Author's personal copy

9.04 QuantitativeMethodsforEcologicalNetworkAnalysisandIts
ApplicationtoCoastalEcosystems*
REUlanowicz, UniversityofFlorida,Gainesville,FL,USA;UniversityofMarylandCenterforEnvironmentalScience,Solomons,MD,
USA
[Link].
9.04.1 AnAlternativetoMechanism 35
9.04.2 Requirements 36
9.04.3 IssuesNeedingAttention 38
9.04.4 InputOutputAnalysis 38
9.04.5 TrophicAnalysis 41
9.04.6 AnalysisofCycling 43
9.04.7 WholeSystemStatus 46
9.04.8 HigherDimensionalConsiderations 53
9.04.9 SummaryandConclusions 54
References 56
Abstract
Inrecentyears,thedescriptionofsystemnetworkshasbeengaininginpopularityasanalternativetoconventionaldynamical
[Link],ecosystemtheoryhasenjoyeda
legacy of analyzing quantitative digraphs. Ecological network analysis spans all levels of system description. Indirect
[Link] cyclingcan
be identified and quantified. Whole system attributes, for ex ample, organization and flexibility can be gauged. It is even
possibletoprovidespecificdirectionsforremediatingecosystemssoastoimprovetheirsustainability.
9.04.1 AnAlternativetoMechanism
Over the years, ecological modeling has produced numerous
successes(Ulanowicz,2009a).Fisheriesmanagers,forexample,
havefoundLesliematrixmodels(Caswell,1989)tobeuseful
for setting harvest quotas for fish stocks and game species.
Models of single species or single processes have also been
useful for purposes of prediction and interpolation (Jassby
andPlatt,1976).Similarly,modelsthataredrivenbyphysical
forces, such as hydrodynamic simulations (e.g., Wang and
Johnson, 2000), have enjoyed widespread application. When
chemical processes drive the distribution of living organisms,
suchasoftenoccursinmodelsofwaterquality(Biswas,1981),
[Link]-
lular automata have been instrumental in testing and
articulatinghypothesesindisciplinessuchaslandscapeecology
(SandersonandHarris,2000)andindividual-basedmodeling
(DeAngelisandGross,1992).
Problemsarise,however,assoonasthescopeofthemod-
eling exercise encompasses more than one biological process.
Inmostcases,modelsofmultipleinteractingprocessesareless
thanrobustandofmarginalutility(Plattetal.,1981).Notthat
such exercises are completely without merit: multiple process
modelsoftentimesgeneratehypothesesaboutwholeecological
[Link],actuallyusingthemtotesthypotheses
is quite another matter, because the dynamics of multiple
process models usually bear little resemblance to actual
*AdaptedfromUlanowicz(2004),withpermissionfromElsevier.
ecosystems behaviors, unless (as mentioned) the community
being modeled is dominated by a single physical or chemical
factor.
Suchdifficulties notwithstanding, thereremains apressing
needforthestudyofwholeecosystembehaviorhowcollec-
tionsofinteractingecologicalprocessesarelinkedtogetherand
affectoneanother(Lewin,1984).Onewaytoaddressmultiple
biological processesistoforegothespecification ofdynamics
andseewhatcanbelearnedaboutthesystembyfocusingon
the observed configurations of system processes. That is, one
effectively truncates the modeling process after the first two
steps identification and parsing. Identification refers to the
choice of the ecosystem elements that are deemed of interest
(Halfon,1979).Thecastinanyecologicalscenariowillneces-
sarily reflect the particular interests of the individual carrying
out the study as will the suite of interactions among those
components(parsing).
The assumptions that comprise these two steps can be
represented as a set of boxes (nodes) that are connected by
lines (arcs) denoting the pertinent interactions. A surprising
amount can be learned about how the system is functioning
fromthetopologyoftheresultinggraphornetworkofinterac-
tions. In the past decade or so, patterns of connections have
been classified for any number of systems (Barabsi, 2002),
andknowingwhetherasystemisconnected,forexample,ina
powerlawfashionorviaasmallworldtopologyrevealsmuch
abouthowthesystembehaves.
Simply knowing that two components interact constitutes
useful information. More is provided by signifying the
35
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
36 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
direction of the interaction. For example, an influence by A
uponBcanbedenotedbyconnectingAwithBbyanarrowthat
terminates in B (AB). Should B also actively affect A, that
interaction would be denoted by a second arrow, BA. The
ensuingpictureofboxesandarrowsiscalledadirectedgraph,
[Link]
classify ecosystems according to their representations as
digraphsworkthatfallsundertherubricoffood-webtheory
(e.g., Yodzis, 1989; Cohen et al., 1990; Polis and Winemiller,
1995).
AdigraphissaidtobebinaryinthataconnectionfromAto
B either exists or does not. All connections in digraphs are
considered equiponderant, which is almost never the actual
[Link],
[Link]
assigned to each connection that quantifies the differences
amongtheconnections,theresultiscalledaweighteddigraph.
Examples of the progression graphdigraphweighted
digraph are presented in Figure 1. The weighted digraph is
obviouslythemostinformation-richdepiction,becauseeither
ofthesimplerformsofgraphscanbederivedfromaweighted
representation simply by ignoring appropriate information.
Becauseonecanalwaysproceedfromtheweighteddigraphto
one of the simpler forms, but not necessarily in the reverse
direction, the remainder of this chapter will be devoted to
analyses of weighted digraphs. It is noted without further
remark that considerations of the simpler graphs could, in
somecases,resultinmisleadinginferences.
Onefinalclassificationofnetworksoccurswhentheweights
assignedtotheinteractionsrepresentthetransferofsomecon-
servativephysicalmedium,suchasenergyoraformofmatter
(e.g., carbon, nitrogen, or phosphorus). Such weighted
digraphs are known as flow networks. Flow networks can be
assumedtobalancearoundeachnode(i.e.,thesumofinputs
to a given node equals the sum of its outputs), but strict
balance is not required by some of the analyses that will be
discussed.
9.04.2 Requirements
The flow networks most familiar to ecologists are trophic in
nature. That is, the network represents the answers to two
questions: (1) who eats whom? and (2) at what rates? Each
nodeofthenetworkwillbesaidtorepresentataxonthatis,
either an individual, a population of a given species or some
aggregationofspeciespopulationsaccordingtoguild,habitat,
[Link]
taxa, one must then determine for each taxon which of the
othertaxaareincludedinitsdiet.
(a) (b) (c)
19
7
19
14
14
Uptothispointthedigraphforntaxacanberepresentedas
[Link]
entryforrowiandcolumnjwhenevermaterialflowsfromprey
i to predator j; otherwise the ijth component is set to zero.
(Someinvestigatorspreferthatthepredatorsappearintherows
andthepreyinthecolumns,i.e.,thematrixisthetransposeof
that described here.) This binary array is called the adjacency
matrix. The advantage of representing networks as matrices is
that one may then invoke operations from linear algebra to
treat in systematic fashion communities of arbitrarily large
dimensionandcomplexity.
A defining characteristic of an ecosystem is that it is never
closed. There is always exchange of material and energy with
[Link]
[Link],allexogenous(external)
inputstoaparticulartaxonwillbebundledintoasingleflow.
(This assumption is made without loss of generality, because
different types of inputs could always be treated separately,
whenever necessary.) Examples of external inputs include pri-
mary production, immigration, and inbound advection of
material or energy. An exogenous input is represented by an
arrow that has no visible taxon as its origin but which termi-
natesintheparticularreceivingnode(Figure2).
Unlike inputs, two different categories of exogenous out-
putswillbedistinguished:thefirsttypeconsistsoftheexportof
material or energy that is still useful to other comparable
[Link],harvest-
ing by humans, or advection out of the system. Similar to
the inputs, export flows will be represented by arrows that
originate in the appropriate taxon but which terminate in no
node(Figure2).Tobedistinguishedfromusefulexports,some
energywillbedissipatedintoheat(respiration),orsomemate-
rial will be degraded into its lowest energetic form
(e.g., reduction of N to N
2
, or C to CO
2
). Following Odum
(1971), dissipations will be depicted by the conventional
groundsymbolofelectricalcircuitdiagrams(Figure2).
Underthistypology,themaximumnumberofflowsamong
ntaxabecomesn
2
+ [Link]>20,thenumberoftrans-
fersisusuallymuchlower,andtypicallyislessthan15%ofn
2
.
That is, real ecosystems are usually represented by sparse
matrices.
With flow matrices, the weightings become the rates of
exchange among the taxa. Typically, the medium is energy or
a particular chemical element, such as C, N, P, or S. The con-
vention in ecological network analysis (ENA) is that each
network represents a particular medium (although some
investigators allow several (e.g., Hannon et al., 1991)).
(The treatment of several parallel networks of interacting
mediawillbetreatedinSection9.04.8.)
Figure1 Thenestedhierarchyofnetworks:(a)graph;(b)digraph;(c)[Link],R.E.,2009b.A
ThirdWindow:[Link],WestConshohocken,PA.
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 37
10.44
0.05
0.17
Filter
feeders
1
Predators
5
Deposited
detritus
6
Deposit
feeders
4
Micro-biota
2
Meio-fauna
3
41.47
0.51
14.72
5.76
3.58
.
0.33
15.79 0.30
.
0.64
1.91
8.17
.
4.24
6.16 0.43
.
0.66
7.27
1.21
1.21
. .
Figure2 Thetrophicexchangesofenergy(kcalm
2
yr
1
)intheOysterReefcommunity(DameandPatten,1981).Arrowsnotoriginatingfromabox
[Link].
To quantify the various rates, a transfer from prey i to
predator j will be denoted as T
ij
. The rate of exogenous input
to taxon i willbe designated asX
i
, and therate of removal of
usefulmediumfromtaxonitotheoutsideworldwillbecalled
E
i
. Dissipation by i will be
i
. Whenever rates are aggregated
over a sufficiently long interval, medium is likely to balance
[Link],aroundeachtaxon,i,
n n
X
i
T
ji
T
ik
E
i

