USGS Science⌕ Search

SEARCH · USGS Science

Results for “Machine Learning with Applications”

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 109 records · Page 6Linked to original sources
USGS2025

Harmonization of a water withdrawal dataset for the conterminous United States

The U.S. Geological Survey is developing nationally consistent water-use modeling approaches to replace previous methods relying on locally specific reported and estimated data. These national assessments require datasets that incorporate water withdrawal variability across the United States and over long periods. However, source data often have unclear definitions, missing or varied units, differing temporal resolutions, varied data quality, and inconsistent formats, which hinder automation and require individualized processing. The public-supply datasets described in this paper were used in machine learning models to estimate annual and monthly public-supply water use for 2000–2020 for the conterminous United States (CONUS) and in a model to estimate public-supply deliveries. Public-supply withdrawal data were acquired for the CONUS and the District of Columbia; however, 11 states had annual data for only 1 year, and 10 states had no monthly data. Annual withdrawal data were acquired for 81% of public-supply water service areas, and monthly withdrawal data were acquired for 47% for at least 1 year from 2000 to 2020. These datasets and methods provide the most comprehensive collection of reported public-supply withdrawals to date and can be used by water-use managers, the scientific community, and the broader public. The extensive data processing described herein can be applicable to datasets representing other categories of water use.

conterminous United States↗
USGS2025

Leveraging an observed-data likelihood improves the use of machine learning labels in a Bayesian hierarchical model for bioacoustic data

Classification of massive datasets by machine learning (ML) algorithms is promising for many scientific domains, especially wildlife monitoring programs that rely on passive acoustic surveys for detecting species. However, treating ML-predicted class labels (e.g., species identity) as truth biases inferences of focal parameters within common modeling frameworks. One solution is to model the misclassification process explicitly using human-validated true-class labels for a subset of observations. Validation by experts can present a substantial bottleneck in otherwise efficient workflows that use ML predictions. Bioacoustics practitioners seek guidance on both the quantity and process for selecting ML-labeled data to validate by an expert. We derive an alternative model formulation that jointly models human-validated and ML-predicted class labels with an observed-data likelihood (ODL) and use empirically informed simulations motivated by a real-data application to explore different probability designs for selecting class labels for validation. Simulation results suggest that with smaller validation sets the ODL formulation increases computational speed and reduces estimation error compared to a default MCMC data augmentation routine. Our methodology is transferable to applications that treat predictions from classification algorithms as the response variable of interest.

Annals of Applied Statistics↗
USGS2021

Subsurface characterization and machine learning predictions at Brady Hot Springs

Subsurface data analysis, reservoir modeling, and machine learning (ML) techniques have been applied to the Brady Hot Springs (BHS) geothermal field in Nevada, USA to further characterize the subsurface and assist with optimizing reservoir management. Hundreds of reservoir simulations have been conducted in TETRAD-G and CMG STARS to explore different injection and production fluid flow rates and allocations and to develop a training data set for ML. This process included simulating the historical injection and production since 1979 and prediction of future performance through 2040. ML networks were created and trained using TensorFlow based on multilayer perceptron, long short-term memory, and convolutional neural network architectures. These networks took as input selected flow rates, injection temperatures, and historical field operation data and produced estimates of future production temperatures. This approach was first successfully tested on a simplified single-fracture doublet system, followed by the application to the BHS reservoir. Using an initial BHS data set with 37 simulated scenarios, the trained and validated network predicted the production temperature for six production wells with the mean absolute percentage error of less than 8%. In a complementary analysis effort, the principal component analysis applied to 13 BHS geological parameters revealed that vertical fracture permeability shows the strongest correlation with fault density and fault intersection density. A new BHS reservoir model was developed considering the fault intersection density as proxy for permeability. This new reservoir model helps to explore under-exploited zones in the reservoir. A data gathering plan to obtain additional subsurface data was developed; it includes temperature surveying for three idle injection wells at which the reservoir simulations indicate high bottom-hole temperatures. The collected data assist with calibrating the reservoir model. Data gathering activities are planned for the first quarter of 2021.

