A Comparative Study of Autoregressive,Autoregressive Moving Average,Gene Expression Programming and Bayesian Networks for Estimating Monthly Streamflow

In the recent years, artificial intelligence techniques have attracted much attention in hydrological studies, while time series models are rarely used in this field. The present study evaluates the performance of artificial intelligence techniques including gene expression programming (GEP), Bayesian networks (BN), as well as time series models, namely autoregressive (AR) and autoregressive moving average (ARMA) for estimation of monthly streamflow. In addition, simple multiple linear regression (MLR) was also used. To fulfill this objective, the monthly streamflow data of Ponel and Toolelat stations located on Shafarood and Polrood Rivers, respectively in Northern Iran were used for the period of October 1964 to September 2014. In order to investigate the models’ accuracy, root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R²) were employed as the error statistics. The obtained results demonstrated that the single AR and ARMA time series models had better performance in comparison with the single GEP, BN and MLR methods. Furthermore, in this study, six hybrid models known as GEP-AR, GEP-ARMA, BN-AR, BN-ARMA, MLR-AR and MLR-ARMA were developed to enhance the estimation accuracy of the monthly streamflow. It was concluded that the developed hybrid models were more accurate than the corresponding single artificial intelligence and time series models. The obtained results confirmed that the integration of time series models and artificial intelligence techniques could be of use to improve the accuracy of single models in modeling purposes related to the hydrological studies.     

Uncertainty Analysis of Climate Change Impacts on Flood Frequency by Using Hybrid Machine Learning Methods

In the present study, for the first time, a new framework is used by combining metaheuristic algorithms, decomposition and machine learning for flood frequency analysis under climate-change conditions and application of HadCM3 (A2 and B2 scenarios), CGCM3 (A2 and A1B scenarios) and CanESM2 (RCP2.6, RCP4.5 and RCP8.5 scenarios) in global climate models (GCM). In the proposed framework, Multivariate Adaptive Regression Splines (MARS) and M5 Model tree are used for classification of precipitation (wet and dry days), whale optimization algorithm (WOA) is considered for training least square support vector machine (LSSVM), wavelet transform (WT) is used for decomposition of precipitation and temperature, LSSVM-WOA, LSSVM, K nearest neighbor (KNN) and artificial neural network (ANN) are performed for downscaling precipitation and temperature, and discharge is simulated under present period (1972–2000), near future (2020–2040) and far future (2070–2100). Log normal distribution is used for flood frequency analysis. Furthermore, analysis of variance (ANOVA) and fuzzy method are employed for uncertainty analysis. Karun3 Basin, in southwest of Iran, is considered as a case study. Results indicated that MARS performed better than M5 model tree. In downscaling, ANN and LSSVM_WOA slightly outperformed other machine learning algorithms. Results of simulating the discharge showed superiority of LSSVM_WOA_WT algorithm (Nash-Sutcliffe efficiency (NSE)?=?0.911). Results of flood frequency analysis revealed that 200-year discharge decreases for all scenarios, except CanESM2 RCP2.6 scenario, in the near future. In the near and far future periods, it is obvious from ANOVA uncertainty analysis that hydrological models are one of the most important sources of uncertainty. Based on the fuzzy uncertainty analysis, HadCM3 model has lower uncertainty in higher return periods (up to 60% lower than other models in 1000-year return period).

     

Emotional ANN (EANN) and Wavelet-ANN (WANN) Approaches for Markovian and Seasonal Based Modeling of Rainfall-Runoff Process

The hydrological time series have three principle components (autoregressive, seasonality and trend) and the performance of the models is strongly related to the nature of these components. The current research examines the accuracy of two Artificial Neural Network (ANN) based approaches for rainfall-runoff (r-r) modeling of two catchments with different geomorphological conditions at monthly and daily time scales. The techniques proposed here are hybrid wavelet-ANN (WANN) model, as a multi-resolution forecasting tool and Emotional Artificial Neural Network (EANN) (a new generation of ANN based models) which serves artificial emotional factors as well as classic bias and weights parameters. The obtained results for monthly modeling show that WANN could perform better than the simple feed forward neural network (FFNN) model up to 40% and 35% in terms of verification and training efficiency criteria due to significant seasonality involved in the monthly time series of the process. On the other hand, the obtained results for daily modeling via FFNN and EANN, both as Markovian models, indicates the superiority of EANN over FFNN because of EANN capability to better learning of extraordinary and extreme conditions of the process in the training phase.     

