Exploring Persistence in Streamflow Forecasting

In this study, the authors explore three persistence approaches in streamflow forecasting motivated by the need for forecasting model skill evaluation. The authors use streamflow observations with 15 min resolution from the year 2008 to 2017 at 140 United States Geological Survey streamflow gauges monitoring the streams and rivers over the State of Iowa. The spatial scale of the basins ranges from about 7 to 37,000 km². The study explores three approaches: simple persistence, gradient persistence, and anomaly persistence. The study shows that persistence forecasts skill has strong dependence on basin scales and weaker but non‐negligible dependence on geometric properties of the river network for a given basin. Among the three approaches explored, anomaly persistence shows highest skill especially for small basins, under about 500 km². The anomaly persistence can serve as a benchmark for model evaluations considering the effect of basin scales and geometric properties of river network of the basin. This study further reiterates that persistence forecasts are hard‐to‐beat methods for larger basin scales at short to medium forecast range.     

Mapping Outlets of Iowa Flood Center and National Water Center River Networks for Hydrologic Model Comparison

River networks based on Digital Elevation Model (DEM) data differ depending on the DEM resolution, accuracy, and algorithms used for network extraction. As spatial scale increases, the differences diminish. This study explores methods that identify the scale where networks obtained by different methods agree within some margin of error. The problem is relevant for comparing hydrologic models built around the two networks. An example is the need to compare streamflow prediction from the Hillslope Link Model (HLM) operated by the Iowa Flood Center (IFC) and the National Water Model (NWM) operated by the National Water Center of the National Oceanic and Atmospheric Administration. The HLM uses landscape decomposition into hillslopes and channel links while the NWM uses the NHDPlus dataset as its basic spatial support. While the HLM resolves the scale of the NHDPlus, the outlets of the latter do not necessarily correspond to the nodes of the HLM model. The authors evaluated two methods to map the outlets of NHDPlus to outlets on the IFC network. The methods compare the upstream areas of the channels and their spatial location. Both methods displayed similar performance and identified matches for about 80% of the outlets with a tolerance of 10% in errors in the upstream area. As the aggregation scale increases, the number of matches also increases. At the scale of 100 km², 90% of the outlets have matches with tolerance of 5%. The authors recommend this scale for comparing the HLM and NWM streamflow predictions.     

PRISM vs. CFSR Precipitation Data Effects on Calibration and Validation of SWAT Models

Precipitation is one of the most important drivers in watershed models. Our objective was to compare two sources of interpolated precipitation data in terms of their effect on calibration and validation of two Soil and Water Assessment Tool (SWAT) models. One model was a suburban watershed in metropolitan Atlanta, Georgia. The precipitation sources were Parameter‐elevation Relationships on Independent Slopes Model (PRISM) data on a 4‐km grid and climate forecast system reanalysis (CFSR) data on a 38‐km grid. The PRISM data resulted in a better fit to the calibration data (Nash Sutcliffe efficiency [NSE] = 0.64, Kling‐Gupta efficiency [KGE] = 0.74, p‐factor = 0.84, and r‐factor = 0.43) than the CFSR data (NSE = 0.47, KGE = 0.53, p‐factor = 0.67, and r‐factor = 0.39). Validation results were similar. Sensitive parameters were similar in both the PRISM and CFSR models, but fitted values indicated more rapid groundwater flow to the streams with the PRISM data. The same comparison was made in the Big Creek watershed located approximately 1,000 km away, in central Louisiana. Results were similar with a more responsive groundwater system indicating PRISM data may produce better predictions of streamflow because of a more accurate estimate of rainfall within a watershed or because of a denser grid. Our study implies PRISM is providing a better estimate than CFSR of precipitation within a watershed when rain gauge data are not available, resulting in more accurate simulations of streamflows at the watershed outlet. Editor's note: This paper is part of the featured series on SWAT Applications for Emerging Hydrologic and Water Quality Challenges. See the February 2017 issue for the introduction and background to the series.     