Nevada↗
USGS2024

A probabilistic approach to training machine learning models using noisy data

Machine learning (ML) models are increasingly popular in environmental and hydrologic modeling, but they typically contain uncertainties resulting from noisy data (erroneous or outlier data). This paper presents a novel probabilistic approach that combines ML and Markov Chain Monte Carlo simulation to (1) detect and underweight likely noisy data, (2) develop an approach capable of detecting noisy data during model deployment, and (3) interpret the reasons why a data point is deemed noisy to help heuristically distinguish between outliers and erroneous data. The new algorithm recognizes that there is no unique way to split the training data into noisy and clean data, and thus produces an ensemble of plausible splits. The algorithm successfully detected noisy data in synthetic benchmark problems with varying complexity and a real-world public supply water withdrawal dataset. The algorithm is generic and flexible, making it suitable for application across a broad range of hydrologic and environmental disciplines.

Environmental Modelling & Software↗
USGS2018

Estimating vegetation biomass and cover across large plots in shrub and grass dominated drylands using terrestrial lidar and machine learning

Terrestrial laser scanning (TLS) has been shown to enable an efficient, precise, and non-destructive inventory of vegetation structure at ranges up to hundreds of meters. We developed a method that leverages TLS collections with machine learning techniques to model and map canopy cover and biomass of several classes of short-stature vegetation across large plots. We collected high-definition TLS scans of 26 1-ha plots in desert grasslands and big sagebrush shrublands in southwest Idaho, USA. We used the Random Forests machine learning algorithm to develop decision tree models predicting the biomass and canopy cover of several vegetation classes from statistical descriptors of the aboveground heights of TLS points. Manual measurements of vegetation characteristics collected within each plot served as training and validation data. Models based on five or fewer TLS descriptors of vegetation heights were developed to predict the canopy cover fraction of shrubs (R 2 = 0.77, RMSE = 7%), annual grasses (R 2 = 0.70, RMSE = 21%), perennial grasses (R 2 = 0.36, RMSE = 12%), forbs (R 2 = 0.52, RMSE = 6%), bare earth or litter (R 2 = 0.49, RMSE = 19%), and the biomass of shrubs (R 2 = 0.71, RMSE = 175 g) and herbaceous vegetation (R 2 = 0.61, RMSE = 99 g) (all values reported are out-of-bag). Our models explained much of the variability between predictions and manual measurements, and yet we expect that future applications could produce even better results by reducing some of the methodological sources of error that we encountered. Our work demonstrates how TLS can be used efficiently to extend manual measurement of vegetation characteristics from small to large plots in grasslands and shrublands, with potential application to other similarly structured ecosystems. Our method shows that vegetation structural characteristics can be modeled without classifying and delineating individual plants, a challenging and time-consuming step common in previous methods applying TLS to vegetation inventory. Improving application of TLS to studies of shrub-steppe ecosystems will serve immediate management needs by enhancing vegetation inventories, environmental modeling studies, and the ability to train broader datasets collected from air and space.

Idaho↗
USGS2024

Application of a workflow to determine the feasibility of using simulated streamflow for estimation of streamflow frequency statistics

Streamflow records from hydrologic models are attractive for use in operational hydrology, such as a streamflow frequency analysis. The amount of bias inherent to simulated streamflow from hydrologic models is often unknown, but it is likely present in derivative products. Therefore, a workflow may help determine where streamflow frequency analysis is credibly feasible from simulated streamflow and allow for a systematic way to assess and correct for bias. The proposed workflow consists of hydrologically matching model output locations with streamflow-gauging station (stream gauge) locations, computing the desired statistic from the simulated and observed streamflow record, computing the differences between the simulated and observed statistic (i.e., the bias), and constructing generalized additive models (GAMs) from the differences to determine bias corrections. The US Geological Survey, in cooperation with the Gulf Coast Ecosystem Restoration Council and the US Environmental Protection Agency, is testing the proposed workflow on a low-streamflow frequency (LFF) analysis. Simulated streamflows for the LFF analysis were sourced from a machine-learning model that estimated daily streamflow at Level-12 hydrologic unit code (HUC12) pour points (outlets) in the southern and southeastern US for 1950–2010. The comparison data set consists of 497 stream gauges that are coincident with a HUC12 outlet. The simulated LFF statistics were being overestimated on average; thus, there are limits to using simulated streamflow for frequency analysis. The magnitude of the overprediction generally increases where no-flow conditions are common. Bias corrections determined from the GAMs decreased the magnitude of bias observed in the simulated LFF statistics on average, suggesting it is feasible to expand the operational use of simulated streamflows to frequency analyses with the proposed workflow. The proposed workflow could be advantageous to practitioners interested in leveraging existing and future simulated streamflow data sets with regional and or global coverage.

