality challenge for highly variant systems surrounds the limited ability to quantify operationally reasonable limits on the data elements being collected and provide reasonable threshold predictions. In many instances, the number of influences that drive a resulting value or operational range is too large to enable physical sampling for each influencer, or is too complicated to accurately model in an explicit simulation. An alternative method to determine reasonable observation thresholds is to employ an automation algorithm that would emulate a human analyst visually inspecting data for limits. Using the visualization technique of self-organizing maps (SOM) on data having poorly understood relationships, a methodology for determining threshold limits was developed. To illustrate this approach, analysis of environmental influences that drive the abundance of a target indicator species (the pink shrimp, Farfantepenaeus duorarum) provided a real example of applicability. The relationship between salinity and temperature and abundance of F. duorarum is well documented, but the effect of changes in water quality upstream on pink shrimp abundance is not well understood. The highly variant nature surrounding catch of a specific number of organisms in the wild, and the data available from up-stream hydrology measures for salinity and temperature, made this an ideal candidate for the approach to provide a determination about the influence of changes in hydrology on populations of organisms.

Conference Paper↗
USGS2007

Consistency of L4 TM absolute calibration with respect to the L5 TM sensor based on near-simultaneous image acquisition

The Landsat archive provides more than 35 years of uninterrupted multispectral remotely sensed data of Earth observations. Since 1972, Landsat missions have carried different types of sensors, from the Return Beam Vidicon (RBV) camera to the Enhanced Thematic Mapper Plus (ETM+). However, the Thematic Mapper (TM) sensors on Landsat 4 (L4) and Landsat 5 (L5), launched in 1982 and 1984 respectively, are the backbone of an extensive archive. Effective April 2, 2007, the radiometric calibration of L5 TM data processed and distributed by the U.S. Geological Survey (USGS) Center for Earth Resources Observation and Science (EROS) was updated to use an improved lifetime gain model, based on the instrument's detector response to pseudo-invariant desert site data and cross-calibration with the L7 ETM+. However, no modifications were ever made to the radiometric calibration procedure of the Landsat 4 (L4) TM data. The L4 TM radiometric calibration procedure has continued to use the Internal Calibrator (IC) based calibration algorithms and the post calibration dynamic ranges, as previously defined. To evaluate the "current" absolute accuracy of these two sensors, image pairs from the L5 TM and L4 TM sensors were compared. The number of coincident image pairs in the USGS EROS archive is limited, so the scene selection for the cross-calibration studies proved to be a challenge. Additionally, because of the lack of near-simultaneous images available over well-characterized and traditionally used calibration sites, alternate sites that have high reflectance, large dynamic range, high spatial uniformity, high sun elevation, and minimal cloud cover were investigated. The alternate sites were identified in Yuma, Iraq, Egypt, Libya, and Algeria. The cross-calibration approach involved comparing image statistics derived from large common areas observed eight days apart by the two sensors. This paper summarizes the average percent differences in reflectance estimates obtained between the two sensors. The work presented in this paper is a first step in understanding the current performance of L4 TM absolute calibration and potentially serves as a platform to revise and improve the radiometric calibration procedures implemented for the processing of L4 TM data.

Conference Paper↗
USGS2008

MODIS and SeaWIFS on-orbit lunar calibration

The Moon plays an important role in the radiometric stability monitoring of the NASA Earth Observing System's (EOS) remote sensors. The MODIS and SeaWIFS are two of the key instruments for NASA's EOS missions. The MODIS Protoflight Model (PFM) on-board the Terra spacecraft and the MODIS Flight Model 1 (FM1) on-board the Aqua spacecraft were launched on December 18, 1999 and May 4, 2002, respectively. They view the Moon through the Space View (SV) port approximately once a month to monitor the long-term radiometric stability of their Reflective Solar Bands (RSB). SeaWIFS was launched on-board the OrbView-2 spacecraft on August 1, 1997. The SeaWiFS lunar calibrations are obtained once a month at a nominal phase angle of 7??. The lunar irradiance observed by these instruments depends on the viewing geometry. The USGS photometric model of the Moon (the ROLO model) has been developed to provide the geometric corrections for the lunar observations. For MODIS, the lunar view responses with corrections for the viewing geometry are used to track the gain change for its reflective solar bands (RSB). They trend the system response degradation at the Angle Of Incidence (AOI) of sensor's SV port. With both the lunar observation and the on-board Solar Diffuser (SD) calibration, it is shown that the MODIS system response degradation is wavelength, mirror side, and AOI dependent. Time-dependent Response Versus Scan angle (RVS) Look-Up Tables (LUT) are applied in MODIS RSB calibration and lunar observations play a key role in RVS derivation. The corrections provided by the RVS in the Terra and Aqua MODIS data from the 412 nm band are as large as 16% and 13%, respectively. For SeaWIFS lunar calibrations, the spacecraft is pitched across the Moon so that the instrument views the Moon near nadir through the same optical path as it views the Earth. The SeaWiFS system gain changes for its eight bands are calibrated using the geometrically-corrected lunar observations. The radiometric corrections to the SeaWiFS data, after more than ten years on orbit, are 19% at 865 nm, 8% at 765 nm, and 1-3% in the other bands. In this report, the lunar calibration algorithms are reviewed and the RSB gain changes observed by the lunar observations are shown for all three sensors. The lunar observations for the three instruments are compared using the USGS photometric model. The USGS lunar model facilitates the cross calibration of instruments with different spectra bandpasses whose measurements of the Moon differ in time and observing geometry.

