Sacramento Soil Moisture Accounting Model Research Articles

Performance Evaluation of Rainfall-Runoff Models for Predictions of Inflows to Bhumibol Reservoir in Thailand

The Bhumibol reservoir in the Ping River basin is the largest reservoir in the Kingdom of Thailand. This reservoir has contributed to economic development of the country by supplying increased electricity and irrigation water demands as well as flood mitigation in riparian areas along the Ping and the Chao Phraya River. The prediction of inflows to the reservoir is crucial for the optimal management of water for irrigation, power generation and flood control. Properly customized rainfall-runoff models of the catchment could provide the basis for predicting the inflows to the reservoir. Hence, five lumped conceptual rainfall-runoff models were developed for the Ping River basin to simulate daily inflows to the Bhumibol reservoir. The rainfall-runoff models are Australian Water Balance Model (AWBM), Sacramento Soil Moisture Accounting Model, Simplified Hydrolog Model (SIMHYD), Soil Moisture Accounting and Routing Model (SMAR) and Tank Model. The evaluation of the performances of these models showed that all models are capable of predicting inflows. However, the SIMHYD, Sacramento, AWBM and Tank models perform better than SMAR model. Hence, these models could be employed for prediction of inflows to the reservoir with acceptable accuracy.

Strategy for Deriving Sacramento Model Parameters Using Soil Properties to Improve Its Runoff Simulation Performances

Physically-based parameter estimations are essential to improve the simulation performance of a hydrologic model and to produce physically reasonable parameters with spatial consistency. This study proposed a parameter derivation strategy to improve the Sacramento Soil Moisture Accounting (SAC-SMA) model simulation performance based on the publicly accessible Harmonized World Soil Database (HWSD). The HWSD soil properties were used to estimate the soil moisture characteristics, and the HWSD soil texture classifications and International Geosphere-Biosphere Programme (IGBP) land cover types were used to identify the Soil Conservation Service (SCS) runoff curve number (CN). After the soil moisture characteristics and CNs were identified, the major parameters of the SAC-SMA model were derived. The simulation results were evaluated using the Nash efficiency coefficient (NSEC), and Free Search (FS) algorithm was used to further adjust and calibrate the parameters. Compared with the simulation accuracy (NSEC = 0.66~0.88) and parameter transferability (NSEC = 0.22~0.83) obtained for the SAC-SMA model using directly calibrated parameters, the HWSD data-derived parameters allowed the SAC-SMA model to achieve a similar simulation accuracy (NSEC = 0.65~0.86) and a better transferability (NSEC = 0.61~0.85).

Evaluation of Temperature-Index and Energy-Balance Snow Models for Hydrological Applications in Operational Water Supply Forecasts

In the western United States, snow accumulation, storage, and ablation affect seasonal runoff. Thus, the prediction of snowmelt is essential to improve the reliability of water supply forecasts to guide water allocation and operational decisions. The current method used at the Colorado Basin River Forecast Center (CBRFC) couples the SNOW-17 temperature-index snow model and the Sacramento Soil Moisture Accounting (SAC-SMA) runoff model in a lumped approach. Limitations in parameter transferability and calibration requirements for changing conditions with the temperature-index model motivated this research, in which new avenues were investigated to assess and prototype the application of an energy-balance snow model in a distributed modeling approach. The Utah Energy Balance (UEB) model was chosen to compare with the SNOW-17 model because it is simple and parsimonious, making it suitable for distributed application with the potential to improve water supply forecasts. Each model was coupled with the SAC-SMA model and the Rutpix7 routing model to simulate basin snowmelt and discharge. All the models were applied on grids over watersheds using the Research Distributed Hydrologic Model (RDHM) framework. Case studies were implemented for two study sites in the Colorado River basin over a period of two decades. The model performance was evaluated by comparing the model output with observed daily discharge and snow-covered area data obtained from remote sensing sources. Simulated evaporative components of sublimation and evapotranspiration were also evaluated. The results showed that the UEB model, requiring calibration of only a snow drift factor, achieves a comparable performance to the calibrated SNOW-17 model, and both provided reasonable basin snow and discharge simulations in the two study sites. The UEB model had the additional advantage of being able to explicitly simulate sublimation for different land types and thus better quantify evaporative water balance components and their sensitivity to land cover change. UEB also has a better transferability potential because it requires calibration of fewer parameters than SNOW-17. The majority of the parameters for UEB are physically based and regarded as constants characterizing spatially invariant properties of snow processes. Thus, the model remains valid for different climate and terrain conditions for multiple watersheds.