Journal of Hydrologic Engineering↗
USGS2026

Predicting streamflow drought in the conterminous United States using machine learning and a donor-gage approach, 1982–2020

Drought is a highly consequential natural disaster that may likely increase in both severity and extent across the conterminous United States (CONUS). The mechanisms affecting the propagation of drought from the atmosphere to streamflow are complex and interactive, making the prediction of streamflow drought difficult with current modeling approaches. Machine learning is an emerging tool in the field of hydrology that may be well-suited to prediction of streamflow drought across large and topographically diverse areas. Here, we train and analyze 3198 random forest models at U.S. Geological Survey streamgages to understand common meteorological drivers of streamflow drought and to define physiographic characteristics of basins sensitive to these drivers. We also develop a novel dynamic regionalization approach using donor gages to predict daily streamflow drought at pseudo-ungaged locations. Our results show that teleconnections, temperature, evaporative demand, and snow–water equivalent are important drivers of streamflow drought in the West, Southwest, and Northern Rocky Mountains (Northern Rockies) regions of the United States, and precipitation and soil moisture are primary drivers of streamflow drought in the Northeast, Southeast, and the Northwest regions. Prediction using dynamic regionalization shows comparable performance to at-site models. This method may be applicable to other hydrologic prediction problems requiring transferability to ungaged locations.

conterminous United States↗
USGS2022

Animal tracing with sulfur isotopes: Spatial segregation and climate variability in Africa likely contribute to population trends of a migratory songbird

Climatic conditions affect animals but range-wide impacts at the population level remain largely unknown, especially in migratory species. However, studying climate–population relationships is still challenging in small migrants due to a lack of efficient and cost-effective geographic tracking method. Spatial distribution patterns of environmental stable isotopes (so called ‘isoscapes’) generally overcome these limitations but none of the currently available isoscapes provide a substantial longitudinal gradient in species-rich sub-Saharan Africa. In this region, sulphur ( δ 34 S) has not been sufficiently explored on a larger scale. We developed a δ 34 S isoscape to trace animal origins in sub-Saharan Africa by coupling known-origin samples from tracked migratory birds with continental remotely sensed environmental data building on environment– δ 34 S relationships using a flexible machine learning technique. Furthermore, we link population-specific nonbreeding grounds with interannual climatic variation that might translate to breeding population trends. The predicted δ 34 S isotopic map featured east–west and coast-to-inland isotopic gradients and was applied to predict nonbreeding grounds of three breeding populations of Eurasian Reed Warblers Acrocephalus scirpaceus with two distinct migratory phenotypes. Breeding populations as well as migratory phenotypes exhibited large-scale segregation within the African nonbreeding range. These regions also differed substantially in the interannual climatic variation, with higher interannual variability in the eastern part of the range during 2001–2012. Over the same period, the eastern European breeding population seemed to have experienced a more steep decline in population size. The link between migratory patterns and large-scale climatic variability appears important to better understand population trajectories in many declining migratory animals. We believe animal tracing using sulphur isotopes will facilitate these efforts and offers manifold ecological and forensic applications in the biodiversity hotspot of sub-Saharan Africa.

Journal of Animal Ecology↗
USGS2026

Listening for extinction: Range-wide occupancy highlights critical risks and priorities for northern spotted owls