Conference Paper↗
USGS2012

Spectroscopic remote sensing for material identification, vegetation characterization, and mapping

Identifying materials by measuring and analyzing their reflectance spectra has been an important procedure in analytical chemistry for decades. Airborne and space-based imaging spectrometers allow materials to be mapped across the landscape. With many existing airborne sensors and new satellite-borne sensors planned for the future, robust methods are needed to fully exploit the information content of hyperspectral remote sensing data. A method of identifying and mapping materials using spectral feature analyses of reflectance data in an expert-system framework called MICA (Material Identification and Characterization Algorithm) is described. MICA is a module of the PRISM (Processing Routines in IDL for Spectroscopic Measurements) software, available to the public from the U.S. Geological Survey (USGS) at http://pubs.usgs.gov/of/2011/1155/. The core concepts of MICA include continuum removal and linear regression to compare key diagnostic absorption features in reference laboratory/field spectra and the spectra being analyzed. The reference spectra, diagnostic features, and threshold constraints are defined within a user-developed MICA command file (MCF). Building on several decades of experience in mineral mapping, a broadly-applicable MCF was developed to detect a set of minerals frequently occurring on the Earth's surface and applied to map minerals in the country-wide coverage of the 2007 Afghanistan HyMap data set. MICA has also been applied to detect sub-pixel oil contamination in marshes impacted by the Deepwater Horizon incident by discriminating the C-H absorption features in oil residues from background vegetation. These two recent examples demonstrate the utility of a spectroscopic approach to remote sensing for identifying and mapping the distributions of materials in imaging spectrometer data.

Conference Paper↗
USGS2025

Long-term dynamics of earthquake swarms in the Yellowstone caldera

The factors controlling the spatial distribution and temporal evolution of earthquake swarms in volcanic systems remain unclear. We leverage leading-edge deep learning algorithms and a detailed three-dimensional velocity model to construct a 15-year high-resolution earthquake catalog of the Yellowstone caldera region. More than half of the region’s earthquakes are clustered into swarm-like families characterized by episodes of hypocenter expansion and migration. Adjacent earthquake swarms, separated by long quiescent periods, are found to be a dominant feature. We suggest that these swarms are controlled by the interplay between slowly diffusing aqueous fluids and rapid episodic fluid injections, which may result from the breaking of permeability seals. Our analyses also indicate that clustered seismicity beneath the caldera occurs on relatively immature, rougher fault structures, compared to more planar faults outside. Our results provide additional context for understanding seismicity in hydrothermal systems, highlighting the key role played by long-term fluid diffusion processes in driving the occurrence of earthquake swarms.

Idaho, Montana, Wyoming↗
USGS2005

Pattern-oriented modeling of agent-based complex systems: Lessons from ecology

Agent-based complex systems are dynamic networks of many interacting agents; examples include ecosystems, financial markets, and cities. The search for general principles underlying the internal organization of such systems often uses bottom-up simulation models such as cellular automata and agent-based models. No general framework for designing, testing, and analyzing bottom-up models has yet been established, but recent advances in ecological modeling have come together in a general strategy we call pattern-oriented modeling. This strategy provides a unifying framework for decoding the internal organization of agent-based complex systems and may lead toward unifying algorithmic theories of the relation between adaptive behavior and system complexity.

Science↗
USGS2020

3D fault architecture controls the dynamism of earthquake swarm

The vibrant evolutionary patterns made by earthquake swarms are incompatible with standard, effectively two-dimensional (2D) models for general fault architecture. We leverage advances in earthquake monitoring with a deep-learning algorithm to image a fault zone hosting a 4-year-long swarm in southern California. We infer that fluids are naturally injected into the fault zone from below and diffuse through strike-parallel channels while triggering earthquakes. A permeability barrier initially limits up-dip swarm migration but ultimately is circumvented. This enables fluid migration within a shallower section of the fault with fundamentally different mechanical properties. Our observations provide high-resolution constraints on the processes by which swarms initiate, grow, and arrest. These findings illustrate how swarm evolution is strongly controlled by 3D variations in fault architecture.

