Abstract We present a verification analysis of the operational short-range streamflow forecasts produced by the Office of Water Prediction with the National Water Model (NWM) from 2019 to 2024. Forecasts were compared to hourly observations at 7500 U.S. Geological Survey (USGS) locations over the conterminous United States (CONUS). The study has two goals: 1) estimate the probability of detection (POD) of high-flow events, focusing on small basins prone to occurrence of flash floods, and 2) quantify the impact that can be attributed to the modeling system components, the streamflow data assimilation, and the quantitative precipitation forecasts from the High-Resolution Rapid Refresh (HRRR) model using the Kling–Gupta efficiency (KGE) score. When the model uses streamflow data assimilation, the median POD reduces from 0.8 at the first hour of the forecast to zero after 7 h. When there is no streamflow data assimilation, the median POD reduces from 0.1 at the first hour of the forecast to zero after 5 h. When examining the individual components of the forecasting system, it was found that the errors in precipitation forecast reduce KGE at headwater basins. Our results suggest that the main factors limiting the skill to predict high-flow events at small basins are states of the model that are used to initialize the forecasts, especially when the model has no access to streamflow assimilation. Under these conditions, the forecasts have a median KGE value of −0.6, which is below the KGE acceptable reference value of −0.41. Significance Statement Flash floods develop quickly and can threaten lives and property, especially in small watersheds that respond rapidly to heavy rain. This study evaluates how well the National Water Model predicts short-term streamflow across the United States, with special attention to these vulnerable basins. We show that forecasts often struggle to anticipate high flows more than a few hours in advance, mainly because of the model’s initial conditions, that is, the model estimates of how much water is already in the landscape are not accurate enough without assimilation of streamflow observations. We also show that errors in rainfall forecasts further reduce skill. These findings point to where future improvements are most needed: better rainfall information, model structures, and combination of both.

ABSTRACT This work presents the development and validation of the River Network Streamflow Temperature Model (RNSTM), which solves the energy balance equations at the air‐water interface within the channels to estimate their temperature. RNSTM considers solar radiation, net longwave radiation, evaporative heat flux, and convective heat transfer. Additionally, it includes sub‐surface heat transfer and the rainfall effects on water temperature. First, we present the formulation and testing of a lumped‐energy balance model. For this test, we used atmospheric forcings from ground‐based observations and the High‐Resolution Rapid Refresh (HRRR) weather forecasting system. Next, we formulated RNSTM for a general river network using the ordinary differential equations (ODE) solver that is part of the Hillslope Link Model (HLM). We tested RNSTM using HRRR meteorological data and discharge simulations from HLM and we validated it using United States Geological Survey (USGS) water temperature observations at the Cedar River at Waverly, Iowa, for 2021. Our model results show potential for large‐scale deployment and water quality‐related applications.

The study evaluates radar-based quantitative precipitation estimations (QPEs) for 10 extreme rain events that occurred between 2013 and 2019 in the Kansas City metropolitan area, United States. These precipitation estimates were derived at hourly and approximately 0.5-km scales using two polarimetric QPE algorithms one based on specific attenuation (A) and the other on specific differential phase (KDP) for the study area covered by two overlapping radars in Topeka, Kansas, and Kansas City, Missouri. The polarimetric QPE assessment for extreme rain events was motivated by improved flood forecasting and precipitation frequency analysis. The analysis utilizes ground reference observations from a dense network of about 170 rain gauges over the study area to quantitatively assess the accuracy of these polarimetric rainfall (R) estimates. The comparison of R(A) and R(KDP) with the conventional algorithm based on radar reflectivity observations reveals that the two polarimetric algorithms outperform the reflectivity-based approach. While R(KDP) shows a systematic conditional feature (i.e., underestimation at high rain rates) with reduced scatter, R(A) appears to be less biased but with relatively large scatter. The significanct overestimation of R(A) for one of the extreme events was attributed to the misestimation of its key parameter (a), which resulted from hail contaminated data samples. To examine the observed underestimation tendency of R(KDP), we characterized the magnitude of underestimation (bias) with rainfall spatial variability as this variability may account for different rainfall regimes or the smoothing effect of KDP to reduce its inherent noisiness. Our result demonstrates that the underestimation tendency of R(KDP) becomes more pronounced as rainfall spatial variability increases.

