Printer Friendly

Evaluation of landuse and rainfall on zayanderood river water quality using ANN.


Rivers have changed the surface of the earth. Accumulation of ecological and human populations along the rivers is one of these changes change. Generally, due to soil fertility, available water for irrigation, industry and drinking, and human gatherings and concentration, River basins are areas with high concentration of pollution [1].The amount of salts is rising in the Rivers and in most Rivers it has risen to 2 - 7 times of the initial salts in average. Nowadays, pollution in water resources in many parts of the world has resulted in useable condition [2]. Currently, water pollution, threaten human health and life and other organisms on the earth. Salinization fresh water resources as a result of indiscriminate harvesting and reduced rainfall is the main factor of water pollution. Drought is one of natural disasters which has very harmful effects on the ecological environment [3].

(ANN) is a computational method that with the help learning process and using processors called neurons, tries to make a connection between input and output layer through recognizing the inherent relationships between the data [4]. Hidden layer or layers process the information received from the input layer and transfer it to the output layer. Each network is trained with examples. Training is a process that will ultimately cause learning. Learning the network occurs when the weights between the layers change in a way that differences between the predicted and calculated values are acceptable. Achieving to these conditions results in the learning process. These weights show memory and knowledge of the networks. Trained neural network can be suitable for predicting outputs with new set of data with regard to the (ANN) structure, its main features, high processing speed, learning ability, the ability to generalize the knowledge after learning, flexibility in unwanted errors against and lack of a significant the knowledge disruption in case of errors in connecting parts of the network is due to the weight distribution [5].

In recent years, several models to estimate water quality have been presented since the water quality depends on many factors, usually the information and data for estimation of water quality do not exist, therefore the use of artificial neural networks is suitable for prediction and estimation of water quality [6].

Many researches have been conducted in the world about the neural networks, predicting rainfall, groundwater modeling and predictingRiver flow can be referred to in this regard (Talib, 2009) [7]. among them, the following can be mentioned: salinity in rivers [8] benthic communities [9] in streams, eutrophication and algal blooms in lakes [10,11] predict rainfall, groundwater modeling and predict river flow [7].

Maier and Dandy [8], in a study of estimate salinity of Mary River with using (ANN), concluded that ANN model is a precise instrument in the estimation of the river quality decrease, and differences between observed and simulated values are between 46 to 53 [micro]mhos per centimeter. They suggested that a comparison be conducted between these models with other physical or mathematical models [8].

Ha [12] estimated the daily dissolved oxygen (DO) in Tualatin River during the years 2000-1991 and successfully predicted the variation with 0.9 mg/L error and developed the network for other parameters such as the biological oxygen demand (BOD) and chemical oxygen demand required (COD), ammonia nitrogen and other cases[12].

Ying [6] found ANN a suitable method in the prediction of pollution (BOD, DO) in basin of Yuqiao China, which was fast and easy with few errors [6].

Mohsenifar et al. [13] simulated the effect of rainfall on the pollution of Leng station in Zayandehrud River with ANN with correlation coefficientof 0.99 [13].

Several researches have been done on water pollution sources and the following factors have been found to affect rivers.A) Landuse and land cover[14,15,16] B) River discharge [17,18] C) rainfall [19]

The main objective of this study is to show and model pollutant (landuse and rainfall) and polluting (Ca, Mg, Na and Etc.) factors in the Zayanderood River with the use of GIS and ANN.

Materials and Methods

ZayanderoodRiver is the only central river in Iran which originates in the West from ZardkoohBakhtiyari (50.130 - 32.454 degrees longitude and latitude respectively) in Chaharmahal&Bakhtiyari province, [20] and discharges in the Gavkhoni swamp East of Esfahan (52.892 longitudes and 32.244 degrees latitude) and is 400.580 km long. Zayanderood River basin, area is 41533.457 km, with a height range of 1470 to 3974 meters and includes aridic to very wet climates.(Fig.1).

The river basin divided to 12 sub basins using the Mapinfo and ArcGIS software. (Table1)

Table 2 shows the positions of the stations along the Zayanderood based on which the following classification has been done.