Pipe Break Rate Assessment While Considering Physical and Operational Factors: A Methodology based on Global Positioning System and Data-Driven Techniques

An accurate prediction of pipes failure rate plays a substantial role in the management of Water Distribution Networks (WDNs). In this study, a field study was conducted to register pipes break and relevant causes in the WDN of Yazd City, Iran. In this way, 851 water pipes were incepted and localized by the Global Positioning System (GPS) apparatus. Then, 1033 failure cases were reported in the eight zones of understudy WDN during March-December 2014. Pipes break rate (BR_P) was calculated using the depth of pipe installation (h_P), number of failures (N_P), the pressure of water pipes in operation (P), and age of pipe (A_P). After completing a pipe break database, robust Artificial Intelligence models, namely Multivariate Adaptive Regression Spline (MARS), Gene-Expression Programming (GEP), and M5 Model Tree were employed to extract precise formulation for the pipes break rate estimation. Results of the proposed relationships demonstrated that the MARS model with Coefficient of Correlation (R) of 0.981 and Root Mean Square Error (RMSE) of 0.544 provided more satisfying efficiency than the M5 model (R?=?0.888 and RMSE?=?1.096). Furthermore, statistical results indicated that MARS and GEP models had comparatively at the same accuracy level. Explicit equations by Artificial Intelligence (AI) models were satisfactorily comparable with those obtained by literature review in terms of various conditions: physical, operational, and environmental factors and complexity of AI models. Through a probabilistic framework for the pipes break rate, the results of first-order reliability analysis indicated that the MARS technique had a highly satisfying performance when MARS-extracted-equation was assigned as a limit state function.

     

Comparison of performance of statistical models in forecasting monthly streamflow of Kizil River,China

This paper presents the application of autoregressive integrated moving average(ARIMA),seasonal ARIMA(SARIMA),and Jordan-Elman artificial neural networks(ANN)models in forecasting the monthly streamflow of the Kizil River in Xinjiang,China.Two different types of monthly streamflow data(original and deseasonalized data)were used to develop time series and Jordan-Elman ANN models using previous flow conditions as predictors.The one-month-ahead forecasting performances of all models for the testing period(1998-2005)were compared using the average monthly flow data from the Kalabeili gaging station on the Kizil River.The Jordan-Elman ANN models,using previous flow conditions as inputs,resulted in no significant improvement over time series models in one-month-ahead forecasting.The results suggest that the simple time series models(ARIMA and SARIMA)can be used in one-month-ahead streamflow forecasting at the study site with a simple and explicit model structure and a model performance similar to the Jordan-Elman ANN models.     

Water Consumption Prediction of Istanbul City by Using Fuzzy Logic Approach

This paper presents a Takagi Sugeno (TS) fuzzy method for predicting future monthly water consumption values from three antecedent water consumption amounts, which are considered as independent variables. Mean square error (MSE) values for different model configurations are obtained, and the most effective model is selected. It is expected that this model will be more extensively used than Markov or ARIMA (AutoRegressive Integrated Moving Average) models commonly available for stochastic modeling and predictions. The TS fuzzy model does not have restrictive assumptions such as the stationarity and ergodicity which are primary requirements for the stochastic modeling. The TS fuzzy model is applied to monthly water consumption fluctuations of Istanbul city in Turkey. In the prediction procedure only lag one is considered. It is observed that the TS fuzzy model preserves the statistical properties. This model also helps to make predictions with less than 10% relative error.     

Performance Assessment of the Linear,Nonlinear and Nonparametric Data Driven Models in River Flow Forecasting

In recent years, the data-driven modeling techniques have gained more attention in hydrology and water resources studies. River runoff estimation and forecasting are one of the research fields that these techniques have several applications in them. In the current study, four common data-driven modeling techniques including multiple linear regression, K-nearest neighbors, artificial neural networks and adaptive neuro-fuzzy inference systems have been used to form runoff forecasting models and then their results have been evaluated. Also, effects of using of some different scenarios for selecting predictor variables have been studied. It is evident from the results that using flow data of one or two month ago in the predictor variables dataset can improve accuracy of results. In addition, comparison of general performances of the modeling techniques shows superiority of results of KNN models among the studied models. Among selected models of the different techniques, the selected KNN model presented best performance with a linear correlation coefficient equal to 0.84 between observed flow data and predicted values and a RMSE equal to 2.64.     

