Potential of support vector regression for prediction of monthly streamflow using endogenous property

In the recent past, a variety of statistical and other modelling approaches have been developed to capture the properties of hydrological time series for their reliable prediction. However, the extent of complexity hinders the applicability of such traditional models in many cases. Kernel‐based machine learning approaches have been found to be more popular due to their inherent advantages over traditional modelling techniques including artificial neural networks(ANNs ). In this paper, a kernel‐based learning approach is investigated for its suitability to capture the monthly variation of streamflow time series. Its performance is compared with that of the traditional approaches. Support vector machines (SVMs) are one such kernel‐based algorithm that has given promising results in hydrology and associated areas. In this paper, the application of SVMs to regression problems, known as support vector regression (SVR), is presented to predict the monthly streamflow of the Mahanadi River in the state of Orissa, India. The results obtained are compared against the results derived from the traditional Box–Jenkins approach. While the correlation coefficient between the observed and predicted streamflows was found to be 0·77 in case of SVR, the same for different auto‐regressive integrated moving average (ARIMA) models ranges between 0·67 and 0·69. The superiority of SVR as compared to traditional Box‐Jenkins approach is also explained through the feature space representation. Copyright © 2009 John Wiley & Sons, Ltd.     

The integrated effects of climate and hydrologic uncertainty on future flood risk assessments

This work examines future flood risk within the context of integrated climate and hydrologic modelling uncertainty. The research questions investigated are (1) whether hydrologic uncertainties are a significant source of uncertainty relative to other sources such as climate variability and change and (2) whether a statistical characterization of uncertainty from a lumped, conceptual hydrologic model is sufficient to account for hydrologic uncertainties in the modelling process. To investigate these questions, an ensemble of climate simulations are propagated through hydrologic models and then through a reservoir simulation model to delimit the range of flood protection under a wide array of climate conditions. Uncertainty in mean climate changes and internal climate variability are framed using a risk‐based methodology and are explored using a stochastic weather generator. To account for hydrologic uncertainty, two hydrologic models are considered, a conceptual, lumped parameter model and a distributed, physically based model. In the conceptual model, parameter and residual error uncertainties are quantified and propagated through the analysis using a Bayesian modelling framework. The approach is demonstrated in a case study for the Coralville Dam on the Iowa River, where recent, intense flooding has raised questions about potential impacts of climate change on flood protection adequacy. Results indicate that the uncertainty surrounding future flood risk from hydrologic modelling and internal climate variability can be of the same order of magnitude as climate change. Furthermore, statistical uncertainty in the conceptual hydrological model can capture the primary structural differences that emerge in flood damage estimates between the two hydrologic models. Copyright © 2014 John Wiley & Sons, Ltd.     

Hydrologic drought prediction under climate change: Uncertainty modeling with Dempster–Shafer and Bayesian approaches

Representation and quantification of uncertainty in climate change impact studies are a difficult task. Several sources of uncertainty arise in studies of hydrologic impacts of climate change, such as those due to choice of general circulation models (GCMs), scenarios and downscaling methods. Recently, much work has focused on uncertainty quantification and modeling in regional climate change impacts. In this paper, an uncertainty modeling framework is evaluated, which uses a generalized uncertainty measure to combine GCM, scenario and downscaling uncertainties. The Dempster–Shafer (D–S) evidence theory is used for representing and combining uncertainty from various sources. A significant advantage of the D–S framework over the traditional probabilistic approach is that it allows for the allocation of a probability mass to sets or intervals, and can hence handle both aleatory or stochastic uncertainty, and epistemic or subjective uncertainty. This paper shows how the D–S theory can be used to represent beliefs in some hypotheses such as hydrologic drought or wet conditions, describe uncertainty and ignorance in the system, and give a quantitative measurement of belief and plausibility in results. The D–S approach has been used in this work for information synthesis using various evidence combination rules having different conflict modeling approaches. A case study is presented for hydrologic drought prediction using downscaled streamflow in the Mahanadi River at Hirakud in Orissa, India. Projections of n most likely monsoon streamflow sequences are obtained from a conditional random field (CRF) downscaling model, using an ensemble of three GCMs for three scenarios, which are converted to monsoon standardized streamflow index (SSFI-4) series. This range is used to specify the basic probability assignment (bpa) for a Dempster–Shafer structure, which represents uncertainty associated with each of the SSFI-4 classifications. These uncertainties are then combined across GCMs and scenarios using various evidence combination rules given by the D–S theory. A Bayesian approach is also presented for this case study, which models the uncertainty in projected frequencies of SSFI-4 classifications by deriving a posterior distribution for the frequency of each classification, using an ensemble of GCMs and scenarios. Results from the D–S and Bayesian approaches are compared, and relative merits of each approach are discussed. Both approaches show an increasing probability of extreme, severe and moderate droughts and decreasing probability of normal and wet conditions in Orissa as a result of climate change.     

