USGS Science⌕ Search

SEARCH · USGS Science

Results for “Water Research”

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 1,009 records · Page 56Linked to original sources
USGS2018

A comparison of methods for streamflow uncertainty estimation

Streamflow time series are commonly derived from stage‐discharge rating curves, but the uncertainty of the rating curve and resulting streamflow series are poorly understood. While different methods to quantify uncertainty in the stage‐discharge relationship exist, there is limited understanding of how uncertainty estimates differ between methods due to different assumptions and methodological choices. We compared uncertainty estimates and stage‐discharge rating curves from seven methods at three river locations of varying hydraulic complexity. Comparison of the estimated uncertainties revealed a wide range of estimates, particularly for high and low flows. At the simplest site on the Isère River (France), full width 95% uncertainties for the different methods ranged from 3 to 17% for median flows. In contrast, uncertainties were much higher and ranged from 41 to 200% for high flows in an extrapolated section of the rating curve at the Mahurangi River (New Zealand) and 28 to 101% for low flows at the Taf River (United Kingdom), where the hydraulic control is unstable at low flows. Differences between methods result from differences in the sources of uncertainty considered, differences in the handling of the time‐varying nature of rating curves, differences in the extent of hydraulic knowledge assumed, and differences in assumptions when extrapolating rating curves above or below the observed gaugings. Ultimately, the selection of an uncertainty method requires a match between user requirements and the assumptions made by the uncertainty method. Given the significant differences in uncertainty estimates between methods, we suggest that a clear statement of uncertainty assumptions be presented alongside streamflow uncertainty estimates.

Water Resources Research↗
USGS2018

Tidal response of groundwater in a leaky aquifer—Application to Oklahoma

Quantitative interpretation of the tidal response of water levels measured in wells has long been made either with a model for perfectly confined aquifers or with a model for purely unconfined aquifers. However, many aquifers may be neither totally confined nor purely unconfined at the frequencies of tidal loading but behave somewhere between the two end‐members. Here we present a more general model for the tidal response of groundwater in aquifers with both horizontal flow and vertical leakage. The model has three independent parameters: the transmissivity ( T ) and storativity ( S ) of the aquifer and the specific leakage ( K′/b′ ) of the leaking aquitard, where K′ and b′ are the hydraulic conductivity and the thickness of the aquitard, respectively. If T and S are known independently, this model may be used to estimate aquitard leakage from the phase shift and amplitude ratio of water level in wells obtained from tidal analysis. We apply the model to interpret the tidal response of water level in a US Geological Survey (USGS) deep monitoring well installed in the Arbuckle aquifer in Oklahoma, into which massive amount of wastewater coproduced from hydrocarbon exploration has been injected. The analysis shows that the Arbuckle aquifer is leaking significantly at this site. We suggest that the present method may be effective and economical for monitoring leakage in groundwater systems, which bears on the safety of water resources, the security of underground waste repositories, and the outflow of wastewater during deep injection and hydrocarbon extraction.

Oklahoma↗
USGS2019

POLARIS properties: 30-meter probabilistic maps of soil properties over the contiguous United States

Soils play a critical role in the cycling of water, energy, and carbon in the Earth system. Until recently, due primarily to a lack of soil property maps of a sufficiently high‐quality and spatial detail, a minor emphasis has been placed on providing high‐resolution measured soil parameter estimates for land surface models and hydrologic models. This study introduces Probabilistic Remapping of SSURGO (POLARIS) soil properties—a database of 30‐m probabilistic soil property maps over the contiguous United States (CONUS). The mapped variables over CONUS include soil texture, organic matter, pH, saturated hydraulic conductivity, Brooks‐Corey and Van Genuchten water retention curve parameters, bulk density, and saturated water content. POLARIS soil properties was assembled by (1) depth harmonizing and aggregating the pedons in the National Cooperative Soil Survey Soil Characterization Database and the components in Soil Survey Geographic Database into a database of 21,481 different soil series, each soil series having its own vertical profiles of different soil properties, (2) pruning the original POLARIS soil series maps using conventional soil maps to improve soil series prediction accuracy, and (3) merging the assembled soil series databases with the pruned POLARIS soil series maps to construct the soil property maps over CONUS. POLARIS soil properties includes 100‐bin histograms for each layer and variable per grid cell and a series of summary statistics at 30‐, 300‐, and 3,000‐m spatial resolution. Evaluation of POLARIS soil properties using in situ measurements shows an average R2 of 0.41, normalized root‐mean‐square error of 12%, and a normalized mean absolute error of 8.8%.

