Flood duration, volume, and peak flow are important considerations in flood risk analysis and man... more Flood duration, volume, and peak flow are important considerations in flood risk analysis and management of hydraulic structures. The conventional flood frequency analysis assumed that the marginal distribution functions of flood parameters follow a certain pattern. However, such assumption is impractical because a flood event is multivariate and the flood parameter distributions can be different. These discrepancies were addressed using bivariate joint distributions and Copula function which allow flood parameters having different marginal distributions to be analyzed simultaneously. The analysis used hourly stream flow data for 45 years recorded at the Rantau Panjang gauging station on the Johor River in Malaysia. It was found that flood duration and volume are best fitted by the generalized extreme value distribution while peak flow by the Generalized Pareto. Inference function for margin (IFM) method was applied to model the joint distributions of correlated flood variables for each pair and the results showed that all the calculated θ values were in acceptable range of Gaussian Copula. By horizontally cutting the joint cumulative distribution function (CDF), a set of contour lines were obtained for Gaussian Copula which represented the occurrence probabilities for the joint variables. Also the joint return period for pair of flood variables was calculated.
Soil and water are the two major resources in the Earth’s hydro biological and geological systems... more Soil and water are the two major resources in the Earth’s hydro biological and geological systems. The hydrology of arid areas has become a topic of interest recently for hydrologists as water shortage at these areas can affect the agriculture, irrigation, and industry as a whole. This has also prompted water resource planners to more thoroughly investigate water resource crisis at arid areas. In this respect, the Soil and Water Assessment Tool (SWAT), a semi-distributed hydrological model, can be a subsidiary tool to be used in the prediction of surface runoff (blue water). This paper presents the application of SWAT on the Roodan watershed, which is located in the southern part of Iran and has 215 mm of annual precipitation. SWAT was engaged to know more about the daily flow and to evaluate the runoff volume. Three continuous scenarios were defined over the 21 years (1988- 1992, 1993-2001, 2002-2008) for the land use map as it was found that continuous update of this layer were basically done during these periods. Results of sensitivity analysis showed that parameters related to transmission losses are most sensitive for this watershed. Furthermore, the SWAT had also visualized from the input data that the sub basins which have been designated for agricultural activities from 1988 to 2008 were at the southwest, center and northeast parts of Roodan watershed. Strength of modeling was evaluated by percentage of observations covered by the 95 Percentage prediction uncertainty (P-factor) and relative width of 95 % probability band (R-factor). The P and R factors in this study were recorded at, for calibration and validation periods, 50 % and 0.18 (calibration), and 50 % and 0.17 (validation) respectively. Nash-Sutcliffe and PBIAS obtained for calibration period were 0.75 and 1.5 %, and those for validation period were 0.64 and 21 %. However, results showed an underestimation trend for most peak flows during the modeling of daily stream flow. Nevertheless, the annual runoff volume for calibration and validation periods depicted a promising performance and thus validated the usage of SWAT as a subsidiary hydrological tool for water management projects attributed with stream flow and runoff volume.
Modelling and Simulation in Engineering, Feb 6, 2014
Rainfall-runoff simulation in hydrology using artificial intelligence presents the nonlinear rela... more Rainfall-runoff simulation in hydrology using artificial intelligence presents the nonlinear relationships using neural networks. In
this study, a hybrid network presented as a feedforward modular neural network (FF-MNN) has been developed to predict the
daily rainfall-runoff of the Roodan watershed at the southern part of Iran. This FF-MNN has three layers—input, hidden, and
output.The hidden layer has two types of neural expert or module. Hydrometeorological data of the catchment were collected for
21 years. Heuristicmethod was used to develop the MNN for exploring daily flow generalization. Two training algorithms, namely,
backpropagation with momentum and Levenberg-Marquardt, were used. Sigmoid and linear transfer functions were employed
to explore the network’s optimum behavior. Cross-validation and predictive uncertainty assessments were carried out to protect
overtiring and overparameterization, respectively. Results showed that the FF-MNN could satisfactorily predict streamflow during
testing period. The Nash-Sutcliff coefficient, coefficient of determination, and root mean square error obtained usingMNN during training and test periods were 0.85, 0.85, and 39.4 and 0.57, 0.58, and 32.2, respectively.The predictive uncertainties for both periods were 0.39 and 0.44, respectively. Generally, the study showed that the FF-MNN can give promising prediction for rainfall-runoff relations.
A copula based methodology is presented in this study for the flood frequency analysis of Johor r... more A copula based methodology is presented in this study for the flood frequency analysis of Johor river basin in southern Malaysia. The objective of this study was to find the best-fit distributions to the flood characteristics and finding their joint probability for flood frequency analysis. The joint dependence structures of the flood characteristics (peak flow (Qp), flood volume (V) and flood duration (D) were modelled using an Archimedean Copula (t-Copula). The distribution methods were tested to identify the best distribution that would fit the distributions of various flood characteristics. Based on Kolmogorov-Smirnov (K-S) test the Generalized Pareto distribution is the best-fit distribution for peak flow. On the other hand, General Extreme Value (GEV) is the best-fit distribution for the flood duration and flood volume. The best fit distributions were then used to develop the joint Cumulative Distribution Function (CDF) of the flood characteristics based on t-Copula. Peak flow-volume, volume-duration and peak flow-duration pairs were found to be negatively related. It is expected that the bivariate distributions formulated is useful for flood risk assessment and design of hydraulic structures in Malaysia
Caspian Journal of Applied Sciences Research, 2013
This study describes the application of a semi-distributed model for flow simulation and assessme... more This study describes the application of a semi-distributed model for flow simulation and assessment of sensitive parameters. Semi-distributed model is a trade-off between fully distributed and lumped models. In this study, the Soil and Water Assessment Tool (SWAT) model was applied for modeling the average monthly flow in Roodan watershed, Iran. This watershed has arid and semi-arid areas critical for development as they have the potential to preserve surface waters in spite of water scarcity. The major purposes of this research were (1) to identify sensitive parameters; and (2) to evaluate the monthly flow at arid region (south of Iran) with low precipitation. To formulate a better model, the impacts of three additional parameters, namely revap coefficient (GW_REVAP), reach evaporation adjustment factor (EVRCH) and length of main channel (CH_L(2)) were reviewed critically. To delineate the watershed, the kind of data used were the digital elevation map (DEM), land use map, soil layers properties and meteorological data. Then, the model was calibrated using the Sequential Uncertainty Fitting (SUFI-2) algorithm. This method is a kind of inverse modeling and considers uniqueness. A modeler defines a large limit range values for every parameter and after every iteration every parameter will get small limit range values. Generally, the model gave satisfactory values of Nash- Sutcliffe (NS) and coefficient of determination (R2). Values of R2 and NS were 0.93 and 0.92 respectively for calibration. For validation, both values were reported at 0.83. Usually, calibration and validation of hydrological models have different accuracy. The main reason is that the model validate for different phenomena. In such cases, the calibration of additional parameters, i.e. GW_REVAP, EVRCH and CH_L(2), cannot be substantially improved as well.
