Numerical modelling of the catastrophic flooding of Santa Fe City, Argentina

Intl. J. River Basin Management Vol. 4, No. 2 (2006), pp. 1–14

© 2006 IAHR, INBO & IAHS

Numerical modelling of the catastrophic flooding of Santa Fe City, ArgentinaC.A. VIONNET, Department of Engineering and Water Resources, Universidad Nacional del Litoral, CC217, 3000 Santa Fe,Argentina; CONICET (National Council for Scientific and Technological Research)

P.A. TASSI, CONICET (National Council for Scientific and Technological Research); Currently at the Applied Math Dept, Univ.Twente, The Netherlands

L.B. RODRIGUEZ, Department of Engineering and Water Resources, Universidad Nacional del Litoral, CC217, 3000 Santa Fe,Argentina

C.G. FERREIRA, Santa Fe State Dept of Water Resources

ABSTRACTThe large plain in the lower basin of the Salado River in west-central Santa Fe State, Argentina, sustains a prolific agricultural activity vital for thelocal economy. In April–May of 2003, the region suffered the most devastating flood on record for the Salado River, triggered by heavy rains in itslower basin. The west side of the State Capital, Santa Fe City, located at the mouth of the Salado River, was suddenly flooded when a protective leveefailed. People living in the floodplains near the city, accustomed to coping with floods characterized by slowly rising water, faced a sudden increaseup to 4 m of water in a matter of hours. During the flooding of one third of the city, nearly 120,000 people were suddenly displaced from their homes,23 people died as a direct result of the flood, and other 43 are believed to have died from post-traumatic distress. This work presents a numericalreconstruction of the event, emphasizing the chain of miss-management decisions made over the years that have a share in the worst environmentaldisaster of Argentina’s recent history.

Keywords: Levee-break; Dam-break; Floodwall failure; Flood-wave propagation; Flooding and drying; Shallow water flows.

1 Introduction

A popular current perception is that man’s activities have dis-rupted most of the earth ecosystems. Hurricane Katrina, whichwrecked New Orleans in 2005, the flood of Mozambique inAfrica in 2000 [15], the landslides of Venezuela in 1999 inSouth America [30], and the Hurricane Mitch that devastatedCentral America in 1998 [15] are just few examples of largelyunexpected natural phenomena with devastating consequencesfor hundreds, or even thousands of people. The magnitudeand characteristics of these catastrophes constitute a clear sig-nal that engineers, scientists and policy makers are compelledto find a range of options to mitigate river and coastal floods[28]. From a scientific point of view it is difficult, however, toattribute those catastrophes to global warming [2]. Most often,a water-related disaster is directly linked to the sudden failureof a flood protection work [43], a dam break [19, 20], or eventsunamis generated in reservoirs by landslides [23]. Neverthe-less, the recent flood of Santa Fe City in Argentina [40] thatcaused exceptional life and property losses has renewed localpublic interest in determining if the disaster was mainly pro-voked by the faulty design of a flood protection work ratherthan by an extreme river flood event that was beyond the design

Received xxxx. Accepted xxxx.

1

capacity of the protection structure. The catastrophic floodingof Santa Fe City was indeed initiated by high intensity rainfallconcentrated in the lower basin of the Salado River (Figure 1).However, the rising waters of the Salado River first overtoppeda precarious 5 m long levee located adjacent to the containmentwall of a properly engineered levee, known as the West levee(Figure 2). Whereas the 7 km long West levee was built with fineand medium-size sand dredged from the Salado River bed andelevated to its final level with a mixture of properly compactedsoils, the short non-engineered levee was constructed with loosesoils and eventually armored with sandbags, as happened dur-ing the big flood of 1998 (Table 1). In brief, the purpose of thesmall non-engineered levee was to provide some protection whilepermitting access to the golf course located behind the levee onthe river floodplain. Meanwhile, the non-engineered levee wasmeant to be reinforced during periods of high waters to close theexisting gap between the West levee and the higher lands nearby(Figure 2).

The water overtopping the non-engineered levee eroded thefoundation of the containment wall of the West levee until itfinally collapsed at 8:30AM on April 29th, 2003. In a few min-utes, the incoming water eroded approximately 120 m of the endportion of the West levee, which resulted in the rapid flooding

2 C. A. Vionnet, P. A. Tassi, L. B. Rodriguez and C. G. Ferreira

Figure 1 Left: Image from Argentinean satellite SAC-C taken on Jun 15/2002, cropped to Santa Fe state limits (see incept). Right: City map showingthe location of several references used in this work.

Figure 2 Left: Schematic of the West levee showing the gap to provideaccess to the golf course. Contour levels are referred to mean sea level.Right: Translation of layout 34 c of the original West levee project [31].The non-engineered levee was 2 m shorter than the West levee.

Table 1 Peak discharges and water elevations of floods (gage locationsare shown in Figure 1, stage readings are referred to mean sea level).

River Year Discharge (m3/s) Stage (m)a,b Stage (m)c

Paraná 1905 50,000a 15.35a

1966 42,000a 15.14a

1982–83 62,500a 15.53a

1992 54,000a 15.63a

1998 47,000a 15.36a

Salado 1973 2,400b 18.52b 13.94c

1998 2,650b 18.43b 14.31c

2003 3,100c–3,800b 19.22b 14.70c

aSanta Fe harbor gage (Paraná River); bSR70 gage (Salado River); cINALI gage(Salado River).

of lowland neighborhoods practically without warning. A fewhours later, the water elevation inside the city at a point locateddownstream of the breached area was 2.48 m higher than thewater elevation on the river side [11].

1.1 Extreme floods in the 20th century

Santa Fe City is located at 31◦ south latitude, and oriented fromsouth to north on the wedge formed by the confluence of theSalado River on the west and the alluvial system of the ParanáRiver on the east (Figure 1). There are approximately one half mil-lion inhabitants living in the metropolitan area today, a significantproportion of whom occupy both floodplains of the Salado andthe Paraná rivers. Extreme discharges of the Paraná River havecaused severe property damage to Santa Fe and nearby townsin the past, most notably the flood of 1982–83. Whereas stagerecords of the Paraná River span a century, the extreme floodof 1905 being the oldest on record, systematic measurements ofthe Salado River started in 1952. Thus, one of the few recordedextreme floods of the Salado River occurred in 1973, with anestimated peak discharge of 2,400 m3/s under the narrow bridgeof the Santa Fe-Rosario freeway (Table 1). On that occasion,and in spite of the destruction of the freeway bridge and muchother state and private property damage as well, the bridge wasrebuilt with the same span. In the meantime, more and more low-income people have moved onto the floodplains of both rivers.This influx has been driven mainly by: (i) the economic crisisof Argentina that forced many residents of rural areas in north-east Argentina to seek better opportunities in larger cities, and(ii) the lenience of official authorities to enforce the State lawthat explicitly prohibits settlements in flood-prone lowlands.