Hydrological models and quantitative precipitation estimation (QPE) are critical elements of fiood forecasting systems. Both are subject to considerable uncertainties. Quantifying their relative contribution to the forecasted streamfiow and fiood uncertainty has remained challenging. Past work documented in the literature focused on one of these elements separately from the other. With this in mind, we present a systematic approach to assess the impact of QPE uncertainty in streamfiow forecasting. Our approach explores the operational Iowa Flood Center (IFC) hydrological model performance after altering two radar-based QPE products. We ran the Hillslope Link Model (HLM) for Iowa between 2015 and 2020, altering the Multi-Radar/Multi-Sensor (MRMS) system and the specific attenuation-based (IFCA) IFC radar-derived product with a multiplicative error term. We assessed the forecasting system performance at 112 USGS streamfiow gauges using the altered QPE products. Our results suggest that addressing rainfall uncertainty has the potential for much-improved fiood forecasting spatially and seasonally. We identified spatial patterns linking prediction improvements to the radar's location and the magnitude of rainfall. Also, we observed seasonal trends suggesting underestimations during the cold season (October-April). The patterns for different radar products are generally similar but also show some differences, implying that the QPE algorithm plays a role. This study's results are a step toward separating modeling and QPE uncertainties. Future work involving larger areas and different hydrological and error models is essential to improve our understanding of the impact of QPE uncertainty. SIGNIFICANCE STATEMENT: This study investigates the impact of radar rainfall on fiood forecasting uncertainty. Previous research focused on rainfall-runoff models, ignoring the errors in rainfall estimation. We used a systematic approach to adjust two radar-rainfall products, forcing a simple hydrological model. Results show the potential improvement in streamfiow prediction by correcting basinwide bias in rainfall. The optimal correction varies with basin size, location, season, and rainfall amount.

Quantitative precipitation forecasting benefits real-time streamflow forecasts by extending the lead time horizon. Uncertainties in QPF compromise these benefits. This study examined the performance of the short-term QPF product known as High-Resolution Rapid Refresh, used as the input to hydrologic models for streamflow forecasting. The models are the National Water Model operated by the National Water Center and the Hillslope Link Model used by the Iowa Flood Center to provide real-time forecasts for Iowa. The National Water Model (NWM) streamflow output is examined at 7162 gauging stations operated by the U.S. Geological Survey. Results of three analyses are discussed. The first analysis compares HRRR QPF to the corresponding quantitative precipitation estimation product known as Multi-Radar Multi-Sensor. Both the QPF and the QPE products represent hourly rainfall accumulations. The comparison is performed in the context of river basins with boundaries defined by the USGS gauging stations using several performance criteria. The second analysis represents a categorical evaluation of the ability of the QPF-driven NWM to detect floods, defined as discharge exceeding the mean annual peak value. The third analysis is limited to the USGS-gauged basins located in Iowa using the Hillslope Link Model (HLM). The HLM is driven by the QPF for the 18 separate lead times in an open-loop configuration mimicking traditional hydrologic model simulation. A control simulation uses the MRMS QPE as the driving input. All analyses are conducted as a function of lead time and spatial scale. Results demonstrate the marginal benefit of the HRRR for streamflow forecast especially for basins smaller than 1000 km2. SIGNIFICANCE STATEMENT: The authors analyzed several years of real-time streamflow forecasts generated by the National Water Model over the CONUS that used high-resolution quantitative precipitation forecasts. They concluded that QPF uncertainty is the dominant source of errors in streamflow forecasts. The forecasting skill for both streamflow and rainfall increases with the basin scale but is poor for basins smaller than about 1000 km2 which represent half of the gauged basins. The study, limited to 18 h of lead time, also shows that forecasting skill quickly decreases with time. The results offer a hydrologic perspective that explains why so many deadly flash floods lack site-specific forecasts. The authors argue for a balanced approach to allocating resources for rainfall estimation and forecasting and hydrologic modeling.