i
1
j1 k1

or
X
i
T
:i
T
i:
E
i

i
where a dot in the subscript indicates summation over that
index. When reckoned over shorter intervals, however, inputs
andoutputsaroundeachtaxonmaynotbalance.(Theanalysis
of cycling and the whole-system information indices that fol-
low does not require balance around each compartment. The
inputoutput (IO) and trophic level analyses, however, will
bedistortedbyanyimbalances.)
Inthebestofallpossibleworlds,aninvestigatorhasaccess
[Link],
suchasprimaryproduction,suchdataareoftenavailable,but
[Link]
resort, therefore, to indirect means of estimating the magni-
tudesofmostrates.
One helpful assumption for estimating unknown flows is
that the inputs and outputs balance around each taxon. Even
then,theestimationofmagnitudesinabalancedflownetwork
remainsanartformthatcannotbeformalizedinallcases.
Becauseitismoredifficulttomeasuremetabolicratesinthe
field than the sizes of populations, a common approach has
beentoassessfirstthestockofataxon(intermsofthechosen
medium)andtomultiplythatdensitybytheappropriatetabu-
lated metabolic quotients to obtain the desired rates. For
example, if the density of microzooplankton is measured as
150mgC m
2
and published metabolic constants show
their average consumption rate to be 160% per diem, then
the total demand by these organisms should be roughly
2 1
87600mgC m yr . Respiration, excretion, and natural
mortality for this population can be similarly reckoned.
Overall consumption by predators would then be estimated
[Link]
estimates of overall demands and availabilities now at hand,
balancingthenetworkbecomesthetaskofmatchingthecorre-
spondingendsoftheindividualflowarrows.
Perhaps the most popular automated balancing routine
available to the public has been ECOPATH (Christensen and
Pauly, 1992), which was built on an optimization routine by
Polovina(1985).Otheroptimizationroutineshavebeenwrit-
tenbyVezinaandPlatt(1988)andKavanaghetal.(2004).The
maindifficultywithoptimizationalgorithms,however,isthat
they almost always nullify one or more of the flows that are
[Link]
keepsallknownexchangesnonzeroandinproperproportion
[Link],oneseeksamethodofbalancing
thenetworkthatimposesthefewestexternalassumptions.
Two such least-inference schemes are now available
(Ulanowicz and Scharler, 2008). In the first routine,
MATBLD, the initial magnitude of a flow is set to be jointly
proportional to the availability of the given prey and the
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
38 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
demand by the particular predator. If the adjacency matrix of
flowswerefullyconnected(allones),thisjointapportionment
[Link]-
tioned, however, ecosystem networks are almost always
sparselyconnected,soonemustinflatetheestimatesaccording
tojointproportionbyacommonfactorinordertoyieldatotal
system activity (T..) that is commensurate with the collective
[Link]
necessarilybalancethenetwork,andonemustbringthesystem
intobalancebysomeautomatedscheme,suchaslineardonor-
control(AllesinaandBondavalli,2003.)
The second, possibly more promising algorithm, is called
[Link]
flow to all nonzero connections and progressively loads the
flowmatrixbyincrementingeachactiveflowbythesame,very
[Link]
ofthatpreyorifitsatisfiesthetotaldemandbythatpredator,
the flow in question is eliminated from further increments.
Matrixloadinghaltswhennofurtherflowremainstobeincre-
mented. This iteration technique resembles the filling of a
[Link]
is achieved via the Allesina and Bondavalli (2003) equilibra-
tionroutine.
Whenstartedwiththesamedata,networksbalancedusing
MATBLDdonotdifferfromthosecreatedbyMATLODtoany
statistically significant degree; neither do the results of either
method diverge from published networks to any statistically
significantdegree(UlanowiczandScharler,2008).
The minimal information required to create a balanced,
weighted digraph oftrophic interactionsthus consists offield
estimatesoftherespectivepopulationdensities,thepredator
preytopology,andtheappropriatevaluesofthephysiological
[Link]
ofecologicalsystems,itshouldnowbepossibletocreateahost
ofquantified digraphs for adiversity [Link]
now becomes how to analyze such networks in order to
respondtotheusualquestionsposedbyecologicalmanagers.
9.04.3 IssuesNeedingAttention
Theestimatedweighteddigraphsjustdiscussedareessentially
snapshots of the suite of trophic transfers. They do not expli-
[Link],however,
inimplicitinformationonhowthesystemisfunctioningjust
asX-raysofthebodyprovidesnapshotsthat,inturn,provide
myriad clues about physiological functions in the body
(e.g., the presence of stones in the kidneys might indicate
some disorder in the metabolism of calcium). Similarly,
weighted digraphs of ecosystems are rich in information
aboutthestatusoftheensembleatanumberoflevels.
Atthe level of the individual population, oneis interested
not only in how that compartment interacts with each of its
directpreyandpredators,butalsoinhoweachofthosepopu-
lations relates in turn to others in the overall ecosystem. In
otherwords,eventhoughaspeciesinteractsimmediatelywith
only a limited number of other organisms, it communicates
indirectly with a much larger number of elements of the
[Link]
ofmatricesprovidesamajoradvantagetowardassessingthese
[Link],matrixalgebracanbeinvoked
toquantifyindirecteffectsacrossamyriadofnetworkpathways
anendeavorthathasbeentermedIO analysis.
In particular, it is possible via successive application of
matrix multiplication to gauge how many times, on average,
mediumhasbeentransferredalongitswayfromprimarypro-
duction to the taxon of interest, that is, the average trophic
position of the given taxon. Obversely, the same calculations
reveal the proportions of that taxons activity that has arrived
[Link],what
percentage of the taxonsactivity is supported by pathways of
lengthone(herbivory),oflengthtwo(carnivory),andsoforth?
Itthenbecomesamatterofbookkeepingtoconsolidateallthe
activitiesofallcompartmentsatanygivenintegerlevelintoa
virtual representation of that trophic stage. In the absence of
cycles,thesestagesallfittogetherasaclassicaltrophicchainor
[Link],thereductionofa
complicatedfoodwebintoadidacticchainisusuallypossible.
Although they complicate trophic analysis, cycles of med-
iumareofspecialinterest,becausetheyareusuallyconnected
with controls in ecosystems. Endogenous control in noncen-
tralizedsystemsisusuallyexpressedviafeedback,andfeedback
implies the existence of cycles. Hence, how many cycles are
containedinaparticularecosystemnetwork,whatarethecon-
stituents of each loop, and how are they all put together
becomepressingquestions,theanswerstowhichshouldhelp
[Link]
analyses,uncoveringthedetailsofcyclinginecosystemsisnot
effected by matrix operations per se, but by representing net-
worksintermsofmatrices,oneprovidesaformatconduciveto
thesystematicdelineationofcyclingstructure.
Finally,thereremainsthequestionofhowtheecosystemas
[Link]?
Canonealsoquantifytheflexibilityofthesystemtopersistin
thefaceofnovelperturbations?Whataretheprospectsforthe
sustainabilityofaparticularsystem?Wherearethecontrolling
bottlenecks in system functioning? How can system dysfunc-
tionberemediated?Toapproachthesequestions,itshouldbe
pointedoutthatbyidentifyingthevariouscompartmentsofan
ecosystem, the investigator articulates the differences among
the taxa. Gregory Bateson defined information as the differ-
ence that makes a difference. It should come as no surprise;
therefore, that information theory is a most appropriate tool
foransweringsuchwhole-systemissues.
9.04.4 InputOutputAnalysis
Thatnetworkscanberepresentedasmatricesisserendipitous,
because diagrams otherwise grow too complex to be able to
[Link]-
venient starting point for IO analysis is the definition of the
matrix of dietary proportions, [G]. The elements of [G] are
calculated using the elements of the flow matrix, [T], and the
inputvector,(X).EachT
ij
of[T]isnormalizedbythetotalinput
tothereceivingcompartment,j:
T
ij
g
ij
_ _ 2
T
:j
X
j
so that the element g
ij
of [G] becomes the fraction which i
comprises of the total input into j. In other words, element i
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
_ _
_ _
_ _
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 39
incolumnjof[G]describesthepercentageofiinthedietofj.
Hence,[G]isknownasthematrixofdietarycoefficients.
Most fortuitously, the algebraic powers of [G] reveal the
fractionsofthedietofjthatflowstoialongpathwaysoflength
corresponding to the integer power to which [G] is raised. Its
algebraicpowershaveverydidacticmeaningsasregardstrophic
[Link]
[Link]
ij
becomethe
sixcorrespondingnonzeroelementsof[G]:
_
0 g
12
g
13
g
14
0 0 g
G
23
g
24
0 0 0 g
34
_

0 0 0 0

Thesecondpowerof[G
_
]isobtainedviamatrix
_
multiplication
of[G]byitself,
_ _
0 0 g
12
g
23
g
12
g
24
g
13
g
34

0 0 0 g
23
g
34

2

G
_ _
0 0 0 0
0 0 0 0
Onenoticesthateachofthenonzeroelementsof[G]
2
consists
of terms that represent all of the pathways of length 2 that
connect i with j. For example, elements 13 of [G]
2
reveal
how much gets to 3 from 1 over the single two-step pathway
123. Element 14 of [G]
2
is made up of two terms, each
representing a two-step pathway from 1 to 4. The first term
givesthefractionoftotalinputto4thatflowsoverthepathway
124, whereas the second term reports the fraction of
inputto4thathastraversedtheroute134.
Multiplying[G]
2
yetagainby[G]producesthematrix[G]
3
:
_ _
0 0 0 g
12
g
23
g
34
3
0 0 0 0


G :
_ _
0 0 0 0
0 0 0 0
Thereisonlyonenonzeroelementof[G]
3
anditrepresentsthe
sole pathway of length 3 in the network, 1234. The
powers of [G] truncate with [G]
4
=[0], because there exist no
pathwayslongerthan3inthisnetwork.
Thereaderisinvitedtoexperimentwithothersimplegraphs
toconvinceherself/himselfthattheelementsofthemthpower
of[G]areformedbycontributionsfromallpathwaysofexactly
lengthmandnoothers.
Thewayinwhich[G]wasnormalizedguaranteesthateach
g
ij
1, making it highly probable that elements in the higher
1
2
4
3
Figure3 Asimplehypotheticalnetworkoffourcomponents.
powers of [G] will grow progressively smaller. The question
thusariseswhetherornottheinfiniteseries
2 3 4

I G G G G
convergestoafinitelimit.(Thematrix[I]{=[G]
0
}iscalledthe
identity matrix. It consists of ones for i=j and zeroes else-
where.) Simon and Hawkins (1949) demonstrated that the
normalizationprocedureineqn[2]guaranteesthattheinfinite
seriesofmatrixpowersdoesconvergetoafinitelimit,namely,

1
lim I G G
2
G
3
G
4
IG
This limit, the matrix inverse of [I G], is called the Leontief
structure matrix, [S]. The ijth component of [S] reveals the
fractionofthetotalinputintojthatleftiandtraveledoverall
pathwaysofalllengthstosatisfyafinaldemanduponjofone
unit. Final demand is a pivotal concept in economic theory,
andthe[S]matrixprovidedeconomistswithestimatesofhow
much production was required by each economic sector in
order to satisfy any prescribed vector of final demands.
Leontief(1951)formulatedthebeginningsofwhateventually
was called IO theory. Hannon (1973) introduced Leontiefs
methodsintoecology.
The homologs of final demands in economics are the
respirations in ecology. Although respiration is important to
ecologists, the intermediate transfers among system compo-
nents interest them more. That is, ecologists give priority to
knowing how much of what arrives at j can be traced to a
particular activity i. Szyrmer and Ulanowicz (1987) demon-
stratedhowsuchintermediatetransferscanbeestimatedbya
simple transformation of the [S] matrix. Szyrmer called the
transformedmatrixthetotaldependencymatrix,[D].Theele-
mentsofthe[D]matrixarecalculatedfrom[S]andtheoriginal
flows[T] as
T
i:
d
ij
S
ij

ij
s
ii
T
:j
where
ij
[Link]
from 0 to (n+2), and the T
0j
represent the exogenous inputs
to j, T
i,(n+1)
are the useable exports from i to other systems of
comparablescale,andT
i,(n+2)
arethedissipativelossesfromi.
Effectively,d
ij
answersthequestion,Whatfractionofthetotal
diet of j passes through i along its way to j? Thus, the jth
column of [D] becomes the indirect diet of j the respective
amountsbywhichjdependsupontheactivityofeachelement
in the ecosystem. Because medium from i can visit several
compartmentsalongitswaytoj,theelementsofthejthcolumn
sum to 1, and that sum also signifies the effective trophic
positionofj.
Itispossibletousethematrixofindirectdietstodifferenti-
ate trophic roles. For example, the Chesapeake mesohaline
ecosystem is host to two piscivorous predators, striped bass
(Morone saxatillis) and bluefish (Pomotatus saltatrix). One
would expect heavy competition between them. Baird and
Ulanowicz (1989), however, calculated their indirect diets,
revealing(amongothers)thefollowingindirectdependencies:
(1)stripedbassonzooplankton:65.8%;(2)bluefishonzoo-
plankton 28.7%; (3) striped bass on polychaetes 1.8%; and
(4)bluefishonpolychaetes48.0%.
These apportionments revealed significant stratification of
[Link]
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
40 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
showed a high dependency on pelagic production, whereas
Bluefish derived their resources more from benthic secondary
production.
Instead ofnormalizing [T] down its columns to get ,one
could have as well normalized across the rows of [T], which
wouldhaveresultedinamatrixofhostcoefficients,[F]:
T
f
ij

ij
T
i:
E
i

That is, the component, f

ij
, of [F] reveals the fraction of the
totalactivityofanycompartmentithatflowsdirectlytosome
other element j. The corresponding output structure matrix,
[] =(I F
T
)
1
, was originally formulated by Augustinovics
(1970) to estimate how much of the activity of any sector j
isgeneratedbyasingleunitofprimaryinputtoi.(Thesuper-
script T indicates matrix transpose, i.e., the matrix that results
when the rows of a matrix are exchanged with its columns
[i.e., (f
ij
)
T
=f
ji
].) [] is the obverse set of indicators to [S].
Primary inputs are of substantial interest to ecologists, but
many might prefer to know the intermediate contributions
of each compartment to all the others, that is, how much of
[Link]
coefficients, Szyrmer defined a total contribution matrix, [C],
with the elements
T
i
c
:
ij

ji
f
ji
_

ii
T
:j
_
where
_ _
ij
isthei-jthelementof[].
It is sometimes useful to regard the elements of the total
contributionmatrixasrepresentingtheefficiencieswithwhich
the medium flows from any one compartment to any given
other. Suppose, for example, that one desired an indicator of
overall ecosystem efficiency, that is, how much primary pro-
duction eventually reaches the upper trophic components. In
thisregard,Ulanowicz(1984)wantedtocomparethetrophic
efficiencies of two tidal marsh ecosystems situated near the
CrystalRivernuclearpowerplantonthewestcoastofFlorida.
Thefirstnetworkstraddledthethermaloutfallfromtheplant
(T6