Water Resources Research↗
USGS2018

Direct observations of hydrologic exchange occurring with less‐mobile porosity and the development of anoxic microzones in sandy lakebed sediments

Quantifying coupled mobile/less‐mobile porosity dynamics is critical to the prediction of biogeochemical storage, release, and transformation processes in the zone where groundwater and surface water exchange. The recent development of fine‐scale geoelectrical monitoring paired with pore‐water sampling in groundwater systems enables direct characterization of hydrologic exchange between more‐ and less‐mobile porosity during tracer tests. We adapt this technique to sandy interface sediments at a groundwater flow‐through kettle lake. Tracer experiments were conducted within controlled‐head permeameters over a range of specified downward flow conditions over several days. Although the bed was predominantly composed of highly permeable sands and gravels, cobble inclusions created less‐mobile flow zones at the centimeter scale. Less‐mobile porosity fractions, residence times, and rates of exchange were inferred from paired bulk and fluid electrical conductivity data, without the need for inverse model calibration. The conservative solute experiments were paired with 15 NO 3 − and other reactive amendments, revealing anaerobic processes occurring at shallow sediment depths where pore‐water sampling indicated bulk‐oxic conditions. The average less‐mobile porosity residence times as evaluated with the geoelectrical method were on 1‐hr timescales, which appear to be biogeochemically important in the context of creating anoxic microzones within less‐mobile porosity of sandy interface sediments.

Water Resources Research↗
USGS2018

Responses of unimpaired flows, storage, and managed flows to scenarios of climate change in the San Francisco Bay-Delta watershed

Projections of meteorology downscaled from global climate model runs were used to drive a model of unimpaired hydrology of the Sacramento/San Joaquin watershed, which in turn drove models of operational responses and managed flows. Twenty daily climate change scenarios for water years 1980–2099 were evaluated with the goal of producing inflow boundary conditions for a watershed sediment model and for a hydrodynamical model of the San Francisco Bay-Delta estuary. The resulting time series of meteorology, snowpack, unimpaired flow, reservoir storage, and managed flow were analyzed for century-scale trends. In the Sacramento basin, which dominates Bay-Delta inflows, all 20 scenarios portrayed warming trends (with a mean of 4.1 °C) and most had precipitation increases (with a mean increase of 9%). Sacramento basin snowpack water equivalent declined sharply (by 89%), which was associated with a major shift toward earlier unimpaired runoff timing (33% more flow arriving prior to 1 April). Sacramento basin reservoirs showed large declines in end-of-September storage. Water-year averaged outflows increased for most scenarios for both unimpaired and impaired flows, and frequency of extremely high daily unimpaired and impaired flows increased (increases of 175% and 170%, respectively). Managed Delta inflows were projected to experience large increases in the wet season and declines in the dry season. Changes in management strategy and infrastructure can mitigate some of these changes, though to what degree is uncertain.

California↗
USGS2018

Time-dependent pore filling

Capillarity traps fluids in porous media during immiscible fluid displacement. Most field situations involve relatively long time scales, such as hydrocarbon migration into reservoirs, resource recovery, nonaqueous phase liquid remediation, geological CO 2 storage, and sediment‐atmosphere interactions. Yet laboratory studies and numerical simulations of capillary phenomena rarely consider the impact of time on these processes. We use time‐lapse microphotography to record the evolution of saturation in air‐ or hydrocarbon‐filled capillary tubes submerged in water to investigate long‐term pore filling phenomena beyond imbibition. Microphotographic sequences capture a lively pore filling history where various concurrent physical phenomena coexist. Dissolution and diffusion play a central role. Observations indicate preferential transport of the wetting liquid along corners, vapor condensation, capillary flow induced by asymmetrical interfaces, and interface pinning that defines the diffusion length. Other processes include internal snap‐offs, fluid redistribution, and changes in wettability as fluids dissolve into each other. Overall, the rate of pore filling is diffusion‐controlled for a given interfacial configuration; diffusive transport takes place at a constant rate for pinned interfaces and is proportional to the square root of time for free interfaces where the diffusion length increases with time.