Flooding is a natural part of a river's life cycle but it is a major disaster affecting many regi... more Flooding is a natural part of a river's life cycle but it is a major disaster affecting many regions around the world, year after year. Malaysia is among the countries that faces potential flooding problems due to rapid development and, improper river systems. The Skudai river basin covers an area of 293.7 km2 in the south-western part of Johor in Malaysia. The Skudai River has come under the spotlight due to the impacts of future development projects. Land clearing for urbanization and, infrastructure construction may increase the magnitude of flood. Flood risk map is one of the best ways to study and understand the flood behavior. To produce flood level at various locations along the river and flood plain, hydraulic modeling is required to carry out the flood simulation. However, analysis a river system requires tremendous amount of data such as rainfall distribution, river properties and, most important, the flood plain topography. This study presents flood mapping results in Skudai River basin in Johor, Malaysia using InfoWorks software 1D modeling. The tasks involved hydrological modeling, hydrodynamic modeling. Ground model and generation of flood risk map. The results show that eighteen locations are affected by flood of 100 years ARI.
Journal of Environmental Science and Technology, 2012
The conception of modeling in hydrology is involved with relationships of water, climate, soil an... more The conception of modeling in hydrology is involved with relationships of water, climate, soil and land use. Moreover, hydrological models include temporal and spatial features. Behavior of each feature controlled by its own and therefore it makes a vast variety for types of hydrological models. Hydrological models are the main tools for hydrologists with different purposes to use such as water resource management, ground water modeling, urban and rural watershed management and so on. Many hydrological models have been developed and refined during the past four decades and it is required to fully understand their characteristics to effortlessly employ them. Therefore, hydrologists need to familiarize themselves with the classification of hydrological models and understand the theoretical definition behind them. However, in regard to this issue, only a few discrete studies had been done. Classification of hydrological models is not exact and different hydrologist may give different definitions. The reason is that the nature of models is often the same but many models have overlapping characteristics. Thus, this study was aimed at showing the dominant classifications for hydrological models alongside the different views from past to present but generally, they have common meaning even though they may be classified under different categories. In addition, although there are overlapping features in different hydrological models, their nature is not that hard to understand.
Journal of Environmental Science and Technology, 2012
Understanding the distribution of flood variable is crucial for the purposes of flood prediction ... more Understanding the distribution of flood variable is crucial for the purposes of flood prediction and hydraulic design. Three major parameters that are useful to describe a flood event are peak flow or flood peak, flood volume and flood duration. This study aimed at exploring the statistical distribution of these parameters for Johor River in south of Malaysia. Hourly data were recorded for 45 years from Rantau Panjang gauging station. The annual flood peak was selected from the maximum flow in each water year (July-June). Five probability distributions namely Gamma, Weibull, Exponential, Gumbel and Generalize Extreme Value (GEV) were used to model the distribution of flood parameters. Kolmogorov-Smirnov and chi-squared goodness-of-fit tests were used to evaluate the best fit. Goodness-of-fit tests at 5% level of significance indicated that all the models can be used to model the distribution of peak flow, duration and volume. However, the Exponential distribution is the most suitable model when tested with the Kolmogorov-Smirnov test whereas GEV is the best when chi-squared test was used. The result can be used as a basis to improve flood frequency analysis for Malaysian rivers.
Modeling of rainfall-runoff relationship in a watershed is of prime importance to water resources... more Modeling of rainfall-runoff relationship in a watershed is of prime importance to water resources engineers and hydrologists in designing hydraulic structures, flood control plans and stormwater management strategies. The study site, Sungai Johor watershed, has annual precipitation of about 2500 mm. This study aimed at developing an Artificial Neural Network (ANN) model for annual peak flow during 1980-2010 without employing exogenous runoff-generating process variables. ANNs have been known as having the ability to model nonlinear mechanisms. In the present study, the sensitivity of input data, namely the initial discharge, average temperature, average evaporation, average wind speed, and average relative humidity were also evaluated. Multi layer perceptron (MLP) network was chosen for modeling the annual flood/peakflow. The data was divided into three sets, notably data for training, cross-validation and testing. The data analysis process involved cleansing, normalization and data division. Next, for the best architecture, the behavior of the input data was assessed separately. Results showed that the most sensitive input data were the initial discharge, relative humidity and temperature. The best architecture was obtained by neurons 3-9-1 (input-hidden-output layer). This was computed by using the linear tangent hyperbolic axon as transfer function. The best learning algorithm for the hidden and output layers were conjugate gradient and momentum (back propagation) respectively. For this study the coefficient of determination (R2), Nash and Sutcliffe (NS) and Root Mean Square error (RMSE) for validation period were 0.64, 0.6 and 48.9 (m3/s) respectively. This study revealed that the application of various inputs data together did not significantly improve the modeling performance in ANN. The use of exogenous variables such as initial flows can be beneficial for primary evaluation when there is significant missing data or when the data accuracy is questionable.
