Long-term Streamflow Forecasting Research Articles

Long-term streamflow forecasting in data-scarce regions: Insightful investigation for leveraging satellite-derived data, Informer architecture, and concurrent fine-tuning transfer learning

Accurate multistep forecasting of the long-term streamflow (Qflow) in poorly gauged basins is pivotal for sustainable water resource management and decision-making. The purpose of this study is to improve the performance of Qflow predictions in data-scarce regions. To achieve this, we employed geo-spatiotemporal satellite-derived datasets, advanced Informer architecture, and transfer learning methodologies. The deep-transfer-learning-based Informer model enables the perception of time dependencies and is specifically tailored to facilitate multistep-ahead Qflow prediction. Initial investigations identified correlations between historical Qflow records and geo-spatiotemporal satellite metrics, and this helped select grids that encapsulated the highest correlation between Qflow and climatic parameters. Subsequent assessments contrasted three main dimensionality reduction techniques—principal component analysis (PCA), t-distributed Stochastic Neighbor Embedding (t-SNE), and multi-dimensional scaling (MDS)—to craft deep-learning-amenable endogenous inputs. The introduction of the current fine-tuning transfer learning (CFTL) methodology, augmented with incremental learning, which was evaluated against conventional transfer learning (CTL) across two distinct target domains with varying similarities to the source domain, was central to this study. A holistic comparison involving a range of univariate multistep-ahead prediction models including DNN, CNN, LSTM, AR-LSTM, and Transformer, was also undertaken. In addition, multivariate models, notably CFTL and CTL, were set against the two former models under different data availability scenarios for each target domain. The findings revealed that the MDS-Informer model significantly outperformed the t-SNE-Informer and PCA-Informer models, showing an improvement of approximately 23 % and 40 % in terms of MAE, respectively. Moreover, the results confirmed that the CFTL approach effectively enhanced long-term Qflow prediction accuracy compared to conventional transfer learning (CTL) and the limited data scenario (LD) without TL by approximately 30 and 47 % improvement in terms of MAE, respectively. This study demonstrates that a target domain possessing a higher similarity index with the source domain is more likely to facilitate the transfer of positive noise, thereby enhancing the predictability of the model. This phenomenon underscores the importance of domain similarity in the efficacy of TL approaches.

Two Strategies for Avoiding Overfitting Long-Term Forecasting Models: Downsampling Predictor Fields and Shrinking Coefficients

Long-term hydrological forecasting based on sea surface temperature (SST) fields faces the large p and small n problem, i.e., too many potential predictors and a limited number of samples. Considering the selection of predictors will also enhance the complexity of models and lead to overfitting, in this study, two strategies are used for building forecasting models for long-term streamflow forecasting. The first strategy is to downsample the SST field and optimize its spatial resolution; the second is to shrink model coefficients based on L1 regularization. We build models based on the downsampled SST fields with different spatial resolutions. It is found that the model based on a proper spatial resolution always performs better than the model based on the raw SST field. This result suggests that it is better to treat the spatial resolution of the predictor field as a hyperparameter, which is similar to hyperparameters for controlling the complexity of many machine learning models. For applying the second strategy, L1 norm regularization models, including (1) least absolute selection and shrinkage operator (LASSO), (2) relaxed LASSO, and (3) two-step approach of LASSO and ordinary least squares regression (LASSO+OLS) are explored. We have found that the relaxed LASSO model always performs better than the ordinary LASSO, indicating that relaxed LASSO is a better shrinkage approach. Furthermore, the skills of the presented models are compared with the stepwise regression, and the lower skill of the stepwise regression based on SST fields with a high spatial resolution suggests that one should not select predictors based on fields with high resolutions considering the limited number of samples.