The northern spotted owl ( Strix occidentalis caurina ) continues to decline across its range, threatened by habitat disturbances and invasive barred owls ( Strix varia ). We conducted a range-wide spotted owl occupancy assessment using 2.1 million h of passive acoustic recordings from 4081 survey stations within 1027 randomly selected 5-km 2 hexagons across six physiographic regions. Applying multistate occupancy models, we estimated probabilities of landscape use (by at least one northern spotted owl) and pair occupancy (occupied by both a male and a female), while accounting for imperfect and sex-biased detection probability. Detection probability of pairs was low in most regions, primarily due to undetected females, but increased with survey effort. If management decisions can only be made when there is a high probability of detecting both sexes when they are present, then longer surveys are one approach to support the decisions. Landscape use and pair occupancy were positively associated with old-growth forest structure and topographic diversity. We also found a pronounced latitudinal gradient in spotted owl landscape use rates with estimated mean landscape use of 0.17 (SE = 0.03) and 0.16 (SE = 0.02) in the two most northern regions, and 0.54 (SE = 0.04) and 0.52 (SE = 0.05) in the two most southern regions. Pair occupancy rates were higher in southern regions (0.29, SE = 0.03, 0.43, SE = 0.05) and very low in northern regions (0.04, SE = 0.01, 0.03, SE = 0.01). In contrast to spotted owls, barred owls were detected in 90% of northern hexagons and 50% of southern hexagons, and barred owl calling intensity was over eight times higher than spotted owls. These results highlight extinction risks in several regions and suggest that management actions, such as spatially targeted barred owl control and protection of structurally complex habitats, have a narrow temporal window in which they can be effective in lowering the probability of regional extirpations. We generated predictive maps of landscape use and pair occupancy, which could help guide regional conservation strategies. This study demonstrates how passive acoustic monitoring and machine learning can be used for broad-scale ecological monitoring, identifying extinction risk thresholds, supporting adaptive management, and improving conservation outcomes for wide-ranging and elusive species that can reliably be detected through unique vocalizations.

California, Oregon, Washington↗
USGS2018

Seismology with dark data: Image-based processing of analog records using machine learning for the rangely earthquake control experiment

Before the digital era, seismograms were recorded in analog form and read manually by analysts. The digital era represents only about 25% of the total time span of instrumental seismology. Analog data provide important constraints on earthquake processes over the long term, and in some cases are the only data available. The media on which analog data are recorded degrades with time and there is an urgent need for cost‐effective approaches to preserve the information they contain. In this study, we work directly with images by constructing a set of image‐based methods for earthquake processing, rather than pursue the usual approach of converting analog data to vector time series. We demonstrate this approach on one month of continuous Develocorder films from the Rangely earthquake control experiment run by the U.S. Geological Survey (USGS). We scan the films into images and compress these into low‐dimensional feature vectors as input to a classifier that separates earthquakes from noise in a defined feature space. We feed the detected event images into a short‐term average/long‐term average (STA/LTA) picker, a grid‐search associator, and a 2D image correlator to measure both absolute arrival times and relative arrival‐time differences between events. We use these measurements to locate the earthquakes using hypoDD. In the month that we studied, we identified 40 events clustered near the injection wells. In the original study, Raleigh et al. (1976) identified only 32 events during the same period. Scanning without vectorizing analog seismograms represents an attractive approach to archiving these perishable data. We demonstrated that it is possible to carry out precision seismology directly on such images. Our approach has the potential for wide application to analog seismograms.

Seismological Research Letters↗
USGS2022

Human-in-the-Loop segmentation of earth surface imagery

Segmentation, or the classification of pixels (grid cells) in imagery, is ubiquitously applied in the natural sciences. Manual methods are often prohibitively time-consuming, especially those images consisting of small objects and/or significant spatial heterogeneity of colors or textures. Labeling complicated regions of transition that in Earth surface imagery are represented by collections of mixed-pixels, -textures, and -spectral signatures, can be especially error-prone because it is difficult to reliably unmix, identify and delineate consistently. However, the success of supervised machine learning (ML) approaches is entirely dependent on good label data. We describe a fast, semi-automated, method for interactive segmentation of N-dimensional (x, y, N) images into two-dimensional (x, y) label images. It uses human-in-the-loop ML to achieve consensus between the labeler and a model in an iterative workflow. The technique is reproducible; the sequence of decisions made by human labeler and ML algorithms can be encoded to file, so the entire process can be played back and new outputs generated with alternative decisions and/or algorithms. We illustrate the scientific potential of segmentation of imagery of diverse settings and image types using six case studies from river, estuarine, and open coast environments. These photographic and non-photographic imagery consist of 1- and 3-bands on regular and irregular grids ranging from centimeters to tens of meters. We demonstrate high levels of agreement in label images generated by several labelers on the same imagery, and make suggestions to achieve consensus and measure uncertainty, ideal for widespread application in training supervised ML for image segmentation.