Uncertainty estimation of a complex water quality model: The influence of Box–Cox transformation on Bayesian approaches and comparison with a non-Bayesian method

In urban drainage modelling, uncertainty analysis is of undoubted necessity. However, uncertainty analysis in urban water-quality modelling is still in its infancy and only few studies have been carried out. Therefore, several methodological aspects still need to be experienced and clarified especially regarding water quality modelling. The use of the Bayesian approach for uncertainty analysis has been stimulated by its rigorous theoretical framework and by the possibility of evaluating the impact of new knowledge on the modelling predictions. Nevertheless, the Bayesian approach relies on some restrictive hypotheses that are not present in less formal methods like the Generalised Likelihood Uncertainty Estimation (GLUE). One crucial point in the application of Bayesian method is the formulation of a likelihood function that is conditioned by the hypotheses made regarding model residuals. Statistical transformations, such as the use of Box–Cox equation, are generally used to ensure the homoscedasticity of residuals. However, this practice may affect the reliability of the analysis leading to a wrong uncertainty estimation. The present paper aims to explore the influence of the Box–Cox equation for environmental water quality models. To this end, five cases were considered one of which was the “real” residuals distributions (i.e. drawn from available data). The analysis was applied to the Nocella experimental catchment (Italy) which is an agricultural and semi-urbanised basin where two sewer systems, two wastewater treatment plants and a river reach were monitored during both dry and wet weather periods. The results show that the uncertainty estimation is greatly affected by residual transformation and a wrong assumption may also affect the evaluation of model uncertainty. The use of less formal methods always provide an overestimation of modelling uncertainty with respect to Bayesian method but such effect is reduced if a wrong assumption is made regarding the residuals distribution. If residuals are not normally distributed, the uncertainty is over-estimated if Box–Cox transformation is not applied or non-calibrated parameter is used.     

Robust vulnerability analysis of nuclear facilities subject to external hazards

Natural hazards have the potential to trigger complex chains of events in technological installations leading to disastrous effects for the surrounding population and environment. The threat of climate change of worsening extreme weather events exacerbates the need for new models and novel methodologies able to capture the complexity of the natural-technological interaction in intuitive frameworks suitable for an interdisciplinary field such as that of risk analysis. This study proposes a novel approach for the quantification of risk exposure of nuclear facilities subject to extreme natural events. A Bayesian Network model, initially developed for the quantification of the risk of exposure from spent nuclear material stored in facilities subject to flooding hazards, is adapted and enhanced to include in the analysis the quantification of the uncertainty affecting the output due to the imprecision of data available and the aleatory nature of the variables involved. The model is applied to the analysis of the nuclear power station of Sizewell B in East Anglia (UK), through the use of a novel computational tool. The network proposed models the direct effect of extreme weather conditions on the facility along several time scenarios considering climate change predictions as well as the indirect effects of external hazards on the internal subsystems and the occurrence of human error. The main novelty of the study consists of the fully computational integration of Bayesian Networks with advanced Structural Reliability Methods, which allows to adequately represent both aleatory and epistemic aspects of the uncertainty affecting the input through the use of probabilistic models, intervals, imprecise random variables as well as probability bounds. The uncertainty affecting the output is quantified in order to attest the significance of the results and provide a complete and effective tool for risk-informed decision making.     