Research Journal of Environmental and Earth Sciences, Feb 20, 2013
A physically-based model namely the Soil Water Assessment Tool (SWAT) was used on the Roodan wate... more A physically-based model namely the Soil Water Assessment Tool (SWAT) was used on the Roodan watershed in southern part of Iran; the watershed has an area of 10570 km2. The main objectives were to simulate monthly discharge and evaluate the base and peak flows separately. Required parameters to run the model were meteorological data, soil type, land use, management practices and topography maps at watershed scale. To find the sensitive parameters, an initial sensitivity analysis was performed using the Latin Hypercube sampling One-at-A-Time (LH-OAT) method embedded in the SWAT model. Then, the model was calibrated and validated for stream flow using the SWAT-CUP program. Generally, the model was assessed using the modified coefficient of determination (bR2), Nash-Sutcliffe (NS) and PBIAS. Values of bR2 and NS were 0.93 and 0.92 for calibration respectively and 0.69 and 0.83, respectively, for validation. For calibration and validation, PBIAS were obtained at 23 and 5%, respectively. Reviewing the results, it seems that simulation of the monthly peak flows has better harmony (fluctuation) than monthly base flows for Roodan watershed. To summarize, the simulated SWAT stream flow was within the acceptable range for Roodan watershed as an arid catchment.
Research Journal of Applied Sciences, Jun 20, 2013
Besides peakflow, a flood event is also characterized by other possibly mutually correlated varia... more Besides peakflow, a flood event is also characterized by other possibly mutually correlated variables. This study was aimed at exploring the statistical distribution of peakflow, flood duration and flood volume for Johor River in south of Peninsular Malaysia. Hourly data were recorded for 45 years from the Rantau Panjang gauging station. The annual peakflow was selected from the maximum flow in each water year (July-June). Five probability distributions, namely Gamma, Generalized Pareto, Beta, Pearson and Generalized Extreme Value (GEV) were used to model the distribution of peakflow events. Anderson-Darling and Chi-squared goodness-of-fit tests were used to evaluate the best fit. Goodness-of-fit tests at 5% level of significance indicate that all the models can be used to model the distribution of peakflow, flood duration and flood volume. However, Generalized Pareto distribution was found to be the most suitable model when tested with the Anderson-Darling-Smirnov test and the Chi-squared test suggested that Generalized Extreme Value was the best for peakflow. The result of this research can be used to improve flood frequency analysis.
Flood duration, volume, and peak flow are important considerations in flood risk analysis and man... more Flood duration, volume, and peak flow are important considerations in flood risk analysis and management of hydraulic structures. The conventional flood frequency analysis assumed that the marginal distribution functions of flood parameters follow a certain pattern. However, such assumption is impractical because a flood event is multivariate and the flood parameter distributions can be different. These discrepancies were addressed using bivariate joint distributions and Copula function which allow flood parameters having different marginal distributions to be analyzed simultaneously. The analysis used hourly stream flow data for 45 years recorded at the Rantau Panjang gauging station on the Johor River in Malaysia. It was found that flood duration and volume are best fitted by the generalized extreme value distribution while peak flow by the Generalized Pareto. Inference function for margin (IFM) method was applied to model the joint distributions of correlated flood variables for each pair and the results showed that all the calculated θ values were in acceptable range of Gaussian Copula. By horizontally cutting the joint cumulative distribution function (CDF), a set of contour lines were obtained for Gaussian Copula which represented the occurrence probabilities for the joint variables. Also the joint return period for pair of flood variables was calculated.
Soil and water are the two major resources in the Earth’s hydro biological and geological systems... more Soil and water are the two major resources in the Earth’s hydro biological and geological systems. The hydrology of arid areas has become a topic of interest recently for hydrologists as water shortage at these areas can affect the agriculture, irrigation, and industry as a whole. This has also prompted water resource planners to more thoroughly investigate water resource crisis at arid areas. In this respect, the Soil and Water Assessment Tool (SWAT), a semi-distributed hydrological model, can be a subsidiary tool to be used in the prediction of surface runoff (blue water). This paper presents the application of SWAT on the Roodan watershed, which is located in the southern part of Iran and has 215 mm of annual precipitation. SWAT was engaged to know more about the daily flow and to evaluate the runoff volume. Three continuous scenarios were defined over the 21 years (1988- 1992, 1993-2001, 2002-2008) for the land use map as it was found that continuous update of this layer were basically done during these periods. Results of sensitivity analysis showed that parameters related to transmission losses are most sensitive for this watershed. Furthermore, the SWAT had also visualized from the input data that the sub basins which have been designated for agricultural activities from 1988 to 2008 were at the southwest, center and northeast parts of Roodan watershed. Strength of modeling was evaluated by percentage of observations covered by the 95 Percentage prediction uncertainty (P-factor) and relative width of 95 % probability band (R-factor). The P and R factors in this study were recorded at, for calibration and validation periods, 50 % and 0.18 (calibration), and 50 % and 0.17 (validation) respectively. Nash-Sutcliffe and PBIAS obtained for calibration period were 0.75 and 1.5 %, and those for validation period were 0.64 and 21 %. However, results showed an underestimation trend for most peak flows during the modeling of daily stream flow. Nevertheless, the annual runoff volume for calibration and validation periods depicted a promising performance and thus validated the usage of SWAT as a subsidiary hydrological tool for water management projects attributed with stream flow and runoff volume.