Earth and Space Science↗
USGS2025

An integrated sensor network and data driven approach to satellite remote sensing of dissolved organic matter

Traditional remote sensing retrieval models for water quality have historically relied on limited, localized data sets due to the prohibitive costs of extensive field campaigns and logistical challenges of collecting match-up data with satellite overpasses. As a result, these models often lack generalizability across seasons, tides, and sites. Furthermore, small field data sets limit the utility of modern machine learning techniques to advance remote sensing retrieval models. In situ optical sensors deployed in a sensor network to continuously monitor larger water bodies can drastically increase the number of measurements, providing the opportunity to develop new approaches for building robust remote sensing retrieval models by leveraging both remote sensing data and in situ networks as an integrated monitoring system. This study leverages a large “ground-to-space” sensor network that combines an in situ optical sensor network with satellite-based remote sensing to overcome these limitations. Utilizing a large-scale data set from the U.S. Geological Survey's Sacramento—San Joaquin River Delta monitoring network, of dissolved organic matter fluorescence measurements, and remote sensing data from the European Space Agency's Sentinel-2A and -2B satellites, this study implemented a data driven approach for dissolved organic matter models. The data set, consisting of 982 samples collected between 2018 and 2021 was used to train and validate a random forest model ( R 2 = 0.76, RMSE = 6.1 Quinine Sulfate Equivalents), with demonstrated applicability across diverse site conditions, tidal stages, and seasons. This work provides a scalable solution to address critical challenges in water quality monitoring and offers a replicable framework for global water quality management.

Calfornia↗
USGS2020

GeoNat v1.0: A dataset for natural feature mapping with artificial intelligence and supervised learning

Machine learning allows “the machine” to deduce the complex and sometimes unrecognized rules governing spatial systems, particularly topographic mapping, by exposing it to the end product. Often, the obstacle to this approach is the acquisition of many good and labeled training examples of the desired result. Such is the case with most types of natural features. To address such limitations, this research introduces GeoNat v1.0, a natural feature dataset, used to support artificial intelligence‐based mapping and automated detection of natural features under a supervised learning paradigm. The dataset was created by randomly selecting points from the U.S. Geological Survey’s Geographic Names Information System and includes approximately 200 examples each of 10 classes of natural features. Resulting data were tested in an object‐detection problem using a region‐based convolutional neural network. The object‐detection tests resulted in a 62% mean average precision as baseline results. Major challenges in developing training data in the geospatial domain, such as scale and geographical representativeness, are addressed in this article. We hope that the resulting dataset will be useful for a variety of applications and shed light on training data collection and labeling in the geospatial artificial intelligence domain.

Transactions in GIS↗
USGS2023

Preliminary machine learning models of manganese and 1,4-dioxane in groundwater on Long Island, New York

