Watershed water quality models are mathematical tools used to simulate processes related to water, sediment, and nutrients. These models provide a framework that can be used to inform decision-making and the allocation of resources for watershed management. Therefore, it is critical to answer the question “when is a model good enough?” Established performance evaluation criteria, or thresholds for what is considered a ‘good’ model, provide common benchmarks against which model performance can be compared. Since the publication of prior meta-analyses on this topic, developments in the last decade necessitate further investigation, such as the advancement in high performance computing, the proliferation of aquatic sensors, and the development of machine learning algorithms. We surveyed the literature for quantitative model performance measures, including the Nash-Sutcliffe efficiency (NSE), with a particular focus on process-based models operating at fine temporal scales as their performance evaluation criteria are presently underdeveloped. The synthesis dataset was used to assess the influence of temporal resolution (sub-daily, daily, and monthly), calibration duration (< 3 years, 3 to 8 years, and > 8 years), and constituent target units (concentration, load, and yield) on model performance. The synthesis dataset includes 229 model applications, from which we use bootstrapping and personal modeling experience to establish sub-daily and daily performance evaluation criteria for flow, sediment, total nutrient, and dissolved nutrient models. For daily model evaluation, the NSE for sediment, total nutrient, and dissolved nutrient models should exceed 0.45, 0.30, and 0.35, respectively, for ‘satisfactory’ performance. Model performance generally improved when transitioning from short (< 3 years) to medium (3 to 8 years) calibration durations, but no additional gain was observed with longer (> 8 years) calibration. Dissolved nutrient models calibrated to load (e.g., kg/s) out-performed those calibrated to concentration (e.g., mg/L), whereas selection of target units was not significant for sediment and total nutrient models. We recommend the use of concentration rather than load as a water quality modeling target, as load may be biased by strong flow model performance whereas concentration provides a flow-independent measure of performance. Although the performance criteria developed herein are based on process-based models, they may be useful in assessing machine learning model performance. We demonstrate one such assessment on a recent deep learning model of daily nitrate prediction across the United States. The guidance presented here is intended to be used alongside, rather than to replace, the experience and modeling judgement of engineers and scientist who work to maintain our collective water resources.

The unit sediment graph approach, analogous to the unit hydrograph method, was rarely applied in the past 50 years, presumably due to limitations from scaling the sediment kernel. We hypothesised that spatially explicit sediment connectivity modelling might be combined with unit sediment graph theory to estimate sediment source zones and time of mobilisation across the watershed and estimate sediment flux for hydrologic events. We formulated the model using the probability of sediment connectivity with log-normal parameterisation of the 1-h unit sediment graph. Simulations were carried out for a sediment transport application in a third-order watershed in Kentucky, USA, using a two-stage calibration procedure assisted by a high-performance computing cluster. Results showed sufficient evidence for the efficacy of the approach, including Nash-Sutcliffe Efficiency as high as 0.87 and 0.84 in Stages 1 and 2, respectively, of calibration and 0.88 for model validation. Results of the probability of connectivity showed variability across and within transport events, and 7.5% connectivity for the high flow isolated event. The log-normal distribution effectively estimated the rising limb and the falling limb of the sediment graphs. Post-processing of modelling results showed the importance of the probability of sediment connectivity, as simulations omitting it produced inadequate results. Post-processing with a shallow artificial neural network model showed that both sediment connectivity and surface runoff control sediment yield at the event scale. Results showed the ability of the hourly time step to capture the onset of sediment connectivity and peak connectivity across the ephemeral network.

