BleedLines

Trim Trim

BleedLines 4.75" (+.031)

120.65mm (±.8mm)

Fold Line

4.75" (+.031)120.65mm (±.8mm)

4.73

"(±

.031

)12

0.14

mm

(±.8

mm

)

Trim TrimFold Line

This inside panel is reserved for installation instructionswhich Omnipress will customize for your project.

This panel is the front of your jewelcase.

BleedLines

BleedLines

OPENING SCREEN

Examining the Confluence ofEnvironmental and Water Concerns

World Environmental andWater Resources Congress2006

• Proceedings Information• Contents

• Author Index• Search

All rights reserved. © 2006 American Society of Civil Engineers

EXAMINING THE CONFLUENCE OF

ENVIRONMENTAL AND WATER

CONCERNS

PROCEEDINGS OF THE 2006 WORLD ENVIRONMENTAL AND WATER RESOURCES CONGRESS

May 21-25, 2006 Omaha, Nebraska

SPONSORED BY Environmental and Water Resources Institute (EWRI)

of the American Society of Civil Engineers

COOPERATING ORGANIZATIONS American Academy of Environmental Engineers (AAEE) International Association for Hydraulic Research (IAHR)

EDITED BY Randall Graham, P.E., M.ASCE

Published by the American Society of Civil Engineers

Notices

ISBN 0-7844-0856-4

Any statements expressed in these materials are those of the individual authors and do not necessarily represent the views of ASCE, which takes no responsibility for any statement made herein. No reference made in this publication to any specific method, product, process or service constitutes or implies an endorsement, recommendation, or warranty thereof by ASCE. The materials are for general information only and do not represent a standard of ASCE, nor are they intended as a reference in purchase specifications, contracts, regulations, statutes, or any other legal document.

ASCE makes no representation or warranty of any kind, whether express or implied, concerning the accuracy, completeness, suitability, or utility of any information, apparatus, product, or process discussed in this publication, and assumes no liability therefore. This information should not be used without first securing competent advice with respect to its suitability for any general or specific application. Anyone utilizing this information assumes all liability arising from such use, including but not limited to infringement of any patent or patents.

Copyright © 2006 by the American Society of Civil Engineers. All Rights Reserved. ISBN 0-7844-0856-4 Manufactured in the United States of America. American Society of Civil Engineers ASCE International Headquarters 1801 Alexander Bell Drive Reston, VA 20191-4400 USA Call Toll-Free in the U.S.: 1-800-548-2723 (ASCE) Call from anywhere in the world: 1-703-295-6300 Internet: http://www.pubs.asce.org

Preface

The world community has begun to recognize and to look for answers to reverse adverse impacts to the environmental and water resource systems induced by an expanding human population. Yet this expansion is still requiring and placing additional demands on our limited environmental and water resources, resulting in their continued degradation. This constant challenge of meeting the human population needs while finding the means to restore our environmental and water resources obligates the civil engineering and scientific professionals to better understand and develop methods and techniques to find a successful compromise between the using and the replacement or restoration of our resources. The manuscripts within these Proceedings address the various hydrologic, climatologic, hydraulic, subsurface, and other disciplines important in the analysis and addressing of our depleting environmental and water resources. The manuscripts in this volume provide keys to assist the environmental professional in daily practice, whether within the academic, governmental or industry arenas. The Proceedings of the 2006 World Environmental and Water Resources Congress, held in Omaha, Nebraska on May 22–25, 2006 covers a wide spectrum of topics important to water resources and environmental engineering professionals. Thanks to the dedicated work of many EWRI Councils, Committees, and members, the Proceedings contains papers from the General Conference portion of the Congress that address a wide variety of the most crucial topics facing water professionals today. Areas of interest to water professionals including:

• Irrigation and Drainage • Urban and Natural Watershed Management • Hydrology • Sustainable Development in Water, Wastewater and Stormwater • Hydraulics; Structures, Waterways and Measurement & Experimental Methods • Hydrology • Adaptive Management in Water and Natural Resources • Planning and Management • Climate, Meteorology and Water Resources • Computational Hydraulics • Environmental Processes • Evolutionary Computations • ASCE/EWRI Standards • Education and Research • Groundwater Hydrology, Quality and Management • International Water Resources Issues • River and Wetlands Restoration • Applied Research in Water, Wastewater and Stormwater

The Editors have accepted the best papers of the approximately 800 abstracts initially submitted to be included in these proceedings. All of these papers are eligible for further discussion in the appropriate ASCE journal and are eligible for ASCE awards.

