Search

This work explores the relationship between catchment size, rainfall duration, and future streamflow increases on 133 North American catchments with sizes ranging from 66.5 to 9886 km2. It uses the outputs from a high spatial (0.11°) and temporal (1-h) resolution single model initial-condition large ensemble (SMILE) and a hydrological model to compute extreme rainfall and streamflow for durations ranging from 1 to 72 h and for return periods of between 2 and 300 years. Increases in extreme precipitation are observed across all durations and return periods. The projected increases are strongly related to duration, frequency, and catchment size, with the shortest durations, longest return periods, and smaller catchments witnessing the largest relative rainfall increases. These increases can be quite significant, with the 100-yr rainfall becoming up to 20 times more frequent over the smaller catchments. A similar duration–frequency–size pattern of increases is also observed for future extreme streamflow, but with even larger relative increases. These results imply that future increases in extreme rainfall will disproportionately impact smaller catchments, and particularly so for impervious urban catchments which are typically small, and whose stormwater drainage infrastructures are designed for long-return-period flows, both being conditions for which the amplification of future flow will be maximized. [ABSTRACT FROM AUTHOR]

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. [ABSTRACT FROM AUTHOR]

Predicting streamflow in ungauged basins is a critical task in many projects. This study investigates the applicability and performance of three common regionalization methods over 314 catchments distributed throughout Africa, using the HMETS hydrological model. This study finds that regionalization methods that work well in other regions of the world also do so in the African catchments, except for multiple linear regression which performs worse than expected. Furthermore, skill-based donor filtering is shown to be problematic in some circumstances, notably when the filter rejects mediocre but still useful donors. This study shows that regionalization performance depends more on the hydrological regime of the ungauged catchment and the capacity of hydrological models to adapt to that regime, rather than on any physiographic or climatological metric. In the context of ungauged basins, this indicates that there is no easy recipe to predict the skill of regionalization methods on any given catchment. [ABSTRACT FROM AUTHOR]

Ensemble forecasting applied to the field of hydrology is currently an established area of research embracing a broad spectrum of operational situations. This work catalogs the various pathways of ensemble streamflow forecasting based on an exhaustive review of more than 700 studies over the last 40 years. We focus on the advanced state of the art in the model‐based (dynamical) ensemble forecasting approaches. Ensemble streamflow prediction systems are categorized into three leading classes: statistics‐based streamflow prediction systems, climatology‐based ensemble streamflow prediction systems and numerical weather prediction‐based hydrological ensemble prediction systems. For each ensemble approach, technical information, as well as details about its strengths and weaknesses, are provided based on a critical review of the studies listed. Through this literature review, the performance and uncertainty associated with the ensemble forecasting systems are underlined from both operational and scientific viewpoints. Finally, the remaining key challenges and prospective future research directions are presented, notably through hybrid dynamical ‐ statistical learning approaches, which obviously present new challenges to be overcome in order to allow the successful employment of ensemble streamflow forecasting systems in the next decades. Targeting students, researchers and practitioners, this review provides a detailed perspective on the major features of an increasingly important area of hydrological forecasting. Key Points: This work summarizes the 40 years of research in the generation of streamflow forecasts based on an exhaustive review of studiesEnsemble prediction systems are categorized into three classes: statistics‐based, climatology‐based and numerical weather prediction‐based hydrological ensemble prediction systemsFor each ensemble forecasting system, thorough technical information is provided [ABSTRACT FROM AUTHOR]

Resampling historical time series remains one of the main approaches used to generate long-term probabilistic streamflow forecasts, while there is a need to develop more flexible approaches taking into account non-stationarities. One possible approach is to use a modelling chain consisting of a stochastic weather generator and a hydrological model. However, the ability of this modelling chain to generate adequate probabilistic streamflows must first be evaluated. The aim of this paper is to compare the performance of a stochastic weather generator against resampling historical meteorological time series in order to produce ensemble streamflow forecasts. The comparison framework is based on 30 years of forecasts for a single Canadian watershed. Forecasts resulting from the two methods are evaluated using the continuous ranked probability score (CRPS) and rank histograms. Results indicate that while there are differences between the methods, they nevertheless perform similarly, thus showing that weather generators can be used as substitutes for resampling the historical past. [ABSTRACT FROM AUTHOR]

