at SciVerse ScienceDirect

Ocean & Coastal Management 69 (2012) 231e242

Contents lists available

Ocean & Coastal Management

journal homepage: www.elsevier .com/locate/ocecoaman

A methodology for the classification of estuary restoration areas:A management tool

Mirian Jiménez a,*, Sonia Castanedo a, Raúl Medina a, Paula Camus a,b

a Environmental Hydraulics Institute IH Cantabria, Universidad de Cantabria, C/Isabel Torres n�15, Parque Científico y Tecnológico de Cantabria, 39011 Santander, SpainbClimate Research Division, Atmospheric Science and Technology Directorate, Science and Technology Branch, Environment Canada, 4905 Dufferin Street, Toronto, Canada

a r t i c l e i n f o

Article history:Available online 7 September 2012

* Corresponding author. Tel.: þ34 942 201616x1347E-mail addresses: mirian.jimenez@unican.

(M. Jiménez).

0964-5691/$ e see front matter � 2012 Elsevier Ltd.http://dx.doi.org/10.1016/j.ocecoaman.2012.08.017

a b s t r a c t

Planning the recovery of estuarine areas represents a major challenge for environmental managers, whomust find a balance between the desired environmental restoration, understood as the return to naturalconditions, and the different socioeconomic uses currently borne by the estuaries. This work presentsa methodology to optimize decision-making in accordance with the objectives which might arise inprojects for the hydrodynamic restoration of estuaries. Socioeconomic issues are not considered in thisstudy. The new approach is based on a classification of the zones to be restored according to charac-teristics representing their hydrodynamic performance and the possible morphodynamic effects of therestoration on the rest of the estuary. To achieve this, the four following parameters were chosen: (1)changes in tidal prism induced by restoration of that zone (DU), (2) the distance between the concessionand the estuary inlet (L), (3) the tidal wave phase lag (4) and (4) the flood potential of the restorationarea (I). The classification combines self-organizing maps (SOM) and the K-means algorithm. Themethodology was applied in a total of 139 areas (concessions) on ten estuaries along the entire coast ofCantabria (Northern Spain) where a Spanish Ministry of the Environment Recuperation Plan is underway. The results classify the 139 areas of restoration into five clusters. Empirical relationships were usedto estimate the effects the restoration of each cluster may have on the estuary’s various morphodynamicelements (cross-sectional area of the estuary mouth, area of tidal flats, volume of tidal channels andvolume of the ebb tidal delta), giving managers an overall view of the potential effects of the restorationin each zone and providing a basis on which to plan these actions.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Estuaries present relatively important gradients in their phys-ical, morphological, chemical and biological characteristics, makingthem systems of great complexity and variability, providinga habitat for a large diversity of flora and fauna species (Barne et al.,1995; Struyf et al., 2009). Moreover, estuaries act as a barrier toerosion and flooding on the coastal front, dissipating swell and tideenergy (Möller and Spencer, 2002) with great capacity to abateflooding and, on the other hand, supporting many socioeconomicservices and activities (fishing, accommodation, recreation, etc.).

However, in spite of these functions, in Spain many estuarieswere negatively impacted by projects officially aimed at improvingpublic health but whose real purpose was the drying and filling ofthese areas (Spanish Coasts Act, Act No. 22/1988). These actions led

; fax: þ34 942 266361.es, jimeneztm@unican.es

All rights reserved.

to a drastic reduction of intertidal zones in the nineteenth andtwentieth centuries when they were subjected to drying anddrainage policies, while the granting of concession titles stimulatetheir conversion toward agricultural, livestock or urban uses (in thiscontext we use the term “concession” to describe an area of anestuary whose exploitation rights are granted, for a prescribedperiod of time, to a public or private entity in order to carry outdifferent activities).

Subsequently, in the second half of the twentieth century,increased understanding and awareness of the importance of thesezones led to the development of diverse protection policies, forexample the international Ramsar Convention (1975) and EuropeanDirectives like the Habitats Directive 92/43/CEE and the WaterFramework Directive 2000/60/CE. The last of these requires allEuropean countries to attain good conditions in the quality of allwater masses, including transition waters (i.e. estuaries).

Currently in Spain, wetland restoration is fomented in thecurrent Coasts Act, Act No. 22/1988, according to which allconcession titles expire 30 years after the Act (2018). This repre-sents a challenge for managers who must plan the restoration of

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242232

these zones, which sometimes account for a high percentage of thetotal estuary area.

Restoration is understood as the recovery for the estuaries offormer intertidal zones currently used for other purposes, and toreturn them as far as possible to their natural state (Mitsch andGosselink, 2000). In principle, the natural state is the name givento the situation in place prior to any human intervention. However,the restoration of these areas often comes into conflict with well-established uses. It may even be that the artificial state ofa wetland has endowed it with characteristics which, for variousreasons, may be of interest to conserve. Thus in planning therestoration of an estuary, not just hydrodynamic and biologicalconsiderations come into play but also those of a legal and socio-economic nature.

A review of the literature shows that restoration techniquesfocus mainly on recovering the estuary long-term tidal leveldistribution by reestablishing its tidal flow (Pethick, 2002; Coxet al., 2006; Jacobs et al., 2009; Yang et al., 2010b). In other cases,authors do not just analyze the hydrodynamic restoration but seekto ecologically restore the area (Mitsch and Day, 2006; Milano,1999). Diefenderfer et al. (2005) and Yang et al. (2010a) evaluatedthe hydrodynamic response and the cumulative effects of variousrestoration projects on specific estuaries. Finally, there are someguidelines for hydrodynamic and biological restorations of inter-tidal zones but they focus mainly either on restoring a specific siteor on technical issues (e.g. Niedowski, 2000; Leggett et al., 2004).However, as far as we know, there has been no study analyzing theplanning of a large-scale (regional) restoration to providemanagerswith a methodology to assist in the decision-making process.Providing such a general methodology is the objective of this work.

With this in mind, this study offers a methodology for theclassification of the areas to be recovered in an estuary according tothe characteristics representing their hydrodynamic behavior andthe possible morphodynamic effects of the restoration on the restof the estuary. Legal and socioeconomic constraints are assumed tobe fixed for the purposes of this study. Thus, depending on theobjective sought, it will be possible to prioritize restoration in theconcessions.

This methodology is applied at the regional level, specifically inCantabria (Northern Spain) where managers must plan the resto-ration of 139 concessions on ten estuaries along over 200 km ofcoast. The outcome of this study is to provide a tool which, togetherwith other aspects such as those of a biological or legal nature, will

Fig. 1. Location of the ten estuaries stu

facilitate decision-making in the process of planning this type ofaction.

This work is novel in its use of advanced clustering techniques(Self-OrganizingMaps, (SOM) and the K-means algorithm) pointingto different types of areas for restoration depending on themagnitude of the changes the restoration will have on the estuary.