C),whilethesecond(control)marshwasbeyondthe
[Link]-
tribution coefficients of primary production to the Gulf
Killifish (Fundulus grandis) and to two Needlefish (Strongylura
marinaandStrongyluranotata)revealedthefollowing:
Control T
GulfKillifish 0.14710
3
0.6710
3
(54%)
Needlefish 0.33810
3
0.14010
3
(59%)
The impact of the thermal effluent was obvious. Heated
waterloweredtheoverallefficiencyoftheecosystemforprodu-
cingtopcarnivoresby5060%.
A frequent criticism of IO methodology is that it pertains
only to steady-state or temporally averaged networks. Several
attempts have been made to expand IO theory to treat time-
[Link].(1979),forexample,regardedany
imbalance simply as another category of input or output
(depending on the direction of the imbalance). Matis and
Patten (1981) scaled the flows by the storages of their source
compartments and inflated the diagonal flows to better reflect
[Link]-
mation-theoreticmethodstoatimeseriesofnetworksnapshots
depictingthedynamicsoveraninterval(SeeSection9.04.8).
StillanotherdeficiencyofIOanalysisintheeyesofmanyis
that it pertains to only positive flows of the medium. Beyond
contributionsanddependencies,ecologistsarealsointerestedin
howthenegativeeffectsofpredationmightpropagatethrough-
out the system. Fortunately, the same linear algebra used to
quantify positive contributions can evaluate as well the net
positiveornegativetrophicimpactthatanyonespecieshason
any other (Ulanowicz and Puccia, 1990). One notes that g
ij
quantifies the positive impact of prey i upon predator j. By
contrast, f
ji
measures the negative impact that i has on j.
Consequently, the overall effect of i on j should be related to
thedifferenceg
ij
f
ji
.Unfortunately,becausef
ij
hasbeennormal-
ized by the total output from i, rather than just the predatory
lossesthati sustains,itunderestimatesthemagnitudeofnega-
[Link],f
ij
shouldberenormalizedby
itssecondaryproductionalone,ratherthanbyitstotaloutflow.
Therenormalizedpredatorimpactsarelabeledf
ij
*,where
f

T
ij

ij
nl

T
im
m1
and the index m is summed over only the subset of living
components (nl in number) of the full suite of n ecosystem
components. As a result, a more accurate measure of the net
directeffectthatihasonjbecomes
q
ij
g
ij
f
ji

where1q
ij
1.
Inalmostallinstances,thepowerseriesof[Q]convergesin
themannerthatthecorrespondinginfiniteseriesofthepowers
of[G][Link],
lim
_
I Q Q
2
Q
3
Q
4

IQ
1
Inrare(usuallydegenerate)instances,thematrix
_
[IQ]canbe
singular,andthentheseriesdoesnotconverge.
The net trophic impact (direct and indirect) of any given
compartment i upon any chosen other, j, is the ijth compo-
nentofthematrixofnettotalimpacts,[M].Itiscalculatedas
{[I Q]
1
[I]}. For any connected graph, all n
2
elements of
[M] will be nonzero. Such a jumble of impacts would be
difficulttointerpret(e.g.,HillebrandandShurin,2005)failing
a systematic way of presenting the results. The original algo-
rithm for calculating these indirect trophic influences
(IMPACTS)allowstheusertofocusuponaparticularcompart-
mentandrankstheimpactsofalltheotherspeciesuponthat
taxon,sothatthemajorinfluencesareeasytospotateitherend
oftheranking.
Patten (1982, On the quantitative dominance of indirect
effects in ecosystems, unpublished paper presented at the
Third International Conference on State-of-the-Art in
Ecological Modeling, Colorado State University, 2428 May,
FortCollins,CO)positedatendencyfornegative,competitive
[Link]
the non-normalized form of f
ij
, Fath and Patten (1998)
counted a higher ratio of positive elements in [M] than are
present in the direct effects matrix, [Q]. As noted above, the
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
2
8
0

2
0

1
45
3 200
15
3
0
5

4
90 15
50
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 41
lack of renormalization (to f
ij
*
) would tend to inflate positive
effectsin[M].Nevertheless,itiseasytoidentifyinstanceswhen
negativedirectinteractionsbecomepositiveindirectinfluences,
even under the more conservative normalization used above
(BondavalliandUlanowicz,1999).
9.04.5 TrophicAnalysis
It has been mentioned how the powers of the [G] matrix
elaborate flow in terms of integral trophic steps. It stands to
reason, therefore, that the powers of [G] can be employed to
quantifythetrophicstatusofanecosystemnetwork.
Ithappensthatthenotionoftrophiclevelbeganwithavery
simple depiction of the ecosystem as a trophic chain or pyr-
amid. At the first level, energy is fixed by primary producers
(plants),andsome(butneverall)ofthatenergyistransmitted
to plant eaters (herbivores) at the second level. Subsequent
transfer is made to carnivores at the third trophic level, then
to the top carnivores at the fourth, etc. Rarely is any mention
made of tertiary carnivores and beyond. Earlier, it was pre-
sumed that thermodynamic losses of energy and material
during each transfer curtailed trophic chains to no more than
five steps, but that assumption was shown to be problematic
(PimmandLawton,1977).
Ofcourse,natureishardlyeverassimpleasastraightchain
orpyramid, and theactual webof interactions usually proves
to be quite complicated (Hillebrand and Shurin, 2005.)
Oneseeks,therefore,asystematicmethodformakingsenseof
[Link],infact,obligate
autotrophs or herbivores, many of the heterotrophs feed at
severallevels,sothatnoassignmentoftaxatointegraltrophic
levels seems obvious. Cousins (1985), for example, remarked
as how a hawk usually feeds at five separate trophic levels.
Although some urge that the concept of a trophic level be
dropped as a useless fiction, the properties of the powers of
[G] give reason to hope that some sense can be made of
Darwinsentangledbank.
Levine(1980)observedthatthesumofanycolumnofthe
structure matrix, [S], was related to the number of trophic
transfers that medium had experienced on its way to that
given compartment. This sum is generally not an integer, but
Levinesuggestedthatoneregarditasanaveragelevelatwhich
the taxon is feeding. In the hypothetical example shown in
Figure 4, compartments 1, 2, and 3 are arrayed in chain-like
fashion. Compartment 4, however, receives only 5 of its
[Link],30%
ofitssustenancearrivesthethirdleveland60%atthesecond.
Hence, its effective trophic level can be reckoned as
(0.62)+(0.33)+(0.14) =2.5. In the [S] matrix corre-
sponding to this network, the first three columns sum to 1.0,
2.0,and3.0,respectively,andthesumdownthefourthcolumn
is 2.5, as just calculated. This calculation applies to the most
[Link](1999)
used Levines notion of effective trophic level to establish the
directionofcausality([Link]-down)inseagrasscom-
[Link],[Link].(1998)usedit
todemonstratehowworldfisheriesinthelastdecadeshavebeen
fishing down stocks, that is, decimating higher trophic level
speciesandrelyingmoreonlowerlevelpopulations.
Now that it is possible to attach several trophic levels to a
particular compartment, one asks whether the reverse might
also be feasible that is, can the activity of a given compart-
ment be apportioned among several integral trophic levels
(Ulanowicz and Kemp, 1979)? In this regard, one notes that
the fraction by which compartment i feeds directly upon pri-
marysourcesis
T
oi
g
oi

n
T
mi
m0

so that the fractions for all n components can be arrayed as a

rowvector,callit(L
1
)
T
=(g
O
)
T
.Eachelementofthisrowvector
reveals the fraction of the activity of the respective taxon that
can be attributed to primary production. To calculate the
amountsbywhicheachtaxonfeedsasaherbivore,itisneces-
sary only to multiply [G] by (L
1
)
T
from the left to generate a
new row vector, (L
2
)
T
= (L
1
)
T
[G]. In like manner, the fractions
by which each compartment feeds as a carnivore (acts at the
third level) become (L
3
)
T
= (L
2
)
T
[G]. By mathematical induc-
tion,thefractionsatthemthtrophiclevelbecome(L
m
)
T
=(L
1
)
T
[G]
m1
. Whenever cycles are absent from a network, the
sequence of row vectors, (L
1
)
T
, (L
2
)
T
, (L
3
)
T
,, truncates
(become all zeroes) within n1 steps. As a result, whenever
cycling isabsent,itbecomespossibletodefineanunambigu-
ousLindemantrophictransformationmatrix,[L],suchthatthe
ith row of [L] is constituted by (L
i
)
T
. The fraction by which
taxonjfeedsattheithtrophiclevelisgivenbymatrixcompo-
nent L
ij
. Furthermore, reading across the row i reveals the
compositionofithvirtualtrophiclevel.
Applying this construction to the simple network in
Figure4yieldsastheLindemantrophictransformationmatrix,
L
_
1: 0: 0: 0:
0:
_

1: 0: 0:6

_
0: 0: 1: 0:3
0: 0: 0: 0:1

Readingdown the columns,one seesthat

_
the first threecom-
partmentsactwhollyatoneofthefirstthreetrophiclevels,but
[Link]
across the rows reveals the apportionments of the various
components to the virtual integer levels. For example, virtual
trophiclevel3consistsofalloftheactivityoftaxon3and30%
oftheactivityoftaxon4.
Reality is not hypothetical, however, and almost all real
trophic networks contain cycles. Although this is true, Pimm
(1982) nevertheless observed that cycles confined strictly to Figure4 [Link].
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy

0
n
(n nl)
nl
L
0
000

0
: : : :
: : : :
000 0

111 1
(n nl) nl
200
I
110 35 5
II III IV
90 75 30 5
42 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
living,feedingtaxaremainquiterareandareusually smallin
magnitude. Reading Pimms statement the other way around,
virtuallyallcyclesinecosystemscontainatleastonenonliving
[Link]
place all living compartments (say there are nl of them) first,
the nlnl submatrix of transfers among the living compart-
mentsshouldcontainveryfewcycles,allconsistingofflowsof
[Link]
be removed from networks without affecting the balance
around any constituent. It becomes possible, then, to remove
such feeding cycles without significantly altering the nlnl
[Link]-
tion matrix [L] can then be constructed using the nlnl
submatrix of [T] from which cycles of predatory exchanges
havebeenremoved(Ulanowicz,1995).
Tosortoutthemanifoldpredatoryflowsintheecosystem,
one regards [L] as a conventional transform in linear algebra.
Thatis,[L]canactupontheknowninputvector(X)todefinea
vectorofcanonicalinputs,()(Greekuppercasepsi):
L X
Similarly,canonicalformsfortheusefulexports,(E),anddis-
sipations,(R),become
Y E L
and
P R L
respectively.
Onethenseekstocreateacanonicalmatrixofflows[] by
transformingthematrixoforiginalflows,[T],inconventional
fashion:
T
L T L :
Unfortunately,[L]isnotorthonormal,sothatthetransformed
matrixofflows,[],[Link]
be redeemed, however, because [] possesses a peculiar form
of symmetry that allows one to rework it into a surrogate
diagonal matrix, []. [] has as its ith diagonal element the
sum of the ith row of [] and zeroes elsewhere, that is,

ii
=
k

ik
, and
ij
=0 otherwise. Then in analogy to eqn [1],
onemaybalanceeachvirtualtrophiclevel,,asfollows:

Proceedinginthisway,theLindemanmatrix[L]shownabove
forthesimplenetworkinFigure4transformsthatnetworkinto
theequivalentstraightchaininFigure5.
Althoughthecanonicaltransformationjustdescribedworks
whenevertherearenocyclesintheentirenetwork,significant
cyclesarepresentinalmosteveryrealecosystemnetwork and
involve mostly the nonliving compartments. Arranging the n
compartments so that the nl living taxa always appear first
facilitatestheremovalofall(insignificant)cyclesamongonly
Figure5 TheLindemanspineofthenetworkinFigure4.
Figure6 Thepartitionedexpansionofthe[L]matrix.
predators. This arrangement allows one to create an nlnl
Lindeman matrix, [L], from the revised nlnlacyclic subma-
[Link],
oneplaces[L]intotheupperlefthandcorneroftheexpanded
nn matrix, [L], as in Figure 6. The remainder of matrix is
filled out with zeroes, except for the final [nnl] elements of
thenthrow,allofwhicharesetequaltoone.
The ensuing matrix [] creates what is called a Lindeman
spine of chain-like flows. The remainder of the [L] matrix
aggregates all the nonliving compartments into a single non-
livingpoolinsuchawaythatmaintainsbalancearoundeach
virtual component. Applying this method to the network in
Figure2producestheLindemanspineinFigure7alongwith
itsassociateddetritalflows.
Transforming general networks into canonical trophic
chains allows the investigator to compare trophic dynamics
among disparate ecosystems or to monitor changes in the
same ecosystem over time. It should be mentioned that one
oftenencounterscanonicaltrophicchainsthatarelongerthan
fivesteps,seeminglyatoddswiththeobservationofPimmand
Lawton (1977) that trophic chains do not exceed five steps.
Closer inspection, however, shows that the magnitudes of
transfers beyond the fifth level are miniscule. Whipple and
Patten(1993)writeaboutactivitiesatveryhightrophiclevels
(e.g., >10), but the magnitudes associated with such levels
make it improbable that more than a few molecules reach
[Link],
one notes that environmental perturbations usually act to
shortenthelengthofthecanonicalchainbydisproportionately
diminishingflowsathigherlevels(Ulanowicz,1984).
In addition to these IO methods, Bernard Patten and
associateshavedevelopedwhattheycalledtheenvirontheory
of ecosystems (Patten, 1978, 1982; Patten and Auble, 1981;
FathandPatten,1999).Byanenvironismeanttheentiretree
of flows antecedent to a particular compartment, or, alterna-
tively, the subsequent dendrite of medium exiting a
compartment. That is, an input environ is the quantitative
description of where the current biomass has appeared since
itenteredthesystemandthepathwaysalongwhichitflowedto
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
10.4 0.038 0.003 0.009 0.0005
0.51 3.03 0.21 .006
16.08
9.16 0.83 .069
4.37 1.98 0.31
6.66
I II III IV V
Detritus