Modelling and Simulation in Engineering, Feb 6, 2014
Rainfall-runoff simulation in hydrology using artificial intelligence presents the nonlinear rela... more Rainfall-runoff simulation in hydrology using artificial intelligence presents the nonlinear relationships using neural networks. In
this study, a hybrid network presented as a feedforward modular neural network (FF-MNN) has been developed to predict the
daily rainfall-runoff of the Roodan watershed at the southern part of Iran. This FF-MNN has three layers—input, hidden, and
output.The hidden layer has two types of neural expert or module. Hydrometeorological data of the catchment were collected for
21 years. Heuristicmethod was used to develop the MNN for exploring daily flow generalization. Two training algorithms, namely,
backpropagation with momentum and Levenberg-Marquardt, were used. Sigmoid and linear transfer functions were employed
to explore the network’s optimum behavior. Cross-validation and predictive uncertainty assessments were carried out to protect
overtiring and overparameterization, respectively. Results showed that the FF-MNN could satisfactorily predict streamflow during
testing period. The Nash-Sutcliff coefficient, coefficient of determination, and root mean square error obtained usingMNN during training and test periods were 0.85, 0.85, and 39.4 and 0.57, 0.58, and 32.2, respectively.The predictive uncertainties for both periods were 0.39 and 0.44, respectively. Generally, the study showed that the FF-MNN can give promising prediction for rainfall-runoff relations.
A copula based methodology is presented in this study for the flood frequency analysis of Johor r... more A copula based methodology is presented in this study for the flood frequency analysis of Johor river basin in southern Malaysia. The objective of this study was to find the best-fit distributions to the flood characteristics and finding their joint probability for flood frequency analysis. The joint dependence structures of the flood characteristics (peak flow (Qp), flood volume (V) and flood duration (D) were modelled using an Archimedean Copula (t-Copula). The distribution methods were tested to identify the best distribution that would fit the distributions of various flood characteristics. Based on Kolmogorov-Smirnov (K-S) test the Generalized Pareto distribution is the best-fit distribution for peak flow. On the other hand, General Extreme Value (GEV) is the best-fit distribution for the flood duration and flood volume. The best fit distributions were then used to develop the joint Cumulative Distribution Function (CDF) of the flood characteristics based on t-Copula. Peak flow-volume, volume-duration and peak flow-duration pairs were found to be negatively related. It is expected that the bivariate distributions formulated is useful for flood risk assessment and design of hydraulic structures in Malaysia
Caspian Journal of Applied Sciences Research, 2013
This study describes the application of a semi-distributed model for flow simulation and assessme... more This study describes the application of a semi-distributed model for flow simulation and assessment of sensitive parameters. Semi-distributed model is a trade-off between fully distributed and lumped models. In this study, the Soil and Water Assessment Tool (SWAT) model was applied for modeling the average monthly flow in Roodan watershed, Iran. This watershed has arid and semi-arid areas critical for development as they have the potential to preserve surface waters in spite of water scarcity. The major purposes of this research were (1) to identify sensitive parameters; and (2) to evaluate the monthly flow at arid region (south of Iran) with low precipitation. To formulate a better model, the impacts of three additional parameters, namely revap coefficient (GW_REVAP), reach evaporation adjustment factor (EVRCH) and length of main channel (CH_L(2)) were reviewed critically. To delineate the watershed, the kind of data used were the digital elevation map (DEM), land use map, soil layers properties and meteorological data. Then, the model was calibrated using the Sequential Uncertainty Fitting (SUFI-2) algorithm. This method is a kind of inverse modeling and considers uniqueness. A modeler defines a large limit range values for every parameter and after every iteration every parameter will get small limit range values. Generally, the model gave satisfactory values of Nash- Sutcliffe (NS) and coefficient of determination (R2). Values of R2 and NS were 0.93 and 0.92 respectively for calibration. For validation, both values were reported at 0.83. Usually, calibration and validation of hydrological models have different accuracy. The main reason is that the model validate for different phenomena. In such cases, the calibration of additional parameters, i.e. GW_REVAP, EVRCH and CH_L(2), cannot be substantially improved as well.
Flooding is a natural part of a river's life cycle but it is a major disaster affecting many regi... more Flooding is a natural part of a river's life cycle but it is a major disaster affecting many regions around the world, year after year. Malaysia is among the countries that faces potential flooding problems due to rapid development and, improper river systems. The Skudai river basin covers an area of 293.7 km2 in the south-western part of Johor in Malaysia. The Skudai River has come under the spotlight due to the impacts of future development projects. Land clearing for urbanization and, infrastructure construction may increase the magnitude of flood. Flood risk map is one of the best ways to study and understand the flood behavior. To produce flood level at various locations along the river and flood plain, hydraulic modeling is required to carry out the flood simulation. However, analysis a river system requires tremendous amount of data such as rainfall distribution, river properties and, most important, the flood plain topography. This study presents flood mapping results in Skudai River basin in Johor, Malaysia using InfoWorks software 1D modeling. The tasks involved hydrological modeling, hydrodynamic modeling. Ground model and generation of flood risk map. The results show that eighteen locations are affected by flood of 100 years ARI.
Journal of Environmental Science and Technology, 2012
The conception of modeling in hydrology is involved with relationships of water, climate, soil an... more The conception of modeling in hydrology is involved with relationships of water, climate, soil and land use. Moreover, hydrological models include temporal and spatial features. Behavior of each feature controlled by its own and therefore it makes a vast variety for types of hydrological models. Hydrological models are the main tools for hydrologists with different purposes to use such as water resource management, ground water modeling, urban and rural watershed management and so on. Many hydrological models have been developed and refined during the past four decades and it is required to fully understand their characteristics to effortlessly employ them. Therefore, hydrologists need to familiarize themselves with the classification of hydrological models and understand the theoretical definition behind them. However, in regard to this issue, only a few discrete studies had been done. Classification of hydrological models is not exact and different hydrologist may give different definitions. The reason is that the nature of models is often the same but many models have overlapping characteristics. Thus, this study was aimed at showing the dominant classifications for hydrological models alongside the different views from past to present but generally, they have common meaning even though they may be classified under different categories. In addition, although there are overlapping features in different hydrological models, their nature is not that hard to understand.