Hydrologic controls on the timing of sediment transport and sediment hysteresis patterns remain an open area of investigation in hydrology, especially for low-gradient watersheds with substantial instream sediment deposition. Sediment hysteresis, which describes the mismatch between hydrograph peak and sedigraph peak, aids with elucidation of the mechanisms of sediment transport in watersheds. Most frequently, the controls of hysteresis are attributed to the proximity of sediment sources to monitoring locations in a watershed. However, this assumption, while widely applied, is infrequently verified. We investigated the controls of sediment hysteresis in a low gradient system located in the Bluegrass Region of central Kentucky, USA. Turbidity and conductivity sensors installed at the basin outlet provided data to quantify sediment hysteresis and separate hydrologic flow pathways (i.e., by describing the source of water delivered to the watershed's outlet) using a tracer-based approach. Predictive hydrologic parameters, including hydrologic pathways, event magnitude, and antecedent conditions, were estimated and grouped based on hydrologic similitude. Thereafter, we identified parameters required to predict sediment hysteresis using a tailored ensemble feature selection approach coupled with three machine learning algorithms-Random Forest, K-Nearest Neighbors, and Gradient Boosted Trees. Results from the analysis of 68 storm events occurring over a two-year period showed that clockwise events accounted for 85 % of the total sediment yield despite comprising only 53 % of the events. The hysteresis index (HI) can be predicted (r = 0.8, RMSE = 0.12) using three, out of the thirty-nine hydrologic parameters considered. The most important predictors of HI reflect the volume of event rainfall and the relative proportions of new water (i.e., water derived from precipitation during the storm event) and old water (i.e., water previously stored in the watershed) comprising the hydrograph. Further analyses reveal that new water timing-which changes with the rainfall volume-and sediment timing are closely linked, suggesting that variations in the hysteresis patterns are controlled by changes in the response time of fast flowing water pathways. This implies that hydrologic pathways, as opposed to sediment proximity to the watershed outlet, control sediment hysteresis in this watershed. These results have important implications for better understanding the mechanisms controlling sediment transport at the watershed scale.

Understanding the physics of nitrate contamination in surface and subsurface water is vital for mitigating downstream water quality impairment. Though high frequency sensor data have become readily available and computational models more accessible, the integration of these two methods for improved prediction is underdeveloped. The objective of this study was to utilize high‐frequency data to advance our understanding and model representation of nitrate transport for an agricultural karst spring in Kentucky, USA. We collected 2‐years of 15‐min nitrate and specific conductance data and analyzed source‐timing dynamics across dozens of events to develop a conceptual model for nitrate hysteresis in karst. Thereafter, we used the sensing data, specifically discharge‐concentration indices, to constrain modeled nitrate prediction bounds as well as the uncertainty of hydrologic and nitrogen processes, such as soil percolation and biogeochemical transformation. Observed nitrate hysteresis behavior at the spring was complex and included clockwise ( n = 11), counterclockwise ( n = 13), and figure‐eight ( n = 10) shapes, which contrasts with surface systems that are often dominated by a single hysteresis shape. Sensing results highlight the importance of antecedent connectivity to nitrate‐rich storages in determining the timing of nitrate delivery to the spring. After integrating hysteresis analysis into our numerical model evaluation, simulated nitrate prediction bounds were reduced by 43 ± 12% and parameter uncertainty by 36 ± 20%. Taken together, this study suggests that discharge‐concentration indices derived from high‐frequency sensor data can be successfully integrated into numerical models to improve process representation and reduce modeled uncertainty.

The -1 power region of the turbulence spectrum is a constant energy region responsible for energy production. However, its scaling and their implications are underreported. Experiment results show that the production region comprises 42% of turbulence kinetic energy for open-channel flows over fine gravel and very coarse gravel beds. Energy of the production region scales similarly for these gravel bed types, and enables its prediction using logarithmic and exponential equations. Similarity scaling is evidenced for the production region's low and high wavenumber boundaries, defined as the macroturbulence and bursting wavenumbers. The macroturbulence and bursting wavenumbers have similar functions, suggesting that the production region is controlled by the same phenomena across its extent. These results are contrary to those of previous research, in which low and high wavenumber boundaries were hypothesized to be scaled by outer and inner scales, respectively. The authors introduce the macroturbulence streampower, which is derived using the energy similarity and macroturbulence wavenumber. The macroturbulence streampower is distributed across the flow depth, is dependent on roughness element, and agrees with Bagnold's streampower near the channel half-height. (C) 2022 American Society of Civil Engineers.