1

Artificial Neural Network approach for streamflow forecasting in India usingENSO and EQUINOO

Rajib Maity1 and D Nagesh Kumar2

1 Research Scholar, PH +91 80 2293 2673; e-mail: rajib@civil.iisc.ernet.in2 Associate Professor, PH +91 80 2293 2666; e-mail: nagesh@civil.iisc.ernet.inWater Resources and Environmental Engineering, Department of Civil EngineeringIndian Institute of Science, Bangalore – 560 012, Karnataka, India;FAX: +91 80 2360 0404

Abstract: It is well recognized that hydrologic time series, like rainfall, streamflowetc. are modulated by various large-scale atmospheric circulation patterns, like ElNiño Southern Oscillation (ENSO) etc., throughout the globe by climaticteleconnection. Indian summer monsoon rainfall (ISMR) is proved to be significantlyinfluenced by ENSO. Recently, it is established that relationship between ISMR andENSO is modulated by the influence of atmospheric circulation pattern over IndianOcean region, like, Equatorial Indian Ocean Oscillation (EQUINOO). Thus, forIndian subcontinent, hydrologic time series are significantly influenced by ENSOalong with EQUINOO. Though influence of these large-scale atmosphericcirculations on rainfall is investigated, their influence on streamflow is yet to beinvestigated. In this paper, information of ENSO from tropical Pacific Ocean andEQUINOO from tropical Indian Ocean are used in terms of their correspondingindices for streamflow forecasting. To model the nonlinear, time-varying anddynamic relationship between streamflow and such large-scale atmosphericcirculation information, Artificial Neural Network (ANN) methodology has beenopted for the present study. Efficient optimization of ANN architecture is obtained byusing evolutionary optimizer based on genetic algorithm. This study concludes thatuse of such large-scale atmospheric circulation information potentially improves theperformance of streamflow prediction which, in turn, helps in better management ofwater resources.

1. Introduction

Prediction of streamflow is a very important aspect for water resources management.It helps in flood control, devising agriculture strategy, reservoir operation etc. Use ofthe inherent statistical properties of time series for prediction is practiced widely.

2

Recently, it is understood that temporal structure of hydrologic time series issignificantly influenced by large scale atmospheric circulation (Jain and Lall, 2001).However, it is scientifically and mathematically challenging to use such signal inhydrologic time series prediction. Moreover, for smaller spatio-temporal scale, suchlarge scale atmospheric influence is yet to be established for the developing countries,like, India. In this study, it is aimed to investigate such influences on streamflow,through hydroclimatic teleconnection for a catchment in India for monthly andseasonal time scale. The catchment, selected for case study, is located in the state ofChhatishgarh in India. The streamflow is measured at Basantpur which is a fewkilometers upstream of Hirakud reservoir, which is located in Orissa and one of themost important multipurpose reservoirs in India (Figure 1). Thus reliable predictionof streamflow at this site will be a valuable input for efficient operation of Hirakudreservoir.

Figure 1. Catchment and sub-basins of Mahanadi River and location of Basantpur.

Before proceeding further, a brief description of two large scale atmosphericcirculations along with the physical mechanism, that influence the temporal structureof hydrologic time series in India, is presented.

