Comparison of local and global approximators in multivariate chaotic forecasting of daily streamflow

ABSTRACT

Although it is conceptually assumed that global models are relatively ineffective in modelling the highly unstable structure of chaotic hydrologic dynamics, there is not a detailed study of comparing the performances of local and global models in a hydrological context, especially with new emerging machine learning models. In this study, the performance of a local model (k-nearest neighbour, k-nn) and, as global models, several recent machine learning models – artificial neural network (ANN), least square-support vector regression (LS-SVR), random forest (RF), M5 model tree (M5), multivariate adaptive regression splines (MARS) – was analysed in multivariate chaotic forecasting of streamflow. The models were developed for Australia’s largest river, the River Murray. The results indicate that the k-nn model was more successful than the global models in capturing the streamflow dynamics. Furthermore, coupled with the multivariate phase-space, it was shown that the global models can be successfully used for obtaining reliable uncertainty estimates for streamflow.     

Application of newly developed ensemble machine learning models for daily suspended sediment load prediction and related uncertainty analysis

ABSTRACT

Ensemble machine learning models have been widely used in hydro-systems modeling as robust prediction tools that combine multiple decision trees. In this study, three newly developed ensemble machine learning models, namely gradient boost regression (GBR), AdaBoost regression (ABR) and random forest regression (RFR) are proposed for prediction of suspended sediment load (SSL), and their prediction performance and related uncertainty are assessed. The SSL of the Mississippi River, which is one of the major world rivers and is significantly affected by sedimentation, is predicted based on daily values of river discharge (Q) and suspended sediment concentration (SSC). Based on performance metrics and visualization, the RFR model shows a slight lead in prediction performance. The uncertainty analysis also indicates that the input variable combination has more impact on the obtained predictions than the model structure selection.     

Hydrologic response to future land use change in the Upper Mississippi River Basin by the end of 21st century

This study demonstrates the spatial variation in hydrologic processes across the Upper Mississippi River Basin (UMRB) by the end of 21st century, by ingesting FOREcasting Scenarios (FORE‐SCE) of Land‐use Change projections into a physics‐based hydrologic model—Soil and Water Assessment Tool. The model is created for UMRB (440,000 km²), using the National Landcover Database of year 2001 and climate data of 1991–2010. Considering 1991–2010 as the baseline reference period, FORE‐SCE projections of year 2091 under three scenarios (A1B, A2, and B1 from the Intergovernmental Panel on Climate Change) are separately assimilated into the calibrated model, whereas climate input is kept the same as in the baseline. Modeling results suggest an increase of 0.5% and 3.5% in the average annual streamflow at the basin outlet (Grafton, Illinois) during 2081–2100, respectively, for A1B and A2, whereas for B1, streamflow would decrease by 1.5%. Under the “worst case” A2 scenario, 6% and 133% increase, respectively, in agricultural and urban areas with 30% depletion of forest and grassland would result into 70% increase in surface runoff, 20% decrease in soil moisture, and 4% decrease in evapotranspiration in certain parts of the basin. Conversion of cropland, forest, or grassland to perennial hay/pasture areas would lower surface runoff by 25% especially in the central region, whereas persistent forest cover in the northern region would cause up to 7% increase in evapotranspiration. The ecosystem in the lower half of UMRB is likely to become adverse, as dictated by a composite water–energy balance indicator. Future land use change extents and resultant hydrologic responses are found significantly different under A2, A1B, and B1 scenarios, which resonates the need for multi‐scenario ensemble assessments towards characterizing a probable future. The spatial variation of hydrologic processes as shown here helps to identify potential “hot spots,” giving ways to adopt more effective policy alternatives at regional level.     