A Hybrid Physics–AI Model to Improve Hydrological Forecasts

Abstract The National Oceanic and Atmospheric Administration has developed a very high-resolution streamflow forecast using National Water Model (NWM) for 2.7 million stream locations in the United States. However, considerable challenges exist for quantifying uncertainty at ungauged locations and forecast reliability. A data science approach is presented to address the challenge. The long-range daily streamflow forecasts are analyzed from December 2018 to August 2021 for Alabama and Georgia. The forecast is evaluated at 389 observed USGS stream gauging locations using standard deterministic metrics. Next, the forecast errors are grouped using watersheds’ biophysical characteristics, including drainage area, land use, soil type, and topographic index. The NWM forecasts are more skillful for larger and forested watersheds than smaller and urban watersheds. The NWM forecast considerably overestimates the streamflow in the urban watersheds. The classification and regression tree analysis confirm the dependency of the forecast errors on the biophysical characteristics. A densely connected neural network model consisting of six layers [deep learning (DL)] is developed using biophysical characteristics, NWM forecast as inputs, and the forecast errors as outputs. The DL model successfully learns location invariant transferrable knowledge from the domain trained in the gauged locations and applies the learned model to estimate forecast errors at the ungauged locations. A temporal and spatial split of the gauged data shows that the probability of capturing the observations in the forecast range improved significantly in the hybrid NWM-DL model (82% ± 3%) than in the NWM-only forecast (21% ± 1%). A trade-off between overly constrained NWM forecast and increased forecast uncertainty range in the DL model is noted. Significance Statement A hybrid biophysical–artificial intelligence (physics–AI) model is developed from the first principle to estimate streamflow forecast errors at ungauged locations, improving the forecast’s reliability. The first principle refers to identifying the need for the hybrid physics–AI model, determining physically interpretable and machine identifiable model inputs, followed by the deep learning (DL) model development and its evaluations, and finally, a biophysical interpretation of the hybrid model. A very high-resolution National Water Model (NWM) forecast, developed by the National Oceanic and Atmospheric Administration, serves as the biophysical component of the hybrid model. Out of 2.7 million daily forecasts, less than 1% of the forecasts can be verified using the traditional hydrological method of comparing the forecast with the observations, motivating the need for the AI technique to improve forecast reliability at millions of ungauged locations. An exploratory analysis followed by the classification and regression tree analysis successfully determines the dependency of the forecast errors on the biophysical attributes, which along with the NWM forecast, are used for the DL model development. The hybrid model is evaluated in a subtropical humid climate of Alabama and Georgia in the United States. Long-term streamflow forecasts from zero-day lead to 30-day lead forecasts are archived and analyzed for 979 days (December 2018–August 2021) and 389 USGS gauging stations. The forecast reliability is assessed as the probability of capturing the observations in its ensemble range. As a result, the forecast reliability increased from 21% (±1%) in the NWM only forecasts to 82% (±3%) in the hybrid physics–AI model.

Study on the evolution law of performance of mid- to long-term streamflow forecasting based on data-driven models

Mid- to long-term streamflow forecasting is of great significance for the sustainable utilization and management of water resources. Currently, data-driven models are mainly used for mid- to long-term streamflow forecasting. However, the evolution law of prediction models’ performance with increasing lead times has not been fully revealed due to too much attention on proposing a novel method to improve prediction accuracy. The main objective of this study was to explore the evolution law of the performance of ten-day streamflow forecasting models with the extension of lead time, including multiple linear regression, support Vector Regression, feedforward neural network, and least-squares Boosting Decision Tree. The upper reaches of the Yangtze River in China were considered as the case study region. The results show that: (1) with the extension of the lead time, the performance of data-driven models first decreases and then gradually levels off; (2) the inclusion of teleconnection climatic factors can improve the performance of each model; and (3) the randomness in streamflow can weaken the prediction ability of the models. These findings can promote improvements in mid- to long-term streamflow forecasting accuracy.

Featured Collection Introduction: Severe Sustained Drought Revisited: Managing the Colorado River System in Times of Water Shortage 25 Years Later — Part I

Featured Collection Introduction: Severe Sustained Drought Revisited: Managing the Colorado River System in Times of Water Shortage 25 Years Later — Part I

Assessing different roles of baseflow and surface runoff for long-term streamflow forecasting in southeastern China

ABSTRACT Accurate long-term streamflow forecast is essential to alleviate and solve the water security problems related to flood and drought disaster warnings. In this study, a new strategy for forecasting monthly streamflow is proposed and four scenarios are designed for the evaluation of different roles of baseflow and surface runoff on performances of long-term streamflow forecasting. The developed models are evaluated at multiple streamflow sites located in the Zhejiang Province of China. The results show that artificial intelligence (AI)-based models with two predictor variables (i.e. baseflow and surface runoff) performed better than that with a single predictor (streamflow) for all the months in a year, and the prediction accuracy of annual peak and monthly streamflow values is improved. Based on the comprehensive evaluations of all the models, the baseflow and surface runoff values are recommended as inputs to AI-based models for an improved prediction accuracy of streamflows.