1.1. El Niño and Southern Oscillation (ENSO)

El Niño Southern Oscillation (ENSO) is the coupled Ocean-atmosphere mode oftropical Pacific Ocean (Cane, 1992). It is a large scale anomalous warming of seasurface temperature (SST) over central and eastern Pacific Ocean with associatedchange in pressure field. In normal years, SST of the western part of equatorial

3

Pacific Ocean remains warmer than that of eastern part and pressure at the easternpart of Pacific Ocean is higher than that of western part. During anomalous years,SST of the eastern part of equatorial Pacific Ocean becomes warmer-than-normal andthe pressure field is reversed, i.e., the anomalous pressure in the eastern part of thePacific Ocean becomes lower than that of western part. A contrary situation may alsooccur. Anomalous warming (cooling) of SST over eastern part of Pacific Ocean isknown as El Niño (La Niña) whereas anomalous sea-saw variation of the pressurefield, between eastern and western parts of Pacific Ocean, is called the SouthernOscillation. Acting together, the oceanic and atmospheric part is jointly known as ElNiño Southern Oscillation (ENSO) phenomenon. It is established that Indian summermonsoon rainfall (ISMR) is significantly related to ENSO through Oceanic-Atmospheric teleconnection (Rasmusson and Carpenter, 1983; Parthasarathy et al.,1988). However, contrary to the long recognized negative correlation between theISMR and ENSO, some discrepancies have been observed in recent years (Kane,1998; Krishna Kumar et al., 1999; Li et al., 2001; Gadgil et al., 2004). Theseunanticipated experiences suggest that the response of monsoon to El Niño is not yetassessed adequately (Gadgil et al., 2003; Gadgil, 2003) or more importantly thatthere are some other causative climate forcing events which are also influencing theIndian rainfall concurrently.

1.2. Equatorial Indian Ocean Oscillation (EQUINOO)

Before describing Equatorial Indian Ocean Oscillation (EQUINOO), it is necessary todefine Indian Ocean Dipole (IOD). Indian Ocean Dipole (IOD) mode is a pattern ofinternal variability with anomalously low sea surface temperatures off Sumatra andhigh sea surface temperatures in the western Indian Ocean, with accompanying windand precipitation anomalies (Saji et al., 1999). There are two coupled components ofIOD – Oceanic and Atmospheric. Dipole mode index (DMI) is the oceaniccomponent (Saji et al., 1999) whereas EQUINOO is the atmospheric component ofthe Indian Ocean Dipole (IOD) mode (Gadgil et al., 2003, 2004). The convectionover the eastern part of the equatorial Indian Ocean (EEIO, 90° – 110° E, 10° S – 0°)is negatively correlated to that over the western part of the equatorial Indian Ocean(WEIO, 50° – 70° E, 10° S – 10° N). The anomalies in the sea level pressure and thezonal component of the surface wind along the equator are consistent with theconvection anomalies. When the convection is enhanced (suppressed) over theWEIO, the anomalous surface pressure gradient is towards the west (east) so that theanomalous surface wind along the equator becomes easterly (westerly). Theoscillation between these two states is known as EQUINOO (Gadgil et al., 2003,2004). Equatorial zonal wind index (EQWIN) is considered as an index ofEQUINOO which is defined as the negative of the anomaly of the zonal componentof surface wind in the equatorial Indian Ocean region (60° – 90° E, 2.5° S – 2.5° N).Northward propagation of large scale convective system generated over Indian Oceanindicates the physical link between ISMR and EQUINOO (Gadgil et al., 2004).Recently it is observed that ISMR is influenced by both ENSO and EQUINOOseasonally (Ashok et al., 2001; Gadgil et al., 2004) as well as monthly (Maity andNagesh Kumar, 2006)

4

Having observed the link of ISMR with ENSO and EQUINOO, it is logical toinvestigate the hydroclimatic teleconnection between streamflow at different parts ofIndia and these large scale atmospheric circulation phenomena. In this study,temporal structure of streamflow at Basantpur site is modeled using ENSO andEQUINOO information in terms of their corresponding indices.

2. Data

Sea surface temperature anomaly from the region Niño 3.4 (5˚ S – 5˚ N, 120˚ – 170˚W) is used as ENSO index in this study. This data is obtained from the website ofNational Weather Service, Climate Prediction Centre of NOAA[http://www.cpc.noaa.gov/data/indices/] for the period January 1972 to December2003.