The remainder of this article is structured as follows: Section 2introduces the area of study; Section 3 describes the method-ology developed for the classification of concessions; Section 4shows the results of the application of the methodology in theCantabria estuaries, Section 5 includes a brief discussion andSection 6 presents the study’s main findings.

2. Study area

Cantabria is a region located in northern Spain (see Fig. 1),a coast mainlymade up of cliffs, interrupted locally by rivermouths,forming estuaries of very diverse sizes (75e2346 ha). Regarding thewave climate, NW sea states are the most frequent (51%); NNW seastates (25%) are also important. 50th and 80th percentiles ofsignificant wave height are 1.5 and 2.5 m, respectively. Tide issemidiurnal with a mean and spring tide range of 3 m and 5 mrespectively.

The estuaries on this coast are of similar hydrological charac-teristics, meaning that they are of a single type characterized bylarge intertidal surfaces and dominated by the tidal dynamic,making them well-mixed estuaries. The rivers flowing north arerelatively short, the longest one being the Deva (65 km e TinaMayor estuary), characterized by relatively pronounced slopes inthe area of the waterhead, which are less marked in the middle andlower sections. The flow contributed by the rivers, Qr, is negligiblecompared with that from tides. Galván et al. (2010) analyze therelation between the volume of river water entering the estuaryduring a half-tide cycle (Qr6) and the tidal prism,U, concluding thatin all these estuaries the tidal dynamic dominates over the fluvialdynamic ((Qr6/U)< 0.2), taking values between (0e0.12), except onthe Tina Mayor estuary which is governed by the fluvial dynamics,where ((Qr6/U) > 0.2), reaching a value of 0.36 (see Table 1).

All these estuaries, and in particular the Oyambre, SantanderBay, and San Martín de la Arena estuaries, have been greatlypressured by numerous human activities in their environments(Galván et al., 2010). A feature of the Oyambre estuary is thepresence of artificial structures which sharply restrict tidal flow,

died (Cantabria, Northern Spain).

Table 1Characteristics of the ten estuaries studied: Area (km2); Perimeter (km); Tidal prism (hm3); Number of concessions on each estuary (Nc); The area of those concessions (km2);The percentage of area restored (%) and the relation between the volume of river water entering the estuary during a half-tide cycle (Qr6) and the tidal prism (u). (*) In Fig. 2a map for each estuary is shown.

Estuary Area, A (km2) Perimeter,P (km)

Tidal prism,U (hm3)

Number ofconcessions, Nc

Total area to berestored, Ar (km2)

Ar (%) Qr6/U

Santander Bay*2b 21.67 85.25 68.19 39 5.46 25.19 0.003Santoña*2a 18.68 76.79 52.38 55 9.15 48.98 0.007San Vicente de la Barquera*2c 4.33 27.28 12.27 6 1.42 32.79 0.004San Martín de la Arena*2g 3.39 30.27 8.5 3 0.15 4.42 0.065Mogro*2i 2.23 26.78 4.2 8 0.93 41.7 0.123Tina Menor*2j 1.35 13.44 2.97 1 0.26 19.25 0.021Ajo*2e 1.28 18.78 2.14 6 0.15 11.71 0.04Tina Mayor*2f 1.17 13.44 1.43 4 0.26 22.22 0.36Oyambre*2d 1.01 13.64 0.8 5 0.45 44.55 0Oriñón *2h 0.57 9.35 0.75 11 0.55 96.49 0.124

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242 233

affecting approximately 50% of the estuary surface. Similarly,Santander Bay has been greatly modified by the urban develop-ment and port activity pursued there.

As a summary, Table 1 presents the main characteristics of eachof the estuaries studied and the number of concessions on eachone, while Fig. 2 shows the location of these areas.

As seen in Fig. 2a, Santoña is the estuary with the largestnumber of concessions (55) and the greatest area for restoration(9.15 km2). At the other extreme is the Tina Menor estuary (Fig. 2j)with just one concession. As Fig. 2 shows, the morphology of theseconcessions varies, and they range from 0.0021 ha to 91 ha in area.

3. Methodology

For a classification which evaluates restoration from a hydrody-namic and morphodynamic standpoint, a set of variables has beenidentified allowing us to describe the hydro-morphodynamicperformance of each concession. The concessions were thengrouped according to the variables selected, combining two tech-niques: (1) Self-Organizing Maps (SOM) (Kohonen, 2000), a tech-nique included in neural networks (ANN’s, Artificial NeuralNetworks), and (2) the K-means algorithm (Hastie et al., 2001).

The following is a description of the variables selected and themethodology used for the classification.

3.1. Selection of variables

According to a large number of authors, the fundamental aim ofa restoration project is to reestablish the tidal flow (Mitsch andGosselink, 2000; Weishar et al., 2005). The effect of the restora-tion of each concession on the morphodynamics of an estuary isrelated to the degree to which this tidal flow is reestablished. Manyauthors have investigated the relation between the tidal prism, U(the volume of water which enters and exits an estuary in one tidalcycle) and the estuary’s morphodynamic elements. O’Brien (1931),Jarret (1976), Van de Kreeke and Haring (1980), Eysink (1990),Gerritsen et al. (1990), Watanabe et al. (1991), Hume andHerdendorf (1993), Kraus (1998), Hughes (2002), Gerritsen et al.(2003), Powell et al. (2006), and Stive and Rakhorst (2008) estab-lish relations of equilibrium between the tidal prism and the cross-sectional area of the estuary mouth. On the other hand, Renger(1976), Eysink (1990) and Eysink and Biegel (1992) develop rela-tions between the volume of the tidal channels and the tidal prism.Walton and Adams (1976) and Marino and Mehta (1987) establishempirical relations with the volume of the ebb tidal delta. Otherauthors have investigated the relation between different morpho-logical elements, e.g. Renger and Partenscky (1974), proposing

a relation between the tidal flat area and the area of the bay, whileEysink (1991) relates the tidal flats area and their volume.

After this review, it is clear that any action involving a change inthe tidal prism will cause changes in the various morphologicalelements of an estuary. This fact led us to adopt the change in tidalprism,DU, as one of the decisive variables in classifying the effect ofa restoration.

Reestablishment of the tidal flow as a consequence of a resto-ration project will mean that a greater volume of water will enterand exit the estuary. This exchange of water, nutrients and sedi-ment flow takes place along an estuary’s tidal channels which, asa consequence of the restoration, will adjust their morphometricsto the new conditions. Thus, restoration of concessions furthestfrom the mouth area will affect a larger extension of each channeland a larger area of the estuary will then be affected by the amountof sedimentary material put in movement. In other words, thefurthest from the mouth a change is made the longer distance ofthe estuary may be affected by the change. However, this fact doesnot imply that the affection will be more or less important; it onlyrefers to the extent of the estuary affected. The variable selectedhere to describe this aspect is the distance between the concessionand the estuary mouth, L.