A data set containing hydrometeorological and hydroclimatological data for 3,540 watersheds in North America is described. The data set contains four main parts: (a) observed hydrometeorological data including daily streamflow observations, precipitation, minimum temperature, and maximum temperature; (b) 20 bias‐corrected climate model projections for two Representative Concentration Pathway (RCP) scenarios and five bias correction methods; (c) hydrological model calibration parameters and simulated streamflow for 4 hydrological models, 2 objective functions, and 10 calibration parameter sets for the reference period; and (d) hydrological simulations for each of the combinations of the abovementioned elements of the climate change impact study chain, for a total of 16,000 combinations. The data set also contains simulations and bias‐corrected climate for 30‐year horizons corresponding to 1.5°C and 2°C temperature increases for a subset of the climate models, for an additional 8,000 combinations. All simulations in the reference period are also provided. Fifty‐one precomputed hydrological indices are made available for each simulation. Overall, 2.89 × 1012 years of simulations are classified, analyzed, compressed, and made available for all researchers. This data set can be used to evaluate the uncertainty of various components in the impact study chain, to establish relationships between catchment properties and hydrological response to climate change, and to evaluate the spatial distribution of hydrological change according to a multitude of hydrological indices. Key Points: A data set containing hydrometeorological observations and hydroclimatological simulations for 3,540 watersheds in North America is presentedThe data set is the first to provide large‐sample hydrological simulations and indices in the context of climate change impacts on hydrologyThis data set can be used for multiple purposes including impact assessment, uncertainty, and hydrological process analysis over large scales [ABSTRACT FROM AUTHOR]

Copyright of Canadian Water Resources Journal / Revue Canadienne des Ressources Hydriques is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

The potential impacts of floods are of significant concern to our modern society raising the need to identify and quantify all the uncertainties that can impact their simulations. Climate simulations at finer spatial resolutions are expected to bring more confidence in these hydrological simulations. However, the impact of the increasing spatial resolutions of climate simulations on floods simulations has to be evaluated. To address this issue, this paper assesses the sensitivity of summer–fall flood simulations to the Canadian Regional Climate Model (CRCM) grid resolution. Three climate simulations issued from the fifth version of the CRCM (CRCM5) driven by the ERA-Interim reanalysis at 0.44°, 0.22° and 0.11° resolutions are analysed at a daily time step for the 1981–2010 period. Raw CRCM5 precipitation and temperature outputs are used as inputs in the simple lumped conceptual hydrological model MOHYSE to simulate streamflows over 50 Quebec (Canada) basins. Summer–fall flooding is analysed by estimating four flood indicators: the 2-year, 5-year, 10-year and 20-year return periods from the CRCM5-driven streamflows. The results show systematic impacts of spatial resolution on CRCM5 outputs and seasonal flood simulations. Floods simulated with coarser climate datasets present smaller peak discharges than those simulated with the finer climate outputs. Smaller catchments show larger sensitivity to spatial resolution as more detail can be obtained from the finer grids. Overall, this work contributes to understanding the sensitivity of streamflow modelling to the climate model's resolution, highlighting yet another uncertainty source to consider in hydrological climate change impact studies. [ABSTRACT FROM AUTHOR]

Weather generators are usually used to produce an ensemble of climate time series for vulnerability assessments and impact studies in hydrological and agricultural communities. Multisite and multivariate weather generators (MMWGs) have many advantages over single-site counterparts in terms of coupling with distributed models for the assessment of spatial variability in various impact sectors. However, the existing MMWGs usually suffer from limitations in preserving multisite and multivariate dependencies at multiple time scales, as well as preserving the low-frequency variability of climate variables. This study proposes a new MMWG which preserves low-frequency climate variability by coupling annual, monthly and daily weather generators into a single model. Specifically, the daily precipitation and temperature time series generated by a widely used multisite daily weather generator is adjusted using monthly and annual climate time series generated by a first-order linear autoregressive model. This combination preserves the multisite and multivariate attributes, as well as low-frequency variability at monthly and annual scales. The performance of the proposed MMWG was evaluated by comparing the baseline model against that of its variants with monthly or annual adjustments for climate generation. The performance was also assessed using a hydrological model over two watersheds with different hydroclimatic characteristics. The results show that the proposed weather generator performs well with respect to reproducing the marginal distributional attributes, multisite and multivariate dependencies, and climate variability at the daily, monthly and annual scales. Weather generators with either monthly or annual adjustments (and not both) only improve the simulations in multisite and multivariate dependencies and low-frequency variability at the corresponding time scales. In terms of hydrological modeling, the proposed model consistently performs better than the baseline model and its variants with only monthly or annual adjustments in representing the mean and variance of monthly and annual streamflows. It also performs better in representing the frequency distribution of mean and extreme streamflow events. Overall, the proposed MMWG can effectively produce multisite and multivariate climate time series with low-frequency variability and has a strong potential for use in climate change impact studies. [ABSTRACT FROM AUTHOR]