Similarly, EQWIN is used as EQUINOO index. Monthly surface wind data for theperiod, January 1972 to December 2003, are obtained from National Center forEnvironmental Prediction [http://www.cdc.noaa.gov/Datasets].

Streamflow data at Basantpur site is obtained from the office of Executive Engineer,Mahanadi Division, Central Water Commission (CWC), Burla, Orissa for the periodJanuary 1972 to December 2003.

3. Methodology

Streamflows are predicted for two different temporal resolutions – monthly andseasonal. Total streamflow during monsoon period (June to October) is considered asseasonal streamflow during monsoon period. While deciding the input variables forseasonal streamflow (or streamflow for different months), a correlation analysis isperformed and best correlated month which is prior to the season (or month) ischosen. Apart from these circulation indices, information of previous monthstreamflow is also used due to significant serial autocorrelation. However, streamflowin the month of June is not correlated with that of May. This is because there is hardlyany streamflow in May. Thus for June, previous month streamflow information is notused. On the other hand, for September, streamflow information of both July andAugust is considered, being significant. The set of input variables for differentperiods is shown in Table 1. Artificial Neural Network (ANN) is used in this study topredict the streamflow using the input variables as described above. All the inputvariables and streamflow values are scaled between 0 and 1 before using ANN.

ANN can capture the nonlinear relationship between two time series, if any, and doesnot depend on the distributional form of the data set. Details of ANN approach and itsapplication in engineering can be found elsewhere (Haykin, 1999; ASCE, 2000a;2000b). One important aspect of ANN methodology is the design of networkarchitecture. In most of the studies, network architecture is decided based onheuristic. However, in this study, evolutionary optimizer based on genetic algorithm

5

(GA) is used to decide optimum architecture of feed forward network and backpropagation algorithm (Yu and Chen, 1997) is used to train the network. Theevolutionary optimizer based on GA approach has been used in some other studiesalso (Nagesh et al., 2005; ). Design of network architecture and its training isperformed based on the data-period 1972 to 1991.

The evolutionary optimizer uses the principle of GA to obtain optimal networksarchitecture. The main steps involved in evolutionary optimizer are explained asfollows.

1. Parameters initialization: All the parameters like population size ( N ), numberof maximum generations, probability of crossover ( cP ) and probability of

mutation ( mP ) are set to specific values. In this study, N , cP and mP are selected

as 50, 0.2 and 0.04 respectively.

2. Generation of initial population: At initial generation, the evolutionaryoptimizer randomly creates N numbers networks for the initial population, whereN is equal to the size of population.

3. Training of the network and fitness evaluation: All the networks within thecurrent generation are trained by back propagation algorithm and their fitnessvalues are determined according to the goals to be achieved. These goals areattributed to the maximum number of neuron used in the network, network’smean square deviation and maximum square deviation on pattern setreproduction.

4. Evolution of new generation: A new generation of networks will be createdfrom the present generation according to the following procedure:

a. Two “parent” networks will be chosen out of the old generation. The selectionalgorithm will choose networks with a high fitness by a higher probability.

b. Two “children” networks will be created from the two “parent” networks.Using the cross over probability cP , the two “parent” networks will be crossed

over, i.e., they will swap a portion of the network with each other.

c. The “children” will be mutated with a mutation probability mP . Here, mutation

means insertion or deletion of a layer and/or insertion or deletion of a neuroninto a layer.

d. A few elitist members of the population in current generation are carried tonext generation. The selection continues until the new generation has Nmembers too. After completion the new generation will be evaluated.

5. Checking for Termination Criteria: Evolution is stopped, if the target goals areachieved at least for one network in the population or maximum number ofgenerations is reached. Otherwise, generation counter is increased by one andsteps 3 and 4 are repeated.

6

The procedure of evolutionary optimizer is applied for seasonal as well as all theindividual monsoon months. For each case, the best network in the final generation,which fulfils the target goals, is shown in Table 1. These trained networks arevalidated for the testing period, 1992 - 2003. Results are presented in the next section.