Journal of Environmental Science and Technology, 2012
Understanding the distribution of flood variable is crucial for the purposes of flood prediction ... more Understanding the distribution of flood variable is crucial for the purposes of flood prediction and hydraulic design. Three major parameters that are useful to describe a flood event are peak flow or flood peak, flood volume and flood duration. This study aimed at exploring the statistical distribution of these parameters for Johor River in south of Malaysia. Hourly data were recorded for 45 years from Rantau Panjang gauging station. The annual flood peak was selected from the maximum flow in each water year (July-June). Five probability distributions namely Gamma, Weibull, Exponential, Gumbel and Generalize Extreme Value (GEV) were used to model the distribution of flood parameters. Kolmogorov-Smirnov and chi-squared goodness-of-fit tests were used to evaluate the best fit. Goodness-of-fit tests at 5% level of significance indicated that all the models can be used to model the distribution of peak flow, duration and volume. However, the Exponential distribution is the most suitable model when tested with the Kolmogorov-Smirnov test whereas GEV is the best when chi-squared test was used. The result can be used as a basis to improve flood frequency analysis for Malaysian rivers.
Modeling of rainfall-runoff relationship in a watershed is of prime importance to water resources... more Modeling of rainfall-runoff relationship in a watershed is of prime importance to water resources engineers and hydrologists in designing hydraulic structures, flood control plans and stormwater management strategies. The study site, Sungai Johor watershed, has annual precipitation of about 2500 mm. This study aimed at developing an Artificial Neural Network (ANN) model for annual peak flow during 1980-2010 without employing exogenous runoff-generating process variables. ANNs have been known as having the ability to model nonlinear mechanisms. In the present study, the sensitivity of input data, namely the initial discharge, average temperature, average evaporation, average wind speed, and average relative humidity were also evaluated. Multi layer perceptron (MLP) network was chosen for modeling the annual flood/peakflow. The data was divided into three sets, notably data for training, cross-validation and testing. The data analysis process involved cleansing, normalization and data division. Next, for the best architecture, the behavior of the input data was assessed separately. Results showed that the most sensitive input data were the initial discharge, relative humidity and temperature. The best architecture was obtained by neurons 3-9-1 (input-hidden-output layer). This was computed by using the linear tangent hyperbolic axon as transfer function. The best learning algorithm for the hidden and output layers were conjugate gradient and momentum (back propagation) respectively. For this study the coefficient of determination (R2), Nash and Sutcliffe (NS) and Root Mean Square error (RMSE) for validation period were 0.64, 0.6 and 48.9 (m3/s) respectively. This study revealed that the application of various inputs data together did not significantly improve the modeling performance in ANN. The use of exogenous variables such as initial flows can be beneficial for primary evaluation when there is significant missing data or when the data accuracy is questionable.
Research Journal of Environmental and Earth Sciences, Feb 20, 2013
A physically-based model namely the Soil Water Assessment Tool (SWAT) was used on the Roodan wate... more A physically-based model namely the Soil Water Assessment Tool (SWAT) was used on the Roodan watershed in southern part of Iran; the watershed has an area of 10570 km2. The main objectives were to simulate monthly discharge and evaluate the base and peak flows separately. Required parameters to run the model were meteorological data, soil type, land use, management practices and topography maps at watershed scale. To find the sensitive parameters, an initial sensitivity analysis was performed using the Latin Hypercube sampling One-at-A-Time (LH-OAT) method embedded in the SWAT model. Then, the model was calibrated and validated for stream flow using the SWAT-CUP program. Generally, the model was assessed using the modified coefficient of determination (bR2), Nash-Sutcliffe (NS) and PBIAS. Values of bR2 and NS were 0.93 and 0.92 for calibration respectively and 0.69 and 0.83, respectively, for validation. For calibration and validation, PBIAS were obtained at 23 and 5%, respectively. Reviewing the results, it seems that simulation of the monthly peak flows has better harmony (fluctuation) than monthly base flows for Roodan watershed. To summarize, the simulated SWAT stream flow was within the acceptable range for Roodan watershed as an arid catchment.
Research Journal of Applied Sciences, Jun 20, 2013
Besides peakflow, a flood event is also characterized by other possibly mutually correlated varia... more Besides peakflow, a flood event is also characterized by other possibly mutually correlated variables. This study was aimed at exploring the statistical distribution of peakflow, flood duration and flood volume for Johor River in south of Peninsular Malaysia. Hourly data were recorded for 45 years from the Rantau Panjang gauging station. The annual peakflow was selected from the maximum flow in each water year (July-June). Five probability distributions, namely Gamma, Generalized Pareto, Beta, Pearson and Generalized Extreme Value (GEV) were used to model the distribution of peakflow events. Anderson-Darling and Chi-squared goodness-of-fit tests were used to evaluate the best fit. Goodness-of-fit tests at 5% level of significance indicate that all the models can be used to model the distribution of peakflow, flood duration and flood volume. However, Generalized Pareto distribution was found to be the most suitable model when tested with the Anderson-Darling-Smirnov test and the Chi-squared test suggested that Generalized Extreme Value was the best for peakflow. The result of this research can be used to improve flood frequency analysis.