Water Resources Research↗
USGS2019

Winter precipitation and summer temperature predict lake water quality at macroscales

Climate change can have strong effects on aquatic ecosystems, including disrupting nutrient cycling and mediating processes that affect primary production. Past studies have been conducted mostly on individual or small groups of ecosystems, making it challenging to predict how future climate change will affect water quality at broad scales. We used a subcontinental-scale database to address three objectives: (1) identify which climate metrics best predict lake water quality, (2) examine whether climate influences different nutrient and productivity measures similarly, and (3) quantify the potential effects of a changing climate on lakes. We used climate data to predict lake water quality in ~11,000 north temperate lakes across 17 U.S. states. We developed a novel machine learning method that jointly models different measures of water quality using 48 climate metrics and accounts for properties inherent in macroscale data (e.g. , spatial autocorrelation). Our results suggest that climate metrics related to winter precipitation and summer temperature were strong predictors of lake nutrients and productivity. However, we found variation in the magnitude and direction of the relationship between climate and water quality. We predict that a likely future climate change scenario of warmer summer temperatures will lead to increased nutrient concentrations and algal biomass across lakes (median ~3%–9% increase), whereas increased winter precipitation will have highly variable effects. Our results emphasize the importance of heterogeneity in the response of individual ecosystems to climate and are a caution to extrapolating relationships across space.

Water Resources Research↗
USGS2018

Predicting geogenic arsenic in drinking water wells in glacial aquifers, north-central USA: Accounting for depth-dependent features

Chronic exposure to arsenic (As) via drinking groundwater is a human health concern worldwide. Probabilities of elevated geogenic As concentrations in groundwater were predicted in complex, glacial aquifers in Minnesota, north‐central USA, a region that commonly has elevated As concentrations in well water. Maps of elevated As hazard were created for depths typical of drinking water supply and with well construction attributes common for domestic wells. Conventional variables describing aquifer properties and materials, position on the hydrologic landscape, and soil geochemistry were among the most influential for predicting the probability of elevated As. We also found that certain well construction attributes were influential in predicting As hazard. Smaller distances between the top of the well screen and overlying aquitard (proximity) and shorter well screen lengths were each associated with higher probabilities of elevated As. Influential predictor variables, which are either mapped across the region or are well construction attributes, are proxies in the model for measurable physical or geochemical causes of elevated As (e.g., redox condition, till or aquifer sediment chemistry, and water chemistry), which are not mapped across the region. Our setting shares some important characteristics with deltaic and other high‐As aquifers in Southeast Asia: late Quaternary age, complex layering of coarse‐ and fine‐grained materials, low‐As sediment concentrations, and geochemical controls on As mobilization. Translating three‐dimensional geologic and geochemical understanding of As mobility to quantifiable variables for modeling with powerful, flexible statistical tools could improve predictions and help identify safer groundwater supply options in the USA, Southeast Asia, and elsewhere.

Minnesota↗
USGS2019

The dependence of hydroclimate projections in snow‐dominated regions of the western United States on the choice of statistically downscaled climate data

We assess monthly temperature and precipitation data produced by four statistically based techniques that were used to downscale general circulation models (GCMs) in the Climate Model Intercomparison Program Phase 5 (CMIP5) (Taylor et al., 2012). We drive a simple water-balance model with the downscaled data to demonstrate the effect of the methods on the cold season hydrology of three, snow dominated regions in the western U.S. Independent of substantial variation among the GCM simulations over the regions (maximum range of ~3.5 °C and 50% change in precipitation), the four methods produce disparate high resolution representations of the magnitude and spatial patterns of future temperature and precipitation simulated by the models that range for up to ~3 °C and 30% change in precipitation that propagate into the hydrologic simulations. Temperature-dependent snowfall, accumulation, and melt in the model are sensitive to how atmospheric lapse rates are applied in the gridded observations that are used to remove the bias in raw GCM temperatures. By the end of the century the same downscaling method (Bias Corrected Spatial Disaggregation) yields a loss of cold-season snowpack of 34% over the Greater Yellowstone Area under a constant lapse rate ( 6.5°C km-1), whereas spatially variable lapse rates nearly double the loss to 66%, highlighting the roll of both lapse rates and high elevation stations in the bias correction dataset. The two newest downscaling methods (Multivariate Adaptive Constructed Analogs and Localized Constructed Analogs) preserve the magnitude of change simulated GCMs better than the other methods and the produce comparable hydrologic projections. Because the downscaled data from the methods vary spatially and by GCM, the downscaled data should be evaluated carefully as part of the process of using downscaled climate products to drive hydrological models over the area of interest.