Neural Network Input Selection for Hydrological Forecasting Affected by Snowmelt1

Abstract: Snowmelt largely affects runoff in watersheds in Nordic countries. Neural networks (NN) are particularly attractive for streamflow forecasting whereas they rely at least on daily streamflow and precipitation observations. The selection of pertinent model inputs is a major concern in NNs implementation. This study investigates performance of auxiliary NN inputs that allow short‐term streamflow forecasting without resorting to a deterministic snowmelt routine. A case study is presented for the Rivière des Anglais watershed (700 km²) located in Southern Québec, Canada. Streamflow (Q), precipitations (rain R and snow S, or total P), temperature (T) and snow lying (A) observations, combined with climatic and snowmelt proxy data, including snowmelt flow (Q_SM) obtained from a deterministic model, were tested. NN implemented with antecedent Q and R produced the largest gains in performance. Introducing increments of A and T to the NNs further improved the performance. Long‐term averages, seasonal data, and Q_SM failed to improve the networks.     

HYDROLOGIC IMPACTS OF A LARGE-SCALE MOUNTAIN PINE BEETLE (DENDROCTONUS PONDEROSAE HOPKINS) EPIDEMIC1

The Jack Creek watershed, a 133 km² (51.5 mi²) drainage in southwestern Montana, was impacted by a mountain pine beetle (Dendroctonus ponderosae Hopkins) epidemic in 1975–1977 which killed an estimated 35 percent of its total timber. Analyses of USGS streamflow data for four years prior to and five years after mortality suggest a 15 percent post-epidemic increase in annual water yield, a two-to three-week advance in the annual hydrograph, a 10 percent increase in low flows and little increase of peak runoff.     

Channel Roughness in North Carolina Mountain Streams

Channel roughness, often described by Manning's n, is used to represent the amount of resistance that flow encounters, and has direct implications on velocity and discharge. Ideally, n is calculated from a long‐term record of channel discharge and hydraulic geometry. In the absence of these data, a combination of photo references and a validated qualitative method is preferable to simply choosing n arbitrarily or from a table. The purpose of this study was to use United States Geological Survey (USGS) streamflow data to calculate roughness coefficients for streams in the mountains of North Carolina. Five USGS gage stations were selected for this study, representing drainage areas between 71.5 and 337 km². Photo references of the study sites are presented. Measured discharges were combined with hydraulic geometry at a cross‐section to calculate roughness coefficients for flows of interest. At bankfull flow, n ranged between 0.039 and 0.064 for the five study sites. Roughness coefficients were not constant for all flows in a channel, and fluctuated over a large range. At all sites, roughness was highest during low‐flow conditions, then quickly decreased as flow increased, up to the bankfull elevation.     

Process-based calibration of WRF-Hydro in a mountainous basin in southwestern U.S.

The National Water Model (NWM) will provide the next generation of operational streamflow forecasts across the United States (U.S.) using the WRF-Hydro hydrologic model. In this study, we propose a strategy to calibrate 10 parameters of WRF-Hydro that control runoff generation during floods and snowmelt seasons, and due to baseflow. We focus on the Oak Creek Basin (820 km²), an unregulated mountainous sub-watershed of the Salt and Verde River Basins in Arizona, which are the largest source of water supply for the Phoenix Metropolitan area. We calibrate the model against discharge observations at the outlet in 2008–2011, and validate it at two stream gauging stations in 2012–2016. After bias correcting the precipitation forcings, we sequentially modify the model parameters controlling distinct runoff generation processes in the basin. We find that capturing the deep drainage to the aquifer is crucial to improve the simulation of all processes and that this flux is mainly controlled by the SLOPE parameter. Performance metrics indicate that snowmelt, baseflow, and floods due to winter storms are simulated fairly well, while flood peaks caused by summer thunderstorms are severely underestimated. We suggest the use of spatially variable soil depth to enhance the simulation of these processes. This work supports the ongoing calibration effort of the NWM by testing WRF-Hydro in a watershed with a large variety of runoff mechanisms that are representative of several basins in the southwestern U.S.     

BASE FLOW RECESSION RATES,LOW FLOWS,AND HYDROLOGIC FEATURES OF SMALL WATERSHEDS IN PENNSYLVANIA,USA1

This paper examines the relationships between measurable watershed hydrologic features, base flow recession rates, and the Q_7,10 low flow statistic (the annual minimum seven‐day average streamflow occurring once every 10 years on average). Base flow recession constants were determined by analyzing hydrograph recession data from 24 small (>130 km²), unregulated watersheds across five major physiographic provinces of Pennsylvania, providing a highly variable dataset. Geomorphic, hydrogeologic, and land use parameters were determined for each watershed. The base flow recession constant was found to be most strongly correlated to drainage density, geologic index, and ruggedness number (watershed slope); however, these three parameters are intercorrelated. Multiple regression models were developed for predicting the recession rate, and it was found that only two parameters, drainage density and hydrologic soil group, were required to obtain good estimates of the recession constant. Equations were also developed to relate the recession rates to Q_7,10 per unit area, and to the Q_7,10/Q₅₀ ratio. Using these equations, estimates of base flow recession rates, Q_7,10, and streamflow reduction under drought conditions can be made for small, ungaged basins across a wide range of physiography.     