Modeling of rainfall-runoff relationship in a watershed is of prime importance to water resources... more Modeling of rainfall-runoff relationship in a watershed is of prime importance to water resources engineers and hydrologists in designing hydraulic structures, flood control plans and stormwater management strategies. The study site, Sungai Johor watershed, has annual precipitation of about 2500 mm. This study aimed at developing an Artificial Neural Network (ANN) model for annual peak flow during 1980-2010 without employing exogenous runoff-generating process variables. ANNs have been known as having the ability to model nonlinear mechanisms. In the present study, the sensitivity of input data, namely the initial discharge, average temperature, average evaporation, average wind speed, and average relative humidity were also evaluated. Multi layer perceptron (MLP) network was chosen for modeling the annual flood/peakflow. The data was divided into three sets, notably data for training, cross-validation and testing. The data analysis process involved cleansing, normalization and data division. Next, for the best architecture, the behavior of the input data was assessed separately. Results showed that the most sensitive input data were the initial discharge, relative humidity and temperature. The best architecture was obtained by neurons 3-9-1 (input-hidden-output layer). This was computed by using the linear tangent hyperbolic axon as transfer function. The best learning algorithm for the hidden and output layers were conjugate gradient and momentum (back propagation) respectively. For this study the coefficient of determination (R2), Nash and Sutcliffe (NS) and Root Mean Square error (RMSE) for validation period were 0.64, 0.6 and 48.9 (m3/s) respectively. This study revealed that the application of various inputs data together did not significantly improve the modeling performance in ANN. The use of exogenous variables such as initial flows can be beneficial for primary evaluation when there is significant missing data or when the data accuracy is questionable.
One of the major issues for semidistributed models is calibration of sensitive parameters. This s... more One of the major issues for semidistributed models is calibration of sensitive parameters. This study compared 3 scenarios for Soil and Water Assessment Tool (SWAT) model for calibration and uncertainty. Roodan watershed has been selected for simulation of daily flow in southern part of Iran with an area of 10 570 km2. After preparation of required data and implementation of the SWAT model, sensitivity analysis has been performed by Latin Hypercube One-factor-At-a-Time method on those parameters which are effective for flow simulation. Then, SWAT Calibration and Uncertainty Program (SWAT-CUP) has been used for calibration and uncertainty analysis. Three schemes for calibration were followed for the Roodan watershed modeling in calibration analysis as evolution. These include the following: the global method (scheme 1), this is a method that takes in all globally adjusted sensitive parameters for the whole watershed; the discretization method (scheme 2), this method considered the do...
The application of modular artificial neural networks (MANN) has been increasing in recent years ... more The application of modular artificial neural networks (MANN) has been increasing in recent years because of their capabilities in simulating hydrological phenomena. This paper documents the successful application of MANN for a semi-arid region situated at the southern part of Iran. The main objective was to assess a hybrid structure derived from two parallel multi-layer perceptrons (MLPs) in a hidden layer for rainfallrunoff prediction. This MANN had three layers, that is, input, hidden and output layers. The hidden layer had a training algorithm and two types of modules (halves) where every module had its unique transfer function. To prevent overtraining, cross validation assessment was done during the development of various topologies. The Nash-Sutcliff coefficient (NS), root and normalised mean square error (RMSE and NRMSE) were used to assess the model’s performance, and the corresponding result values were 0.85, 42 and 2.6 for training period; 0.71, 27.5 and 5.7 for cross validation period; and finally 0.57, 32.2 and 6.2 for test period. Generally, the simulation results are promising, but further evaluation is needed for large and arid catchments.
Rainfall-runoff simulation in hydrology using artificial intelligence presents the nonlinear rela... more Rainfall-runoff simulation in hydrology using artificial intelligence presents the nonlinear relationships using neural networks. In this study, a hybrid network presented as a feedforward modular neural network (FF-MNN) has been developed to predict the daily rainfall-runoff of the Roodan watershed at the southern part of Iran. This FF-MNN has three layers—input, hidden, and output. The hidden layer has two types of neural expert or module. Hydrometeorological data of the catchment were collected for 21 years. Heuristic method was used to develop the MNN for exploring daily flow generalization. Two training algorithms, namely, backpropagation with momentum and Levenberg-Marquardt, were used. Sigmoid and linear transfer functions were employed to explore the network’s optimum behavior. Cross-validation and predictive uncertainty assessments were carried out to protect overtiring and overparameterization, respectively. Results showed that the FF-MNN could satisfactorily predict strea...
Flood duration, volume, and peak flow are important considerations in flood risk analysis and man... more Flood duration, volume, and peak flow are important considerations in flood risk analysis and management of hydraulic structures. The conventional flood frequency analysis assumed that the marginal distribution functions of flood parameters follow a certain pattern. However, such assumption is impractical because a flood event is multivariate and the flood parameter distributions can be different. These discrepancies were addressed using bivariate joint distributions and Copula function which allow flood parameters having different marginal distributions to be analyzed simultaneously. The analysis used hourly stream flow data for 45 years recorded at the Rantau Panjang gauging station on the Johor River in Malaysia. It was found that flood duration and volume are best fitted by the generalized extreme value distribution while peak flow by the Generalized Pareto. Inference function for margin (IFM) method was applied to model the joint distributions of correlated flood variables for each pair and the results showed that all the calculated θ values were in acceptable range of Gaussian Copula. By horizontally cutting the joint cumulative distribution function (CDF), a set of contour lines were obtained for Gaussian Copula which represented the occurrence probabilities for the joint variables. Also the joint return period for pair of flood variables was calculated.
Abstract: A physically-based model namely the Soil Water Assessment Tool (SWAT) was used on the R... more Abstract: A physically-based model namely the Soil Water Assessment Tool (SWAT) was used on the Roodan watershed in southern part of Iran; the watershed has an area of 10570 km 2 . The main objectives were to simulate monthly discharge and evaluate the base and peak flows separately. Required parameters to run the model were meteorological data, soil type, land use, management practices and topography maps at watershed scale. To find the sensitive parameters, an initial sensitivity analysis was performed using the Latin Hypercube sampling One-at-ATime (LH-OAT) method embedded in the SWAT model. Then, the model was calibrated and validated for stream flow using the SWAT-CUP program. Generally, the model was assessed using the modified coefficient of determination (bR 2 ), Nash-Sutcliffe (NS) and PBIAS. Values of bR 2 and NS were 0.93 and 0.92 for calibration respectively and 0.69 and 0.83, respectively, for validation. For calibration and validation, PBIAS were obtained at 23 and 5%,...