Arizona, California, Colorado, Idaho, Montana,Neva↗
USGS2019

Temporal variability in nitrate – discharge relationships in large rivers as revealed by high frequency data

Little is known about temporal variability in nitrate concentration responses to changes in discharge on intraannual time scales in large rivers. To investigate this knowledge gap, we used a six‐year data set of daily surface water nitrate concentration and discharge averaged from near‐continuous monitoring at U.S. Geological Survey gaging stations on the Connecticut, Potomac, and Mississippi Rivers, three large rivers that contribute substantial nutrient pollution to important estuaries. Interannually, a comparison of nitrate concentration‐discharge (c‐Q) relationships between a traditional discrete grab sample data set and the near‐continuous data set revealed differing c‐Q slopes, which suggests that sample frequency can impact how we ultimately characterize hydrologic systems. Intraannually, we conducted correlation analyses over 30‐day windows to isolate the strength and direction of monthly c‐Q relationships. Monthly c‐Q slopes in the Potomac were positive (enrichment/mobilization response) in summer and fall and negative (dilution response) and weakly chemostatic (nonsignificant near‐zero c‐Q slope) in winter and spring, respectively. The Connecticut displayed a dilution response year‐round, except summer when it was weakly chemostatic. Mississippi c‐Q slopes were weakly chemostatic in all seasons and showed inconsistent responses to discharge fluctuations. The c‐Q dynamics in the Potomac and Connecticut were correlated ( R > 0.3) to river temperature, flow percentile, and calendar day. Minimal correlation in the Mississippi suggests that the large basin area coupled with spatiotemporally variable anthropogenic forcings from substantial land use development created stochastic short‐term c‐Q relationships. Additional work using high‐frequency sensors across large river networks can improve our understanding of spatial source input dynamics in these natural‐human coupled systems.

Water Resources Research↗
USGS2019

Remote sensing of river bathymetry: Evaluating a range of sensors, platforms, and algorithms on the upper Sacramento River, California, USA

Remote sensing has become an increasingly viable tool for characterizing fluvial systems. In this study, we used field measurements from a 1.6 km reach of the upper Sacramento River, CA, to evaluate the potential of mapping water depths from a range of platforms, sensors, and depth retrieval methods. Field measurements of water column optical properties also were compared to similar data sets from other rivers to provide context for our results. We considered field spectra, a multispectral satellite image, hyperspectral data collected from conventional and unmanned aircraft, and a bathymetric LiDAR and applied a generalized version of Optimal Band Ratio Analysis (OBRA) and the K nearest neighbors regression (KNN) machine learning algorithm. Linear, quadratic, exponential, power, and lowess OBRA models enabled more flexible curve-fitting in calibrating spectrally based quantities to depth; an exponential formulation avoided artifacts associated with other model types. KNN increased observed vs. predicted R2 values, particularly for the satellite image; we also found that pre-processing of satellite images was unnecessary and that a basic data product could be used for depth retrieval. Bathymetric LiDAR was highly accurate and precise in shallow water, but a lack of bottom returns from areas greater than 2 m deep resulted in large gaps in coverage. The maximum detectable depth imposes an important constraint on fluvial remote sensing and a hybrid approach combined with field surveys of deep areas might be a more realistic operational strategy for bathymetric mapping. Future work will focus on scaling up from short reaches to long river segments.

California↗
USGS2019

Soil physical, hydraulic, and thermal properties in interior Alaska, USA: Implications for hydrologic response to thawing permafrost conditions

Boreal forest regions are a focal point for investigations of coupled water and biogeochemical fluxes in response to wildfire disturbances, climate warming, and permafrost thaw. Soil hydraulic, physical, and thermal property measurements for mineral soils in permafrost regions are limited, despite substantial influences on cryohydrogeologic model results. This work expands mineral soil property quantification in cold regions through soil characterization from the discontinuous permafrost zone of interior Alaska, USA. Values extend beyond the range of prior measurement magnitudes in analogous regions, highlighting the importance of this dataset. Rocky and silty upland soil landscape classifications and wildfire disturbance provided guiding frameworks for the sampling and analysis for potential implications for the hydrologic response to thawing permafrost. Bulk density, soil organic matter, soil–particle size distributions (sand, silt, and gravel fractions), and soil hydraulic properties of van Genuchten parameters alpha and N had moderate evidence of differences between silty and rocky classifications. Burned and unburned sites had only moderate evidence of differences for silt fraction. Field-saturated hydraulic conductivity (Kfs) was more variable at burned sites compared to unburned sites, which corresponded to observations of greater rooting depths at burned sites and observations of root paths in soil cores for Kfs measurement. Soil thermal properties suggested that gravel content may reduce the accuracy of commonly used estimation methods for thermal conductivity . This work provides soil parameter constraints necessary for hypothesis testing and site-specific prediction with cryohydrogeologic models to examine controls on active layer and permafrost dynamics in upland boreal forests.