Decoupling Streamflow Responses to Climate Variability and Land Use/Cover Changes in a Watershed in Northern China

Restored annual streamflow (Q_r) and measured daily streamflow of the Chaohe watershed located in northern China and associated long‐term climate and land use/cover data were used to explore the effects of land use/cover change and climate variability on the streamflow during 1961‐2009. There were no significant changes in annual precipitation (P) and potential evapotranspiration, whereas Q_r decreased significantly by 0.81 mm/yr (p < 0.001) over the study period with a change point in 1999. We used 1961‐1998 as the baseline period (BP) and 1999‐2009 the change period (CP). The mean Q_r during the CP decreased by 39.4 mm compared with that in the BP. From 1979 to 2009, the grassland area declined by 69.6%, and the forest and shrublands increased by 105.4 and 73.1%, respectively. The land use/cover change and climate variability contributed for 58.4 and 41.6% reduction in mean annual Q_r, respectively. Compared with the BP, median and high flows in the CP decreased by 38.8 and up to 75.5%, respectively. The study concludes that large‐scale ecological restoration and watershed management in northern China has greatly decreased water yield and reduced high flows due to the improved land cover by afforestation leading to higher water loss through evapotranspiration. At a large watershed scale, land use/cover change could play as much of an important role as climate variability on water resources.     

Modeling Discharge Rates Using a Coupled Modeled Approach for the Merced River in Yosemite National Park

This study describes the application of the NASA version of the Carnegie‐Ames‐Stanford Approach (CASA) ecosystem model coupled with a surface hydrologic routing scheme previously called the Hydrological Routing Algorithm (HYDRA) to model monthly discharge rates from 2000 to 2007 on the Merced River drainage in Yosemite National Park, California. To assess CASA‐HYDRA's capability to estimate actual water flows in extreme precipitation years, the focus of this study is the 2007 water year, which was very dry, and the 2005 water year, which was a moderately wet year in the historical record. Prior to comparisons to gauge records, CASA‐HYDRA snowmelt algorithms were modified with equations from the U.S. Department of Agriculture Snowmelt‐Runoff Model (SRM), which has been designed to predict daily streamflow in mountain basins where snowmelt is a major runoff factor. Results show that model predictions closely matched monthly flow rates at the Pohono Bridge gauge station (USGS#11266500), with R² = 0.67 and Nash‐Sutcliffe (E) = 0.65. By subdividing the upper Merced River basin into subbasins with high spatial resolution in the gridded modeling approach, we were able to determine which biophysical characteristics in the Sierra differed to the largest degree in extreme low‐flow and high‐flow years. Average elevation and snowpack accumulation were found to be the most important explanatory variables to understand subbasin contributions to monthly discharge rates.     

Generation of Ensemble Streamflow Forecasts Using an Enhanced Version of the Snowmelt Runoff Model1

Harshburger, Brian J., Von P. Walden, Karen S. Humes, Brandon C. Moore, Troy R. Blandford, and Albert Rango, 2012. Generation of Ensemble Streamflow Forecasts Using an Enhanced Version of the Snowmelt Runoff Model. Journal of the American Water Resources Association (JAWRA) 48(4): 643‐655. DOI: 10.1111/j.1752‐1688.2012.00642.x Abstract: As water demand increases in the western United States, so does the need for accurate streamflow forecasts. We describe a method for generating ensemble streamflow forecasts (1‐15 days) using an enhanced version of the snowmelt runoff model (SRM). Forecasts are produced for three snowmelt‐dominated basins in Idaho. Model inputs are derived from meteorological forecasts, snow cover imagery, and surface observations from Snowpack Telemetry stations. The model performed well at lead times up to 7 days, but has significant predictability out to 15 days. The timing of peak flow and the streamflow volume are captured well by the model, but the peak‐flow value is typically low. The model performance was assessed by computing the coefficient of determination (R²), percentage of volume difference (Dv%), and a skill score that quantifies the usefulness of the forecasts relative to climatology. The average R² value for the mean ensemble is >0.8 for all three basins for lead times up to seven days. The Dv% is fairly unbiased (within ±10%) out to seven days in two of the basins, but the model underpredicts Dv% in the third. The average skill scores for all basins are >0.6 for lead times up to seven days, indicating that the ensemble model outperforms climatology. These results validate the usefulness of the ensemble forecasting approach for basins of this type, suggesting that the ensemble version of SRM might be applied successfully to other basins in the Intermountain West.     