A fusion-based data assimilation framework for runoff prediction considering multiple sources of precipitation

ABSTRACT A fusion-based framework, in which a particle filter Markov chain Monte Carlo (PFMCMC) data assimilation method was coupled with the hydrological Sacramento Soil Moisture Accounting Model (SAC-SMA), was developed to improve the model’s capacity to predict one-day-ahead runoff. A case study was applied where mean daily precipitation from multiple sources served as forcing data in the data assimilation procedure, while ground station and multiple bias-corrected satellite-based precipitation datasets served as precipitation input datasets. The model training period used six years (2002–2007) of data to determine optimal weights through a genetic algorithm optimization model, while two years (2008–2009) were used to test the model. The proposed framework, applied to a real case study, improved SAC-SMA runoff prediction accuracy by incorporating precipitation datasets from multiple sources in the data assimilation procedure. On average, the PFMCMC-based data assimilation procedure led to a 13.7% improvement in SAC-SMA model performance metrics (NSE, MAB, RMSE, RMSRE, RMRE).

Evaluation of distributed process-based hydrologic model performance using only a priori information to define model inputs

Fully distributed, integrated surface–subsurface hydrological models (ISSHMs) have seen renewed interest due to availability of better software, high performance computing facilities, and high-resolution, spatially extensive data products. ISSHMs are valuable as tools for advancing system understanding as they can resolve multiple processes defined on the plot scale including three-dimensional interaction of surface water and groundwater. Here, we evaluated the performance of an ISSHM, the Advanced Terrestrial Simulator (ATS), on seven diverse catchments across the continental US using widely available data products to define model inputs without calibration. We compare the ATS-simulated streamflow and evapotranspiration with gauge observations and MODIS-derived evapotranspiration, respectively. Using the Kling-Gupta Efficiency (KGE) as metric, ATS with default data products performed reasonably well at 6 of 7 catchments for streamflow. However, in one of those 6 catchments ATS had poor performance on baseflow and ATS’s overall performance was thus judged to be inadequate despite the acceptable KGE. ATS performance for evapotranspiration was good in all 7 catchments using default data products. In the two catchments where ATS streamflow performance using default data products was not acceptable, the performance was significantly improved by using local information on subsurface properties below the soil. We also compare the model-simulated streamflow and evapotranspiration with the Sacramento soil moisture accounting (SAC-SMA) model, a semi-distributed model that was calibrated on a catchment-by-catchment basis. Uncalibrated ATS performance is comparable to the calibrated SAC-SMA model in terms of streamflow while ATS performance is similar to or better (much better in certain catchments) in reproducing MODIS-derived evapotranspiration. Reasonably good performance of ATS without catchment-specific calibration provides new confidence in the ISSHM class of models and community data products as tools for advancing understanding of watershed function in a changing environment.