Application of the SPARROW model in watersheds with limited information: a Bayesian assessment of the model uncertainty and the value of additional monitoring

How can spatially explicit nonlinear regression modelling be used for obtaining nonpoint source loading estimates in watersheds with limited information? What is the value of additional monitoring and where should future data‐collection efforts focus on? In this study, we address two frequently asked questions in watershed modelling by implementing Bayesian inference techniques to parameterize SPAtially Referenced Regressions On Watershed attributes (SPARROW), a model that empirically estimates the relation between in‐stream measurements of nutrient fluxes and the sources/sinks of nutrients within the watershed. Our case study is the Hamilton Harbour watershed, a mixed agricultural and urban residential area located at the western end of Lake Ontario, Canada. The proposed Bayesian approach explicitly accounts for the uncertainty associated with the existing knowledge from the system and the different types of spatial correlation typically underlying the parameter estimation of watershed models. Informative prior parameter distributions were formulated to overcome the problem of inadequate data quantity and quality, whereas the potential bias introduced from the pertinent assumptions is subsequently examined by quantifying the relative change of the posterior parameter patterns. Our modelling exercise offers the first estimates of export coefficients and delivery rates from the different subcatchments and thus generates testable hypotheses regarding the nutrient export ‘hot spots’ in the studied watershed. Despite substantial uncertainties characterizing our calibration dataset, ranging from 17% to nearly 400%, we arrived at an uncertainty level for the whole‐basin nutrient export estimates of only 36%. Finally, we conduct modelling experiments that evaluate the potential improvement of the model parameter estimates and the decrease of the predictive uncertainty if the uncertainty associated with the current nutrient loading estimates is reduced. Copyright © 2012 John Wiley & Sons, Ltd.     

Predicting river water temperatures using stochastic models: case study of the Moisie River (Québec,Canada)

Successful applications of stochastic models for simulating and predicting daily stream temperature have been reported in the literature. These stochastic models have been generally tested on small rivers and have used only air temperature as an exogenous variable. This study investigates the stochastic modelling of daily mean stream water temperatures on the Moisie River, a relatively large unregulated river located in Québec, Canada. The objective of the study is to compare different stochastic approaches previously used on small streams to relate mean daily water temperatures to air temperatures and streamflow indices. Various stochastic approaches are used to model the water temperature residuals, representing short‐term variations, which were obtained by subtracting the seasonal components from water temperature time‐series. The first three models, a multiple regression, a second‐order autoregressive model, and a Box and Jenkins model, used only lagged air temperature residuals as exogenous variables. The root‐mean‐square error (RMSE) for these models varied between 0·53 and 1·70 °C and the second‐order autoregressive model provided the best results. A statistical methodology using best subsets regression is proposed to model the combined effect of discharge and air temperature on stream temperatures. Various streamflow indices were considered as additional independent variables, and models with different number of variables were tested. The results indicated that the best model included relative change in flow as the most important streamflow index. The RMSE for this model was of the order of 0·51 °C, which shows a small improvement over the first three models that did not include streamflow indices. The ridge regression was applied to this model to alleviate the potential statistical inadequacies associated with multicollinearity. The amplitude and sign of the ridge regression coefficients seem to be more in agreement with prior expectations (e.g. positive correlation between water temperature residuals of different lags) and make more physical sense. Copyright © 2006 John Wiley & Sons, Ltd.     

Assessing rainfall–runoff models for the management of Lake Verbano

Growing human pressure and potential change in precipitation pattern induced by climate change require a more efficient and sustainable use of water resources. Hydrological models can provide a fundamental contribution to this purpose, especially as increasing availability of meteorological data and forecast allows for more accurate runoff predictions. In this article, two models are presented for describing the flow formation process in a sub‐alpine catchment: a distributed parameter, physically based model, and a lumped parameter, empirical model. The scope is to compare the two modelling approaches and to assess the impact of hydrometeorological information, either observations or forecast, on water resources management. This is carried out by simulating the real‐time management of the regulated lake that drains the catchment, using the inflow predictions provided by the two models. Copyright © 2010 John Wiley & Sons, Ltd.     