Science↗
USGS2025

An empirical test of maximum depositional age through paired LA-ICP-MS and CA-ID-TIMS detrital zircon analysis with implications for the “Nevadan orogeny” in the Sierra Nevada foothills, California, U.S.A.

The interpretation of maximum depositional age (MDA) from U-Pb detrital zircon data acquired via laser ablation–inductively coupled plasma–mass spectrometry (LA-ICP-MS) or secondary ion mass spectrometry (SIMS) is now routine; however, to date only a few studies have presented tests of such MDAs by subsequently analyzing a subset of the same detrital zircon grains with the more accurate and precise chemical abrasion–isotope dilution–thermal ionization mass spectrometry (CA-ID-TIMS) method. We first generated LA-ICP-MS dates and MDAs from three turbidite sandstone samples of the Late Jurassic Mariposa Formation (Sierra Nevada foothills, California) containing large proportions of young detrital zircon grains (at the time of sediment deposition). We then removed 4–5 of the apparent youngest grains from the epoxy mount and analyzed them with CA-ID-TIMS. Despite a lack of low Th/U or high U (ppm) characteristics that might indicate U-Pb system disturbance, all LA-ICP-MS dates are younger than corresponding CA-ID-TIMS dates, on the same grain, by up to 11.8 m.y. when considered as point estimates. Only two out of 13 paired analyses overlap within error at 95% confidence. We interpret MDAs from our CA-ID-TIMS dates to be the youngest CA-ID-TIMS dates (YTDs). LA-ICP-MS MDAs based on fewer dates from the young tail of the youngest date distribution perform the worst (i.e., furthest from CA-ID-TIMS MDAs), whereas LA-ICP-MS MDA methods that incorporate more dates from the youngest date distribution perform better; the best performing method in all cases is the maximum likelihood algorithm–minimum. The performance of tested LA-ICP-MS MDA methods is improved by removing statistical outlier dates and by removing visually young “outlier” dates that drift away from the primary distribution of young dates. Our paired LA-ICP-MS/CA-ID-TIMS MDA workflow shows that the accuracy of MDA can be significantly improved by conducting CA-ID-TIMS on as few as four grains from a sample. The combination of our new CA-ID-TIMS MDAs with published CA-ID-TIMS analyses and petrochronology on the nearby Guadalupe igneous complex provides sub-million-year resolution of contemporaneous igneous and sedimentary systems during deformation and clarifies the timing of regional deformation that defines the local “Nevadan orogeny.” Rocks in the upper plate of the Bear Mountains fault zone have a CA-ID-TIMS YTD MDA of 151.71 ± 0.23 Ma, and rocks of the lower plate have a CA-ID-TIMS YTD MDA of 149.92 ± 0.11 Ma. These new MDAs, in conjunction with the observation of fabric-bearing Mariposa Formation xenoliths in the ca. 149.65 ± 0.10 Ma Guadalupe igneous complex, suggest that rocks in the upper plate of the Bear Mountains fault zone represent a slightly older (~2 m.y.) section of Mariposa Formation that was deformed and intruded prior to being juxtaposed against, and further deformed with, a slightly younger lower-plate section of Mariposa Formation in actively deforming, fault-bounded basins. Our observations are not consistent with traditional models that require that sedimentation of the Mariposa Formation ended by ca. 155 Ma. Instead, we interpret our data to be consistent with other evidence for a continuum of deformation in Late Jurassic to Early Cretaceous time and document that the regional “slatey cleavage” observed in the greater Mariposa Formation and used to define the “Nevadan orogeny” in our study area is largely younger than 149.92 ± 0.11 Ma.