Evaluation of IMERG,TMPA, ERA5, and CPC precipitation products over mainland China: Spatiotemporal patterns and extremes

A comprehensive assessment of representative satellite-retrieved (Integrated Multi-satellite Retrievals for Global Precipitation Measurement (IMERG) and Tropical Rainfall Measuring Mission Multi-satellite Precipitation Analysis (TMPA)), reanalysis-based (fifth generation of atmospheric reanalysis by the European Centre for Medium Range Weather Forecasts (ERA5)), and gauge-estimated (Climate Prediction Center (CPC)) precipitation products was conducted using the data from 807 meteorological stations across mainland China from 2001 to 2017. Error statistical metrics, precipitation distribution functions, and extreme precipitation indices were used to evaluate the quality of the four precipitation products in terms of multi-timescale accuracy and extreme precipitation estimation. When the timescale increased from daily to seasonal scales, the accuracy of the four precipitation products first increased and then decreased, and all products performed best on the monthly timescale. Their accuracy ranking in descending order was CPC, IMERG, TMPA, and ERA5 on the daily timescale and IMERG, CPC, TMPA, and ERA5 on the monthly and seasonal timescales. IMERG was generally superior to its predecessor TMPA on the three timescales. ERA5 exhibited large statistical errors. CPC provided stable estimated values. For extreme precipitation estimation, the quality of IMERG was relatively consistent with that of TMPA in terms of precipitation distribution and extreme metrics, and IMERG exhibited a significant advantage in estimating moderate and heavy precipitation. In contrast, ERA5 and CPC exhibited poor performance with large systematic underestimation biases. The findings of this study provide insight into the performance of the latest IMERG product compared with the widely used TMPA, ERA5, and CPC datasets, and points to possible directions for improvement of multi-source precipitation data fusion algorithms in order to better serve hydrological applications.     

Enhancing Long-Term Streamflow Forecasting and Predicting using Periodicity Data Component: Application of Artificial Intelligence

Streamflow forecasting and predicting are significant concern for several applications of water resources and management including flood management, determination of river water potentials, environmental flow analysis, and agriculture and hydro-power generation. Forecasting and predicting of monthly streamflows are investigated by using three heuristic regression techniques, least square support vector regression (LSSVR), multivariate adaptive regression splines (MARS) and M5 Model Tree (M5-Tree). Data from four different stations, Besiri and Malabadi located in Turkey, Hit and Baghdad located in Iraq, are used in the analysis. Cross validation method is employed in the applications. In the first stage of the study, the heuristic regression models are compared with each other and multiple linear regression (MLR) in forecasting one month ahead streamflow of each station, individually. In the second stage, the models are evaluated and compared in predicting streamflow of one station using data of nearby station. The research investigated also the influence of the periodicity component (month number of the year) as an external sub-set in modeling long-term streamflow. In both stages, the comparison results indicate that the LSSVR model generally performs superior to the MARS, M5-Tree and MLR models. In addition, it is seen that adding periodicity as input to the models significantly increase their accuracy in forecasting and predicting monthly streamflows in both stages of the study.     

A Hybrid VMD-SVM Model for Practical Streamflow Prediction Using an Innovative Input Selection Framework

Some previous studies have proved that prediction models using traditional overall decomposition sampling (ODS) strategy are unreasonable because the subseries obtained by the ODS strategy contain future information to be predicted. It is, therefore, necessary to put forward a new sampling strategy to fix this defect and also to improve the accuracy and reliability of decomposition-based models. In this paper, a stepwise decomposition sampling (SDS) strategy according to the practical prediction process is introduced. Moreover, an innovative input selection framework is proposed to build a strong decomposition-based monthly streamflow prediction model, in which sunspots and atmospheric circulation anomaly factors are employed as candidate input variables to enhance the prediction accuracy of monthly streamflow in addition to regular inputs such as precipitation and evaporation. Meanwhile, the partial correlation algorithm is employed to select optimal input variables from candidate input variables including precipitation, evaporation, sunspots, and atmospheric circulation anomaly factors. Four basins of the U.S. MOPEX project with various climate characteristics were selected as a case study. Results indicate that: (1) adding teleconnection factors into candidate input variables helps enhance the prediction accuracy of the support vector machine (SVM) model in predicting streamflow; (2) the innovative input selection framework helps to improve the prediction capacity of models whose candidate input variables interact with each other compared with traditional selection strategy; (3) the SDS strategy can effectively prevent future information from being included into input variables, which is an appropriate substitute of the ODS strategy in developing prediction models; (4) as for monthly streamflow, the hybrid variable model decomposition-support vector machine (VMD-SVM) models, using an innovative input selection framework and the SDS strategy, perform better than those which have not adopted this framework in all study areas. Generally, the findings of this study showed that the hybrid VMD-SVM model combining the SDS strategy and innovative input selection framework is a useful and powerful tool for practical hydrological prediction work in the context of climate change.