The modeling of rainfall-runoff relationship in a watershed is very important in designing hydrau... more The modeling of rainfall-runoff relationship in a watershed is very important in designing hydraulic structures, controlling flood and managing storm water. Artificial Neural Networks (ANNs) are known as having the ability to model nonlinear mechanisms. This study aimed at developing a Generalized Feed Forward (GFF) network model for predicting annual flood (depth) of Johor River in Peninsular Malaysia. In order to avoid over training, cross-validation technique was performed for optimizing the model. In addition, predictive uncertainty index was used to protect of over parameterization. The governing training algorithm was back propagation with momentum term and tangent hyperbolic types was used as transfer function for hidden and output layers. The results showed that the optimum architecture was derived by linear tangent hyperbolic transfer function for both hidden and output layers. The values of Nash and Sutcliffe (NS) and Root mean square error (RMSE) obtained 0.98 and 5.92 for the test period. Cross validation evaluation showed 9 process elements is adequate in hidden layer for optimum generalization by considering the predictive uncertainty index obtained (0.14) for test period which is acceptable.
Flood duration, volume, and peak flow are important considerations in flood risk analysis and man... more Flood duration, volume, and peak flow are important considerations in flood risk analysis and management of hydraulic structures. The conventional flood frequency analysis assumed that the marginal distribution functions of flood parameters follow a certain pattern. However, such assumption is impractical because a flood event is multivariate and the flood parameter distributions can be different. These discrepancies were addressed using bivariate joint distributions and Copula function which allow flood parameters having different marginal distributions to be analyzed simultaneously. The analysis used hourly stream flow data for 45 years recorded at the Rantau Panjang gauging station on the Johor River in Malaysia. It was found that flood duration and volume are best fitted by the generalized extreme value distribution while peak flow by the Generalized Pareto. Inference function for margin (IFM) method was applied to model the joint distributions of correlated flood variables for each pair and the results showed that all the calculated θ values were in acceptable range of Gaussian Copula. By horizontally cutting the joint cumulative distribution function (CDF), a set of contour lines were obtained for Gaussian Copula which represented the occurrence probabilities for the joint variables. Also the joint return period for pair of flood variables was calculated.
Modeling of rainfall-runoff relationship in a watershed is of prime importance to water resources... more Modeling of rainfall-runoff relationship in a watershed is of prime importance to water resources engineers and hydrologists in designing hydraulic structures, flood control plans and stormwater management strategies. The study site, Sungai Johor watershed, has annual precipitation of about 2500 mm. This study aimed at developing an Artificial Neural Network (ANN) model for annual peak flow during 1980-2010 without employing exogenous runoff-generating process variables. ANNs have been known as having the ability to model nonlinear mechanisms. In the present study, the sensitivity of input data, namely the initial discharge, average temperature, average evaporation, average wind speed, and average relative humidity were also evaluated. Multi layer perceptron (MLP) network was chosen for modeling the annual flood/peakflow. The data was divided into three sets, notably data for training, cross-validation and testing. The data analysis process involved cleansing, normalization and data division. Next, for the best architecture, the behavior of the input data was assessed separately. Results showed that the most sensitive input data were the initial discharge, relative humidity and temperature. The best architecture was obtained by neurons 3-9-1 (input-hidden-output layer). This was computed by using the linear tangent hyperbolic axon as transfer function. The best learning algorithm for the hidden and output layers were conjugate gradient and momentum (back propagation) respectively. For this study the coefficient of determination (R2), Nash and Sutcliffe (NS) and Root Mean Square error (RMSE) for validation period were 0.64, 0.6 and 48.9 (m3/s) respectively. This study revealed that the application of various inputs data together did not significantly improve the modeling performance in ANN. The use of exogenous variables such as initial flows can be beneficial for primary evaluation when there is significant missing data or when the data accuracy is questionable.
Uploads
this study, a hybrid network presented as a feedforward modular neural network (FF-MNN) has been developed to predict the
daily rainfall-runoff of the Roodan watershed at the southern part of Iran. This FF-MNN has three layers—input, hidden, and
output.The hidden layer has two types of neural expert or module. Hydrometeorological data of the catchment were collected for
21 years. Heuristicmethod was used to develop the MNN for exploring daily flow generalization. Two training algorithms, namely,
backpropagation with momentum and Levenberg-Marquardt, were used. Sigmoid and linear transfer functions were employed
to explore the network’s optimum behavior. Cross-validation and predictive uncertainty assessments were carried out to protect
overtiring and overparameterization, respectively. Results showed that the FF-MNN could satisfactorily predict streamflow during
testing period. The Nash-Sutcliff coefficient, coefficient of determination, and root mean square error obtained usingMNN during training and test periods were 0.85, 0.85, and 39.4 and 0.57, 0.58, and 32.2, respectively.The predictive uncertainties for both periods were 0.39 and 0.44, respectively. Generally, the study showed that the FF-MNN can give promising prediction for rainfall-runoff relations.
engineers and hydrologists in designing hydraulic structures, flood control plans and stormwater
management strategies. The study site, Sungai Johor watershed, has annual precipitation of about
2500 mm. This study aimed at developing an Artificial Neural Network (ANN) model for annual
peak flow during 1980-2010 without employing exogenous runoff-generating process variables.
