InflowForecastofIranamaduReservoir,SriLanka,under

Research ArticleInflow Forecast of Iranamadu Reservoir Sri Lanka underProjected Climate Scenarios Using Artificial Neural Networks

Chamaka Karunanayake Miyuru B Gunathilake and Upaka Rathnayake

Department of Civil Engineering Faculty of Engineering Sri Lanka Institute of Information Technology Malabe Sri Lanka

Correspondence should be addressed to Upaka Rathnayake upakasanjeewagmailcom

Received 10 September 2020 Revised 12 November 2020 Accepted 16 November 2020 Published 7 December 2020

Academic Editor Wan Hanna Melini

Copyright copy 2020 Chamaka Karunanayake et al +is is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in anymedium provided the original work isproperly cited

Prediction of water resources for future years takes much attention from the water resources planners and relevant authoritiesHowever traditional computational models like hydrologic models need many data about the catchment itself Sometimes theseimportant data on catchments are not available due to many reasons +erefore artificial neural networks (ANNs) are useful softcomputing tools in predicting real-world scenarios such as forecasting future water availability from a catchment in the absenceof intensive data which are required for modeling practices in the context of hydrology +ese ANNs are capable of buildingrelationships to nonlinear real-world problems using available data and then to use that built relationship to forecast future needsEven though Sri Lanka has an extensive usage of water resources for many activities including drinking water supply irrigationhydropower development navigation and many other recreational purposes forecasting studies for water resources are not beingcarried out +erefore there is a significant gap in forecasting water availability and water needs in the context of Sri Lanka +usthis paper presents an artificial neural network model to forecast the inflows of one of the most important reservoirs in northernSri Lanka using the upstream catchmentrsquos rainfall Future rainfall data are extracted using regional climate models for the years2021ndash2050 and the inflows of the reservoir are forecasted using the validated neural network model Several training algorithmsincluding LevenbergndashMarquardt (LM) BFGS quasi-Newton (BFG) scaled conjugate gradient (SCG) have been used to find thebest fitting training algorithm to the prediction process of the inflows against the measured inflows Results revealed that the LMtraining algorithm outperforms the other tests algorithm in developing the prediction model In addition the forecasted resultsusing the projected climate scenarios clearly showcase the benefit of using the forecasting model in solving future water resourcemanagement to avoid or to minimize future water scarcity +erefore the validated model can effectively be used for properplanning of the proposed drinking water supply scheme to the nearby urban city Jaffna in northern Sri Lanka

1 Introduction

Water is the livelihood of civilization +e demand forusable water is always high in any community thereforethis has made many conflicts in the history Whether theyare recorded or not in a reliable way the history of conflictsfor water extends up to the civilization of humans [1ndash4]Demand for various water usages made these conflictsmore significant Negotiation among different stakeholdersbrings different aspects to water usage and they can initiatedifferent issues For example some people might argue thatthe priority should be given to drinking water rather thanagriculture However farmers might argue that the priority

should be given to them as they are living closer to theparticular water resources +erefore conflicts for watermay even happen between communities of the samecountry

On the other hand Sri Lanka as a country has facedmany conflicts due to water over the history [5ndash8] eventhough the country has many water resources It is a water-rich country It has managed its water resources well over thepast However there are several incidents where the conflictsare brought to the society by various stakeholders Waterresources management in the Iranamadu reservoir which isin the northern part of Sri Lanka is one of such cases toshowcase the water conflicts in Sri Lanka

HindawiApplied Computational Intelligence and So ComputingVolume 2020 Article ID 8821627 11 pageshttpsdoiorg10115520208821627

At the same time climate change and climate vari-abilities have done many adverse impacts to the water re-sources not only to Sri Lanka but also all over the worldSome countries like Ethiopia Somalia Djibouti EritreaKenya and Slovakia have experienced arid weather patterns[9 10] while some other countries like the United King-dom Sri Lanka Lithuania and Myanmar have experiencedintensified precipitation events [11 12] +e adverse impactsof these extreme climate events have influenced water supplysystems [13] sewer systems [14] agriculture [15] hydro-power development [16] and transportation [17] Howevertriggering natural disasters like landslides floods and tor-rential winds brings more weight to the problem

In addition the impact of climate change has beenwitnessed in many other multiple sectors For examplesignificantly higher temperatures which were the cause ofheat waves have damaged the ecosystems human healthand agriculture in European regions [18ndash20] In addition itwas showcased that increasing atmospheric temperatureshas significantly decreased the agricultural productivity inPakistan [21] On the other hand freezing temperatures dueto climate change have damaged the agricultural yield[22 23] Not only the losses in agriculture due to climatechange but also there is plenty of evidence in many otherareas for the impact of climate change including biodi-versity water resources and hydropower

+erefore proper planning and management of waterresources is a must for the future world +e projectedclimatic scenarios can be used to identify the available waterresources for future years and thus appropriate measurescan be preplanned +ese measures might not have to beperfect however an acceptable accuracy would grab theinterest of many stakeholders +erefore prediction orforecast of inflows to a reservoir is highly important

Time series inflow prediction in reservoirs is an im-portant task (probably the most important) in controlling thestorage of a reservoir +us it is a necessary activity tomanage the water resources for various downstream tasksincluding drinking water supply hydropower developmentand irrigation +ere are several ways of predicting theinflow to the systems starting from simplified tank models[24ndash27] to unsteady hydrological routing models [28ndash32]Literature shows these approaches are clustered into em-pirical conceptual physical and data-driven models[33ndash36] However usage of artificial neural networks hasintroduced a reasonably new approach (compared to thetraditional computational models) in predicting catchmentinflows without using many physical catchment data

Forecasting techniques are varied from one catchment toanother catchment depending on the available catchmentcharacteristics In addition to that data availability plays amajor role in the accuracy of the forecasting models [37 38]Furthermore different computational models have differentmodeling accuracies in the forecasting process

Some researchers have combined genetic algorithms andANN to predict the inflows to reservoirs [39] Results fromSedki et al [39] have shown that the hybrid models com-bining genetic algorithms to ANN outperform the tradi-tional ANNs However a well-trained ANN model can still

be a useful tool to predict the inflow to a reservoir at lowcomputational cost In addition ANNs do not need sta-tionary variables to be stationary and normally distributedcompared to classical stochastic modeling [40 41] Fur-thermore ANNs are comparably stable for the variousnoises in the measured data [42] Also ANNs are capable ofmodeling complex and nonlinear hydrological processes[43] having a self-adjusting ability to the data [44] needing alimited knowledge of the hydrological process [45] easyintegration to other models [43] and less demand incomputational cost are few other advantages of ANNs inhydrological processes

Nevertheless inflow predictions using ANNs have somedisadvantages too Nonstationary data have some impacts tothe predictions [46 47] +e extreme events in meteorologymay not be captured in the predictions+erefore the inflowpredictions may misguide the planners and authorities Inaddition the seasonal effects of rainfall can adversely impactthe quality of the predictions [48]

However despite several disadvantages literature givesmany examples of inflow prediction using ANNs [49ndash54]due to their advantages Many researchers have used dif-ferent types of algorithms in predicting inflows usingANNs For example Kisi [55] has used four different al-gorithms including backpropagation conjugate gradientcascade correlation and LevenbergndashMarquardt in theirstreamflow predictions However the usage of ANN topredict inflows in Sri Lanka is quite new Only a handful ofnumber of research has been done to Sri Lanka in the samemanner [56 57] Basnayake et al [56] have developed amodel to forecast the monthly inflow to one of the majorreservoirs in the wet zone of Sri Lanka +ey have usednonlinear autoregressive with exogenous input (NARX-ANN) neural network however they have not presentedtheir model with the projected climatic cases for futureyears Apart from the studies by Basnayake et al [56 57]the authors could not find related research in the context ofSri Lanka In addition the importance of such study isvisible when there are conflicts in the water usages in thedry zone of the country