Applying the NWS’s Distributed Hydrologic Model to Short-Range Forecasting of Quickflow in the Mahantango Creek Watershed

Abstract Accurate and reliable forecasts of quickflow, including interflow and overland flow, are essential for predicting rainfall–runoff events that can wash off recently applied agricultural nutrients. In this study, we examined whether a gridded version of the Sacramento Soil Moisture Accounting model with Heat Transfer (SAC-HT) could simulate and forecast quickflow in two agricultural watersheds in east-central Pennsylvania. Specifically, we used the Hydrology Laboratory–Research Distributed Hydrologic Model (HL-RDHM) software, which incorporates SAC-HT, to conduct a 15-yr (2003–17) simulation of quickflow in the 420-km2 Mahantango Creek watershed and in WE-38, a 7.3-km2 headwater interior basin. We directly calibrated HL-RDHM using hydrologic observations at the Mahantango Creek outlet, while all grid cells within Mahantango Creek, including WE-38, were calibrated indirectly using scalar multipliers derived from the basin outlet calibration. Using the calibrated model, we then assessed the quality of short-range (24–72 h) deterministic forecasts of daily quickflow in both watersheds over a 2-yr period (July 2017–October 2019). At the basin outlet, HL-RDHM quickflow simulations showed low biases (PBIAS = 10.5%) and strong agreement (KGE″ = 0.81) with observations. At the headwater scale, HL-RDHM overestimated quickflow (PBIAS = 69.0%) to a greater degree, but quickflow simulations remained satisfactory (KGE″ = 0.65). When applied to quickflow forecasting, HL-RDHM produced skillful forecasts (>90% of Peirce and Gerrity skill scores above 0.5) at all lead times and significantly outperformed persistence forecasts, although skill gains in Mahantango Creek were slightly lower. Accordingly, short-range quickflow forecasts by HL-RDHM show promise for informing operational decision-making in agriculture. Significance Statement Daily runoff forecasts can alert farmers to rainfall–runoff events that have the potential to wash off recently applied fertilizers and manures. To gauge whether daily runoff forecasts are accurate and reliable, we used runoff monitoring data from a large agricultural watershed and one of its headwater tributaries to evaluate the quality of short-term runoff forecasts (1–3 days ahead) that were generated by a National Weather Service watershed model. Results showed that the accuracy and reliability of daily runoff forecasts generally improved in both watersheds as lead times increased from 1 to 3 days. Study findings highlight the potential for National Weather Service models to provide useful short-term runoff forecasts that can inform operational decision-making in agriculture.

Variational Bayesian dropout with a Gaussian prior for recurrent neural networks application in rainfall–runoff modeling

Recurrent neural networks (RNNs) are a class of artificial neural networks capable of learning complicated nonlinear relationships and functions from a set of data. Catchment scale daily rainfall–runoff relationship is a nonlinear and sequential process that can potentially benefit from these intelligent algorithms. However, RNNs are perceived as being difficult to parameterize, thus translating into significant epistemic (lack of knowledge about a physical system) and aleatory (inherent randomness in a physical system) uncertainties in modeling. The current study investigates a variational Bayesian dropout (or Monte Carlo dropout (MC-dropout)) as a diagnostic approach to the RNNs evaluation that is able to learn a mapping function and account for data and model uncertainty. MC-dropout uncertainty technique is coupled with three different RNN networks, i.e. vanilla RNN, long short-term memory (LSTM), and gated recurrent unit (GRU) to approximate Bayesian inference in a deep Gaussian noise process and quantify both epistemic and aleatory uncertainties in daily rainfall–runoff simulation across a mixed urban and rural coastal catchment in North Carolina, USA. The variational Bayesian outcomes were then compared with the observed data as well as with a well-known Sacramento soil moisture accounting (SAC-SMA) model simulation results. Analysis suggested a considerable improvement in predictive log-likelihood using the MC-dropout technique with an inherent input data Gaussian noise term applied to the RNN layers to implicitly mitigate overfitting and simulate daily streamflow records. Our experiments on the three different RNN models across a broad range of simulation strategies demonstrated the superiority of LSTM and GRU approaches relative to the SAC-SMA conceptual hydrologic model.