Karst aquifers are susceptible to contamination by pathogenic microorganisms, such as those found in human and animal waste, because the surface and subsurface drainage are well integrated through dissolution features. Fecal contamination of water is commonly assessed by the concentration of thermotolerant coliform bacteria, especially E. coli. This method is time-consuming, taking ≥18 h between the start of incubation and subsequent enumeration, as well as the time required to collect and transport samples. We examined the utility of continuous monitoring of tryptophan-like fluorescence (TLF) as a real-time proxy for E. coli in a mixed-land-use karst basin in the Inner Bluegrass region of central Kentucky (USA). Two logging fluorometers were sequentially deployed at the outlet spring. During storm flow, TLF typically peaked after discharge, which suggests that TLF transport in the phreatic conduit is likely related to sediment transport. The ability of TLF and other parameters (48 h antecedent precipitation, turbidity, and air temperature) to predict E. coli concentrations was assessed using the Akaike information criterion (AIC) applied to linear regression models. Because both the models and baseline concentrations of TLF differed between fluorometers, TLF and instrument interaction were accounted for in the AIC. TLF was positively correlated with E. coli and, in conjunction with antecedent precipitation, was the best predictor of E. coli. However, a model that included air temperature and antecedent precipitation but not TLF predicted E. coli concentrations similarly well. Given the expense of the fluorometers and the performance of the alternate model, TLF may not be a cost-effective proxy for E. coli in this karst basin.

&lt;p&gt;Karst characterizes almost 15% of the worlds terrain, however the mechanics of sediment transport and its prediction in karst river and cave systems remains underdeveloped. Hysteresis analysis has recently been used more to investigate the behaviors of sediments during storm events in surface systems and to some extent in karst systems. Historically, clockwise and counter-clockwise hysteresis typically refer to proximal and distal sourcing for streams. For karst systems, clockwise and counter-clockwise hysteresis has been identified to refer to an saturated and unsaturated aquifer prior to the event.&lt;/p&gt; &lt;p&gt;However, most interpretation of hysteresis assumes a single dominant water source, for example runoff, and assumes that baseflow is not contributing to the sediment load. One aspect of sediment hysteresis and its interpretation that has received less attention is the occurrence of several, significant water sources, eroding and delivering sediment to the watershed outlet. It is common for both surface stream systems &amp;#160;and karst subsurface systems to have multiple water sources contributing to the total sediment load. Each of the sources carries their own sediment time distribution, and often lead to complex hysteresis looping behavior after mixing. The primary goals of this work are to (1) study how the complex source water-sediment mixing processes impact hysteresis results and (2) to carry out solutions to the water-sediment mixing processes for karst streams, caves, and springs and show the utility and uncertainty of the method.&lt;/p&gt; &lt;p&gt;Several high-resolution sensors have collected data at a karst spring in central Kentucky, USA, for a 2.5 year period. Water unmixing was performed using electrical conductivity as a tracer to separate the groundwater from the surface water and infer sediment sources. Theoretical analyses have shown that not only timing and magnitude of sedigraphs influence the result of the hysteresis loops, but also timing and magnitude of each of the multiple water sources have a strong effect on the resulting hysteresis loop. The groundwater flow shows to have dominant counter-clockwise hysteresis loop, surface water shows to have clockwise loops dominating. Depending on the timing and magnitude of the water sources, the hysteresis loop at the karst spring varies from strictly counterclockwise, to a figure-8 loop, to a complex pattern.&lt;/p&gt;