• The impact of internal variability and climate change on streamflow is studied in 133 catchments of different sizes. • Internal climate variability is important for masking anthropogenic climate change, especially in the near future. • The size of the catchment used to have a significant impact on the timing of the emergence of the extreme flow. • Smaller catchments experience floods (droughts) at earlier (later) times. • Instead of KS-test and standard error of the mean (SEM), signal-to-noise (SNR) approaches are too conservative to detect the timing of emergence. Uncertainty about the future impacts of climate change represents a significant barrier to implementing adaptation measures. This work explores the impact of internal climate variability on streamflow projections for 133 catchments across the eastern and northeastern United States. Using data from a single model initial-condition large ensemble (SMILE) at high spatial and temporal resolution, this work assesses the magnitude of anthropogenic climate change and internal climate variability on projected future streamflow. The impact of catchment size is studied by grouping catchments into three different size classes (less than500 km2, between 500 and 1000 km2, and greater than 1000 km2). Results show that in a warmer climate, low to middle quantiles of future streamflow will systematically decrease, while the upper quantiles will increase. Increases are largest for more extreme streamflow indices. Using three different approaches, the role of internal variability is studied to estimate the time of emergence (TOE). In this case, results show that the climate change signal of extreme floods and droughts emerges later than that of median flow quantiles, even though the changes for floods as droughts is more significant. There is a clear relationship between catchment size and TOE, with small catchments seeing an earlier TOE for floods, and a later one for droughts. These results provide insight into adaptation times for small to large watersheds. [ABSTRACT FROM AUTHOR]

Bias correction is usually applied to climate model outputs before they are used as inputs to environmental models for impact studies. Every climate model is post-processed independently of others to account for biases originating from model structure and internal variability. To better understand the role of internal variability, multi-member ensembles (multiple runs of a single climate model, with identical forcing but different initial conditions) have now become common in the modeling community. Bias correcting such ensembles requires specific considerations. Correcting all members of such an ensemble independently would force all of them to the target distribution, thus removing the signature of natural variability over the calibration period. How this undesirable effect would propagate onto subsequent time periods is unknown. This study proposes three bias correction variants of a multi-member ensemble and compares their performances against an independent correction of each individual member of the ensemble. The comparison is based on precipitation and temperature, as well as on resulting streamflows simulated by a hydrological model. Two multi-member ensembles (5-member CanESM2 and 10-member CSIRO-MK3.6) were used for a subtropical monsoon watershed in China. The results show that all bias correction methods reduce precipitation and temperature biases for all ensemble members. As expected, independent correction reduces the spread of each ensemble over the calibration period. This is, however, followed by an overestimation of the spread over the subsequent validation period. Pooling all members to calculate common bias correction factors produces the best results over the calibration period; however, the difference among three bias correction variants becomes less clear over the validation period due to internal variability, and even less so when considering streamflows, as the impact model adds its own uncertainty. [ABSTRACT FROM AUTHOR]

Highlights • Calibration on full time series is shown to be more robust than split-sample methods. • 30 Bootstrapping tests on 6 cases provide evidence towards this method being optimal. • Verification on 10 independent catchment-model pairs support the conclusions. • Caveats of split-sampling on model performance are demonstrated. • Length of the calibration period is proportional to the parameter set robustness. Abstract This paper investigates the issues related to the use of validation in hydrological model calibration. Traditionally, models are calibrated and then assessed on an independent period (split-sample) to determine their adequacy in simulating streamflow as compared to observations. In this study, two hydrological models and three North American catchments are used to evaluate the effects of using validation to assess the model parameters' robustness on the model's actual simulation capabilities and accuracy in simulating streamflow. The length of the calibration period is increased from 1 to 16 years, and for each case a large number of randomly selected combinations of years are used for calibration and for validation using the Nash-Sutcliffe Efficiency metric. The calibrated model is then run on an independent 8-year test-period to assess the model's actual performance in simulation mode in unknown conditions. The process is bootstrapped 30 times to ensure the robustness of the results. The tests pit the calibration/validation methods on increasing calibration period lengths against a full calibration on the entire available dataset. Results show that the calibration on the full dataset is the optimal strategy as it generates the most robust parameter sets, provides the best model accuracy on an independent testing period and does not require assumption-making on the modeler's part. The calibrated parameter sets for each test-case were evaluated using the relative bias and correlation metrics, which revealed that the method transfers well to these two other metrics. Results also demonstrate the pitfalls of the commonly used split-sampling strategy, where good parameter sets may be discarded due to model performance discrepancies between calibration and validation periods. The conclusions point to the need to use as many years as possible in the calibration step and to entirely disregard the validation aspect under certain conditions. [ABSTRACT FROM AUTHOR]

Hydrological models are commonly used to forecast streamflow and for climate change impact studies. There is a wide range of hydrology models using lumped conceptual approaches all the way to more physically based distributed algorithms. Most of these models come with a steep learning curve before they can be used efficiently by the end user, and they can be tricky to calibrate appropriately. Only a small number of hydrology models can be considered easy to set up and use, and even fewer provide their source code for easy modification to be tailored to individual needs. These drawbacks make it difficult to use these models in educational applications. The goal of this paper is to introduce a very simple, yet efficient, lumped-conceptual hydrological model designed to address the above problems. The MATLAB-based HMETS hydrological model is simple and can be easily and quickly set up on a new watershed, including automatic calibration using state of the art optimization algorithms. Despite its simplicity, the model has proved to perform well against two other lumped-conceptual hydrological models over 320 watersheds. HMETS obtained a median Nash-Sutcliffe Efficiency of 0.72 in validation, compared to 0.64 for MOHYSE (similar structure) and 0.77 for HSAMI (more complex structure). The model's source code is freely available and includes an optional simplified user interface. A climate change impacts simulation tool using the constant scaling downscaling method is also incorporated to the interface. HMETS has been tested in the Construction Engineering Final-Year Project for a group of 60 undergraduate students. [ABSTRACT FROM AUTHOR]