Little River Experimental Watershed,a keystone in understanding of coastal plain watersheds

The US Department of Agriculture-Agricultural Research Service Southeast Watershed Research Laboratory (SEWRL) initiated a hydrologic research program on the Little River Experimental Watershed (LREW) in 1967. Long-term (52 years) streamflow data are available for nine sites, including rainfall-runoff relationships and hydrograph characteristics regularly used in research on interactive effects of climate, vegetation, soils, and land-use in low-gradient streams of the US EPA Level III Southeastern Plains ecoregion. A summary of prior research on the LREW illustrates the impact of the watershed on building a regional understanding of hydrology and water quality. Climatic and streamflow data were used to make comparisons of scale across the nine nested LREW watersheds (LRB, LRF, LRI, LRJ, LRK, LRO, LRN, LRM, and LRO3) and two regional watersheds (Alapaha and Little River at Adel). Annual rainfall for the largest LREW, LRB, was 1200 mm while average annual streamflow was 320 mm. Annual rainfall, streamflow, and the ratio between annual streamflow and rainfall (Sratio) were similar (α = 0.05) across LREWs LRB, LRF, LRI, LRJ, LRK, and LRO. While annual rainfall within the 275 ha LRO3 was found to be similar to LRO and LRM (α = 0.05), annual streamflow and Sratio were significantly different (α = 0.05). Comparisons of annual rainfall, streamflow, and Sratio between LRB and the regional watersheds indicated no differences (α = 0.05). Based upon this analysis, most regional watersheds shared similar hydrologic characteristics. LRO3 was an exception, where increases in row crops and decreases in forest coverage resulted in increased streamflow. LREW data have been instrumental in building considerable scientific understanding of flow and transport processes for these stream systems. Continued operation of the LREW hydrologic network will support hydrologic research as well as environmental quality and riparian research programs that address emerging and high priority natural resource and environmental issues.     

Deep echo state network: a novel machine learning approach to model dew point temperature using meteorological variables

ABSTRACT

The potential of different models – deep echo state network (DeepESN), extreme learning machine (ELM), extra tree (ET), and regression tree (RT) – in estimating dew point temperature by using meteorological variables is investigated. The variables consist of daily records of average air temperature, atmospheric pressure, relative humidity, wind speed, solar radiation, and dew point temperature (T_dew) from Seoul and Incheon stations, Republic of Korea. Evaluation of the model performance shows that the models with five and three-input variables yielded better accuracy than the other models in these two stations, respectively. In terms of root-mean-square error, there was significant increase in accuracy when using the DeepESN model compared to the ELM (18%), ET (58%), and RT (64%) models at Seoul station and the ELM (12%), ET (23%), and RT (49%) models at Incheon. The results show that the proposed DeepESN model performed better than the other models in forecasting T_dew values.     

A process-based,distributed hydrologic model based on a large-scale method for surface–subsurface coupling

Process-based watershed models are useful tools for understanding the impacts of natural and anthropogenic influences on water resources and for predicting water and solute fluxes exported from watersheds to receiving water bodies. The applicability of process-based hydrologic models has been previously limited to small catchments and short time frames. Computational demands, especially the solution to the three-dimensional subsurface flow domain, continue to pose significant constraints. This paper documents the mathematical development, numerical testing and the initial application of a new distributed hydrologic model PAWS (Process-based Adaptive Watershed Simulator). The model solves the governing equations for the major hydrologic processes efficiently so that large scale applications become relevant. PAWS evaluates the integrated hydrologic response of the surface–subsurface system using a novel non-iterative method that couples runoff and groundwater flow to vadose zone processes approximating the 3D Richards equation. The method is computationally efficient and produces physically consistent solutions. All flow components have been independently verified using analytical solutions and experimental data where applicable. The model is applied to a medium-sized watershed in Michigan (1169 km²) achieving high performance metrics in terms of streamflow prediction at two gages during the calibration and verification periods. PAWS uses public databases as input and possesses full capability to interact with GIS datasets. Future papers will describe applications to other watersheds and the development and application of fate and transport modules.     

Potential impact of climate change on flooding in the Upper Great Miami River Watershed,Ohio, USA: a simulation-based approach

Abstract

This study aims to assess the potential impact of climate change on flood risk for the city of Dayton, which lies at the outlet of the Upper Great Miami River Watershed, Ohio, USA. First the probability mapping method was used to downscale annual precipitation output from 14 global climate models (GCMs). We then built a statistical model based on regression and frequency analysis of random variables to simulate annual mean and peak streamflow from precipitation input. The model performed well in simulating quantile values for annual mean and peak streamflow for the 20th century. The correlation coefficients between simulated and observed quantile values for these variables exceed 0.99. Applying this model with the downscaled precipitation output from 14 GCMs, we project that the future 100-year flood for the study area is most likely to increase by 10–20%, with a mean increase of 13% from all 14 models. 79% of the models project increase in annual peak flow.