Then various landuses were specified (based on Water Organization of cheharmahal&bakhtiyari and Esfahan provinces) in each sub-basin and the whole basin. Table 3 shows different landuses and the area for total basin.

It should be mentioned that only data with satellite images in this yare (2005) existed and that is why this year was analyzed. The daily rainfall was collected.

Data on daily rainfalls of these stations in 2005 from the Water provinces of Isfahan and Chaharmahal&Bakhtiyari were converted to monthly data of the rainfall [13].

Data about water including pH, EC, Cl, CO3, SO4, HCO3, K, Na, Mg, Ca, SAR and TDS in hydrometric stations related to the 2005 year were collected.

The most important factors among them are: 1) type and area of land use 2) rainfall

Therefore, these two factors was determined as input and each water quality parameters of Zayanderood river (includes: pH, EC, Cl, CO3, SO4, HCO3, K, Na, Mg, Ca, SAR, TDS and discharge) were considered as output. (Fig.2).

After normalized the data, they were entered in network using the Matlab software 2009. With change of the number of neurons and layers, the Mean Square Error (Eq. 1), correlation coefficient ([R.sup.2]) and error (Eq. 2) for each parameter were calculated.

MSE=1/n3([y.sub.j](t)-[d.sub.j][(t)).sup.2] (1)

[E.sub.t] = 1/2 [SIGMA]([y.sub.j](t)-[d.sub.j][(t)).sup.2] (2)


Et: is Error function at the time t, MSE: Is Mean Square Error, q: Is number of neurons in output layer (S2), j: Is number of neurons in yj(t), yj(t): Is real values in the output layer at the time t, dj(t): Is estimated values by the ANNs in the output layer and n: is number of inputs or outputs[21,6].

To determine the effects of land use and rainfall on the pollution of the Zayanderood river MLP (Multilayer Perceptron) network was used. The network includes an input layer, one or more hidden layers and an output layer. For training the network was used Multilayer perceptron (BP).In training of the Network, BP was used. In training MLP Network with learning algorithm BP, first was calculation of the input network to the output network was done. Then the calculated error values were applied to the prior layers.

Learning procedures using this algorithm are as follows:

A) Assigning random weight matrix to each connection

B) Selecting the appropriate input and output

C) Calculating the output neurons in each layer, and thus calculating the output neurons in the output layer

D) updating weights with propagation error network method in which the error is due to the difference is the actual output and the calculated output.

E) Evaluating the performance of the trained network with some per-defined factors such as Root Mean Squared Error, and finally returning to the final part of training.

F) Selecting the appropriate input and output [6].



The model used in this paper is a BP neural network model with a single hidden layer. In Figure 3.shows the structure of an ANN with a single layer. Where R: input layer, S1: the hidden layer, [S.sup.2]: the output layer, [IW.sup.1.1]: the weight matrix of the input layer, [LW.sup.2.1]: the weight matrix from the hidden layer to the output layer, [b.sup.1] and [b.suP.2] threshold values of the hidden and output layers respectively and [f.sup.1 ]and [f.sup.2] the neuron transfer functions of the hidden and output layers respectively [6]

After determining the area of 18 different land uses and the amount of rainfall in the 12 sub-basins monthly, the total number of 2736 as input and 13 pollutants in 12 river stations in the total number as 1872 output network were entered to the neural network. Of these 60 percent were used as the training network and the remaining 40 percent were used for testing the network.

Result and discussion

Fig 4 illustrates the correlation coefficient of the network for training, test and the whole network. According to this figure, correlation coefficient for the three parameters pH, CO3, and discharge are low (39.38, 31.34 and 6553 respectively), but for the other outputs regression coefficient is above 80 percent.

It seems this isdue to low changes of CO3 all along the river. In fact land use did not affect the two factors pH and CO3. Mas (2007) using artificial neural network did modeling pollution in the Charles River watershed in Massachusetts[22]. He pointed to the limited research on the use of ANNs for predicting indicator organism concentrations in freshwater and surface waters under dry weather conditions. The results show that, in this watershed setting, ANNs were unable to correctly generate precise predictions. And modeling results show that ANNs are the weakest at predicting extreme values, but are capable of correctly classifying indicator organism concentrations relative to a threshold.