This paper presents an analysis of three common hydrological regionalization methods (multiple linear regression, spatial proximity and physical similarity) in a virtual-world setting, using a 15 km resolution regional climate model to eliminate uncertainty due to measurement errors and missing data. It was found that in many cases the best donor is neither the most similar nor the closest watershed to the ungauged site, indicating a need for better hydrologically relevant catchment descriptors. Results show that using the closest donors yields satisfactory results only if they share similar characteristics with the ungauged basin, confirming that the proximity method is a good proxy only if there is reason to believe that the basins are physically similar. It was also shown that the ability to predict whether a method will succeed or fail is limited by the quality of catchment descriptors and the inherent probabilistic nature of the problem. A method to determine whether a regionalization method will fail or succeed based on the ungauged catchment’s characteristics failed to recognize a successful candidate 20% of the time, whereas it incorrectly classified a poor candidate in 30% of cases. The results indicate that there are unknown properties or processes that contribute to the hydrological behaviour of ungauged basins.EDITOR D. Koutsoyiannis; ASSOCIATE EDITOR F. Pappenberger [ABSTRACT FROM PUBLISHER]

This paper assesses the possibility of using multi-model averaging techniques for continuous streamflow prediction in ungauged basins. Three hydrological models were calibrated on the Nash-Sutcliffe Efficiency metric and were used as members of four multi-model averaging schemes. Model weights were estimated through optimization on the donor catchments. The averaging methods were tested on 267 catchments in the province of Québec, Canada, in a leave-one-out cross-validation approach. It was found that the best hydrological model was practically always better than the others used individually or in a multi-model framework, thus no averaging scheme performed statistically better than the best single member. It was also found that the robustness and adaptability of the models were highly influential on the models’ performance in cross-verification. The results show that multi-model averaging techniques are not necessarily suited for regionalization applications, and that models selected in such studies must be chosen carefully to be as robust as possible on the study site.Editor M.C. Acreman; Associate editor S. Grimaldi [ABSTRACT FROM PUBLISHER]

Over the last decades, different methods have been used by hydrologists to extend observed hydro-climatic time series, based on other data sources, such as tree rings or sedimentological datasets. For example, tree ring multiproxies have been studied for the Caniapiscau Reservoir in northern Québec (Canada), leading to the reconstruction of flow time series for the last 150 years. In this paper, we applied a new hydro-climatic reconstruction method on the Caniapiscau Reservoir and compare the obtained streamflow time series against time series derived from dendrohydrology by other authors on the same catchment and study the natural streamflow variability over the 1881-2011 period in that region. This new reconstruction is based not on natural proxies but on a historical reanalysis of global geopotential height fields, and aims firstly to produce daily climatic time series, which are then used as inputs to a rainfall-runoff model in order to obtain daily streamflow time series. The performances of the hydro-climatic reconstruction were quantified over the observed period, and showed good performances, in terms of both monthly regimes and interannual variability. The streamflow reconstructions were then compared to two different reconstructions performed on the same catchment by using tree ring data series, one being focused on mean annual flows and the other on spring floods. In terms of mean annual flows, the interannual variability in the reconstructed flows was similar (except for the 1930-1940 decade), with noteworthy changes seen in wetter and drier years. For spring floods, the reconstructed interannual variabilities were quite similar for the 1955-2011 period, but strongly different between 1880 and 1940. The results emphasize the need to apply different reconstruction methods on the same catchments. Indeed, comparisons such as those above highlight potential differences between available reconstructions and, finally, allow a retrospective analysis of the proposed reconstructions of past hydro-climatological variabilities. [ABSTRACT FROM AUTHOR]

This paper investigates the potential of reanalyses as proxies of observed surface precipitation and temperature to force hydrological models. Three global atmospheric reanalyses (ERA-Interim, CFSR, and MERRA), one regional reanalysis (NARR), and one global meteorological forcing dataset obtained by bias-correcting ERA-Interim [Water and Global Change (WATCH) Forcing Data ERA-Interim (WFDEI)] were compared to one gridded observation database over the contiguous United States. Results showed that all temperature datasets were similar to the gridded observation over most of the United States. On the other hand, precipitation from all three global reanalyses was biased, especially in summer and winter in the southeastern United States. The regional reanalysis precipitation was closer to observations since it indirectly assimilates surface precipitation. The WFDEI dataset was generally less biased than the reanalysis datasets. All datasets were then used to force a global conceptual hydrological model on 370 watersheds of the Model Parameter Estimation Experiment (MOPEX) database. River flows were computed for each watershed, and results showed that the flows simulated using NARR and gridded observations forcings were very similar to the observed flows. The simulated flows forced by the global reanalysis datasets were also similar to the observations, except in the humid continental and subtropical climatic regions, where precipitation seasonality biases degraded river flow simulations. The WFDEI dataset led to better river flows than reanalysis in the humid continental and subtropical climatic regions but was no better than reanalysis-and sometimes worse-in other climatic zones. Overall, the results indicate that global reanalyses have good potential to be used as proxies to observations to force hydrological models, especially in regions with few weather stations. [ABSTRACT FROM AUTHOR]