Citation Wu, S.-Y. (2010) Potential impact of climate change on flooding in the Upper Great Miami River Watershed, Ohio, USA: a simulation-based approach. Hydrol. Sci. J. 55(8), 1251–1263.     

Water quality monitoring at a virtual watershed monitoring station using a modified deep extreme learning machine

ABSTRACT

A new deep extreme learning machine (ELM) model is developed to predict water temperature and conductivity at a virtual monitoring station. Based on previous research, a modified ELM auto-encoder is developed to extract more robust invariance among the water quality data. A weighted ELM that takes seasonal variation as the basis of weighting is used to predict the actual value of water quality parameters at sites which only have historical data and no longer generate new data. The performance of the proposed model is validated against the monthly data from eight monitoring stations on the Zengwen River, Taiwan (2002–2017). Based on root mean square error, mean absolute error, mean absolute percentage error and correlation coefficient, the experimental results show that the new model is better than the other classical spatial interpolation methods.     

HydroBlocks: a field‐scale resolving land surface model for application over continental extents

Land surface spatial heterogeneity plays a significant role in the water, energy, and carbon cycles over a range of temporal and spatial scales. Until now, the representation of this spatial heterogeneity in land surface models has been limited to over simplistic schemes because of computation and environmental data limitations. This study introduces HydroBlocks – a novel land surface model that represents field‐scale spatial heterogeneity of land surface processes through interacting hydrologic response units (HRUs). HydroBlocks is a coupling between the Noah‐MP land surface model and the Dynamic TOPMODEL hydrologic model. The HRUs are defined by clustering proxies of the drivers of spatial heterogeneity using high‐resolution land data. The clustering mechanism allows for each HRU's results to be mapped out in space, facilitating field‐scale application and validation. The Little Washita watershed in the USA is used to assess HydroBlocks' performance and added benefit from traditional land surface models. A comparison between the semi‐distributed and fully distributed versions of the model suggests that using 1000 HRUs is sufficient to accurately approximate the fully distributed solution. A preliminary evaluation of model performance using available in situ soil moisture observations suggests that HydroBlocks is generally able to reproduce the observed spatial and temporal dynamics of soil moisture. Model performance deficiencies can be primarily attributed to parameter uncertainty. HydroBlocks' ability to explicitly resolve field‐scale spatial heterogeneity while only requiring an increase in computation of one to two orders of magnitude when compared with existing land surface models is encouraging – ensemble field‐scale land surface modelling over continental extents is now possible. Copyright © 2016 John Wiley & Sons, Ltd.     

Diagnostic calibration and cross‐catchment transferability of a simple process‐consistent hydrologic model

The transferability of hydrologic models is of ever increasing importance for making improved hydrologic predictions and testing hypothesized hydrologic drivers. Here, we present an investigation into the variability and transferability of the recently introduced catchment connectivity model (Smith et al., 2013 ). The catchment connectivity model was developed following extensive experimental observations identifying the key drivers of streamflow in the Tenderfoot Creek Experimental Forest (Jencso et al., 2009 ; Jencso et al., 2010 ), with the goal of creating a simple model consistent with internal observations of catchment hydrologic connectivity patterns. The model was applied across seven catchments located within Tenderfoot Creek Experimental Forest to investigate spatial variability and transferability of model performance and parameterization. The results demonstrated that the model resulted in historically good fits (based on previous studies at the sites) to both the hydrograph and internal water table dynamics (corroborated with experimental observations). The impact of a priori parameter limits was also examined. It was observed that enforcing field‐based limits on model parameters resulted in slight reductions to streamflow hydrograph fits, but significant improvements to model process fidelity (as hydrologic connectivity), as well as moderate improvement in the transferability of model parameterizations from one catchment to the next. Copyright © 2016 John Wiley & Sons, Ltd.     