     

A Novel Hybrid Method for River Discharge Prediction

Accurate prediction of river discharge is essential for the planning and management of water resources. This study proposes a novel hybrid method named HD-SKA by integrating two decomposition techniques (termed as HD) with support vector regression (SVR), K-nearest neighbor (KNN) and ARIMA models (combined as SKA) respectively. Firstly, the proposed method utilizes local mean decomposition (LMD) to decompose the original river discharge series into sub-series. Next, ensemble empirical mode decomposition (EEMD) is employed to further decompose the LMD-based sub-series into intrinsic mode functions. Further, the EEMD decomposed components are used as inputs in three data-driven models to predict river discharge respectively. The prediction of all components is then aggregated to obtain the results of HD-SVR, HD-KNN and HD-ARIMA models. The final prediction is obtained by taking the average prediction of these models. The proposed method is illustrated using five rivers in Indus Basin System. In five case studies, six models were built to compare the performance of the proposed HD-SKA model. The data analysis results show that the HD-SKA model performs better than all other considered models. The Diebold-Mariano test confirms the superiority of the proposed HD-SKA model over ARIMA, SVR, KNN, EEMD-ARIMA, EEMD-KNN, and EEMD-SVR models.

     

Forecasting Surface Water Level Fluctuations of Lake Van by Artificial Neural Networks

Lake Van in eastern Turkey has been subject to water level rise during the last decade and, consequently, the low-lying areas along the shore are inundated, giving problems to local administrators, governmental officials, irrigation activities and to people's property. Therefore, forecasting water levels of the Lake has started to attract the attention of the researchers in the country. An attempt has been made to use artificial neural networks (ANN) for modeling the temporal change water levels of Lake Van. A back-propagation algorithm is used for training. The study indicated that neural networks can successfully model the complex relationship between the rainfall and consecutive water levels. Three different cases were considered with the network trained for different arrangements of input nodes, such as current and antecedent lake levels, rainfall amounts. All of the three models yields relatively close results to each other. The neural network model is simpler and more reliable than the conventional methods such as autoregressive (AR), moving average (MA), and autoregressive moving average with exogenous input (ARMAX) models. It is shown that the relative errors for these two different models, are below 10% which is acceptable for engineering studies. In this study, dynamic changes of the lake level are evaluated. In contrast to classical methods, ANNs do not require strict assumptions such as linearity, normality, homoscadacity etc.     

Probabilistic Assessment of Monthly River Discharge using Copula and OSVR Approaches

In this study, two efficient approaches for bivariate simulation are presented, which include meteorological and hydrological variables. For this purpose, the applicability of support vector regression (SVR) model optimized by Ant colony and Copula-GARCH (Generalized Autoregressive Conditional Heteroscedasticity) algorithms were investigated and compared in simulating the river discharge based on total monthly rainfall in Talezang Basin, Iran. Entropy theory was used to select a suitable meteorological station corresponding to a hydrometric station. The vector autoregressive model was also used as the base model in Copula-GARCH simulations. According to the 99% confidence intervals of the simulations, the accuracy of both models was confirmed. The simulation results showed that the Copula-GARCH model was more accurate than the optimized SVR (OSVR) model. Considering the 90% efficiency (NSE=0.90) of the Copula-GARCH approach, the results show a 36% improvement of RMSE statistics by the Copula-GARCH model compared to the OSVR model in simulating the river discharge on a monthly scale. The results also showed that by combining nonlinear ARCH models with the copula-based simulations, the reliability of the simulation results increases, which was also confirmed using the violin plot. The results also showed an increase in the accuracy of the Copula-GARCH model at the minimum and maximum values of the data.