Hydrologic models are robust tools for estimating key parameters in the management of water resources, including water inputs, storage, and pathway fluxes. The selection of process-based versus data-driven modeling structure is an important consideration, particularly as advancements in machine learning yield potential for improved model performance but at the cost of lacking physical analogues. Despite recent advancement, there exists an absence of cross-model comparison of the tradeoffs between process-based and data-driven model types in settings with varying hydrologic controls. In this study, we use physically-based (SWAT), conceptually-based (LUMP), and deep-learning (LSTM) models to simulate hydrologic pathway contributions for a fluvial watershed and a karst basin over a twenty-year period. We find that, while all models are satisfactory, the LSTM model outperformed both the SWAT and LUMP models in simulating total discharge and that the improved performance was more evident in the groundwater-dominated karst system than the surface-dominated fluvial stream. Further, the LSTM model was able to achieve this improved performance with only 10–25% of the observed time-series as training data. Regarding pathways, the LSTM model coupled with a recursive digital filter was able to successfully match the magnitude of process-based estimates of quick, intermediate, and slow flow contributions for both basins (ρ ranging from 0.58 to 0.71). However, the process-based models exhibited more realistic time-fractal scaling of hydrologic flow pathways compared to the LSTM model which, depending on project objectives, presents a potential drawback to the use of machine learning models for some hydrologic applications. This study demonstrates the utility and potential extraction of physical-analogues of LSTM modeling, which will be useful as deep learning approaches to hydrologic modeling become more prominent and modelers look for ways to infer physical information from data-driven predictions.

Nitrogen removal rates can vary with time, space, and external environmental drivers, but are underreported for karst environments. We carried out a multi‐year study of a karst conduit where we: (a) measured inputs and outputs of sediment nitrogen (SN and δ15NSed) and nitrate (NO3− and δ15NNO3); (b) developed, calibrated, and applied a numerical model of nitrogen physics and biogeochemistry; and (c) forecasted the impacts of climate and land use changes on nitrate removal and export. Data results from conduit inputs (SN = 0.43% ± 0.07%, δ15NSed = 5.07‰ ± 1.01‰) and outputs (SN = 0.36% ± 0.09%, δ15NSed = 6.45‰ ± 0.71‰) indicate net‐mineralization of SN and increase of δ15NSed (p < 10−2). However, δ15NSed increase cannot be explained by SN mineralization alone and is instead accompanied by immobilization of isotopically heavier mineral nitrogen (δ15NNO3 = 11.25‰ ± 6.96‰). Modeled SN and δ15NSed sub‐routines provided a boundary condition for DIN simulation and improved NO3− model performance (from NSE = 0.06 to NSE = 0.68). Modeled spatial zones of removal occur in close proximity to conduit entrances, where deposition of labile organic matter promotes a three‐fold increase in denitrification (∼60 mg N m−2 d−1). Modeled temporal periods of removal occur during the dry‐season where longer residence times cause up to 90% removal of NO3− inputs. Projected effects of environmental drivers suggest an increase in denitrification (+14.1%); however, this removal is largely offset by greater nitrate soil leaching (+28.1%) from wetter regional climate. Results suggest that conduits underlying mature karst terrain experience spatiotemporal removal gradients, which are modulated by solute and sediment delivery.

The fate of sedimentary carbon in rivers is determined by a combination of mineral protection and transit time. Along the fluvial journey from headwaters to sea, biogeochemical transformations control whether carbon is buried or returned to the atmosphere as CO2.

Forecasting change in sediment transport using global climate model ensembles is under-developed in the hydrology community due to a lack of knowledge regarding the variance structure of predictions. We investigate the uncertainty of forecast sediment transport from sources, including global climate model realizations, global climate model ensemble design, forecasted hydrologic inputs, and hydrologic modeling parameterization. We then forecast sediment transport for the gently rolling watershed in Kentucky, USA. Contrary to past research forecasting hydrology with global climate models, hydrologic model parameterization was the most significant source of variance impacting forecasted sediment transport variables. Forecast of sediment transport responses shows the propagation of uncertainty from the hydrologic model parameterization was over two times greater than the uncertainty from the selected global climate model realizations. This result emphasizes researchers focused on forecasting sediment transport with global climate models may need to give as much, or more, consideration to their water-sediment linkages as climate-water-sediment linkages. Hydrologic inputs from climate change, including forecast precipitation, temperature, relative humidity, solar radiation, and wind speed, impacted sediment transport. Considering changes in precipitation and temperature alone under-predicts streamflow and sediment transport by 49% and 35%, respectively, compared to including all meteorological variables inputs. Variance introduced across the three different global climate model ensembles was a relatively small source of variance impacting forecasted streamflow or sediment yield. The results suggest a quantitative effort by the researcher to design the global climate model ensemble by considering representativeness, historical performance, and independency will lead to robust results. Ensemble average forecasts forecast streamflow and sediment yield to increase by 19% and 14%, respectively, for the lowland watershed. The sediment transport forecasts reflect the shear-limited landscape of the watershed and the transport-limited conditions in the stream channel.