Time‐varying,nonlinear suspended sediment rating curves to characterize trends in water quality: An application to the Upper Hudson and Mohawk Rivers,New York

This work proposes two modelling frameworks for diagnosing temporal variations in nonlinear rating curves that describe suspended sediment–discharge relationships. A variant of the weighted regression on time, discharge, and season model is proposed and is compared against dynamic nonlinear modelling, a newly developed nonlinear time series filter based on sequential Monte Carlo sampling. Both approaches estimate a time series of rating curve parameters, with uncertainty, that can be used to diagnose variability in the sediment–discharge relationship over time. We evaluate the models with a variety of synthetic scenarios to highlight their ability to estimate signals of known rating curve change. Results reveal important bias‐variance trade‐offs unique to each approach, and in general, suggest that dynamic nonlinear modelling is better suited for rapid rating curve changes, whereas the weighted regression on time, discharge, and season variant more precisely estimates slow change. The techniques are then applied in two case studies in the Upper Hudson and Mohawk Rivers in New York. We conclude with a discussion of the implications of dynamic rating curves for the management of water quality in riverine and estuary systems.     

Modelling Freshwater Resources at the Global Scale: Challenges and Prospects

Quantification of spatially and temporally resolved water flows and water storage variations for all land areas of the globe is required to assess water resources, water scarcity and flood hazards, and to understand the Earth system. This quantification is done with the help of global hydrological models (GHMs). What are the challenges and prospects in the development and application of GHMs? Seven important challenges are presented. (1) Data scarcity makes quantification of human water use difficult even though significant progress has been achieved in the last decade. (2) Uncertainty of meteorological input data strongly affects model outputs. (3) The reaction of vegetation to changing climate and CO₂ concentrations is uncertain and not taken into account in most GHMs that serve to estimate climate change impacts. (4) Reasons for discrepant responses of GHMs to changing climate have yet to be identified. (5) More accurate estimates of monthly time series of water availability and use are needed to provide good indicators of water scarcity. (6) Integration of gradient-based groundwater modelling into GHMs is necessary for a better simulation of groundwater–surface water interactions and capillary rise. (7) Detection and attribution of human interference with freshwater systems by using GHMs are constrained by data of insufficient quality but also GHM uncertainty itself. Regarding prospects for progress, we propose to decrease the uncertainty of GHM output by making better use of in situ and remotely sensed observations of output variables such as river discharge or total water storage variations by multi-criteria validation, calibration or data assimilation. Finally, we present an initiative that works towards the vision of hyperresolution global hydrological modelling where GHM outputs would be provided at a 1-km resolution with reasonable accuracy.     

Ensemble prediction of regional droughts using climate inputs and the SVM–copula approach

In this study, the climate teleconnections with meteorological droughts are analysed and used to develop ensemble drought prediction models using a support vector machine (SVM)–copula approach over Western Rajasthan (India). The meteorological droughts are identified using the Standardized Precipitation Index (SPI). In the analysis of large‐scale climate forcing represented by climate indices such as El Niño Southern Oscillation, Indian Ocean Dipole Mode and Atlantic Multidecadal Oscillation on regional droughts, it is found that regional droughts exhibits interannual as well as interdecadal variability. On the basis of potential teleconnections between regional droughts and climate indices, SPI‐based drought forecasting models are developed with up to 3 months' lead time. As traditional statistical forecast models are unable to capture nonlinearity and nonstationarity associated with drought forecasts, a machine learning technique, namely, support vector regression (SVR), is adopted to forecast the drought index, and the copula method is used to model the joint distribution of observed and predicted drought index. The copula‐based conditional distribution of an observed drought index conditioned on predicted drought index is utilized to simulate ensembles of drought forecasts. Two variants of drought forecast models are developed, namely a single model for all the periods in a year and separate models for each of the four seasons in a year. The performance of developed models is validated for predicting drought time series for 10 years' data. Improvement in ensemble prediction of drought indices is observed for combined seasonal model over the single model without seasonal partitions. The results show that the proposed SVM–copula approach improves the drought prediction capability and provides estimation of uncertainty associated with drought predictions. Copyright © 2013 John Wiley & Sons, Ltd.     