After a major flood of the Paraná River in 1992, the City ofSanta Fe, in concordance with State authorities, decided to buildseveral flood-control structures on the Paraná River in the east partof the city, as well as along the Salado River on the west side. Withthe assistance of loans from the World Bank and the governmentof Kuwait, the State invested about 74 million dollars in newroads, bridges, and levees all over the region. The main flood-control structure for Santa Fe City against floods of the SaladoRiver was the so-called West levee, a northern continuation of anexisting levee named Irigoyen (Figure 1). After completion, the

Numerical modelling of the catastrophic flooding of Santa Fe City, Argentina 3

whole levee ran parallel to the river for about 7 km, rising 5.2 mon average above the floodplain level. While the first two stagesof construction of the West levee were completed, the third andlast stage was never completed. The 3-km long third stage ofthe levee was meant to close a protective ring around the city’snorthwest side. In the meantime, lured by the sense of securitytransmitted by the new flood-control structure combined with lowreal state values, an increasing number of people moved into thewest side of the city. Simultaneously, very low income peopledirectly settled onto public lands in the river floodplain.

1.2 Extreme flood of the Salado River in April 2003

Extremely high precipitation saturated the lower basin of the Sal-ado River during the last quarter of 2002 and the first quarter of2003. From April 20 to April 29, 2003, around 400 mm of rainfell in some regions of the basin (Figure 3), whose annual averageprecipitation is about 1000 mm. The storm return period at oneselected station was estimated to be over 100 years [11]. Thisprecipitation generated a flood wave whose peak discharge wasestimated at approximately 4,000 m3/s, according to measure-ments taken with an Acoustic Doppler Current Profiler upstreamof the State Road 70 (see SR70 gage location in Figure 1). Thedischarge data were obtained on a floodplain-main channel crosssection located upstream and far away from the State Road 70’s

46042038034030026022018014010060

SC

M

FO

Rai

nfal

l [m

m]

0

100

200

300

400

SE

P

OC

T

NO

V

DE

C

JAN

FE

B

MA

R

AP

R

MA

Y

JUN

JUL

AU

G

Fortin Olmos (FO)

Rai

nfal

l [m

m]

0

100

200

300

400

SE

P

OC

T

NO

V

DE

C

JAN

FE

B

MA

R

AP

R

MA

Y

JUN

JUL

AU

G

Margarita (M)

Rai

nfal

l [m

m]

0

100

200

300

400

SE

P

OC

T

NO

V

DE

C

JAN

FE

B

MA

R

AP

R

MA

Y

JUN

JUL

AU

G

San Cristobal (SC)

Rainfall averageRainfall 2002-2003

Figure 3 Spatial distribution of the heavy precipitation (mm) that fell over the lower basin of the Salado River from April 20th to April 29th. Thedots point location of the rain gages. Right: Average vs precipitation fell during 2002–2003 on three key gage stations.

bridge to avoid the typical high-speed approaching flow prevail-ing in its vicinity. Submerged trees, and submerged wire fencesused in Argentina to limit land properties, could have posed theonly hazardous situation faced during measurements. The returnperiod for the flood was estimated to be over 250 years [1].

The river overtopped its narrow and wandering channel andspread over its floodplain with unusual strength, destroyingbridges and roads, and isolating some small towns in its runtoward its outlet to the Paraná River alluvial system near Santa FeCity. People living in west Santa Fe, accustomed to coping withfloods from slowly rising water levels, faced a sudden increase inwater level of about 4 m in a few hours after the breaching of theWest levee. The flooding took a heavy toll, with 23 people dieingin the flood, and 43 reportedly dieing in the following weeks dueto post-traumatic distress [8]. The new 20 million dollar facil-ity for the Children’s Hospital, a pride of the city and the region,had 1.5 m of water in its first floor, flooding very expensive equip-ment and forcing an emergency evacuation of dozens of in-patientchildren. In the chaos of the first 24 to 48 hours of the flood, anestimated 120,000 people were displaced. Schools, clubs, gov-ernment facilities, churches, and non-government organizationssheltered about 50,000 people, while approximately 70,000 peo-ple are believed to have sought shelter with friends and familyall over the city. One third of the city was under water at somepoint, including middle and upper-class neighborhoods. One of


the biggest challenges during the emergency was to control theoutbreak of diseases related to the stagnant water trapped insidethe city for about 3 to 4 weeks. In total, 162 cases of hepatitis and111 cases of leptospirosis were reported in the weeks followingthe catastrophe [8]. The total loss attributed to the flood is beingestimated on the order of 1,000 million dollars [8], though a largeproportion of that is being borne by individuals and, as such, isstill unreported.

1.3 Objectives

The use of models to test “what-if” failure scenarios of floodprotection works was not a widespread practice at the local waterauthorities’ level before the event of 2003. Right after April 29of 2003, the city flooding seemed to provide a remarkable studycase for water resources planning students and prospective engi-neers as well [13], given the apparent cumulative managementerrors made over the years that magnified the consequences ofan extreme rainfall event in the lower basin of the Salado River.Consequently, a research project was initiated as part of a broadereffort to extract some lessons that may be later useful for deci-sion makers on water related issues at the State level. The ultimategoal of the project was to stress that only improving the ability toquantify and manage different flood risk scenarios could lead toeffective mitigations actions to protect people, private propertyand public infrastructure during the occurrence of such extremeevents.

As an initial step towards that goal, this work was particu-larly designed to: (1) reproduce numerically the most strikingfeatures of the wave front that flooded one third of the city ina few hours, seeking a favorable comparison with field obser-vations of inundated depth and flooded area, and (2) compilethe errors made over the years by water resources planners andrelated authorities that contributed to the worst environmentaldisaster of Argentina’s recent history.

A significant part of this study relies on the vast research andapplications accumulated over the years in the simulation of time-varying inundation modelling due to natural river floods [3, 21,22, 17], coastal inundation by storm surges, dam breaks [19, 20],or failure of protective works [16, 43]. These type of problems areusually characterized by a wave front, or water elevation jump,propagating either on initially dry or wet bed conditions. Thenumerical techniques employed in such studies are often basedupon finite difference [7, 29], finite element [18], and variants ofthe finite volume methods [38].

In the next section, the main geomorphologic characteris-tics of the lower basin of the Salado River are given first.Then, the underlying hypotheses that support the computations,which closely resemble the steps adopted for the Malpasset dam-break study [20], are summarized. Next, the 2D Shallow WaterEquations (SWE) are briefly introduced, followed by the treat-ment of the topographical and boundary data required for themodelling purposes. The results produced with the code Telemac-2D, and their comparison with field data collected during theevent, are then discussed. Finally, the work closes by examining

some cumulative management errors made over the years thatdramatically worsen the outcome of the flood.

2 Materials and methods

The lower basin of the Salado River, whose loose limits com-prise a relatively flat area of approximately 30,000 km2 locatedmostly within the State of Santa Fe boundaries (Figure 1), sus-tains considerable industrial and agricultural activity that is vitalfor the regional economy, as reflected by the following numbers:in 2002, the region yielded 450 thousands tons of corn, 2 milliontons of soya beans, 50 thousands tons of sunflower, 510 thousandtons of wheat, 450 thousand tons of sorghum, 250 million kg ofmeat, and 1 thousand million litres of milk, totalling a productioncapable of feeding up to thirteen million people [6].