+erefore it is timely important to conduct a hydro-logical catchment analysis to predict the inflow of the Ira-namadu reservoir and thus to investigate the cultivationcapacity of the Iranamadu reservoir in the upcoming de-velopment projects under the ongoing climate changescenarios +us this paper presents the soft computingtechniques in finding the relationships between the catch-mentrsquos rainfall and the inflow to the Iranamadu reservoir+e conventional method of flow calculation hydrologicalcatchment modeling was not conducted here due to twoimportant reasons unavailability of various data andcomplexity of the real-world hydrological models in theabsence of various required data including a dense networkof rain gauges timelymeasured streamflows soil infiltrationand evaporation +e obtained relationship was then used toforecast the Iranamadu inflow for the years 2021ndash2050 fromthe projected climate scenarios +e future climate data wereextracted from regional climate models using the Coordi-nated Regional Downscaling Experiment (CORDEX)

2 Applied Computational Intelligence and Soft Computing

+e results presented here reveal the importance of sucha study in the context of data unavailability and the usage ofthe developed model in future water management in theIranamadu reservoir

2 Study Area and Its Water Scarcity Issues

Iranamadu reservoir is the largest reservoir in northern SriLanka and can be considered one of the most importantreservoirs in that area +e reservoir is nutrified by theKanakarayan Aru a 86 km long river which starts fromSemamadu Kulam in Vavuniya district As it was alreadystated conflicts have aroused in the area due to the waterresources of this reservoir +e irrigational water supplyfrom the reservoir is comparably significant as many otherreservoirs in Sri Lanka +e reservoir caters around8000ndash10000 ha of paddy fields in the region

Sri Lanka practices two cultivation seasons namelyMaha and Yala seasons Maha is the major cultivating seasonwhereas Yala is the minor season Maha season starts inSeptember and lasts until March of the next year HoweverYala season starts in May and ends in August +e seasonitself is treated to be a minor season due to low irrigationalwater availability during the drier months

Figure 1 shows the cultivated land areas and the re-spective paddy harvests over the years for Maha (seeFigure 1(a)) and Yala (see Figure 1(b)) seasons (Maha andYala are the two major agricultural seasons of the country)Even though there is no connectivity from one year toanother dashed lines are used to showcase the variationclearly +ere were no cultivations during the 2008 to 2009years due to intensified war activities in the region Howeversince 2009 the area was well cultivated+is is an advantagewhich Sri Lanka is enjoying because of the end of the warconflict +e figures reveal a greater correlation between thecultivated land areas and paddy harvest However there aredrops in both cultivated land areas and harvest (for examplein 2012 and 2016 in Maha season and in 2014 2016 2017and 2018 in Yala season) +e main reason for this was theunavailability of water resources in the Iranamadu reservoirdue to the severe drought conditions in the area

Iranamadu tank was the first new tank to be built by theDepartment of Irrigation Sri Lanka other than restorationof ancient irrigational tanks [59]+is tank with a catchmentarea of 588 km2 was built damming two low-lying swamps ofthe Kanakarayan Aru +e construction of Iranamadu tankwith a capacity of 49 million cubic meters (MCM) wasinitiated in 1902 by the Department of Irrigation and thetank dam was raised to hold a capacity of 88 MCM in 1951+e dam was further raised by 2 feet in 1954 and that in-creased the capacity to 101 MCM +e dam was raised fromtime to time and now it is at the height of 34m (from meansea level (MSL)) +e tank can now hold water up to 148MCM [60] It is clearly evident the importance of the tank toKilinochchi area from these time-to-time dam height rises+e water is mainly used for irrigational purposes and thedemand for the irrigational water has increased with time(see Figure 1)

3 Future Climate Data Extraction

Regional climate models (RCMs) are numerical predictionmodels which use the boundary conditions of global climatemodels (GCMs) RCMs are developed in the interests oflocalized regions rather than globally +ere have been in-creasing trends in using regional climate models (RCMs) forclimate change impact studies due to the fine resolution inbetween 12 and 50 km when compared to coarse resolutionin between 100 and 250 km of global climate models(GCMs) Yet these RCMs inherit significant biases due toimperfect conceptualization internal climatic variabilitydiscretization and spatial averaging within grid cells[61ndash63] Bias correction methods including linear scalingmethod quantile mapping and local intensity scaling can beused to reduce the biases in climatic datasets [62] +e biascorrection process is usually carried out because of limitedspatial resolution and simplified physics of the models Inaddition the physics of the world is yet to be revealed andthus the models usually produced some biased future cli-mate data

ACCESS_CCAM MPI_ESM_CCAM CNRM_CCAMand REMO2009 are four RCMs which were developed bythe Commonwealth Scientific and Industrial Research Or-ganization (CSIRO) of spatial resolution 05 deg times 05 deg [64 65]from Coordinated Regional Downscaling Experiment(CORDEX) +e driving GCMs of these RCMs are ACCESS10 MPI_ESM_LR CNRM_CM5 and ECHAM-4 GCM[66 67] respectively Climatic data of CORDEX are avail-able through httpswwwcordexorg +ese RCMs arewidely used for climate studies +e climatic data for theyears 1976ndash2100 are available in RCMs comprising a his-torical period in between 1976 and 2005 and future periodbetween 2006 and 2100

4 Forecasting Model Development

Rainfall data for three rain gauging stations placed inside thecatchment area of Iranamadu reservoir (naming OlumaduIranamadu and Mankulam) were obtained from the De-partment of Irrigation Sri Lanka In addition the inflows tothe reservoir were collected from the same department for 30years from 1959 to 1989 Most of the rainfall data are missingafter 1989 due to various reasons mainly the war in this areauntil 2009 In addition the Mankulam rain gauge has notfunctioned well in recent years +erefore the authors haveto rely on the monthly rainfall data and reservoir inflow datafor the period of 1959 to 1989 During this period themissing data percentage is less than 10 (Olumadumdash0Iranamadumdash69 and Mankulammdash19) +e annualvariations of the rainfall in three gauges are given in Figure 2As they were expected zigzag variations can be observed inall three annual rainfall variations +e annual rainfalls wereinterconnected along the time axis to show the temporalvariations even though the annual rainfalls are notinterconnected

Inflow to the Iranamadu reservoir from KanakarayanAru based on the rainfall of three rain gauges was modeled

Applied Computational Intelligence and Soft Computing 3

using catchmentrsquos rainfall in the ANN environment +erelationship is given in the following equation

Inflow ϕ Rain falli( 1113857 (1)

where ϕ is the nonlinear function in between inflow andrainfall and i represents the rain gauges 70 of the data wasfed to ANN architecture to train the neural network whereas15 was used to validate the trained neural network and

300Mankulam

Mankulam

Olumadu

Olumadu

Iranamadu

Iranamadu

250

200

150

100

50

01959 1964 1969 1974 1979

Year

Ave

rage

rain

fall

(mm

)

300

250

200

150

100

50

0

Ave

rage

rain

fall

(mm

)300

250

200

150

100

50

0

Ave

rage

rain

fall

(mm

)

1984 1989 1959 1964 1969 1974 1979Year

1984 1989

1959 1964 1969 1974 1979Year

1984 1989

Figure 2 Rain gauges and rainfall variation over the years

100000 25000

20000

15000

10000

5000

0

80000

60000

Har

vest

(MT)

Har

veste

d ar

ea (h

a)

40000

20000

02004 2008 2012

Time (years)2016 2020

Harvest (MT)Area (ha)

(a)

25000

20000

15000

10000

5000

0

Har

veste

d ar

ea (h

a)

100000

80000

60000

Har

vest

(MT)

40000

20000

02004 2008 2012



(b)

Figure 1 Paddy harvest with respect to the cultivated area in Kilinochchi district (data source [58]) (a) For Maha season and (b) for Yala season


another 15 to test the validated neural network +efeedforward neural network (FFNN) was in this studyFFNN works in one direction (forward) from inputs tooutcome through the hidden nodes +ere is no feedback inthe FFNN and it is the simplest form of ANNs Howeverthey can still be effectively used to model complex scenarios

+e developed ANN model was trained using sevenalgorithms namely BFGS quasi-Newton (BFG) Fletch-erndashReeves conjugate gradient algorithm (CGF) scaledconjugate gradient (SCG) PolakndashRibiere conjugate gradient(CGP) conjugate gradient with PowellBeale restarts (CGB)LevenbergndashMarquardt (LM) and resilient backpropagation(RP) algorithms +ese algorithms are widely used in real-world problems [68ndash70] +e problem was tested on theseseven ANN algorithms to check their suitability and stabilityin the specified application More information on thesealgorithms can be found in Perera et al [71]

+e trained ANN model was saved for the future fore-casting process +e ANN prediction process was validatedusing the data from 1999 to 2000+is process was carried outto revalidate the developed model using available rainfall datafor a future period from themodel development+emonthlyrainfall data of the stations Iranamadu and Olumadu werereadily available whereas the data for Mankulam station werenot available In addition Iranamadu inflow data wereavailable for this period+erefore 1999ndash2000 two years wereselected to do the prediction validation of the ANN model+e missing rainfall days were filled with gridded precipi-tation datasets from APHRODITErsquos (Asian PrecipitationHighly Resolved Observational Data Integration TowardsEvaluation of Water Resources) precipitation dataset [72 73]+e precipitation data related to the location of Mankulam(913700N 8044520E) were extracted from APHRODITErsquosdataset (available at httpaphroditesthirosaki-uacjp) us-ing ARCGIS (version 104) +e validity of the APHRODITEdataset for Mankulam was checked before the application+e data from APHRODITE and the observed data werecompared for 10 years (1979ndash1989) +e resulting Pearsoncorrelation coefficient (07916) obtained was greater than075 and hence the dataset from APHRODITE was used toextract monthly rainfall data of Mankulam station for theyears 1999 and 2000

+e monthly rainfall data for 1999ndash2000 were fed to thedeveloped ANN model and derived the streamflows to thecorresponding months as the predicted streamflows +esepredicted streamflows were then compared against themeasured streamflows for the 1999ndash2000 time

After the prediction validation rainfall data for 30 yearsfrom 2020 to 2050 were extracted from four regional climatemodels (namely ACCESS_CCAM MPI_ESM_CCAMCNRM_CCAM and REMO2009) for Iranamadu Olumaduand Mankulam Two Representative Concentration Path-ways (RCP45 and RCP85) were used for rainfall data ex-traction +e RCPs are greenhouse gas emissionconcentration trajectories adopted by the IPCC [74] +esetwo RCPs were selected based on their emission scenariosRCP45 is a medium emission scenario and RCP85 highemission scenario More details on RCP scenarios can befound in Van Vuuren et al [75]

Linear scaling method (LS) [76] was used to removebiases in the rainfall data +e linear scaling approach as-sumes that the correction algorithm and parametrization ofhistorical climate will remain stationary for future climaticconditions Previous studies demonstrate that LS approachperforms well for coarse temporal scale analysis as morecomplicated methods such as quantile mapping deltachange and power transformation [77 78] Mathematicalformulations for bias correction in precipitation are given inequations (2) and (3) +e linear scaling correction wasapplied at all stations individually

Plowasthis(d) Phis(d) times

μm Pobs(d)( 1113857

μm Phis(d)( 1113857 (2)

Plowastsim(d) Psim(d) times

μm Pobs(d)( 1113857

μm Phis(d)( 1113857 (3)

where lowast his obs sim P and d stand for bias-corrected dataraw RCM hindcast observed data raw RCM-corrected dataprecipitation and daily respectively μm is the long-termmonthly mean of the precipitation data +e correctedrainfall data were then fed to the saved ANN to predict thefuture inflows to the Iranamadu reservoir

5 Results and Discussion

Table 1 presents the coefficient of correlation and perfor-mances of each training algorithms in the ANN analyses Itcan be clearly observed that all algorithms have performedwell in the context of correlation coefficients +ey are al-most 1 for validation and more than 08 for all other cases+ese values are acceptable in the time-dependent highlynonlinear climate analyses [79ndash82] However out of themLM algorithm has outperformed other training algorithmsin the computational performance (highlighted in grey) Ithas given better results in less computational time (8 epochs)and lowest mean squared error (MSE 614) +erefore LMtraining algorithm was selected for the prediction analyses

Figure 3 gives the illustrations of the correlation coef-ficients of LM algorithm in training validation test and allcases It can be understood that the prediction is well val-idated in the developed ANN architecture +e results arescattered along the 45deg line (ideal match) for all cases+erefore this observation validates the procedure whichwas used in this analysis

Figure 4 presents the predicted (from ANNmodel) inflowto the Iranamadu reservoir against the actual measuredinflow for the years 1999ndash2000 +e trend line drawn to thedata scatter has a coefficient of determination of 0954(which is almost 1) +at means the trend line almost givesthe real feeling of the data scatter In addition the equationto the linear trend line drawn is given in the figure itself Infact if the predicted inflow has a perfect match to themeasured inflow the equation of the line should be y x(predicted inflowmeasured inflow) However this is theideal match +e trend line gives an inclination of 0826(which is equal to 395 degrees) +erefore the prediction isslightly lower than the actual measured inflow In addition


the graph has an MSE of 1611 MCM Even though there is aslight deviation the results show a validation to a greaterextent

+e bias-corrected annual rainfalls for three rain gaugingstations are given in Figure 5 (Iranamadu (Figure 5(a))Mankulam (Figure 5(b)) and Olumadu (Figure 5(c))) +efigures give the projected rainfall for two RCPs which wereconsidered in this analysis +e variations are given in

dashed lines for a clear understanding of the temporalvariations of the rainfall even though there is no connec-tivity of rainfall from one year to the next year As it isexpected annual rainfall variations show a zigzag patternYears 2041 and 2045 are two possible extreme years forRCP45 scenario Heavy rainfalls are projected for the year2041 for all three stations whereas lowest rainfalls are pre-dicted for the year 2045 However interestingly RCP85

80

60

40

20

00

R = 0989

20 40Actual inflow (MCM)

Pred

icte

d in

flow

(MCM

)

60 80

(a)

R = 0989

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)

0 20 40Actual inflow (MCM)

60 80

(b)

R = 0947

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(c)

R = 0983

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(d)

Figure 3 Correlation coefficients for LM algorithm (a) for training (b) for validation (c) for test and (d) for all (combining trainingvalidation and testing)

Table 1 ANN results from different training algorithms

AlgorithmCorrelation coefficients Performance

Training Validation Test All MSE EpochsBFG 0986 0978 097 0983 1017 21CGF 0847 0978 0825 0861 4701 15SCG 098 0988 0981 0981 741 66CGP 0978 0957 0957 0971 2119 12CGB 0973 0977 0961 0972 1333 17LM 0989 0989 0947 0983 614 8RP 0977 097 0947 0971 1152 40


scenario gives a different projection for the year 2045 In-stead of the lowered rainfall projections higher rainfalls canbe projected in all three stations +ese are projected ob-servations where a detailed research is essential to commentHowever planners can use these projections for their futureactivities based on the rainfall

In addition to the annual projections monthly rainfallprojections are also obtained +e projected rainfall for the

year 2021 over the months (see Figures 6(a) for RCP45 and6(b) for RCP85) are given under the secondary data in theAppendix section Heavy rainfalls can be expected in themonths of October to December in both RCP scenarios +enorthern area of Sri Lanka receives its maximum rainfallduring the northeastern monsoon which is practiced fromDecember to February However the higher rainfalls areprojected in the 2nd intermonsoon period (October to

2200

1850

1500

1150

8002020 2025 2030 2035

Time (years)

Ann

ual r

ainf

all (

mm

)

2040 2045 2050

RCP45RCP85

(a)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(b)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(c)

Figure 5 Projected annual rainfall variation from 2021 to 2050 (a) for Iranamadu (b) for Mankulam and (c) for Olumadu

y = 08265x ndash 00697R2 = 09542

50

40

30

20

10

00 10 20

Actual inflow (MCM)

Pred

icte

d in

flow

(MCM

)

30 40 50

Figure 4 Inflow validations for the years 1999ndash2000


November) +erefore this is interesting as farmers aregetting ready for the new cultivation season

Local farmers usually state that the rainy seasons areshifted in recent years +is is not scientifically proved yetHowever that statement with the observations in monthlyvariations is to be researched in detail for sound conclu-sions Interestingly RCP85 scenario projects anotherrainfall peak in April for all three stations +is is in the 1stintermonsoon (March to April) where usually a direrperiod is observed in northern Sri Lanka +erefore itwould be better to conduct a detailed research along thelines of the impact of climate change on water resources inthe Iranamadu area

+e most important finding of this research inflow tothe Iranamadu reservoir forecast is illustrated in Figure 7(see Figure 7(a) for RCP45 scenario and Figure 7(b) forRCP85) +e projected inflows from the years 2021 to 2050are presented Annual inflows are not interconnectedhowever they were presented by the dashed lines for clarityof the variations Not only the annual forecasts but also the

monthly inflow forecasts are obtained in this researchHowever they are not presented here due to spacelimitations

Inflow forecasts clearly showcase the importance offuture water management in the Iranamadu reservoirSeveral peaks and troughs can be visible in the inflow Peakscan easily lead to flooding in the area due to the non-contoured area However the drought seasons would betough in managing water as there is water scarcity in thedrier seasons +erefore proper planning is required andthe results of this analysis can be used

+e available statistics show that the average annualinflow to the Iranamadu reservoir between the years 1958and 2014 is 143MCM However the forecasts projected thatto be 154 MCM for RCP45 and 178 MCM for RCP85+erefore the average annual inflow (from 2021 to 2050) tothe Iranamadu reservoir can be higher compared to thatfrom 1958 to 2014 Nevertheless as stated the demand forthe water from the Iranamadu reservoir has been increasedover the years and projected to increase

202050

100

150

200

2025 2030 2035Time (years)

2040 2045 2050

Pred

icte

d in

flow

(MCM

)

(a)

2020 2025 2030 2035Time (years)

2040 2045 2050100

150

200

Pred

icte

d in

flow

(MCM

)

(b)

Figure 7 Inflow prediction to the Iranamadu reservoir from 2021 to 2050 (a) For RCP45 and (b) for RCP85

Jan0

100

200

300

400

Rain

fall

(mm

)

500

600

IranamaduMankulamOlumadu

Feb Mar Apr May JunMonths

Jul Aug Sep Oct Nov Dec

(a)

0

100

200

300

400

Rain

fall

(mm

)


Jan Feb Mar Apr May JunMonths


(b)

Figure 6 Monthly rainfall variations in the year 2021 (a) For RCP45 and (b) for RCP85


In addition there is a proposal to transfer a significantamount of water (27000m3day) from the Iranamadu res-ervoir to northern city Jaffna +erefore the water man-agement strategies would have to be rearranged in theIranamadu reservoir to cater to these new demands If notthere may be some conflicts among the farmers and thewater distribution planners for the water in the Iranamadureservoir

Conflicts may be further increased with the populationgrowth More agricultural lands would be utilized to cater tothe food industry whereas more drinking water will berequired in increasing population +erefore these arecritical issues to be handled at extremely careful levels

6 Conclusions

Artificial neural network model under the LM algorithm hasoutperformed all other training algorithms in the inflowprediction model +erefore the inflow prediction modelfrom Kanakarayan Aru to the Iranamadu reservoir innorthern Sri Lanka is set under the LM training algorithm+e validation process of the inflows from the predicted andmeasured data reveals the accuracy and robustness of theprediction model +erefore the model can be used toforecast the inflows to Iranamadu under any future climatescenarios +e results reveal its applicability only for 2 testedRCPs however depending on the requirement the modelcan be successfully used +e inflow forecast in the coming30 years is highly important to the water resources managersto deal with the conflicts among various stakeholders underthe water scarcities+erefore the forecast can be used in theplanning of future arrangements to develop the water dis-tribution network to the Jaffna city In addition the waterresources in this dry area can be utilized in a sustainablemanner among various stakeholders including farmers inthe area Proper crop management can be introduced to thefarmers based on the future inflow availability +ereforethe results from this research are highly important to thewater resources managers and planners in the provincialcouncil and the local authorities

Data Availability

+e climatic data and the analysis data are available from thecorresponding author upon request

Disclosure

+e research was carried out in the Sri Lanka Institute ofInformation Technology environment

Conflicts of Interest

+e authors declare no conflicts of interest

Acknowledgments

+e authors would like to acknowledge and appreciate DrArulanantham Anburuvel Senior Lecturer in Faculty ofEngineering University of Jaffna Sri Lanka and Eng

N Suthakaran the Deputy Director of the Department ofIrrigation Kilinochchi Sri Lanka for their effort inobtaining the rainfall and inflow data for the Iranamadureservoir +is research received no external funding

References

[1] L J Del Giacco R Drusiani L Lucentini and S MurtasldquoWater as a weapon in ancient times considerations oftechnical and ethical aspectsrdquo Water Supply vol 17 no 5pp 1490ndash1498 2017

[2] H Hatami and P H Gleick ldquoConflicts over water in themyths legends and ancient history of the Middle EastrdquoEnvironment Science and Policy for Sustainable Developmentvol 36 no 3 pp 10-11 1994

[3] A Janku ldquoChina a hydrological historyrdquo Nature vol 536no 7614 pp 28-29 2016

[4] D K Kreamer ldquo+e past present and future of water conflictand international securityrdquo Journal of Contemporary WaterResearch amp Education vol 149 no 1 pp 87ndash95 2012

[5] R Meinzen-Dick and M Bakker ldquoWater rights and multiplewater usesmdashframework and application to Kirindi Oya irri-gation system Sri Lankardquo Irrigation and Drainage Systemsvol 15 no 2 pp 129ndash148 2001

[6] F Molle P Jayakody R Ariyaratne and H S SomatilakeldquoIrrigation versus hydropower sectoral conflicts in southernSri Lankardquo Water Policy vol 10 no S1 pp 37ndash50 2008

[7] K Nandalal and S Simonovic ldquoResolving conflicts in watersharing a systemic approachrdquo Water Resources Researchvol 39 pp 1ndash11 2003

[8] M Samad ldquoWater institutional reforms in Sri Lankardquo WaterPolicy vol 7 no 1 pp 125ndash140 2005

[9] M G Ghebrezgabher T Yang and X Yang ldquoLong-termtrend of climate change and drought assessment in the horn ofafricardquo Advances in Meteorology vol 2016 Article ID8057641 12 pages 2016

[10] J Vido P Nalevankova J Valach Z Sustek and T TadesseldquoDrought analyses of the horne pozitavie region (Slovakia) inthe period 1966ndash2013rdquo Advances in Meteorology vol 2019Article ID 3576285 10 pages 2019

[11] B Khaniya I Jayanayaka P Jayasanka and U RathnayakeldquoRainfall trend analysis in Uma Oya basin Sri Lanka andfuture water scarcity problems in perspective of climatevariabilityrdquo Advances in Meteorology vol 2019 Article ID3636158 10 pages 2019

[12] D Kum K J Lim C H Jang et al ldquoProjecting future climatechange scenarios using three bias-correction methodsrdquo Ad-vances in Meteorology vol 2014 Article ID 704151 12 pages2014

[13] W Mo H Wang and J M Jacobs ldquoUnderstanding theinfluence of climate change on the embodied energy of watersupplyrdquo Water Research vol 95 pp 220ndash229 2016

[14] M Kleidorfer M Moderl R Sitzenfrei C Urich andW Rauch ldquoA case independent approach on the impact ofclimate change effects on combined sewer system perfor-mancerdquo Water Science and Technology vol 60 no 6pp 1555ndash1564 2009

[15] W Wang M W Ertsen M D Svoboda and M HafeezldquoPropagation of drought from meteorological drought toagricultural and hydrological droughtrdquo Advances in Meteo-rology vol 2016 Article ID 6547209 5 pages 2016

[16] B Kichonge ldquo+e status and future prospects of hydropowerfor sustainable water and energy development in Tanzaniardquo


Journal of Renewable Energy vol 2018 Article ID 657035812 pages 2018

[17] D Jaroszweski L Chapman and J Petts ldquoAssessing thepotential impact of climate change on transportation theneed for an interdisciplinary approachrdquo Journal of TransportGeography vol 18 no 2 pp 331ndash335 2010

[18] R Garcıa-Herrera J Dıaz R M Trigo J Luterbacher andEM Fischer ldquoA review of the European summer heat wave of2003rdquo Critical Reviews in Environmental Science and Tech-nology vol 40 no 4 pp 267ndash306 2010

[19] G A Meehl and C Tebaldi ldquoMore intense more frequentand longer lasting heat waves in the 21st centuryrdquo Sciencevol 305 no 5686 pp 994ndash997 2004

[20] C Schar P L Vidale D Luthi et al ldquo+e role of increasingtemperature variability in European summer heatwavesrdquoNature vol 427 no 6972 pp 332ndash336 2004

[21] U Shakoor A Saboor I Ali and A Q Mohsin ldquoImpact ofclimate change on agriculture empirical evidence from aridregionrdquo Pakistan Journal of Agricultural Sciences vol 48no 4 pp 327ndash333 2011

[22] P K Jha E Xanthakis S Chevallier V Jury and A Le-BailldquoAssessment of freeze damage in fruits and vegetablesrdquo FoodResearch International vol 121 pp 479ndash496 2019

[23] D B Lobell A Torney and C B Field ldquoClimate extremes inCalifornia agriculturerdquo Climatic Change vol 109 no S1pp 355ndash363 2011

[24] Y Fukushima ldquoA model of river flow forecasting for a smallforested mountain catchmentrdquo Hydrological Processes vol 2no 2 pp 167ndash185 1988

[25] M G Kang J H Lee and K W Park ldquoParameter region-alization of a tank model for simulating runoffs fromungauged watershedsrdquo Journal of Korea Water ResourcesAssociation vol 46 no 5 pp 519ndash530 2013

[26] K Mizumura ldquoRunoff prediction by simple tank model usingrecession curvesrdquo Journal of Hydraulic Engineering vol 121no 11 pp 812ndash818 1995

[27] P-S Yu and T-C Yang ldquoUsing synthetic flow durationcurves for rainfall-runoff model calibration at ungaugedsitesrdquoHydrological Processes vol 14 no 1 pp 117ndash133 2000

[28] D Che and L W Mays ldquoDevelopment of an optimizationsimulation model for real-time flood-control operation ofriver-reservoirs systemsrdquo Water Resources Managementvol 29 no 11 pp 3987ndash4005 2015

[29] T Haktanir and H Ozmen ldquoComparison of hydraulic andhydrologic routing on three long reservoirsrdquo Journal of Hy-draulic Engineering vol 123 no 2 pp 153ndash156 1997

[30] K P Singh and A Snorrason ldquoSensitivity of outflow peaksand flood stages to the selection of dam breach parametersand simulationmodelsrdquo Journal of Hydrology vol 68 no 1ndash4pp 295ndash310 1984

[31] B G Tassew M A Belete and K Miegel ldquoApplication ofHEC-HMS model for flow simulation in the lake tana basinthe case of gilgel abay catchment upper blue nile basinEthiopiardquo Hydrology vol 6 no 1 p 21 2019

[32] S-C Yang and T-H Yang ldquoUncertainty assessment reser-voir inflow forecasting with ensemble precipitation forecastsand HEC-HMSrdquo Advances in Meteorology vol 2014 ArticleID 581756 11 pages 2014

[33] J Adamowski K Adamowski and A Prokoph ldquoA spectralanalysis based methodology to detect climatological influ-ences on daily urban water demandrdquo Mathematical Geo-sciences vol 45 no 1 pp 49ndash68 2012

[34] M A Gad ldquoA useful automated rainfall-runoff model forengineering applications in semi-arid regionsrdquo Computers ampGeosciences vol 52 pp 443ndash452 2013

[35] M Paudel E J Nelson C W Downer and R HotchkissldquoComparing the capability of distributed and lumped hy-drologic models for analyzing the effects of land use changerdquoJournal of Hydroinformatics vol 13 no 3 pp 461ndash473 2010

[36] A K Verma M K Jha and R K Mahana ldquoEvaluation ofHEC-HMS and WEPP for simulating watershed runoff usingremote sensing and geographical information systemrdquo Paddyand Water Environment vol 8 no 2 pp 131ndash144 2009

[37] M U Kale M B Nagdeve and S B Wadatkar ldquoReservoirinflow forecasting using artificial neural networkrdquo HydrologyJournal vol 35 no 1and2 p 52 2012

[38] F Othman andM Naseri ldquoReservoir inflow forecasting usingartificial neural networkrdquo International Journal of the PhysicalSciences vol 6 no 3 pp 434ndash440 2011

[39] A Sedki D Ouazar and E El Mazoudi ldquoEvolving neuralnetwork using real coded genetic algorithm for daily rainfall-runoff forecastingrdquo Expert Systems with Applications vol 36no 3 pp 4523ndash4527 2009

[40] L Burke ldquoClustering characterization of adaptive resonancerdquoNeural Networks vol 4 no 4 pp 485ndash491 1991

[41] O R Dolling and E A Varas ldquoArtificial neural networks forstreamflow predictionrdquo Journal of Hydraulic Research vol 40no 5 pp 547ndash554 2002

[42] C Zealand D Burn and S Simonovic ldquoShort termstreamflow forecasting using artificial neural networksrdquoJournal of Hydrology vol 214 no 1ndash4 pp 32ndash48 1999

[43] O Oyebode and J Adeyemo ldquoReservoir inflow forecastingusing differential evolution trained neural networksrdquo inEVOLVEmdashA Bridge between Probability Set Oriented Nu-merics and Evolutionary Computation V A A TantarE Tantar J-Q Sun et al Eds Springer Cham Switzerland2014

[44] E Sharghi V Nourani H Najafi and AMolajou ldquoEmotionalANN (EANN) and wavelet-ANN (WANN) approaches formarkovian and seasonal based modeling of rainfall-runoffprocessrdquo Water Resources Management vol 32 pp 3441ndash3456 2018

[45] C Bennett R A Stewart and C D Beal ldquoANN-based res-idential water end-use demand forecasting modelrdquo ExpertSystems with Applications vol 40 no 4 pp 1014ndash1023 2013

[46] B Cannas A Fanni L See and G Sias ldquoData preprocessingfor river flow forecasting using neural networks wavelettransforms and data partitioningrdquo Physics and Chemistry ofthe Earth Parts ABC vol 31 no 18 pp 1164ndash1171 2006

[47] T Partal and O Kisi ldquoWavelet and neuro-fuzzy conjunctionmodel for precipitation forecastingrdquo Journal of Hydrologyvol 342 no 1-2 pp 199ndash212 2007

[48] S Kumar M K Tiwari C Chatterjee and A MishraldquoReservoir inflow forecasting using ensemble models based onneural networks wavelet analysis and bootstrap methodrdquoWater Resources Management vol 29 no 13 pp 4863ndash48832015

[49] P Coulibaly F Anctil and B Bobee ldquoMultivariate reservoirinflow forecasting using temporal neural networksrdquo Journalof Hydrologic Engineering vol 6 no 5 pp 367ndash376 2001

[50] S Debbarma and P Choudhury ldquoRiver flow prediction withmemory-based artificial neural networks a case study of theDholai river basinrdquo International Journal of Advanced In-telligence Paradigms vol 15 no 1 pp 51ndash62 2020


[51] H M Ibrahim ldquoPrediction of darbandikhan reservoir inflowusing ANFIS modelsrdquo Journal of Zankoy SulaimanimdashPart Avol 19 no 2 pp 127ndash142 2017

[52] S K Jain A Das and D K Srivastava ldquoApplication of ANNfor reservoir inflow prediction and operationrdquo Journal ofWater Resources Planning and Management vol 125 no 5pp 263ndash271 1999

[53] R B Magar and V Jothiprakash ldquoIntermittent reservoirdaily-inflow prediction using lumped and distributed datamulti-linear regression modelsrdquo Journal of Earth SystemScience vol 120 no 6 pp 1067ndash1084 2011

[54] A Moradi A Dariane G Yang and P Block ldquoLong-rangereservoir inflow forecasts using large-scale climate predictorsrdquoInternational Journal of Climatology vol 40 no 13 2020

[55] O Kisi ldquoStreamflow forecasting using different artificialneural network algorithmsrdquo Journal of Hydrologic Engi-neering vol 12 no 5 pp 532ndash539 2007

[56] N Basnayake D Attygalle L Liyanage and K NandalalldquoEnsemble forecast for monthly reservoir inflow a dynamicneural network approachrdquo in Proceedings of the 4th AnnualInternational Conference on Operations Research and Statistics(ORS 2016) pp 1ndash8 Singapore January 2016

[57] N Basnayake L Liyanage D Attygalle and K NandalalldquoEffective water management in the mahaweli reservoirsystem analyzing the inflow of the upmost reservoirrdquo inProceedings of the International Symposium for Next Gener-ation Infrastructure MART Infrastructure Facility pp 1ndash6Wollongong Australia October 2013

[58] Department of Census and Statistics Paddy Statistics httpwwwstatisticsgovlkagriculturePaddy20StatisticsPaddyStatshtm 2020

[59] S Arumugum Water Resources of Ceylon its Utilization andDevelopment Water Resources Board Sri Lanka ColomboSri Lanka 1st edition 1969

[60] NPC Sri Lanka Status Report of Iranamadu Tank as of 28012019httpsnpgovlkstatus-report-of-iranamadu-tank-as-of-28-01-2019Northern Provincial Council Vavuniya Sri Lanka 2019 httpsnpgovlkstatus-report-of-iranamadu-tank-as-of-28-01-2019

[61] J Christensen F Boberg O Christensen and P Lucas-Picher ldquoOn the need for bias correction of regional climatechange projections of temperature and precipitationrdquo Geo-physical Research Letters vol 35 pp 1ndash6 2008

[62] C Teutschbein and J Seibert ldquoBias correction of regionalclimate model simulations for hydrological climate-changeimpact studies review and evaluation of different methodsrdquoJournal of Hydrology vol 456-457 pp 12ndash29 2012

[63] O Varis T Kajander and R Lemmela ldquoClimate and waterfrom climate models to water resources management and viceversardquo Climatic Change vol 66 no 3 pp 321ndash344 2004

[64] W Cabos D V Sein A Duran-Quesada et al ldquoDynamicaldownscaling of historical climate over CORDEX CentralAmerica domain with a regionally coupled atmosphere-oceanmodelrdquoClimate Dynamics vol 52 no 7-8 pp 4305ndash4328 2018

[65] V P Pandey S Dhaubanjar L Bharati and B R +apaldquoHydrological response of Chamelia watershed in MahakaliBasin to climate changerdquo Science of Ie Total Environmentvol 650 pp 365ndash383 2019

[66] D Jacob ldquoA note to the simulation of the annual and inter-annual variability of the water budget over the Baltic Seadrainage basinrdquoMeteorology and Atmospheric Physics vol 77no 1ndash4 pp 61ndash73 2001

[67] D Jacob B J J M Van den Hurk U Andraelig et al ldquoAcomprehensive model inter-comparison study investigatingthe water budget during the BALTEX-PIDCAP periodrdquo

Meteorology and Atmospheric Physics vol 77 no 1ndash4pp 19ndash43 2001

[68] M T Hagan and M B Menhaj ldquoTraining feedforwardnetworks with the Marquardt algorithmrdquo IEEE Transactionson Neural Networks vol 5 no 6 pp 989ndash993 1994

[69] S Singh S Jain and A Bardossy ldquoTraining of artificial neuralnetworks using information-rich datardquo Hydrology vol 1no 1 pp 40ndash62 2014

[70] J Wang P Shi P Jiang et al ldquoApplication of BP neuralnetwork algorithm in traditional hydrological model for floodforecastingrdquo Water vol 9 no 1 pp 48ndash63 2017

[71] A Perera H Azamathulla and U Rathnayake ldquoComparisonof different artificial neural network (ANN) training algo-rithms to predict atmospheric temperature in Tabuk SaudiArabiardquo MAUSAM Quarterly Journal of Meteorology Hy-drology amp Geophysics vol 71 no 2 pp 551ndash560 2020

[72] APHRODITErsquos Water Resources httpswwwchikyuacjpprecipenglishfbclidIwAR0B13K0v2eaZpwdvIKItjmhbo5fz0_-EXOJ5ZnENbzhrXq0l5xNciZFjHU 2020

[73] A Yatagai K Kamiguchi O Arakawa A HamadaN Yasutomi and A Kitoh ldquoAPHRODITE constructing along-term daily gridded precipitation dataset for asia basedon a dense network of rain gaugesrdquo Bulletin of theAmerican Meteorological Society vol 93 no 9 pp 1401ndash1415 2012

[74] IPCC ldquoClimate change 2014 synthesis reportrdquo in Fifth As-sessment Report of the Intergovernmental Panel on ClimateChange R K Pachauri and L A Meyer Eds pp 1ndash151Intergovernmental Panel on Climate Change Geneva Swit-zerland 2014

[75] D P Van Vuuren J Edmonds M Kainuma et al ldquo+erepresentative concentration pathways an overviewrdquo Cli-matic Change vol 109 no 1-2 pp 5ndash31 2011

[76] G Lenderink A Buishand and W van Deursen ldquoEstimatesof future discharges of the river Rhine using two scenariomethodologies direct versus delta approachrdquo Hydrology andEarth System Sciences vol 11 no 3 pp 1145ndash1159 2007

[77] U Ghimire G Srinivasan and A Agarwal ldquoAssessment ofrainfall bias correction techniques for improved hydrologicalsimulationrdquo International Journal of Climatology vol 39no 4 pp 2386ndash2399 2018

[78] P Stanzel and H Kling ldquoFrom ENSEMBLES to CORDEXevolving climate change projections for upper Danube riverflowrdquo Journal of Hydrology vol 563 pp 987ndash999 2018

[79] A B Dariane and S Azimi ldquoStreamflow forecasting bycombining neural networks and fuzzy models using advancedmethods of input variable selectionrdquo Journal of Hydro-informatics vol 20 no 2 pp 520ndash532 2017

[80] S Nacar M A Hınıs and M Kankal ldquoForecasting dailystreamflow discharges using various neural network modelsand training algorithmsrdquo KSCE Journal of Civil Engineeringvol 22 no 9 pp 3676ndash3685 2017

[81] T Peng J Zhou C Zhang and W Fu ldquoStreamflow fore-casting using empirical wavelet transform and artificial neuralnetworksrdquo Water vol 9 no 6 pp 1ndash20 2017 406

[82] G Uysal A A ſorman and A ſensoy ldquoStreamflow forecastingusing different neural network models with satellite data for asnow dominated region in Turkeyrdquo Procedia Engineeringvol 154 pp 1185ndash1192 2016


At the same time climate change and climate vari-abilities have done many adverse impacts to the water re-sources not only to Sri Lanka but also all over the worldSome countries like Ethiopia Somalia Djibouti EritreaKenya and Slovakia have experienced arid weather patterns[9 10] while some other countries like the United King-dom Sri Lanka Lithuania and Myanmar have experiencedintensified precipitation events [11 12] +e adverse impactsof these extreme climate events have influenced water supplysystems [13] sewer systems [14] agriculture [15] hydro-power development [16] and transportation [17] Howevertriggering natural disasters like landslides floods and tor-rential winds brings more weight to the problem

In addition the impact of climate change has beenwitnessed in many other multiple sectors For examplesignificantly higher temperatures which were the cause ofheat waves have damaged the ecosystems human healthand agriculture in European regions [18ndash20] In addition itwas showcased that increasing atmospheric temperatureshas significantly decreased the agricultural productivity inPakistan [21] On the other hand freezing temperatures dueto climate change have damaged the agricultural yield[22 23] Not only the losses in agriculture due to climatechange but also there is plenty of evidence in many otherareas for the impact of climate change including biodi-versity water resources and hydropower

+erefore proper planning and management of waterresources is a must for the future world +e projectedclimatic scenarios can be used to identify the available waterresources for future years and thus appropriate measurescan be preplanned +ese measures might not have to beperfect however an acceptable accuracy would grab theinterest of many stakeholders +erefore prediction orforecast of inflows to a reservoir is highly important

Time series inflow prediction in reservoirs is an im-portant task (probably the most important) in controlling thestorage of a reservoir +us it is a necessary activity tomanage the water resources for various downstream tasksincluding drinking water supply hydropower developmentand irrigation +ere are several ways of predicting theinflow to the systems starting from simplified tank models[24ndash27] to unsteady hydrological routing models [28ndash32]Literature shows these approaches are clustered into em-pirical conceptual physical and data-driven models[33ndash36] However usage of artificial neural networks hasintroduced a reasonably new approach (compared to thetraditional computational models) in predicting catchmentinflows without using many physical catchment data

Forecasting techniques are varied from one catchment toanother catchment depending on the available catchmentcharacteristics In addition to that data availability plays amajor role in the accuracy of the forecasting models [37 38]Furthermore different computational models have differentmodeling accuracies in the forecasting process

Some researchers have combined genetic algorithms andANN to predict the inflows to reservoirs [39] Results fromSedki et al [39] have shown that the hybrid models com-bining genetic algorithms to ANN outperform the tradi-tional ANNs However a well-trained ANN model can still

be a useful tool to predict the inflow to a reservoir at lowcomputational cost In addition ANNs do not need sta-tionary variables to be stationary and normally distributedcompared to classical stochastic modeling [40 41] Fur-thermore ANNs are comparably stable for the variousnoises in the measured data [42] Also ANNs are capable ofmodeling complex and nonlinear hydrological processes[43] having a self-adjusting ability to the data [44] needing alimited knowledge of the hydrological process [45] easyintegration to other models [43] and less demand incomputational cost are few other advantages of ANNs inhydrological processes

Nevertheless inflow predictions using ANNs have somedisadvantages too Nonstationary data have some impacts tothe predictions [46 47] +e extreme events in meteorologymay not be captured in the predictions+erefore the inflowpredictions may misguide the planners and authorities Inaddition the seasonal effects of rainfall can adversely impactthe quality of the predictions [48]

However despite several disadvantages literature givesmany examples of inflow prediction using ANNs [49ndash54]due to their advantages Many researchers have used dif-ferent types of algorithms in predicting inflows usingANNs For example Kisi [55] has used four different al-gorithms including backpropagation conjugate gradientcascade correlation and LevenbergndashMarquardt in theirstreamflow predictions However the usage of ANN topredict inflows in Sri Lanka is quite new Only a handful ofnumber of research has been done to Sri Lanka in the samemanner [56 57] Basnayake et al [56] have developed amodel to forecast the monthly inflow to one of the majorreservoirs in the wet zone of Sri Lanka +ey have usednonlinear autoregressive with exogenous input (NARX-ANN) neural network however they have not presentedtheir model with the projected climatic cases for futureyears Apart from the studies by Basnayake et al [56 57]the authors could not find related research in the context ofSri Lanka In addition the importance of such study isvisible when there are conflicts in the water usages in thedry zone of the country

+erefore it is timely important to conduct a hydro-logical catchment analysis to predict the inflow of the Ira-namadu reservoir and thus to investigate the cultivationcapacity of the Iranamadu reservoir in the upcoming de-velopment projects under the ongoing climate changescenarios +us this paper presents the soft computingtechniques in finding the relationships between the catch-mentrsquos rainfall and the inflow to the Iranamadu reservoir+e conventional method of flow calculation hydrologicalcatchment modeling was not conducted here due to twoimportant reasons unavailability of various data andcomplexity of the real-world hydrological models in theabsence of various required data including a dense networkof rain gauges timelymeasured streamflows soil infiltrationand evaporation +e obtained relationship was then used toforecast the Iranamadu inflow for the years 2021ndash2050 fromthe projected climate scenarios +e future climate data wereextracted from regional climate models using the Coordi-nated Regional Downscaling Experiment (CORDEX)


















300Mankulam

Mankulam

Olumadu

Olumadu

Iranamadu

Iranamadu

250

200

150

100

50

01959 1964 1969 1974 1979

Year

Ave

rage

rain

fall

(mm

)

300

250

200

150

100

50

0

Ave

rage

rain

fall

(mm

)300

250

200

150

100

50

0

Ave

rage

rain

fall

(mm

)

1984 1989 1959 1964 1969 1974 1979Year

1984 1989

1959 1964 1969 1974 1979Year

1984 1989


100000 25000

20000

15000

10000

5000

0

80000

60000

Har

vest

(MT)

Har

veste

d ar

ea (h

a)

40000

20000

02004 2008 2012



(a)

25000

20000

15000

10000

5000

0

Har

veste

d ar

ea (h

a)

100000

80000

60000

Har

vest

(MT)

40000

20000

02004 2008 2012



(b)










μm Pobs(d)( 1113857

μm Phis(d)( 1113857 (2)


μm Pobs(d)( 1113857

μm Phis(d)( 1113857 (3)










80

60

40

20

00

R = 0989


Pred

icte

d in

flow

(MCM

)

60 80

(a)

R = 0989

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(b)

R = 0947

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(c)

R = 0983

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(d)









2200

1850

1500

1150

8002020 2025 2030 2035

Time (years)

Ann

ual r

ainf

all (

mm

)

2040 2045 2050

RCP45RCP85

(a)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(b)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(c)


y = 08265x ndash 00697R2 = 09542

50

40

30

20

10

00 10 20

Actual inflow (MCM)

Pred

icte

d in

flow

(MCM

)

30 40 50









202050

100

150

200

2025 2030 2035Time (years)

2040 2045 2050

Pred

icte

d in

flow

(MCM

)

(a)

2020 2025 2030 2035Time (years)

2040 2045 2050100

150

200

Pred

icte

d in

flow

(MCM

)

(b)


Jan0

100

200

300

400

Rain

fall

(mm

)

500

600




(a)

0

100

200

300

400

Rain

fall

(mm

)




(b)





6 Conclusions


Data Availability


Disclosure




Acknowledgments



References








































































































300Mankulam

Mankulam

Olumadu

Olumadu

Iranamadu

Iranamadu

250

200

150

100

50

01959 1964 1969 1974 1979

Year

Ave

rage

rain

fall

(mm

)

300

250

200

150

100

50

0

Ave

rage

rain

fall

(mm

)300

250

200

150

100

50

0

Ave

rage

rain

fall

(mm

)

1984 1989 1959 1964 1969 1974 1979Year

1984 1989

1959 1964 1969 1974 1979Year

1984 1989


100000 25000

20000

15000

10000

5000

0

80000

60000

Har

vest

(MT)

Har

veste

d ar

ea (h

a)

40000

20000

02004 2008 2012



(a)

25000

20000

15000

10000

5000

0

Har

veste

d ar

ea (h

a)

100000

80000

60000

Har

vest

(MT)

40000

20000

02004 2008 2012



(b)










μm Pobs(d)( 1113857

μm Phis(d)( 1113857 (2)


μm Pobs(d)( 1113857

μm Phis(d)( 1113857 (3)










80

60

40

20

00

R = 0989


Pred

icte

d in

flow

(MCM

)

60 80

(a)

R = 0989

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(b)

R = 0947

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(c)

R = 0983

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(d)









2200

1850

1500

1150

8002020 2025 2030 2035

Time (years)

Ann

ual r

ainf

all (

mm

)

2040 2045 2050

RCP45RCP85

(a)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(b)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(c)


y = 08265x ndash 00697R2 = 09542

50

40

30

20

10

00 10 20

Actual inflow (MCM)

Pred

icte

d in

flow

(MCM

)

30 40 50









202050

100

150

200

2025 2030 2035Time (years)

2040 2045 2050

Pred

icte

d in

flow

(MCM

)

(a)

2020 2025 2030 2035Time (years)

2040 2045 2050100

150

200

Pred

icte

d in

flow

(MCM

)

(b)


Jan0

100

200

300

400

Rain

fall

(mm

)

500

600




(a)

0

100

200

300

400

Rain

fall

(mm

)




(b)





6 Conclusions


Data Availability


Disclosure




Acknowledgments



References



























































































300Mankulam

Mankulam

Olumadu

Olumadu

Iranamadu

Iranamadu

250

200

150

100

50

01959 1964 1969 1974 1979

Year

Ave

rage

rain

fall

(mm

)

300

250

200

150

100

50

0

Ave

rage

rain

fall

(mm

)300

250

200

150

100

50

0

Ave

rage

rain

fall

(mm

)

1984 1989 1959 1964 1969 1974 1979Year

1984 1989

1959 1964 1969 1974 1979Year

1984 1989


100000 25000

20000

15000

10000

5000

0

80000

60000

Har

vest

(MT)

Har

veste

d ar

ea (h

a)

40000

20000

02004 2008 2012



(a)

25000

20000

15000

10000

5000

0

Har

veste

d ar

ea (h

a)

100000

80000

60000

Har

vest

(MT)

40000

20000

02004 2008 2012



(b)










μm Pobs(d)( 1113857

μm Phis(d)( 1113857 (2)


μm Pobs(d)( 1113857

μm Phis(d)( 1113857 (3)










80

60

40

20

00

R = 0989


Pred

icte

d in

flow

(MCM

)

60 80

(a)

R = 0989

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(b)

R = 0947

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(c)

R = 0983

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(d)









2200

1850

1500

1150

8002020 2025 2030 2035

Time (years)

Ann

ual r

ainf

all (

mm

)

2040 2045 2050

RCP45RCP85

(a)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(b)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(c)


y = 08265x ndash 00697R2 = 09542

50

40

30

20

10

00 10 20

Actual inflow (MCM)

Pred

icte

d in

flow

(MCM

)

30 40 50









202050

100

150

200

2025 2030 2035Time (years)

2040 2045 2050

Pred

icte

d in

flow

(MCM

)

(a)

2020 2025 2030 2035Time (years)

2040 2045 2050100

150

200

Pred

icte

d in

flow

(MCM

)

(b)


Jan0

100

200

300

400

Rain

fall

(mm

)

500

600




(a)

0

100

200

300

400

Rain

fall

(mm

)




(b)





6 Conclusions


Data Availability


Disclosure




Acknowledgments



References































































































μm Pobs(d)( 1113857

μm Phis(d)( 1113857 (2)


μm Pobs(d)( 1113857

μm Phis(d)( 1113857 (3)










80

60

40

20

00

R = 0989


Pred

icte

d in

flow

(MCM

)

60 80

(a)

R = 0989

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(b)

R = 0947

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(c)

R = 0983

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(d)









2200

1850

1500

1150

8002020 2025 2030 2035

Time (years)

Ann

ual r

ainf

all (

mm

)

2040 2045 2050

RCP45RCP85

(a)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(b)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(c)


y = 08265x ndash 00697R2 = 09542

50

40

30

20

10

00 10 20

Actual inflow (MCM)

Pred

icte

d in

flow

(MCM

)

30 40 50









202050

100

150

200

2025 2030 2035Time (years)

2040 2045 2050

Pred

icte

d in

flow

(MCM

)

(a)

2020 2025 2030 2035Time (years)

2040 2045 2050100

150

200

Pred

icte

d in

flow

(MCM

)

(b)


Jan0

100

200

300

400

Rain

fall

(mm

)

500

600




(a)

0

100

200

300

400

Rain

fall

(mm

)




(b)





6 Conclusions


Data Availability


Disclosure




Acknowledgments



References



























































































80

60

40

20

00

R = 0989


Pred

icte

d in

flow

(MCM

)

60 80

(a)

R = 0989

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(b)

R = 0947

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(c)

R = 0983

80

60

40

20

0

Pred

icte

d in

flow

(MCM

)


60 80

(d)









2200

1850

1500

1150

8002020 2025 2030 2035

Time (years)

Ann

ual r

ainf

all (

mm

)

2040 2045 2050

RCP45RCP85

(a)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(b)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(c)


y = 08265x ndash 00697R2 = 09542

50

40

30

20

10

00 10 20

Actual inflow (MCM)

Pred

icte

d in

flow

(MCM

)

30 40 50









202050

100

150

200

2025 2030 2035Time (years)

2040 2045 2050

Pred

icte

d in

flow

(MCM

)

(a)

2020 2025 2030 2035Time (years)

2040 2045 2050100

150

200

Pred

icte

d in

flow

(MCM

)

(b)


Jan0

100

200

300

400

Rain

fall

(mm

)

500

600




(a)

0

100

200

300

400

Rain

fall

(mm

)




(b)





6 Conclusions


Data Availability


Disclosure




Acknowledgments



References



























































































2200

1850

1500

1150

8002020 2025 2030 2035

Time (years)

Ann

ual r

ainf

all (

mm

)

2040 2045 2050

RCP45RCP85

(a)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(b)

2200

1850

1500

1150

800

Ann

ual r

ainf

all (

mm

)

2020 2025 2030 2035Time (years)

2040 2045 2050

RCP45RCP85

(c)


y = 08265x ndash 00697R2 = 09542

50

40

30

20

10

00 10 20

Actual inflow (MCM)

Pred

icte

d in

flow

(MCM

)

30 40 50









202050

100

150

200

2025 2030 2035Time (years)

2040 2045 2050

Pred

icte

d in

flow

(MCM

)

(a)

2020 2025 2030 2035Time (years)

2040 2045 2050100

150

200

Pred

icte

d in

flow

(MCM

)

(b)


Jan0

100

200

300

400

Rain

fall

(mm

)

500

600




(a)

0

100

200

300

400

Rain

fall

(mm

)




(b)





6 Conclusions


Data Availability


Disclosure




Acknowledgments



References






























































































202050

100

150

200

2025 2030 2035Time (years)

2040 2045 2050

Pred

icte

d in

flow

(MCM

)

(a)

2020 2025 2030 2035Time (years)

2040 2045 2050100

150

200

Pred

icte

d in

flow

(MCM

)

(b)


Jan0

100

200

300

400

Rain

fall

(mm

)

500

600




(a)

0

100

200

300

400

Rain

fall

(mm

)




(b)





6 Conclusions


Data Availability


Disclosure




Acknowledgments



References


























































































6 Conclusions


Data Availability


Disclosure




Acknowledgments



References
































































































































































































Documents

InflowForecastofIranamaduReservoir,SriLanka,under