The National Weather Service (NWS) operates the National Water Model (NWM) to provide continental-scale streamflow forecasting across the United States. Despite the broad scope of NWM, it faces limitations in delivering operational-level predictions. To overcome these limitations, the NWS embarked on development of the Next Generation Water Resources Modeling Framework (NextGen). However, a key shortcoming of the NextGen and NWM is the lack of robust data assimilation (DA) step. This study provides a DA module that incorporates the Ensemble Kalman Filter (EnKF), and the Particle Filter (PF) for use within the NextGen framework. The effectiveness of the developed module is evaluated by assimilating the in-situ observations to the Conceptual Functional Equivalent model, a simplified version of the current NWM, demonstrating the first advanced DA application on this model. The results show that both DA methods effectively enhance the performance of the model prediction, while the PF outperforms the EnKF.

The study evaluated radar-derived polarimetric rainfall estimates for extreme rain events that occurred in the Kansas City Metropolitan area in the United States. To derive quantitative precipitation estimates (QPE), we implemented two polarimetric algorithms based on specific attenuation $(A)$ and specific differential phase ($K_{D P}$), along with the reflectivity ($\boldsymbol{Z}$) based one using data from two radars in the study area. The analysis to assess radar-rainfall estimates ($R$) utilizes ground observations from a dense network of about 170 rain gauges. Based on our analysis results, the two polarimetric estimates from $R(A)$ and $R\left(K_{D P}\right)$ outperform the conventional estimation $R(Z)$. $R(A)$ appeared to be less biased with relatively large scatter while $R\left(K_{D P}\right)$ underestimates at high rainfall rate with less scatter compared to $R(A)$. To generate robust rainfall estimates by accounting for the error structure of the individual algorithms, we decomposed the errors into systematic and random components, conditioned on the magnitude of radar estimates. These conditional features were then used to generate compositeweighted rainfall estimates. The composite estimates derived from two polarimetric algorithms, $R(A)$ and $R\left(K_{D P}\right)$, showed significant improvement, particularly for reduction in bias and variability. The spatial averaging of these composite estimates over an experimental domain demonstrates their potential for streamflow prediction.

The frequency of extreme flood events is increasing throughout the world. Daily, high-resolution (30m) Flood Inundation Maps (FIM) observed from space play a key role in informing mitigation and preparedness efforts to counter these extreme events. However, the temporal frequency of publicly available high-resolution FIMs, e.g., from Landsat, is at the order of two weeks thus limiting the effective monitoring of flood inundation dynamics. Conversely, global, low-resolution ( 300m) Water Fraction Maps (WFM) are publicly available from NOAA VIIRS daily. Motivated by the recent successes of deep learning methods for single image super-resolution, we explore the effectiveness and limitations of similar data-driven approaches to downscaling low-resolution WFMs to high-resolution FIMs. To overcome the scarcity of high-resolution FIMs, we train our models with high-quality synthetic data obtained through physics-based simulations. We evaluate our models on real-world data from flood events in the state of Iowa. The study indicates that data-driven approaches exhibit superior reconstruction accuracy over non-data-driven alternatives and that the use of synthetic data is a viable proxy for training purposes. Additionally, we show that our trained models can exhibit superior zero-shot performance when transferred to regions with hydroclimatological similarity to the U.S. Midwest.

An increase in extreme rainfall frequency across the midwestern United States has been accompanied by an increase in damaging floods. The US has over 90,000 dams, more than 75% of which are small and rarely used for flood mitigation. Recent research focused on operating these ponds for flood reduction using gated outlets, a technique known as activated distributed storage, has confirmed its potential for reducing flood impacts. Here, the authors build upon this work by developing a hydrologic model to simulate the active management of a distributed network of 130 ponds that employs up to 18 h of forecasted rainfall for operational decision making, a process known as forecast-informed reservoir operation (FIRO). Using five observed rainfall events and a single dam operations scheme, the effects of using FIRO for real-time gate operations on both downstream peak flows and basin wide storage utilization are evaluated. Simulation results that use the high-resolution rapid refresh (HRRR) product, were compared to those that (1) use no rainfall forecasts for decision making; and (2) use 18 h of observed rainfall mimicking an ideal forecast. Regardless of forecast accuracy or rainfall accumulation, shorter forecast lead times result in operational decisions that release water early in an event, vacating storage, while longer lead times result in increased storage throughout an event, thus reducing downstream flows. These results indicate that rainfall forecasts may not be solely capable of addressing the complexities governing a distributed storage network's ability to release water. This suggests that a more nuanced approach, utilizing optimal control of the storage network is required to unlock the technique's full potential.