The role of hydrological modelling uncertainties in climate change impact assessments of Irish river catchments

This study attempts to assess the uncertainty in the hydrological impacts of climate change using a multi-model approach combining multiple emission scenarios, GCMs and conceptual rainfall-runoff models to quantify uncertainty in future impacts at the catchment scale. The uncertainties associated with hydrological models have traditionally been given less attention in impact assessments until relatively recently. In order to examine the role of hydrological model uncertainty (parameter and structural uncertainty) in climate change impact studies a multi-model approach based on the Generalised Likelihood Uncertainty Estimation (GLUE) and Bayesian Model Averaging (BMA) methods is presented. Six sets of regionalised climate scenarios derived from three GCMs, two emission scenarios, and four conceptual hydrological models were used within the GLUE framework to define the uncertainty envelop for future estimates of stream flow, while the GLUE output is also post processed using BMA, where the probability density function from each model at any given time is modelled by a gamma distribution with heteroscedastic variance. The investigation on four Irish catchments shows that the role of hydrological model uncertainty is remarkably high and should therefore be routinely considered in impact studies. Although, the GLUE and BMA approaches used here differ fundamentally in their underlying philosophy and representation of error, both methods show comparable performance in terms of ensemble spread and predictive coverage. Moreover, the median prediction for future stream flow shows progressive increases of winter discharge and progressive decreases in summer discharge over the coming century.     

Probabilistic and ensemble simulation approaches for input uncertainty quantification of artificial neural network hydrological models

Artificial neural network (ANN) has been demonstrated to be a promising modelling tool for the improved prediction/forecasting of hydrological variables. However, the quantification of uncertainty in ANN is a major issue, as high uncertainty would hinder the reliable application of these models. While several sources have been ascribed, the quantification of input uncertainty in ANN has received little attention. The reason is that each measured input quantity is likely to vary uniquely, which prevents quantification of a reliable prediction uncertainty. In this paper, an optimization method, which integrates probabilistic and ensemble simulation approaches, is proposed for the quantification of input uncertainty of ANN models. The proposed approach is demonstrated through rainfall-runoff modelling for the Leaf River watershed, USA. The results suggest that ignoring explicit quantification of input uncertainty leads to under/over estimation of model prediction uncertainty. It also facilitates identification of appropriate model parameters for better characterizing the hydrological processes.     

Assessing the uncertainties of hydrologic model selection in climate change impact studies

The uncertainties associated with atmosphere‐ocean General Circulation Models (GCMs) and hydrologic models are assessed by means of multi‐modelling and using the statistically downscaled outputs from eight GCM simulations and two emission scenarios. The statistically downscaled atmospheric forcing is used to drive four hydrologic models, three lumped and one distributed, of differing complexity: the Sacramento Soil Moisture Accounting (SAC‐SMA) model, Conceptual HYdrologic MODel (HYMOD), Thornthwaite‐Mather model (TM) and the Precipitation Runoff Modelling System (PRMS). The models are calibrated based on three objective functions to create more plausible models for the study. The hydrologic model simulations are then combined using the Bayesian Model Averaging (BMA) method according to the performance of each models in the observed period, and the total variance of the models. The study is conducted over the rainfall‐dominated Tualatin River Basin (TRB) in Oregon, USA. This study shows that the hydrologic model uncertainty is considerably smaller than GCM uncertainty, except during the dry season, suggesting that the hydrologic model selection‐combination is critical when assessing the hydrologic climate change impact. The implementation of the BMA in analysing the ensemble results is found to be useful in integrating the projected runoff estimations from different models, while enabling to assess the model structural uncertainty. Copyright © 2011 John Wiley & Sons, Ltd.     

A sample reconstruction method based on a modified reservoir index for flood frequency analysis of non-stationary hydrological series

Flood extremes, affected by climate change and intense human activities, exhibit non-stationary characteristics. As a result, the stationarity assumption of traditional flood frequency analysis (FFA) cannot be satisfied. Generally, the impacts of human activities, especially water conservancy projects (i.e., reservoirs), on extreme flood series are much greater than those of climate change; therefore, new FFA methods must be developed to address the non-stationary flood extremes associated with large numbers of reservoirs. In this study, a new sample reconstruction method is proposed to convert the reservoir-influenced annual maximum flow (AMF) series from non-stationary to stationary, thus warranting the feasibility of the traditional FFA approach for non-stationary cases. To be more specifically, a modified reservoir index (MRI(t)) is proposed and the original non-stationary AMF series are converted to stationary series by multiplying by a scalar factor 1/(1 ? MRI(t)), and thus traditional FFA can be adopted. Besides, Bayesian theory was applied to analyze the effect of uncertainty on the designed reconstructed AMF. As an example, the proposed method was applied to observations from Huangzhuang station located on the Hanjiang River. The original AMF observations from Huangzhuang displayed nonstationarity for the continuous construction of reservoirs in the basin. After applying the new method of sample reconstruction, the original AMF observations became stationary, and the designed AMFs were estimated using the reconstructed series and compared with those estimated based on the original observation series. In addition, Bayesian theory is adopted to quantify the uncertainty of designed reconstructed AMF and provide the expectation of the sampling distribution.     