Manganese and 1,4-dioxane in groundwater underlying Long Island, New York, were modeled with machine learning methods to demonstrate the use of these methods for mapping contaminants in groundwater in the Long Island aquifer system. XGBoost, a gradient boosted, ensemble tree method, was applied to data from 910 wells for manganese and 553 wells for 1,4-dioxane. Explanatory variables included soil properties, groundwater flow, land use, and other features that describe the hydrogeology and geochemistry of the aquifer system. Four models were developed to predict the probability of manganese concentrations greater than a detection level of 10 micrograms per liter (μg/L) and greater than three threshold concentrations (50, 150, and 300 μg/L) relevant to drinking-water quality. One model was developed to predict the probability of 1,4-dioxane concentrations greater than a detection level of 0.07 μg/L. The 1,4-dioxane model was limited geographically to Suffolk County because of data availability. Predictions were made for two layers in the upper glacial aquifer and three layers in the Magothy aquifer, which are the upper two of the three major aquifers of the Long Island aquifer system. The objective of the study described in this report was to demonstrate the application of the methods rather than to develop precise estimates of manganese or 1,4-dioxane concentrations at any given location. The predictive models developed in the study are considered preliminary in the sense that they are an initial effort at developing these kinds of models specifically for Long Island. The models could be improved by the inclusion of additional data, by the use of methods to improve the modeling of infrequent high concentrations of manganese and 1,4-dioxane (above threshold concentrations), and by including more explanatory variables that specifically describe conditions and contaminant sources on Long Island. Nonetheless, the distribution of model predictions and the influence of explanatory variables in the models were consistent with the expected relations between contaminant concentrations and groundwater-flow-system characteristics and the distribution of manmade sources. Mapped predictions indicated that manganese detections were more probable in the upper glacial aquifer and along the southern shore of Long Island, consistent with the distribution of anoxic conditions in groundwater in the Long Island aquifer system. Manganese was infrequently predicted at concentrations greater than thresholds of concern for drinking-water quality in any of the aquifer layers. Detections of 1,4-dioxane were predicted in the western, more highly developed parts of Suffolk County, in the upper glacial aquifer and the top and middle layers of the Magothy aquifer, and in northwestern Suffolk County in the bottom layer of the Magothy aquifer. Although preliminary in nature and based on limited data, these mapped predictions can be used to generally identify areas where manganese and 1,4-dioxane may be present at concentrations of concern to prioritize areas for future monitoring and to guide future modeling and mapping efforts.

New York↗
USGS2021

Machine learning can assign geologic basin to produced water samples using major ion geochemistry

Understanding the geochemistry of waters produced during petroleum extraction is essential to informing the best treatment and reuse options, which can potentially be optimized for a given geologic basin. Here, we used the US Geological Survey’s National Produced Waters Geochemical Database (PWGD) to determine if major ion chemistry could be used to classify accurately a produced water sample to a given geologic basin based on similarities to a given training dataset. Two datasets were derived from the PWGD: one with seven features but more samples (PWGD7), and another with nine features but fewer samples (PWGD9). The seven-feature dataset, prior to randomly generating a training and testing (i.e., validation) dataset, had 58,541 samples, 20 basins, and was classified based on total dissolved solids (TDS), bicarbonate (HCO 3 ), Ca, Na, Cl, Mg, and sulfate (SO 4 ). The nine-feature dataset, prior to randomly splitting into a training and testing (i.e., validation) dataset, contained 33,271 samples, 19 basins, and was classified based on TDS, HCO 3 , Ca, Na, Cl, Mg, SO 4 , pH, and specific gravity. Three supervised machine learning algorithms—Random Forest, k-Nearest Neighbors, and Naïve Bayes—were used to develop multi-class classification models to predict a basin of origin for produced waters using major ion chemistry. After training, the models were tested on three different datasets: Validation7, Validation9, and one based on data absent from the PWGD. Prediction accuracies across the models ranged from 23.5 to 73.5% when tested on the two PWGD-based datasets. A model using the Random Forest algorithm predicted most accurately compared to all other models tested. The models generally predicted basin of origin more accurately on the PWGD7-based dataset than on the PWGD9-based dataset. An additional dataset, which contained data not in the PWGD, was used to test the most accurate model; results suggest that some basins may lack geochemical diversity or may not be well described, while others may be geochemically diverse or are well described. A compelling result of this work is that a produced water basin of origin can be determined using major ions alone and, therefore, deep basinal fluid compositions may not be as variable within a given basin as previously thought. Applications include predicting the geochemistry of produced fluid prior to drilling at different intervals and assigning historical produced water data to a producing basin.

Natural Resources Research↗
USGS2023

A multimodal data fusion and deep learning framework for large-scale wildfire surface fuel mapping

Accurate estimation of fuels is essential for wildland fire simulations as well as decision-making related to land management. Numerous research efforts have leveraged remote sensing and machine learning for classifying land cover and mapping forest vegetation species. In most cases that focused on surface fuel mapping, the spatial scale of interest was smaller than a few hundred square kilometers; thus, many small-scale site-specific models had to be created to cover the landscape at the national scale. The present work aims to develop a large-scale surface fuel identification model using a custom deep learning framework that can ingest multimodal data. Specifically, we use deep learning to extract information from multispectral signatures, high-resolution imagery, and biophysical climate and terrain data in a way that facilitates their end-to-end training on labeled data. A multi-layer neural network is used with spectral and biophysical data, and a convolutional neural network backbone is used to extract the visual features from high-resolution imagery. A Monte Carlo dropout mechanism was also devised to create a stochastic ensemble of models that can capture classification uncertainties while boosting the prediction performance. To train the system as a proof-of-concept, fuel pseudo-labels were created by a random geospatial sampling of existing fuel maps across California. Application results on independent test sets showed promising fuel identification performance with an overall accuracy ranging from 55% to 75%, depending on the level of granularity of the included fuel types. As expected, including the rare—and possibly less consequential—fuel types reduced the accuracy. On the other hand, the addition of high-resolution imagery improved classification performance at all levels.

California↗
USGS2025

Interrogating process deficiencies in large-scale hydrologic models with interpretable machine learning

Large-scale hydrologic models are increasingly being developed for operational use in the forecasting and planning of water resources. However, the predictive strength of such models depends on how well they resolve various functions of catchment hydrology, which are influenced by gradients in climate, topography, soils, and land use. Most assessments of hydrologic model uncertainty have been limited to traditional statistical methods. Here, we present a proof-of-concept approach that uses interpretable machine learning techniques to provide post hoc assessment of model sensitivity and process deficiency in hydrologic models. We train a random forest model to predict the Kling–Gupta efficiency (KGE) of National Water Model (NWM) and National Hydrologic Model (NHM) streamflow predictions for 4383 stream gauges in the conterminous United States. Thereafter, we explain the local and global controls that 48 catchment attributes exert on KGE prediction using interpretable Shapley values. Overall, we find that soil water content is the most impactful feature controlling successful model performance, suggesting that soil water storage is difficult for hydrologic models to resolve, particularly for arid locations. We identify nonlinear thresholds beyond which predictive performance decreases for NWM and NHM. For example, soil water content less than 210 mm, precipitation less than 900 mm yr −1 , road density greater than 5 km km −2 , and lake area percent greater than 10 % contributed to lower KGE values. These results suggest that improvements in how these influential processes are represented could result in the largest increases in NWM and NHM predictive performance. This study demonstrates the utility of interrogating process-based models using data-driven techniques, which has broad applicability and potential for improving the next generation of large-scale hydrologic models.

conterminous United States↗
USGS2023

Beyond overlap: Considering habitat preference and fitness outcomes in the umbrella species concept

Umbrella species and other surrogate species approaches to conservation provide an appealing framework to extend the reach of conservation efforts beyond single species. For the umbrella species concept to be effective, populations of multiple species of concern must persist in areas protected on behalf of the umbrella species. Most assessments of the concept, however, focus exclusively on geographic overlap among umbrella and background species, and not measures that affect population persistence (e.g. habitat quality or fitness). We quantified the congruence between the habitat preferences and nesting success of a high-profile umbrella species (greater sage-grouse, Centrocercus urophasianus , hereafter ‘sage-grouse’), and three sympatric species of declining songbirds (Brewer's sparrow Spizella breweri , sage thrasher Oreoscoptes montanus and vesper sparrow Pooecetes gramineus ) in central Wyoming, USA during 2012–2013. We used machine-learning methods to create data-driven predictions of sage-grouse nest-site selection and nest survival probabilities by modeling field-collected sage-grouse data relative to habitat attributes. We then used field-collected songbird data to assess whether high-quality sites for songbirds aligned with those of sage-grouse. Nest sites selected by songbirds did not coincide with sage-grouse nesting preferences, with the exception that Brewer's sparrows preferred similar nest sites to sage-grouse in 2012. Moreover, the areas that produced higher rates of songbird nest survival were unrelated to those for sage-grouse. Our findings suggest that management actions at local scales that prioritize sage-grouse nesting habitat will not necessarily enhance the reproductive success of sagebrush-associated songbirds. Measures implemented to conserve sage-grouse and other purported umbrella species at broad spatial scales likely overlap the distribution of many species, however, broad-scale overlap may not translate to fine-scale conservation benefit beyond the umbrella species itself. The maintenance of microhabitat heterogeneity important for a diversity of species of concern will be critical for a more holistic application of the umbrella species concept.

Animal Conservation↗