     

A Comparative Assessment of Artificial Neural Network,Generalized Regression Neural Network,Least-Square Support Vector Regression,and K-Nearest Neighbor Regression for Monthly Streamflow Forecasting in Linear and Nonlinear Conditions

Monthly forecasting of streamflow is of particular importance in water resources management especially in the provision of rule curves for dams. In this paper, the performance of four data-driven models with different structures including Artificial Neural Network (ANN), Generalized Regression Neural Network (GRNN), Least Square-Support Vector Regression (LS-SVR), and K-Nearest Neighbor Regression (KNN) are evaluated in order to forecast monthly inflow to Karkheh dam, Iran, in linear and non-linear conditions while the optimized values of the model parameters are determined in the same condition via the Leave-One-Out Cross Validation (LOOCV) method. Results show that the performance of the models is different in linear and nonlinear conditions; the cumulative ranking of the models according to the three assessment criteria including NSE, RMSE and R² indicates that ANN performs best in linear conditions while LS-SVR, GRNN and KNN are in the next ranks, respectively. But in nonlinear conditions, the best performance belongs to LS-SVR, followed by KNN, ANN, and GRNN models.     

Comparing ANN and ARIMA model in predicting the discharge of River Opeki from 2010 to 2020

Many attempts have been made in the recent past to model and forecast streamflow using various techniques with the use of time series techniques proving to be the most common. Time series analysis plays an important role in hydrological research. Traditionally, the class of autoregressive moving average techniques models has been the statistical method most widely used for modelling water discharge, but it has been shown to be deficient in representing nonlinear dynamics inherent in the transformation of runoff data. In contrast, the relatively newly improved and efficient soft computing technique artificial neural networks has the capability to approximate virtually any continuous function up to an arbitrary degree of accuracy, which is not otherwise true of other conventional hydrological techniques. This technique corresponds to human neurological system, which consists of a series of basic computing elements called neurons, which are interconnected together to form networks. The aim of the study is to compare the artificial neural network and autoregressive integrated moving average to model River Opeki discharge (1982–2010) and to use the best predictor to forecast the discharge of the river from 2010 to 2020. The performance of the two models was subjected to statistical test based on correlation coefficient (r) and the root‐mean‐square error. The result showed that autoregressive integrated moving average performs better considering the level of root‐mean‐square error and higher correlation coefficient.     

基于CEEMD-GRNN组合模型的月径流预测方法

针对径流序列的噪声因素与非线性特性,采用互补集合经验模态分解法(Complete Ensemble Empirical Mode Decomposition, CEEMD)与广义回归神经网络(Generalized Regression Neural Networks, GRNN)的组合模型,对汾河上游上静游站、汾河水库站、寨上站、兰村站1958～2000年的月径流序列进行实例研究,探究3种不同建模方式下的组合模型对预测精度的影响,其中组合模型1使用加权平均集成法将各分量预测结果相加,组合模型2去除高频分量后再使用加权平均集成法将剩余分量预测结果相加,组合模型3去除高频分量后将剩余分量预测结果直接相加;再将组合模型与单一GRNN模型进行对比。结果表明:各模型的确定性系数(NS)均大于0.5,预测结果均具有可信度;不同的月径流资料适用不同的建模方法,对于极差较小的月径流序列,组合模型1预测效果最好,与另外两种组合模型及单一模型相比,平均绝对误差(MAE)、平均绝对百分比误差(MAPE)、均方根误差(RMSE)分别平均减少26%,17%,23%;对于极差较大的径流序列,组合模型2预测效果最好,与另外两种组合模型及单一模型相比,MAE,MAPE,RMSE分别平均减少30%,28%,33%。组合模型2预测误差总小于组合模型3,即加权平均集成法对提高预测精度有一定作用。三种建模过程的CEEMD-GRNN组合模型预测误差均比单一GRNN模型小,说明组合模型较单一模型更适用于月径流预测。     

Precipitation Forecasting Using Wavelet-Genetic Programming and Wavelet-Neuro-Fuzzy Conjunction Models

Forecasting precipitation as a major component of the hydrological cycle is of primary importance in water resources engineering, planning and management as well as in scheduling irrigation practices. In the present study the abilities of hybrid wavelet-genetic programming [i.e. wavelet-gene-expression programming, WGEP] and wavelet-neuro-fuzzy (WNF) models for daily precipitation forecasting are investigated. In the first step, the single genetic programming (GEP) and neuro-fuzzy (NF) models are applied to forecast daily precipitation amounts based on previously recorded values, but the results are very weak. In the next step the hybrid WGEP and WNF models are used by introducing the wavelet coefficients as GEP and NF inputs, but no satisfactory results are produced, even though the accuracies increased to a great extent. In the third step, the new WGEP and WNF models are built; by merging the best single and hybrid models’ inputs and introducing them as the models inputs. The results show the new hybrid WGEP models are effective in forecasting daily precipitation, while the new WNF models are unable to learn the non linear process of precipitation very well.     