Changes to the hydrodynamics and morphodynamics of anestuary resulting from interventions do not depend only ondistance L, but also on the channel configuration (Rinaldo et al.,1999; French and Stoddar, 1992) which will influence the way thetidal wave propagates through the estuary. The variable whichprovides information on this aspect is the tidal phase lag, 4, definedas the time between the instant of the high tide outside the estuaryand the instant of the high tide at a point inside the estuary. Thisparameter is responsible for the asymmetry of current velocities,which has major implications for the sediment transport (Pingreeand Griffiths, 1979; Aubrey and Speer, 1985) and consequently inthe estuary’s evolutionary tendencies. Changes in the asymmetry ofcurrents may modify the tendency in sedimentation or erosionwhich will alter an estuary’s morphology locally and overall.

As well as focusing on the reestablishment of the originalhydrodynamics present in the area, restoration efforts aim toecologically rehabilitate it as well. Although this document does notset out to establish the criteria for ecological recuperation, it hasbeen thought that, as a consequence of the change in the floodablesurface which may be caused by recovery of the concessions, theflora and fauna in these intertidal zones will be affected, impactingspecies distribution. The long-term distribution of flooding level isclosely related to hydrological and biological processes and deter-mines the distribution of vegetation and fauna (Mitsch andGosselink, 2000; Roman et al., 1994; Montalvo and Steenhuis,2002; Todd et al., 2010) and is linked to sedimentation processes

Fig. 2. Location of the concessions on the estuaries. (a) Santoña, (b) Santander Bay, (c) San Vicente de la Barquera, (d) Oyambre, (e) Ajo, (f) Tina Mayor, (g) San Martín de la Arena, (h)Oriñón, (i) Mogro and (j) Tina Menor. The characteristics of each estuary are shown in Table 1.

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242234

(Anisfield and Benoit, 1999; Pasternack and Brush, 2001). To takethis aspect into account, flood potential, I, was considered, definedas the change in the number of hours per annum during which eachconcession is flooded, another of the variables selected in this work.

Consequently, the four variables selected for this study to createa general classification of concessions to be restored are: changes intidal prism, DU (m3), distance to the estuary’s mouth, L (m), tidalwave phase lag, 4 (hours) and flood potential, I (h/year). For eachconcession, L was measured on a 20 m resolution DTM (digitalterrain model) using a Geographical Information System (ArcGIS9.2 by ESRI) and DU, 4 and I were calculated using a two-dimensional hydrodynamic model, H2D, which resolves the well-known vertical-averaged SWEs. This model has been employed

and calibrated in numerous estuaries in Northern Spain (Bárcenaet al., 2011). The model provides the free surface level and thevertically averaged currents resulting from the propagation of thetidal wave into the estuary.

3.2. Classification of the estuarine zones to be restored

3.2.1. Statistical analysisPrior to the classification, the variables selected (DU, L, 4 and I)

must be analyzed to ensure that they fulfill the requirement ofindependence. This was done using Spearman’s non-parametricrank-correlation method. Spearman’s correlation coefficient, r,can have values between�1.00 and 1.00. At 0 it indicates that there

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242 235

is no correlation between the two variables. A positive coefficientindicates that the value of variable A increases with that of variableB, and the value þ1.00 is identified as a perfectly rising linearrelation. If the coefficient is negative, variables A and B vary inopposite directions, and value �1 is identified as a perfect inverserelation.

3.2.2. Application of SOMs and K-meansSOMs (self-organizing maps) constitute a technique which arose

in the field of neuronal computation used to work in high-dimension spaces and to project them into a 2D-dimensionalspace (Kohonen, 2000). It is a classification method which detectspatterns or classes in a set of data, preserving the neighboringrelations. This means that similar clusters in the multidimensionalspace are located together in the projection lattice. Data projectionon a 2D grid allows the data to be visualized intuitively. In thisstudy, the map is in the form of a rectangular 2D grid with l by mneurons laid out on a hexagonal lattice (C ¼ l � m neurons in theoutput layer).

The SOM is “trained” using an iterative learning algorithm. Thisprocess includes a self-organizing neighborhood mechanism, soneighboring clusters of the winning reference vector in the 2Dlattice space are also adapted toward the sample vector, thus pro-jecting the topological neighborhood relationships of the high-dimensional data space onto the lattice. The starting point of thistechnique is a data samplewhere N is the total number of data to beclassified.

Each neuron of the output layer Ck, is associated with twovectors. The first one has a lattice-position vector Ck ¼ (lk, mk)associated to it, which describes the position of the cluster on thelattice. On the other hand a reference vector vk ¼ (vk1,.,vk2)represents the position of the cluster centroid in the data space. Thegoal of the algorithm is to minimize an overall within e clusterdistance d(Ck) from the data vectors xi within the cluster to corre-sponding reference vector vk, for each cluster Ck.

Xk¼1;:::m

dðCkÞ ¼X

k¼1;:::;m

Xxi˛Ck

kxi � vkk2 (1)

The least distant output neuron is declared the “winner”. Theprocess of training of this classification algorithm includes a spatialneighborhood nucleus in the projection grid which means not onlythat the winning centroid is displaced toward the input vector, butthat the neighboring centroids in the 2D grid are also modified. Thislearning process is iterative and ongoing, and ends when a quasi-stable state is reached.

This technique has been used recently in many different disci-plines, one of them ecology where Giraudel and Lek (2001) usedthe SOM to observe the distribution of the abundance of treespecies in southern Wisconsin (USA). Meanwhile, Gevrey et al.(2006) used the procedure to measure the risk of invasion ofcertain insect species in specific geographical areas. Another field inwhich this technique has been used is meteorology, whereHewitson and Crane (2002) and Gutiérrez et al. (2005) used SOMsto detect patterns of atmospheric circulation and relate them toprecipitation series. In biology, Ainsworth and Jones (1999) usedthe SOM to classify data on the concentration of chlorophyllthroughout the Pacific Ocean from satellite temperature and colordata. More recent work has been done in the field of oceanographywhere, since the introduction and demonstration of the use ofSOMs by Richardson et al. (2003), the technique has beenincreasingly employed (Risien et al., 2004; Liu and Weisberg, 2005,2007; Méndez et al., 2009; Camus et al., 2011). The procedure hasalso been used in other disciplines such as hydrology (Kwang-Seuket al., 2010).

It must be emphasized that the SOM begins with a given shape(topology) and dimension: it may be rectangular or hexagonal inshape (with 4 and 6 neighbors respectively). In our work, todetermine the dimension of the grid, various tests were carried out,to find that the adequate dimension for the number of availabledata (556: 139 concessions and four variables for each) was 6 � 6.