A weights combined model for middle and long-term streamflow forecasts and its value to hydropower maximization

Stochastic nature of streamflow poses significant challenges in attaining a reliable forecasting model. In general, variational mode decomposition (VMD) can improve the forecast performance but easily expose the sub-signals to boundary effect. Furthermore, one model is not able to adapt all properties of different sub-series. Accordingly, we have two aims in this study. One is to propose an adaptive weight combined forecasting model to improve the middle and long-term streamflow forecast skill. It adapts the boundary effect in such a way that its inputs come from decomposition during calibration sets, and outputs are extracted from decomposition during the entire series. The other one is to link system performance improvement, i.e., the forecast skill, to the forecast value to address the gap in methodologies appropriate for data-limited locations (only hydrological time series collected). Four artificial intelligence-based models coupled with adaptive appendant and parameter optimization are developed as hybrid adaptive (HA) for forecasting each sub-signal decomposed by VMD in the proposed forecast model. The multi-objective grey wolf optimization (MOGWO) algorithm is then employed to combine individual forecasts based on performance-based weighting strategies and provide the Pareto-optimal ensemble forecasts. The proposed model is applied to forecast streamflow 1 to 6 months ahead of two stations in the Yellow River, China, and the results show that the ensemble forecasts can increase the values of Nash-Sutcliffe efficiency coefficient by 0.10 ~ 7.96% and reducing the values of root mean squared error by 1.08 ~ 32.11% compared to the HA model. The relationship between the forecast skill and its value can be strongly affected by decision-makers priorities, but the relative improvement in hydropower generation obtained by the compromised forecasts going from 0.02% to 3.39% indicates that improved forecasts are potentially valuable for informing strategic decisions.

Prediction of Yangtze River streamflow based on deep learning neural network with El Ni\xf1o\u2013Southern Oscillation

Accurate long-term streamflow and flood forecasting have always been an important research direction in hydrology research. Nowadays, climate change, floods, and other anomalies occurring more and more frequently and bringing great losses to society. The prediction of streamflow, especially flood prediction, is important for disaster prevention. Current hydrological models based on physical mechanisms can give accurate predictions of streamflow, but the effective prediction period is only about 1 month in advance, which is too short for decision making. The artificial neural network (ANN) has great potential for predicting runoff and is not only good at handling non-linear data but can also make long-period forecasts. However, most of ANN models are unstable in their predictions when faced with raw flow data, and have excessive errors in predicting extreme flows. Previous studies have shown a link between the El Niño–Southern Oscillation (ENSO) and the streamflow of the Yangtze River. In this paper, we use ENSO and the monthly streamflow data of the Yangtze River from 1952 to 2018 to predict the monthly streamflow of the Yangtze River in two extreme flood years and a small flood year by using deep neural networks. In this paper, three deep neural network frameworks are used: stacked long short-term memory, Conv long short-term memory encoder–decoder long short-term memory and Conv long short-term memory encoder–decoder gate recurrent unit. The results show that the use of ConvLSTM improves the stability of the model and increases the accuracy of the flood prediction. Besides, the introduction of ENSO to the experimental data resulted in a more accurate prediction of the time of the occurrence of flood peaks and flood flows. Furthermore, the best results were obtained on the convolutional long short-term memory + encoder–decoder gate recurrent unit model.

Middle- and Long-Term Streamflow Forecasting and Uncertainty Analysis Using Lasso-DBN-Bootstrap Model

Middle-term and long-term streamflow forecasting is of great significance for water resources planning and management, cascade reservoirs optimal operation, agriculture and hydro-power generation. In this work, a framework was proposed which integrates least absolute shrinkage and selection operator (lasso), DBN and bootstrap to improve the performance and the stability of streamflow forecasting with the lead-time of one month. Lasso helps to screen the appropriate predictors for the DBN model, and the DBN model simulates the complex relationship between the selection predictors and streamflow, and then bootstrap with the DBN model contributes to evaluate the uncertainty. The Three-River Headwaters Region (TRHR) was taken as a case study. The results indicated that lasso-DBN-bootstrap model produced significantly more accurate forecasting results than the other three models and provides reliable information on the forecasting uncertainty, which will be valuable for water resources management and planning.