Reanalyses have the potential to provide meteorological information in areas where few or no traditional observation records are available. The terrestrial branch of the water cycle of CFSR, MERRA, ERA-Interim, and NARR is examined over Quebec, Canada, for the 1979-2008 time period. Precipitation, evaporation, runoff, and water balance are studied using observed precipitation and streamflows, according to three spatial scales: 1) the entire province of Quebec, 2) five regions derived from a climate classification, and 3) 11 river basins. The results reveal that MERRA provides a relatively closed water balance, while a significant residual was found for the other three reanalyses. MERRA and ERA-Interim seem to provide the most reliable precipitation over the province. On the other hand, precipitation from CFSR and NARR do not appear to be particularly reliable, especially over southern Quebec, as they almost systematically showed the highest and the lowest values, respectively. Moreover, the partitioning of precipitation into evaporation and runoff from MERRA and NARR does not agree with what was expected, particularly over southern, central, and eastern Quebec. Despite the weaknesses identified, the ability of reanalyses to reproduce the terrestrial water cycle of the recent past (i.e., 1979-2008) remains globally satisfactory. Nonetheless, their potential to provide reliable information must be validated by comparing reanalyses directly with weather stations, especially in remote areas. [ABSTRACT FROM AUTHOR]

Weather data are usually required at several locations over a large watershed, especially when using distributed models for hydrological simulations. In many applications, spatially correlated weather data can be provided by a multisite stochastic weather generator, which considers the spatial correlation of weather variables. Prior to using a multisite weather generator for hydrological modeling, its ability to adequately represent the proper hydrological response needs to be assessed. This study assesses the effectiveness of a new multisite weather generator (MulGETS) for hydrological modeling over a Canadian watershed in the Province of Québec. Prior to hydrological modeling, MulGETS is first evaluated with respect to reproducing the spatial correlation and statistical characteristics of precipitation and temperature for the studied watershed. Hydrological simulations obtained from MulGETS-generated precipitation and temperature are then compared with those obtained from a single-site weather generator (WeaGETS) and a WeaGETS-based lumped approach (WeaGETS-lumped) that averages the climate series over all stations in a watershed before running the single-site weather generator. The hydrology is simulated using two hydrological models: the conceptually lumped model HSAMI and the physically based distributed model CEQUEAU. When using the conceptually lumped model, the weather time series is first averaged over all stations in the watershed. The results show that the monthly mean discharge is accurately represented by both MulGETS-generated and WeaGETS-lumped-generated precipitation and temperature, whereas it is considerably underestimated by WeaGETS data for the snowmelt period. The MulGETS and WeaGETS-lumped data also show significant advantages in representing the monthly streamflow variability, which is underestimated by the WeaGETS outputs. Additionally, MulGETS and WeaGETS-lumped consistently perform better than WeaGETS for simulating extreme flows (snowmelt high flows and summer-autumn high and low flows). However, no obvious difference in performance was found between MulGETS and WeaGETSlumped data for hydrological modeling. Moreover, the use of a physically based distributed model with MulGETS did not result in any significant performance gain compared with the much simpler combination of WeaGETS-lumped with a lumped hydrological model for the studied watershed. Overall, this study indicates that a single-site weather generator combined with a lumped hydrological model is sufficient for accurate hydrological simulations, even in the case of a large watershed. [ABSTRACT FROM AUTHOR]

The bias correction of climate model outputs is based on the main assumption of the time invariance of the bias, in which the statistical relationship between observations and climate model outputs in the historical period stays constant in the future period. The present study aims to assess statistical bias correction under nonstationary bias conditions and its implications on the simulated streamflow over two snowmelt-driven Canadian catchments. A pseudoreality approach is employed in order to derive a proxy of future observations. In this approach, CRCM-ECHAM5 ensemble simulations are used as pseudoreality observations to correct for bias in the CRCM-CGCM3 ensemble simulations in the reference (1971-2000) period. The climate model simulations are then bias corrected in the future (2041-70) period and compared with the future pseudoreality observations. This process demonstrates that biases (precipitation and temperature) remain after the bias correction. In a second step, the uncorrected and bias-corrected CRCM-CGCM3 simulations are used to drive the Soil and Water Assessment Tool (SWAT) hydrological model in both periods. The bias correction decreases the error on mean monthly streamflow over the reference period; such findings are more mixed over the future period. The results of various hydrological indicators show that the climate change signal on streamflow obtained with uncorrected and bias-corrected simulations is similar in both its magnitude and its direction for the mean monthly streamflow only. Regarding the indicators of extreme hydrological events, more mixed results are found with site dependence. All in all, bias correction under nonstationary bias is an additional source of uncertainty that cannot be neglected in hydrological climate change impact studies. [ABSTRACT FROM AUTHOR]