Evaluation of Short-to-Medium Range Streamflow Forecasts Obtained Using an Enhanced Version of SRM1

Harshburger, Brian J., Karen S. Humes, Von P. Walden, Brandon C. Moore, Troy R. Blandford, and Albert Rango, 2010. Evaluation of Short-to-Medium Range Streamflow Forecasts Obtained Using an Enhanced Version of SRM. Journal of the American Water Resources Association (JAWRA) 46(3):603-617. DOI: 10.1111/j.1752-1688.2010.00437.x Abstract: As demand for water continues to escalate in the western United States, so does the need for accurate streamflow forecasts. Here, we describe a methodology for generating short-to-medium range (1 to 15 days) streamflow forecasts using an enhanced version of the Snowmelt Runoff Model (SRM), snow-covered area data derived from MODIS products, data from Snow Telemetry stations, and meteorological forecasts. The methodology was tested on three mid-elevation, snowmelt-dominated basins ranging in size from 1,600 to 3,500 km². To optimize the model performance and aid in its operational implementation, two enhancements have been made to SRM: (1) the use of an antecedent temperature index method to track snowpack cold content, and (2) the use of both maximum and minimum critical temperatures to partition precipitation into rain, snow, or a mixture of rain and snow. The comparison of retrospective model simulations with observed streamflow shows that the enhancements significantly improve the model performance. Streamflow forecasts generated using the enhanced version of the model compare well with the observed streamflow for the earlier leadtimes; forecast performance diminishes with leadtime due to errors in the meteorological forecasts. The three basins modeled in this research are typical of many mid-elevation basins throughout the American West, thus there is potential for this methodology to be applied successfully to other mountainous basins.     

Effects of Land Use and Sample Location on Nitrate‐Stream Flow Hysteresis Descriptors during Storm Events

The U.S. Geological Survey's New Jersey and Iowa Water Science Centers deployed ultraviolet‐visible spectrophotometric sensors at water‐quality monitoring sites on the Passaic and Pompton Rivers at Two Bridges, New Jersey, on Toms River at Toms River, New Jersey, and on the North Raccoon River near Jefferson, Iowa to continuously measure in‐stream nitrate plus nitrite as nitrogen (NO₃ + NO₂) concentrations in conjunction with continuous stream flow measurements. Statistical analysis of NO₃ + NO₂ vs. stream discharge during storm events found statistically significant links between land use types and sampling site with the normalized area and rotational direction of NO₃ + NO₂‐stream discharge (N‐Q) hysteresis patterns. Statistically significant relations were also found between the normalized area of a hysteresis pattern and several flow parameters as well as the normalized area adjusted for rotational direction and minimum NO₃ + NO₂ concentrations. The mean normalized hysteresis area for forested land use was smaller than that of urban and agricultural land uses. The hysteresis rotational direction of the agricultural land use was opposite of that of the urban and undeveloped land uses. An r² of 0.81 for the relation between the minimum normalized NO₃ + NO₂ concentration during a storm vs. the normalized NO₃ + NO₂ concentration at peak flow suggested that dilution was the dominant process controlling NO₃ + NO₂ concentrations over the course of most storm events.     

Nitrate Dynamics in Two Streams Impacted by Wastewater Treatment Plant Discharge: Point Sources or Sinks?

We examined nitrate processing in headwater stream reaches downstream of two wastewater treatment plant outfalls during low streamflow. Our objectives were to quantify nitrate mass flux before and after effluent discharge and to use field and laboratory techniques to assess the mechanism of nitrate uptake. Microcosm experiments were utilized to determine the location of nitrate processing, and molecular biomarkers were used to detect and quantify microbial denitrification. At one site, downstream nitrate mass flux was significantly (p = 0.01) lower than sum of upstream and wastewater effluent fluxes, indicating rapid stream assimilation of incoming nitrate in the vicinity of the point source. Microcosm experiments supported the theory that nitrate processing occurs in sediments. Molecular assays for denitrifcation‐associated functional genes nosZ, nirS, and nirK, provided evidence that effluent contained enriched denitrifying communities relative to ambient stream water. Nitrate loss at the site with greater uptake was correlated with sulfate loss (p < 0.01; r² = 0.86), suggesting a possible link between sulfate reducing bacteria and denitrifying bacterial communities. Results suggest there is an opportunity to better understand nitrate dynamics in cases where point sources may act as point sinks under specific sets of conditions.     

