Predictions In Ungauged Basins Research Articles

Streamflow prediction in ungauged basins: How dissimilar are drainage basins?

Hydrological processes leading to flow in river channels are extremely complex, which is why we cannot use any fundamental law to fully explain river flow dynamics at basin scale. The general approach for hydrological modelling considers that every catchment’s response to precipitation input is unique, which means different catchments respond differently to the same precipitation input. In other words, river basins are sufficiently dissimilar to warrant having hydrological models with multiple free-parameters for predicting discharge. It is assumed that free-parameters are related to static catchment characteristics. However, past efforts to relate model parameters with observable catchment characteristics have not been very successful. We thus need a framework to quantify catchment dissimilarities exhibited by catchment characteristics because there is no need of free-parameters if we decide to ignore dissimilarities. Thus, relative role of climate and catchment characteristics in streamflow response of the basin needs to be quantified. In this study, we have considered daily data from 3534 gauging stations spread across the globe to compare HBV, a widely used hydrological model, with Dynamic Budyko (DB) model, a recently proposed calibration-free model. The HBV model assumes that both climate and catchment characteristics are important in modeling streamflow response, whereas a calibration-free DB model essentially assumes that streamflow response is largely governed by climate. The performance of HBV run with regionalized parameters and DB in terms of median coefficient of determination (R2) and weighted R2 (WR2) are (0.43, 0.45) and (0.24, 0.27), respectively. The corresponding values of R2 and WR2 from quantile-quantile comparison are (0.94, 0.94) and (0.60, 0.59). Results here seem to support of idea of developing calibration-free models by appreciating similarities between basins, particularly if the objectives are prediction in ungauged basins and assessment of climate change impact on streamflow.

Optimal combinations of global evapotranspiration and terrestrial water storage products for catchment water balance

ABSTRACT During the last two decades, a great number of global products about evapotranspiration ( E ) and terrestrial water storage ( S ) have been released. This has led to numerous combinations for describing catchment water balance, and hence determining an appropriate combination has become an important issue. The main objective of this study is to evaluate various global products of E and S at the catchment scale and determine the most appropriate combination, taking two Korean catchments as examples. For E , we evaluated global evapotranspiration (Global ET), Global Land Evaporation Amsterdam Model (GLEAM), and Penman–Monteith–Leuning (PML). For S , six individual datasets of Gravity Recovery and Climate Experiment (GRACE) were evaluated, with Global Land Data Assimilation System (GLDAS) products used for comparison. We found that the performance is sensitive to the choice of E product while various S products displayed minimal differences in results. Based on four evaluation criteria, the combination of PML and GRACE-SH-GFZ is suggested as the most suitable pair. For the period between 2003 and 2013, these data exhibit noteworthy trends of increasing E , offset by decreasing S . Warming climate is suspected to be behind these trends. Our research presents an approach that allows for the estimation of monthly streamflow exclusively with global data products, which is an advancement in hydrological analysis and particularly useful for regions that lack in-situ data networks. This approach provides a new perspective in the application of global datasets for the assessment of water balance and could significantly improve predictions in ungauged basins.

Universal recession constants and their potential to predict recession flow

Streamflow prediction during dry or recession periods is crucial for water resources management. The storage-discharge (S-Q) relationship is generally used to model recession flow in the basin. Since storage is practically unobservable, the S-Q relationship is often quantified by analyzing the relationship between -dQ/dt and Q, which Brutsaert and Nieber (1977) observed to be of a power-law type: -dQ/dt=kQα, for natural basins. Integration of the above equation yields an expression for the streamflow at any time, t, during the recession period. However, this approach relies on the availability of the streamflow data and, therefore, cannot be applied directly to ungauged or data-scarce basins. Thus, there exists a necessity for an alternative approach to predict streamflow in the ungauged basins during the recession period. In this study, we investigated recession curves from 413 basins in the United States and determined the existence of a universal relationship (k=kuQN-λu) between the storage coefficient (k) and the average past discharge (QN) that can serve as a proxy for basin storage. We tested the potential of universal recession constants (ku and λu) in predicting the recession flow in both gauged and ungauged basins. It is observed that the median NSEBC, i.e., NSE calculated using the Box-Cox transformation of discharge, was 0.78 for the gauged basins and 0.55 for the ungauged basins. Our findings provide a model-independent method for predicting the recession discharge, which can pave new avenues towards improving the predictions in ungauged basins.