The basin itself is part of the large north Pampa system ofArgentina [27], which has been undergoing a climate transitionto more humid conditions in the last 30 years [14]. Today, theregion experiences a subtropical humid climate with a meanannual temperature of 18◦C, and with a precipitation gradientfrom east to west ranging from 900 to 1,200 mm, respectively, inthe period 1971–2000 [14]. The main geomorphological featureof the lower basin of the Salado River is the presence of an ele-vated block bounded to the west by the Tostado Selva fault [27],inducing a topographical gradient of 10 cm/km in WE directionand 5 cm/km in N-S direction; therefore, the river runs in a NW-SE direction when entering the Santa Fe State limits until nearits junction with the Calchaquí River, then turns south toward itsmouth near Santa Fe City. By contrast, the hydraulic gradient ofthe Salado River in overbank-flow condition is on the order of20 cm/km. The Calchaquí River drains the region known as BajosSubmeridionales (BS), which is a markedly flat region that is fre-quently flooded by heavy rains (Figure 1). The Calchaquí River isestimated to have contributed with about 1,000 m3/s to the floodwave that devastated one third of Santa Fe City onApril 29, 2003.There are four other tributaries to the Salado River, among themthe so-called Pantanoso River, which contributed approximatelyanother 500 m3/s to the flood peak [11].

The Salado River is a meandering river with an average chan-nel width of 150 m, bounded by very eroded banks, and flowingwithin a floodplain belt between 1500 and 2000 m wide. Thefloodplains are covered with sparse grassland and bushes dueto the salty soil left by the periodic flooding of the SaladoRiver, which normally carries an extremely high concentrationof salts. The annual average discharge is 135 m3/s for the period1954–2002, increasing to 176 m3/s if the period 1971–2002 isconsidered instead [14]. The mean discharge of the Salado Rivercorresponds to water levels about 1 m below the elevation of theriverbanks. Thus, the bank-full situation corresponds to higherdischarges of approximately 300 m3/s [1, 25].

2.1 Mathematical model

At the time of selecting a mathematical model to study the flood-ing of a city there are a number of issues that may be worth


considering first. One dimensional (1D) models have the advan-tage of being relatively simple, not very data-demanding andcomputationally efficient [16] in comparison to 2D models. Nev-ertheless, a 2D computational code with the ability to handleflooding and drying land processes has the added advantage ofsimulating the spreading of the flood wave inside a city in a muchmore realistic situation. Consequently, a 2D model was chosenfor this work.

The 2D computations presented here were achieved in twostages or modes: first, in steady-state mode, and second, in dam-break mode, as is commonly done in simulation studies of flood-wave propagation over initially dry land [20]. The steady-statemode refers to the situation where water was not allowed to enterthe city through the levee breach, and the transient or dam-breakmode refers to the condition where water was allowed to flowthrough the breaching area over initially dry land.

In spite of the fact that the collapse of the protection workitself occurred in two stages, as explained in the introduction,the computations performed for this study were based on thefollowing assumptions: (i) for simplicity in the computations, asudden collapse of the protection levee, totalling a breach widthof approximately 150 m, was considered, (ii) the water enteredthe city only through the levee breach (i.e, the incoming overlandflow water from higher lands to the north of the breach (Figure 4)was ignored in the computations), (iii) the conveyance capacityof the main channel of the river was negligible relative to the con-veyance capacity of the river floodplains (i.e., if Qmc representsthe bank-full discharge of the main-channel andQfp the dischargeof the whole-channel floodplain system considering an even mainchannel, it follows that the computation assumed Qmc/Qfp � 1for all practical purposes), and consequently, (iv) the hydraulicresistance of the main-channel floodplain system was uniformfor the steady-state computation and increased for the transientsimulation, differentiating river from urban areas [21, 17], (v)the upstream river discharge, imposed as an inflow boundarycondition onto the modelled area, remained constant during thetransient computation, (vi) the flowing water was bounded from

Figure 4 Spot image taken on May 3, 2003, showing Santa Fe city surrounded by water. Flow discharges measured and/or estimated during the eventare indicated on the image. The inset delimits the area of the breached levee as shown on the companion areal photograph, whose view is from northto south (image provided by INA – National Water Institute, Buenos Aires, Argentina; photograph provided by “El Litoral” newspaper, Santa Fe,Argentina).

above by a free surface free of wind effects and from below by arigid and smooth bottom, and (vii) the water motion was predom-inantly horizontal, where small-scale effects acted as an effectiveeddy viscosity on the large-scale motion [5].

For the transient simulation, the hydraulic resistance withinthe main channel-floodplain system is assumed to preserve thevalue used for the steady-state simulation, albeit higher withinthe city to represent the augmented surface roughness effect typ-ical of an urban area. Other assumptions are justified on thegrounds that some simplifications are required in order to per-form the computations (e.g., a sudden levee failure is muchmore tractable than a progressive failure [43]), whereas othersare justified on the grounds of available data. First, a crude esti-mate done during the emergency yielded an incoming overlandflow of about 50 m3/s (Figure 4), compared with the estimatedincoming 600 m3/s through the levee breach [11]. Secondly, theSalado River has a bank-full discharge of the order of 300 m3/s[1], compared with the 4,000 m3/s flowing throughout the wholemain-channel floodplain system for the flood studied here. Anindependent check using the Lateral Distribution Method [41]yielded a slightly higher bank full discharge than 300 m3/s albeitconsistent with the value cited by Acuña [1]. All these esti-mates yield Qmc/Qfp � 0.08 and, so adding the main-channelconveyance capacity should have no major effects on the com-putations. The validity of this hypothesis was also reported byHervouet [20] for the Malpasset dam-break study. In other words,as long as the constraint Qmc/Qfp � 1 is satisfied, any errorthat could be embedded in the topographical representation ofthe river main channel becomes, within some prescribed toler-ance, irrelevant. An added advantage is that there is no majorneed to distinguish hydraulic resistance and eddy viscosity valuesbetween the main channel and the floodplains, as is normallyrequired in situations where the conveyance capacity of bothcomponents are of the same order [41] (e.g. the Paraná Riveris characterized by a ratio Qmc/Qfp � 1 [35]).

With the aforementioned hypotheses in mind, if the verticalextent of a flowing water layer, h, is assumed small in compar-


ison with the horizontal length of the wavelike motion of thefluid, l (i.e., if h/l � 1), the motion of a free-surface flowcan be analyzed by a horizontal 2D formulation known as thelong wave approximation [34]. The resulting equations, knownas the Shallow Water Equations (SWE) or as the Saint Venantequations, can be obtained from the depth-integrated form of theNavier–Stokes equations of motion, previously averaged overturbulence:

∂h

∂t+ ∇ · uh = 0 (1)

∂u∂t

+ u · ∇u g + ∇h = −g∇η − τ b

ρh+ 1

h∇ · (hνt∇u), (2)

where h is the water depth, u is the depth-averaged velocity vectorwith (u, v) components in the (x, y) horizontal directions, respec-tively, t is the time, η is the elevation of the stream bed abovedatum, g is the acceleration due to gravity, ρ is the fluid density,τ b is the shear stress vector acting on the stream bed, and νt is theturbulent eddy viscosity that models the lateral stress effects thatinclude viscous friction, the so-called Reynolds stresses, and thedifferential dispersion terms originating from the lack of verticaluniformity of the horizontal velocity field [42]. In the absenceof wind effects, two additional relationships must be specifiedto close the problem: (i) the bed resistance, τ b = (τbx , τby ), and(ii) the turbulent eddy viscosity. For the former, the classicalsquared function dependency on the depth-averaged velocity isused

τbx = ρCF|u|u, τby = ρCF|u|v, |u| = (u2 + v2)1/2, (3)

where the friction coefficient CF,

CF = u2∗|u|2 , (4)

can be specified with either Manning or Keulegan relations

CF = gn2h−1/3, CF =[

2.5 ln

(11h

ks

)]−2

, (5)

respectively, where n represents the Manning roughness coef-ficient, and ks represents an effective roughness height. In theabove equations, u∗ denotes the shear velocity, defined by

u∗ = √|τ b|/ρ, |τ b| = (τ2bx

+ τ2by

)1/2(6)

For the second closure relationship, a proper estimate of thedepth-averaged eddy viscosity νt is given by the Elder model [12].