Alaska↗
USGS2019

Salinity yield modeling of the Upper Colorado River Basin using 30-meter resolution soil maps and random forests

Salinity loading in the Upper Colorado River Basin (UCRB) costs local economies upwards of $300 million US dollars annually. Salinity source models have generally included coarse spatial data to represent non‐agriculture sources. We developed new predictive soil property and cover maps at 30 m resolution to improve source representation in salinity modeling. Salinity loading erosion risk indices were also created based on soil properties, remotely sensed bare ground exposure, and topographic factors to examine potential surface soil erosion drivers. These new maps and data from previous SPARROW models were related to recently updated records of salinity at 309 stream gauges in the UCRB using random forest regressions. Resulting salinity yield predictions indicate more diffuse salinity sources, with slightly higher yields in more arid portions of the UCRB, and less overall load coming from irrigated agricultural sources. Model simulations still indicate irrigation to be the major human source of salinity (661,000 Mg, or 12%), but also suggest that 75,000 Mg (1.4%) of annual salinity in the UCRB is coming from areas with excessive exposed bare ground in high elevation mountain areas. Model inputs allow for field scale screening of locations that could be targeted for salinity control projects. Results confirm recent studies indicating limited surface erosional influence on salinity loading in UCRB surface waters, but impacts of monsoonal runoff events are still not fully understood, particularly in drylands. The study highlights the utility of new predictive soil maps and machine learning for environmental modeling.

Arizona, Colorado, New Mexico, Utah, Wyoming↗
USGS2020

Deposition potential and flow-response dynamics of emergent sandbars in a braided river

Sandbars are ubiquitous in sandy‐braided rivers throughout the world. In the Great Plains of the United States, recovery and expansion of emergent sandbar habitat (ESH) has been a priority in lowland rivers where the natural extent of sandbars has been degraded. Recovery efforts are aimed at protection of populations of the interior least tern ( Sterna antillarum ) and piping plover ( Charadrius melodus ). But quantitative observations of deposition and erosion dynamics of populations of sandbars across long segments of rivers are rare. We present a three‐part case study which used Bayesian regression models to examine relations between hydrology, channel morphology, and ESH responses in the Platte River, eastern Nebraska. Logistic regression indicates presence of ESH is positively related to the Parker, (1976) stability criterion and a gradient in sediment transport mode, and negatively related to presence of vegetation. Hierarchical linear regression modeling shows direct coupling between sandbar top‐surface height and formative flood magnitude, but the gap between formative flood stage and sandbar top‐surface increases with increasing discharge. Finally, linear regression modeling of sandbar erosion demonstrates rates of ESH erosion are on the order of 10 −1 ha/day during high‐flow periods and 10 −2 during low‐flow periods, but sandbar persistence is largely a function of sandbar starting size. The collective observations highlight the importance of large floods (>3‐year recurrence) in creating very large sandbars that persist as high‐quality ESH over periods of years whereas lower‐magnitude, more‐frequent flood events create lower‐quality ESH that typically does not persist into the following nesting season.

Nebraska↗
USGS2019

Morphological computation of dune evolution with equilibrium and non-equilibrium sediment-transport models

This paper presents an exploratory study that comprises the implementation and comparison of different approaches and parameterization of sediment transport mechanisms in a process-based morphological model for simulating river dunes. The purpose of this study was to assess the underlying physical processes associated with sediment transport and dune evolution simulated by the model with two different bedload transport models: a non-equilibrium approach considering saltation distance and an equilibrium approach with the incorporation of the effect of the local bed slope. The advancement of this study is that it comprised detailed analysis of the bed shear stress and sediment transport over a dune during time varying flows, comparing both transport models and revealing distinctive transient features of bedload transport and dune evolution. We also improved a non-equilibrium transport model by incorporating a formulation for sediment saltation distance as a length-scale. Using both sediment transport approaches, the morphological model was applied to replicate a large-scale field experiment. Notably, the model yielded the quasi-equilibrium dune feature reasonably well using either sediment-transport formulation. However, detailed analysis of simulated spatial and temporal features of sediment transport and the dune evolution process were found to be noticeably different. Even though both sediment transport mechanisms appear to work adequately at a basic level, the major difference between the two approaches is the underlying transport process over the dunes and time-scale of the dune evolution process.