California↗
USGS2016

Persistent slip rate discrepancies in the eastern California (USA) shear zone

Understanding fault slip rates in the eastern California shear zone (ECSZ) using GPS geodesy is complicated by potentially overlapping strain signals due to many sub-parallel strike-slip faults and by inconsistencies with geologic slip rates. The role of fault system geometry in describing ECSZ deformation may be investigated with total variation regularization, which algorithmically determines a best-fitting geometry from an initial model with numerous faults, constrained by a western United States GPS velocity field. The initial dense model (1) enables construction of the first geodetically constrained block model to include all ECSZ faults with geologic slip rates, allowing direct geologic-geodetic slip rate comparisons, and (2) permits fault system geometries with many active faults that are analogous to distributed interseismic deformation. Beginning with 58 ECSZ blocks, a model containing 10 ECSZ blocks is most consistent with geologic slip rates, reproducing five of 11 within their reported uncertainties. The model fits GPS observations with a mean residual velocity of 1.5 mm/yr. Persistent geologic-geodetic slip rate discrepancies occur on the Calico and Garlock faults, on which we estimate slip rates of 7.6 mm/yr and <2 mm/yr, respectively, indicating that inconsistencies between geology and geodesy may be concentrated on or near these faults and are not due to pervasive distributed deformation in the region. Discrepancies may in part be due to postseismic relaxation following the A.D. 1992 M w 7.3 Landers and 1999 M w 7.1 Hector Mine earthquakes. Otherwise, resolving geologic-geodetic discrepancies would require as much as 11.4 mm/yr of off-fault deformation within <10 km of the main ECSZ faults, with ∼5 mm/yr concentrated near the Calico fault.

California↗
USGS2011

Surface (sea floor) and near-surface (box cores) sediment mineralogy in Baffin Bay as a key to sediment provenance and ice sheet variations

To better understand the glacial history of the ice sheets surrounding Baffin Bay and to provide information on sediment pathways, samples from 82 seafloor grabs and core tops, and from seven box cores were subjected to quantitative X-ray diffraction weight percent (wt.%) analysis of the <2 mm sediment fraction. The samples were collected between 67°N and 78°N, in water depths of 155 to 2375 m and were retrieved on cruises between A.D. 1964 and 2009. Grain size, magnetic characteristics, and colour reflectance data were also obtained on many of the samples. Twenty-one non-clay and 10 clay mineral species were identified; the average wt.% of the non-clay minerals was 70% and was dominated by quartz, various feldspars, and dolomite, whereas the dominant clay minerals were 1 M illite, biotite, and chlorite. Cluster analysis on principal component scores identified three main mineral groups, which also had strong associations with grain size and sediment magnetic properties. Box cores from the deep central basin (>2000 m) all show an abrupt drop in calcite wt.% (post-5 cal ka BP?) following a major peak in detrital carbonate (mainly dolomite). This dolomite-rich detrital carbonate (DC) event in JR175BC06 is possibly coeval with the Younger Dryas cold event. Four possible glacial-sourced end members were employed in a compositional unmixing algorithm to gain insight into down core changes in sediment provenance at the deep central basin. Estimates of the rates of sediment accumulation in the central basin are only in the range of 2 to 4 cm/cal ka, surprisingly low given the glaciated nature of the surrounding land.

Canadian Journal of Earth Sciences↗
USGS2011

Geostatistical modeling of riparian forest microclimate and its implications for sampling

Predictive models of microclimate under various site conditions in forested headwater stream - riparian areas are poorly developed, and sampling designs for characterizing underlying riparian microclimate gradients are sparse. We used riparian microclimate data collected at eight headwater streams in the Oregon Coast Range to compare ordinary kriging (OK), universal kriging (UK), and kriging with external drift (KED) for point prediction of mean maximum air temperature (Tair). Several topographic and forest structure characteristics were considered as site-specific parameters. Height above stream and distance to stream were the most important covariates in the KED models, which outperformed OK and UK in terms of root mean square error. Sample patterns were optimized based on the kriging variance and the weighted means of shortest distance criterion using the simulated annealing algorithm. The optimized sample patterns outperformed systematic sample patterns in terms of mean kriging variance mainly for small sample sizes. These findings suggest methods for increasing efficiency of microclimate monitoring in riparian areas.

Canadian Journal of Forest Research↗
USGS2022

Modeling morphodynamics of coastal response to extreme events: What shape are we in?

This review focuses on recent advances in process-based numerical models of the impact of extreme storms on sandy coasts. Driven by larger-scale models of meteorology and hydrodynamics, these models simulate morphodynamics across the Sallenger storm-impact scale, including swash, collision, overwash, and inundation. Models are becoming both wider (as more processes are added) and deeper (as detailed physics replaces earlier parameterizations). Algorithms for wave-induced flows and sediment transport under shoaling waves are among the recent developments. Community and open-source models have become the norm. Observations of initial conditions (topography, land cover, and sediment characteristics) have become more detailed, and improvements in tropical cyclone and wave models provide forcing (winds, waves, surge, and upland flow) that is better resolved and more accurate, yielding commensurate improvements in model skill. We foresee that future storm-impact models will increasingly resolve individual waves, apply data assimilation, and be used in ensemble modeling modes to predict uncertainties.

Annual Review of Marine Science↗