A General Method to Improve Runoff Prediction in Ungauged Basins Based on Remotely Sensed Actual Evapotranspiration Data

The availability of remotely sensed (RS) actual evapotranspiration (ET) provides a possibility for improving runoff prediction in ungauged basins. To develop a general practical method to improve runoff prediction by directly incorporating RS-ET into rainfall-runoff (RR) models, two modeling schemes are proposed: (i) using RS-ET as direct input; and (ii) using RS-ET as partial direct input. The principle is to use RS-ET in cases where the runoff prediction can be improved. The two schemes are compared in over 200 basins using three RR models (Xinanjiang model, SIMHYD, and GR4J) and RS-ET inverted from AVHRR, and the modeling results in ungauged basins are assessed using the spatial proximity method. Results show that: (i) it is beneficial to incorporate RS-ET into the Xinanjiang model for over 85% of the basins, but this is not the case for SIMHYD and GR4J models; (ii) further model improvements can be obtained by using RS-ET as partial direct input, and are achieved in 91.1%, 59.0%, and 53.2% of the basins for Xinanjiang, SIMHYD, and GR4J, respectively; and (iii) incorporation of RS-ET is more applicable for Xinanjiang while less so for GR4J, and the efficacy is superior for basins that are relatively arid and were originally poorly simulated. Overall, using RS-ET as partial direct input is recommended.

Ecohydrologic model with satellite-based data for predicting streamflow in ungauged basins

Information on water availability in basins can be crucial for making decisions for effective water resource management in basins. As the operation of hydrometric stations in Korea is mainly focused on flood season and large rivers, most basins have lack or no observed data. Consequently, this complicates water resource planning and management. Remote sensing data is emerging as a powerful alternative to hydrological information in ungauged basins. This study investigated the applicability of Satellite-Remote Sensed Data (SRSD) as a source for model calibration in Prediction in Ungauged Basins (PUB) through modeling. Remote sensed leaf area index (LAI), actual evapotranspiration, and soil moisture data were used. Each SRSD was used alone to calibrate a hydrologic model to predict the daily streamflow for 28 basins in Korea. A vegetation module was added to the existing hydrologic model to use LAI. Among the SRSDs tested, the model calibrated with LAI had the most robust performance, predicting streamflow with acceptable accuracy compared to the traditional calibration based on streamflow. In particular, since the model account for vegetation actively interacting with evapotranspiration and soil moisture in the season of low flow, the LAI-calibrated model showed an advantage in improving the flow prediction performance. Although further research is required to utilize evapotranspiration and soil moisture data, the overall results of the LAI-based calibration were promising for predicting streamflow in ungauged basins where observations are scarce or absent, given that the satellite-derived LAI data were used alone without any preprocessing such as a bias correction. However, the prediction performance of the LAI-calibrated model was found to have a statistically significant relationship with local conditions. Therefore, by evaluating and improving the potential of SRSD in different region and climatic conditions, it is expected that the application of the SRSD-only calibration method can be extended to various ungauged basins.

Streamflow Predictions in Ungauged Basins Using Recurrent Neural Network and Decision Tree-Based Algorithm: Application to the Southern Region of the Korean Peninsula

River runoff predictions in ungauged basins are one of the major challenges in hydrology. In the past, the approach using a physical-based conceptual model was the main approach, but recently, a solution using a data-driven model has been evaluated as more appropriate through several studies. In this study, a new data-driven approach combining various recurrent neural networks and decision tree-based algorithms is proposed. An advantage of recurrent neural networks is that they can learn long-term dependencies between inputs and outputs provided to the network. Decision tree-based algorithms, combined with recurrent neural networks, serve to reflect topographical information treated as constants and can identify the importance of input features. We tested the proposed approach using data from 25 watersheds publicly available on the Korean government’s website. The potential of the proposed approach as a regional hydrologic model is evaluated in the view that one regional model predicts river runoff in various watersheds using the leave-one-out cross-validation regionalization setup.

Exploring Signature‐Based Model Calibration for Streamflow Prediction in Ungauged Basins

AbstractCalibration of precipitation‐streamflow models to streamflow signatures is a promising approach for streamflow prediction in ungauged basins (PUB). The estimation of parameter and prediction uncertainty in this case is not trivial because: (a) calibration takes place in the signature domain, while predictions are required in the time domain, and (b) streamflow signatures are estimated (e.g., from donor catchments) rather than “observed” (computed from observed streamflow in the target catchment), and therefore particularly uncertain. This study investigates model calibration using estimated signatures in an Approximate Bayesian Computation framework. First, we construct a stochastic signature transfer model, based on seasonal flow duration curves. Second, we calibrate a precipitation‐streamflow model to the estimated signatures, accounting for their uncertainty. The proposed method is tested in six catchments of the Thur basin, Switzerland. Three data availability scenarios are considered: (a) concomitant scenario, where signatures are “observed,” (b) non‐concomitant scenario, where signatures are transferred from a different time period, and (c) regionalization scenario, where signatures are transferred from (neighboring) donor catchments. In this study, the switch from observed to regionalized signatures increases predictive streamflow uncertainty by 38% and worsens the (deterministic) fit to observations by 17% (in terms of Nash‐Sutcliffe efficiency). Despite this deterioration, posterior predictive uncertainty remains lower than prior predictive uncertainty generated using uniform priors over representative parameter ranges (“uncalibrated” model), which demonstrates the effectiveness of the proposed signature‐based calibration. More importantly, uncertainty is reliably estimated at the ungauged catchments, which represents a key advance in stochastic streamflow PUB.

Optimal Postprocessing Strategies With LSTM for Global Streamflow Prediction in Ungauged Basins

AbstractStreamflow prediction in ungauged basins (PUB) is challenging, and Long Short‐Term Memory (LSTM) is widely used to for such predictions, owing to its excellent migration performance. Traditional LSTM forced by meteorological data and catchment attribute data barely highlight the optimum data integration strategy for LSTM and its migration from data‐rich basins to ungauged ones. In this study, we experimented with 1,897 global catchments and found that LSTM‐corrected Global Hydrological Models (GHMs) outperformed uncorrected GHMs, improving the median Nash‐Sutcliff efficiency (NSE) from 0.03 to 0.66. Notably, there was a large gap between traditional LSTM modeling in ungauged basins and autoregressive modeling in data‐rich basins, and GHM‐forced LSTM were an effective way to close this gap in ungauged basins. The spatial heterogeneity of the performance of GHM‐forced LSTM was mainly influenced by three metrics (dryness, the leaf area index and latitude), which described the hydrological similarity among catchments. Weaker hydrological similarity among continental catchments results in larger variability in GHM‐forced LSTM, with the best performance in Siberia (NSE, 0.54) and the worst in North America (NSE, 0.10). However, the migration performance of GHM‐forced LSTM was significantly improved (NSE, 0.63) in ungauged basins when hydrological similarity was considered. This study stressed the advantages of GHM‐forced LSTM and due significance should be attached to hydrological similarities among catchments to improve hydrological prediction in ungauged catchments.

The suitability of differentiable, physics-informed machine learning hydrologic models for ungauged regions and climate change impact assessment

Abstract. As a genre of physics-informed machine learning, differentiable process-based hydrologic models (abbreviated as δ or delta models) with regionalized deep-network-based parameterization pipelines were recently shown to provide daily streamflow prediction performance closely approaching that of state-of-the-art long short-term memory (LSTM) deep networks. Meanwhile, δ models provide a full suite of diagnostic physical variables and guaranteed mass conservation. Here, we ran experiments to test (1) their ability to extrapolate to regions far from streamflow gauges and (2) their ability to make credible predictions of long-term (decadal-scale) change trends. We evaluated the models based on daily hydrograph metrics (Nash–Sutcliffe model efficiency coefficient, etc.) and predicted decadal streamflow trends. For prediction in ungauged basins (PUB; randomly sampled ungauged basins representing spatial interpolation), δ models either approached or surpassed the performance of LSTM in daily hydrograph metrics, depending on the meteorological forcing data used. They presented a comparable trend performance to LSTM for annual mean flow and high flow but worse trends for low flow. For prediction in ungauged regions (PUR; regional holdout test representing spatial extrapolation in a highly data-sparse scenario), δ models surpassed LSTM in daily hydrograph metrics, and their advantages in mean and high flow trends became prominent. In addition, an untrained variable, evapotranspiration, retained good seasonality even for extrapolated cases. The δ models' deep-network-based parameterization pipeline produced parameter fields that maintain remarkably stable spatial patterns even in highly data-scarce scenarios, which explains their robustness. Combined with their interpretability and ability to assimilate multi-source observations, the δ models are strong candidates for regional and global-scale hydrologic simulations and climate change impact assessment.

Predicting streamflow with LSTM networks using global datasets

Streamflow predictions remain a challenge for poorly gauged and ungauged catchments. Recent research has shown that deep learning methods based on Long Short-Term Memory (LSTM) cells outperform process-based hydrological models for rainfall-runoff modeling, opening new possibilities for prediction in ungauged basins (PUB). These studies usually feature local datasets for model development, while predictions in ungauged basins at a global scale require training on global datasets. In this study, we develop LSTM models for over 500 catchments from the CAMELS-US data base using global ERA5 meteorological forcing and global catchment characteristics retrieved with the HydroMT tool. Comparison against an LSTM trained with local datasets shows that, while the latter generally yields superior performances due to the higher spatial resolution meteorological forcing (overall median daily NSE 0.54 vs. 0.71), training with ERA5 results in higher NSE in most catchments of Western and North-Western US (median daily NSE of 0.83 vs. 0.78). No significant changes in performance occur when substituting local with global data sources for deriving the catchment characteristics. These results encourage further research to develop LSTM models for worldwide predictions of streamflow in ungauged basins using available global datasets. Promising directions include training the models with streamflow data from different regions of the world and with higher quality meteorological forcing.

Regional streamflow prediction in northwest Spain: A comparative analysis of regionalisation schemes

Study Region: The present study was conducted in 24 watersheds located in the region of Galicia, in the northwest of Spain, covering an extension of approximately 13,000 km2.Study focus: This study is focused on the application and evaluation of different schemes for streamflow Prediction in Ungauged Basins (PUB). The MHIA model (Spanish acronym for Modelo HIdrológico Agregado), is first used to reproduce the observed time series of discharge in several gauged basins. Then, six different regionalisation schemes are applied to transfer the hydrological model parameters to ungauged catchments. For that purpose, we explore and compare two physical similarity, two spatial proximity and two regression-based regionalisation schemes. Output averaging (also known as ensemble modelling) as well as parameter averaging implementations of the physical similarity and spatial proximity methods are analysed.New hydrological insights: The most efficient methods are those based on output averaging, with acceptable success rates (SR) in 88% of the cases. On the other hand, the parameter averaging-based methods have the lowest SR. The methods based on spatial proximity output averaging provide the best performance when the receptor basin has a sufficient number of nearby donor basins. On the other hand, the methods based on physical similarity output averaging show a better performance in areas where there is a low density of donor catchments. The regression-based methods showed the lowest performance in all cases. The existence of correlations between the performance of the regionalisation schemes and the area of the receptor catchments was observed, with higher performances in large basins than in small basins.

Runoff predictions in new-gauged basins using two transformer-based models

In hydrology, runoff predictions are challenging when the data is lacking (e.g., predictions in un-gauged basins (PUB) and predictions with limited data (PLD)). Here, PLD refers to the case that the data is not enough for training or fine-tuning a data-driven model well (e.g., a new-gauged basin). We also name PLD as PNB (predictions in new-gauged basins). The difference between PNB and PUB is that the new-gauged basins can provide some data (e.g., runoff observations) while the un-gauged basins cannot. The long short-term memory (LSTM)-based models have shown good performance in runoff predictions due to their advanced structures. However, those structures have low level of flexibility and two nonadjacent positions cannot communicate directly. For high level of flexibility and better performance, we propose a simple Transformer-based rainfall-runoff model named RRS-Former, and want to show its power and also the power of the previously proposed RR-Former in PNB compared with LSTM-based models. The main part of RRS-Former is attention modules in which two arbitrary positions can be connected directly. Four hydrological units including 241 basins in the Catchment Attributes and Meteorology for Large-Sample Studies (CAMELS) dataset are used to compare the performance of our Transformer-based models and that of LSTM-based models. Besides using the k-fold validation to test performance for PNB, we propose a new way named unit-fold validation, in which we train the models by using the basins in three hydrological units and then test the performance using basins in the rest one hydrological unit. The results based on both k-fold and unit-fold show that our RRS-Former and RR-Former have better performance and are more reliable compared with the LSTM-based models.

Beyond river discharge gauging: hydrologic predictions using remote sensing alone

This study suggests a radical approach to hydrologic predictions in ungauged basins, addressing the long standing challenge of issuing predictions when in-situ river discharge does not exist. A simple but powerful rationale for measuring and modeling river discharge is proposed, using coupled advances in hydrologic modeling and satellite remote sensing. Our approach presents a Surrogate River discharge driven Model (SRM) that infers Surrogate River discharge (SR) from remotely sensed microwave signals with the ability to mimic river discharge in varying topographies and vegetation cover, which is then used to calibrate a hydrological model enabling physical realism in the resulting river discharge profile by adding an estimated mean of river discharge via the Budyko framework. The strength of SRM comes from the fact that it only uses remotely sensed data in prediction. The approach is demonstrated for 130 catchments in the Murray Darling Basin (MDB) in Australia, a region of high economic and environmental importance. The newly proposed SR (SRL, representing L-band microwave) boosts the Nash-Sutcliffe Efficiency (NSE) of modeled flow, showing a mean NSE of 0.54, with 70% of catchments exceeding NSE 0.4. We conclude that SRM effectively predicts high-flow and low-flow events related to flood and drought. Overall, this new approach will significantly improve catchment simulation capacity, enhancing water security and flood forecasting capability not only in the MDB but also worldwide.

Continuous streamflow prediction in ungauged basins: long short-term memory neural networks clearly outperform traditional hydrological models

Abstract. This study investigates the ability of long short-term memory (LSTM) neural networks to perform streamflow prediction at ungauged basins. A set of state-of-the-art, hydrological model-dependent regionalization methods are applied to 148 catchments in northeast North America and compared to an LSTM model that uses the exact same available data as the hydrological models. While conceptual model-based methods attempt to derive parameterizations at ungauged sites from other similar or nearby catchments, the LSTM model uses all available data in the region to maximize the information content and increase its robustness. Furthermore, by design, the LSTM does not require explicit definition of hydrological processes and derives its own structure from the provided data. The LSTM networks were able to clearly outperform the hydrological models in a leave-one-out cross-validation regionalization setting on most catchments in the study area, with the LSTM model outperforming the hydrological models in 93 % to 97 % of catchments depending on the hydrological model. Furthermore, for up to 78 % of the catchments, the LSTM model was able to predict streamflow more accurately on pseudo-ungauged catchments than hydrological models calibrated on the target data, showing that the LSTM model's structure was better suited to convert the meteorological data and geophysical descriptors into streamflow than the hydrological models even calibrated to those sites in these cases. Furthermore, the LSTM model robustness was tested by varying its hyperparameters, and still outperformed hydrological models in regionalization in almost all cases. Overall, LSTM networks have the potential to change the regionalization research landscape by providing clear improvement pathways over traditional methods in the field of streamflow prediction in ungauged catchments.