Rainfall-runoff modeling using long short-term memory based step-sequence framework

Rainfall-runoff modeling, a nonlinear time series process, is challenging and important in hydrological sciences. Among the data-driven approaches, those ones based on the long short-term memory (LSTM) network show their promising performance. In this paper, for rainfall-runoff modeling, we propose a novel data-driven framework named long short-term memory based step-sequence (LSTM-SS) framework, which contains m specific models for m-step-ahead runoff predictions. This model uses the sequential information of runoff series and follows the causality in practice: the current runoff is not affected by the later meteorological data. To show its performance and advantages, we test it on 241 basins of the Catchment Attributes and Meteorology for Large-Sample Studies (CAMELS) data set and predict the 7-day-ahead runoff. The results show that our rainfall-runoff models outperform the benchmark (physically-based or data-driven) models significantly employing for the CAMELS data set, including the Sacramento Soil Moisture Accounting Model (SAC-SMA) coupled with the Snow-17 snow routine, a two-layer LSTM network, and a LSTM-based sequence-to-sequence network. For 1-day-ahead runoff predictions, the median of Nash–Sutcliffe model efficiency for the 241 basins provided by our model is 0.85, while that provided by the two-layer LSTM network is 0.65. Furthermore, the results also show that our proposed LSTM-SS framework not only can significantly improve the performance of a single daily runoff prediction, but also has good performance for multiple-step-ahead runoff predictions.

Increasing the Efficiency of the Sacramento Model on Event Basis in a Mountainous River Basin

An evaluation of the Sacramento Soil Moisture Accounting (SAC-SMA) model was conducted to be used in flood event simulations with datasets at a time step up to one hour. The SAC-SMA model is a complex conceptual model which integrates two soil zones, the upper and lower zone, in order to provide current soil moisture conditions and generated streamflow. However, in flood events, where time intervals are small, the generated flood hydrograph is usually the product of only the upper soil layer runoff generation mechanism while the lower zone and baseflow have little impact. In this context, a modified version of the original SAC-SMA model was introduced, where only the upper zone processes are kept in order to reduce the parameter count and the overall model uncertainty involved, and a comparison was made against the original model output. The two models were calibrated and validated for a series of flood events occurred at the Karitaina basin of the Alfeios river, located in southern Greece. The results show that both model versions were able to reproduce the observed runoff with success. The simplified model showed high consistency with the original model in all cases, which is an obvious improvement to the original model, since it provided results of equal quality, while lowering significantly the total parameter count and the computing time. This contributes against the overall model generated uncertainty which is crucial for real-time data processing applications and flood forecasting systems. • Presentation of the SAC-SMA model concept, variables, parameters and flowchart. • Introduction of a modified – simplified version of the original SAC-SMA model to be used for event-based rainfall runoff applications. • Calibration and validation of the SAC-SMA using fine temporal scale datasets in a mountainous basin in Greece.

Flood spatial coherence, triggers, and performance in hydrological simulations: large-sample evaluation of four streamflow-calibrated models

Abstract. Floods cause extensive damage, especially if they affect large regions. Assessments of current, local, and regional flood hazards and their future changes often involve the use of hydrologic models. A reliable hydrologic model ideally reproduces both local flood characteristics and spatial aspects of flooding under current and future climate conditions. However, uncertainties in simulated floods can be considerable and yield unreliable hazard and climate change impact assessments. This study evaluates the extent to which models calibrated according to standard model calibration metrics such as the widely used Kling–Gupta efficiency are able to capture flood spatial coherence and triggering mechanisms. To highlight challenges related to flood simulations, we investigate how flood timing, magnitude, and spatial variability are represented by an ensemble of hydrological models when calibrated on streamflow using the Kling–Gupta efficiency metric, an increasingly common metric of hydrologic model performance also in flood-related studies. Specifically, we compare how four well-known models (the Sacramento Soil Moisture Accounting model, SAC; the Hydrologiska Byråns Vattenbalansavdelning model, HBV; the variable infiltration capacity model, VIC; and the mesoscale hydrologic model, mHM) represent (1) flood characteristics and their spatial patterns and (2) how they translate changes in meteorologic variables that trigger floods into changes in flood magnitudes. Our results show that both the modeling of local and spatial flood characteristics are challenging as models underestimate flood magnitude, and flood timing is not necessarily well captured. They further show that changes in precipitation and temperature are not always well translated to changes in flood flow, which makes local and regional flood hazard assessments even more difficult for future conditions. From a large sample of catchments and with multiple models, we conclude that calibration on the integrated Kling–Gupta metric alone is likely to yield models that have limited reliability in flood hazard assessments, undermining their utility for regional and future change assessments. We underscore that such assessments can be improved by developing flood-focused, multi-objective, and spatial calibration metrics, by improving flood generating process representation through model structure comparisons and by considering uncertainty in precipitation input.