Ensemble streamflow predictions (ESPs) offer great potential benefits for water resource management, as they contain key probabilistic information for analyzing prediction uncertainty. Ensemble weather forecasts (EWFs) are usually incorporated into ESPs to provide climate information. However, there is no simple way to combine both of them, since EWFs are generally biased and under-dispersed. This study presents a new short-term (1 to 7 lead days) probabilistic streamflow prediction system combining stochastic weather generation and EWFs. The bias and under-dispersion of EWFs were first corrected using a weather generator-based post-processing approach (GPP). The corrected weather forecasts were then coupled with a hydrological model for streamflow forecasts. The proposed GPP forecast was compared against two other forecasts, one using the raw EWF (GFS), and the other using a stochastic weather generator (WG). The comparison was carried out over two Quebec watersheds, using a set of deterministic and probabilistic verification metrics. The deterministic metrics showed that the GPP forecast is consistently the best at predicting the ensemble mean streamflow for both watersheds and for all the leads ranging between 1 and 7 days, followed by the WG forecast. The probabilistic metrics showed negative or near zero skill retained by the GFS forecast for the first 7 lead days. The WG system was much more skillful than the GFS. The GPP forecast consistently displayed the highest skill and reliability in terms of all the metrics applied. With increasing lead days, the skill and reliability of the GPP forecast tend to converge toward that of the WG forecast, indicating that the short-term GPP forecast could easily be linked to a pure WG forecast to extend the forecast horizon. [ABSTRACT FROM AUTHOR]

This paper focuses on evaluating the uncertainty of three common regionalization methods for predicting continuous streamflow in ungauged basins. A set of 268 basins covering 1.6 million km2 in the province of Quebec was used to test the regionalization strategies. The multiple linear regression, spatial proximity, and physical similarity approaches were evaluated on the catchments using a leave-one-out cross-validation scheme. The lumped conceptual HSAMI hydrological model was used throughout the study. A bootstrapping method was chosen to further estimate uncertainty due to parameter set selection for each of the parameter set/regionalization method pairs. Results show that parameter set selection can play an important role in regionalization method performance depending on the regionalization methods (and their variants) used and that equifinality does not contribute significantly to the overall uncertainty witnessed throughout the regionalization methods applications. Regression methods fail to consistently assign behavioral parameter sets to the pseudoungauged basins (i.e., the ones left out). Spatial proximity and physical similarity score better, the latter being the best. It is also shown that combining either physical similarity or spatial proximity with the multiple linear regression method can lead to an even more successful prediction rate. However, even the best methods were shown to be unreliable to an extent, as successful prediction rates never surpass 75%. Finally, this paper shows that the selection of catchment descriptors is crucial to the regionalization strategies' performance and that for the HSAMI model, the optimal number of donor catchments for transferred parameter sets lies between four and seven. [ABSTRACT FROM AUTHOR]

Summary: Statistical and dynamical downscaling techniques have been proposed to bridge the gaps between coarse-scale and generally biased climate model outputs and the point-scale requirements of impact model inputs. Amongst the various statistical approaches, empirical downscaling methods are the most commonly used due to their ease of implementation. Several empirical downscaling approaches have been proposed and need to be assessed as to which method contributes (or not) to the overall climate change uncertainty. Accordingly, this work aims at assessing the uncertainty of six empirical downscaling methods in quantifying the hydrological impact of climate change over two North American river basins under different climate conditions. The six empirical downscaling methods are grouped into change factor (two methods) and bias correction (four methods) approaches. The uncertainty related to the choice of an empirical downscaling method is compared to that associated with the choice of climate simulation, through the use of two Regional Climate Models (RCMs) driven by three different General Circulation Models (GCMs), totaling four RCM simulations, taken from the NARCCAP inter-comparison project. The future (2041–2065) hydrological regimes simulated with an empirical lumped hydrology model (HSAMI) are compared to the reference period (1971–1995) using a set of hydrology criteria which includes statistics of both mean and extreme values. The results show a large uncertainty envelope associated with the choice of a given empirical downscaling method, as well as for the choice of an RCM simulation. The uncertainty due to empirical downscaling and RCM simulation was more significant in projecting extreme streamflow than in projecting mean flows. Comparing the uncertainty envelope of empirical downscaling methods to the envelope resulting from four RCM simulations indicates that both are similar, even though the latter was slightly larger for some statistics. Finally, the uncertainty linked to the choice of an empirical downscaling approach (change factor vs. bias correction) was much larger than within each type. Overall, this work emphasizes the importance of using several climate projections and empirical downscaling approaches to delineate uncertainty when assessing the climate change impacts on hydrology. [Copyright &y& Elsevier]