41.5
14.8 .003
15.8
.003
6.16
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 43
Figure7 ThecanonicalrepresentationoftheOysterReefnetwork(Figure2).
arrive at the node in question. Similarly, an output environ
describeswhichothernodeswillreceivethematerial inques-
[Link]
canbedecomposedintoasuiteofinputandoutputenvirons.
Environs are useful for scoping out the possible causes and
effects of any changes in a particular population. One signifi-
cant result of environ theory is that indirect effects are very
frequentlymoresignificantthandirectinteractions.
A useful survey of IO and trophic methods of network
analysis can be found in Higashi and Burns (1991) and
Higashietal.(1991).
9.04.6 AnalysisofCycling
Asmentionedinthelastsection,mediumdoesnotsimplypass
straight through an ecosystem. Scarcity and crowding make it
inevitable that some of the same material be reused by the
[Link],chancesarethat
the same medium will be used more than once by the same
[Link],thesystemengagesinrecycling,ormore
simply,cyclingofmedium.
Words such as scarcity and inevitable hint that cycling in
ecosystemsisassociatedwithcontrolsthatguidetheprocessing
of medium by biotic species. Ecosystems are likely to develop
meansforretainingchemicalelementsthatarebothscarceand
[Link],whichrewardseach
participantforitsroleinkeepingthescarcemediumincircula-
tion(Ulanowicz,2009b).Hence,asignificantfocusofthestudy
ofecosystemdynamicsshouldbethepatternandmagnitudesby
[Link],oneneedstoknowthe
topologyofpathwaysoverwhichmediumisrecycledaswellas
someideaofhowmuchmaterialiscycledaroundeachloop.
It was the latter, quantitative issue that was first received
attention in ecology. Odum (1969) identified the magnitude
ofcyclingasoneofhis24criteriadefiningmatureordeveloped
[Link].(1976)andFinn(1976), therefore, set
about to use IO analysis (earlier introduced to ecology by
Hannon,1973)toquantifytheaggregateintensityofcyclingin
a system. Finn reasoned that each diagonal element in the
Leontiefstructurematrix,[S],isrelatedtotheprobabilitythata
quantum of medium will visit the same designated compart-
[Link]
inthegivennetworkbequantifiedbymultiplyingeachdiagonal
element by the total activity (throughput) of that particular
taxonandthensummingallsuchproducts(i.e.,T
cyc
=
i
[T
i.
S
ii
]).
Overtimetheratioofthissumtothetotalsystemthroughflow
(T..)becameknownastheFinncyclingindex(FCI=[T
cyc
/T..]).
Asnotedearlier,SzyrmerandUlanowicz(1987)elaborated
uponLeontiefsaccountingschemeandnotedthatthediagonal
elementsofboththetotaldependencymatrix,[D],andthetotal
contributionmatrix,[C],[Link],thesediag-
onals,ratherthanthediagonalsof[S],moreaccuratelyrepresent
the probability that a given quantum leaves a particular taxon
[Link]
of either [C] or [D] be used in lieu of those of [S] in Finns
calculations.(Han,1997independentlyrecapitulatedSzyrmers
calculations.) Allesina and Ulanowicz (2004) remarked how
previous estimates of cycling ignored some of the compound
[Link]
a tedious method to calculate the full amount involved in
cycling. Fortunately, their revised estimates amounted to the
same 14% in all the cases examined. In effect, the original
index continues to be a useful estimate, when appropriately
modified([Link],personalcommunication).
TheproblemwiththeFinnindexisthatitsmagnitudedoes
not always reflect the developmental status of an ecosystem.
Ulanowicz (1984), for example, showed that perturbing an
ecosystem could lead to an increase the FCI (see also Wulff
andUlanowicz, 1989).Ulanowicz (1984)suggested thataug-
mentedcyclingwasahomeostaticresponseofanecosystemto
[Link]
thebiomassofhighertaxa,andthesystemrespondsinhomeo-
static fashion to prevent, insofar as possible, the loss of the
resourcebycyclingitamongthelowertrophictaxa.
To understand better the increase of the Finn index in
response to stress requires detailed knowledge of how the
pattern of recycling changes under stress. It happens that
longer, slower cycles involving the higher taxa are usually
replaced under perturbation by shorter, faster cycles among
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy

44 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
[Link]-
ingtheexacttopologyofcycling(Ulanowicz,1983,1986).
The first task in elaborating the topology of cycling is the
identification of all simple cycles in a network. Simple cycles
are those in which any element appears only once. Writing an
algorithmtofindallthesimplecyclesinanetworkwouldappear,
at first glance, to be straightforward. For example, one could
perform a depth-first search with backtracking (Tarjan 1972;
Mateti and Deo, 1976). A backtracking algorithm starts with
ordering the nodes in some convenient way (to be described)
andthenrepeatsthesameorderofnnodesatnlevelsasdepicted
inTable1.
Thesearchbeginsatthefirstnodeinthezerothlevel(desig-
nated as the pivot element) and the nodes in the next higher
levelareexaminedfromlefttorightuntilanetworkconnection
[Link]
initiatessearching(lefttoright)thelevelabovetofindacon-
[Link],
thelastnodevisitedineachpreviouslevelistemporarilystored
[Link]-
cing to any higher level, one checks the destination there to
makesurethatthisnodedoesnotappearinthecurrentpath-
way.(Thesearchisforsimplecycles.)
Oneclimbstoashighalevelaspossible,untilhaltedbyone
oftwoconditions:(1)Ifthelinkencounteredinthenextlevelis
tothepivot(starting)element,thenasimple-directedcyclehas
[Link]
pathway,andthesearchcontinues.(2)Ifthesearchisfromnode
k at level m and all possibilities in level m+1 have been
exhausted(i.e.,onereachesthefar-rightextreme),thenoneback-
trackstothenodeinthecurrentpathwayatthe(m1)thlevel
and reinitiates searching the mth level, beginning with the
(k+1)
st
node. When further backtracking becomes impossible,
allcyclespassingthroughthepivotelementhavebeenidentified.
Thepivotelementmaybeeliminatedfromfurtherconsideration
(therebydecreasingthedimensionofthesubsequentsearch).
Asanillustration,thebacktrackingalgorithmisappliedtothe
OysterReef(DameandPatten,1981) network in Figure2. Cycles
canbefoundmostefficientlyifthenodesareorderedinapropi-
[Link],onewishestoconsiderfirstthosenodes
[Link]
likelihoodistocountthenumberofcyclearcsincidenttoeach
[Link](Knuth,
1973).Tocountuptheincidentcyclearcs,onefirstdeterminesall
oftheothernodesthatcanbereachedfromthegivennode.(The
reachable nodescan be located in n
2
orfewersteps.)Onethen
countsuptheconnectionsfromeachreachablenodedirectlyback
tothegivennode(anothernsteps).Theconsequentsumisthe
numberofcyclearcsincidenttothegivennode.
Table1 Theorderofthenodesinadepth-firstsearch
n | [1] [2] [3] [n]
. | . . . .
L . | . . . .
e . | . . . .
v 2 | [1] [2] [3]

[n]
e 1 | [1] [2] [3]

[n]
l 0 | [1] [2] [3]

[n]
|__________________________________
Compartment
Performing these operations on the Oyster Reef network,
onecounts3cyclearcseachincidenttocompartments4and6.
There are 2 cycle arcs into compartment 3 and one each into
compartments2and5.Compartment1possessesnoincident
[Link]-
tious order in which to consider the compartments thus
becomes 4, 6, 3, 2, 5. The following mnemonic array may
helptokeeptrackoftheorderofoperations:
L 4 j 4 6 3 2 5
e 3 j 4 6 3 2 5
v 2 j 4 6 3 2 5
e 1 j 4 6 3 2 5
l 0 j 4 6 3 2 5