Comparing Flood Projection Approaches Across Hydro‐Climatologically Diverse United States River Basins

AbstractThe challenge of estimating design flood magnitude under climate change has led to the development of multiple approaches to long‐term flood projection: stationarity, informed‐parameter (composed of both trend informed and climate informed), and hydrologic simulation. This is the first study to compare these approaches across a large set of hydro‐climatologically diverse basins located throughout the contiguous United States, grouped into distinct clusters based on catchment and flood characteristics. The comparison is achieved using a split‐sample test conducted over the historic period; the climate (specifically, precipitation) informed and hydrologic simulation (specifically, Sacramento Soil Moisture Accounting model) approaches are forced with observations and downscaled general circulation model (GCM) simulations. The results provide a quantitative perspective on key long‐term flood projection issues. The precipitation informed approach can be informative for projections of regional change; the hydrologic simulation approach can be informative for direction of change for a single basin. When forced with GCM outputs, precipitation informed is both more accurate and more precise than hydrologic simulation, after accounting for model error. For hydrologic simulation, forcing with GCMs will add positive bias for basins in the Rocky Mountains and Southwest and will add negative bias with large uncertainty across GCMs for basins in the lower Midwest, the South, and the mid‐Atlantic seaboard. The results illustrate the need for continued improvement in long‐term flood projection approaches and for design paradigms incorporating uncertainty.

Mean Field Bias-Aware State Updating via Variational Assimilation of Streamflow into Distributed Hydrologic Models

When there exist catchment-wide biases in the distributed hydrologic model states, state updating based on streamflow assimilation at the catchment outlet tends to over- and under-adjust model states close to and away from the outlet, respectively. This is due to the greater sensitivity of the simulated outlet flow to the model states that are located more closely to the outlet in the hydraulic sense, and the subsequent overcompensation of the states in the more influential grid boxes to make up for the larger scale bias. In this work, we describe Mean Field Bias (MFB)-aware variational (VAR) assimilation, or MVAR, to address the above. MVAR performs bi-scale state updating of the distributed hydrologic model using streamflow observations in which MFB in the model states are first corrected at the catchment scale before the resulting states are adjusted at the grid box scale. We comparatively evaluate MVAR with conventional VAR based on streamflow assimilation into the distributed Sacramento Soil Moisture Accounting model for a headwater catchment. Compared to VAR, MVAR adjusts model states at remote cells by larger margins and reduces the Mean Squared Error of streamflow analysis by 2–8% at the outlet Tiff City, and by 1–10% at the interior location Lanagan.