Data for "Exploring Signature-Based Calibration of Hydrological Models for Streamflow Prediction in Ungauged Basins"

Data for "Exploring Signature-Based Calibration of Hydrological Models for Streamflow Prediction in Ungauged Basins"

Estimating the Routing Parameter of the Xin’anjiang Hydrological Model Based on Remote Sensing Data and Machine Learning

The parameters of hydrological models should be determined before applying those models to estimate or predict hydrological processes. The Xin’anjiang (XAJ) hydrological model is widely used throughout China. Since the prediction in ungauged basins (PUB) era, the regionalization of the XAJ model parameters has been a subject of intense focus; nevertheless, while many efforts have targeted parameters related to runoff yield using in-site data sets, classic regression has predominantly been applied. In this paper, we employed remotely sensed underlying surface data and a machine learning approach to establish models for estimating the runoff routing parameter, namely, CS, of the XAJ model. The study was conducted on 114 catchments from the Catchment Attributes and MEteorology for Large-sample Studies (CAMELS) data set, and the relationships between CS and various underlying surface characteristics were explored by a gradient-boosted regression tree (GBRT). The results showed that the drainage density, stream source density and area of the catchment were the three major factors with the most significant impact on CS. The best correlation coefficient (r), root mean square error (RMSE) and mean absolute error (MAE) between the GBRT-estimated and calibrated CS were 0.96, 0.06 and 0.04, respectively, verifying the good performance of GBRT in estimating CS. Although bias was noted between the GBRT-estimated and calibrated CS, runoff simulations using the GBRT-estimated CS could still achieve results comparable to those using the calibrated CS. Further validations based on two catchments in China confirmed the overall robustness and accuracy of simulating runoff processes using the GBRT-estimated CS. Our results confirm the following hypotheses: (1) with the help of large sample of catchments and associated remote sensing data, the ML-based approach can capture the nonstationary and nonlinear relationships between CS and the underlying surface characteristics and (2) CS estimated by ML from large samples has a robustness that can guarantee the overall performance of the XAJ mode. This study advances the methodology for quantitatively estimating the XAJ model parameters and can be extended to other parameters or other models.

A Non‐Parametric Circular Statistics‐Based Framework for Predicting Peakflow Seasonality at Ungauged Sites

AbstractRegional frequency analysis (RFA) serves as a useful tool for flood estimation at a regional level, to improve the precision of estimated flood quantiles or for prediction in ungauged basins. RFA is used extensively in hydrologic literature for estimation of the magnitude of the hydrologic variable (annual maximum flows), pooling data from several sites within a homogeneous region. However, in areas that experience a strong seasonality of the climate, the timing of a hazardous event, such as flood, may be equally important. Therefore, in such regions, improved estimation of annual peakflow seasonality is imperative for efficient management of water infrastructure or for undertaking timely preventive measures against floods. In this study, we propose a novel RFA framework for estimation of annual peakflow seasonality based on circular statistics. The framework consists of (a) selection of attributes based on catchment similarity using seasonality descriptors as a response, (b) formation of homogeneous regions based on the region of influence method, (c) homogeneity tests adapted to directional data on timing of annual peakflows, and (d) probabilistic prediction of annual peakflow seasonality using the non‐parametric regional circular density. Applicability of the proposed approach is first demonstrated on two synthetically generated homogeneous regions, with unimodal and bimodal annual peakflow seasonality, respectively. The proposed homogeneity test outperforms the existing test yielding H value < 1 for homogeneous regions. Further, real‐world applications are illustrated on prediction of flood peak timing in the Northwest USA, revealing nearly 75% of the sites with absolute bias <16 days.

Utilization of the Long Short-Term Memory network for predicting streamflow in ungauged basins in Korea

Hydrological prediction is essential for managing and preserving headwater, wetland, and rural basins, yet it is difficult due to a lack of data. The Long Short-Term Memory (LSTM) network is a promising deep learning approach and has demonstrated excellent performance in streamflow prediction. However, due to the characteristics that require abundant and high-quality observations, most LSTM applications focus on investigating performance for data-rich basins. To investigate the applicability of LSTM in ungauged basins without hydrological observations, we introduce and test an approach to predict streamflow in ungauged basins using the LSTM network that has learned the integration of data from multiple gauged basins. Four learning strategies are constructed using various datasets such as ground-observed meteorological data, satellite data, and hydro-geomorphological characteristics. As a result, the LSTM network that learned the meteorological data in multiple gauged basins satisfactorily predicted streamflow in ungauged basins. The LSTM network that learned the satellite data further showed affirmative results. On the other hand, the LSTM, which additionally learned the hydro-geomorphological characteristics of the basins, exposed the need for improvement, such as securing additional various types of training data. In addition, it was recognized that additional efforts were needed to solve overfitting and out-of-distribution prediction problems. However, this approach achieved model performance deriving metrics above threshold values. These results show the potential of LSTM for streamflow prediction in ungauged basins including wetlands.

Applying non-parametric Bayesian networks to estimate maximum daily river discharge: potential and challenges

Abstract. Non-parametric Bayesian networks (NPBNs) are graphical tools for statistical inference widely used for reliability analysis and risk assessment and present several advantages, such as the embedded uncertainty quantification and limited computational time for the inference process. However, their implementation in hydrological studies is still scarce. Hence, to increase our understanding of their applicability and extend their use in hydrology, we explore the potential of NPBNs to reproduce catchment-scale hydrological dynamics. Long-term data from 240 river catchments with contrasting climates across the United States from the Catchment Attributes and Meteorology for Large-sample Studies (CAMELS) data set will be used as actual means to test the utility of NPBNs as descriptive models and to evaluate them as predictive models for maximum daily river discharge in any given month. We analyse the performance of three networks, one unsaturated (hereafter UN-1), one saturated (hereafter SN-1), both defined only by hydro-meteorological variables and their bivariate correlations, and one saturated network (hereafter SN-C), consisting of the SN-1 network and including physical catchments' attributes. The results indicate that the UN-1 network is suitable for catchments with a positive dependence between precipitation and discharge, while the SN-1 network can also reproduce discharge in catchments with negative dependence. The latter can reproduce statistical characteristics of discharge (tested via the Kolmogorov–Smirnov statistic) and have a Nash–Sutcliffe efficiency (NSE) ≥0.5 in ∼40 % of the catchments analysed, receiving precipitation mainly in winter and located in energy-limited regions at low to moderate elevation. Further, the SN-C network, based on similarity of the catchments, can reproduce discharge statistics in ∼10 % of the catchments analysed. We show that once a NPBN is defined, it is straightforward to infer discharge and to extend the network itself with additional variables, i.e. going from the SN-1 network to the SN-C network. However, the results also suggest considerable challenges in defining a suitable NPBN, particularly for predictions in ungauged basins. These are mainly due to the discrepancies in the timescale of the different physical processes generating discharge, the presence of a “memory” in the system, and the Gaussian-copula assumption used for modelling multivariate dependence.

Strength and Memory of Precipitation's Control Over Streamflow Across the Conterminous United States

AbstractHow precipitation (P) is translated into streamflow (Q) and over what timescales (i.e., “memory”) is difficult to predict without calibration of site‐specific models or using geochemical approaches, posing barriers to prediction in ungauged basins or advancement of general theories. Here, we used a data‐driven approach to identify regional patterns and exogenous controls on P–Q interactions. We applied an information flow analysis, which quantifies uncertainty reduction, to a daily time series of P and Q from 671 watersheds across the conterminous United States. We first demonstrated that information transfer from P to Q primarily reflects the quickflow component of water‐budgets, based on a watershed model. Readily quantifiable information flows show a functional relationship with model parameters, suggesting utility for model calibration. Second, applied to real watersheds, P–Q information flows exhibit seasonally varying behavior within regions in a manner consistent with dominant runoff generation mechanisms. However, the timing and the magnitude of information flows also reflect considerable subregional heterogeneity, likely attributable to differences in watershed size, baseflow contributions, and variation in aerial coverage of preferential flow paths. A regression analysis showed that a combination of climate and watershed characteristics are predictive of P–Q information flows. Though information flows cannot, in most cases, uniquely determine dominant runoff mechanisms, they provide a means to quantify the heterogeneous outcomes of those mechanisms within regions, thereby serving as a benchmarking tool for models developed at the regional scale. Last, information flows characterize regionally specific ways in which catchment connectivity changes from the wet to dry season.