Compartment
Onebeginswithpivotelement4inlevel0andsearcheslevel1
fromlefttoright,followingtheinstructionsinthetextabove.
The order of the significant operations in the search and the
currentpathwaystheygenerateareasfollows:
Operation Current
pathway
Beginatpivotelementinlevel0 4
Advancetolevel1 46
Reportcycle1 464
Advancetolevel2 463
Reportcycle2 4634
Backtracktolevel1 46
Advancetolevel2 462
Reportcycle3 4624
Advancetolevel3 4623
Reportcycle4 46234
Backtracktolevel2 462
Backtracktolevel1 46
Backtracktolevel0 4
Advancetolevel1 45
Advancetolevel2 456
Reportcycle5 4564
Advancetolevel3 4563
Reportcycle6 45634
Advanceinlevel3 4562
Reportcycle7 45624
Advancetolevel4 45623
Reportcycle8 456234
Backtracktolevel3 4562
Backtracktolevel2 456
Backtracktolevel1 45
Backtracktolevel0 4
Advanceinlevel0 6
Allcycleswith4havebeenfound
Advancetolevel1 63
Reportcycle9 636
Backtracktolevel0 6
Advancetolevel1 62
Advancetolevel2 623
Reportcycle10 6236
Backtracktolevel1 62
Backtracktolevel0 6
Advanceinlevel0 3
Allcycleswith6havebeenfound
Advanceinlevel0 2
Advanceinlevel0 5
Furtherbacktrackingimpossible: END
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
(a)
A
5
35
Smallest
D B
arc
25
20
30
40
C
15
25
100
70
(b)
10
10
15
A B C D
100
65 35 25
35 20 25 25
(c)
A
D
C
B
5
5
5 5
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 45
One sees that the Oyster Reef network contains 10 simple
cycles. If the components had been considered in the order
given in Figure 2, the first pivot element would have been 1,
and considerable time would have been wasted searching for
[Link]
pivot elements can significantly diminish the time it takes to
completethefullsearch.
Inprinciple,thetimespentsearchingnetworksiscombina-
toricinmagnitude(i.e.,itvariesasn-factorial).Thismeansthat
yearsofcomputertimecanbespentsearchingnetworksofeven
[Link]
be guaranteed for the method just described. As mentioned,
however, ecological networks of moderate dimension are
rathersparse,thatis,theytypicallypossess15%orfewernon-
zero arcs. This, in combination with the component ordering
scheme outlined above, makes the identification of simple
cyclesfeasibleformostestimatedecosystemflownetworks.
Having identified all simple cycles in a network, one now
seeks to separate cycled from straight-through flows, that is,
[Link]
an isolated cycle from out of its supporting network is rather
[Link]-
nitude(DA in Figure8(a)).
Thenthemagnitudeofthatsmallestlinkissubtractedfrom
the weight of each link in the cycle. Doing so separates the
target cycle (e.g., ABCDA) from the supporting
Figure8 (a)Hypotheticalnetworkconsistingofonesimplecycle.(b)Single-passflowsthroughthenetwork.(c)[Link]
(a)=(b)+(c).
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
46 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
straight-throughpathway(Figure8(b)).Eachlinkofthecycle
bears the magnitude of the smallest arc in the original cycle
(Figure8(c))andthebalancearoundeachofthenodesinthe
residualfragment(Figure8(b))hasbeenmaintained.
Thismethodofcycleextractionwillworkonlywhennone
of the simple cycles in the network overlaps. In most whole-
system networks, however, the number of simple cycles far
outnumbersthetotalnumberofflowscomprisingthenetwork.
Hence, some cycles must share the same smallest link. The
collection of simple cycles that shares the same smallest link
is henceforth called a nexus. It is also likely that all cycles
comprising a nexus are controlled by their shared smallest
link. Removing that smallest link by subtracting it from any
onecycleofthenexuswouldalsonegatealltheothercyclesin
the nexus, but choosing any particular cycle is likely to be
[Link]
arationalmethodforapportioningthemagnitudeofthesmal-
lestlinkoverallthecyclesofthenexus.
Somereflectionwillrevealthatthereareaninfinitenumber
of ways of apportioning the least flow over all cycles of the
[Link],onecouldsimplydividethemagnitudeof
the smallest link by the number of cycles in the nexus and
uniformlysubtractthatamountfromeacharcofallthenexus
cycles. However, this uniform distribution seems arbitrary as
[Link]
leastflowtoeachnexuscycleinrelationtothefractionofthe
total nexus activity that is comprised by that cycle. Perhaps
morereasonablestillwouldbetousethe[F]matrixtocalculate
the probability that a quantum of medium anywhere in a
simplecycle willcomplete thatparticular pathway andreturn
toitsstartingpoint.(Thatprobabilityissimplytheproductof
allthef
ij
intheparticularcyclepathway.)Onethendividesthat
probability of completion by the sum of all other such
[Link]
thesmallestmagnitudeisthensubtractedfromeacharcinthe
given cycle. Once the subtraction is done for all cycles in the
nexus, thesmallestlink willzero-out, but thebalancearound
all nodes will remain intact (William Silvert, personal
communication).
After removing all cycles from the starting network, one is
[Link]
procedureensuresthatnoneoftheinputs,exports,ordissipa-
tionsinthesystemhavebeenaltered.(Foreveryamountthat
hasbeensubtractedfromaninputtoanode,anequalmagni-
tude was also removed from an output of the same node.)
Anetwork of allthe medium beingcycled canbeconstructed
eitherbysimpleadditionofthecyclesthathavebeenremoved,
or by subtracting the residual tree from the starting network.
Figure9(a)showstheresidual,acyclicnetworkfromtheOyster
Reef example (Figure 2) and Figure 9(b) the complementary
webofpurelycyclicalflows.
Although the partitioning of a graph into its cyclic and
acyclic components is anabstract exercise, italsohas itsprac-
[Link](1989),forexample,extracted
allthecyclesfromthemesohalineChesapeakeBayecosystem
(Figure 10) and were surprised to find that the aggregated
cycling was comprised of two nonoverlapping graphs. One
cluster cycled among only planktonic species (those that
move passively with water currents), while the other encom-
passed both benthic (bottom-dwelling) compartments and
nektonic (actively moving through the water) species. The
two constellations of cycling represent independent domains
of control anapparently common characteristic of estuaries
(Allesinaetal.,2005).Anotheritemofinterestwasthatnoneof
the filter-feeding compartments, be they on the bottom or
amongthefishes,[Link],
their function in this context appeared to be the transfer of
resources from the pelagic realm into the benthicnektonic
domain. It was also of particular interest that two compart-
ments of pelagic microbes that are normally included among
themicrobialloopofopenoceanicwaterswerenotamongany
of the cycles in the Chesapeake system. In the Chesapeake,
these compartments seemed instead to be transferring carbon
(which is fixed by excessive plant growth in Chesapeake Bay)
outoftheecosystem.
9.04.7 WholeSystemStatus
The methods discussed thus far treat either particular system
components,orasubsystemofthewhole,butecosystemman-
agers increasingly are wanting to know how an ecosystem is
performing in toto. Funding agencies more often are seeking
researchthataddressesenvironmentalproblemsinthecontext
of the whole ecosystem (NSF, 1999). As mentioned in the
introduction, the conventional response to the need for
whole-ecosystem research has been to simulate the system
using a mechanistic mathematical model. Such models, how-
ever,areusuallylackinginpredictionabilityandoftendonot
provide a convenient metric by which to gauge system status.
As an alternative, investigators have developed several whole-
system indices of an ad hoc nature, for example, the popular
anduseful indexof bioticintegrity (IBI)ofKarr etal.(1986).
More suitable would be indices that are founded upon solid
theoretical considerations. As the subject of this chapter is
ecological networks, it is only appropriate to ask whether
some quantitative attribute of ecosystem networks might be
appropriatewithwhichtogaugeecosystemperformance.
Beforeembarkinguponthequestforwhole-systemmetrics,
itshouldberemarkedthatecosystemsaresignificantlyaffected
by what is missing, very often as much as by what is present
(Ulanowicz et al., 2009). Whether or not a given predator is
missing in a habitat will have great bearing on whether a
particular preywillthrivethere. Thesignificanceofabsenceis
mentioned, because most treatments of ecosystem behaviors
derive from methodology created for physics and chemistry.
Bateson(1972)pointedout,however,thatphysicstreatsover-
whelmingly only that which is present. Very little in physics
addresses that which is missing. (Rare exceptions include the
PauliexclusionprincipleandHeisenbergsuncertainty.)
Theapproachtoquantifyingthatwhichismissingiscalled
information theory (IT). Conventional descriptions of
information theory usually begin with the rubrics of commu-
nication. IT, however, has applications far beyond
communications theory, and can be invoked to measure the
relative degrees of constraint and flexibility inherent in a
system.
Thatinformationtheorybeginsbyquantifyingthatwhichis
missing is evident in Boltzmanns famous definition of
surprisal:
s klog p
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
Filter
feeders
1
Predators
5
Deposited
detritus
6
Deposit
feeders
4
Micro-biota
2
Meio-fauna
3
(a)
41.47
14.76
5.76 3.58
10.44
0.51 0.05
0.16
15.76
.
0.30
.
6.18
6.16
.
0.43
.
3.58
0.43
. .
(b)
Filter
Predators
feeders
5
1
0.17
0.17
0.64
1.99
0.66
3.69
0.78
Micro-biota Meio-fauna
2 3
1.21
Deposited Deposit
detritus 1.91 feeders
6 4
4.24
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 47
Figure9 (a)TheresidualthroughflowintheOysterReefecosystem(Figure2).(b)Theaggregatecycledflowsinthesamesystem.
where s is ones surprisal at seeing an event that occurs with
probabilityp, and kisanappropriate(positive)scalarconstant.
Becausetheprobability,p,fallsbetween0and1,wemaycon-
cludethatthenegativesigninthedefinitionisamathematical
conveniencethatguaranteesthatswillremainpositive(andsuch
mayhavebeenBoltzmannsmotivation).Howeverintermsof
sheerlogic,thisequationclearlydefinesstomeasurethedegree
to which p is not. That is, if p is the weight one gives to the
presenceofsomething,thensbecomesameasureofitsabsence.
(Here the reader might ask why the absence of event i is not
representedmoredirectlyby(1p
i
).Theadvantageandnecessity
ofusingthelogarithmwillsoonbecomeapparent.)Ifpisvery
small,thentheensuinglargemagnitudeofsreflectsthecircum-
stancethatmostofthetimewedonotseetheeventinquestion.
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
43229.2
22774
14639.2 30.
2.1
20.
4.9
25. 7.5 7.5
5. 6. 7. blue
fish
0.3
1. hetero micro croaker fish
phyto
free
larvae
2.5
microflag zooplank
plankton
bacteria
3439
7555
579.8 6157
21.
34.
26.
14 50.3 31.
alewife&
DOC
170162
8.
hog
14
blue
weak
choker
zoo
9. 154.8
10.
447 herring
22.3 fish
plankton
4072
cteno sea
phore nettle
suspended
2293.1
35. 292.2
POC
2.
bac
21986 22.
97
27.
teria 32.
2
bay 0.9
105.4 spot 251.9 summer
anchovy 0.9
flounder
55
11.
12. 13.
other
suspen Mya oyster
feeders
28. 33.
23.
59 78.8
2.4
120289 striped
24062.9
men
white
19.8 2.4
sediment perch bass
haden
23528 16.
36.
137437
14.
15.
3. Macoma
POC other
bac Nereis spp.
poly
teria
chaetes
107.5 286.9
24. 22.6 29.
113.8
5013 293
shad
43.1 catfish
4. 19.
30209 18. 48.1
benthic 17.
crusta 352 blue
diatoms meio
deposit crab
fauna 967
feeders
137437 1711 23721
30209 4046 23176
219847
194.7
60.3
198
510.5 10.8
_ _ _ _
48 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
Figure10 [Link],D.,Ulanowicz,R.E.,
[Link].EcologicalMonographs59,329364.
Boltzmanns gift to science the feasibility of quantifying
[Link]
thecontributionoftheArabianmathematicianswhoinvented
thenumber [Link] upon Boltzmannsinvention to
demonstratethatITliterallyopensnewvistastowhichclassical
physics remains blind. Moreover, the interplay between pre-
senceandabsencebecomescrucialtowhetherasystemsurvives
ordisappears,thatis,[Link],itisthe
veryabsenceoforder(intheformofadiversity ofprocesses)
thatmakesitpossibleforasystemtopersist(sustainitself)over
thelongrun(Ulanowicz,2009c).
That the Boltzmann definition actually quantifies absence
provides an insight that few appreciate namely, that the
product of the measure of the presence of event i (p
i
) by the
magnitude of its absence (s
i
) yields a quantity that represents
theindeterminacy(h
i
)oftheeventinquestion,
h
i
kp
i
log p
i
3
When p
i
1, event i is almost certain, and h
i
0; then when
p
i
0, the event is almost surely absent, so that again h
i
0. It
is only for intermediate, less determinate values of p
i
that h
i
becomesappreciable,achievingitsmaximumatp
i
=(1/e).
Itishelpfultoreinterpreteqn[3]asitrelatestoevolutionary
[Link]
i
1,theeventinquestionis
almost constantly present and exhibits little change (h
i
0).
Conversely,whenp
i
0,theeventinquestionhasgreatpoten-
tialtochangematters(s
i
1),butithardlyeverappearsinthe
system dynamics (so that, again, h
i
0). It is only when p
i
is
intermediate thattheeventisbothpresent frequentlyenough
and has sufficient potential for change. Whence, h
i
represents
the capacity for event i to be a significant player in system
change or evolution. With regard to the entire ensemble of
events,onecanaggregatealltheindeterminacies:

H h
i
k p
i
log p
i
4
i i
to achieve a metric of the total capacity of the system to
undergo change. Whether such change will be coordinated or
mostly stochastic depends upon whether or not the various
eventsicanaffecteachotherandbyhowmuch.
Mostreaderswillrecognizeeqn[4]asthefamiliarShannon
Wiener index of diversity. The quest for measures of whole-
ecosystem status was provided major impetus by MacArthur
(1955) when he applied Shannons information measure to
quantifythediversityofflows,H,inanecosystemnetwork:

T
ij
T
ij
H k log
T
::
T
::
i;j
wherekisascalarconstant,andT
..
denotesthesumofT
ij
over
[Link],thep
i
ineqn[4]becomes
theprobabilitythatarandomlyselectedquantumofmedium
isintransitbetweencompartmentsiandj,anditisestimated
bythequotient(T
ij
/T
..
).
MacArthursgoalwastoprobeforaconnectionbetweenthe
diversity of a systems flows, as measured by eqn [4],and the
[Link]
of efforts to link a systems diversity with its stability
(WoodwellandSmith,1969)thatdominatedsystemsecology
duringthe1960s.
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 49
Not long after MacArthurs effort, attention shifted away
from the difficult-to-measure flows to the more accessible
stocksofbiomass,andbiologicaldiversity,D,wasrecastas
B
k

_ _
i
B
D log
i

B
:
B
i
_ _
:
whereB
i
isthebiomass(ornumber)incompartmenti,andB
.
isthesumofalltheB
i
.
Unfortunately, thefervortoestablish atheoretical connec-
tionbetweenthetwosystemsattributescametoasuddenhalt
as soon as May (1972) demonstrated that in randomly
assembledlineardynamicalsystems,ahigherbiodiversitywas
[Link]-
ment, most ecologists immediately foreswore any further
efforts to apply information theory to ecology a prejudice
that lingers among many ecologists today. As a consequence,
when Rutledge et al. (1976) returned to MacArthurs original
focusuponflows,andreinterpretedtheirdiversityinaBayesian
manner, hardly anyone was paying attention. Rutledge et al.
(1976) were able to decompose MacArthurs index into two
complementary terms using the notion of conditional prob-
ability. Taking (T
ij
/T
..
) as the estimate of the unconditional
probabilitythataflowoccursfromitoj, (T
ij
/T
.j
)thenbecomes
theestimatoroftheconditionalprobabilitythatanyquantum
of flow continues on to compartment j, given that it had
[Link]
as
HAMIH
c
5
where