The state of Iowa in the Central United States has experienced increasing flooding, with major events occurring most recently in 1993, 2008, 2011, and 2019. These floods caused over $23B in damage despite Iowa's three flood control reservoirs and expansive levee systems, suggesting the need for additional solutions. Iowa is home to over 4,000 small dams whose cumulative capacity more than double the state's current flood storage. These locations are operated passively, i.e., without the use of gated outlets to control basin storage utilization, thus limiting their flood mitigating potential. Here, the authors simulate gated outlets at 130 small dams within a 660 km2 watershed to (1) evaluate how effectively these storages can be activated across a watershed using gated outlets; and (2) quantify the utilization capacity of an activated distributed storage system for flow reduction. The authors used stochastic storm transposition to generate thousands of spatially variable rainfall events using Stage IV rainfall data within the Iowa domain at durations of 6, 12, 24, and 48 h and annual exceedance probabilities (AEPs) of 0.2, 0.1, 0.02, and 0.01. This expands the effective period of record, providing storms of various durations, intensities, and spatiotemporal distributions. An active management scheme was defined within the reservoir module of the hillslope link model designed to store water within the ponding locations. The study calculated the flow reductions that were achieved through this active scheme and found that flows were reduced for every rainfall duration and probability regardless of basin spatial scale. Reductions reached as high as 70% for a 6 h, 0.2 AEP event at a 93 km2 drainage area, while flows were reduced by roughly 12% for a 48 h, 0.01 AEP event at the basin outlet. This work establishes activated distributed storage as a meaningful flood reduction measure under realistic rainfall conditions at a variety of spatial scales.

The Upper Mississippi Information System (UMIS) is a cyberinfrastructure framework designed to support large-scale real-time water quality data integration, analysis, and visualization for the Upper Mississippi River Basin (UMRB). UMIS is intended to directly address three of the Grand Challenges for Engineering including: 1) understanding access to clean drinking water, 2) management of the nitrogen cycle, and 3) engineering the tools of scientific discovery. The UMIS is designed to provide significant immediate and long-term impacts including a central platform for data access, integration, discovery, and adoption of cyberinfrastructure tools and services. The UMIS demonstrates that public data aggregators and central repositories can provide important services to anyone interested in water quality research or education. In addition, working across multiple scales (e.g., state, region, county, or watershed) allows researchers to understand broad and narrow effects of water quality strategies. Exploration of data across these scales encourages the development of problem-based research questions that can eventually provide feedback to public policies.

Significant errors often arise when measuring streamflow during high flows and flood events. Such errors conflated by short records of observations may induce bias in the flood frequency estimates, leading to costly engineering design mistakes. This work illustrates how observational (measurement) errors affect the uncertainty of flood frequency estimation. The study used the Bulletin 17 C (US standard) method to estimate flood frequencies of historical peak flows modified to represent the measurement limitations. To perform the modifications, we explored, via Monte Carlo simulation, four hypothetical scenarios that mimic measurement errors, sample size limitations, and their combination. We used a multiplicative noise from a log-normal distribution to simulate the measurement errors and implemented a bootstrap approach to represent the sampling error. Then, we randomly selected M samples from the total N records of the observed peak flows of four gauging stations in Iowa in central USA. The observed data record ranges between 76 and 119 years for watersheds with drainage areas between 500 and 16,000 km 2 . According to the results, measurement errors lead to more significant differences than sampling limitations. The scenarios exhibited differences with median magnitudes of up to 50%, with some cases reaching differences up to 100% for return periods above 50 years. The results raise a red flag regarding flood frequency estimation that warrants looking for further research on observational errors.

New flood records are being set across the world as precipitation patterns change due to a warming climate. Despite the presence of longstanding water management infrastructure like levees and reservoirs, this rise in flooding has been met with property damage, loss of life, and hundreds of billions in economic impact, suggesting the need for new solutions. In this work, the authors suggest the active management of distributed networks of ponds, wetlands and retention basins that already exist across watersheds for the mitigation of flood damages. As an example of this approach, we investigate optimal control of the gated outlets of 130 such locations within a small watershed using linear programming, genetic algorithms, and particle swarm optimization, with the objective of reducing downstream flow and maximizing basin storage. When compared with passive operation (i.e., no gated outlets) and a uniformly applied active management scheme designed to store water during heavy rainfall, the optimal control techniques (1) reduce the magnitudes of peak flow events by up to 10%, (2) reduce the duration of flood crests for up to several days, and (3) preserve additional storage across the watershed for future rainfall events when compared with active management. Combined, these findings provide both a better understanding of dynamically controlled distributed storage as a flood fighting technique and a springboard for future work aimed at its use for reducing flood impacts.