Identification of Combined Hydrological Models and Numerical Weather Predictions for Enhanced Flood Forecasting in a Semiurban Watershed

Flood forecasting in urban and semiurban catchments is often limited by the capability of the combined hydrological models and forecast inputs to predict floods accurately. The objective of this research is to develop an approach (1) to identify the best model forecast from multiple integrations of various hydrological models and numerical weather predictions (NWP), and (2) to find the best forecast combination method for an improved short-range flood forecasting. Seven selected hydrological models were coupled, each with two high-resolution NWP forecasts to provide several alternatives of deterministic hydrological forecasts at a catchment outlet. As such, the different model-input combinations were used to generate 14 hydrological forecasts. Hydrological forecast verification was then carried out over a one-year hindcast period. A comparison between six forecast combination methods, including a benchmark Bayesian model averaging (BMA) method, was also performed for the multiple available short-term streamflow forecasts. Results indicate that the coupling of the Sacramento soil moisture accounting (SACSMA) model with both High-Resolution Deterministic Precipitation System and High-Resolution Rapid Refresh inputs outperformed other model-input integrations. Maximum forecast errors in all model-input integration outputs occurred at forecast lead times of 12–14 h, corresponding to the time of concentration of the catchment. Providing constraints on the estimation of model weights was found to be a significant factor for obtaining an improved combined streamflow forecast. In general, the regression-based forecast combination method of the constrained ordinary least squares (CLS) has emerged as a possible alternative to the widely used BMA method for hydrology application.

Evaluation of Snowmelt Estimation Techniques for Enhanced Spring Peak Flow Prediction

Water resources management and planning requires accurate and reliable spring flood forecasts. In cold and snowy countries, particularly in snow-dominated watersheds, enhanced flood prediction requires adequate snowmelt estimation techniques. Whereas the majority of the studies on snow modeling have focused on comparing the performance of empirical techniques and physically based methods, very few studies have investigated empirical models and conceptual models for improving spring peak flow prediction. The objective of this study is to investigate the potential of empirical degree-day method (DDM) to effectively and accurately predict peak flows compared to sophisticated and conceptual SNOW-17 model at two watersheds in Canada: the La-Grande River Basin (LGRB) and the Upper Assiniboine river at Shellmouth Reservoir (UASR). Additional insightful contributions include the evaluation of a seasonal model calibration approach, an annual model calibration method, and two hydrological models: McMaster University Hydrologiska Byrans Vattenbalansavdelning (MAC-HBV) and Sacramento Soil Moisture Accounting model (SAC-SMA). A total of eight model scenarios were considered for each watershed. Results indicate that DDM was very competitive with SNOW-17 at both the study sites, whereas it showed significant improvement in prediction accuracy at UASR. Moreover, the seasonally calibrated model appears to be an effective alternative to an annual model calibration approach, while the SAC-SMA model outperformed the MAC-HBV model, no matter which snowmelt computation method, calibration approach, or study basin is used. Conclusively, the DDM and seasonal model calibration approach coupled with the SAC-SMA hydrologic model appears to be a robust model combination for spring peak flow estimation.

Ensemble Streamflow Forecasting Using an Energy Balance Snowmelt Model Coupled to a Distributed Hydrologic Model with Assimilation of Snow and Streamflow Observations

AbstractIn many river basins across the world, snowmelt is an important source of streamflow. However, detailed snowmelt modeling is hampered by limited input data and uncertainty arising from inadequate model structure and parametrization. Data assimilation that updates model states based on observations, reduces uncertainty and improves streamflow forecasts. In this study, we evaluated the Utah Energy Balance (UEB) snowmelt model coupled to the Sacramento Soil Moisture Accounting (SAC‐SMA) and rutpix7 stream routing models, integrated within the Research Distributed Hydrologic Model (RDHM) framework for streamflow forecasting. We implemented an ensemble Kalman filter for assimilation of snow water equivalent (SWE) observations in UEB and a particle filter for assimilation of streamflow to update the SAC‐SMA and rutpix7 states. Using leave one out validation, it was shown that the modeled SWE at a location where observations were excluded from data assimilation was improved through assimilation of data from other stations, suggesting that assimilation of sparse observations of SWE has the potential to improve the distributed modeling of SWE over watershed grid cells. In addition, the spatially distributed snow data assimilation improved streamflow forecasts and the forecast volume error was reduced. On the other hand, the assimilation of streamflow observations did not provide additional forecast improvement over that achieved by the SWE assimilation for seasonal forecast volume likely due to there being little information content in streamflow at the forecast date prior to its rising during the melt period and this application of particle filter being better suited for shorter timescales.