νt = αu∗h (7)

= α√

CF|u|h (8)

The model (7) is used here to specify an average eddy viscos-ity value, held constant during the simulations, in which casethe mean shear velocity is estimated according to the proce-dure of Nezu and Nakawaga [32]. Alternatively, (8) can be usedwhenever the friction coefficient CF is known. The parameter α,which could be considered the dimensionless eddy viscosity, mayrange from approximately 0.07 to about 0.30 [41]. The standarddepth-averaged value for an infinitely wide channel obtained withequilibrium turbulent boundary layer theory is α = κ/6, where κ

is the Von Kármán constant (see [39], for example). This estimate

yields the lower bound of 0.07. Higher values of α are found inthe work of Fischer et al. [12].

2.2 Numerical solution of the SWE

The Telemac-2D code is used in this study to solve the primitiveform of the SWE given by Eqs (1) and (2) with the finite elementmethod [18]. The standard version of the computational codeTelemac-2D is based on a fractional step approach, where thevelocity components (u, v) are solved first along the characteris-tic curves, and the Streamline Upwind Petrov Galerkin (SUPG)method is applied to h in the continuity equation to ensure massconservation. In the SUPG formulation, the standard Galerkinweighting functions are usually set equal to the basis functionswith the exception of the convective terms, where perturbedPetrov-Galerkin functions with balancing tensor diffusivity areemployed [26, 4]. Briefly, the Telemac-2D code starts from somegiven initial and boundary conditions, and advances the solutionone time step, �t, for u, v, and δh, since the mass conservationconstraint is treated explicitly in the form δh+�t∇ · (uhn) = 0,where hn represents the water depth at the previous time step,and δh = hn+1 − hn is the computed free-surface increment ateach time step.

All results presented in this work were obtained with the prim-itive form of the SWE using a computational domain partitionedinto linear triangular elements. Thus, once the computationaldomain is partitioned into a structured or non-structured meshwith Ne finite elements and N nodes, the dependent variables(u, v) and h are expanded using equal-order interpolation func-tions. In the second step, the remaining terms not included in thefirst step are coupled through an implicit time scheme, and theresulting linear system is solved with a variant of the conjugategradient method [18].

2.3 Model area and boundary conditions

The digital representation of the topography of the affectedarea for the post-flood assessment was developed using offi-cial cartographic maps provided by the Geographic Institute ofthe Argentinean Army (IGM). Local topographic features whichwere not represented in the maps, such as levees in the floodplain,were later integrated into the digital terrain model (DTM) usedfor the numerical simulation. The lack of information was partic-ularly severe in the northern part of the modeled area, near SR70,where a few contour levels 0.5 m apart were available for sucha low gradient river. Data from a few cross sections of the mainchannel of the Salado River were used, though none were avail-able on the area near the bridge of the Santa Fe-Rosario freeway.So, in this case, an old data survey of the cross section at the free-way bridge was used to recreate the riverbed topography. In brief,the interpolation of the topography data to a high-resolution finiteelement mesh (Figure 5) was based on high resolution elevationdata inside the city, particularly within the flooded urban area,and on sparse data in the northern part of the modeled area. Nev-ertheless, the few data available on the floodplain were deemedadequate under the premise that the conveyance capacity of the


(a) (b)

Q

P

Q

(c)

=3800 m3/s

η + h = const

η+h= [14.0-14.5] m

Salado

river

SF

eha

rbou

rga

ge

INALI

gage

Figure 5 (a) High resolution mesh used to define the topography of the modeled area (mesh 1), (b) intermediate mesh used during the computations(mesh 2), (c) final mesh (mesh 4); Inset above: detail of the mesh around the breached levee; Below, schematic of gage positions and outflow BC.

main channel is negligible in comparison with the conveyance ofthe floodplains.

A series of post-processing procedures were applied to detectpotential discrepancies between the DTM and ground obser-vations, particularly near the area of the breached levee. Bothresolutions, the vertical and horizontal scales of the DTM, werebased upon the topographic information available. The result-ing DTM was estimated to have few centimetres of uncertaintyin the lower half portion, which covers the zone of interest inthis study with the exception of the riverbed near the freewaybridge.A maximum vertical uncertainty of±30 cm was estimatedat the northern half of the modelled area, where contour levelswere available at 50 cm intervals. These estimates arose fromthe comparison of the interpolated DTM with few topographicprofiles surveyed by the Santa Fe Water Resources Departmentduring the emergency. The horizontal spatial resolution is pro-portional to the size of the finite elements used for interpolatingthe topography, i.e., 25–40 m in the river channel, 50–75 in thefloodplain, and 250–350 m in the northern portion of the mod-elled area (Figure 5a). The high-resolution finite element meshcontaining topography data (Figure 5) was, in turn, used to inter-polate bed elevation data to several coarser meshes used for thecomputations (Table 2).

The limits adopted in the final computational domain resultedfrom trial and error, where one of the main objectives was todetermine the approximate extent of the flooded area in order tosave memory and computing time. To that aim, a very coarse

Table 2 Different meshes used for the computations.

Mesh Nodes Elements Used for

1 5,212 9,848 topography2 2,075 3,744 hydrodynamics3 4,190 7,765 ”4 4,464 8,289 ”

mesh was first used, as indicated in Table 2. The final compu-tational domain occupies about 88 km2, extending 25 km in N-Sdirection from 3 km upstream of the SR70 up to the point wherethe Salado River discharges into the alluvial system of the ParanáRiver (Figure 5). The average width of the floodplain along themodeled reach was 2,000 m. Included within the model domainwere the lowland areas in the western part of the city. Meshes 3and 4 were defined over the same computational domain, thelatter being a refinement of the former. So only one of them isshown in Figure 5c.

Figure 4 included an aerial photograph of the breached leveeas well as a Spot image of Santa Fe City and its surroundingstaken on May 3, 2003. The image shows different locations wherethe river discharge was measured and/or estimated: (a) 200 m3/sdiverted from the Salado River to the alluvial system of theParaná River in the northern part of the city (at one point, thecity was completely surrounded by water), (b) 3,800 m3/s com-ing down from the north near SR70, and measured with an


ADCP, (c) 600 m3/s coming into the city trough the levee breach,(d) 50 m3/s entering the city as overland flow from relatively highland situated north of the breached levee, and (e) 3,100 m3/s dis-charging into the Paraná alluvial system at the Salado River’soutlet, and measured with an ADCP. The peak flood dischargewas estimated at 4,000 m3/s upstream of SR70 before reachingthe flow diversion to the alluvial system of the Paraná River.Thus, the combined flows illustrated in Figure 4 yield an esti-mated error on the flux mass balance on the order of 50 m3/s,which represents a relative error of 1.25% with respect to thepeak flood discharge [11].

Boundary conditions were defined at the upstream and down-stream borders of the computational domain. Upstream, apiecewise constant-inflow condition was set until the flow ratereached the observed peak discharge value of 3,800 m3/s. Thisvalue was later decreased to 3,600 m3/s to determine the sensi-tivity of the model response to changes in the prescribed boundaryconditions. The downstream boundary condition was defined asa constant water level ranging from 14.0 to 14.5 m along theoutflow cross section (see Figure 5).

3 Results

Several Landsat 5 and 7 images, as well as topographic data,were processed with ENVI [10] to define the computationaldomain. Mesh generation and modifications to bed topographydata were treated with the SMS interface [33], and later convertedto Telemac-2D portable files with the Stbtel interface [37]. In turn,the appropriateness of the results was assessed through the use ofthe Rubens interface [37] and the visualization software Tecplot[36]. It was necessary to write a simple code to make Telemac’soutputs compatible with Tecplot.

The flow through the breached levee is a function of the dif-ference between water levels behind the levee (in the river) andimmediately downstream. At the beginning, the breached leveebehaves as a lateral weir, branching the main flow discharge andinducing changes in the local water surface elevation of the rivernearby the gap. A much simpler set up would be given by freezingthe water surface elevation behind the levee without solving theriver flow behind the levee. However, the comparison of solutionsobtained by different modelling strategies is not within the scopeof the paper. The fully coupled flow simulation between the riverflow behind the breached levee and the forward wave propagat-ing on initially dry land after breaching was chosen instead. Thisscenario closely mimics the actual conditions observed duringthe emergency.

Consequently, the computations were performed in two steps,a steady-state and a transient simulation. The steady-state sim-ulation was carried out with the aim of obtaining a reasonabledistribution of water elevations and flow velocities on theriver side that would drive the transient scenario. Achievinga steady-state condition within the floodplain limits (i.e., withthe levee breach closed), was actually much more computation-ally demanding than the transient simulation. Instead of startingfrom zero initial conditions, which converged extremely slowly

for a solver accuracy of 10−7 with the available computationalresources, the computation was initiated assuming a water eleva-tion distribution (say zw = h + η, = const) compatible with thebed slope along the river over all of the computational domain.This initial interpolation was such that the condition h > 0was satisfied for all nodes pertaining to the river domain (i.e.,within the floodplains limits). Consequently, it was necessary toraise the ground elevation, η, of the mesh nodes inside the cityin order to avoid an artificially flooded situation. Thus, wher-ever the condition η > zw was satisfied, a zero water depthwas assigned to the corresponding node, as required. This initialcondition speeded up the convergence to the sought steady-statesolution quite remarkably by imposing a staircase inflow hydro-graph, starting with 500 m3/s and progressing up to the required3,800 m3/s with increments of about 1,000 m3/s. Once the steadystate solution was available, the ground elevations of the city wererestored to their original values just by trading out the topographyinput files.

Because some numerical instabilities were observed near theSanta Fe-Rosario freeway bridge, the mesh around that area washighly refined to model accurately the effects of the flow contrac-tion and the sharp increase on the free surface elevation upstreamof the structure. Small elements were also placed at the breachingzone and all along the inner side of the West levee, since the wavefront initially propagated over the freeway belt that runs parallel tothe levee (see the city map in Figure 1 for details). In brief, severalmeshes with different levels of refinement were used (Table 2)until the mass flux discrepancy was less than 1.5%, which wasdeemed negligible and, therefore, likely mesh-independent.

Uniform values of the hydraulic resistance, in terms of theManning’s roughness coefficient n, and the eddy viscosity wereused across over all of the computational domain. The roughnesscoefficient was later increased for the dam-break mode simula-tion in order to assess the additional resistance effects due tothe urban area [21, 17]. Consequently, initial values of n were setbetween 0.021 and 0.025 for the steady-state simulation, and laterincreased to 0.030 for the transient simulation, whereas the eddyviscosity was set equal to 0.1 m2/s throughout the simulations.Usually, the value of the eddy viscosity has rather limited impacton the SWE results, with the model output being more responsiveto changes in the hydraulic resistance coefficient [20, 41]. How-ever, instead of calibrating the model by adjusting the hydraulicresistance coefficient to achieve the best fit to the few availabledata, the model was calibrated by varying the water elevationat the outflow boundary within the physical range 14.0–14.5 m,until the observed stage value at the INALI gage was matchedat the moment of interest, as it is explained next. This strategywas adopted to work around the uncertainty present in the waterelevation value at the river confluence with the alluvial system ofthe Paraná River during the flood peak. At that moment, there wasabout a 1 m difference between the readings of the SFe harborand the INALI gages (Figures 1 and 5).

Figure 6 shows the free surface at the end of the steady-state simulation. Note the considerable difference in the freesurface elevations upstream and downstream of the Santa Fe-Rosario freeway bridge caused by the severe flow contraction.


Figure 6 3D projection of the steady-state simulation solution with an inflow BC of 3,800 m3/s.

The simulated difference was 1.4 m while the observed differ-ence was estimated to be around 1.2 m, although this figure isnot known to be reliable. This discrepancy, if any, could beattributed in part to the inaccurate representation of the alluvialvalley topography adopted for the simulation in the upper portionof the computational domain, upstream of the freeway bridge,which was on the order of ±30 cm as explained in Section 2.3,and more likely to the combined effect of a poor representationof the riverbed topography and the hydraulic resistance on thebridge.

Figure 7 shows the first stages of the wave front propagat-ing inside the city at different times when the gap closure wasremoved for the transient simulation, obtained with a �t = 1 safter starting with the initial condition depicted in Figure 6. Theinset shows a detail of the velocity field and water depth contours.The gap was assumed to be 150 m wide and to extend down to

t=5 s

t=100 s

t=50 s

t=25 s

t=400 st=200 s

t=500 s

t=1000 s

Figure 7 Wave front propating over initially dry land inside the city, after a sudden breach of the West levee.

ground level as shown in Figure 2. To complete the transientsimulation, larger �t = 2, 5, 25 and 50 s, were used. From somepreliminary computations, it was found that the solution obtainedwith �t = 25 s reproduced the features of the solution obtainedwith finer time increments quite remarkably.

The river discharge to the city through the breach was com-puted with the following line integral, defined over a polylineacross the breach joining the two extreme points P and Q

(Figure 5c)

Flux|P̂Q =∫

P̂Q

(uh · n)ds (9)

where the unit vector normal to the differential line elementδs = (δx, δy) is n = (−δy/δs, δx/δs), where δs = (δx2 +δy2)1/2.Consequently, it was advantageous to introduce the standardstream function concept (i.e., an unknown scalar function ψ(x, t))


such that

uh = −ψy, vh = ψx (10)

Then, in the limit δs → 0, the above line integral becomes

Flux|P̂Q =∫

P̂Q

−uh dy + vh dx

= ψ(xQ, t) − ψ(xP, t) (11)

The line integral (11) was evaluated for each time step of thetransient computation using, for compatibility with Telemac-2Dsolution method, 1D linear piecewise functions locally definedover each side of the finite elements belonging to the polylineP̂Q. The outcome of this evaluation is depicted in Figure 8 forthe same elapsed time as shown in Figure 7. The graph showsthat the discharge entering the city attains a maximum value fewseconds after the breaching of the levee, then smooths out toa constant value of approximately 540 m3/s. This value is quiteclose to the estimated value of 600 m3/s obtained by trackingfloats, and reported by the Water Resources State authorities [11].These first 1,000 s of the transient simulation were obtained with�t = 1 s after starting from the steady-state solution depicted inFigure 6, whereas the remainder of the simulation was continuedwith �t = 25 s. Though not shown here, the inflow rate depictedin Figure 8 diminishes as the lowland neighborhoods filled upwith water with increasing time. The inflow rate is quite sensitiveto the assumed shape of the gap. Nevertheless, the computedvalues of the incoming flow were obtained assuming a breach ofrectangular cross-section whose lower lip had the bed elevationshown in Figure 2, which was deemed reasonable given the lackof topographic data in and around the breaching area.

A 3D projection of the progression of the flood wave insidethe city at different times is plotted in Figure 9. The view point

Figure 9 3D projection of the flooding of west Santa Fe at different times. Left: Views from the southeast, Right: Views from the northeast, markedVSE and VNE in Figure 10, respectively.

t (s)

Flu

x of

vol

ume

on b

reac

h (m

3 /s)

0 200 400 600 800 10000

100

200

300

400

500

600

700

0.0

0.5

1.0

1.5

2.0

2.5Qups = 3600 m3/s , n = 0.025

Qups = 3800 m3/s , n = 0.025

Qups = 3600 m3/s , n = 0.030

h (m)

Figure 8 Flux of volume entering the city after the sudden breach ofthe West levee. Water depth is also shown as time progresses on a pointlocated nearby the breach.

on the right part of Figure 9 was reversed to get a better sight ofthe large extent of the flood wave in the southern part of the city.The flood extent 25 hr after the breaching compares favorablywith the affected area as captured by the Spot image on May 3(Figure 4). The comparison is just qualitative, since the satelliteimage was taken 3 days after levees were ruptured in six differentplaces to release the trapped water inside the city. One featurethat emerged from the simulation is that the railroad levees builtby French engineers at the end of the nineteenth century (one ofthem still in use) delayed the propagation of the flood wave.

The delay in the flood wave propagation due to the presenceof the railroad levees is also expressed in Figure 10, where thewater depth evolution along the inner side of the West levee isdepicted at different points approximately equally spaced. Thewater elevations measured on the INALI and FyG gages are also


t (hr)

Sta

ge

(m)

20 30 40 5012

13

14

15

16

17

18

FyG

INALI

1 2 3 4 5

propagationstage

reservoirstage 2.48m

a

bFyG

1

2

3

4

5baInali

VNE

VSE

Figure 10 Left: Numerical results of the evolution of the water elevationat different locations inside the city (solid lines); numerical results of theevolution of the water elevation at FyG and INALI (dashed lines); stagereadings (bullets). Right: Location of selected points within modelingboundaries whose water levels time evolution are shown on the left. VNE

and VSE are the viewpoints of the 3D projections of Figure 9. FyG standsfor Furlong and Gorriti streets.

plotted on the same figure for comparison with the numericalresults. The mild depletion of the simulated water level at FyGat the beginning of the transient computation is in response tothe huge amount of water entering the city. Meanwhile, the sim-ulated rising curves of zw inside the city are bounded by the FyGreadings at the end of the simulation time. Around 27 hr afterthe breaching, a measurement taken by the local division of theNational Water Institute (INA) near the river outflow reported awater level difference of 2.48 m at both sides of the West levee. Itwas clear by then that the city had become a reservoir, as shownin Figure 9. In response, 29 hr after the breaching, the Armyproceeded to blow up the West levee in four places and the MarArgentino levee in 2 places (see Figure 1), to release part of thehuge amount of water trapped inside the city. The reported dif-ference of 2.48 m was numerically reached 30 hr after assumingthe sudden breaching (Figure 10). Part of this phase lag can beattributed to the misrepresentation in the model of culverts andbridges of the railroad levees and some other waterways exist-ing within the urban area that could have facilitated the waterspreading along west Santa Fe. The constant value of the simu-lated water elevations at INALI, and point ‘a’ as well, depictedin Figure 10, reflects the time-independent boundary conditionsimposed further downstream. The model was calibrated to matchthe observed values of INALI stage at the end of the simulation.It can be concluded that, in spite of the slow time variation of theflood levels, the numerical results obtained within the bounds ofthe hypotheses are in good agreement with the observed values.

Two stages are clearly identified in Figure 10. The first stageis that of propagation, where the water depth remains more orless uniform while the wave front moves forward relatively fast.The second stage is that of a reservoir filling as characterized bywater levels just moving upward.

3.1 Sensitivity analysis

Without loss of generality, the model response for some giveninput data at t = tm = m�t, m = 0, 1, . . ., and for some

Table 3 Sensitivity coefficients in response to changes in the inputparameter values.

Param. change Model change |δq| Sens. coeff. θ

|δn| = 20.0% 6.7% 0.34|δQups| = 5.3% 4.6% 0.87

set of parameter values a, can be represented by the vectory(u, v, δh; a), solution of the system of algebraic equations pro-duced by the finite element discretization method [18]. Therefore,for a given discretization, the vector y(t; a) represents the vec-tor of unknown nodal values, from which a dependent quantitysuch as the inflow rate to the city through the breach can becomputed. Then, the vector a = (a1, a2, . . . , ak) represents theparametric space embedded into the mathematical model (i.e.,a1 = CF, a2 = νt, a3 = Qups, which is the upstream boundarycondition, and so on [41]). It follows that a precise measure ofthe model sensitivity can be obtained by considering the changesin model responses with respect to the base (or calibrated) statedue to changes (or departure) from the calibrated value ao

k ofits kth parameter ak. Thus, denoting the base state with a zerosuperscript, and upon linearizing the model response functionabout the base state, a change �a in the calibrated value of theparameter ao produces a change in the model output given by�y � (a − ao) · ∂y/∂a|o. The entries of the Jacobian matrixare known as the sensitivity coefficients or condition numbers,θij = ∂yi

(t; ao)/∂aj [9]. Taking the discharge q entering the city asthe model output of paramount interest, the sensitivity coefficientcan be written in terms of the absolute value of the relative errorθ(ao

k) � |δq|/|δak|, where |δq| = |�q/qo| and |δak| = |�ak/aok|.

The model is said to be sensitive if θ > 1, particularly if θ islarge, neutral if θ = 1, or robust if θ < 1. The uncertainty of themodel output attributed to the inherent uncertainty with regardto the precise value of ao

k grows, remains bounded, or attenuates,respectively .

Figure 8 shows the effects of globally increasing the calibratedvalue of the roughness coefficient by 20% and the inflow bound-ary condition by 5.3%. For this analysis, the adopted base statecorresponded to no = 0.025 and Qo

ups = 3,600 m3/s. The result-ing sensitivity coefficients are included in Table 3, where qmax

(i.e., the maximum discharge coming in through the breach) waschosen for estimating the change in model response. Note thatθ < 1 in both cases.

4 Conclusions

Slowly rising waters typically characterize the Salado Riverfloods, whereas the failure of the flood protection work in April2003 forced people to confront a sudden increase in water levels ina matter of a few hours. This transformation of the river behavioris well captured in Figure 10, where the evolution of the recordedwater levels within the floodplain outside the city at the flood peakconfirms not only the validity of the hypothesis (v) used to set upthe mathematical model, but also is in sharp contrast with the sud-den increase of water depths inside the city. Still, the simulated


levels are well within the bounds of observed stages. The sim-ulated inflow rate through the gap, � 540 m3/s, compares fairlywell with the estimated value of 600 m3/s obtained by trackingfloats. Additionally, the simulation shows that the catastrophicflooding of west Santa Fe City occurred in two stages: the wavefront propagation stage, and the fateful reservoir-filling stage.Consequently, a tool like Telemac-2D could have been used dur-ing the design stage of the flood-protection work to assess the riskassociated with the failure of the non-engineered levee used toclose the gap left between the properly designed levee and higherlands nearby.

There are many well documented issues regarding levees andearth dam breaches that could be considered in order to mitigatesituations like the sudden flood experienced by one third of thepopulation of Santa Fe City, Argentina. These issues range fromthe need for a river forecasting system to the need to enforce theState law prohibiting the occupation of flood-prone lowlands.Nevertheless, the authors believe that the main factor behind thecatastrophe was cumulative management errors made over theyears, which in turn magnified the consequences of an extremerainfall event.

There are indications that a similar extreme flood eventoccurred in 1914. An ongoing research program to reconstructit is underway based upon the historical accounts and archivesof the French company that once operated the railroad networkthroughout the State. However, the only available record dataindicates that the 2003 flood was triggered by a 100-year stormover a completed saturated basin [11]. This rainfall caused a floodwith a return period estimated above 250 years [1]. In spite of theflood wave severity running down the Salado River, no portion ofthe levee around Santa Fe that was built to engineering standardsfailed. Moreover, no portion of the engineered West levee wasovertopped with the exception of the lateral weir formed wherethe railroad crossed the protection structure, rapidly blocked withsandbags by neighbors and State workers.

Among the chain of miss-management decisions that havecontributed to the risk of flooding of west Santa Fe City, asexplicitly or implicitly mentioned in this work, are: (1) the free-way bridge was rebuilt with the same span, 155 m, in spite ofbeing washed away during the big flood of 1973. The State neverdecided to increase its span, ignoring technical studies indicatingthe urgency to do so [24]. Currently, the bridge is being rebuiltwith a span of 550 m, (2) the West levee was constructed on thefloodplain itself, diminishing the conveyance capacity of the riverin periods of high waters by approximately a factor of 2, from itsoriginal average width of 2,000 m to the current 1,000–1,200 m,(3) the provisional closure of the West levee was located near anacute turn of the river (Figure 6), where the streambed strongererosive capacity due to recirculation and turbulence related toflow deceleration could have been anticipated, (4) the West leveewas terminated just few meters before reaching higher lands,which serve as a natural levee constructed by the river over geo-logic time (Figure 2). A continuation of the properly designedlevee for others 100 m or nearly so would have diminished theinflow rate to the city by a factor of approximately ten (i.e.,about the ratio of incoming overland flow to incoming discharge

through the gap (Figure 4)). Today, the third stage of the Westlevee to close the protective ring for the city is under construction,(5) the State agencies in charge of West levee maintenance andrelated aspects showed a total lack of coordination. Whereas theState Highway Administration held responsibility for the Westlevee maintenance, and the State Water Resources Authoritieswere responsible for issuing flood warnings, no one was assignedthe responsibility of reinforcing the non-engineering levee dur-ing floods. This lack of coordination could be attributed in part tothe success of the non-engineered portion of the flood protectionlevee during the big flood of 1998 (Table 1), and (6) speciallydesigned devices or gates, to release excess water trapped insidethe city in case of a levee breach were not available. The Westlevee had to be blown up in four places and the city highway beltat two other points. Those sites had to be reconstructed afterwardsat a considerable cost.

Finally, the 20 million dollar new Children Hospital was sup-posed to be built in lands with zero risk of flooding with theassistance of a World Bank loan. A close up look of the suddenend of the West levee by a team of experts could have recognizedthe high likelihood of failure of the non-engineered levee in caseof extreme floods, and helped to put in the public eye the lack ofany contingency plan to reinforce the gap during the emergency,among other deficiencies.

Acknowledgments

This work was made possible thanks to the financial supportof the National Agency for Promoting Science and Technol-ogy (ANPCyT) through the research grant number PID74, andthe National Institute of Scientific and Technological Research(CONICET) through a PIP98-616 research grant. Their sup-port is gratefully acknowledged. The authors also want to thankDr. Laurel Lacher for her thoughtful comments that helped toimprove the first draft of the paper.

Notation

Qmc = Bank-full discharge of the main channelQfp = Discharge of the whole channel-floodplain

systemQups = Discharge imposed at the upstream boundary

h = Water depth in the streamu = (u, v) = Depth-average velocity vectorx = (x, y) = Horizontal direction vector

t = time∇ = (∂x, ∂y) = Gradient operator

η = Streambed elevation above datumg = Acceleration due to gravityρ = Fluid density

τ b = (τbx , τby ) = Shear stress vector acting on the streambedνt = Eddy viscosity

CF = Friction coefficientks = Effective roughness heightn = Manning roughness coefficient


κ = Von Kármán constantu∗ = Shear velocityα = Dimensionless eddy viscosity

zw = Water elevationN = Number of nodes of the finite

element meshNe = Number of finite elements

n = Normal unit vectorδs = Differential line element

ψ(x, t) = Stream functiony(u, v, δh) = Vector of unknowns of the

discretized shallow water modela(a1, . . . , ak) = Parametric space of the discretized

shallow water modelθ = Sensitivity coefficient of the

discretized shallow water modelq = Discharge entering the city through

the breachqmax = Maximum discharge entering the city

through the breach

References

1. Acuña, J.C. (2003). Emergency program for the recupera-tion of flood affected zones (in Spanish). Report submittedto Ministry of Public Works (MOSPyV), Santa Fe State,September 2003.

2. Arnel, N. (1996). Global Warming, River Flows and WaterResources, Wiley, Chichester.

3. Bates, P.D., Horritt, M.S., Smith, C.N. and Mason, D.(1997). “Integrating Remote Sensing Observations ofFlood Hydrology and Hydraulic Modelling,” HydrologicalProcesses, 11, 1777–1795.

4. Brooks, A.N. and Hughes, T.J.R. (1982). “Stream-line Upwind/Petrov-Galerkin Formulations for ConvectiveDominated Flows with Particular Emphasis on the Incom-pressible Navier-Stokes Equations,” Computer Methods inApplied Mechanics and Engineering, 32, 199–259.

5. Carrasco, A. and Vionnet, C.A. (2004). “Separation ofScales on a Broad, Shallow Turbulent Flow,” J. HydraulicResearch, 42(6), 630–638.

6. Castignani, M.I., Cursack, A.M., D’Angelo, C.,Tosolini, R. and Travadero, M. (2003). Soil and produc-tivity aspects; evolution in land used (in Spanish); in: Thelower basin of the Salado River, towards its water manage-ment. Report ProCIFE funded by the United Nations Pro-gram for Development (PNUD), Dec. 2003, Sec. ExtensiónUNL, Santa Fe, Argentina.

7. Casulli, V. (1990). “Semi-implicit Finite DifferenceMethods for Two-dimensional Shallow Water Equations,”J. Comput. Physics, 86, 56–74.

8. CEPAL (2003). Impact evaluation of the Salado Riverflood in the Santa Fe State, Republic of Argentina,2003 (in Spanish). United Nations – Economic Coun-cil for Latin American and the Caribbean (CEPAL),

Rep. LC/BUE/R.246, http://www.eclac.cl/publicaciones/Buenosaires/6/LCBUER246/infostafe1.pdf

9. Chapra, S.C. (1997). Surface Water Quality Modeling.McGraw Hill.

10. ENVI (2002). The Environment for Visualizing Images,v.4.0, Exploring ENVI, Training Course Manual, ResearchSystem Inc., Illinois, USA.

11. Ferreira, C.G. (2004). Hydrology of the Salado Riverlower basin during January–May 2003 (in Spanish). AnnexoI Report Dept. Water Res. (Memo. 40/2003), Ministry ofPublicWorks (MOSPyV), Santa Fe State.

12. Fischer, H.B., List, J.E., Koh, R.C.Y., Imberger J. andBrooks, N.H. (1979). Mixing in Inland and Coastal Waters.Academic Press, San Diego.

13. Gaddi, M., Sexton, T., Muste, M., Patel, V.C. andAlvarez, P.J.J. (2003). “International Perspectives in WaterResources Management: The Paraná River Watershed,”World Transactions on Engineering and Technology Edu-cation, 2, 435–439.

14. Garcia, N. and Vargas, W. (1998). “The Temporal Cli-matic Variability in the Río de la Plata basin displayed bythe River Discharges,” Climatic Change, 38, 359–379.

15. Gardiner, B. (2003). Report: Extreme Weather on the Rise.Associated Press, London, February 27, 2003.

16. Han, K-Y., Lee, J.-T. and Park, J.-H. (1998). “Flood Inun-dation Analysis Resulting from Levee-Break,” J. HydraulicResearch, 36, 747–759.

17. Haider, S., Paquier, A., Morel, R. and Champagne,J.-Y. (2002). Urban flood modelling using computationalfluid dynamics, Proc. Institution of Civil Engineering, Waterand Maritime Engs. pp. 1–8.

18. Hervouet, J.M and Van Haren, L. (1996). RecentAdvances in Numerical Methods for Fluid Flows. InFloodplain Processes, M.G. Anderson, D.E. Wallingand P.D. Bates (eds). JohnWiley and Sons: Chichester,pp. 183–214.

19. Hervouet, J.M. and Petitjean, A. (1999). “MalpassetDam-Break Revisited with Two-Dimensional Computa-tion,” J. Hydraulic Research 37, 777–788.

20. Hervouet, J.M. (2000). “A High Resolution 2-D Dam-Break Model using Parallelization,” Hydrological Pro-cesses, 14, 2211–2230.

21. Hervouet, J.M., Samie, R. and Moreau, B. (2000).Modelling urban areas in dam-break flood-wave numericalsimulations. Proceedings RESCDAM Workshop, Seinajoki,Finland, Finnish Environmental Institute.

22. Horrit, M.S. and Bates, P.D. (2001). “Predicting Flood-plain Inundation: Raster-based Modelling versus the Finite-Element Approach,” Hydrological Processes, 15, 825–842.

23. Hunt, B. (1988). “Water Waves Generated by Distant Land-slides,” J. Hydraulic Research, 26,307–322.

24. INA (1998). Hydraulic redesign of the Santa Fe-Rosariofreeway bridge (in Spanish). Report National Institute forthe Water and the Environment – INA, local division CRL,Santa Fe, Argentina.


25. INCOCIV (2003). Hydraulic study for flood-control struc-tures of Santa Fe city (in Spanish), Final Report fromConsulting Co. on Civil Engineering – INCOCIV, Paraná,State of Entre Ríos, Argentina.

26. Kelly, D.W., Nakazawa, S., Zienkiewics, O.C.and Heinrich, J.C. (1980). “A Note on Upwindingand Anisotropic Balancing Dissipation in Finite ElementApproximations to Convective Diffusion Problems,” Int. J.Num. Meth. Eng., 15, 1705–1711.

27. Krohling, D.M. and Iriondo, M. (1999). “Upper Quater-nary Palaeoclimates of the Mar Chiquita area, North Pampa,Argentina,” Quaternary Int., 57/58, 149–163.

28. Lewin, J., Macklin, M.G. and Newson, M.D. (1988).Regime theory and environmental change – irreconcil-able concepts? International Conference on River Regime,W.R. White, (ed.) Wiley, Chichester. pp. 431–445.

29. Lin, B. and Falconer, R.A. (2001). “Numerical Modellingof 3D Tidal Currents and Water Quality Indicators in theBristol Channel,” Proc. Institution of Civil Engineering,Water & Maritime Engs., 48, 155–166.

30. López, J.L., García-Martínez, R. and Pérez-

Hernández, D. (2000). The torrential landslides of Decem-ber 1999 in Venezuela (in Spanish), Proceedings XIXLatin-American Hydraulic Conf., Córdoba, Argentina,pp. 417–426.

31. MOSP (1994). Project for the Highway belt in west Santa FeCity, Reach: Interstate Road 11 – Blas Parera Ave., Section:Freeway AP 01-Blas Parera Ave. (in Spanish), State High-wayAdministration, Ministry of Works, Public Services andHousing – MOSP, State of Santa Fe, 2nd Volume, 1994.

32. Nezu, I. and Nakawaga, H. (1993). Turbulence in open-channel flows. IAHR Monograph, Balkema. Rotterdam.

33. SMS (2000). SurfaceWater Modelling System, User’s Man-ual. Brigham Young University, Environmental ModelingResearch Laboratory, Utah, USA.

34. Stoker, J.J. (1958). WaterWaves. The Mathematical Theorywith Applications. JohnWiley and Sons: New York.

35. Tassi, P.A., Vionnet, C.A. and Rodríguez, L. (2000).Integrating technologies for 2D numerical modelling of theParaná River, in Proc. Hydroinformatics 2000 Conf., CedarRapids, Iowa, USA, 23–27 July 2000.

36. Tecplot (2004), Tecplot User’s Manual v.10.0, Amtec Engi-neering Inc., Bellevue, Washington, USA.

37. Telemac Modelling System (1998). 2D HydrodynamicsTELEMAC-2D Software v.4.0. User’s Manual, Electricitéde France, Départment Laboratorie National d’Hydraulic.

38. Toro, E. (2001). Shock-Capturing Methods for Free-Surface Flows, Wiley, Chichester.

39. Uijttewaal, W.S.J. and Jirka, G.H. (2003). “Grid Turbu-lence in Shallow Flows,” J. Fluid Mech. 489, 325–344.

40. Vionnet, C.A. and Garcia, M.H. (2003). “CatastrophicFlooding in Santa Fe, Argentina, IARH Newsletter, 20,”Supplement to J. Hydraulic Research, 41(4).

41. Vionnet, C.A., Tassi, P.A. and Martin-Vide, J.P. (2004).“Estimates of Flow Resistance and Eddy Viscosity Coeffi-cients for 2D Modelling on Vegetated Floodplains,” Hydro-logical Processes, 18, 2907–2926.

42. Vreugdenhil, C.B. (1994). Numerical Methods for Shal-low Water Flow. 2nd edition. Kluwer Academic Publishers:Dordrecht, Netherlands.

43. Wahl, T. (1997). Predicting embankment dam breachparameters – a needs assessment, in Proc. XXVII IAHRConf., San Francisco, CA, USA, Aug. 10–15, 1997.

Numerical modelling of the catastrophic flooding of Santa Fe City, Argentina

Documents

Transcript of Numerical modelling of the catastrophic flooding of Santa Fe City, Argentina