Monthly Rainfall Forecasting Using EEMD-SVR Based on Phase-Space Reconstruction

Rainfall links atmospheric and surficial processes and is one of the most important hydrologic variables. We apply support vector regression (SVR), which has a high generalization capability, to construct a rainfall forecasting model. Before construction of the model, a self-adaptive data analysis methodology called ensemble empirical mode decomposition (EEMD) is used to preprocess a rainfall data series. In addition, the phase-space reconstruction method is implemented to design input vectors for the forecasting model. The proposed hybrid model is applied to forecast the monthly rainfall at a weather station in Changchun, China as a case study. To demonstrate the capacity of the proposed hybrid model, a typical three-layer feed-forward artificial neural network model, an auto-regressive integrated moving average model, and a support vector regression model are constructed. Predictive performance of the models is evaluated based on normalized mean squared error (NMSE), mean absolute percent error (MAPE), Nash–Sutcliffe efficiency (NSE), and the coefficient of correlation (CC). Results indicate that the proposed hybrid model has the lowest NMSE and MAPE values of 0.10 and 14.90, respectively, and the highest NSE and CC values of 0.91 and 0.83, respectively, during the validation period. We conclude that the proposed hybrid model is feasible for monthly rainfall forecast and is better than the models currently in common use.     

Bivariate Simulation of Potential Evapotranspiration Using Copula-GARCH Model

Developing statistical period and simulating the required values in case of data shortage increases certainty and reliability of simulations and statistical analyses, which is very important in studies on hydrology and water resources. Therefore, in this study, for simulating values of potential evapotranspiration at Birjand Station located in eastern Iran, contemporaneous autoregressive moving average (CARMA), CARMA-generalized autoregressive conditional heteroskedasticity (GARCH), and Copula-GARCH models were used in statistical period of 1984–2019. The potential evapotranspiration and relative humidity time series were simulated using these three models. CARMA model has acceptable accuracy for simulating potential evapotranspiration values due to the effect of the second parameter on simulations. Nash–Sutcliffe efficiency (NSE) coefficient of CARMA model for simulating potential evapotranspiration values was estimated as 0.85. NSE coefficient of CARMA-GARCH model was obtained as 0.87 through extracting residuals of CARMA model and simulating variance of data using GARCH model. Comparing the CARMA and CARMA-GARCH models with each other, it was concluded that a combination of two linear and non-linear time series models increases simulation accuracy to some extent. Using Clayton copula (the selected copula from the studied copulas), the mentioned values were simulated by Copula-GARCH model. The results showed that among the three models used, Copula-GARCH model reduced root mean square error of bivariate simulation compared to CARMA and CARMA-GARCH models by 15 and 13%, respectively. The results also showed that the proposed model simulates the average, first, and third quarters and range of changes in the data by 5 and 95% better than the two CARMA and CARMA-GARCH models.

     

Reservoir Inflow Modeling Using Temporal Neural Networks with Forgetting Factor Approach

In this paper, a recursive training procedure with forgetting factor is proposed for on-line calibration of temporal neural networks. The forgetting factor discounts old measurements through an on-line model calibration. The forgetting factor approach enables the recursive algorithm to reduce the effect of the older error data by multiplying the error data by a discounting factor. The proposed procedure is used to calibrate a temporal neural network for reservoir inflow modeling. The mean monthly inflow of the Karoon-III reservoir dam in the south-western part of Iran is used to test the performance of the proposed approach. An autoregressive moving average (ARMA) model is also applied to the same data. The temporal neural network, which is trained with the proposed approach, has shown a significant improvement in the forecast accuracy in comparison with the network trained by the conventional method. It is also demonstrated that the neural network trained with forgetting factor results in better forecasts compared to the statistical ARMA model, which has been calibrated through this approach.