Abstract We explore the projected changes in flood impacts across Iowa (central United States) and the associated uncertainties by forcing a hydrologic model with downscaled global climate model outputs and four Shared Socioeconomic Pathways. Our results point to projected increasing magnitude and variability in flooding across the state, especially for high‐emission scenarios. Next, we partition the flood impacts' projections into: (a) the response of the global climate models to anthropogenic forcing, (b) scenario uncertainty due to emissions, and (c) internal climate variability. We find scenario uncertainty plays a small role, while climate model uncertainty and internal climate variability dominate the flood impacts' projections, with the contribution of model uncertainty increasing toward the end of this century. Insights from our work can be utilized by stakeholders to understand the current limitations of flood impact projections and provide suggestions about where modelers should focus efforts to reduce uncertainty.

Data assimilation (DA) techniques such as the Ensemble Kalman filter (EnKF) and its extensions allow for real-time corrections of state-space models and model parameters based on an assumption of Gaussian error. The hydrological DA literature primarily documents applications of the EnKF to solve sequential state estimation problems. Recent advances in the DA literature demonstrate the potential of applying EnKF-based methods as efficient, derivative-free algorithms to solve various general Bayesian inverse problems, such as parameter estimation, while simultaneously providing Uncertainty Quantification (UQ). In this paper, the authors employ the Ensemble Kalman Inversion (EKI) algorithm to infer the distribution of a set of routing parameters. Through this correction, we improve streamflow at locations upstream of the gauged site in a virtual catchment setting. The algorithm enables learning spatially distributed routing parameters with observations available only at the outlet. The study reveals that this method sufficiently improves model performance throughout the basin. The performance of this method is demonstrated in a virtual catchment for three different model/data configurations. Favorable results, even with model misspecification, indicate that this method holds promise for operational application and more general hydrologic parameter estimation problems.

Understanding the projected changes in annual maximum peak discharge is important to improve resiliency in water resource planning and design at the community level. Currently, much of the literature on climate change impacts focuses on analyses at the regional scale. In this study we use a hydrologic model to evaluate the projected changes in annual maximum peak discharge at the community-level across Iowa under two emission scenarios, Representative Concentration Pathway 4.5 and 8.5 (RCP4.5 and RCP8.5). We utilize climate forcings from global climate models part of the Coupled Model Intercomparison Projected Phase 5 (CMIP5) from 1950 to 2100. Our simulations show a detectable increase in annual maximum discharge for 27% of Iowa's communities under RCP8.5. However, under RCP4.5 none of the communities are projected to face an increase in annual maximum discharge by the end of the 21st century; furthermore, precipitation intensity under this scenario is not projected to increase in the latter half of the 21st century. Our results point to a larger increase in precipitation intensity and annual maximum discharge under RCP8.5 compared to RCP4.5, especially in the second half of this century. The projected flood peak distributions tend to become statistically different from the historical ones later in the 21st century under RCP8.5 than under RCP4.5. This study provides a basis for analyzing climate change impacts at a local decision-making scale.

The High‐Resolution Model Intercomparison Project (HighResMIP) experiments from the Coupled Model Intercomparison Phase 6 represent a broad effort to improve the resolution, and performance of climate models. The HighResMIP suite provides high spatial resolution (i.e., 25‐ and 50‐km) forcings that have been shown to improve the representation of climate processes. However, little is known about their suitability for hydrologic applications. We use outputs from the HighResMIP suite to simulate annual maximum discharge with the Hillslope‐Link Model (HLM) at ∼1,000 river communities across Iowa. First, we assess whether the runoff from the climate models can be directly routed through the river network model in HLM to estimate annual maximum discharge. Runoff‐based simulations can capture the empirical distribution of flood peaks in five of the 10 models/members assessed. Next, we force the HLM with precipitation, temperature, and potential evapotranspiration from HighResMIP models to simulate flood peaks, finding all models/members produce empirical distributions similar to our reference. However, significant biases exist in the model/member forcings as correct flood response is being generated for the wrong reason. To improve their suitability for community‐level assessment, we use nine statistical approaches to bias‐correct and downscale HighResMIP precipitation to a 4‐km resolution. The bias‐correction and downscaling of climate model precipitation performs well for all models/members. Furthermore, we do not find significant changes in the magnitude flood peak projections for Iowa based on the HLM forced with HighResMIP outputs, or based on routed runoff, while there are indications that the variability in flood peaks is projected to increase across the state.