Effects of climatic variability on the hydrologic response of a freshwater watershed

A generalized watershed model was used to evaluate the effects of global climate changes on the hydrologic responses of freshwater ecosystems. The Enhanced Trickle Down (ETD) model was applied to W-3 watershed located near Danville, Vermont. Eight years of field data was used to perform model calibration and verification and the results were presented in Nikolaidis et al., (1993). Results from the Goddard Institute for Space Studies (GISS) and the Geophysical Fluid Dynamics Laboratory (GFDL) general circulation models which simulated the doubling of present day atmospheric CO₂ scenarios were used to perform the hydrologic simulations for the W-3 watershed. The results indicate that the W-3 watershed will experience increases in annual evapotranspiration and decreases in annual outflow and soil moisture. Stochastic models that simulate collective statistical properties of meteorological time series were developed to generate data to drive the ETD model in a Monte-Carlo fashion for quantification of the uncertainty in the model predictions due to input time series. This coupled deterministic and stochastic model was used to generate probable scenarios of future hydrology of the W-3 watershed. The predicted evapotranspiration and soil moisture under doubling present day atmospheric CO₂ scenarios exceed the present day uncertainty due to input time series by a factor greater than 2. The results indicate that the hydrologic response of the W-3 watershed will be significantly different than its present day response. The Enhanced Trickle Down model can be used to evaluate land surface feedbacks and assessing water quantity management in the event of climate change.     

Estimation of uncertainty sources in the projections of Lithuanian river runoff

Particular attention is given to the reliability of hydrological modelling results. The accuracy of river runoff projection depends on the selected set of hydrological model parameters, emission scenario and global climate model. The aim of this article is to estimate the uncertainty of hydrological model parameters, to perform sensitivity analysis of the runoff projections, as well as the contribution analysis of uncertainty sources (model parameters, emission scenarios and global climate models) in forecasting Lithuanian river runoff. The impact of model parameters on the runoff modelling results was estimated using a sensitivity analysis for the selected hydrological periods (spring flood, winter and autumn flash floods, and low water). During spring flood the results of runoff modelling depended on the calibration parameters that describe snowmelt and soil moisture storage, while during the low water period—the parameter that determines river underground feeding was the most important. The estimation of climate change impact on hydrological processes in the Merkys and Neris river basins was accomplished through the combination of results from A1B, A2 and B1 emission scenarios and global climate models (ECHAM5 and HadCM3). The runoff projections of the thirty-year periods (2011–2040, 2041–2070, 2071–2100) were conducted applying the HBV software. The uncertainties introduced by hydrological model parameters, emission scenarios and global climate models were presented according to the magnitude of the expected changes in Lithuanian rivers runoff. The emission scenarios had much greater influence on the runoff projection than the global climate models. The hydrological model parameters had less impact on the reliability of the modelling results.     

The influence of dependence in characterizing multi-variable uncertainty for climate change impact assessments

Few approaches exist that explicitly use the uncertainty associated with the spread of climate model simulations in assessing climate change impacts. An approach that does so is second-order approximation (SOA). This incorporates quantification of uncertainty to ascertain its impact on the derived response using a Taylor series expansion of the model. This study uses SOA in a statistical downscaling model of monthly streamflow, with a focus on the influence of dependence in the uncertainty of multiple atmospheric variables. Uncertainty is quantified using the square root error variance concept with a new extension that allows the inter-dependence terms among the atmospheric variable uncertainty to be specified. Applying the model to selected point locations in Australia, it is noted that the downscaling results differ considerably from downscaling that ignores uncertainty. However, when the effects of dependence in uncertainty are incorporated, the results differ according to the regional variations in dependence structure.