_
T
ij
_ _
T
ij
T
AMIk log
::
T
i;j
::
T
i:
T
:j
_
and
_ _
T
T
2

ij

ij
H
c
k log
T
i;j
::
T
i:
T
:j
_ _
AMI is called the average mutual information inherent in the
flowstructure,whileH
c
istheresidual(conditional)diversity/
freedom (commonly known as the conditional entropy in
informationtheory).Botharenon-negative.
Itisimportanttostressthatinformationisnotbeingused
hereintheusualcontextofcommunicationstheory,butrather
in the more generalized sense of constraint. In that context,
MacArthurs overall complexity of flow structure has been
decomposed into a component that gauges how orderly and
coherentlytheflowsareconnected(constrained)andonethat
measurestheresidualdisorderand/orfreedom.
Rutledgeetal.(1976)focused uponH
c
asamore didactic
measure of ecosystem maturity (sensu Odum, 1969) than
[Link](1980),however,
becamemoreinterestedintheAMI,becausehewasimpressed
by Atlans (1974) argument that AMI can provide a useful
measure of the organization inherent in a system. Ulanowicz
suggestedthatAMIisanindicatorofthedevelopmentalstatus
of anecosystem, because it reflects the average degreeof con-
straint that channels an arbitrary quantum of currency from
any one compartment to the next (Ulanowicz, 1997; Latham
andScully,2002).
Unfortunately,[Link]
particularvalueoftheAMI,onecannottellwhethertheindexis
representing a microbial community in a Petri dish or an
ecological community on the Serengeti Plain. Tribus and
McIrvine (1971) had commented that the lack of physical
dimensions on information measures gives rise to ambiguity
about their importance. As a remedy, they suggested that the
scalarconstant,k(usuallysetequaltooneandthenforgotten),
becomes the vehicle that imparts physical dimensions to the
[Link],Ulanowiczsetk=T
..
(thetotalsystemthroughput)andnamedtheresultingproduct
thesystemnetworkascendency,A,where
A

T
ij
T
T
::
ij
log
i;j
_
T
i:
T
:j
_
Ascendency was formulated to combine the total activity, or
powergeneratedbythesystem(T
..
),withitsorganizationinthe
senseofhoweffectivelycomponentprocessesarelinked(AMI)
(LathamandScully,2002).(Thealternativespellingascendency
is used here to distinguish the quantitative measure from the
conventionalmeaningofascendancy.)Asaresult,itquantifies
[Link]
easiesttothinkofascendencyasorganizedpower,becausethe
measure represents power that is flowing within the system
towardparticularends,asdistinctfrompowerthatisdissipated
willy-nilly (Ulanowicz, 2009b). Almost half a century earlier,
Lotka (1922) had suggested that the capacity of a system to
prevail in evolution is related to its ability to capture useful
power. Ascendency can be regarded as a refinement of Lotkas
suppositionthattakesintoaccounthowpowerisactuallybeing
channeledwithinasystem.InFigure11(a),forexample,poweris
exchangedinrelativelyconfusedfashionamongthecomponents
(nodes)[Link],inFigure11(b),theflows
out of or into any given compartment are more selective and
determinate (constrained). In addition, the magnitudes of the
flowsaregreater(asindicatedbythickerarrows).Theincreased
powerflowinginFigure11(b)isrealizedinamoreorganized
way;whencetheascendencyofnetworkbexceedsthatofa.
When he first created the ascendency measure, Ulanowicz
(1980)wasthinkingthatanecosystemwoulddevelopsoasto
maximize its ascendency. He later abandoned that presump-
tion as being too mechanical and deterministic (Mueller and
Leupelt,1998).Henowreferstothepropensityforecosystems
toincreaseinascendency(Ulanowicz,1997).
(a) (b)
Figure11 (a)Anetworkofflowshavingarelativelyconfusedpatternof
transfers.(b)Anothernetworkwheretheflowsofpoweraremoredeter-
minate(greaterascendency).Reproducedwithpermissionfrom
Ulanowicz,R.E.,[Link]:NaturalLifebeyondNewtonand
[Link],WestConshohocken,PA.
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
50 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
One may impart physical dimensions to H
c
(the comple-
mentofAMI)byalsomultiplyingitby(T
..
).Theresultiswhat
UlanowiczandNorden(1990)calledthesystemoverhead,:
T
2

ij

;
_ _

T
ij
log
T
i
i j
:
T
:j
Infact,HitselfcanbescaledbyT
..
tocreatewhatistermedthe
systemsdevelopmentcapacity,C:
C

T
ij
log
i;j
_
T
ij
_
6
T
::
Accordingly,thescaledversionofrelationshipeqn[5]becomes
C A 7
Thecomplementaryrelationshipofthetermsineqn[7] suggests
thatanyincreaseinascendencyusuallycomesattheexpenseof
overhead,.Oneisledtoaskwhetherthereareanylimitsasto
how far ascendency can displace system overhead without
endangering system persistence. Systems with high ascendency
that retain little overhead appear rigidly linked and almost
[Link](1986)callssuchsystemsbrittleandvul-
nerable tocollapse. Such vulnerability proceedsfroma lackof
sufficient freedom and flexibility (overhead) remaining to the
[Link]
orintegrityofanecosystemrequiresadequateamountsofboth
ascendencyandoverhead(Ulanowicz,2000a).
Tohelpgaugewhereabalanceofconstraintandflexibility
mightlie,onenotesthat
0A
C1
The ratio a=A/C, therefore, becomes a convenient (and nor-
malized) measure of the degree of system order. The actual
pattern of order in any instance is the result of two opposing
tendencies (Ulanowicz, 2009a). In an inchoate system
(onewithlowa),therearemanifoldopportunitiesforselective
constraints(suchasautocatalyticcycles)[Link]
for atoincrease persistsoverall [Link]-
head, , however, changes over the range of a. As just
mentioned, in systems with little order (low a), the random-
ness inherent in is what provides the opportunities for
constraintstoappear(Ulanowicz,2009b).Insystemsthatare
alreadyhighlydeveloped(a1),however,thedominantaction
ofistodisruptestablishedconstraints,oftenresultinginthe
sudden loss of organized performance (the system resets to
muchalowera).Athigha,then,stronglyopposesadditional
[Link],acriticalbalancebetweenthecoun-
tervailingrolesofexistsforsomevalueofa.
Itisimportanttostressthatecologicalrealityisnotsimply
[Link]
opposition between order-generating tendencies and dissipa-
[Link],inefficiencies,andincoherenciesplay
aslargearoleinhowecosystemsbehaveasdoconstraintsand
efficientperformance.
Exactly how this agonism plays out and where the balance
betweenAandliesaremattersofboththeoreticalandpractical
[Link]
(2009c). As for actual data on how ecosystems are distributed
over the domain of a, quantified networks of real ecosystems
0.2
0
1
0.8
e
s
s

0.6
n
t
i
F0.4
0 0.2 0.4 0.6 0.8 1
Degreeoforder
Figure12 Thedegreesoforderandcorrespondingmagnitudesof
fitnessforthesubsetof17ecosystemflownetworksthatconsistofmore
[Link],R.E.,[Link]:Heat
deathorperpetualharmonies?InternationalJournalofDesignandNature
andEcodynamics42(2),8396.
scatteracrosstheinterval0<a<1,skewedslightlytowardhigher
values,butwithanotableconcentrationbetweena=0.3and0.5
(Ulanowicz, 2009c). Robert Christian (personal communica-
tion) noted that the outlying systems (those closer to either 0
or 1) are all represented by networks of small dimension.
Wheneverecosystemswereparsedinmoredetail(say,n >12),
however,theirvaluesofawouldjointhecluster.
Figure12isaplotofdataon17ecosystems(mostlyfrom
estuarineandcoastalhabitats).Theordinateisahypothesized,
normalizedmeasureofthefitnessforanecosystemtopersist,F,
andisdefinedasF=ealn(a),whereeisthenaturalbasefor
logarithms(seeUlanowicz,2009c).
An implication that can be drawn from Figure 12 is that
networks distant from the maximum at a=(1/e) are not sus-
tainable. They either have too little organization (a small) or
aretooinflexible(alarge.)Inorderforanecosystemtopersist,
itshouldexhibitavalueofaintheneighborhoodofmaximum
[Link],ifasystemisnotatthemax-
imum,howcanitbedirected towardthemaximumthatis,
towardamoresustainableconfiguration?Thisquestioncanbe
answered by defining the robustness, R, of a system as the
productofitsfitness,F,anditstotalactivity(T..):
R T::F
Now,takingthederivativesoftherobustness,R,withrespectto
eachflow,T
ij
,yieldsthemarginalcontributionofeachunitof
flowT
ij
towardthesustainabilityofthesystem:
R F

F T
T
::
ij
T
ij
R
F

a
F T
T
::
ij

T
ij
T T
T
2
R T F

T
ij
F
::
_
_
ij ::
log

ij
alog
C T
i:
T
: j
_
T
i:
T
:j
__
WhereFisthederivativeofFwithrespecttoa,thatis,
_
log a
F

e 1
log e
_
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 51
OnenoticesthatbecauseF=1andF= 0 at a=(1/e),themar-
ginal contribution of each T
ij
at the maximum is 1. If a
particular derivative is less than 1.0, then the system would
bemovedinthedirectionofthemaximumbydecreasingthat
flow. Conversely, if the marginal contribution of T
ij
is greater
than one, systemic sustainability should be improved by
increasingthemagnitudeofthatarc.
As an example, the derivatives of the robustness are calcu-
lated for each flow in the Oyster Reef network of Figure 2 as
depicted in Figure 13. Because a=0.436, the system is only
slightlybeyondthemaximum,andincrementalchangesinthe
various flows would move the system to the left toward the
maximuminFigure12.
Amoredidacticdemonstrationoftheapplicationofthese
vectors toward sustainability would be to hypothetically
inflate primary production in the system and shunt all the
extra production through the detritus on to the microbiota,
where it is respired out of the system that is, it makes the
[Link]
value of 0.687. The corresponding values of the marginal
contributions are given in parentheses below the values of
[Link] seesfromthosemarginalvalues
thateachlinkintheprimaryrouteofeutrophicationpossesses
negative marginal values (especially the one for primary
production).Mostoftheotherflowsnowhavepositivemar-
ginal values, and some, such as feeding by predators and
deposit feeders, are quite significant. According to these
results,remediationshouldentaildepressingprimaryproduc-
tion in combination with channeling material away from
the eutrophication pathway and into the consumer ecosys-
[Link]
for eutrophic estuaries (Newell, 1988; Ulanowicz and
Tuttle,1992).
10.44
(1.03)
0.05
(0.675)
5.76
.
(0.997)
0.17
(1.05)
Filter
feeders
1
Predators
5
Deposited
detritus
6
Deposit
feeders
4
Micro-biota
2
Meio-fauna
3
41.47
14.76
3.58
(0.917)
0.51
(1.08) (0.879)
0.33
15.76

0.30
(1.00)
.
(0.885)
.
(0.949)
(0.829)
0.64
(0.831)
1.91
(1.01)
4.24
8.17 6.16
.
0.43
0.66
7.27
(0.985)
.
(1.09)
(0.885) (0.726) (0.955)
(1.07)
1.21
(1.06)
1.21
(0.906)
.
It should be clear that the relative values of A and are
significantindicators ofthestatus ofanecosystem. Forexam-
ple, a key property of flow networks is their degree of
connectivity,orhowmanylinks,onaverage,flowintoorout
of a typical node (sometimes called the link density).
Ulanowicz and Wolff (1991) demonstrated that a convenient
measure of the effective link density is achieved by raising e
(thenaturalbaseoflogarithms)tothepower(H
c
/2).Theywere
able to demonstrate that this index, which is valid for any
weighteddigraph,convergestointuitiveintegervaluesforsim-
ple examples. Bersier et al. (2002) saw major significance in
how properties of weighted digraphs can generalize those of
binary topologies. They used the link-density relationship to
bridge quantitative network analysis with the more popular
food-web analysis that deals only in binary networks. For
eachandeveryindexthathasbeendefinedinfood-webanaly-
sis,[Link]-theoretic
weighted homolog. Of special importance, they were able to
demonstratethatthemeasurespertainingtoweighteddigraphs
convergemore rapidlyto theintendedproperties than dothe
original,adhocfood-webmeasures.
ZorachandUlanowicz(2003)havebeenabletopursuethe
connectionbetweenH
c
andlinkdensitytoevengreaterdepth.
They were able to demonstrate that raising e to the power
(H
c
/2) yields precisely the weighted geometric mean of the
link density that one calculates using conventional algebra.
Followingdimensionalconsiderations,theyinferredthatrais-
ing e to the power AMI should provide an estimate of the
number of trophic roles in the network (i.e., it should corre-
spondroughlytotheeffectivenumberoftrophiclevels,orthe
trophicdepthofthenetwork).
AlthoughAandquantifythestatusofthewholenetwork,
oneisofteninterestedindetailsabouthowsubgroupsoftaxa
Figure13 Thevaluesofthemarginalcontributions(inparentheses)oftheflowsintheOysterReefcommunity(Figure2).
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
10.44
(1.14)
0.05
141.47 0.51
(0.49) (3.46)
(4.48)
115.76
0.30
.
0.17
(0.28) (4.49) (0.35)
108.17 0.66
(0.55) (0.594)
.
1.21
(5.05)
.
Filter
feeders
1
Predators
5
Deposit
feeders
4
Meio-fauna
3
.
0.33
(4.18)
14.76

(4.27)
0.64
(4.75)
Deposited
detritus
1.91
6
(2.39)
4.24
6.16
.
0.43
7.27
(2.54)
.
(5.93)
(5.71)
(1.37)
1.21
(3.19)
Micro-biota
2
105.76 3.58
(0.39) (3.01)
_ _
52 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
Figure14 HypotheticaleutrophicationintheOysterReefcommunity,[Link]
marginalcontributionsappearinparentheses.
[Link]-
tions,[Link],one
maybeinterestedinonlytheinternalworkingsofthesystem.
To focus on the inside of the system, one defines an internal
ascendency,A
I
,thatisconfinedsolelytotheinternalexchanges
betweenthensystemcomponents:
n

T
ij
T
::
A
I
T
ij
log
T
i:
T
:j
i;j1
ThefractionofthedevelopmentcapacitythatiscomprisedbyA
I
isthenpurportedtobeanindexoftheinternaldevelopmentof
thegiven [Link] has beenusedtocompare eco-
systemsthatweresimilarlyparsed(WulffandUlanowicz,1989;
Bairdetal.,1991;Christianetal.,2003;Bairdetal.,2004).
Likewise,itissometimeshelpfultoknowthecontributions
[Link]
noted earlier, there are four basic categories of flow in ecosys-
tems:internalexchanges,exogenousinputs,usefuloutputs,and
dissipations. Accordingly, the overhead separates neatly into
fourrespectivecomponents(UlanowiczandNorden,1990):
_ _
n n
T
ij
log
ij
T
i:
T
:j
T
2
i1 j1
_ _
n

I
T
0j
log
0j
T
0:
T
:j

T
2
j1
_ _
n
i;n1

T
2

E
T
i;n1
log
i1
T
i:
T
:;n1
_ _
n

T
2
i;n2

D
T
i;n2
log
i1
T
i:
T
:;n2
where is called the redundancy of flows in the system,
becauseitprovidesameasureofthemultiplicityof,orparalle-
lism in, flow pathways between any two arbitrary system
components.
I
is the overhead due to imports, and
E
and

D
aretheoverheadsthatcanbetracedtoexportsanddissipa-
tions, respectively. Often, disturbances will increase the ratio
/C(Ulanowicz, 1984), indicating that theresponse ofa sys-
temtodisturbanceusuallyincreasesitsabilitytoresistfurther
perturbations (the LeChtelierBraun principle as applied to
ecology). Heymans (2003) and Heymans et al. (2007)
employed the internal overhead to gauge the potential of a
systemforadaptationandresilience.
One criticism of network analysis in general is that all
attentionseemstobefocusedonflowstotheneglectofstocks.
It has been argued here that ecosystem dynamics depend lar-
gelyontheconfigurationsofflowswithinthem,buttheremost
certainly remain situations where the magnitudes of stocks
affect system dynamics. In order to introduce stock size into
theinformationdynamics,thestockincompartmentiwillbe
denotedbyB
i
.TorelateB
i
toflowsintoandoutofi,onenotes
that the a priori probability that any quantum of medium in
transithasoriginatedfromcompartmentiis(B
i
/B
.
).Similarly,
the corresponding (unconditional) probability that a particle
offlowingmediumwillenterjis(B
j
/B
.
).Thus,theapriorijoint
probability that any quantum of flowing medium is en route
from i to j becomes (B
i
B
j
/B
.
2
). The observed, or a posteriori
frequency ofsuchflowfrom itojismeasuredas(T
ij
/T
..
).The
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy

1
AMI
3

2

3
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 53
Kullback (1959) index can be employed to calculate the
average difference between these a priori and a posteriori prob-
abilities,I
B
, to be
T T B
2

I
B

ij
k

_
ij
log
:
T
::
_ _
T B
i
:: i
B
j
;j
_
Aswiththeflowascendency,onemayscaleI
B
byT
..
toyieldthe
biomass-inclusiveformoftheascendency,A
B
(Ulanowiczand
Abarca-Arenas,1997),as

ij
B
A
ij
log
:
B

T
2
T

T B
j
i
::
B
i
;j
_ _
Howtheoverallsystemstatusmightchangewithinfinitesimal
additionstothebiomassoftaxonp,B
p
,isreckonedbytaking
thepartialderivativeofA
B
withrespecttoB
p
:
A
B

2
B
p
_
T
::
1 T

:p
T
p:

B
:
2 B
p
_
Examinationofthisresultshowsthatthesensitivityispropor-
tional to the amount by which the average turnover rate of
mediumincompartmentpisexceededbytheoverallturnover
rate (T
..
/B
.
). The system appears to be more sensitive to those
componentshavingslowerturnoverrates.
One might ask further, Which prey resource, r, is most
important as a source of medium to predator p? As with the
biomasses, thelimiting source canbeidentified bytaking the
partialderivativeofA
B
withrespecttoeachprey,r,oftaxonp.
Because A
B
is a first-order homogeneous function in the T
ij
(Courant,1936:108),onemayinvokeEulersrelationtoarrive
atthesimpleresult:
A
B
T
rp
B
2
log
:
T
rp

_
T
::
B
r
B
p
_
Accordingtothisformula,onearrivesattheintuitivelysatisfy-
ing conclusion that the controlling resource is the one that is
being depleted of medium at the fastest rate relative to its
available remaining stock (i.e., the one with the highest ratio
(T
rp
/B
r
)).Inthenextsection,thesesensitivitieswillbeusedto
identify points of control by various nutrients in an estuarine
ecosystem.
9.04.8 HigherDimensionalConsiderations
Theutilityofthesensitivitiesjustpresentedwouldbeenhanced
if they could be used to discriminate among the dynamics of
[Link],untilnow
networkanalysishasbeenappliedonlytoasinglesnapshotof
the system. It is not difficult, however, to depict temporal
ecosystem dynamics as a motion picture or a time series of
such network snapshots. Nor is it far-fetched to envision a
tessellation of networks, each pertaining to a small segment
of the physical space occupied by the ecosystem. While it
remains difficult in general to extend the IO, trophic and
cycle analyses to encompass heterogeneities in time, space,
and media, widening the scope of information-theoretic
indicespresentsnoconceptualobstacles.
If, for example, T
ijk
denotes the flow of medium from
taxon i to taxon j during time interval k, then the temporal
capacity, C
T
, immediately follows as a straightforward gener-
alization of eqn [6]:

T
C
T
T
ijk
log
_
ijk
T
i

;j;k
_
Indefiningthecorresponding temporalascendency,however,
one must choose among several possibilities. In schematic
terms, C
T
can be decomposed into numerous terms, as sug-
gested by Venn diagrams (Blachman, 1961). In two
dimensionsthesituationisunambiguous(Figure15).
TheleftcircleinFigure15representsthediversityofsource
flows (from prey), whereasthe right one represents the diver-
sity of sinks (to predators). The overlap between the circles
[Link]-
prisetheconditionaluncertainty,H
c
.
The overlaps in three dimensions are multiple, however.
Figure 16 consists of the pattern of Figure 15 repeated 3
times. If the three circles represent the diversities of sources,
sinks,andtime,thentherearethreebilateralregionsofoverlap,
representing the mutual information between sources and
sinks, sources and time, and sinks and time, respectively. In
addition,however,theseoverlapsthemselvesintersect,yielding
atleastfourdistinctregions:thereisthecentralareawhereall
threedomainsintersect(AMI
3
)andthethreeadjoiningregions
where only a single pair each overlaps (
1
,
2
, and
3
in the
figure). As to which region should represent the information
factor in three-dimensional (3D) ascendency, one could
choose the innermost circular triangle (AMI
3
; which is desig-
natedasthe3DAMIininformationtheory),butthisformula
hasthedisadvantageofsometimesbeingnegativeinvalue.
In terms of the T
ijk
, AMI
3
can be calculated (Abramson,
1963) as
H
c1 AMI
H
c2
Figure15 2DVenndiagramshowingtheAMIastheintersectionofthe
[Link],
H
c
=H
c1
+H
c2
.
Figure16 Theintersectionofthethreedomainsofsources,sinks,and
[Link](AMI
3
)representsthe3Daveragemutual
information,butthethreeareasofbilateraloverlap(
1
,
2
,
3
)also
contributetotheoverallsystemorganization.
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
_ _
_ _
_ _
_ _ __
_ _
54 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
_ _
T
2
T
T
ijk ijk
AMI
3
log
T

T
i::
T
:j:
T
::k
i;j;k
It represents the simultaneous coherence among sources and
[Link],however,encompassallthecoher-
[Link],theleaflets(
1
,
2
,
3
)
that adjoin AMI
3
represent bilateral coherencies that do not
correlate withthe thirdremaining variable. To beinclusive of
allcoherenciesinthesystem,Pahl-Wostl(1995)suggestedthat
the appropriate measure is the overlap of all bilateral AMIs
(i.e.,thetrefoil-likeregionAMI
3
+
1
+
2
+
3
).Callingthiscon-
junctionAMI
T
,onecalculatesitsmagnitudeintermsoftheT
ijk
as
_ _
T
2
T
T
ijk ijk
AMI
T
log
T

T
ij:
T
i:k:
T
:jk
i;j;k
UnlikeAMI
3
,AMI
T
[Link]
T
byT

yieldsthetemporalascendencyA
T
:
_

_
T
2
T

A
T
T
ijk
log
ijk
T
ij:
T
i:k:
T
:jk
i;j;k
Havingresolved3Dambiguities,theextensionofcapacityand
ascendency by induction into four or more dimensions
becomes relatively straightforward. Probably, the most inclu-
sivedomainanecologistislikelytoconsiderwouldconsistof
fivedimensionsdonor,recipient,time,space,andmedium.
Accordingly,transferT
ijklm
willrepresenttheflowfromtaxoni
totaxonjduringtimeintervalkatspatiallocationlofmedium
m.(Spaceisconsideredasonlyasingledimension,becauseany
2D and 3D partitioning of a finite domain of physical space
always can be enumerated by a single index (in the same
fashion that multi- dimensional arrays are stored as a single
sequencebycomputermachinecode).)Thecorresponding5D
ascendency,A
O
,wouldthenbe
_

_
T
4
ijklm
T
::
A
O
T
ijklm
log
T
:jklm
T
i:klm
T
ij:lm
T
ijk:m
T
ijkl:
i;j;k;l;m
This ascendency remains homogeneous (sensu Euler) in the
flows. As before, this property allows one to calculate readily
the5Dsensitivitiesas
_ _
T
4
A
O pqrst
T
::
log
T
pqrst
T
:qrst
T
p:rst
T
pq:st
T
pqr:t
T
pqrs:
Thecorrespondingbiomass-inclusiveoverallascendency,A
OB
,
thentakestheform