Flood frequency estimation forms the basis for engineering design of hydraulic structures, including bridges and culverts, local and regional development planning, and flood insurance. In the United States, the Water Resources Council recommends using the Log-Pearson Type III (LP3) distribution as a standard for use with the annual peak flow data. However, researchers have argued for the use of more than one streamflow value in a year thus increasing the sample size and decreasing the sampling error in the estimates of the flood quantiles. In this study, conducted over Iowa, the authors revisit the method proposed by Donald Turcotte and others to use power-law distribution applied to streamflow peak values for events separated by a time window. In contrast to those earlier studies, the authors applied formal statistical approach based on the maximum likelihood method and Kolmogorov-Smirnov statistic for parameter estimation. They also propose a novel simulation framework for the estimation of the sampling uncertainty of the power-law distribution. They apply the methodology to streamflow data from 62 USGS stream gauges in Iowa. The key finding of the study is that low-probability quantile estimates using Turcotte's method result in conservative estimates when compared with LP3 distribution confirming the earlier outcomes.

IMERG provides state-of-the-art satellite-based precipitation estimates that combine observations from multiple satellite platforms. This study evaluates IMERG products by examining hydrologic simulations of streamflow at a range of spatial scales. The main objective of this study is to assess the predictive utility of the near-real-time product (IMERG-Early). The assessment also includes the IMERG-Final product that is not available in real time. The authors used MRMS precipitation estimates and USGS streamflow observation data as references for the precipitation and stream -flow evaluations during a 5-yr period (2016-20). The precipitation evaluation results show that IMERG-Early yields signifi-cant overestimations, particularly during warm months, with higher variability in its conditional distributions, whereas the performance of IMERG-Final seems unbiased. The authors performed hydrologic simulations using the Iowa Flood Cen-ter's Hillslope Link Model with three precipitation forcing products, i.e., MRMS, IMERG-Early, and IMERG-Final. The simulation results reveal that IMERG-Early leads to high hit and false alarm rates due to its overestimation in precipitation and has almost no skill, as measured by the overall performance metric Kling-Gupta efficiency (KGE), in streamflow pre-diction regarding basin scales ranging from 10 to 30000 km2. This indicates that the product requires a bias correction be-fore it is useful for real-time flood prediction. The streamflow prediction performance of IMERG-Final seems comparable to that of MRMS at spatial scales greater than 100 km2. This scale limitation is attributable to the IMERG's product spatial resolution that is inadequate to capture the small-scale variability of precipitation.

Extreme rainfall in midwestern United States has gotten more common over the last half century, thus increasing flooding events across the region. As a result, traditional flood mitigation measures are commonly overwhelmed by highwater events, illustrating the need for new solutions. Of the 91,000 dams in the US, the vast majority are small and go unused for flood mitigation. Among those that are utilized in flood peak reduction, few are actively managed in which outflows are manipulated through a gated outlet. Instead, small storage locations typically use passive control, allowing impounded water levels to fluctuate without the use of a gated outlet, possibly squandering some of their flow-reduction potential. In this paper, we have evaluated actively managed storage within a distributed network of 130 small dams in a 660-km2 watershed in southeastern Iowa using three operation schemes to increase storage utilization and reduce downstream flows. We developed a module to simulate the dam operation into a distributed hydrologic model that is forced with soil conservation service (SCS) 24-h design storms distributed uniformly across the watershed with 0.2, 0.1, 0.02, and 0.01 exceedance probabilities to evaluate flow reductions. When compared with passive operation, outlet flows were reduced under each proposed iteration of the 24-h design storm. Using the most aggressive operation scheme, outlet flows were reduced by over 70%. These results showcase the need for better understanding of activated flood storage across midwestern watersheds and encourage further work in optimizing this technique for real-time management.