The Value of Long-Term Streamflow Forecasts in Adaptive Reservoir Operation: The Case of the High Aswan Dam in the Transboundary Nile River Basin

AbstractTransboundary river basins are experiencing extensive dam development that challenges future water management, especially for downstream nations. Thus, adapting the operation of existing reservoirs is indispensable to cope with alterations in flow regime. We proposed a Forecast-Based Adaptive Reservoir Operation (FARO) framework to evaluate the use of long-term climate forecasts in improving real-time reservoir operations. The FARO approach was applied to the High Aswan Dam (HAD) in the Nile River basin. Monthly precipitation and temperature forecasts at up to 12 months of lead time are used from a suite of eight North American Multi-Model Ensemble (NMME) models. The value of NMME-based forecasts to reservoir operations was compared with perfect and climatology-based forecasts over an optimization horizon of 10 years from 1993 to 2002. Our results indicated that the forecast horizon for HAD operation ranges between 5 and 12 months lead time at low- and high-demand scenarios, respectively, beyond which the forecast information no longer improves the release decision. The forecast value to HAD operation is more pronounced in the months following the flooding season (October–December). During these months, the skill of streamflow forecasts using NMME forcings outperforms the climatology-based forecasts. When considering the operation of upstream Grand Ethiopian Renaissance Dam (GERD), using streamflow forecasts minimally helps to maintain current target objectives of HAD operation and therefore result in higher operation costs as opposed to current conditions without GERD. Our study underlined the importance of deriving a new adaptive operating policy for HAD to improve the value of available forecasts while considering GERD filling and operation phases.

Evaluation of a stochastic weather generator for long-term ensemble streamflow forecasts

ABSTRACT 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.

Coupling large-scale climate indices with a stochastic weather generator to improve long-term streamflow forecasts in a Canadian watershed

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.

Long-term streamflow forecasting for the Cascade Reservoir System of Han River using SWAT with CFS output

Abstract Due to the large uncertainties of long-term precipitation prediction and reservoir operation, it is difficult to forecast long-term streamflow for large basins with cascade reservoirs. In this paper, a framework coupling the original Climate Forecasting System (CFS) precipitation with the Soil and Water Assessment Tool (SWAT) was proposed to forecast the nine-month streamflow for the Cascade Reservoir System of Han River (CRSHR) including Shiquan, Ankang and Danjiangkou reservoirs. First, CFS precipitation was tested against the observation and post-processed through two machine learning algorithms, random forest and support vector regression. Results showed the correlation coefficients between the monthly areal CFS precipitation (post-processed) and observation were 0.91–0.96, confirming that CFS precipitation post-processing using machine learning was not affected by the extended forecast period. Additionally, two precipitation spatio-temporal distribution models, original CFS and similar historical observation, were adopted to disaggregate the processed monthly areal CFS precipitation to daily subbasin-scale precipitation. Based on the reservoir restoring flow, the regional SWAT was calibrated for CRSHR. The Nash–Sutcliffe efficiencies for three reservoirs flow simulation were 0.86, 0.88 and 0.84, respectively, meeting the accuracy requirement. The experimental forecast showed that for three reservoirs, long-term streamflow forecast with similar historical observed distribution was more accurate than that with original CFS.

Identification of Hydrologic Models, Optimized Parameters, and Rainfall Inputs Consistent with In Situ Streamflow and Rainfall and Remotely Sensed Soil Moisture

Abstract An increased understanding of the uncertainties present in rainfall time series can lead to improved confidence in both short- and long-term streamflow forecasts. This study presents an analysis that considers errors arising from model input data, model structure, model parameters, and model states with the objective of finding a self-consistent set that includes hydrological models, model parameters, streamflow, remotely sensed (RS) soil moisture (SM), and rainfall. This methodology can be used by hydrologists to aid model and satellite selection. Taking advantage of model input data reduction and model inversion techniques, this study uses a previously developed methodology to estimate areal rainfall time series for the study catchment of Warwick, Australia, for multiple rainfall–runoff models. RS SM observations from the Soil Moisture Ocean Salinity (SMOS) and Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E) satellites were assimilated into three different rainfall–runoff models using an ensemble Kalman filter (EnKF). Innovations resulting from the observed and predicted SM were analyzed for Gaussianity. The findings demonstrate that consistency between hydrological models, model parameters, streamflow, RS SM, and rainfall can be found. Joint estimation of rainfall time series and model parameters consistently improved streamflow simulations. For all models rainfall estimates are less than the observed rainfall, and rainfall estimates obtained using the Sacramento Soil Moisture Accounting (SAC-SMA) model are the most consistent with gauge-based observations. The SAC-SMA model simulates streamflow that is most consistent with observations. EnKF innovations obtained when SMOS RS SM observations were assimilated into the SAC-SMA model demonstrate consistency between SM products.