To establish the relative importance of each variable, Xi, theywere adimensionalized as follows:

Xi ¼xcxe

(2)

Where:

Xi ¼ {DU, L, 4, I} ¼ variable i adimensionalized.xc ¼ {DUc, Lc, 4c, Ic} ¼ value of the variables for each concession,c.xe ¼ {DUe, Le, 4e, Ie} ¼ value of the variables for the estuary, e, Ue

(tidal prism in the estuary), Le (total length estuary), 4e

(maximum time lag in the estuary), Ie (maximum flood potentialin the estuary) where the concession, c, is located.

These variables were then normalized at an interval of [0, 1] (bXi)by linear transformation, the minimum andmaximumvalues beingnecessary for each variable in each estuary.

bXi ¼Xi � Xmin

Xmax � Xmin(3)

Where

Xi ¼ variable i adimensionalized.Xmin ¼ minimum value of the variable adimensionalized.Xmax ¼ maximum value of the variable adimensionalized.

As will be shown in the results, the number of groups obtainedwith the application of the SOMs, 36, is high for the creation ofa simple, manageable classification. Thus the K-means algorithmwas applied to these groups. The classification procedure startswith random initialization of the centroids. On each interaction, thedata nearest to each centroid are identified and the centroid is thenredefined as the mean of the corresponding data. The algorithm isiteratively moved until the intragroup distance is minimal and theprocess converges (Hastie et al., 2001).

Finally, N groups were obtained, each represented by a centroidwith one value of the four variables considered wherefDbUi;

bLi; b4i;bI ig where i ¼ 1,., N.

Many authors in various fields have combined these techniques(SOM and K-means) to obtain a classification. For example, Solidoroet al. (2007) classified water quality in the Northern Adriatic Seabased on multiple biochemical variables and Gevrey et al. (2006)measured the risk of plagues of various insect species in differentparts of the planet.

3.3. Effects of estuarine restoration

Once the data have been grouped together, the decision torecover one group or another depends on the objective set by therestoration project manager. As alreadymentioned, because each ofthe groups has particular characteristics, each one will havedifferent morphodynamic effects on the estuary. Predicting longterm morphological changes is still an unsolved matter. Existingmethods use either process-based models (e.g. Lesser et al., 2004;Marciano et al., 2005) or aggregated models that make use ofempirical regime theory to define the morphological equilibrium

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242236

state (Kragtwijk et al., 2004; Rossington et al., 2011). Some recentpublications compare the performance of process-based modelsand the empirical relationships (Van der Wegen et al., 2010; Tranet al., 2012).

Several authors have demonstrated that inlet equilibrium iscontrolled mainly by tidal prism, durations of the flood and ebbphases of the tides, fresh water discharge and sediment transport(e.g. Gao and Collins, 1994; Lanzoni and Seminara, 2002). In thisstudy, because of the time and spatial scales of interest, the long-term morphological changes induced by the restoration of eachgroup were estimated using the empirical or equilibrium rela-tionships available in the literature. These expressions allow us toestimate gross morphological changes caused in the differentelements of an estuary based on a change in its tidal prism. Thisapproach has to be taken as a first step in the assessment ofmorphological effects caused by hydrodynamic changes. Theinfluence of other factors such as extra sediments supplied by therivers should be considered in detailed studies for each estuarywhich are outside the scope of this research.

To evaluate the effects of the restoration project on the cross-sectional area of the estuary mouth, the empirical relation devel-oped by O’Brien (1969) was used.

Ac present ¼ Be$10�2*Ue (4)

Ac future ¼ Be$10�2*ðUe þ DUiÞ (5)

Where:

Be ¼ proportionality constant obtained in the e estuary for thecurrent situation.Ac ¼ cross-sectional area of the estuary’s mouth (m2).Ue ¼ tidal prism during spring tides in the e estuary (m3).DUi ¼ tidal prism provided by the i group centroid.

Tidal channels are another of the morphological elementsaffected by restoration. In this study, this is analyzed by means ofthe empirical relation developed by Renger (1976).

VMLW present ¼ De*U0:1566e (6)

VMLW future ¼ De*ðUe þ DUiÞ0:1566 (7)

Where:

De ¼ proportionality constant obtained in the e estuary for thecurrent situation.VMLW ¼ tidal channel volume below low tide level (m3).

To determine the changes to the ebb tidal delta, we used therelation between ebb tidal delta volume and the tidal prism asproposed by Walton and Adams (1976).

Vpresent ¼ Ee*U1:23e (8)

Vfuture ¼ Ee*ðUe þ DUiÞ1:23 (9)

Where:

Ee ¼ proportionality constant obtained in each estuary for thecurrent estimate.V ¼ ebb tidal delta volume (m3).

Note that coefficients Be, De and Ee in Eqs. (4)e(9) were obtainedassuming that the estuaries are inmorphodynamic equilibrium and

that the different restoration activities will not change the equi-librium parameters of the estuary.

It is worth mentioning that changes were assessed per conces-sion. So, as the problem is non linear, a similar analysis should becarried out when a concession is restored (Yang et al., 2010a).

4. Results

The methodology described above was applied to a total of tenestuaries with 139 concessions for restoration. For each of theconcessions, the changes in tidal prism produced by their restora-tion, DUc, was calculated, along with the distance between therestoration area and the estuary mouth, Lc, the tidal wave phase lag,4c, and flood potential, Ic. Spearman’s method was then used toexamine the dependence among the variables, the result of whichis shown in Fig. 3.

In all cases, the correlation coefficient, r, was found to be lessthan 0.3, except between the tidal prism, DU, and flood potential, I,where r is 0.7, pointing to a clear correlation between these twovariables. However, I was included in the classification because, aswill be seen in the results below, the consideration of this variableallows establishing a convenient differentiation among the groupscreated.

With the value of the four variables characterizing eachconcession, the data set was used to train the SOM and wassubsequently projected onto a two-dimensional (2D) map.Different grid dimensions were tested and a 6 � 6 SOM waseventually selected as the most appropriate given the number ofdata studied.

The input layer is composed of a total of 139 neurons and theoutput layer of 36 neurons organized in a 6� 6matrix, arranged ona hexagonal grid (see Fig. 4), where each data vector is assigned toa particular centroid (neuron). Similar centroids are located adja-cently on the projection plane, so that the magnitudes of thevariables defining the centroids vary gradually from one cell to theadjoining one. Fig. 5 shows the distribution of each of the variablesDbUi (Fig. 5a), bLi (Fig. 5b), b4i (Fig. 5c) and bI i (Fig. 5d). The highestvalues are shown in red and the lowest by the range of blues.

Centroids with maximum values for variable DbUi, rangingbetween 0.51 and 0.66, are those located in centroid numbers 31,32, 25 and 26. This distribution is similar to that of bI i reachingvalues between 0.73 and 0.78 in the centroids (25, 26, 31, 21 and33). In relation to the distance to the mouth, the concessionsassociated with centroids 3, 4 and 5 present values for this variablebetween 0.77 and 0.8, i.e. matching those furthest from the mouth.And centroids 1, 2, 7 and 13 have the highest values registered fortidal wave phase lag, of between 0.59 and 0.63.

Fig. 5e shows the frequency of presentation or probability ofeach centroid. In this specific case, cells 5 and 18 are those withmore concessions (greater probability) while for cells 8, 15, 20, 27and 28, the presentation frequency value is zero (blank hexagon),meaning that none of the concessions is represented by the char-acteristics of these centroids.

Fig. 6 shows all the information simultaneously, with each cellrepresenting a cluster defined by the four variables used in theclassification. In each cell, the largest hexagon shows the distri-bution of variable DbUi which appears in a scale of blues, the darkestfor maximumvalues reached by the centroids while the lightest arefor the minimum values. The smallest hexagon shows floodpotential distribution, this parameter appearing on a scale varyingbetween reds (associated with the highest values) and degradinguntil reaching yellow for the lowest values. The vectors containtriple information. The length of the arrow is proportional to thevalue of bLi. The vector’s angle to the horizontal axis shows infor-mation on the value of the tide phase lag, b4i, in each centroid.

Fig. 3. Correlation between the variables. a)Time lag vs. Length; b)Time lag vs. Flood potential; c)Time lag vs. Tidal prism; d) Flood potential vs. Length; e) Flood potential vs. Tidalprism; d) Length vs. Tidal prism.

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242 237

Finally, the color of the arrow indicates the frequency of presen-tation of that centroid, that is the number of concessions repre-sented in that cell (maximum values in black).

Each cell on the map includes the number of the concessions(see Fig. 2) associated with each centroid.

Fig. 4. The number in each cell refers to the output neuron number.

Note in Fig. 6 that the concessions in which a restoration willcause a greater increase in the total estuary tidal prism (DbUibetween 0.55 and 0.66) are located in centroids 31 and 32,matching those with high flood potential (bI i of 0.76 and 0.78respectively). However, as shown by the length of the vectorlocated in those centroids, these are concessions where bLi is small,i.e. they are close to the estuary mouths. Moreover, the tidal wavephase lag, b4i, in the concessions in centroid 32 is shorter in theirimmediate surroundings than for those in centroid 31. Theconcessions in these two centroids are located in seven of the tenestuaries considered. There are two concessions on the Santoñaestuary (numbers 1 and 2), four in Santander Bay (56, 57, 58 and59), one in San Vicente de la Barquera (number 97), one on theOyambre estuary (102), number 107 on the Ajo estuary, number 116in Oriñón and two concessions on Mogro estuary (132 and 133).

On the other hand, concessions located further from the estuarymouth (greater bLi), belong to centroids 3, 4 and 5, reaching bLi valuesbetween 0.77 and 0.80. These centroids offer minimum values forDbUi (0.07 and 0.16) and for flood potential, bI i (0.06 and 0.12). Theconcessions are located in these three centroids on six of the tenestuaries evaluated. Santander estuary is the one with mostconcessions in these centroids, with a total of 13 (76, 79, 81, 82, 83,84, 85, 86, 87, 88, 89, 90 and 91), all small concessions (<3 ha),located well inside the estuary. This pattern is also found in theconcessions on Santoña estuary (44 and 50), Oyambre (103 and104), San Vicente de la Barquera (100), Ajo (111 and 112) and Oriñón(120, 121, 122 and 123), included in these centroids.

Concessions with a greater phase lag, b4i, are found in centroids1, 7 and 13 (values between 0.60 and 0.63). Although there is no

Fig. 5. 6 � 6 SOM. Each plan shows the distribution of one of the variables. a) DU distribution on the hexagonal grid, b) distribution of L, c) distribution of variable 4, d) distributionof variable I, e) frequency of presentation on the two-dimension grid.

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242238

variation for variable bLi in these three centroids (values between0.66 and 0.68) the remaining variables perform differently. Thevalue for DbUi in centroid 7 is greater than its neighbors. Centroidnumber 13 has a greater bI i (0.51) than centroids 1 and 7 (0.27 and0.33) for variable bLi. These concessions are in the estuaries ofSantoña (9, 19, 37, 40, 41, 42 and 43), Santander (61, 62, 63 and 70),San Martín de la Arena (126) and Mogro (131 and 138).

Fig. 6. 6 � 6 SOM, all variable

As can be seen in the results shown in Fig. 6, SOM techniquemakes the clustering of the data set simple so that it is possible toobserve very intuitively how the concessions are grouped accord-ing to their characteristics. However, for a more manageable andsimplified classification, the K-means technique was applied to thegroups obtained from the SOM. During the analysis of the results,and considering the number of variables, it was found that 5 was

s shown simultaneously.

Fig. 7. 6 � 6 SOM. Delimitation of the groups obtained using the K-means algorithm.

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242 239

the most adequate number of groups (one group for each variableanalyzed, plus one permitting the transition between two verydifferent groups). The results obtained with application of the K-means algorithm are shown in Figs. 7 and 8.

Fig. 7 shows the limits of the groups obtained with the K-meanstechnique, and which concessions belong to each of the groups.Fig. 8 shows the characteristics which are representative of the fivegroups, each represented by a single centroid with a characteristicvalue for each of the four variables analyzed.

Fig. 8. 6 � 6 SOM. Characteristics of each centroid associated with the groups obtainedusing K-means.

As Fig. 8 shows, the group with the highest value for increasedrelative prism (0.49) is group 1 (red). Restoration of the concessionsin this group will cause a greater increase in the tidal prism andconsequently will produce the greatest changes in the morphody-namics of the estuaries they belong in. The maximum floodpotential value (bI i ¼ 0.72) is also found in this group.

Group 4 (blue) takes in the concessions furthest from theestuary mouth (bLi ¼ 0.68). Should these concessions be restored,the changes will affect a larger area of the estuary. Water flow willmodify existing channels and possibly create new ones. On theother hand, these concessions will not produce a significantincrease in the tidal prism (DbUi ¼ 0.03).

The centroid of Group 3 (orange) represents the concessionslocated in zones where the phase lag of the tidal wave is maximum.Thus the restoration of these concessions may involve a change inthe erosion-sedimentation pattern around them. In this group, thevalue for the relative distance variable (bLi ¼ 0.66) is also high. Theauthors found that the concessions with a greater phase lag coin-cide with those furthest from the mouth.

Group 2 (green) and Group 5 (yellow) are transition groupsbetween the groups referred to above.

To examine the relative importance of the variables within eachcluster, a ranking was established assigning each variable a valuebetween 1 and 5 depending on its value considering all the groups(see Table 2).

Table 2Ranking of each variable within the groups.

Group 1 Group 2 Group 3 Group 4 Group 5

DbUi 1 4 3 5 2bLi 3 4 2 1 5b4i 2 5 1 4 3bI i 1 3 4 5 2

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242240

Table 2 shows that Group 1 brings together the concessionswhich contribute more tidal prism to the estuary and with highflood potential, while in Group 4 both variables are in last place.Group 2 is the onewith the lowest positions in the ranking for threeof the four variables, showing that these concessions will causefewer morphodynamic changes in the estuary should the restora-tion take place.

Note how in Groups 2 and 3 the variables DbUi andbI i do not havethe same ranking within each group despite the correlation foundbetween the two variables. It is confirmed that the inclusion ofvariable bI i allows differentiating the effects of the different groupsmore clearly. For example, were parameter bI i not included, the

Fig. 9. Cluster distribution for every concession in every estuary. It indicates the most importla Barquera, (d) Oyambre, (e) Ajo, (f) Tina Mayor, (g) Suances, (h) Oriñón, (i) Mogro and (j) Tin4 and yellow to group 5. (For interpretation of the references to colour in this figure legen

effects of the recovery of groups 1 and 3 considered overall(combining the values of the ranking in each group except for bI i)would score the same. However, Group 1 will produce a largerfloodable zone (bI i ¼ 1) than Group 3 (bI i ¼ 4).

Fig. 9 shows the distribution of groups in each estuary.Concessionswith greaterDbUi are shown in red (Group 1) and these,on the San Vicente de la Barquera (see Fig. 9c), Oyambre (seeFig. 9d), Oriñón (see Fig. 9h) and Tina Menor (see Fig. 9i) estuaries,are the ones with the greatest surface area.

On the other hand, the authors have found that the Group 4(blue) concessions are mostly on the inner part of Santander Bay(see Fig. 9b), far from the estuary mouth and with low DbUi. In

ant areas for the restoration activities. (a) Santoña, (b) Santander Bay, (c) San Vicente dea Menor. Red color refers to group 1, green to group 2, orange to group 3, blue to groupd, the reader is referred to the web version of this article.)

Table 3Effects on the different elements of Santander Bay produced by recuperation of oneconcession in each group.

Group 1 Group 2 Group 3 Group 4 Group 5

V (%) þ0.35 þ0.07 þ0.11 þ0.03 þ0.23VMLW (%) þ0.42 þ0.09 þ0.14 þ0.04 þ0.28Ac (%) þ0.28 þ0.06 þ0.09 þ0.03 þ0.19

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242 241

general, the concessions are of different types throughout all theestuaries.

5. Discussion

Because of the current situation in Spain, with the expiration in2018 of numerous concessions located inside estuaries and grantedunder the current Coasts Act, new methodologies must be estab-lished to be applied to the management of the restoration of theseintertidal zones. Prior work has defined methodologies for thecomprehensive classification of estuaries according to ecological,biological or physical criteria, or to morphological parameters. Butthere has so far been no methodology which allows the planning ofestuary restorations on a regional scale.

The methodology proposed in this work is based on classifyingrestorations according to the morphological effects of their incor-poration into the estuary dynamics. As pointed out, the objective ofthis study was to propose the use of concession classification asa planning tool making it possible, depending on the aims ofa restoration, to decide in which group those aims will be best met.

To illustrate this proposal, an estimate was made of the mor-phodynamic changes which will be caused by recuperation in eachgroup obtained in previous classification, using the equilibriumrelationships available from the literature. These expressions allowus to estimate gross morphological changes caused in the differentelements of an estuary based on a change in the estuary’s tidalprism.

By way of example, the effects which would be caused by oneconcession in each group were assessed in the case of a restorationin the Santander Bay, Table 3 showing the results of this exercise.

It is seen how an increased DU due to a restoration in theconcessions will lead to an increase in tidal channel volume ofa maximum of 0.42% compared with the current situation (Group 1concession). Exchange of water between the ocean and the tidalinlet takes place through the mouth of the estuary whose crosssectional area, Ac, will be more or less increased as a result of therestoration. For restoration in a Group 1 concession, Ac would beincreased by 0.28%, a rise of 41m2 on its current dimensions. On theother hand, if the concession to be restored is in Group 2, theestuarine morphological elements would hardly be disturbed asthe changes caused by that concession would amount to less than0.1%.

This classification provides managers with an overall view of thevarious restoration options for each group and the potential effectsof restoration in one group or another on the various elements ofthe estuary, optimizing the decision making process as to whichconcessions to restore, depending on the aims of a restorationproject. The inclusion of legal and socioeconomic considerations inthis analysis should be explored in future works.

6. Conclusions

Currently in Spain, wetland restoration is encouraged in thecurrent Coasts Act, Act No. 22/1988, according to which allconcession titles expire 30 years after the Act (2018). This repre-sents a challenge for managers who must plan the restoration of

these zones, which sometimes account for a high percentage of thetotal estuary area.

The methodology proposed in this work is based on the classi-fication of restoration processes according to the morphologicaleffects produced on the estuary dynamics. The nature of the clas-sification (classifying not the estuaries but future restorationzones), the spatial scope of the study (10 estuaries), and the clas-sification techniques used are the main innovations found in thisstudy.

The use of clustering techniques such as self-organizing maps(SOM) and the K-means algorithm to carry out the classificationallows the original data set to be grouped into a reduced number ofclusters. SOMs make it possible, according to the four selectedvariables, to visualize how 139 restoration areas are grouped into36 clusters. To design a manageable tool, the number of clusterswas reduced using the K-means technique, classifying all theconcessions into a total of five groups, each group with hydro-morphological characteristics determining the effects on eachestuary element. An evaluation of these effects will providemanagers with an overall view of the various restoration options ofeach group, thereby enabling decision-making on the concessionsto be optimized.

In summary, the methodology developed in this study opens upa wide range of applications. Each concession can be evaluatedtaking account not only the hydrodynamic parameters as proposedin this study, but also others (biological, economic, legal, etc.)depending on the objectives. Thus we may conclude that thedevelopment of this methodology is a first step toward theimplementation of adequate estuary restoration management.

Acknowledgments

The support of the European Commission through FP7.2009-1,Contract 244104 e THESEUS ("Innovative technologies forsafer European coasts in a changing climate"), is gratefullyacknowledged.

References

Ainsworth, E.J., Jones, S.F., 1999. Radiance spectra classification from the ocean colorand temperature scanner on ADEOS. IEEE Transactions on Geoscience andRemote Sensing 37, 1645e1656.

Anisfield, S.C., Benoit, G., 1999. Sedimentation rates in flow restricted and restoredsalt marshes in Long Island. Estuaries 22, 231e244.

Aubrey, D.G., Speer, P.E., 1985. A study of non-linear tidal propagation in shallowinlet/estuaries systems. Estuarine, Coastal and Shelf Science 21, 185e205.

Barne, J.H., Robson, C.F., Kaznowska, S.S., Doody, J.D., Davison, N.C., 1995. Coast andSeas of the United Kingdom, Region 6: Flamborough Head to Great Yarmouth.Joint Nature Committee, Peterborough, UK.

Bárcena, J.F., García, A., Gómez, A., Álvarez, C., Juanes, J.A., Revilla, J.A., 2011. Spatialand temporal flushing time approach in estuaries influenced by river and tide.An application in Suances Estuary (Northern Spain). Estuarine, Coastal and ShelfScience. http://dx.doi.org/10.1016/j.ecss.2011.08.013.

Camus, P., Méndez, F.J., Medina, R., Cofiño, A.S., 2011. Analysis of clustering andselection algorithms for the study of multivariate wave climate. Coastal Engi-neering 58, 453e462.

Cox, T., Maris, T., Vleeschauwer, P.D., Mulder, T.D., Soetaert, K., Meire, P., 2006. Floodcontrol areas as an opportunity to restore estuarine habitat. Ecological Engi-neering 28, 55e63.

Diefenderfer, H.L., Roegner, C., Thom, R.M., Dawley, E.M., Whiting, A.H.,Johnson, G.E., Sobocinski, K.L., Anderson, M.G., Ebberts, B., 2005. EvaluatingCumulative Ecosystem Response to Restoration Projects in the Columbia RiverEstuary. Pacific Northwest National Laboratory, Richland, Washington.

Eysink, W.D., 1990. Morphological response of tidal basin to changes. In: Proceed-ings 22nd Coastal Engineering Conference. ASCE Delf, pp. 1948e1961.

Eysink, W.D., 1991. Simple morphologic relationships for estuaries and tidal chan-nels e handy tools for engineering. In: Proceedings of the InternationalConference on Coastal and Port Engineering in Developing Countries, II (Beijing,China), pp. 1003e1013.

Eysink, W.D., Biegel, E.J., 1992. ISOS*2 Project, Phase 4, Impact of Sea Level Rise onthe Morphology of the Wadden Sea in the Scope of its Ecological Function.Investigation on Empirical Morphological Relations. WLj Delft Hydraulics.

M. Jiménez et al. / Ocean & Coastal Management 69 (2012) 231e242242

French, J.R., Stoddar, D.R., 1992. Hydrodynamics of salt marsh creek systems:implications for marsh morphological development and material exchange.Earth Surface Processes Landforms 17, 235e252.

Galván, C., Juanes, J.A., Puente, A., 2010. Ecological classification of European tran-sitional waters in the North-East Atlantic eco-region. Estuarine, Coastal andShelf Science 87, 442e450.

Gao, S., Collins, M.B., 1994. Analysis of grain size trends, for defining sedimenttransport pathways in marine environments. Journal of Coastal Research 10,70e78.

Gerritsen, F., De Jong, H., Langerak, A., 1990. Cross-section stability of estuarychannels in the Netherlands. In: Proceedings 22nd Coastal EngineeringConference. ASCE Delf, pp. 2922e2935.

Gerritsen, F., Dunsbergen, D.W., Israel, C.G., 2003. A relational stability approach fortidal inlets, including analysis of the effects of wave action. Journal of CoastalResearch 19, 1066e1081.

Gevrey, M., Worner, S., Kasebov, N., Pitt, J., Girandel, J.L., 2006. Estimating risk ofevents using SOM models: a case of study on invasive specie establishment.Ecological Modelling 127, 361e372.

Giraudel, J.L., Lek, S., 2001. A comparison of self-organizing map algorithm andsome conventional statistical analysis of the effect of wave action. EcologicalModelling 146, 329e339.

Gutiérrez, J.M., Cano, R., Cofiño, A.S., Sordo, C., 2005. Analysis and downscalingmulti-model seasonal forecasts in Peru using self-organizing maps. Tellus,Series A: Dynamic Meteorology and Oceanography 57, 435e447.

Hastie, T., Tibshirani, R., Friedman, J., 2001. The Elements of Statistical Learning.Springer, New York.

Hewitson, B.C., Crane, R.G., 2002. Self-organizing maps: applications of synopticclimatology. Climate Research 22, 13e26.

Hughes, S.A., 2002. Equilibrium cross sectional area at tidal inlets. Journal of CoastalResearch 18, 160e174.

Hume, T.M., Herdendorf, C.E., 1993. On the use of empirical stability relationshipsfor characterising estuaries. Journal of Coastal Research 9, 413e422.

Jacobs, S., Beauchard, O., Struyf, E., Cox, T., Maris, T., Meire, P., 2009. Restoration oftidal freshwater vegetation using controlled reduced tide (CRT) along theSchelde Estuary (Belgium). Estuarine, Coastal and Shelf Science 85, 368e376.

Jarret, J.T., 1976. Tidal Prism-inlet Area Relationship, CERC-WES General Investiga-tion of Tidal Inlets, Dept. of the Army, US Corps of Engineers, 3, pp. 1e32.

Kohonen, T., 2000. Self-Organizing Maps, third ed. Springer-Verlag, Berlin.Kragtwijk, N.G., Zitman, T.J., Stive, J.F., Wang, Z.B., 2004. Morphological response of

tidal basins to human interventions. Coastal Engineering 51, 207e221.Kraus, N.C., 1998. Inlet cross-sectional area calculated by process-based model. In:

Proceedings 26th Conference on Coastal Engineering. ASCE 3, pp. 3265e3278.Kwang-Seuk, J., Dong-Gyun, H., Myeoung-Seop, B., Jon-chul, J., Han-Gook, K., Dong-

Kyun, K., Gea-Jae, J., 2010. Stream modifications patterns in a river basin: fieldsurvey and self-organizing maps (SOM) application. Ecological Informatics 5,293e303.

Lanzoni, S., Seminara, G., 2002. Long-term evolution and morphodynamic equi-librium of tidal channels. Journal of Geophysical Research 107 (C1). http://dx.doi.org/10.1029/2000JC000468.

Leggett, D.J., Cooper, N.J., Harvey, R., 2004. Coastal and Estuarine ManagedRealignment: Design Issues. CIRIA Publications. RP681.

Lesser, G.R., Roelvink, J.A., Van Kester, J.A.T.M., Stelling, G.S., 2004. Development andvalidation of a three-dimensional morphological model. Coastal Engineering 51,883e915.

Liu, Y., Weisberg, R., 2005. Patterns of ocean current variability on the West FloridaShelf using the self-organizing map. Journal of Geophysical Research. ISSN:0148-0227 110. ISSN: 0148-0227, C06003. http://dx.doi.org/10.1029/2004JC002786.

Liu, Y., Weisberg, R.H., Shay, L.K., 2007. Current patterns on the West Florida Shelffrom joint self-organizing map analyses of HF radar and ADCP data. Journal ofAtmospheric and Oceanic Technology. ISSN: 0739-0572 24 (4). ISSN: 0739-0572, 702e712.

Marciano, R., Wang, Z.B., Hibma, A., de Vriend, H.J., Defina, A., 2005. Modeling ofchannel patterns in short tidal basins. Journal of Geophysical Research, F01001.http://dx.doi.org/10.1029/2003JF000092.

Marino, J.H., Mehta, A.J., 1987. Inlet ebb shoals related to coastal parameters. In:Proceedings Coastal Sediments ’87, ASCE pp. 1608e1623.

Méndez, F.J., Camus, P., Medina, R., Cofiño, A.S., 2009. Analyzing the multidimen-sional wave climate with self-organizing maps. In: OCEANS’09 IEE Bremen:balancing Technology with Future Needs art. n� 5278285.

Milano, G.R., 1999. Restoration of coastal wetlands in southeastern Florida. WetlandJournal 11, 15e24.

Mitsch, C.F., Gosselink, J., 2000.Wetlands, third ed. JohnWiley & Sons, Inc, New York.Mitsch, W.J., Day, J.W., 2006. Restoration of wetlands in the MississippieOhioe

Missouri (MOM) River Basin: experience and needed research. EcologicalEngineering 26.

Möller, I., Spencer, T., 2002. Wave dissipation over macro-tidal saltmarshes: effectsof marsh edge typology and vegetation change. Journal of Coastal Research 36,506e521.

Montalvo, F., Steenhuis, S., 2002. The link between hydrology and restoration oftidal marshes in the New York/New Jersey estuary. Wetlands 24, 414e425.

Niedowski, N.L., 2000. New York State Salt Marsh Restoration and MonitoringGuideline. New York State Department of State, Division of Coastal Resourcesand New York State Department of Environmental Conservation.

O’Brien, M.P., 1931. Estuary tidal prism related to entrances areas. Civil Engineering1, 738e739.

O’Brien, M.P., 1969. Equilibrium flow areas of inlets on sandy coast. Journal of theWaterway and Harbour Division ASCE 95 (WW1), 43e45.

Pasternack, G.B., Brush, G.S., 2001. Seasonal variations in sedimentation and organiccontent in five plant associations on a Chesapeake Bay Tidal Freshwater Delta.Estuarine, Coastal and Shelf Science 53, 93e106.

Pethick, J., 2002. Estuarine and tidal wetland restoration in the United Kingdom:policy versus practice. Restoration Ecology 10, 431e437.

Pingree, R.D., Griffiths, D.K., 1979. Sand transport paths around the British Islesresulting from M2 and M4 tidal interactions. Journal of the Marine BiologicalAssociation of the U.K. 59, 497e513.

Powell, M.A., Thiecke, R.J., Metha, A.J., 2006. Morphodynamic relationships for ebband flood delta volumes at Florida’s entrances. Ocean Dynamics 56, 295e307.

Renger, E., 1976. Quantitative Analyse der Morphologie von Watteinzugsgebietenund Tidebecken. Mitteilungen des Franzius Institus der TU Hannover Heft 43,1e160.

Renger, E., Partenscky, H.M., 1974. Stability criteria for tidal basin. In: Proceedings14th International Coastal Engineering Conference. ASCE (Copenhagen,Denmark), pp. 1605e1618.

Richardson, A.J., Risien, C., Shillington, F.A., 2003. Using self-organizing maps toidentify patterns in satellite imagery. Progress in Oceanography 59, 223e239.

Rinaldo, A., Fagherazzi, S., Lanzoni, S., Marini, M., 1999. Tidal network 2: waterheaddelineation and comparative network morphology. Water Resources Research35, 3905e3917.

Risien, C.M., Reason, C.J.C., Shillington, F.A., Chelton, D.B., 2004. Variability insatellite winds over the Benguela upwelling system during 1999-2000. Journalof Geophysical Research. ISSN: 0148-0227 109 (C3). ISSN: 0148-0227, C03010.http://dx.doi.org/10.1029/2003JC001880.

Roman, C.T., Niegering, W.A., Warren, R.S., 1994. Salt marsh vegetation change inresponse to tidal restoration. Environmental Management 8, 141e150.

Rossington, S.K., Nicholls, R.J., Stive, M.J.F., Wang, Z.B., 2011. Estuary schematisationin behaviour-oriented modelling. Marine Geology 281, 27e34.

Solidoro, C., Bandelj, V., Barbieri, P., Gossarini, G., Umani, S.F., 2007. Understandingdynamic of biochemical properties in the northern Adriatic Sea by using self-organizing maps and K-means clustering. Journal of Geophysical Research112, C07S90.

Stive, M.J.F., Rakhorst, R.D., 2008. Review of empirical relationships between inletcross-section and tidal prism. Journal of Water Resources and EnvironmentalEngineering, 2627e2639. ASCE Press, NY.

Struyf, E., Jacobs, S., Meire, P., Jensen, K., Barenderegt, A., 2009. Plant communitiesin European tidal freshwater wetland. Tidal Freshwater Wetlands, 59e70.

Todd, M.J., Muneepeerakul, R., Pumo, D., Azaele, S., Miralles-Wilhelm, F., Rinaldo, A.,Rodríguez-Iturbe, I., 2010. Hydrological drivers of wetland vegetationcommunity distribution within Everglades National Park, Florida. Advances inWater Resources 33, 1279e1289.

Tran, T.T., Van de Kreeke, J., Stive, M.J.F., Walstra, D.J.R., 2012. Cross-sectionalstability of tidal inlets: a comparison between numerical and empiricalapproaches. Coastal Engineering 60, 21e29.

Van de Kreeke, L., Haring, J., 1980. Stability of Estuary mouths in the Rhine-MeuseDelta. In: Proceedings International Conference on Coastal Engineering. ASCEPress, NY, p. 2627.

Van der Wegen, M., Dastgheib, A., Roelvink, J.A., 2010. Morphodynamic modeling oftidal channel evolution in comparison to empirical PA relationship. CoastalEngineering 57 (9), 827e837.

Walton, T.L., Adams, W.D., 1976. Capacity of inlet outer bars to store sand. In:Proceedings 15th the ICCE. ASCE, Honululu, pp. 1919e1937.

Watanabe, A., Shimuzu, T., Kondo, K., 1991. Field application of a numerical model ofbeach topography response. In: Proceedings of Coastal Sediment ’91. ASCEPress, NY, pp. 2627e2639.

Weishar, L.L., Teal, J.M., Hinhle, R., 2005. Designing large-scale wetland restorationfor Delaware Bay. Ecological Engineering 25, 231e239.

Yang, Z., Khangaonkar, T., Calvi, M., Nelson, K., 2010a. Simulation of cumulativeeffects of nearshore restoration projects on estuarine hydrodynamics. Ecolog-ical Modelling 221, 969e977.

Yang, Z., Sobocinski, K., Heatwole, D., Khangaonkar, T., Thom, R., Fuller, R., 2010b.Hydrodynamic and ecological assessment of nearshore restoration: a modellingstudy. Ecological Modelling 221, 1043e1053.