Based on researches by Hafizan (2004) for Dissolved Oxygen (DO), Biochemical Oxygen Demand (BOD), Suspended Solids (SS) and Ammoniacal-Nitrate (AN). (R=0.735, 0.786, 0.809, 0,821 respectively), Najah, A. (2009) for EC, TDS and Turbidity (r=0.767, 0.767 and 0.91 respectively) and Talib, A. (2009) for monitoring a month data of Pulau Pinang River for BOD (R=0.72), have been done. They point that the ANN is a suitable method for predicating water pollution.

But in this research they selected the whole Zayanderood basin and 12 hydrology stations in the river side and considered landuse as the most important factor on the river pollution. Any change on the landuse of the river basin, can effect elements of river. The dam and factories along the river, are effective on discharge and increase of the elements in water of the river. Because these two factors are not considered in the network, it cannot naturally establish relationship whit river pollution. Also, because pH changes are low in the long of the river and pH is more affected by lithology of basin, ANN dose not establish a relationship between these two factors.

So EC, Cl, SO4, K, Na, Mg, Ca, SAR and TDS factors, have direct effect on landuse and rainfall, ANN has been simulated with high accuracy in this study.

Fig. 5 shows changes of these two factors throughout the glen of river. But regarding the discharge, because its value depends on the dam rather than the rainfall, discharge is not a function of land use and rainfall. Figure 6 shows the position of Zayanderood dam in the Zayanderood River Basin.


Fig. 7 shows the MSE. Value of pH, Cl and discharge is high, (0.102, 0.162 and 0.078 respectively)Error related to the actual and simulation values were calculated and are shown in the fig 8 and 9 in percentage. As can be seen in Figure 8 for Mg, Na, SO4, TDS and SAR there is a similar pattern. Error value until the 11th station is less than 18 percent but suddenly at the 12th station, it increases to 100 percent. The reason is the discharge of agriculture and municipal (urban) sewage of Isfahan into the river. From the 1th station to the 7th station the error value is less than 2 percent, but at the 8th station error reaches to twice that error at least, and increase to 4 percent. Because at the 8th Station is Zobahan factory and it discharges the sewage into the river. But for discharge, pH, HCO3, and K, the percentage error for most stations is high (Fig. 9). For discharge until station number 3, error value is 26 percent at maximum, but in station No. 4 it increases to 100 percent because the Zayanderood dam is in distance and discharge is not affected by the rainfall rate but on the dam.

Land use has a direct relation with river Pollution and for factors (EC, Cl, SO4, K, Na, Mg, Ca and TDS), correlation coefficient is over the 0.81 in Zayanderood River basin. As Sandhu and Phinch (1995), Maier and Dandy (1996) and Noroziyan[23](1380) for EC, Sandhu and Phinch (1995) for TDS and ShadmanMotlagh (2007) for runoff of discharge, was simulated the ANN with R up to 98% [24,17].





Many factors affect the river pollution; landuse is most important one which is considered in this study. However, it is suggested that other factors such as geology, soil texture and etc. Also be studied.

Summary and Conclusion:

With attention to the effective factors on River pollution as in this article study, ANN is a suitable factor for predication of water quality of the Zayanderood River.

More facts EC, Cl, SO4, K, Na, Mg, Ca and TDS (R>81%) high dependency, and discharge, pH, HCO3, and K factors are less dependent on variety of land use and rainfall in the catchment considered.

In the end stations due to industrial pollutants entry (large industries: Isfahan melt steel and Mobarakeh steel) and urban (Isfahan city), reduced this correlation.

In addition to review factors in this article, also it is necessary to consider the effect of other factors such as soil texture, amount of industrial pollution.




[1.] Hafizan, J., 2004. Application of artificial neural network models for predicating water quality index, Journal Kejuruteraan Awam, 16(2): 42-55.

[2.] Mohammadi, S., 2007. Drought, effects and struggle manner with it. First regional Conference optimal utilization of water resources and watersheds ZayandehrudKarun. Shahrekord University, Iran, 796.