Table 1 Set of input variables for different cases

StreamflowPeriod

InputVariables

BestNetwork

CFDTP

Seasonal(June to October)

Nov -1 ENMar EQ

2--3--1 0.65

JunMar ENMar EQ

2--4--5--3--1 0.68

JulJun SFJun ENJun EQ

3--3--5--1 0.88

AugJul SFJul ENJul EQ

3--6--5--1 0.97

Sep

Jul SFAug SFAug ENAug EQ

4--3--2--1 0.87

OctSep SFSep ENSep EQ

3--5--1 0.87

Nov -1 – Previous year November; SF – Streamflow; EN – ENSOindex; EQ – EQUINOO index; CFDTP – Correlation coefficientduring training period

For seasonal streamflow, it is noticed that observed and predicted streamflows are notclose for few years. Please refer top-left panel of Figure 2. However, in ordinal scale,categorical prediction provides a better result. By categorical prediction, it is meantwhether the ensuing season will have high, medium or low streamflow. In top-leftpanel of Figure 2, two lines are shown indicating the range high, medium and lowstreamflows. These demarcations are obtained by ‘Mean + Standard Deviation’ and‘Mean - Standard Deviation’. Performance of categorical prediction is summarized inTable 2. It can be observed that categorical streamflow information for the ensuingseason can be provided by this approach with reasonable accuracy.

7

Figure 2. Prediction performances during testing period.

Prediction performances of streamflow for individual months are also shown inFigure 2. It is observed that prediction performance is reasonably good for allmonths. However, for June, it is not as good as that for other months. Reason behindthis poor performance can be attributed to high amount of variance associated withstreamflow during this month. Coefficient of variation, γ , ( µσγ = , where σ is thestandard deviation and µ is the mean) for this month, is observed to be 1.91 whereasfor the months July through October values of γ are 0.90, 0.49, 0.68 and 0.97respectively. This indicates that uncertainty associated with June streamflow is muchhigher than that with streamflow during other months. However, it is also true that thestreamflow during June is lower than other months. Thus, a bit poorer predictionperformance is not very problematic from water resources management point of view.

Table 2 Categorical prediction of seasonal streamflows

PredictedStreamflowCategory Low Medium High

Low 1 1 0

Medium 1 6 2

Obs

erve

d

High 0 0 1

8

For July, vary high streamflow (1994) as well as very low streamflow (1992, 1993and 1997) are successfully captured. However, streamflows are over predicted during2001, 2002 and 2003. Most successful prediction is observed for the month ofAugust. More or less accurate streamflows are predicted almost for all the years.Also, streamflows are predicted with reasonable accuracy for the months ofSeptember and October too except for September 2003.

Performances of streamflow prediction for the month of July, August, September andOctober was also performed using only streamflow information of previous month(s),i.e., information of large scale circulation indices are discarded. Much poorerprediction performances were obtained. This is because the autocorrelation structurevaries considerably for different parts of the streamflow series. In other words, high(low) streamflow of previous month does not always assure high (low) streamflowfor the present months. Thus, the overall observation of this study indicates thepotential power of prediction of large scale circulation indices from tropical Pacificand Indian Ocean. Use of such information improves the prediction performancewhich is very useful for better water resources management.

4. Conclusions

In this study, information of two large scale circulation phenomena, ENSO andEQUINOO from tropical Pacific Ocean and Indian Ocean respectively, issuccessfully used for prediction of streamflow at Basantpur in Orissa, India. ArtificialNeural Network (ANN) approach is shown to be a useful tool to capture the unknownrelationship between streamflow and the large scale circulation phenomena. Insteadof heuristic, evolutionary optimizer, which is used in the present study, can be used toidentify the optimum network architecture for a particular pattern set. Following thisapproach, prior information of streamflow for ensuing monsoon period can be madeavailable. Monthly streamflows are also predicted with reasonable accuracy. Thus,use of such large scale circulation indices improves the performance of streamflowprediction which is very useful for better water resources management.

Acknowledgement. Authors acknowledge the partial support from Dept. of Scienceand Technology, Govt. of India, through a project with reference no. ES/48/010/2003.