Water Resources Research↗
USGS2019

Fishway entrance gate experiments with adult American Shad

The goal of this multiyear study was to examine how changes to an upstream fishway entrance impacted the passage rate of adult American shad ( Alosa sapidissima ). We evaluated a total of nine treatment conditions that consisted of three fishway entrance gate types and three submergence depths (i.e., the water surface elevation of the tailwater relative to the height of the gate crest). Approximately 2,000 wild, actively migrating shad participated in one of the 64 total trials (6–8 trials per treatment) that were conducted for this experiment. The three fishway entrance gate types were the vertical gate (the most common entrance gate type used in fishways), the overshot gate (less common in fishways), and the reversed overshot gate (novel to fishways). The submergence depths ranged from 30.5 to 91.4 cm above the gate crest. Increases to the submergence depth were shown to be the most influential predictor variable of passage time, followed by gate type and river temperature. The reversed overshot and overshot gates outperformed the vertical gate with the best performance occurring for the newly introduced reversed overshot gate type. The results of this study provide guidance on methods to improve fishway attraction and entry rates to numerous state and federal resource agencies and the hydropower industry.

Water Resources Research↗
USGS2019

Physical controls on salmon redd site selection in restored reaches of a regulated, gravel-bed river

Large‐scale river restoration programs have emerged recently as a tool for improving spawning habitat for native salmonids in highly altered river ecosystems. Few studies have quantified the extent to which restored habitat is utilized by salmonids, which habitat features influence redd site selection, or the persistence of restored habitat over time. We investigated fall‐run Chinook salmon spawning site utilization and measured and modeled corresponding habitat characteristics in two restored reaches: a reach of channel and floodplain enhancement completed in 2013 and a reconfigured channel and floodplain constructed in 2002. Redd surveys demonstrated that both restoration projects supported a high density of salmon redds, 3 and 14 years following restoration. Salmon redds were constructed in coarse gravel substrates located in areas of high sediment mobility, as determined by measurements of gravel friction angles and a grain entrainment model. Salmon redds were located near transitions between pool‐riffle bedforms in regions of high predicted hyporheic flows. Habitat quality (quantified as a function of stream hydraulics) and hyporheic flow were both strong predictors of redd occurrence, though the relative roles of these variables differed between sites. Our findings indicate that physical controls on redd site selection in restored channels were similar to those reported for natural channels elsewhere. Our results further highlight that in addition to traditional habitat criteria (e.g., water depth, velocity, and substrate size), quantifying sediment texture and mobility, as well as intragravel flow, provides a more complete understanding of the ecological benefits provided by river restoration projects.

Water Resources Research↗
USGS2019

Prediction and inference of flow-duration curves using multi-output neural networks

We develop multi-output neural network models (MNNs) to predict flow-duration curves (FDCs) in 9,203 ungaged locations in the Southeastern United States for six decades between 1950-2009. The model architecture contains multiple response variables in the output layer that correspond to individual quantiles along the FDC. During training, predictions are made for each quantile, and a combined loss function is used for back propagation and parameter updating. The loss function accounts for the covariance between the quantiles and generates physically consistent outputs (i.e., monotonically increasing quantiles with increasing nonexceedance probabilities). We use neural-network dropout to generate posterior-predictive distributions for FDCs, and test model performance under cross validation. Finally, we demonstrate how local surrgotate models, via the Local Interpretable Model-agnostic Explanations (LIME) method, can be used to infer the relation between basin characteristics and the predicted FDCs. Results suggest that MNNs can learn the monotonic relations between adjacent quantiles on an FDC, they result in better predictions than single output neural-network models that predict each quantile independently, and basin characteristics are most useful for predicting smaller quantiles, whereas bias terms from neighboring quantiles are most informative for predicting higher quantiles.

Water Resources Research↗