MODIS Biophysical States and NEXRAD Precipitation in a Statistical Evaluation of Antecedent Moisture Condition and Streamflow1

Abstract: The potential of remotely sensed time series of biophysical states of landscape to characterize soil moisture condition antecedent to radar estimates of precipitation is assessed in a statistical prediction model of streamflow in a 1,420 km² watershed in south‐central Texas, Moderate Resolution Imaging Spectroradiometer (MODIS) time series biophysical products offer significant opportunities to characterize and quantify hydrologic state variables such as land surface temperature (LST) and vegetation state and status. Together with Next Generation Weather Radar (NEXRAD) precipitation estimates for the period 2002 through 2005, 16 raw and deseasoned time series of LST (day and night), vegetation indices, infrared reflectances, and water stress indices were linearly regressed against observed watershed streamflow on an eight‐day aggregated time period. Time offsets of 0 (synchronous with streamflow event), 8, and 16 days (leading streamflow event) were assessed for each of the 16 parameters to evaluate antecedent effects. The model results indicated a reasonable correlation (r² = 0.67) when precipitation, daytime LST advanced 16 days, and a deseasoned moisture stress index were regressed against log‐transformed streamflow. The estimation model was applied to a validation period from January 2006 through March 2007, a period of 12 months of regional drought and base‐flow conditions followed by three months of above normal rainfall and a flood event. The model resulted in a Nash‐Sutcliffe estimation efficiency (E) of 0.45 for flow series (in log‐space) for the full 15‐month period, ?0.03 for the 2006 drought condition period, and 0.87 for the 2007 wet condition period. The overall model had a relative volume error of ?32%. The contribution of parameter uncertainties to model discrepancy was evaluated.     

Heterogeneity of Hydrologic Response in Urban Watersheds1

Meierdiercks, Katherine L., James A. Smith, Mary Lynn Baeck, and Andrew J. Miller, 2010. Heterogeneity of Hydrologic Response in Urban Watersheds. Journal of the American Water Resources Association (JAWRA) 46(6):1221–1237. DOI: 10.1111/j.1752-1688.2010.00487.x Abstract: The changing patterns of streamflow associated with urbanization are examined through analyses of discharge and rainfall records for the study watersheds of the Baltimore Ecosystem Study (BES). Analyses utilize a decade (1999-2008) of observations from a dense U.S. Geological Survey stream gaging network and Hydro-NEXRAD radar rainfall fields. The principal study watershed of the BES is Gwynns Falls (171 km²). Focus is given to two Gwynns Falls basins with contrasting patterns and histories of development, Dead Run and Upper Gwynns Falls. The sharp contrasts in streamflow properties between the basins reflect the differences in urban development prior to implementation of stormwater management regulations (much of Dead Run) and development for which stormwater management is an integral part of the hydrologic system (Upper Gwynns Falls). The mean annual runoff in Dead Run (558 mm) is 35% larger than that of Upper Gwynns Falls; larger contrasts in runoff properties typify the “warm season” and are linked to storm event hydrologic response. Spatial heterogeneities in storm event response are reflected in seasonal and diurnal properties of streamflow. Analyses of storm event response are presented for June 2006, during which monthly rainfall over the BES region ranged from less than 150 to more than 500 mm. Baisman Run, the BES forest reference watershed, and Moores Run, a highly urbanized watershed in Baltimore City, provide “end-member” representations of urban impacts on streamflow.     

Assessing the Impact of Areal Precipitation Input on Streamflow Simulations Using the SWAT Model1