ANNs have been known as having the ability to model nonlinear mechanisms. In the present
study, the sensitivity of input data, namely the initial discharge, average temperature, average
evaporation, average wind speed, and average relative humidity were also evaluated. Multi layer
perceptron (MLP) network was chosen for modeling the annual flood/peakflow. The data was
divided into three sets, notably data for training, cross-validation and testing. The data analysis
process involved cleansing, normalization and data division. Next, for the best architecture, the
behavior of the input data was assessed separately. Results showed that the most sensitive input
data were the initial discharge, relative humidity and temperature. The best architecture was
obtained by neurons 3-9-1 (input-hidden-output layer). This was computed by using the linear
tangent hyperbolic axon as transfer function. The best learning algorithm for the hidden and
output layers were conjugate gradient and momentum (back propagation) respectively. For this
study the coefficient of determination (R2), Nash and Sutcliffe (NS) and Root Mean Square error
(RMSE) for validation period were 0.64, 0.6 and 48.9 (m3/s) respectively. This study revealed that
the application of various inputs data together did not significantly improve the modeling
performance in ANN. The use of exogenous variables such as initial flows can be beneficial for
primary evaluation when there is significant missing data or when the data accuracy is
questionable.
this study, a hybrid network presented as a feedforward modular neural network (FF-MNN) has been developed to predict the
daily rainfall-runoff of the Roodan watershed at the southern part of Iran. This FF-MNN has three layers—input, hidden, and
output.The hidden layer has two types of neural expert or module. Hydrometeorological data of the catchment were collected for
21 years. Heuristicmethod was used to develop the MNN for exploring daily flow generalization. Two training algorithms, namely,
backpropagation with momentum and Levenberg-Marquardt, were used. Sigmoid and linear transfer functions were employed
to explore the network’s optimum behavior. Cross-validation and predictive uncertainty assessments were carried out to protect
overtiring and overparameterization, respectively. Results showed that the FF-MNN could satisfactorily predict streamflow during
testing period. The Nash-Sutcliff coefficient, coefficient of determination, and root mean square error obtained usingMNN during training and test periods were 0.85, 0.85, and 39.4 and 0.57, 0.58, and 32.2, respectively.The predictive uncertainties for both periods were 0.39 and 0.44, respectively. Generally, the study showed that the FF-MNN can give promising prediction for rainfall-runoff relations.
engineers and hydrologists in designing hydraulic structures, flood control plans and stormwater
management strategies. The study site, Sungai Johor watershed, has annual precipitation of about
2500 mm. This study aimed at developing an Artificial Neural Network (ANN) model for annual
peak flow during 1980-2010 without employing exogenous runoff-generating process variables.
ANNs have been known as having the ability to model nonlinear mechanisms. In the present
study, the sensitivity of input data, namely the initial discharge, average temperature, average
evaporation, average wind speed, and average relative humidity were also evaluated. Multi layer
perceptron (MLP) network was chosen for modeling the annual flood/peakflow. The data was
divided into three sets, notably data for training, cross-validation and testing. The data analysis
process involved cleansing, normalization and data division. Next, for the best architecture, the
behavior of the input data was assessed separately. Results showed that the most sensitive input
data were the initial discharge, relative humidity and temperature. The best architecture was
obtained by neurons 3-9-1 (input-hidden-output layer). This was computed by using the linear
tangent hyperbolic axon as transfer function. The best learning algorithm for the hidden and
output layers were conjugate gradient and momentum (back propagation) respectively. For this
study the coefficient of determination (R2), Nash and Sutcliffe (NS) and Root Mean Square error
(RMSE) for validation period were 0.64, 0.6 and 48.9 (m3/s) respectively. This study revealed that
the application of various inputs data together did not significantly improve the modeling
performance in ANN. The use of exogenous variables such as initial flows can be beneficial for
primary evaluation when there is significant missing data or when the data accuracy is
questionable.
flood and managing storm water. Artificial Neural Networks (ANNs) are known as having the ability to model nonlinear
mechanisms. This study aimed at developing a Generalized Feed Forward (GFF) network model for predicting annual
flood (depth) of Johor River in Peninsular Malaysia. In order to avoid over training, cross-validation technique was
performed for optimizing the model. In addition, predictive uncertainty index was used to protect of over parameterization.
The governing training algorithm was back propagation with momentum term and tangent hyperbolic types was used as
transfer function for hidden and output layers. The results showed that the optimum architecture was derived by linear
tangent hyperbolic transfer function for both hidden and output layers. The values of Nash and Sutcliffe (NS) and Root
mean square error (RMSE) obtained 0.98 and 5.92 for the test period. Cross validation evaluation showed 9 process
elements is adequate in hidden layer for optimum generalization by considering the predictive uncertainty index obtained
(0.14) for test period which is acceptable.
engineers and hydrologists in designing hydraulic structures, flood control plans and stormwater
management strategies. The study site, Sungai Johor watershed, has annual precipitation of about
2500 mm. This study aimed at developing an Artificial Neural Network (ANN) model for annual
peak flow during 1980-2010 without employing exogenous runoff-generating process variables.
ANNs have been known as having the ability to model nonlinear mechanisms. In the present
study, the sensitivity of input data, namely the initial discharge, average temperature, average
evaporation, average wind speed, and average relative humidity were also evaluated. Multi layer
perceptron (MLP) network was chosen for modeling the annual flood/peakflow. The data was
divided into three sets, notably data for training, cross-validation and testing. The data analysis
process involved cleansing, normalization and data division. Next, for the best architecture, the
behavior of the input data was assessed separately. Results showed that the most sensitive input
data were the initial discharge, relative humidity and temperature. The best architecture was
obtained by neurons 3-9-1 (input-hidden-output layer). This was computed by using the linear
tangent hyperbolic axon as transfer function. The best learning algorithm for the hidden and
output layers were conjugate gradient and momentum (back propagation) respectively. For this
study the coefficient of determination (R2), Nash and Sutcliffe (NS) and Root Mean Square error
(RMSE) for validation period were 0.64, 0.6 and 48.9 (m3/s) respectively. This study revealed that
the application of various inputs data together did not significantly improve the modeling
performance in ANN. The use of exogenous variables such as initial flows can be beneficial for
primary evaluation when there is significant missing data or when the data accuracy is
questionable.