[3.] Foth, D. Henry, 1998. Fundamentals of soil sciences, New York.

[4.] Dayhoff, J.E., 1990. Neural Network Principles, Prentice-Hall International, U.S.A.

[5.] Menhaj, M.B., 2007. Fundamental of Neural Networks.

[6.] Ying, Z., 2007. Water quality forecast through application of BP neural network at Yuqiao reservoir, Journal of Zhejiang University SCIENCE, 8(9): 1482-1487.

[7.] Talib, A., 2009. Predicting Biochemical Oxygen Demand as Indicator of River Pollution Using Artificial Neural Networks. 1<M World IMACS, MODSIM Congress, Cairns, Australia, 824.

[8.] Maier, H.R., G.C. Dandy, 2000. The use of artificial neural networks for the prediction of water quality parameters, Water Resource Research, 32(4): 1013-1022.

[9.] Chon, T.S., Y.S. Park, J.H. Park, 2000. Determining temporal pattern of community dynamics by using unsupervised learning algorithms, Ecological Modelling, 132: 151-166.

[10.] Karul, C., S. Soyupak, C. Yurteri, 1999. Neural network models as a management tool in lakes, Hydrobiologia, 408(409): 139-144.

[11.] Recknagel, F., 2003. Understanding Ecology By Biologically Inspired Computation, Ecological Informatics, Springer Verlag, 233.

[12.] Ha, H., M.K. Stenstorm, 2003. Identification of land use with water quality data in storm water using a neural network". Water Research, 37: 4222-4230.

[13.] Mohsenifar, K., S.H. Tabatabaei, et al., 2010. Effect of drought on pollution of LENJ station of Zayndedood River by Artificial neural network XVIIth World Congress of the International Commission of Agricultural Engineering (CIGR), CSBE/SCGAB, Quebec City, Canada, 341.

[14.] Massoud, M., M. El-Fadel, M.D. Scrimshaw, J.N. Lester, 2006. Factors influencing development of management strategies for the Abou Ali River in North Lebanon I: Spatial variation and land use, Science of the Total Environment, 362: 15-30.

[15.] Anguelov, G., 2009. Assessment of landuse effect on trace elements concentrations in soil solution from Ultisols in North Florida. Agriculture, Ecosystems and Environment, 130: 59-66.

[16.] Ouyang, W., 2008. Regional nonpoint source organic pollution modeling and critical area identification for watershed best environmental management. Water Air Pollute, 187: 251-261.

[17.] Shadman, A., Motlagh, 2007. Flood forecasting of Sefidrood River using Artificial Neural Networks. 6th hydraulic conference.Shahr-kord, Iran, 103.

[18.] Erfanmanesh, M., M. Afuni, 2008. Water, Soil and Air pollution, Propagation of Esfahan Industrial University, Iran, 89.

[19.] Abedikohpaie, J., M. Bagheri, 2001. Biological Effect of Irrigation with purification swage on underground sources water, Biological Effect agriculture swage on surface & underground water congress, Persian, Iran, 138.

[20.] Society, J., 1999. Precise Planning water in Iran country & Zaynderood basin. The ministry of power, 1(37).

[21.] Erenturk, S., K. Erenturk, 2007. Comparison of genetic algorithm and neural network approaches for the drying process of carrot, Journal of Food Engineering, 78: 905-912.

[22.] Mas, D. and D. Ahlfeld, 2007. The development and evaluation of ANNs for modeling indicator organism concentration, Marston Hall, Amherst MA, USA, Wat. Sci. Tech., 43(12): 125-132.

[23.] Noroziyan, K., M. Tajrichi, 2001. Forecasting River of water quality with classification of phasyanalysis, Esteghlal Journal, Persian, Iran, 14.

[24.] Sandhu, N., R. Finch, 1995. Methodology for Flow and Salinity Estimates in the SacramentoSan Joaquin Delta and Suisun Marsh, Annual Progress Report, 16(4).

[25.] Najah, A., 2009. Prediction of Johor River Water Quality Parameters Using Artificial Neural Networks. European Journal of Scientific Research.ISSN 1450-216X, 28(3): 422-435.