Toward Improved Predictions in Ungauged Basins: Exploiting the Power of Machine Learning

AbstractLong short‐term memory (LSTM) networks offer unprecedented accuracy for prediction in ungauged basins. We trained and tested several LSTMs on 531 basins from the CAMELS data set using k‐fold validation, so that predictions were made in basins that supplied no training data. The training and test data set included ∼30 years of daily rainfall‐runoff data from catchments in the United States ranging in size from 4 to 2,000 km2 with aridity index from 0.22 to 5.20, and including 12 of the 13 IGPB vegetated land cover classifications. This effectively “ungauged” model was benchmarked over a 15‐year validation period against the Sacramento Soil Moisture Accounting (SAC‐SMA) model and also against the NOAA National Water Model reanalysis. SAC‐SMA was calibrated separately for each basin using 15 years of daily data. The out‐of‐sample LSTM had higher median Nash‐Sutcliffe Efficiencies across the 531 basins (0.69) than either the calibrated SAC‐SMA (0.64) or the National Water Model (0.58). This indicates that there is (typically) sufficient information in available catchment attributes data about similarities and differences between catchment‐level rainfall‐runoff behaviors to provide out‐of‐sample simulations that are generally more accurate than current models under ideal (i.e., calibrated) conditions. We found evidence that adding physical constraints to the LSTM models might improve simulations, which we suggest motivates future research related to physics‐guided machine learning.

Optimization of the Multi-Start Strategy of a Direct-Search Algorithm for the Calibration of Rainfall–Runoff Models for Water-Resource Assessment

Calibration of conceptual rainfall–runoff models (CRRM) for water-resource assessment (WRA) is a complicated task that contributes to the reliability of results obtained from catchments. In recent decades, the application of automatic calibration techniques has been frequently used because of the increasing complexity of models and the considerable time savings gained at this phase. In this work, the traditional Rosenbrock (RNB) algorithm is combined with a random sampling method and the Latin hypercube (LH) to optimize a multi-start strategy and test the efficiency in the calibration of CRRMs. Three models (the French rural-engineering-with-four-daily-parameters (GR4J) model, the Swedish Hydrological Office Water-balance Department (HBV) model and the Sacramento Soil Moisture Accounting (SAC-SMA) model) are selected for WRA at nine headwaters in Spain in zones prone to long and severe droughts. To assess the results, the University of Arizona’s shuffled complex evolution (SCE-UA) algorithm was selected as a benchmark, because, until now, it has been one of the most robust techniques used to solve calibration problems with rainfall–runoff models. This comparison shows that the traditional algorithm can find optimal solutions at least as good as the SCE-UA algorithm. In fact, with the calibration of the SAC-SMA model, the results are significantly different: The RNB algorithm found better solutions than the SCE-UA for all basins. Finally, the combination created between the LH and RNB methods is detailed thoroughly, and a sensitivity analysis of its parameters is used to define the set of optimal values for its efficient performance.

Improving flood forecasting using conditional bias-penalized ensemble Kalman filter

We present a novel ensemble extension of the conditional bias-penalized Kalman filter, referred to herein as the conditional bias-penalized ensemble Kalman filter (CBEnKF), and apply it to flood forecasting. The CBEnKF differs from most data assimilation (DA) techniques in that it minimizes a weighted sum of the error variance and the expected value of the Type-II conditional bias squared for improved estimation and prediction of extremes. To assess the ability of the CBEnKF to improve flood prediction, we carried out real-world DA experiments in which the CBEnKF and the EnKF were applied under identical conditions for assimilation of hourly streamflow observations into the lumped Sacramento soil moisture accounting and unit hydrograph models. Ten headwater basins in Texas, whose drainage areas and times-to-peak range from 137 to 1037 km2 and from 3 to 21 hrs, respectively, were used in the twin experiments. All events in a 10-yr period with peak flow exceeding 100 m3/s were used. Verification of the ensemble mean predictions indicates that the CBEnKF improves the multi-basin mean of the mean square error skill score over the EnKF by about 0.15 over lead times of up to the time-to-peak of the basin. Verification of the ensemble predictions indicates that the CBEnKF improves the mean continuous ranked probability skill score by about 0.2 on average over the EnKF for all ranges of flow within the significant events, and by about 0.3 for flows exceeding the 95th percentile in those events. That the gain in skill is larger for larger flows makes the CBEnKF very appealing despite the significantly higher computational cost.

Enhancing Predictions in Ungauged Basins Using Machine Learning to Its Full Potential

In ungauged basins, long short-term memory (LSTM) networks provide unparalleled precision in prediction. Using k-fold validation, we trained and tested various LSTMs on 531 basins from the CAMELS data set, allowing us to make predictions in basins with no training data. The training and test data set contained 30 years of daily rainfall-runoff data from US catchments ranging in size from 4 to 2,000 km2, with aridity indexes ranging from 0.22 to 5.20, and 12 of the 13 IGPB vegetated land cover classes. Over a 15-year validation period, this effectively "ungauged" model was compared to the Sacramento Soil Moisture Accounting (SAC-SMA) model as well as the NOAA National Water Model reanalysis. Each basin's SAC-SMA was calibrated separately using 15 years of daily data. Across the 531 basins, the out-of-sample LSTM exhibited greater median Nash-Sutcliffe Efficiencies (0.69) than either the calibrated SAC-SMA (0.64) or the National Water Model (0.64). (0.58). This means that there is usually enough information in available catchment attributes data about similarities and differences between catchment-level rainfall-runoff behaviors to generate out-of-sample simulations that are generally more accurate than current models under ideal (i.e., calibrated) conditions. We discovered evidence that adding physical restrictions to the LSTM models improves simulations, which we believe should be the focus of future physics-guided machine learning research.

Legacy, Rather Than Adequacy, Drives the Selection of Hydrological Models

AbstractThe findings of hydrological modeling studies depend on which model was used. Although hydrological model selection is a crucial step, experience suggests that hydrologists tend to stick to the model they have experience with, and rarely switch to competing models, although these models might be more adequate given the study objectives. To gain quantitative insights into model selection, we explored the use of seven rainfall‐runoff models based on the abstract of 1,529 peer‐reviewed papers published between 1991 and 2018. The models selected were the Hydrologiska Byråns Vattenbalansavdelning model (HBV), the Variable Infiltration Capacity model (VIC), the mesoscale Hydrological model (mHM), the TOPography‐based hydrologic model (TOPMODEL), the Precipitation Runoff Modelling System (PRMS), the Génie Rural model à 4 paramètres Journaliers (GR4J), and the Sacramento soil moisture accounting model. We provide quantitative evidence of regional preferences in model use across the world and demonstrate that specific models are consistently preferred by certain institutes. Model attachment is particularly strong. In ~74% of the studies, the model selected can be predicted solely based on the affiliation of the first author. The influence of adequacy on the model selection process is less clear. Our data reveal that each model is used across a wide range of purposes, landscapes, and temporal and spatial scales (i.e., as a model of everything and everywhere). Model intercomparisons can provide guidance for model selection and improve model adequacy, but they are still rare (because each model must usually be setup individually) and the insights they provide are currently limited (because they are rarely controlled experiments). We suggest that moving from fixed‐structure models to modular modeling frameworks (master templates for model generation) can overcome these issues, enable a more collaborative and responsive model development environment, and result in improved model adequacy.