Hybrid Models Combining EMD/EEMD and ARIMA for Long-Term Streamflow Forecasting

Long-term streamflow forecast is of great significance for water resource application and management. However, accurate monthly streamflow forecasting is challenging due to its non-stationarity and uncertainty. Time series analysis methods have been proved to perform well in stationary time series forecasting, which can be derived from decomposition of the non-stationary sequence. As common decomposition methods in time domain, Empirical Mode Decomposition (EMD) and Ensemble Empirical Mode Decomposition (EEMD) are selected to decompose the components with different time-scale characteristics in the original hydrological time series in this study. The derived components are proved to be stationary by the stationarity test. Thus, Autoregressive Integrated Moving Average (ARIMA) model, a simple and effective time series analysis method, is used to forecast the components. A hybrid EMD/EEMD-ARIMA model is proposed in this study for long-term streamflow forecasting, which is applied to the upper stream of the Yellow River. The original daily streamflow time series of six years at the Tangnaihai station are firstly decomposed by EMD/EEMD into several stationary or simple non-stationary sub-series to explore detailed data information with different time scales. ARIMA models with appropriate parameters are then established for each subsequence to forecast the stream flow of the next year. Predicted ten-day and monthly stream flow is finally obtained combing the predictions of all the components. The EMD-ARIMA hybrid model performs best in forecasting high and moderate value of streamflow and fits best with the observation compared with EEMD-ARIMA and ARIMA models. The results not only verify the effectiveness of the proposed hybrid EMD/EEMD-ARIMA model in exploiting comprehensive information to improve the prediction but also indicate that the EMD-ARIMA model with end points disposal performs the best and can be used for long-term hydrological forecasting.

Evaluation of ocean-atmospheric indices as predictors for summer streamflow of the Yangtze River based on ROC analysis

Antecedent anomalies of sea surface temperature and atmospheric circulation are important signals for making long-term streamflow forecasts. In this study, four groups of ocean-atmospheric indices, i.e, El Nino Southern Oscillation (ENSO), the Northern Hemisphere atmospheric circulation, the Southern Hemisphere atmospheric circulation (SAC), and the Western Pacific and Indian Ocean SST (WPI), are evaluated for forecasting summer streamflow of the Yangtze River. The gradient boosting regression tree (GBRT) is used to forecast streamflow based on each group of indices. The score based on receiver operating characteristics (ROC) curves, i.e., area under the ROC curve (AUC), is used to evaluate skills of models for identifying the high category and the low category of summer streamflow. It is found that the ENSO group and the SAC group show higher AUC values. Furthermore, both AUC values of GBRT models and individual indices show that the low flow years are easier to be identified than the high flow years. The result of this study highlights the skill from the Southern Hemisphere circulation systems for forecasting summer streamflow of the Yangtze River. Results of relative influences of predictors in GBRT models and AUC of individual indices indicate some key ocean-atmospheric indices, such as the Multivariate ENSO Index and the 500-hPa height of the east of Australia.

Evaluation of Seasonal Streamflow Forecasting

Long-term streamflow forecasts are essential for optimal management of water resources for various demands, including irrigation, fisheries management, hydropower production and flood warning. In this paper, a probabilistic forecast framework based on Ensemble Streamflow Prediction (ESP) technique is presented, with the basic assumption that future weather patterns will reflect those experienced historically. Hence, past forcing data (input to hydrological model) can be used with the current initial condition of a catchment to generate an ensemble of flow predictions. The present study employs the ESP-based approach using the TopNet hydrological model. The objective of this present paper is to evaluate and assess the uncertainty due to initial condition of the catchments and forcing (meteorological input to the model) data for (ESP) based streamflow forecasting using the TopNet hydrological model in New Zealand catchments. An ensemble of streamflow predictions which provide probabilistic hydrological forecasts, reflecting the intrinsic uncertainty in climate, with lead time up to three months is presented for the four catchments on New Zealand’s South Island. Verification of the forecast over the period 2000-2010 indicates a Ranked Probability Skill Score of 23% to 69% (over climatology) across the four catchments. In general, improvement in ESP forecasting skill over climatology is greatest in summer for all catchments studied. The major uncertainty associated with ESP forecast is combination of uncertainty due to initial state and climate forcing. The analysis indicates that the sensitivity of flow forecast to initial condition uncertainty depends on the hydrological regime experienced by the basin during the forecast period. On average, the relative importance of initial condition is greatest within two weeks to months of the start of the simulation for all catchment and all season. After this time period uncertainty in forecast is mainly due to uncertainty in forcing data. Finding of this study can be valuable tool for water resource managements.