(1) Kamran Mohsenifar, (2) Ebrahim Pazira, (3) Sayed Hassan Tabatabaei, (4) Payam Najafi, (5) Nabiallah Yarali.

(1) Ph.D Student, Department of Soil Science, Islamic Azad University, Tehran Science and ResearchBranch.Tehran. Iran.

(2) Professor, Department of Soil Science, Faculty of Agriculture, Islamic Azad University, Science and Research Branch, Tehran, Iran.

(3) Assistant Professor, Department of Water Engineering, Faculty of Agriculture, Shahrekord University, Shahrekord, Iran.

(4) Assistant Prof. IAU, Khorasgan Branch, Isfahan, Iran.

(5) Assistant Professor, Department of Forest Science, Faculty of Agriculture, Shahrekord University, Shahrekord, Iran.

Kamran Mohsenifar, Ebrahim Pazira, Sayed Hassan Tabatabaei, Payam Najafi, Nabiallah Yarali; Evaluation of landuse and rainfall on Zayanderood river water quality using ANN

Corresponding Author

Kamran Mohsenifar, Ph.D Student, Department of Soil Science, Islamic Azad University, Tehran Science and ResearchBranch.Tehran. Iran.

Table 1: Area of Zayanderoodsub basin in each hydrometric station

Row SubBasin name Area ([km.sup.2])

1 Toonel 1 11.637
2 Toonel2 21.280
3 CheshmeDimeh 300.844
4 GhaleShahrokh 1431.903
5 Sad Tanzimi 4022.513
6 Pol Zamankhan 4430.925
7 Pol kalle 5357.822
8 Diziche 8014.828
9 Lenj 8084.196
10 Mosiyan 11368.60
11 Pol Choom 20974.53
12 Varzaneh 41533.46

Table 2: Location of stations in ZayanderoodRiver

No. Stations longitude latitude

1 Toonel1 50.130 32.456
2 Toonel2 50.151 32.457
3 CheshmeDimeh 50.216 32.501
4 GhaleShahrokh 50.458 32.658
5 Sad Tanzimi 50.787 32.715
6 Pol Zamankhan 50.894 32.498
7 Pol kalle 51.230 32.373
8 Diziche 51.519 32.373
9 Lenj 51.563 32.392
10 Mosiyan 51.525 32.576
11 Pol Choom 51.77 32.585
12 Varzaneh 52.647 32.427

Table 3: Different landuses and area of Zayanderoodriver basin

Row Land Use Area [Km.sup.2]

1 pasturedense 1453.953
2 Semi-dense pastures 2016.097
3 Water farming 5515.327
4 grassland 337.347
5 Garden land 8.279
6 Forest with moderate canopy 0.006
7 Dry farming 482.570
8 Low-dense pasture 22275.39
9 Urban 870.572
10 Fallow 116.794
11 Forest with low canopy 12.299
12 Surface water 91.600
13 Canebrake 50.919
14 Rocky outcrops 5914.358
15 Saline soil 322.732
16 Desert 1514.48
17 Swampland 487.054
18 Sandy area 64.376

Fig. 1: Location of ZayanderoodBasin and stations.

No. Stations

1 Toonel 1
2 Toonel2
3 CheshmeDimeh
4 GhaleShahrokh
5 Sad Tanzimi
6 Pol Zamankhan
7 Pol kalle
8 Diziche
9 Lenj
10 Mosiyan
11 Pol Choom
12 Varzaneh
COPYRIGHT 2011 American-Eurasian Network for Scientific Information
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2011 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:Original Article; artificial neural networks
Author:Mohsenifar, Kamran; Pazira, Ebrahim; Tabatabaei, Sayed Hassan; Najafi, Payam; Yarali, Nabiallah
Publication:Advances in Environmental Biology
Article Type:Report
Geographic Code:7IRAN
Date:Mar 1, 2011
Previous Article:Sunflower tissue culture.
Next Article:Effect of canola oil on mucosal alkaline phosphatase activity enzymes in small intestine of Turkey chicks.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters