***proof of your article attached, please...

UNCO

RREC

TED

PROO

FS

JOHN WILEY & SONS, LTD., THE ATRIUM, SOUTHERN GATE, CHICHESTER P019 8SQ, UK

***PROOF OF YOUR ARTICLE ATTACHED, PLEASE READ CAREFULLY***

After receipt of your corrections your article will be published initially within the online version of the journal.

PLEASE AIM TO RETURN YOUR CORRECTIONS WITHIN 48 HOURS OF RECEIPT OF YOUR PROOF, THIS WILL ENSURE THAT THERE ARE NO UNNECESSARY DELAYS IN THE PUBLICATION OF YOUR ARTICLE

READ PROOFS CAREFULLY

ONCE PUBLISHED ONLINE OR IN PRINT IT IS NOT POSSIBLE TO MAKE ANY FURTHER CORRECTIONS TO YOUR ARTICLE

This will be your only chance to correct your proof Please note that the volume and page numbers shown on the proofs are for position only

ANSWER ALL QUERIES ON PROOFS (Queries are attached as the last page of your proof.)

Please annotate this file electronically and return by email to the production contact as detailed in the covering email. Guidelines on using the electronic annotation tools can be found at the end of the proof. If you are unable to correct your proof using electronic annotation, please list all corrections and send back via email to the address in the covering email, or mark all corrections directly on the proofs and send the scanned copy via email. Please do not send corrections by fax or post.

Acrobat Reader & Acrobat Professional You will only be able to annotate the file using Acrobat Reader 8.0 or above and Acrobat Professional. Acrobat Reader can be downloaded free of charge at the following address:http://www.adobe.com/products/acrobat/readstep2.html

CHECK FIGURES AND TABLES CAREFULLY

Check size, numbering, and orientation of figures All images in the PDF are downsampled (reduced to lower resolution and file size) to facilitateInternet delivery. These images will appear at higher resolution and sharpness in the printedarticleReview figure legends to ensure that they are complete Check all tables. Review layout, title, and footnotes

COMPLETE COPYRIGHT TRANSFER AGREEMENT (CTA) if you have not already signed one

Please send a scanned signed copy with your proofs by e-mail. Your article cannot bepublished unless we have received the signed CTA

OFFPRINTS

Free access to the final PDF offprint or your article will be available via Author Services only.Please therefore sign up for Author Services if you would like to access your article PDFoffprint and enjoy the many other benefits the service offers.

Additional reprint and journal issue purchases

Should you wish to purchase additional copies of your article, please click on the link and follow the instructions provided: http://offprint.cosprinters.com/cos/bw/Corresponding authors are invited to inform their co-authors of the reprint options available. Please note that regardless of the form in which they are acquired, reprints should not beresold, nor further disseminated in electronic form, nor deployed in part or in whole in any marketing, promotional or educational contexts without authorization from Wiley. Permissions requests should be directed to mailto: [email protected]

UNCO

RREC

TED

PROO

FS

HYDROLOGICAL PROCESSESHydrol. Process. 24, 0–0 (2010)Published online in Wiley Online Library(wileyonlinelibrary.com) DOI: 10.1002/hyp.7926

Large-scale modelling of channel flow and floodplaininundation dynamics and its application

to the Pantanal (Brazil)

Adriano Rolim da Paz,1,2* Walter Collischonn,1 Carlos E. M. Tucci1 and Carlos R. Padovani31 Instituto de Pesquisas Hidráulicas, Universidade Federal do Rio Grande do Sul, Av. Bento Gonçalves 9500, Agronomia, CEP 91501-970, Porto

Alegre-RS, Brazil2 EMBRAPA Monitoramento por Satélite, Av. Soldado Passarinho 303, Fazenda Chapadão, CEP 13070-115, Campinas-SP, Brazil

3 EMBRAPA Pantanal, Rua 21 de Setembro, 1880, Bairro Nossa Senhora de Fátima, CEP 79320-900, Corumbá-MS, Brazil

Abstract:

For large-scale sites, difficulties for applying coupled one-dimensional (1D)/2D models for simulating floodplain inundationmay be encountered related to data scarcity, complexity for establishing channel–floodplain connections, computationalcost, long duration of floods and the need to represent precipitation and evapotranspiration processes. This paper presentsa hydrologic simulation system, named SIRIPLAN, developed to accomplish this aim. This system is composed by a 1Dhydrodynamic model coupled to a 2D raster-based model, and by two modules to compute the vertical water balance overfloodplain and the water exchanges between channel and floodplain. Results are presented for the Upper Paraguay River Basin(UPRB), including the Pantanal, one of the world’s largest wetlands. A total of 3965 km of river channels and 140 000 km2 offloodplains are simulated for a period of 11 years. Comparison of observed and calculated hydrographs at 15 gauging stationsshowed that the model was capable to simulate distinct, complex flow regimes along main channels, including channel-floodplain interactions. The proposed system was also able to reproduce the Pantanal seasonal flood pulse, with estimatedinundated areas ranging from 35 000 km2 (dry period) to more than 120 000 km2 (wet period). Floodplain inundation mapsobtained with SIRIPLAN were consistent with previous knowledge of Pantanal dynamics, but comparison with inundationextent provided by a previous satellite-based study indicates that permanently flooded areas may have been underestimated.The results obtained are promising, and further work will focus on improving vertical processes representation over floodplainsand analysing model sensitivity to floodplain parameters, time step and precipitation estimates uncertainty. Copyright 2010John Wiley & Sons, Ltd.

KEY WORDS hydrologic modelling; hydrodynamic model; Pantanal; lateral water exchange

Received 18 March 2010; Accepted 11 October 2010

123456789

1011121314151617181920212223

INTRODUCTION

Mathematical models have been developed and appliedfor simulating the hydrologic regime of rivers since thenineteenth century (Chow, 1959; Abbott, 1979; Cungeet al., 1981). The common approach consists of assumingthat the flow is one-dimensional (1D) along the longitu-dinal axis of the river and employing the Saint Venant’sdynamic and continuity equations for flow routing. Theseequations are used in their complete form (hydrodynamicmodel) or disregarding some terms, which give rise tothe diffusive, kinematic or storage models. The choiceof which model, approach and discretization to use isdependent on several factors such as the characteristicsof the study area, available data sets, purposes of thestudy, available time, computational and human resources(Fread, 1992).

When dealing with rivers with floodplains, the twousual approaches are to consider the 1D model with

* Correspondence to: Adriano Rolim da Paz, Instituto de PesquisasHidráulicas, Universidade Federal do Rio Grande do Sul, Av. BentoGonçalves 9500, Agronomia, CEP 91501-970, Porto Alegre-RS, Brazil.E-mail: [email protected]

2425262728293031323334353637383940414243444546

extended cross sections representing both main channeland floodplain or to consider explicitly storage areasconnected to the 1D model representing major wateraccumulation regions during floods. These methods areable to reproduce the main channel flow regime in asatisfactory way for most cases. Inundation maps maybe further derived from the model results by interpolatingcross sections of water levels and using a digital elevationmodel (DEM). However, if the study aims at representingthe floodplain inundation patterns, these methods maynot be suitable and a more recent approach consists ofcoupling a 1D model for simulating the main channelflow and a 2D model for simulating floodplain inundation(Verwey, 2001; Gillan et al., 2005; Hunter et al., 2007;Chatterjee et al., 2008).

Floodplain inundation plays a key role for severalecological processes and phenomena, such as ecosystemproductivity, species occurrence and distribution andnutrient and sediment dynamics (Junk et al., 1989; Poffet al., 1997; Postel and Richter, 2003). Hence, beingable to simulate the spatial inundation patterns throughmathematical modelling provides a valuable tool to watermanagement and prediction of climate change effects as

Copyright 2010 John Wiley & Sons, Ltd.

UNCO

RREC

TED

PROO

FS

2 A. R. PAZ ET AL.

123456789

1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859

well the effects of human interventions such as waterwithdrawals, embankments, dykes and dredging projects.

In the 1D/2D coupled approach, the floodplain maybe modelled by a full 2D hydrodynamic model (depth-averaged Navier–Stokes equations) or by simpler meth-ods such as 2D diffusive and kinematic approximations.Most of the latter are regular grid models, which arecommonly referred as raster-based models.

Modelling floodplain with a 2D hydrodynamic codemay be infeasible due to numerical instabilities related tosmall water depths and the wetting and drying process aswell as excessive computational costs. The use of raster-based models overcomes these difficulties and providesa way to work with a large number of floodplain gridelements. Additionally, this approach has the advantagesof taking into account the spatial variability of floodplainphysical characteristics (elevation and roughness) andof being easily integrated into a geographic informationsystem (GIS). Reasonable results have been obtained byseveral authors with this modelling approach in terms ofreproducing floodplain spatial inundation patterns (Horrittand Bates, 2001a; Bates et al., 2006; Wilson et al., 2007).

The majority of literature examples of river-floodplainmodelling using the 1D/2D coupled approach encom-passes relative small-scale sites (single river reaches oflength less than 100 km), for which there was largeamount of available data such as high-resolution DEMand inundation maps for calibrating model results (Hor-ritt and Bates, 2001a; Bradbrook et al., 2004; Bates et al.,2006; Tayefi et al., 2007). The few exceptions include thestudy reported by Biancamaria et al. (2009), which mod-elled a single reach of 900 km length of the Ob river(Siberia), and the studies carried out by Wilson et al.(2007) and Trigg et al. (2009), which modelled a 285 kmreach of the main stem of the Amazon (Solimões) riverand a 107 km reach of Purus tributary. If the study sitecomprises an even larger and complex network of chan-nels, junctions and floodplains (over hundreds of squarekilometers), difficulties may be encountered related todata scarcity and complexity for establishing main chan-nel and floodplain connections.

Additionally, the flood pulse may last for months longin large-scale floodplains, which considerably increasethe computational cost by necessitating more modelgrid elements and model time steps. Moreover, forsimulating these long duration floods the representationof the vertical water processes such precipitation andevapotranspiration may be required (Wilson et al., 2007).

In spite of the difficulties for modelling large-scalerivers and floodplains, this is the major scale of interestfor assessing how climate change and variability willaffect water resources. As an increase in accuracy andreliability of flow and inundation predictions is desirablefor better decisions concerning land use and watermanagement in light of climate scenarios, it motivates thedevelopment and improvement of methods for large-scalehydrologic modelling.

This paper presents a hydrologic simulation system,named SIRIPLAN, developed for large-scale river and

60616263646566676869707172737475767778798081828384858687888990919293949596979899

100101102103104105106107108109110111112113114115116117118

floodplains drainage networks. This simulation systemis based on coupling a 1D hydrodynamic model toa 2D raster model and considering the precipitation,evapotranspiration and infiltration processes over thefloodplain. Results are presented from the applicationof the SIRIPLAN to the Upper Paraguay River Basin(UPRB), including the Pantanal, one of the world’slargest wetlands. Results are evaluated by comparingobserved and calculated hydrographs at available gaugingstations and by comparing seasonal inundation areas andinundation patterns provided by previous satellite-basedstudies.

THE SIRIPLAN HYDROLOGIC SIMULATIONSYSTEM

Overview

The SIRIPLAN hydrologic simulation system is com-posed by a 1D hydrodynamic model coupled to a 2Draster-based inundation model (Figure 1). The 1D modelsimulates the flow routing along the river drainage sys-tem, considering cross sections restricted to the mainchannels. The raster-based model simulates the wateraccumulation and the 2D propagation of inundation overthe floodplains. A water exchange scheme is used to sim-ulate the interactions between channel and floodplain. Ifthe water level in a cross section of the main channel risesabove the levee, it spills over and inundates the flood-plain. Analogously, if the inundation propagation overfloodplain reaches the main channel pathway, water istransferred to the channel.

Additionally, the vertical processes of precipitation,evapotranspiration and infiltration are simulated by a thirdmodule, coupled with the raster-based model. Water con-tributions from upstream of the modelled river drainagesystem are considered as boundary conditions set using

1D hydrodynamicmodel (IPH4)

Raster-basedinundation model

Vertical balanceover floodplain

1D flow routing along mainchannels

2D inundation modeling overfloodplain

Simulation of vertical hydrologicprocesses over floodplain

Connectionmodule

Updating of vertical input/ouput overfloodplain

Connectionmodule

Water exchanges between mainchannels and floodplains

Rainfall-runoff modelOR observed data

Precipitation andevapotranspiration data

Meteorologicalboundary conditions

SIRIPLAN

Figure 1. Conceptual overview of the SIRIPLAN hydrologic simulationsystem

Copyright 2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 0–0 (2010)

UNCO

RREC

TED

PROO

FS

LARGE-SCALE RIVER AND FLOODPLAIN MODELLING 3

123456789

101112131415161718192021222324252627282930313233343536373839404142

observed discharge data or by off-line coupling of arainfall-runoff hydrologic model.

Channel flow routing

Flow routing along main channels is simulated withthe 1D hydrodynamic model called IPH4 (Tucci, 1978).This model solves the full Saint Venant equations througha finite difference method, with an implicit scheme basedon a modified version of the Gauss elimination process:

∂h

∂tC 1

b

∂Q

∂xD q �1�

∂Q

∂tC ∂

∂t

(Q2

A

)C gA∂h

∂xC gA�Sf � S0� D 0 �2�

where h is the water level, t is time, Q is the discharge,x is the longitudinal distance along the river, b and A arethe cross section width and area, respectively, g is thelocal gravitational celerity, q is the lateral contributionto discharge per unit of distance, S0 is the channelbotton slope and Sf is the energy friction slope, whichis parameterized through Manning resistance equation.

Cross-section data represented in the IPH4 model isrestricted to the level which characterizes the transitionbetween main channel and floodplain (levees). For eachriver reach between two cross sections, length and slopemust be specified. Manning coefficients may assume dis-tinct values for each river reach, and may also be consid-ered variable as a function of the water level in a givencross section. The discharge exchanged between mainchannel and floodplains is considered as lateral contribu-tion in the continuity equation (term q in Equation (1)).

Floodplain inundation modelling

The floodplain model is a raster-based inundationmodel, which was developed following the approach ofthe LISFLOOD-FP model (Bates and De Roo, 2000;Horritt and Bates, 2001b), but with adaptations mainlyconcerning the water exchange between channel andfloodplain, flow among floodplain elements, water storage

434445464748495051525354555657585960616263646566676869707172737475767778798081828384

in soil reservoirs and water input/loss on floodplain dueto vertical water balance.

Floodplain is discretized by a regular grid of intercon-nected elements, which may change flow with neighbour-ing elements and with the main channel, in the case ofelements directly connected to the channel (Figure 2a).The volume variation along time in a given element ofthe raster model is the following:

V

tplanD Qup C Qdown C Qleft C Qright C Qcf

C Qvert C Qres �3�

where V is the volume variation during time intervaltplan; Qup, Qdown, Qleft and Qright are the dischargesbetween the element and its up, down, left and rightneighbours, respectively; Qcf is the discharge betweenchannel and floodplain element; Qvert is the result of thevertical water balance and Qres represents the volume ofwater flowing to the soil reservoir.

A numerical scheme explicit on time and progressiveon space is used to solve Equation (3), considering thewater level represented in the center of the element andthe exchanges in its interfaces (Figure 2b). As a result, thewater level in the time instant t C tplan in a floodplainelement (i, j) is determined by:

tCthi,j D thi,j C

�tQi�1,jx � tQi,jx C tQi,j�1y � tQi,jyC tQi,jcf� Ð tplan

x Ð yC thi,jvert C thi,jres �4�

where thi,j is the water level in time instant t, tQi,jx isthe discharge in x direction between elements i, j andi C 1, j; tQi,jy is the discharge in y direction betweenelements i, j and i, j C 1; thi,jvert is the result of the verticalwater balance and thi,jres is the available volume of soilreservoir, both expressed in water depth; x and yare the element dimensions in the x and y directions,respectively.

main channel

elements of the floodplain model

elements connected to the main channel

(a)

i

j

j-1

j+1

i+1i-1

hi,j

Qxi,j

Qyi,j

Qyi,j-1

Qxi-1,j

(b)

Bch

Lch

(c)

Zb1Zb2

Zw2Zw1

hflow

flow

1 2

Figure 2. (a) Floodplain elements of the raster-based model; (b) numerical discretization of water level and discharges between elements of thefloodplain, which are calculated through linkage channels of width Bch and length Lch and (c) indication of hflow between two elements (Zw and

Zb refer to water level and botton elevation, respectively), where hflow D Max(Zw1,Zw2)-Max(Zb1,Zb2) (adapted from Bates et al., 2005)


UNCO

RREC

TED

PROO

FS

4 A. R. PAZ ET AL.

123456789

10111213141516171819202122232425262728293031323334353637

In the soil reservoir scheme, a floodplain element isinundated, i.e. with surface water accumulation, only afterthe soil reservoir is full (Figure 3). The term hres is givenby:

hres D hsub � Hsmax �5�where hsub is the current water content of the soilreservoir, which has a maximum capacity of Hsmax(model parameter), both variables being expressed inwater depth; hres always assumes non-positive values,varying from hres D �Hsmax when the reservoir is emptyto hres D 0 when it is full.

If the result of the water balance in a floodplain element(Equation (4)) is positive, the soil reservoir is filled andthere is surface water in this element. On the contrary,a negative result means that the element was dried (interms of surface water). The available water content inthe soil reservoir is updated as follows:

if tCthi,j > 0 ) tCthi,jres D 0 �6�if tCthi,j < 0 )

tCthi,jres D tCthi,j, if∣∣∣ tCthi,j∣∣∣ < Hsmax

tCthi,jres D �Hsmax, if∣∣∣ tCthi,j∣∣∣ > Hsmax

tCthi,j D 0�7�

The discharge between two neighbour floodplain ele-ments is determined by Manning equation with a numericand spatial discretization similar to the used by Batesand De Roo (2000). However, we consider that the flowbetween each two elements occurs along straight chan-nels of width Bch and length Lch (Figure 2c), and thus thedischarge is given by:

tQi,jx D šth5/3fluxoni,j

∣∣∣ thi,j � thiC1,j∣∣∣Lch

1/2

Ð Bch �8�

38394041424344454647484950515253545556575859606162636465666768697071727374

where tQi,jx is the discharge in the x direction betweenelements (i, j) and (i C 1, j) in time instant t; ni,j isManning roughness of the channel linking these elementsand thflow is the water depth available to the flowbetween these elements; flow in y direction is determinedanalogously.

The water depth hflow is defined as the differencebetween the highest water level and the highest bot-ton elevation between the two floodplain elements(Figure 2c), following Horritt and Bates (2001a) andBates et al. (2005).

When modelling large-scale floodplains, model dis-cretization may result in elements with dimensions ofhundreds or thousands of meters to reduce computa-tional cost. If discharge along the floodplain is calculatedconsidering the flow spilling over the whole elementwidth, small differences in the water level may gener-ate huge and unrealistic volumes of water exchangedbetween two elements, causing numerical instabilities andartificially accelerating the inundation propagation. Theadoption of channels with controlled dimensions to rep-resent the hydraulic linkage between each two floodplainelements aims at overcoming this problem. In the flowequation between elements of the floodplain, there arethree parameters related to the linkage channel (Man-ning roughness, width and channel), which may be com-bined into only one, called hydraulic conductivity fac-tor (fhc) (Equation (9)). Albeit indeed inundation overlarge, vegetated floodplains such as Pantanal may prop-agate along preferential pathways, the disadvantage ofthe proposed approach is the increase in the number ofmodel parameters and the difficulty to parameterize themphysically. This may cause parameter equifinality, i.e.different parameter sets leading to same results (Bevenand Freer, 2001). Further study may focus on evaluatingmodel sensitivity to these parameters and the associated

Zf

(a) (b) (c) (d)

floodplain wetting process

Hsmax

element dry;reservoir dry

element dry;reservoirwith water

there is awater demandequals toHsmax;element hasno watercontent tolose

water demandbetween0 and Hsmax;element maylose waterfrom the soilreservoir (ET)but no horizon-tal flow occurs

element dry;reservoir filled

element wet;reservoir filled

floodplain drying process

Za

hahres

water demandhres = 0;element maylose waterfrom the soilreservoir (ET)but no horizon-tal flow occurs

water demandhres = 0;element maylose surfacewater andgenerate hori-zontal flow

elementsurface

botton ofsoil reservoir

hsub

Figure 3. Wetting [(a)–(d)] and drying [(d)–(a)] processes of a floodplain element of the raster model (Zf is floodplain elevation; Za is water level;ha is surface water depth over the element; hsub is water depth of soil reservoir; hres is the available volume of soil reservoir, which has a maximum

capacity equals to Hsmax)


UNCO

RREC

TED

PROO

FS


123456789

1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859

uncertainties.

fi,jhc DBi,jch

ni,j√

Li,jch

�9�

Vertical water balance on floodplain

The vertical water balance on each floodplain elementis performed as a balance between precipitation andevapotranspiration. This balance is updated at a specifictime step (tvert) (Figure 4), which is commonly severaltimes greater than time steps used in 1D and 2D models.At each tvert, this simple water balance is calculated fora given floodplain element (i, j):

tCthi,jvert D tCtPi,j � tCtETi,jactual �10�

where P is precipitation, ETactual is the actual evapotran-spiration and hvert is the resultant of this balance, all ofthem expressed in terms of water depth.

If hvert > 0, it represents a source of water to the waterbalance of the element in the 2D model (Equation (4)),while a negative value means a sink (definite loss) ofwater from the modelling system. As tvert >> tplan,the result of the vertical balance is considered constantalong the following npv number of floodplain timesteps, where npv D tvert/tplan, but after convertingto corresponding units by hvert D hvert/npv.

Actual evapotranspiration is calculated according towet or dry condition of the floodplain element in eachtvert. If the element has surface water, actual evapotran-spiration occurs at the maximum rate equal to potentialevapotranspiration (Equation (11)). If the element is dry,actual evapotranspiration is less than the potential rate,being linearly proportional to water content of the soil

Main channel simulation withthe1D hydrodynamic model

along 1∆tch

Floodplain simulationwith the raster modelalong np.∆tfl(=1∆tch)

Updating time instantt = t + ∆tch

Determination of flowexchanges between channel

and floodplain (Qcf)

Models initializationt = 0

Update of the vertical waterbalance in the floodplain

Completed 1 ∆tvert?

YesNo

Figure 4. Scheme of coupled running of hydrodynamic and raster inun-dation models and vertical water balance

60616263646566676869707172737475767778798081828384858687888990919293949596979899

100101102103104105106107108109110111112113114115116117118

reservoir (Equation (12)).

if thi,j > 0 ) tCtETi,jactual D tCtETi,jpot �11�

if thi,j D 0 ) tCtETi,jactual D tCtETi,jpot

Ð1 �

∣∣∣ thi,jres∣∣∣Hsmax

�12�

Channel–floodplain water exchanges

Every floodplain element located under the mainchannel longitudinal axis is connected with it. Waterexchanges between channel and floodplain are deter-mined as a function of the difference between waterlevels. For the points located between two cross sectionsof the main channel, the water level is calculated by alinear approximation.

Occurrence of flow between channel and floodplain ina given location is triggered by the condition of waterlevel in floodplain and/or main channel higher than thespill elevation (Zspill). This elevation is the maximumvalue between channel levee height and floodplain bottomelevation.

When the water level in the main channel or in thefloodplain reaches Zspill, there is hydraulic connectionand flow occurs. This flow is calculated using simpleor flooded weir-type equations. Analogously to the dis-charge between floodplain elements, if the weir width isconsidered equal to the element width, unrealistic exag-gerated flow may be calculated for small water depthsover the weir in case of elements with large dimensions.Therefore, the weir width is considered a model parame-ter, usually taken in the range 10–100 m, which may beregarded as the typical width values over which occurslateral flows in large natural rivers. As previously statedregarding parameters related to channels linking flood-plain elements, considering the weir width as a modelparameter may lead to equifinality and increase the uncer-tainties. Further study will evaluate this issue, investigat-ing model sensitivity to each parameter.

A decoupled 1D/2D time-step approach is considered(Trigg et al., 2009), in which different time steps are setto the 1D and 2D models. The 1D time step (tchan)is usually several times greater than the 2D time step(tplan), as the 1D model uses an implicit numericscheme while the 2D model is explicitly solved. Thus,the 1D model is run by 1tchan and then the 2D modelis run by np times tplan, where np D tchan/tplan.After a time interval of tchan, the water exchanges(Qcf) between channel (1D model) and floodplain (2Dmodel) are calculated. For the channel, Qcf is convertedinto lateral contribution to discharge per unit of distancefor calculation of the continuity equation (Equation (1))at the next tchan. For the floodplain, Qcf is directlyused into the water level updating equation (Equation (4))throughout a time interval of tchan, i.e. for the next nptplan.


UNCO

RREC

TED

PROO

FS

6 A. R. PAZ ET AL.

123456789

1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859

Code and parallelization

The SIRIPLAN hydrologic simulation system wasdeveloped using FORTRAN 90 programming languageand OpenMP (Open specifications for Multi-Processing)Application Programming Interface (API). The OpenMPrepresents a collection of directives, library routines andenvironment variables that enables programs to run inparallel on shared memory processors (Hermanns, 2002;Chapman et al., 2008). The main advantages of thisapproach relative to other parallel techniques are theease of implementation and requirements of minimalmodification to the code. Recently, Neal et al. (2009)implemented a parallel version of the LISFLOOD-FPmodel using OpenMP, achieving parallel efficiencies ofup to 0Ð75 on four and eight processor cores.

Two loops of the raster inundation model were par-allelized through OpenMP: the calculation of dischargebetween floodplain elements and the calculation of waterdepth in each element (general water balance). The 1Dhydrodynamic model has an implicit numerical scheme,and tests for parallelizing its code with OpenMP hasproven not to be advantageous in terms of run-time reduc-tion (Paiva, 2009).

INPUT DATA REQUIREMENTS ANDPREPARATION

Main channel data

For the hydraulic modelling of channel flow, datarequirements includes channel vector lines, length andslope, cross section profiles and boundary conditions.Among these data, the profiles are the most difficult toobtain. To overcome this issue, a simple linear schemeis adopted for cross-section profiles interpolation whennecessary. Given an upstream and a downstream sectionwith available profiles, for each intermediate cross sectionto be created, the horizontal and vertical location of itsith point is determined through linear interpolation of theith upstream and downstream points.

Main channel georeferenced vector lines may beobtained from available maps or by digitizing satelliteimages, while length and slope of main channels arederived from cross-section data and channel vector lines,taking into account a floodplain DEM as auxiliary data.

Floodplain data

The raster-based model requires a floodplain mask anda DEM to represent floodplain topography. The maskdefines the modelled domain, which is established basedon the main channel network, floodplain topography andcontributing drainage areas of the boundary conditions ofthe channels. As a no flow boundary condition is imposedto the floodplain in the raster model, the floodplain maskmust comprise the full extent of the inundation area.Areas which certainly are not flooded and which do notsignificantly contribute to flooding may be excluded fromfloodplain domain to reduce computational cost.

60616263646566676869707172737475767778798081828384858687888990919293949596979899

100101102103104105106107108109110111112113114115116117118

Additionally, precipitation and potential evapotranspi-ration data are required for the vertical water balance onfloodplain. Point specific data such as rainfall gaugingstation observations or data provided by other sourcessuch as precipitation estimates from atmospheric mod-els are interpolated to the raster model grid using theinverse distance square method. This procedure is carriedout before simulation to reduce model run time. Thesedata are required with a discretization on time equalto tvert. Alternatively, seasonal monthly estimates ofpotential evapotranspiration may be used if more detaileddata are not available.

Channel–floodplain connection

The largest effort on input data preparation involvesestablishing the topological connections between channeland floodplain discretization elements. This is not a trivialtask when dealing with several tributaries, junctionsand hundreds of cross sections, and where the largedimensions of the floodplain elements contrast withrelative small channel meanders.

The main channel drainage network must be repre-sented in terms of raster model grid elements, identifyingwhich floodplain elements are connected to each chan-nel reach, and which cross section or intermediate pointof the reach is connected to each element. A four-stepprocedure was developed to accomplish this task.

The first step is the conversion of vector channelnetwork to raster format with spatial resolution and extentequal to the floodplain discretization (Figure 5a). Theresulting image is composed by pixels representing ornot the channel network (Figure 5b).

(b)

(c)

(d)

(a)

Figure 5. (a) Main channel vector drainage (VD); (b) VD converted toraster (grey pixels); (c) flow directions and (d) raster drainage with aunique pixel-to-pixel flow path (dark pixels were excluded from the

original raster drainage)


UNCO

RREC

TED

PROO

FS


123456789

1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859

Derivation of flow directions is the second step(Figure 5c). Considering the set of non-zero pixels asa mask, the direction water flows out of each pixelis determined based on floodplain DEM, through thewell-known D8 (deterministic eight-neighbour) algorithm(Mark, 1984; Burrough and McDonnel, 1998; Jenson andDomingue, 1988). This algorithm approximates the localflow direction by the direction of the steepest downhillslope within a 3 ð 3 window of pixels over a raster DEM.As this algorithm has a tendency of generating paralleldrainage paths on flat areas, a stochastic factor as pro-posed by Fairfield and Leymarie (1991) was introducedto lessen this problem.

Thirdly, starting from the most upstream pixel of eachchannel reach, trace the downstream path following flowdirections and mark every pixel reached. These markedpixels form the main channel network representation interms of a unique pixel-to-pixel flow path. Pixels non-marked are eliminated from the raster representation ofmain channels (Figure 5d).

Every floodplain element corresponding to the rasterpixel-to-pixel channel network is connected with mainchannel, while none of the other elements are connected.The fourth step is the identification of to which crosssection each element is associated.

The cross sections with available profile and geo-graphic coordinate data are associated to the pixel corre-sponding to these coordinates. For the interpolated crosssections, albeit their longitudinal position along the mainchannels is known, a rescaling procedure is performedbefore locating them, due to the tendency of underesti-mating distances on a coarse-resolution raster representa-tion of meandering channel networks (Fekete et al., 2001;Paz et al., 2008).

The distances along the raster channel representationare measured between each of the cross sections alreadylocated. The flow path is followed pixel by pixel,summing a distance equal to pixel side for an orthogonalstep and equal to 1Ð414 times pixel side for a diagonalstep. For each reach defined by two of these crosssections, the ratio between the distances measured onthe raster and on the vector drainages is calculated. Thisratio is applied to convert the longitudinal position alongthe main channel of the interpolated cross sections intodistances along the raster channel representation, definingthe location of these sections.

EXAMPLE OF APPLICATION: UPRB

Site description and simulation period

The study site comprises the Pantanal area of theUPRB that has an estimated drainage area of600 000 km2, extending over three South American coun-tries (Figure 6): Brazil, Paraguay and Bolivia. The UPRBis part of the La Plata basin and has three distinctregions: Planalto (260 000 km2), Pantanal (140 000 km2)and Chaco (200 000 km2). The Planalto region encom-passes the uplands of the basin mainly in the North and

60616263646566676869707172737475767778798081828384858687888990919293949596979899

100101102103104105106107108109110111112113114115116117118

East portions. Located in the West part of the UPRB, theChaco is a region characterized by low annual rainfalland an endorheic and undefined drainage system.

The Pantanal region is located in the central portion ofthe UPRB and presents very low and flat relief, with acomplex drainage system. Rivers seasonally inundate thefloodplains and flood waters create an intricate drainagesystem, including vast lakes, divergent and endorheicdrainage networks. Annual rainfall is less than thepotential evaporation and drainage is very slow becauseof shallow gradients (Bordas, 1996; Tucci et al., 1999).

The Pantanal region was modelled with the SIRIPLANhydrologic simulation system, considering the contribu-tion of the Planalto area as boundary condition, as flood-plain inundation is negligible in this part of the basin. TheChaco region was not modelled due to data scarcity andbecause its contribution to Paraguay River is consideredinsignificant (Tucci et al., 2005). A period of 11 yearsand 4 months from 1 September 1995 to 31 December2006 was selected for simulation, as this is a more recentperiod with reliable available data (žTable I). AQ2

The Pantanal is considered one of the largest wet-lands of the world, with extraordinary biodiversity (Harriset al., 2005) and of great global ecologic value (Pottand Pott, 2004; Junk et al., 2006). Modelling its hydro-logic regime and floodplain dynamics is imperative forunderstanding, predicting and mitigating possible effectsof anthropogenic activities that currently threaten itsintegrity, such as dam building, agriculture and cattle rais-ing (Tucci and Clarke, 1998; Hamilton, 1999; Hamiltonet al., 2002; Da Silva and Girard, 2004; Junk and Cunha,2005).

1D hydrodynamic model application

The river drainage system modelled with the 1Dhydrodynamic model covers 3965 km of river channels:1250 km of the Paraguay River and 2715 km of its maintributaries. The flow path of each channel was obtainedby manually digitizing Landsat7 ETMC satellite images.

For the Paraguay River, a total of 288 detailed cross-section profiles was available, with distances betweenconsecutive profiles varying from 0Ð5 to 10 km. On thecontrary, only 19 profiles were available for all the trib-utaries together and a linear interpolation procedure wasperformed to generate profiles at about 5 km intervals.Further information concerning river morphology andslopes available in former studies (DNOS, 1974; Brasil,1997; Tucci et al., 2005) as well as elevation valuesextracted from SRTM-90m DEM were used as auxil-iary data for the vertical positioning of cross sections.Detailed description of data preparation for cross sectionsis presented in Paz et al. (2010).

Streamflow gauging stations with available observeddischarge time series were defined as the upstreamboundary conditions of the 1D hydrodynamic model.Missing data were replaced by values calculated withthe distributed hydrologic model MGB-IPH (Collischonnet al., 2007). This model was previously applied and


UNCO

RREC

TED

PROO

FS

8 A. R. PAZ ET AL.

-55° W

-55° W

-60° W

-60° W

-15°

S-2

0° S

0 200 400km100

Argentina

Paraguay

Brazil

Bolivia

Para

na ri

ver

a

b

c

f

g

hi

de

j

123 5467

8

11

1213

1415

16

1718

19

910

Planalto

Pantanal

Chaco

Country border

Modeled channel network

Modeled floodplain

Control point

Boundary condition

Piquiri r.

Cuiab

á r.

S.Lourenço r.

Aquid. r.

Miranda r.

Taquari r.

Negro

r.

Par

agua

y r.

Jauru r.

Boundary conditionsa Cuiabáb A. C. Grandec S. Jerônimod P. Esperidiãoe Cáceresf Coximg P. Bocaínah Aquidauanai Mirandaj Upstream of Apa River

Control points1 B. Melgaço2 P. Cercado3 S. João4 I. Camargo5 S. J. Borireu6 S. J. Piquiri7 P. Taiamã8 P. Alegre9 S. Gonçalo10 P. Rolom11 F. R. Negro12 P. Ciríaco13 T. Fogo14 Descalvados15 P. Conceição16 Amolar17 S. Francisco18 P. Manga19 P. Murtinho

Figure 6. Location of Upper Paraguay River Basin and indication of modelled channel network and floodplain, and of streamflow gauging stationsused as control points or boundary conditions

Table I. List of boundary conditions with drainage area and observed daily discharge data availability during the simulation period(1 September 1995–31 December 2006)

Streamflow gauging station defining the boundarycondition (reference to Figure 6)

River Drainage area(km2)

Observed discharge data availability(% of simulation period)a

a Cuiabá Cuiabá 24 668 100b A. C. Grande S. Lourenço 23 327 94c S. Jerô nimo Piquiri 9215 99Ð7d P. Espiridião Jauru 6221 96Ð5e Cáceres Paraguay 32 574 96Ð4f Coxim Taquari 28 688 99Ð5g P. Bocaı́na Negro 2807 0h Aquidauana Aquidauana 15 350 97Ð1i Miranda Miranda 15 502 99Ð7j ž Upstream of Apa Riverb Paraguay 594 092 bAQ1

a Data available from the Brazilian Water Agency (ANA).b Downstream boundary condition defined by the Paraguay River section upstream of the affluence of Apa river, considering the energy slope parallelto average bed slope.

123456789

adjusted for all the sub-basins of the Planalto region ofthe UPRB in the study reported by Tucci et al. (2005). Avery reasonable fit of the MGB-IPH model was achievedby these authors, with Nash–Suttcliffe (NS) coefficientsranging from 0Ð56 to 0Ð88.

The Paraguay River section upstream of the affluenceof Apa River, about 60 km downstream from Porto Murt-inho, was taken as the downstream boundary conditionof the modelled network, considering the energy slope

101112131415161718

parallel to average bed slope. The time step of chan-nel flow modelling (tchan) was adopted as 1 h, andthe initial conditions were determined considering steadybackwater flow approximation.

2D raster-based model application

The floodplain modelled area was defined accordingto earlier studies that delimited the Pantanal and theSRTM-90m DEM, but also taking into account that a no


UNCO

RREC

TED

PROO

FS


123456789

1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859

flow boundary condition is imposed to the raster model.For this reason, the modelled area was traced overesti-mating the area subject to inundation, which is roughlyabout 140 000 km2. The raster model domain comprises219 514 km2 (Figure 6), discretized into 46 741 elementson a 0Ð02° ð 0Ð02° grid. In planar units, each element isapproximately 2 km wide, with surface area ranging from4Ð58 to 4Ð78 km2 depending on its latitude.

Floodplain topography was represented by the SRTM-90m DEM resampled to the raster-based model dis-cretization, using the nearest neighbour interpolationmethod. Following the data preparation procedures, atotal of 1081 floodplain elements were identified asdirectly connected to the main channels.

The inundation model was run with a 120-s time step,which was selected after testing different values and ver-ifying that this value avoided numerical instabilities. A1-day time step was selected for the vertical water bal-ance, due to precipitation and potential evapotranspirationdata availability on a daily basis and also because this isadequate to represent the modelled processes in this studyarea. Observed precipitation data available from 105 rain-fall gauging stations were interpolated to the 0Ð02° gridresolution of the floodplain model using the inverse dis-tance squared method. Although this rain gauge networkis sparse, for instance it is sufficient to provide precip-itation estimates for testing the proposed model. Futurework will try to investigate model sensitivity to precipi-tation estimates and also the combination of pluviometermeasures with satellite-based estimates, such as thosegenerated by the Tropical Rainfall Measuring Mission(TRMM; Kummerow et al., 2000).

The estimates of potential evapotranspiration producedby the MGB-IPH distributed hydrological model appliedto the entire UPRB in a earlier study (Tucci et al., 2005)were used as input data. The MGB-IPH model calculatespotential evapotranspiration through Penman–Monteithmethod as presented by Shuttleworth (1993) and follow-ing the approach proposed by Wigmosta et al. (1994).Distinct combinations of land cover and soil type arerepresented inside each model cell through patches withspecific parameter values. This model was applied tothe UPRB considering a 0Ð1° ð 0Ð1° regular grid and a1-day time step. The simulation period was from 1968 to2006, and the estimates of potential evapotranspirationused as input data for the floodplain model correspond tothe patch representing surface water, which were interpo-lated to the 0Ð02° floodplain model grid using the inversedistance squared method.

Calibration procedure and model skill assessment

To evaluate the performance of the 1D hydrodynamicmodel, 15 streamflow gauging stations with available datawere used as control points for comparing calculatedand observed discharges along the main channel network(Figure 6). Floodplain inundation dynamics simulated bythe raster model was compared with estimates of totalinundated area provided by Hamilton et al. (1996) and

60616263646566676869707172737475767778798081828384858687888990919293949596979899

100101102103104105106107108109110111112113114115116117118

with estimates of inundation extent produced by Padovani(2007).

Hamilton et al. (1996) estimated the total of floodedareas of Pantanal in the period 1979–1987 throughanalysis of data obtained by the scanning multichannelmicrowave radiometer (SMMR) sensor of the Nimbus-7satellite. Despite the related uncertainties mostly due tocoarse resolution of satellite images (27 km), vegetationcover heterogeneity, and of being relative to a time perioddistinct from the one simulated in this article, the studyof Hamilton et al. (1996) presented to date the mostcomplete time series of seasonal floods in the Pantanal.

Padovani (2007) classified images of the sensor wide-field imager (WFI) on board of the CBERS-2 satellite(China–Brazil Earth Resources Satellite) to distinguishbetween flooded and non-flooded areas of Pantanal forthe dates 6 October 2004 (dry period) and 13 February2005 (wet period). These images have a spatial resolutionof 260 m and, as the WFI has a ground swath of890 km, a unique scene covering the entire area ofinterest for each date was used (path 165, row 116).These images were classified by an unsupervised method,the Iterative Self-Ordering Data Analysis (ISODATA)algorithm, as implemented in the ERDAS Imagine 8Ð5software. The resulting classes were grouped into floodedor non-flooded areas, taking the RGB color compositeof Landsat 7 ETMC images for the year 2000 anddigital aerial photographs of the region as ancillary data.Undoubtedly these estimates have uncertainties, mostlyassociated to inundated areas covered by vegetationand areas with wet saturated soil, which may lead tounder- and overestimation of flooded extent, respectively.However, this is the only readily available inundationextent mapping of the entire Pantanal area for comparisonwith our results.

A simplified approach was adopted for adjusting modelparameters, as the calibration process of coupled 1D/2Dmodels is not straightforward. For instance, some stud-ies indicate that it is not possible to find a unique setof parameters of the raster model that provide acceptableadjustments for both channel flow and floodplain inun-dated area (Horritt and Bates, 2001b). žAnother ques- AQ3tion concerns whether using constant or spatially varyingparameters on 2D floodplain models (Werner et al., 2005;Hunter et al., 2007). Albeit several efforts have been con-ducted to estimate friction parameters based on remotesensing data (Bates et al., 2004), in the case of sim-plified models, such as the proposed in this article, theparameters are related to aggregated hydraulic processdescriptions (Hunter et al., 2007), weakening the rela-tion of them with floodplain physical characteristics. Inlight of this discussion and due to the large extent of thestudy case and scarce available data sets, in this studythe calibration process focused primarily on reproducingmain channel flow, but also trying to reproduce generalaspects of floodplain dynamics. Further study may focusparticularly on adjusting model parameters for reproduc-ing inundation patterns.


UNCO

RREC

TED

PROO

FS

10 A. R. PAZ ET AL.

123456789

1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859

Initially, a constant Manning coefficient was adoptedfor all main channel reaches in the 1D hydrodynamicmodel, and several runs of the hydrologic simulationsystem were performed with varying floodplain modelparameter values. The Manning channel roughness wasselected as 0Ð035 following a recommendation for largenatural rivers (Chow, 1959, 1964). The parameters fhcand Hsmax were varied in each simulation run, butassuming constant values along the floodplain.

This rough sensitivity analysis of floodplain param-eters lead to the selection of the values fhc D 50 andHsmax D 1Ð0 m, based on channel hydrograph compar-isons and the modelled general inundation patterns, bothin terms of total inundated area and inundation extent.Adopting these values for the floodplain parameters, anew set of simulation runs was carried out for adjust-ing main channel roughness. This was done in a trialand error process, by manually varying the Manningcoefficient values and comparing calculated and recordedhydrographs through visual inspection and using as statis-tical measures the NS model efficiency coefficient (NS),the NS coefficient for logarithms of discharge values(NSlog), the relative streamflow volume error (V) andthe root mean square error (RMSE). The calibration pro-cedure was realized first for the tributaries and then forthe Paraguay River, from upstream to downstream alongeach river.

Finally, assessment of floodplain inundation dynam-ics, through comparison with results of Hamilton et al.(1996) and Padovani (2007), was carried out consideringthe simulation run using the adjusted main channel Man-ning coefficients and the selected values for floodplainparameters. It is worth noting that those authors con-sidered distinct delimitations for defining the Pantanalarea in their studies, albeit in general these delimita-tions are very similar between them. The Pantanal’s areafollowing the outline of Hamilton et al. (1996) is about138 139 km2, while the one used in the study of Padovani(2007) has 138 437 km2. The major difference betweenthem regards to the west portion, where the delimitationused by Padovani (2007) follows the Brazilian countryborder, as this sketch defines the Pantanal region offi-cially adopted by Brazilian Government.

Simulated total inundated area was converted into aver-age seasonal values for comparison with the results ofHamilton et al. (1996), considering the Pantanal delimi-tation adopted by those authors and adopting the depththreshold of 2 cm to distinguish between dry and inun-dated condition of each element of the raster-basedmodel.

The comparison between simulated and Padovani’sestimates of inundation extent was carried out through apixel-to-pixel basis, and considering the Pantanal delim-itation used by that author. We aggregated the 260 minundation maps of Padovani (2007) to the spatial reso-lution of the raster-based model (2 km). Each pixel of thePantanal area was compared whether wet or dry on bothsimulated and estimated inundation maps. A 2 ð 2 con-tingency table was built as shown in Figure 7, where ‘a’

a

Satellite-basedestimate

Mod

elsi

mul

atio

n

b

c d

Wet

Dry

Dry

Wet a + b + c + d

a + dPC =

a + b + ca

CSI =

a + ca

POD =

a + bb

FAR =

Figure 7. Contingency table (2 ð 2) for comparison between inundationmaps resultant from satellite-based estimates and floodplain modelsimulation, where ‘a’ and ‘b’ are the number of pixels which were wet onboth maps, ‘c’ is the number of pixels which were wet on estimated mapbut dry on simulated map and ‘d’ is the number of pixels which weredry on estimated map but wet on simulated map; and four derived skillscores: proportion correct (PC, critical success index (CSI, probability of

detection (POD and false alarm ratio (FAR)

60616263646566676869707172737475767778798081828384858687888990919293949596979899

100101102103104105106107108109

and ‘d’ correspond to the number of wet and dry pixels,respectively, simultaneously on both simulated and esti-mated maps. The number of pixels which were estimatedas wet but simulated as dry are summed in ‘c’, while‘d’ is the number of pixels that were wrongly simulatedas wet (they were estimated as dry). Four skill scoreswere then derived: proportion correct (PC), critical suc-cess index (CSI), probability of detection (POD) and falsealarm ratio (FAR) (Figure 7). Each of these measures offit suggests distinct analysis of the results (Wilks, 2006).

The index PC is simply the fraction of the total amountof pixels in agreement between model simulation andPadovani’s estimate, indistinctly whether wet or dry. Itranges from 0 (no agreement) to 1 (perfect agreement),and means the area correctly predicted by the model.For instance, the PC was used as a measure of fit ofinundation models by Bates and De Roo (2000) andPearson et al. (2001).

The CSI is similar to PC, but accounting for onlythe agreement of wet pixels and disregarding the correctsimulation of dry pixels, under the assumption that it isrelatively easier to correctly predict non-flooded areas.The CSI may also be interpreted as the ratio betweenthe intersection of simulated and estimated flooded areasand the combination of them. It ranges from 0, when nooverlap occurs between flooded areas of simulated andestimated inundation maps, to 1, when there is exactly acoincidence. This is by far the most widely used measureof fit for evaluating simulated inundation extent againstestimates from others sources (Bates and De Roo, 2000;Horritt and Bates, 2001a; Bates et al., 2005; Tayefi et al.,2007; Wilson et al., 2007).

The POD skill score, also known as hit rate, meansthe fraction of the pixels estimated as wet which werecorrectly simulated as so, ranging from 0 to 1 (the higherthe value the better the performance). The FAR meansthe fraction of the pixels estimated as dry which werewrongly simulated as wet, also ranging from 0 to 1, butthe smaller the value the better the performance. Theseindices are mostly used for comparing spatial fields ofprecipitation and other meteorological variables (Wilks,2006), but also provide interesting analysis for floodplaininundation assessment.


UNCO

RREC

TED

PROO

FS


123456789

10111213141516171819202122232425262728

RESULTS AND DISCUSSION

Computation time and performance

To evaluate the gain of introducing the parallelizationscheme via OpenMP for part of the floodplain model, theSIRIPLAN was run for the UPRB in a sequential modeand further considering two and four processor cores inthe parallelization. The three runs were performed in aquad core Intel processor 3 GHz with 4 GB RAM.

The computation time required in each run is shownin Table II. When running sequentially, the run time wasgreater than 4 h. This run time was reduced by 45% whenadopting a two cores parallelization and by 67% whenparallelizing with four cores. Parallel speedup (run timeof parallel execution divided by run time of sequentialexecution) equal to 1Ð82 and 3Ð07 was obtained for twoand four cores parallelization, respectively. In terms ofparallel efficiency (speedup divided by the number ofprocessor cores), running in parallel with two and fourcores resulted in values of 0Ð91 and 0Ð77, respectively.

The values of parallel speedup and efficiency obtainedwith SIRIPLAN in this study were similar to the bestresults presented by Neal et al. (2009), who ran theLISFLOOD-FP model applied to several different studycases considering the OpenMP parallelization technique.

Flow regime along main channels

A very reasonable model fit was obtained in terms ofreproducing main channel flow along the Paraguay River

29303132333435363738394041424344454647484950515253545556

and its tributaries, as indicated by the performance mea-sures comparing observed and calculated hydrographsshown in Table III, relative to the period from 1 Decem-ber 1997 to 31 December 2006 (the antecedent periodwas disregarded due to initial conditions influence).

For the gauging stations located at the tributaries, theadjusted Manning coefficients ranged from 0Ð02 to 0Ð055,and were obtained NS and NSlog coefficients rangingfrom 0Ð75 to 0Ð94 and from 0Ð80 to 0Ð97, respectively.The volume error for these stations was less than 10%in absolute value, except for the Ilha Camargo station(V D �13Ð5%), while the RMSE ranged from lessthan 20 m3/s at P. Cirı́aco (Aquidauana River) to near100 m3/s at P. Taiamã (Cuiabá River).

The model was capable to reproduce the general shapeof observed hydrographs at the tributaries, as illustratedby visually comparing observed and calculated hydro-graphs at P. Cercado, P. Taiamã and P. Cirı́aco gaugingstations (Figure 8a–c, respectively). For instance, thesethree cases exemplify the complexity of flow regime ofrivers flowing along Pantanal. There is a small over-estimation trend on calculated seasonal peak flows atP. Cercado station, of about 10% for the wettest years,while at P. Taiamã and P. Cirı́aco there is an underesti-mation trend of up to 15% and 5% on calculated seasonalpeak flows, respectively. For these three gauging stations,there are insignificant differences between observed andcalculated recession flows.

Table II. Run time and performance of the SIRIPLAN hydrologic system applied to the Upper Paraguay River Basin

Run type Run time Performance relative to single core

Run-time reduction Speedup Efficiency

Sequentially 4 h 23 min 47 s — — —Parallel two cores 2 h 25 min 10 s 45% 1Ð82 0Ð91Parallel four cores 1 h 26 min 25 s 67% 3Ð07 0Ð77

Table III. Performance measures of SIRIPLAN hydrologic system in simulating main channel flow along Paraguay River and itstributaries

Reference to Figure 6 Station names River Drainage area (km2) Statisticsa

RMSE (m3/s) NS NSlog V (%)

1 B. Melgaço Cuiabá 27 050 70Ð2 0Ð94 0Ð97 �5Ð82 P. Cercado Cuiabá 38 720 46Ð1 0Ð91 0Ð92 �4Ð63 S. João Cuiabá 39 908 50Ð2 0Ð82 0Ð84 �8Ð84 I. Camargo Cuiabá 40 426 85Ð3 0Ð78 0Ð80 �13Ð55 S. J. Borireu S. Lourenço 24 989 26Ð6 0Ð92 0Ð94 4Ð96 S. J. Piquiri Piquiri 28 871 89Ð2 0Ð75 0Ð82 8Ð97 P. Taiamã Cuiabá 96 492 98Ð5 0Ð90 0Ð92 �2Ð18 P. Alegre Cuiabá 104 408 79Ð8 0Ð82 0Ð85 8Ð39 P. Cirı́aco Aquidauana 19 204 18Ð0 0Ð76 0Ð83 �3Ð510 Descalvados Paraguay 48 360 79Ð3 0Ð91 0Ð92 �5Ð011 P. Conceição Paraguay 65 221 80Ð1 0Ð63 0Ð62 7Ð612 Amolar Paraguay 246 720 180Ð7 0Ð67 0Ð72 6Ð313 P. S. Francisco Paraguay 251 311 258Ð7 0Ð70 0Ð73 �2Ð014 P. Manga Paraguay 331 114 191Ð3 0Ð82 0Ð76 2Ð515 P. Murtinho Paraguay 581 667 343Ð5 0Ð61 0Ð65 �6Ð1a To exclude the effect of initial conditions, statistics were calculated for the period from 1 December 1997 to 31 December 2006.


UNCO

RREC

TED

PROO

FS

12 A. R. PAZ ET AL.

-800

-600

-400

-200

0

200

400

600

800

98 99 00 01 02 03 04 05 06 07

Qobs Qcalc Qlat(a)

(b)

(c) (f)

P. Cercado (Cuiabá River)

P. Ciríaco (Aquidauana River)

P. Conceição (Paraguay River)

P. Manga (Paraguay River)

Dai

ly fl

ow (

m3 /

s)D

aily

flow

(m

3 /s)

P. Taiamã (Cuiabá River)

(d)

(e)Amolar (Paraguay River)

Dai

ly fl

ow (

m3 /

s)

-200

0

200

400

600

800

1000

1200

1400

1600

98 99 00 01 02 03 04 05 06 07

Qobs Qcalc Qlat

-800

-600

-400

-200

0

200

400

600

800

98 99 00 01 02 03 04 05 06 07

Qobs Qcalc Qlat

0

500

1000

1500

2000

2500

98 99 00 01 02 03 04 05 06 07

Qobs Qcalc Qlat

0

500

1000

1500

2000

2500

3000

3500

98 99 00 01 02 03 04 05 06 07

Qobs Qcalc Qlat

-600

-500

-400

-300

-200

-100

0

100

200

98 99 00 01 02 03 04 05 06 07

Qobs Qcalc Qlat

Figure 8. Comparison of calculated (Qcalc) and observed (Qobs) hydrographs at three gauging stations located at tributaries and three stations ofParaguay river; Qlat is the lateral flow exchanged between main channel and floodplain along the following river reaches: (a) from B. Melgaçoto P. Cercado; (b) from the confluence of Piquiri and Cuiabá Rivers to P. Taiamã; (c) from Aquidauana to P. Cirı́aco; (d) from Descalvados to P.Conceição; (e) from the confluence of Cuiabá and Paraguay Rivers to Amolar and (f) from P. S. Francisco to P. Manga; Qlat 0 means flow in the opposite direction

123456789

1011

In the graphs of Figure 8, Qlat means the calculatedlateral flow exchanged between main channel and flood-plain along the upstream river reach specified on the cap-tion of the figure for each case, being negative if flowingfrom the channel to floodplain and positive if flowing inthe opposite direction. Along the 107 km reach of CuiabáRiver upstream of P. Cercado, was simulated a huge lossof water from channel to floodplain during rising limb offlood hydrograph, with Qlat achieving up to �600 m3/s(around 8% greater than flood peak along main chan-nel), and a gain of water after flood peak flow of up

1213141516171819202122

to 180 m3/s. Meanwhile, no water exchanges betweenchannel and floodplain were simulated for the river reachupstream of P. Taiamã station.

At P. Cirı́aco station, located on the AquidauanaRiver 230 km downstream from Aquidauana station(boundary condition), the observed hydrograph presentsa marked maximum value of 150 m3/s. At Aquidauanastation, observed peak flow reaches up to 700 m3/s. Thisenormous reduction of channel flow in this river reachwas well represented by the model, which simulatedlateral exchanges of water from channel to floodplain of


UNCO

RREC

TED

PROO

FS


0

200

400

600

800

1000

1200

98 99 00 01 02 03 04

Qobs Qcalc Qlat

0

200

400

600

800

98 99 00 01 02 03 04

Qobs

-800

-600

-400

-200

0

200

400

600

98 99 00 01 02 03 04

Qobs Qcalc Qlat

0

200

400

600

800

1000

1200

98 99 00 01 02 03 04

Qobs

0 100 km50

A. C. Grande (boundary condition) S. J. Borireu

S. Jerônimo (boundary condition) S. J. Piquiri

Dai

ly fl

ow (

m3 /

s)D

aily

flow

(m

3 /s)

(a) (b)

(c) (d)

Figure 9. (a) and (c) Observed hydrographs at the boundary conditions of S. Lourenço (A. C. Grande station) and Piquiri (S. Jerônimo) rivers and(b) and (d) comparison between calculated (Qcalc) and observed (Qobs) hydrographs at their downstream gauging stations, also showing lateral flow

exchanged between main channel and floodplain along each river reach between the boundary condition and the downstream station

123456789

101112131415161718192021222324252627

up to 500 m3/s during flood peaks. The maximum lateraldischarge simulated corresponds to 3Ð3 times peak flowalong main channel at P. Cirı́aco. During the dry period,no water drainage from the floodplain was simulated andthe observed recession flow at this station was also wellreproduced.

As at P. Cirı́aco, a marked maximum flow (of about400 m3/s) on observed hydrograph is also seen at S. J.Borireu station, located on the S. Lourenço River, whichwas well reproduced by the model (difference less than5%) (Figure 9a and b). Along the 250 km long reachbetween this station and the upstream boundary condition(A. C. Grande station), the model simulated lateral flowsof up to 750 m3/s from main channel to floodplain.

In the reach of the Piquiri River upstream of S. J.Piquiri station (80 km downstream from S. Jerônimo,taken as boundary condition), the exchanges of waterbetween floodplain and main channel was simulated asoccurring in the opposite direction of that reported tothe S. Lourenço River (Figure 9c and d). A gain of waterfrom the floodplains to the main channel was simulated inthis reach of Piquiri River, totalling up to 400 m3/s duringthe floods. This gain of water represents almost 50% ofthe water flowing along the main channel at S. J. Piquiristation. In fact, while at S. Jerônimo observed peak flowranges between 400 and 700 m3/s, at S. J. Piquiri thisrange is between 400 and 1100 m3/s. The increase in

282930313233343536373839404142434445464748495051525354

observed peak flow from upstream to downstream isdue to lateral floodplain contribution, which the modelwas capable to simulate. The estimated hydrograph ofthis lateral gain of water to main channel presents asmall time delay relative to channel flood peak. Duringdry periods, this hydrograph reached null values, whichallowed recession flow at S. J. Piquiri to be quite wellreproduced. Most interestingly is that the major part ofthe contribution of floodplain to main channel of PiquiriRiver at this location during floods was resultant fromthe volume of water lost by the main channel of theS. Lourenço River, 35 km to North, which flowed alongfloodplains.

Owing to large drainage areas and complexity ofprocesses involved, including contributions of tributariesthat may occur both through main channel and floodplainflows, reproduction of flow regimes along the ParaguayRiver is even more difficult than along its tributaries.However, the model was able to reproduce the seasonalflow regime along the Paraguay River, as illustratedby the performance measures comparing observed andcalculated flows at six gauging stations (Table III). TheNS and NSlog coefficients ranged from 0Ð61 to 0Ð91 andfrom 0Ð62 to 0Ð92, respectively. RMSE were obtainedbetween 80 and 343 m3/s, which seem to be largeerrors in absolute terms, but correspond roughly to lessthan 13% of average peak flow in each station: 7%


UNCO

RREC

TED

PROO

FS

14 A. R. PAZ ET AL.

98 99 00 01 02 03 04 05 06 07

98 99 00 01 02 03 04 05 06 07

98 99 00 01 02 03 04 05 06 07

Cáceres - Descalvados

P. S. Francisco - P. Manga

Descalvados - P. Conceição

P. Manga - P. Murtinho

Dai

ly fl

ow (

m3 /

s)D

aily

flow

(m

3 /s)

Dai

ly fl

ow (

m3 /

s)

P. Conceição - Amolar Amolar - P. S. Francisco-800

-600

-400

-200

0

200

400

600

800

98 99 00 01 02 03 04 05 06 07

-800

-600

-400

-200

0

200

400

600

800

98 99 00 01 02 03 04 05 06 07

-800

-600

-400

-200

0

200

400

600

800

98 99 00 01 02 03 04 05 06 07

Qlat < 0: flow from channel to floodplain

Qlat > 0: flow from floodplain to channel

Figure 10. Lateral exchanges of water between main channel and floodplain simulated by SIRIPLAN along the modelled reach of Paraguay River,separated into six river reaches between each, two consecutive gauging stations: Cáceres, Descalvados, P. Conceição, Amolar, P. S. Francisco,

P. Manga and P. Murtinho

123456789

10111213141516

at Descalvados, 12% at P. Conceição, 11% at Amolar,13% at P. S. Francisco, 9% at P. Manga and 13%at P. Murtinho. In terms of volume error, the resultsobtained ranged from �6Ð1% at P. Murtinho to 7Ð6% atP. Conceição station. Manning coefficients ranged from0Ð012 to 0Ð055.

Hydrographs along Paraguay River have marked sea-sonality, as can be seen on Figure 8d (P. Conceiçãostation), Figure 8e (Amolar) and Figure 8f (P. Manga),which were quite well reproduced by the developedmodel, despite some discrepancies between observed andestimated hydrographs, as the overestimation of recessionflows and underestimation of peak flows in some years.

It is important to highlight the model ability fordifferentiating the intensity of the seasonal flood amongyears. For instance, at P. Manga station, which has a

17181920212223242526272829303132

drainage area greater than 330 000 km2, the SIRIPLANwas able to estimate the reduced peak flows (less than1800 m3/s) of the floods of the years 2001 and 2005 andthe large flood of 2002 (peak flow around 2700 m3/s).

The simulated lateral flow in the Paraguay River reachfrom P. S. Francisco to P. Manga (almost 200 km length)was negligible, while a loss of water from main channelto floodplain achieving peak flows up to 600 m3/s wasestimated for the reach between Descalvados and P.Conceição (¾120 km). Along the 21-km long reachdownstream of the confluence of Cuiabá River up toAmolar station, a gain of water from floodplain to mainchannel was simulated. This gain occurred throughoutthe entire year, with peak flows up to 330 m3/s in theperiod June–July and flows up to 30 m3/s in the othermonths.


UNCO

RREC

TED

PROO

FS


123456789

1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859

To better analyse the channel–floodplain waterexchanges along the modelled reach of the ParaguayRiver, the estimates of lateral flows for each reach delim-ited by two consecutive streamflow gauging stations isshown in Figure 10. This figure shows distinct patternsof lateral water exchanges along the upper, middle andlower reaches of the Paraguay River. A loss of waterfrom channel to floodplain prevails in the most upperpart of the Paraguay River, from Cáceres (boundary con-dition) to Descalvados station. Simulated lateral flowsfrom channel to floodplain achieved peaks of up to650 m3/s in the reach between Cáceres and Descalvados,and up to 590 m3/s in the reach between Descalvados andP. Conceição. In the reach Cáceres–Descalvados, resultsshow that water flows from channel to floodplain mostlyduring the period December–April and in the oppositedirection during the period May–July, with null flowsfrom August to November. In the downstream reach(Descalvados–P.Conceição), null lateral flows were sim-ulated from July to November, with a loss of water fromchannel to floodplain over the rest of the year.

In the middle part of the Paraguay River, downstreamof P. Conceição station and upstream of P. S. Fran-cisco, the simulated lateral exchanges of water werepredominantly a gain from floodplains to main chan-nel. Indeed, the model simulated that this reach of theParaguay River receives contribution propagated from itsupstream floodplains and also drained by the floodplainsof Cuiabá River. The simulated lateral peak flows wereup to 800 m3/s in the reach between P. Conceição andAmolar, and up to 620 m3/s in the reach between Amolarand P. S. Francisco. In the former reach, lateral water lossfrom channel to floodplain was simulated in the periodDecember–March, with flows in the opposite directionduring the following months. In the latter reach, a gainof water from floodplain to channel was simulated asoccurring over the entire year.

For the lower part of the Paraguay River, from P. S.Francisco to P. Murtinho station, simulated lateral flowswere relatively small, in comparison to the flows ofthe upstream reaches. Along the reach between P. S.Francisco and P. Manga, these flows were approximatelynull, while a gain of water less than 200 m3/s wassimulated along the reach between P. Manga and P.Murtinho stations.

Floodplain inundation

Typical inundation maps of a dry and wet periodare shown in Figure 11, relative to the dates 6 October2004 and 13 February 2005, respectively. The estimatesof inundation extent produced by Padovani (2007) forthese same dates are also shown in this figure. Thecorrespondent measures of fit between simulated (ourresults) and estimated (Padovani’s results) inundationmaps are given in Table IV.

The model was capable to reproduce part of the majorpermanent inundated areas during the dry period, whichare exclusively due to water spilling from main chan-nels and flowing along floodplain. These areas are located

Simulated

0 100 20050 km

6 October 2004

13 February 2005

Simulated

Estimated by Padovani (2007)

Estimated by Padovani (2007)

Figure 11. Inundation maps of Pantanal simulated and estimated byPadovani (2007), for two dates: 6 October 2004 (dry period) and 13

February 2005 (wet period)

Table IV. Skill scores of the comparison between inundationmaps estimated by Padovani (2007) and simulated with SIRI-

PLAN, at two dates

Accuracymeasure

Dry period(6 October 2004)

Wet period(13 February 2005)

PC 0Ð60 0Ð57CSI 0Ð24 0Ð51POD 0Ð37 0Ð59FAR 0Ð60 0Ð23

606162636465666768697071727374

along the north and central portions of Paraguay River,in the reach between Descalvados and P. Manga gaug-ing stations, along the floodplains of the lower reachof Cuiabá River and along both margins of the TaquariRiver. Also, the inundation along Taquari floodplains isconsistent with the expected pattern, as this region com-prises the distributary fan lobe of the Taquari alluvialmegafan (Assine, 2005). However, considering the esti-mates of Padovani (2007) as correct, these major flooded


UNCO

RREC

TED

PROO

FS

16 A. R. PAZ ET AL.

123456789

10111213141516171819202122232425262728293031323334353637383940414243444546

areas were underestimated by the model, as is clear byvisual comparison of both maps. This underestimationresulted in the low CSI and POD skill scores. About 60%(PC D 0Ð60) of the pixels were in agreement betweenthese two inundation maps, i.e. 60% of the area was wetor dry simultaneously on both maps. However, disregard-ing the coincident dry pixels on both maps, the agreementbetween them reaches 24% (CSI D 0Ð24). From the areaestimated as flooded in Padovani’s work, 37% was alsoflooded in the simulated map (POD D 0Ð37). On thecontrary, the obtained FAR score means that, from thearea simulated as flooded, 60% was estimated as dryby Padovani (2007), and this relatively high value ismostly due to dispersed isolated pixels wrongly simu-lated as flooded by the model. In terms of total area,the model simulated 40 491 km2 as flooded areas, whichcorresponds to 29Ð2% of the Pantanal, while the esti-mates of Padovani (2007) indicate an inundation extentof 45 135 km2 (32Ð6% of total) (Table V).

During floods, the loss of water from main channels tofloodplains is increased and the most important floodedareas identified in the dry period become larger anddeeper. However, the major difference between inunda-tion maps of dry and wet periods is that in the wet periodthe flooded areas cover a much larger extension along thewhole domain. Although with prevailing shallow waterdepths, the simulated flooded area on 13 February 2005covers almost twice the extent estimated at 6 October2004, i.e. a flooded area of about 76 406 km2 or 55Ð2%of the entire Pantanal. The estimates of Padovani (2007)show an even larger flooded area, of about 100 393 km2

(72Ð5% of total), and indicate again an underestimationtrend on model results, but weaker than that for the dryperiod. In terms of skill scores, the general agreementbetween simulated and estimated inundation maps wasincreased in comparison to the dry period. Although thePC index was almost equal between the two periods, theCSI and POD indices were quite improved at this time,with CSI D 0Ð51 and POD D 0Ð59. Also, the FAR hasdecreased (FAR D 0Ð23), meaning that only 23% of thearea simulated as flooded was dry in the inundation mapof Padovani (2007).

In comparison to others studies of floodplain inunda-tion modelling, our CSI scores are relatively similar withthem. For instance, the greater difficulty to reproduce theinundation extent during the dry period is also pointed

47484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192

out by Wilson et al. (2007), which was the unique previ-ous study žwe found that assessed inundation map during AQ4dry period. Those authors used the LISFLOOD-FP modelto simulate part of the Amazon River and Purus trib-utary, obtaining CSI D 0Ð23, approximately the samescore we achieved. They state that their model inabil-ity to simulate low water inundation extent is mostlydue to not including floodplain vertical hydrological pro-cesses and the SRTM DEM aggregation, which makesdifficult the representation of complex, small-scale topog-raphy controlling part of the floodplain drying out pro-cess. Although we have included representation of evap-otranspiration and infiltration processes, the simplicity ofadopted schemes together with the aggregation of SRTMDEM to the 2 km resolution may have reduced modelcapability on reproducing the full drainage of the flood-plain. The sparse pluviometer network and uncertaintieson precipitation estimates may also have contributed tothis model inability. For the wet period, our CSI scoreof 0Ð51 is similar to the lower limit of the range ofresults obtained by others authors varying model param-eters or structure, such as Wilson et al. (2007), Tayefiet al. (2007), Horritt and Bates (2001b) and Bates andDe Roo (2000).

As stated before, during the dry period, the inundationextent was almost limited to the major permanent floodedareas resultant from water spilling from main channel tofloodplains. During the wet period, regions not directlyconnected to overbank flow from main channels wereflooded due to delayed drainage of precipitation. Thisinput of water to the floodplain gives origin to localwater accumulation which drains slowly, or is evaporatedin the following dry period, resulting in a markedseasonal variation in total inundated area as illustratedin Figure 12. Peaks of total inundated areas simulatedby the model ranged from 100 000 to 126 000 km2 alongthe simulation period, which are similar to the maximumvalues of inundation estimated by Hamilton et al. (1996)for a different period (1979–1987). The total inundatedareas during dry periods simulated with SIRIPLANranged from 35 000 to 45 000 km2, while the mentionedstudy estimated much smaller minimum inundated areas,of up to 11 000 km2. This result could indicate anoverestimation of our inundated area during dry period.However, given that the estimate of inundation extentof Padovani (2007) for the date 6 October 2004 (dry

Table V. Flooded and dry total areas over Pantanal on two dates simulated by SIRIPLAN and estimated by Padovani (2007)

Floodplain Dry period (6 October 2004) Wet period (13 February 2005)

Simulated Estimated byPadovani (2007)

Simulated Estimated byPadovani (2007)

Area(km2)

Percentage oftotal area

Area(km2)


Area(km2)


Area(km2)


Flooded 40 491 29Ð2 45 135 32Ð6 76 406 55Ð2 100 393 72Ð5Dry 97 946 70Ð8 93 302 67Ð4 62 032 44Ð8 38 044 27Ð5Total 138 437 100Ð0 138 437 100Ð0 138 437 100Ð0 138 437 100Ð0


UNCO

RREC

TED

PROO

FS


(a)D

aily

inun

date

d ar

ea (

× 10

00 k

m2 )

0

20

40

60

80

100

120

140

98 99 00 01 02 03 04 05 06 07

0

20

40

60

80

100

N J M M J A

Hamilton et al. (1996) Simulated

(b)

Ave

rage

mon

thly

inun

date

dar

ea (

× 10

00 k

m2 )

D F A J OS

Figure 12. (a) Daily inundated areas simulated over Pantanal [the horizontal grey lines represent the maximum and minimum values estimated byHamilton et al. (1996) for the period 1979–1987] and (b) average monthly inundated areas simulated along the period from 1 January 1998 to 31

December 2006 and estimated by Hamilton et al. (1996) for the period 1979–1987

Flood occurrence during>5% of simulation period

0 100 20050 km



Figure 13. Maps showing areas subject to inundation during frequencies greater than 5%, 25% or 75% of simulation period

123456789

10111213141516171819202122

period) corresponds to an area of about 45 000 km2

and seems consistent to expected inundation patterns ofPantanal, may be the results of Hamilton et al. (1996)are underestimated or their period of analysis was muchmore drier than our area.

Comparison of average monthly estimates shows thatin our study the peak of flooding occurred between 1 and2 months in advance relative to the results of Hamiltonet al. (1996) (Figure 9b). Again, it can be noted thedifference on inundated areas in the dry period betweenthe two studies. Nevertheless, it is worth noting theimportance of including the vertical water balance onfloodplain modelling and the capability of SIRIPLAN tosimulate the Pantanal seasonal flood pulse.

The model capability to simulate the major permanentflooded areas are also highlighted by maps shown inFigure 13, which provides an analysis of simulatedinundation frequency spatially distributed over Pantanal.The maps in this figure show the areas that wereinundated during time periods greater than 5%, 25% and75% of the simulation period (considering the 9 years

23242526272829303132333435363738394041424344

from 1 January 1998 to 31 December 2006). Theseinundation frequencies were calculated regardless ofbeing during consecutive days or not. Approximately32% (43 624 km2) of the Pantanal was flooded duringmore than 75% of the simulation period, while 58%(80 330 km2) of Pantanal was flooded during more than25% of the simulation period. This area increases to115 033 km2 (83% of total) when the 5% frequencythreshold is considered, and it goes to the limit of100% of Pantanal area as the threshold approaches zero,i.e. the entire Pantanal was flooded in at least 1 dayof the simulation period. On the contrary, when thefrequency threshold approaches 100%, i.e. consideringsolely pixels which were strictly permanently inundated,the area covers roughly 22% of entire Pantanal (¾30 000km2).

SUMMARY AND CONCLUSIONS

This paper presents the hydrologic simulation systemSIRIPLAN, developed for simulating the flow regime


UNCO

RREC

TED

PROO

FS

18 A. R. PAZ ET AL.

123456789

1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859

and spatial inundation over large-scale networks of riversand floodplains. The SIRIPLAN couples the 1D hydro-dynamic model IPH4 for simulating main channel flowto a 2D raster-based floodplain model, which simulatesthe floodplain inundation dynamics. Auxiliary modulessimulate the vertical water processes of precipitation,infiltration and evapotranspiration over floodplains andwater exchanges between channels and floodplains.

The application example of the SIRIPLAN to theUPRB, which includes the Pantanal, one of the largestwetlands of the world, showed the viability and adequacyof the proposed approach. A total of 3965 km of mainchannels and 140 000 km2 of floodplains were simulatedfor a time period of 11 years. The computational routinesdeveloped for establishing the topological connectionsbetween channel and floodplain discretization elementsstrongly reduced the effort and time needed on inputdata preparation. Additionally, the use of a parallelizationscheme through OpenMP method for two loops of thefloodplain model has proven to be a satisfactory wayto reduce run time, which may allow higher level offloodplain spatial discretization.

The model was capable to reproduce the flow regimealong main channels of Paraguay River and its tributaries.Distinct cases were satisfactorily simulated, such as riversthat present enormous loss of water from main channelto floodplain during the floods, rivers where this lossoccurs during both the flood and dry periods, rivers wherethere is a gain of water from floodplains to main channeland rivers which do not exchange water laterally. Forinstance, it must be emphasized that the ability of theproposed model to simulate the complex behaviour ofchannel–floodplain interactions specifically in the regionof the S. Lourenço and Piquiri Rivers, in which thewater spills over the channel of the S. Lourenço River,inundates the floodplain and propagates over it untilreaching and contributing to the flow of the main channelof the Piquiri River.

The SIRIPLAN was also able to reproduce the Pantanalseasonal flood pulse, with estimates of inundated areavarying from 35 000 to 45 000 km2 in the dry period andranging from 100 000 to 126 000 km2 in the wet period.These estimates were consistent with the results obtainedby a earlier study, which was based on coarse-resolutionsatellite images and analysed a distinct period of time,but with

***proof of your article attached, please...

Documents