Karst pathways and fluvial pathways control hydrology in shallow fluviokarst basins, and numerical modelling of fluviokarst is seldom reported. In this study, we developed a combined discrete-continuum fluviokarst numerical model by simulating surface river routing, in-stream swallet sources and sinks, epikarst storage and dynamic transfer, matrix bedrock interactions, and closed conduit phreatic flow. We applied the model to the Cane Run-Royal Spring basin in Kentucky, USA. Model evaluation indicated that spring discharge alone inadequately constrained model pathways and uncertainty. Instead multi-objective calibration, integrating riverine discharge and well-head data from multiple locations, assisted in identifying sensitive parameters (p < 0.05). Multi-objectives improved representation of stream-cave connectivity and limited prior knowledge biases of the system but was computationally expensive with 168 h required on a high-performance cluster. Results provided evidence for a mature fluviokarst basin with well-defined fracture-conduit network and phreatic aquifer. Residence times of karst pathways vary by five orders of magnitude, ranging from less than one hour in vertical swallets, 12.7 h in longitudinal conduits, 12.7 days in the vadose zone and epikarst, and 142.7 days in the bedrock matrix. Results suggest arrival of source waters to the subsurface systems is disconnected in time from the springflow response. Model simulations show a dimensionless vertical to longitudinal conveyance ratio helps predict swallets linking the fluvial and karst systems. Transferability of the developed model, and other karst models, is discussed relative to availability of information for karst basins.

Coupling erosion formulae with sediment connectivity methods is one promising approach to better represent structural and functional variability of sediment processes. To advance this goal, the probability of connectivity approach is coupled with the revised universal soil loss equation (RUSLE) in a basin with forest and reclaimed mine landuses. Model evaluation showed unforeseen codependency between connectivity formulae and RUSLE. For example, the RUSLE P factor was codependent with the probability of downstream transport within the connectivity formula. Researchers should use feedback calibration schemes to resolve lack of model independence. Connectivity modelling advanced prediction of sediment processes because it simulated the unforeseen impact of legacy terracing on sediment connectivity and soil loss. Structural control dominates connectivity in this study, and soil loss and connectivity are self-similar. The structural control is contrary to recent suggestions that functional, dynamic processes control sediment connectivity in all landscapes. Self-similarity also remains an open topic because a number of studies show poor correlation between soil loss and connectivity. On average 12% of forested land and 47% of reclaimed mine land was connected for events studied. Predicted soil loss rates in the reclaimed mine were approximately 30 times greater than the forest land despite the fact that the reclamation is classified as phase 3. Spatially explicit results highlight pathways that should be targeted for remediation, and this study supports the idea of the Forestry Reclamation Approach for remediation of excess soil loss.

Integrating connectivity theory within watershed modelling is one solution to overcome spatial and temporal shortcomings of sediment transport prediction, and Part I and II of these companion papers advance this overall goal. In Part I of these companion papers, we present the theoretical development of probability of connectivity formula considering connectivity's magnitude, extent, timing and continuity that can be applied to watershed modelling. Model inputs include a high resolution digital elevation model, hydrologic watershed variability, and field connectivity assessments. We use the model to investigate the dependence of the probability of connected timing and spatial connectivity on sediment transport predictors. Results show the spatial patterns of connectivity depend on both structural and functional characteristics of the catchment, such as hillslope gradient, upstream contributing area, soil texture, and stream network configuration (structural) and soil moisture content and runoff generation (functional). Spatial connectivity changes from catchment-to-catchment as a function of soil type and drainage area; and it varies from event-to-event as a function of runoff depth and soil moisture conditions. The most sensitive connected pathways provide the stencil for the probability of connectivity, and pathways connected from smaller hydrologic events are consistently reconnected and built upon during larger hydrologic events. Surprisingly, we find the probability of connected timing only depends on structural characteristics of catchments, which are considered static over the timescales analyzed herein. The timing of connectivity does not statistically depend on functional characteristics, which relaxes the parameterization across events of different magnitudes. This result occurs because the pathway stencil accumulates sediment from adjacent soils as flow intensity increases, but this does not statistically shift the frequency distribution.