References

ASCE Task Committee. (2000a). “Artificial neural networks in hydrology. I:Preliminary Concepts.” J. Hydrologic Engineering, 5(2), 115-123.

ASCE Task Committee. (2000b). Artificial neural networks in hydrology. II:Hydrologic applications, J. Hydrologic Engineering, 5(2), 124-137.

Ashok, K., Z. Guan, and T. Yamagata (2001), Impact of Indian Ocean dipole on therelationship between the Indian monsoon rainfall and ENSO, Geophys. Res. Lett.,28, 4499 – 4502.

9

Cane, M. A. (1992). “Tropical Pacific ENSO models: ENSO as a mode of thecoupled system.” Climate System Modeling, K. E. Trenberth, Ed., CambridgeUniversity Press, 583 – 689.

Gadgil, S. (2003). “The Indian Monsoon and its variability.” Annu. Rev. EarthPlanet. Sci., 31, 429-67, doi: 10.1146/annurev.earth.31.100901.141251.

Gadgil, S., Vinayachandran P. N., and Francis P. A. (2003). “Droughts of the Indiansummer monsoon: Role of clouds over the Indian ocean.” Current Science,85(12), 1713 – 1719.

Gadgil, S., Vinayachandran P. N., Francis P. A., and Gadgil S. (2004). “Extremes ofthe Indian summer monsoon rainfall, ENSO and equatorial Indian Oceanoscillation.” Geophys. Res. Lett., 31, L12213, doi:10.1029/2004GL019733.

Haykin, S. (1999). “Neural Networks: A Comprehensive Foundation.” Prentice Hall,New Jersey.

Jain, S., and Lall U. (2001). “Floods in a changing climate: Does the past representthe future?” Water Resour. Res., 37(12), 3193 – 3205.

Kane, R. P. (1998). “Extremes of the ENSO phenomenon and Indian summermonsoon rainfall.” Int. J. Climatol., 18, 775 – 791.

Krishna Kumar K., Rajagopalan, B., and Cane, M. A. (1999). “On the weakeningrelationship between the Indian Monsoon and ENSO.” Science, 284, 5423, doi:10.1126/science.284.5423.2156.

Li, T., Zhang, Y. S., Chang, C. P., and Wang, B. (2001). “On the relationshipbetween Indian Ocean sea surface temperature and Asian summer monsoon.”Geophys. Res. Lett., 28, 2843-2846.

Maity, R. and Nagesh Kumar, D. (2006), “Bayesian dynamic modeling for monthlyIndian summer monsoon rainfall using ENSO and EQUINOO.” J. Geophys. Res.,In print.

Nagesh Kumar, D., Reddy, M. J. and Maity, R. (2005). “Regional rainfall forecastingusing large scale climate teleconnections and evolutionary algorithm.” InProcedings of the 2nd Indian International conference on artificial intelligence(IICAI-05), WR 105, Pune, India.

Parthasarathy, B., Diaz, H.F. and Eischeid, J.K. (1988). “Prediction of all/Indiasummer Monsoon Rainfall with Regional and large-scale parameters.” J.Geophys. Res., 93(5), 5341-5350.

Rasmusson, E. M., and Carpenter, T. H. (1983). “The relationship between easternequatorial Pacific sea surface temperature and rainfall over India and Sri Lanka.”Mon. Weather Rev., 111, 517 – 528.

Saji, N. H., Goswami, B. N., Vinayachandran, P. N., and Yamagata, T. (1999). “A dipole mode in the tropical Indian Ocean.” Nature, 401, 360 – 363.

See, L. M., Abrahart, R. J. and Heppenstall, A. J. (2005). “Using JavaSANE toevolve neural network rainfall-runoff models.” In Procedings of the 2nd IndianInternational conference on artificial intelligence (IICAI-05), WR 103, Pune,India.

Yu, X. H., and Chen, G. A. (1997). “Efficient back propagation learning usingoptimal learning rate and momentum.” Neural Networks, 10(3), 517 – 527.