Impact of model development,calibration and validation decisions on hydrological simulations in West Lake Erie Basin

Watershed simulation models are used extensively to investigate hydrologic processes, landuse and climate change impacts, pollutant load assessments and best management practices (BMPs). Developing, calibrating and validating these models require a number of critical decisions that will influence the ability of the model to represent real world conditions. Understanding how these decisions influence model performance is crucial, especially when making science‐based policy decisions. This study used the Soil and Water Assessment Tool (SWAT) model in West Lake Erie Basin (WLEB) to examine the influence of several of these decisions on hydrological processes and streamflow simulations. Specifically, this study addressed the following objectives (1) demonstrate the importance of considering intra‐watershed processes during model development, (2) compare and evaluated spatial calibration versus calibration at outlet and (3) evaluate parameter transfers across temporal and spatial scales. A coarser resolution (HUC‐12) model and a finer resolution model (NHDPlus model) were used to support the objectives. Results showed that knowledge of watershed characteristics and intra‐watershed processes are critical to produced accurate and realistic hydrologic simulations. The spatial calibration strategy produced better results compared to outlet calibration strategy and provided more confidence. Transferring parameter values across spatial scales (i.e. from coarser resolution model to finer resolution model) needs additional fine tuning to produce realistic results. Transferring parameters across temporal scales (i.e. from monthly to yearly and daily time‐steps) performed well with a similar spatial resolution model. Furthermore, this study shows that relying solely on quantitative statistics without considering additional information can produce good but unrealistic simulations. Copyright © 2015 John Wiley & Sons, Ltd.     

Impacts of climate change on the Qingjiang Watershed’s runoff change trend in China

Qingjiang River, the second largest tributary of the Yangtze River in Hubei Province, has taken on the important tasks for power generation and flood control in Hubei Province. The Qingjiang River watershed has a subtropical monsoon climate and, as a result, has dramatic diversity in its water resources. Recently, global warming and climate change have seriously affected the Qingjiang watershed’s integrated water resources management. In this article, general circulation model (GCM) and watershed hydrological models were applied to analyze the impacts of climate change on future runoff of Qingjiang Watershed. To couple the scale difference between GCM and watershed hydrological models, a statistical downscaling method based on the smooth support vector machine was used to downscale the GCM’s large-scale output. With the downscaled precipitation and evaporation, the Xin-anjiang hydrological model and HBV model were applied to predict the future runoff of Qingjiang Watershed under A2 and B2 scenarios. The preformance of the one-way coupling approach in simulating the hydrological impact of climate change in the Qingjiang watershed is evaluated, and the change trend of the future runoff of Qingjiang Watershed under the impacts of climate change is presented and discussed.     

Streamflow forecasting using least-squares support vector machines

Abstract

This paper investigates the ability of a least-squares support vector machine (LSSVM) model to improve the accuracy of streamflow forecasting. Cross-validation and grid-search methods are used to automatically determine the LSSVM parameters in the forecasting process. To assess the effectiveness of this model, monthly streamflow records from two stations, Tg Tulang and Tg Rambutan of the Kinta River in Perak, Peninsular Malaysia, were used as case studies. The performance of the LSSVM model is compared with the conventional statistical autoregressive integrated moving average (ARIMA), the artificial neural network (ANN) and support vector machine (SVM) models using various statistical measures. The results of the comparison indicate that the LSSVM model is a useful tool and a promising new method for streamflow forecasting.

Editor D. Koutsoyiannis; Associate editor L. See

Citation Shabri, A. and Suhartono, 2012. Streamflow forecasting using least-squares support vector machines. Hydrological Sciences Journal, 57 (7), 1275–1293.     