The Bluegrass Region is an area in north-central Kentucky with unique natural and cultural significance, which possesses some of the most fertile soils in the world. Over recent decades, land use and land cover changes have threatened the protection of the unique natural, scenic, and historic resources in this region. In this study, we applied a fragmentation model and a set of landscape metrics together with the satellite-derived USDA Cropland Data Layer to examine the shrinkage and fragmentation of grassland in the Bluegrass Region, Kentucky during 2008–2018. Our results showed that recent land use change across the Bluegrass Region is characterized by grassland decline, cropland expansion, forest spread, and suburban sprawl. The grassland area decreased by 14.4%, with an interior (or intact) grassland shrinkage of 5%, during the study period. Land conversion from grassland to other land cover types has been widespread, with major grassland shrinkage occurring in the west and northeast of the Outer Bluegrass Region and relatively minor grassland conversion in the Inner Bluegrass Region. The number of patches increased from 108,338 to 126,874. The effective mesh size, which represents the degree of landscape fragmentation in a system, decreased from 6629.84 to 1816.58 for the entire Bluegrass Region. This study is the first attempt to quantify recent grassland shrinkage and fragmentation in the Bluegrass Region. Therefore, we call for more intensive monitoring and further conservation efforts to preserve the ecosystem services provided by the Bluegrass Region, which has both local and regional implications for climate mitigation, carbon sequestration, diversity conservation, and culture protection.

Integrating connectivity theory within watershed modelling is one solution to overcome spatial and temporal shortcomings of sediment transport prediction, and Part I and II of these companion papers advance this overall goal. In Part II of these companion papers, we investigate sediment flux via connectivity formula discretized over many catchments and then integrated via sediment routing; and we advance model evaluation technology by using hysteresis of sensor data. Model evaluation with hysteresis indices provides nearly a 100% increase in model statistics. Hysteresis loop evaluation shows a shift from near linear behavior at low to moderate events and then clock-wise loops for larger events indicating the importance of proximal sediment sources. Catchment-scale sediment flux varies as function of the probability of timing and extent of connectivity of an individual catchment. Watershed-scale sediment flux shows self-similarity for the main stem of the river channel as the 181 catchments are integrated moving down gradient. Sediment flux varies from event-to-event as a function of the most sensitive connected pathways, including ephemeral gullies and roadside ditches in this basin. These sensitive pathways contribute disproportionately large amounts to overall sediment yield regardless of the total rainfall depth. Prediction requires the connectivity formula, erosion formula and sediment routing formula; and the probability of connectivity alone was a poor predictor for sediment transport. The result highlights the importance of coupling connectivity simulations with sediment transport formula, and our method provides one such approach.

Excessive nitrate threatens a wide range of water resources, aquatic habitats, and sensitive infrastructure. Despite this problem, tracing a nutrient from its eventual fate back to its origin remains an elusive challenge due to heterogeneity in how nutrient sources and hydrologic pathways are connected. Typically, this problem is underdetermined (i.e., too many unknowns, not enough equations) and cannot be solved with existing methodologies. The theory of optimal transport allows for the solution of underdetermined systems, and here we construct a novel formulation for its use in water quality modeling. Our objective was to develop an optimal transport modeling framework—coupled to Bayesian source unmixing, loadograph pathway separation, and geospatial connectivity analysis—to apportion nitrate loading from three sources (soil, fertilizer, and manure) across three pathways (quick, intermediate, and slow), resulting in nine possible source‐pathway couplings (soil‐quick, soil‐intermediate, …, manure‐slow). We apply this model to a 30 month elemental (NO3−) and isotopic (δ15N and δ18O) nitrate data set from a karst watershed in Kentucky, USA. Modeling results indicate that—of the nine possible source‐pathway couplings—nearly 60% of nitrate export is facilitated by just three: fertilizer‐quick (16.4%), manure‐intermediate (15.4%), and soil‐slow (27.2%). Further, we reinforce the need to explicitly consider heterogeneity in source‐pathway connectivity as homogeneous assumptions lead to erroneous inferences. The applicability of the model, its input requirements, and transferability to other sites is discussed. Lastly, we simulated two land management scenarios (field buffers and septic repair) and demonstrate how optimal transport can be used to test nutrient reduction strategies.