Khalili, Malika, François Brissette, and Robert Leconte, 2011. Effectiveness of Multi-site Weather Generator for Hydrological Modeling. Journal of the American Water Resources Association (JAWRA) 1-12. DOI: 10.1111/j.1752-1688.2010.00514.x A multi-site weather generator has been developed using the concept of spatial autocorrelation. The multi-site generation approach reproduces the spatial autocorrelations observed between a set of weather stations as well as the correlations between each pair of stations. Its performance has been assessed in two previous studies using both precipitation and temperature data. The main objective of this paper is to assess the efficiency of this multi-site weather generator compared to a uni-site generator with respect to hydrological modeling. A hydrological model, known as Hydrotel, was applied over the Chute du Diable watershed, located in the Canadian province of Quebec. The distributed nature of Hydrotel accounts for the spatial variations throughout the watershed, and thus allows a more in-depth assessment of the effect of spatially dependent meteorological input on runoff generation. Simulated streamflows using both the multi-site and uni-site generated weather data were statistically compared to flows modeled using observed data. Overall, the hydrological modeling using the multi-site weather generator significantly outperformed that using the uni-site generator. This latter combined to Hydrotel resulted in a significant underestimation of extreme streamflows in all seasons. [ABSTRACT FROM AUTHOR]

• Stochastic weather generator is used a main computational tool for taking climate non-stationarity into account. • Parameters of stochastic weather generator are conditioned on the most relevant subset of large-scale climate indices. • Long-term streamflow forecast can be improved, particularly around spring flood. This paper aims at improving long-term streamflow forecasts by implementing a novel technique based on conditioning the parameters of a stochastic weather generator on large-scale climate indices, with varying lengths of training periods during the establishment of correlations. The most important climate indices are identified by looking at yearly correlations between a set of 40 indices and meteorological data (precipitation and temperature) at the watershed scale. A linear model is then constructed to identify precipitation and temperature anomalies to induce perturbations in the stochastic weather generator. Time windows of 5, 10, 15, 20 and 30 years are used in determining the optimal linear model. The performance of the proposed approach is assessed against that of a resampling of past climatology and using the same stochastic weather generator unconditioned on climate indices. Each member of the ensemble weather forecast is then fed to a hydrological model to create the Ensemble Streamflow Forecasts (ESF) with a one-year forecasting horizon. The three approaches are tested in hindcast mode over a 30-year period at 12 forecast dates. Results show that temperatures are significantly correlated with large-scale climate indices, whereas precipitation is only weakly related to the same indices. The length of the time window has a considerable impact on the prediction ability of the linear models. The precipitation models based on short duration time windows performed better than those based on longer windows, while the reverse was found for the temperature models. A comparison between all three Ensemble Streamflow Forecast approaches is assessed using the Continuous Ranked Probability Score (CRPS) metric. Results show that the proposed method improves long-term streamflow forecasting, particularly the volumetric bias and the peak flows during the spring flood. [ABSTRACT FROM AUTHOR]

This paper evaluated the effects of parametric reduction of a hydrological model on five regionalization methods and 267 catchments in the province of Quebec, Canada. The Sobol’ variance-based sensitivity analysis was used to rank the model parameters by their influence on the model results and sequential parameter fixing was performed. The reduction in parameter correlations improved parameter identifiability, however this improvement was found to be minimal and was not transposed in the regionalization mode. It was shown that 11 of the HSAMI models’ 23 parameters could be fixed with little or no loss in regionalization skill. The main conclusions were that (1) the conceptual lumped models used in this study did not represent physical processes sufficiently well to warrant parameter reduction for physics-based regionalization methods for the Canadian basins examined and (2) catchment descriptors did not adequately represent the relevant hydrological processes, namely snow accumulation and melt. [ABSTRACT FROM AUTHOR]

• A Long Short Term Memory neural network is applied to forecast streamflow. • The LSTM is compared to a distributed hydrological model using data assimilation. • The forecast skill of the LSTM is better for lead-times up to 7 days. • The LSTM does not require data assimilation when fed recent observed flow data. Streamflow forecasting is crucial in water planning and management. Physically-based hydrological models have been used for a long time in these fields, but improving forecast quality is still an active area of research. Recently, some artificial neural networks have been found to be effective in simulating and predicting short-term streamflow. In this study, we examine the reliability of Long Short-Term Memory (LSTM) deep learning model in predicting streamflow for lead times of up to ten days over a Canadian catchment. The performance of the LSTM model is compared to that of a process-based distributed hydrological model, with both models using the same weather ensemble forecasts. Furthermore, the LSTM's ability to integrate observed streamflow on the forecast issue date is compared to the data assimilation process required for the hydrological model to reduce initial state biases. Results indicate that the LSTM model forecasted streamflows are more reliable and accurate for lead-times up to 7 and 9 days, respectively. Additionally, it is shown that the LSTM model using recent observed flows as a predictor can forecast flows with smaller errors in the first forecasting days without requiring an explicit data assimilation step, with the LSTM model generating a median value of mean absolute error (MAE) for the first day of lead-time across all forecast issue dates of 25 m3/s compared to 115 m3/s for the assimilated hydrological model. [ABSTRACT FROM AUTHOR]