An un-mixing model to study watershed erosion processes

An un-mixing model is formulated within a Bayesian Markov Chain Monte Carlo framework for use within land-use fingerprinting to study watershed erosion processes. The model has two new components: (1) An equation and erosion process parameter are used to weight tracer signatures from each erosion process within a land-use. (2) An extra tracer distribution and episodic erosion parameter are used to represent soil eroded throughout the sampling duration and thus include the episodic nature of erosion. To test specification of these new parameters, the un-mixing model is applied in the 15 km² Jerome Creek Watershed in the Palouse Region of Northwestern Idaho. Erosion processes include surface erosion upon mountain slopes due to logging in the forest land-use and rill/interrill erosion on cultivated slopes and headcut erosion in riparian floodplains of the agricultural land-use (winter wheat/peas rotation and hay pasture). Episodic erosion occurs for the event where the model is applied. A sensitivity analysis shows that the smallest Bayesian credible set results when the new parameters are specified using hydrologic data and process-based models. The un-mixing model predicts that 90% of the eroded-soil originated from the agricultural land-use and 10% originated from the forest land-use. A comparative study is performed that estimates 90.5% and 9.5% of eroded-soil originated from the agricultural and forest land-uses. Successful performance of the un-mixing model highlights future application as a standalone probabilistic tool to monitor watershed erosion processes that exhibit non-equilibrium conditions and provide calibration data for process-based watershed models.     

A new method for establishing hydrologic fidelity of snow depth measurements based on snowmelt–runoff hydrographs

For snowmelt-driven flood studies, snow water equivalent (SWE) is frequently estimated using snow depth data. Accurate measurements of snow depth are important in providing data for continuous hydrologic simulations of such watersheds. A new hydrologic fidelity metric is proposed in this study to evaluate the potential contribution of particular snow depth datasets to flow characteristics using observed data and hydrologic modeling using the Variable Infiltration Capacity (VIC) model. Data-based hydrologic fidelity of snow depth measurements is defined as a categorical skill score between the snow depth in the watershed and the hydrograph peak or volume at the watershed outlet. Similarly, model-based hydrologic fidelity is defined as a categorical skill score between the model-simulated snow depth and the model-simulated hydrograph peak or volume. The proposed framework is illustrated using the Pecatonica River watershed in the USA, indicating which sites have a higher hydrologic fidelity, which is preferred in hydrologic studies.     

Performance evaluation of interpolation methods for incorporating rain gauge measurements into NEXRAD precipitation data: a case study in the Upper Guadalupe River Basin

High spatial and temporal resolution of precipitation data is critical input for hydrological budget estimation and flash flood modelling. This study evaluated four methods [Bias Adjustment (BA), Simple Kriging with varying Local Means (SKlm), Kriging with External Drift (KED), and Regression Kriging (RK)] for their performances in incorporating gauge rainfall measurements into Next Generation Weather Radar (NEXRAD) multi‐sensor precipitation estimator (MPE; hourly and 4 × 4 km²). Measurements from a network of 50 gauges at the Upper Guadalupe River Basin, central Texas and MPE data for the year 2004 were used in the study. We used three evaluation coefficients percentage bias (PB), coefficient of determination (R²), and Nash–Sutcliffe efficiency (NSE) to examine the performance of the four methods for preserving regional‐ and local‐scale characteristics of observed precipitation data. The results show that the two Kriging‐based methods (SKlm and RK) are in general better than BA and KED and that the PB and NSE criteria are better than the R² criterion in assessing the performance of the four methods. It is also worth noting that the performance of one method at regional scale may be different from its performance at local scale. Critical evaluation of the performance of different methods at local or regional scale should be conducted according to the different purposes. The results obtained in this study are expected to contribute to the development of more accurate spatial rainfall products for hydrologic budget and flash flood modelling. Copyright © 2011 John Wiley & Sons, Ltd.     

Combining regression and spatial proximity for catchment model regionalization: a comparative study

Abstract

Spatial error regression is employed to regionalize the parameters of a rainfall–runoff model. The approach combines regression on physiographic watershed characteristics with a spatial proximity technique that describes the spatial dependence of model parameters. The methodology is tested for the monthly abcd model at a network of gauges in southeast United States and compared against simpler regression and spatial proximity approaches. Unlike other comparative regionalization studies that only evaluate the skill of regionalized streamflow predictions in ungauged catchments, this study also examines the fit between regionalized parameters and their optimal (i.e. calibrated) values. Interestingly, the spatial error model produces parameter estimates that better resemble the optimal parameters than either of the simpler methods, but the spatial proximity method still yields better hydrologic simulations. The analysis suggests that the superior streamflow predictions of spatial proximity result from its ability to better preserve correlations between compensatory hydrological parameters.

Editor D. Koutsoyiannis; Associate editor Y. Gyasi-Agyei     

An experiment on reservoir representation schemes to improve hydrologic prediction: coupling the national water model with the HEC-ResSim

ABSTRACT

This study experiments with reservoir representation schemes to improve the ability to model active water management in the National Water Model (NWM). For this purpose, we developed an integrated water management model, NWM-ResSim, by coupling the NWM with HEC-ResSim, and two reservoir representation schemes are tested: simulation of reservoir operations and retrieval of scheduled operations. The experiments focus on a pilot reservoir domain in the Russian River basin – Lake Mendocino, California – and its contributing watershed. The evaluation results suggest that the NWM-ResSim improves the simulation performance of reservoir outflow from this managed reservoir over the NWM default level pool routing scheme. The degree of this improvement depends on the suitability of the operation guidance; the reservoir operations simulation scheme could have acceptable errors for the purposes of water resources management, but not for flood operations. Results of the retrieval scheme of scheduled operations demonstrated better performance for sub-daily flood operations.     

Seasonal Short‐Term Prediction of Dissolved Oxygen in Rivers via Nature‐Inspired Algorithms

This study challenges the use of three nature‐inspired algorithms as learning frameworks of the adaptive‐neuro‐fuzzy inference system (ANFIS) machine learning model for short‐term modeling of dissolved oxygen (DO) concentrations. Particle swarm optimization (PSO), butterfly optimization algorithm (BOA), and biogeography‐based optimization (BBO) are employed for developing predictive ANFIS models using seasonal 15 min data collected from the Rock Creek River in Washington, DC. Four independent variables are used as model inputs including water temperature (T), river discharge (Q), specific conductance (SC), and pH. The Mallow's Cp and R² parameters are used for choosing the best input parameters for the models. The models are assessed by several statistics such as the coefficient of determination (R²), root‐mean‐square error (RMSE), Nash–Sutcliffe efficiency, mean absolute error, and the percent bias. The results indicate that the performance of all‐nature‐inspired algorithms is close to each other. However, based on the calculated RMSE, they enhance the accuracy of standard ANFIS in the spring, summer, fall, and winter around 13.79%, 15.94%, 6.25%, and 12.74%, respectively. Overall, the ANFIS‐PSO and ANFIS‐BOA provide slightly better results than the other ANFIS models.     

Hydrologic‐response simulations for the North Fork of Caspar Creek: second‐growth,clear‐cut,new‐growth,and cumulative watershed effect scenarios

This study demonstrates that comprehensive hydrologic‐response simulation can be a useful tool for studying cumulative watershed effects. The simulations reported here were conducted with the Integrated Hydrology Model (InHM). The location of the 473 ha study site is the North Fork of the Caspar Creek Experimental Watershed, near Fort Bragg, California. Existing information from a long‐term monitoring programme and new soil‐hydraulic property measurements made for this study were used to parameterize InHM. Long‐term continuous wet‐season simulations were conducted for the North Fork catchments and main stem for second‐growth, clear‐cut and new‐growth scenarios. The simulation results show that the increases and decreases, respectively, for throughfall and potential evapotranspiration related to clear‐cutting had quantifiable impacts on the simulated hydrologic response at both the catchment and watershed scales. Model performance was best for the new‐growth simulation scenarios. To improve upon the simulations reported here would require additional soil‐hydraulic property information from across the study area. Although principally focused on the integrated hydrologic response, the effort reported here demonstrates the potential for characterizing distributed responses with physics‐based simulation. The search for a comprehensive understanding of hydrologic response will require both data‐intensive discovery and concept‐development simulation, from both integrated and distributed perspectives. Copyright © 2013 John Wiley & Sons, Ltd.