Nitrate (NO3-) fate estimates in turbulent karst pathways are lacking due, in part, to the difficulty of accessing remote subsurface environments. To address this knowledge and methodological gap, we collected NO3-, (delta N-15(NO3), and delta O-18(NO3) data for 65 consecutive days, during a low-flow period, from within a phreatic conduit and its terminal end-point, a spring used for drinking water. To simulate nitrogen (N) fate within the karst conduit, the authors developed a numerical model of NO3- isotope dynamics. During low-flow, data show an increase in NO3- (from 1.78 to 1.87 mg N L-1; p < 10(-4)) coincident with a decrease in delta N-15(NO3)(from 7.7 to 6.8 parts per thousand; p < 10(-3)) as material flows from within the conduit to the spring. Modeling results indicate that the nitrification of isotopically-lighter ammonium (delta N-15(NH4)) acts as a mechanism for an increase in NO3- that coincides with a decrease in delta N-15(NO3). Further, numerical modeling assists with quantifying isotopic overprinting of nitrification on denitrification (i.e., coincident NO3- production during removal) by constraining the rates of the two processes. Modeled denitrification fluxes within the karst conduit (67.0 +/- 19.0 mg N m(-2) d (-1)) are an order-of-magnitude greater than laminar ground water pathways (1-10 mg N m(-2) d(-1)) and an order-of-magnitude less than surface water systems (100-1000 mg N m(-2) d(-1)). In this way, karst conduits are a unique interface of the processes and gradients that control both surface and ground water end-points. This study shows the efficacy of ambient N stable isotope data to reflect N transformations in subsurface karst and highlights the usefulness of stable isotopes to assist with water quality numerical modeling in karst. Lastly, we provide a rare, if not unique, estimate of N fate in subsurface conduits and provide a counterpoint to the paradigm that karst conduits are conservative source-to-sink conveyors. (C) 2019 Elsevier Ltd. All rights reserved.

Atmospheric rivers and tropical cyclones originate in the tropics and can transport high rainfall amounts to inland temperate regions. The purpose of this study was to investigate the response of nitrate (NO3−) pathways, concentration peaks, and stable isotope (δ15NNO3, δ18ONO3, δ2HH2O, δ18OH2O, and δ13CDIC) measurements to these extreme events. A tropical cyclone and atmospheric river produced the number one and four ranked events in 2017, respectively, at a Kentucky USA watershed characterized by mature karst topography. Hydrologic responses from the two events were different due to rainfall characteristics with the tropical cyclone producing a steeper rising limb of the spring hydrograph and greater runoff generation to the surface stream compared to the atmospheric river. Local minima and maxima of specific conductance, δ2HH2O, δ18OH2O, and δ13CDIC coincided with hydrograph peaks for both events. Minima and maxima of NO3−, δ15NNO3, δ18ONO3, and temperature lagged behind the hydrograph peak for both events, and the values continued to be impacted by diffuse recharge during hydrograph recession. Quick-flow pathways accounted for less than 20% of the total NO3− yield, while intermediate (30%) and slow-flow (50%) pathways composed the remaining load. However, hydrograph separation into quick-, intermediate-, and slow-flow pathways was not able to predict the timing of NO3− concentration peaks. Rather, the intermediate-flow pathway is conceptualized to experience a shift in porosity, associated with a change from epikarst macropores and fissures to soil micropores, with the arrival of water from the latter component likely causing peak NO3− concentration at the spring. Our results suggest that a more discretized conceptual model of pathways may be needed to predict peak nutrient concentration in rivers draining karst topography.