Long-term streamflow forecasting using SWAT through the integration of the random forests precipitation generator: case study of Danjiangkou Reservoir

Abstract Long-term streamflow forecasting is of great significance to the optimal management of water resources. However, the forecast lead time of long-term streamflow forecasting is relatively long and the forecasted precipitation within the forecast lead time has inherent uncertainty, so long-term streamflow forecasting has major challenges. In this paper, a hybrid forecasting model is developed to improve accuracy of long-term streamflow forecasting by combining random forests (RF) and the Soil and Water Assessment Tool (SWAT). The RF model is used to forecast monthly precipitation which is further downscaled to a daily dataset according to the hydrological similarity principle for use in the SWAT model of the Danjiangkou Reservoir basin, China. Performance of this hybrid model is compared to that of seasonal autoregressive (SAR (P)) model. Results show the RF precipitation generator yields accurate predictions at the monthly scale and the hybrid model produces acceptable streamflow series in long-term forecasting cases. In addition, the comparison shows that in the Danjiangkou Reservoir basin, the hybrid model performs better than the SAR (P) model, with average Nash–Sutcliffe efficiency (NSE) values of 0.94 and 0.51, which is better when it is closer to 1. This study provides a method of improving accuracy of long-term streamflow forecasting.

Informing the operations of water reservoirs over multiple temporal scales by direct use of hydro-meteorological data

Water reservoir systems may become more adaptive and reliable to external changes by enlarging the information sets used in their operations. Models and forecasts of future hydro-climatic and socio-economic conditions are traditionally used for this purpose. Nevertheless, the identification of skillful forecasts and models might be highly critical when the system comprises several processes with inconsistent dynamics (fast and slow) and disparate levels of predictability. In these contexts, the direct use of observational data, describing the current conditions of the water system, may represent a practicable and zero-cost alternative. This paper contrasts the relative contribution of state observations and perfect forecasts of future water availability in improving multipurpose water reservoirs operation over short- and long-term temporal scales. The approach is demonstrated on the snow-dominated Lake Como system, operated for flood control and water supply. The Information Selection Assessment (ISA) framework is adopted to retrieve the most relevant information to be used for conditioning the operations. By explicitly distinguishing between observational dataset and future forecasts, we quantify the relative contribution of current water system state estimates and perfect streamflow forecasts in improving the lake regulation with respect to both flood control and water supply. Results show that using the available observational data capturing slow dynamic processes, particularly the snow melting process, produces a 10% improvement in the system performance. This latter represents the lower bound of the potential improvement, which may increase to the upper limit of 40% in case skillful (perfect) long-term streamflow forecasts are used.

Long-Term Streamflow Forecasting Based on Relevance Vector Machine Model

Long-term streamflow forecasting is crucial to reservoir scheduling and water resources management. However, due to the complexity of internally physical mechanisms in streamflow process and the influence of many random factors, long-term streamflow forecasting is a difficult issue. In the article, we mainly investigated the ability of the Relevance Vector Machine (RVM) model and its applicability for long-term streamflow forecasting. We chose the Dahuofang (DHF) Reservoir in Northern China and the Danjiangkou (DJK) Reservoir in Central China as the study sites, and selected the 500 hpa geopotential height in the northern hemisphere and the sea surface temperatures in the North Pacific as the predictor factors of the RVM model and the Support Vector Machine (SVM) model, and then conducted annual streamflow forecasting. Results indicate that forecasting results in the DHF Reservoir is much better than that in the DJK Reservoir when using SVM, because streamflow process in the latter basin has a magnitude bigger than 1000 m3/s. Comparatively, accurate forecasting results in both the two basins can be gotten using the RVM model, with the Nash Sutcliffe efficiency coefficient bigger than 0.7, and they are much better than those gotten from the SVM model. As a result, the RVM model can be an effective approach for long-term streamflow forecasting, and it also has a wide applicability for the streamflow process with a discharge magnitude from dozen to thousand cubic meter per second.