• Reducing initial forecasting errors using the EnKF data assimilation technique. • Improving real-time streamflow forecasts by an optimal EnKF implementation. • A seasonal adjustment of the EnKF hyperparameters improves forecast accuracy. • The vadose zone state variable is the most impactful and should be prioritized. • Data assimilation is most sensitive to temperature uncertainty for forecasting. Data Assimilation (DA) is an important step for improving prediction accuracy and real-time correction of hydrological models for operational forecasting purposes. The Ensemble Kalman filter (EnKF) is one of the most popular techniques used to address the issues of updating the model states and parameters by creating a novel set of initial conditions in real-time. This study aims at identifying optimal seasonal EnKF parametrizations to reduce the uncertainty associated with the initial conditions in a semi-distributed hydrological model of a snow-dominated catchment in Canada. Sensitivity analysis is performed to evaluate the effects of the EnKF individual hyperparameters (temperature, precipitation and inflow uncertainties) and the updating of three state variables (vadose zone, saturated zone and snowpack) on the skill of short-term (up to 9 days lead-time) forecasts. Results show that the performance of the forecasts is sensitive to the individual hyperparameters and particularly so to the temperature uncertainty, which varies between seasons. Additionally, the forecast skill is related to the choice of the state variables to be updated depending on the season. The vadose zone state variable displays higher importance and sensitivity than the other states, and the results indicate that all state variables should not be systematically updated on this catchment and with the implemented hydrological model. Finally, combining the best hyperparameter values with the optimal combination of state variables to update, insight is provided on the success of the data assimilation scheme that is evaluated on a rolling horizon using a set of seasonal rules to gain forecast performance over the span of multiple years. [ABSTRACT FROM AUTHOR]

Summary This study aims to test whether a weighted combination of several hydrological models can simulate flows more accurately than the models taken individually. In addition, the project attempts to identify the most efficient model averaging method and the optimal number of models to include in the weighting scheme. In order to address the first objective, streamflow was simulated using four lumped hydrological models (HSAMI, HMETS, MOHYSE and GR4J-6), each of which were calibrated with three different objective functions on 429 watersheds. The resulting 12 hydrographs (4 models × 3 metrics) were weighted and combined with the help of 9 averaging methods which are the simple arithmetic mean (SAM), Akaike information criterion (AICA), Bates–Granger (BGA), Bayes information criterion (BICA), Bayesian model averaging (BMA), Granger–Ramanathan average variant A, B and C (GRA, GRB and GRC) and the average by SCE-UA optimization (SCA). The same weights were then applied to the hydrographs in validation mode, and the Nash–Sutcliffe Efficiency metric was measured between the averaged and observed hydrographs. Statistical analyses were performed to compare the accuracy of weighted methods to that of individual models. A Kruskal–Wallis test and a multi-objective optimization algorithm were then used to identify the most efficient weighted method and the optimal number of models to integrate. Results suggest that the GRA, GRB, GRC and SCA weighted methods perform better than the individual members. Model averaging from these four methods were superior to the best of the individual members in 76% of the cases. Optimal combinations on all watersheds included at least one of each of the four hydrological models. None of the optimal combinations included all members of the ensemble of 12 hydrographs. The Granger–Ramanathan average variant C (GRC) is recommended as the best compromise between accuracy, speed of execution, and simplicity. [ABSTRACT FROM AUTHOR]

• The impacts of hydrological model components on performance and uncertainty of streamflow simulations are investigated. • A multi-model approach over 698 North America catchments is used. • No clear relationship between the structural and parametric complexity of the hydrological components and model performance is found. • No single model structure performs best in all climate zones and for all streamflow metrics. The impacts of hydrological model components on performance and uncertainty of streamflow simulations are investigated using a multi-model approach over 698 catchments in North America. An ensemble of 180 different hydrological model structures is obtained by combining five snow models (SM), four potential evapotranspiration formulations (PET), three vertical (VFS) and three horizontal (HFS) flow rounting schemes. This study aims to first compare the multiple component combinations of SM, PET, VFS and HFS in terms of their ability at simulating streamflows for ten Köppen climate zones. A secondary goal is to evaluate the contribution of each model component to the total variance of the streamflow response according to the model structure for three streamflow metrics representing mean, high and low flows. Results indicate that there is no clear relationship between the structural and parametric complexity of the hydrological components and model performance for all three streamflow metrics. There is no single model structure which performs best in all climate zones and for all streamflow metrics. Streamflow uncertainty generated by the 180 synthetic hydrological model structures depends primarily on the metrics under study, but also on the climate zone. The evapotranspiration formulation is only critical for the mean flow metric. For the low-flow criterion, the VFS generates the most uncertainty and is therefore the most critical, whereas for the high-flow metric, performance is mostly related to the HFS component and SM for Nordic catchments. A better understanding of how hydrological model components generate uncertainty for a given climate and streamflow metric will lead to more robust model structures and therefore reduce the uncertainty of hydrological simulations. [ABSTRACT FROM AUTHOR]