Masih Ilyas, Shreedhar Maskey, Stefan Uhlenbrook, and Vladimir Smakhtin, 2011. Assessing the Impact of Areal Precipitation Input on Streamflow Simulations Using the SWAT Model. Journal of the American Water Resources Association (JAWRA) 47(1):179‐195. DOI: 10.1111/j.1752‐1688.2010.00502.x Abstract: Reduction of input uncertainty is a challenge in hydrological modeling. The widely used model Soil Water Assessment Tool (SWAT) uses the data of a precipitation gauge nearest to the centroid of each subcatchment as an input for that subcatchment. This may not represent overall catchment precipitation conditions well. This paper suggests an alternative – using areal precipitation obtained through interpolation. The effectiveness of this alternative is evaluated by comparing its simulations with those based on the standard SWAT precipitation input procedure. The model is applied to mountainous semiarid catchments in the Karkheh River basin, Iran. The model performance is evaluated at daily, monthly, and annual scales by using a number of performance indicators at 15 streamflow gauging stations each draining an area in the range of 590‐42,620 km². The comparison suggests that the use of areal precipitation improves model performance particularly in small subcatchments in the range of 600‐1,600 km². The modified areal precipitation input results in increased reliability of simulated streamflows in the areas of low rain gauge density. Both precipitation input methods result in reasonably good simulations for larger catchments (over 5,000 km²). The use of areal precipitation input improves the accuracy of simulated streamflows with spatial resolution and density of rain gauges having significant impact on results.     

Application of SWAT with and without Variable Source Area Hydrology to a Large Watershed

In this study, two different versions of the Soil and Water Assessment Tool (SWAT) model were used to simulate the hydrology and biogeochemical response of the Cannonsville Reservoir watershed, in New York. The first version distributes overland flow in ways that are consistent with variable source area (VSA) hydrology driven by saturation excess runoff, whereas the second version is the standard version of SWAT. These two models were each calibrated for streamflow (Flow), particulate phosphorus (PP), total dissolved phosphorus (TDP), and sediment (Sed) against measured data from the 1,200 km² Cannonsville watershed. The standard version of the model yielded an r² between the measured and simulated data of 0.85, 0.73, 0.70, and 0.72 for Flow, Sed, TDP, and PP, respectively. The VSA version yielded an r² of 0.84, 0.69, 0.72, and 0.53 for Flow, Sed, TDP, and PP, respectively. The two models were then used to determine the maximum upper bound on the reduction in phosphorus loading by removing all of the corn in the watershed. The average reductions between the two models were 65 and 37% for PP and TDP, respectively. The VSA version was also used to estimate the effect of moving corn land in the watershed from the wettest, most runoff prone areas to the driest, least runoff prone areas, which cannot be done directly with the standard SWAT model.     

Development and evaluation of the channel routing model and parameters within the National Water Model

The National Water Model (NWM) was deployed by the National Oceanic and Atmospheric Administration to simulate operational forecasts of hydrologic states across the continental United States. This paper describes the geospatial river network (“hydro-fabric”), physics, and parameters of the NWM, elucidating the challenges of extrapolating parameters a large scale with limited observations. A set of regression-based channel geometry parameters are evaluated for a subset of the 2.7 million NWM reaches, and the riverine compound channel scheme is described. Based on the results from regional streamflow experiments within the broader NWM context, the compound channel reduced the root mean squared error by 2% and improved median Nash–Sutcliffe efficiency by 16% compared with a non-compound formulation. Peak event analysis from 910 peak flow events across 26 basins matched from the US Flash Flood Observation Database revealed that the mean timing error is 3 h lagged behind the observations. The routing time step was also tested, for 5-min (default, operational setting) and 1-h increments. The model was computationally stable and able to convey the flood peaks, although the hydrograph shape and peak timing were altered.     

Assessing Retrospective National Water Model Streamflow with Respect to Droughts and Low Flows in the Colorado River Basin

Streamflow monitoring in the Colorado River Basin (CRB) is essential to ensure diverse needs are met, especially during periods of drought or low flow. Existing stream gage networks, however, provide a limited record of past and current streamflow. Modeled streamflow products with more complete spatial and temporal coverage (including the National Water Model [NWM]), have primarily focused on flooding, rather than sustained drought or low flow conditions. Objectives of this study are to (1) evaluate historical performance of the NWM streamflow estimates (particularly with respect to droughts and seasonal low flows) and (2) identify characteristics relevant to model inputs and suitability for future applications. Comparisons of retrospective flows from the NWM to observed flows from the United States Geological Survey stream gage network over 22 years in the CRB reveal a tendency for underestimating low flow frequency, locations with low flows, and the number of years with low flows. We found model performance to be more accurate for the Upper CRB and at sites with higher precipitation, snow percent, baseflow index, and elevations. Underestimation of low flows and variable model performance has important implications for future applications: inaccurate evaluations of historical low flows and droughts, and less reliable performance outside of specific watershed/stream conditions. This highlights characteristics on which to focus future model development efforts.