T
ijklm
B
2
:
A
OB
T
ijklm
log
B
iklm
B
jklm
T
::
i;j;k;l;m
whereB
iklm
isthebiomassofmediummintaxoniattimekin
[Link]
A
OB
T
::
1 T
p:rst
T:
prst
2 8
B
prst
B
:
2 B
prst
B
prst
and
A
OB
T
pqrst
B
:
2

log 9
T
pqrst
B
prst
B
qrst
T
::
Inthose cases for which dataare available onmultiplemedia,
Ulanowicz and Baird (1999) have demonstrated analytically
that the largest biomass sensitivity coefficients eqn [8] corre-
spond to that medium which is present in least proportionate
amount,thatis,inthesenseofvonLiebig(1854).Liebigs law,
however, provides no way of identifying the most limiting
source of that nutrient. The controlling source is given by the
largest value of eqn [9]. Ulanowicz and Baird (1999) applied
eqns[8] and [9]toparallelnetworksofseveralmediatoeluci-
dateanewandsystematicwayofdepictingnutrientdynamicsin
ecosystems. They assembled sufficient data to estimate the
trophic exchanges of carbon, nitrogen, and phosphorus (C, N,
andP)intheecosystemofthemesohalinereachofChesapeake
Bayoverthefourseasonsofatypicalyear(nospatialvariation).
They used the 4D version of eqn [8] to identify the limiting
mediumforeachtaxonandfoundthatNwaslimitingprimary
production during the summer season (as has long been
known). During that same season, however, P was controlling
bacterial production and secondary production by nekton
(mobilefishes).Applyingeqn[9]totheirnetworks,theinvesti-
gators were able to diagram (graphical) trees that depict the
hierarchyofnutrientcontrolsduringeachseason(Figure17).
The nutrient limitation analysis can be used as well for
purposes of comparison. For example, data at very similar
resolutiontothoseestimatedfortheChesapeakeareavailable
on the chief components of the Sylt-Romero Bight ecosystem
(DanielBaird,personalcommunication).TheanalysisofC,N,
and P for this system reveals that a very similar pattern of
nutrient controls is operating in both estuarine ecosystems,
although they are separated by thousands of kilometers and
[Link]
is limited by nitrogen, whereas most heterotrophs in both
habitats are controlled by phosphorus. There is more of a
tendency in the Sylt system for bottom dwellers to be starved
forN:crabsanddemersalfishappeartobecontrolledbyNin
theSylt,whiletheircounterpartsintheChesapeakearelimited
by carbon and phosphorus, respectively. Likewise, sediment
[Link]
is curious that the nutrient limitation of crabs is reversed the
nexttrophicleveldown,thatis,crabslackCintheChesapeake
andNintheSylt.
A
OB
was also applied by Ulanowicz (2000b) to animal
migration across a simple 1010 spatial grid. He simulated
(usingcellularautomata)thehypotheticalmigrationofpopu-
lation across this landscape to generate data to show how
informationtheoreticmeasurescouldbeappliedtolandscape
ecology. Kikawada (1998) extended Ulanowiczs analysis to
include simultaneous trophic interactions, and Krivov and
Ulanowicz (2003) addressed the most general configuration
of network dynamics pertaining to several chemical elements
distributed across a landscape. Of note to hydrodynamicists,
thecalculusdiscussedherecanevenbeappliedtoquantifythe
degreeoforganizationinherentinfluiddynamics(Ulanowicz
andZickel,2005).
9.04.9 SummaryandConclusions
Thenetworkperspectivewaspromptedbytheobservedshort-
comings of the mechanical worldview how mechanical
descriptions of coupled processes usually fail to mimic
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
6
5
Hetero
Frewe
micro
Bacteria
34 flagellates
DOP
4
Benthic
Diatoms
25
Croaker
3 15 27
36
Bacteria
Seciment
onSed
Nereis Spot
PON
PON
14 28
Other
White
Polychaetes
Prech
16 26
Macoma Hogchoker
18
Crust
29
Dep.
Catfish
Feeders
17 19
Meio Blue
fauna Crab
35
Susp
pop
11
Other
Susp
Feeders
7
8
Meso2000
Ciliates
plankton
12
Mya
1 13
34
Phylo
Oyster
DON plankton
2
Bacteria
onSusp
pop
9 10
Cteno Sea
phores Nettles
22
Bay
Anchovy
21
Aliwile&
Herring
23
Menhaden
24
Shed
20
Fish
Larvae
31 32
Weak Summer
fish Rounder
33
Striped
Bass
30
Blue
fish
Summer
Nitrogen
Phosphorus
Carbon
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 55
Figure17 [Link]
controlledbynitrogenareindicatedbysolidlines;thosecontrolledbyphosphorousbyshortdottedline;andthosebycarbonbylongdottedlines.
ecosystem behaviors. In the course of searching for ways
aroundthisobstacle,networktoolshavebeendevelopedthat
do not merely circumvent the problem, but actually open
entirely new perspectives on the process of life (Ulanowicz,
1997).
Inretrospect,thisoutcomeshouldnothavebeenasurprise.
An electron, for example, possesses its own internal nature
(SungchulJi,personalcommunication).Ifoneusesonesetof
apparatus to observe an electron, the entity will appear as a
particle with a certain mass. If one focuses a different set of
observationaltoolsuponanelectron,itwillappearratherasa
wave packet. As every physicist knows, the nature of the elec-
[Link]
observation.
Inamoregeneralway,thesamethingistrueofanecosys-
tem. If one approaches a living community with the tools of
quiescent material and laws, one can observe only the more
mechanical aspects of the system. If, however, one adopts a
process-oriented perspective (such as is afforded by network
analysis), then the dynamics of life can be observed more
directly, and the (restricted) role of noise becomes far more
[Link]
descriptionofwhattranspiresinlivingsystems.
Inparticular,livingsystemsnowappeartobetheoutcome
of the opposing tendencies toward order and dissipation, not
simply the unfolding of a mechanical scheme (Ulanowicz,
2009b).Thefocusinthenetworkperspectiveisuponcausality
via configurations of processes, rather than laws working on
inertmaterial.
That the lens of process affords a most exciting new phi-
losophical viewpoint on living systems does not at all detract
from the utilitarian value of network analysis. In the absence
of mechanical models, one can still say much, both qualita-
tive and quantitative, about how estuarine and coastal
systems are operating. One can use networks to identify
configurations that order dynamics, to locate bottlenecks in
the sequence of processes, to quantify the developmental
status of a particular system in relation to others, to trace
indirect effects, to evaluate trophic status, to assess how the
ecological community as a whole might be made more sus-
tainable, and so on.
Network analysisbelongs inthetoolbox ofevery manager
ofestuarineandcoastalecosystems.
Acknowledgment
The author is grateful to Michael Barfield for editorial
comments.
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
56 QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems
References
Abramson,N.,[Link]-Hill,NewYork,201pp.
Allesina,S.,Bodini,A.,Bondavalli,C.,[Link]:
theroleofstronglyconnectedcomponents.Oikos110(1),164176.
Allesina,S.,Bondavalli,C.,[Link]:a
comparisonbetweenbalancingprocedures.EcologicalModelling165,221229.
Allesina,S.,Ulanowicz,R.E.,[Link]:Finnsindex
revisited.ComputationalBiologyandChemistry28,227233.
Atlan,H.,[Link].JournalofTheoreticalBiology45,
295304.
Augustinovics,M.,[Link]
[Link]:Carter,A.P.,Brody,A.(Eds.),ContributionstoInputOutputAnalysis.
North-Holland,Amsterdam,vol.I,pp.249269.
Barabsi,A.-L.,[Link]:[Link],Cambridge,MA.
Baird,D.,Christian,R.R.,Peterson,C.H.,Johnson,G.A.,[Link]
hypoxiaonEstuarineecosystemfunction:energydiversionfromconsumersto
microbes.EcologicalApplications14(3),805822.
Baird,D.,McGlade,J.M.,Ulanowicz,R.E.,[Link]
ecosystems.PhilosophicalTransactionsoftheRoyalSocietyB333,1529.
Baird,D.,Ulanowicz,R.E.,[Link]
ecosystem.EcologicalMonographs59,329364.
Bateson,G.,[Link],NewYork.
Bersier,L.-F.,Banaek-Richter,C.,Cattin,M.-F.,[Link]
webmatrices.Ecology83,23942407.
Biswas,A.K.,[Link]-Hill,NewYork,
348pp.
Blachman,N.M.,[Link]
49(8),13311332.
Bondavalli,C.,Ulanowicz,R.E.,[Link]:
thecaseoftheAmericanalligator.Ecosystems2,4963.
Caswell,H.,[Link],Sunderland,MA,328pp.
Christensen,V.,Pauly,D.,[Link]-state
modelsandcalculatingnetworkcharacteristics.EcologicalModelling61,169185.
Christian,R.R.,Dame,J.K.,Johnson,G.,Peterson,C.H.,Baird,D.,[Link]
andModelingoftheNeuseRiverEstuary,Phase2:FunctionalAssessmentof
[Link]-WRRI2003-343-E,
DepartmentofBiology,EastCarolinaUniversity,Greenville,NC,111pp.
Christian,R.R.,Luczkovich,J.J.,[Link]
[Link]
117,99124.
Cohen,J.E.,Briand,F.,Newman,C.M.(Eds.),[Link]:Dataand
[Link],Heidleberg,308pp.
Courant,R.,[Link],[Link],NewYork,
681pp.
Cousins,S.,[Link].NewScientist107(1463),5054.
Dame,R.F.,Patten.,B.C.,[Link].
MarineEcologyProgressSeries5,115124.
DeAngelis,D.L.,Gross,L.J.(Eds.),[Link]-BasedModelsandApproachesin
[Link],London,525pp.
Fath,B.D.,Patten,B.C.,[Link]:emergenceofpositiverelationsin
ecologicalsystems.EcologicalModelling107,127143.
Fath,B.D.,Patten,B.C.,[Link].
Ecosystems2,167179.
Finn,J.T.,[Link]
flows.JournalofTheoreticalBiology56,363380.
Halfon,E.A.,[Link],NewYork,516pp.
Han,B.-P.,[Link].
EcologicalModelling104,289302.
Hannon,B.M.,[Link].JournalofTheoreticalBiology41,
535546.
Hannon,B.M.,Costanza,R.,Ulanowicz,R.E.,[Link]
ecologicalsystems:[Link]
PopulationBiology40,78104.
Heymans,J.J.,[Link]
[Link]:Heymans,J.J.(Ed.),Ecosystem
ModelsofNewfoundlandandSoutheasternLabrador(2J3KLNO):Additional
[Link]
Reports,TheUniversityofBritishColumbia,Vancouver,BC,vol.11(5),pp.6271.
Heymans,J.J.,Gunette,S.,VillyChristensen,V.,[Link]
indicatorsofecosystemstatusintheGulfofAlaska.Ecosystems10(3),488502.
Higashi,M.,Burns,T.D.(Eds.),[Link]:TheNetwork
[Link],Cambridge,364pp.
Higashi,M.,Patten,B.C.,Burns,T.D.,[Link]:anemerging
[Link]:Higashi,M.,Burns,T.D.(Eds.),Theoretical
StudiesofEcosystems:[Link],
Cambridge,pp.117154.
Hillebrand,H.,Shurin,J.B.,[Link]:Belgrano,
A.,Scharler,U.,Dunne,J.,Ulanowicz,R.E.(Eds.),AquaticFoodWebs:An
[Link],Oxford,ch.14,pp.184197.
Holling,C.S.,[Link]:localsurpriseandglobal
[Link]:Clark,W.C.,Munn,R.E.(Eds.),SustainableDevelopmentofthe
[Link],Cambridge,pp.292317.
Jassby,A.D.,Platt,T.C.,[Link]
photosynthesisandlightforphytoplankton.LimnologyandOceanography21(4),
540547.
Karr,J.R.,Fausch,K.D.,Angermeier,P.L.,Yant,P.R.,Schlosser,I.J.,[Link]
ofBiologicalIntegrityinRunningWater:[Link]
HistorySurvey,Publication5,Champaign,IL.
Kavanagh,P.,Newlands,N.,Christensen,V.,Pauly,D.,[Link]
optimizationforEcopathecosystemmodels.EcologicalModelling172,141149.
Kikawada,H.,[Link]
[Link],UniversityofMaryland,CollegePark,MD,181pp.
Knuth,D.E.,[Link]-Wesley,Reading,MA,vol.1,
228pp.
Krivov,S.,Ulanowicz,R.E.,[Link]-agent
systems.BioSystems69,3954.
Kullback,S.,[Link],Gloucester,MA,
399pp.
Latham,L.G.,Scully,E.P.,[Link]
ecologicalnetworks.EcologicalModelling154,2544.
Leontief,W.,[Link],19191939,Seconded.
OxfordUniversityPress,NewYork,257pp.
Levine,S.,[Link].
JournalofTheoreticalBiology83,195207.
Lewin,R.,[Link]?Science224,13271329.
Liebig,J.,[Link]
Walton,London,401pp.
Lotka,A.J.,[Link]
NationalAcademyofSciencesoftheUnitedStatesofAmerica8,147150.
MacArthur,R.,[Link],andameasureofcommunity
stability.Ecology36,533536.
Mateti,P.,Deo,N.,[Link]
JournalonScientificComputing5,9099.
Matis, J.H., Patten, B.C., 1981. Environ analysis of linear compartmental systems:
the static, time invariant case. Bulletin of the International Statistical Institute 48,
527565.
Matis,J.H.,Patten,B.C.,White,G.C.,[Link]
[Link],Fairland,MD,368pp.
May,R.M.,[Link]?Nature238,413414.
Mueller,F.,Leupelt,M.(Eds.),[Link],GoalFunctions,andOrientors.
Springer,Berlin,619pp.
Newell,R.I.E.,[Link]:aretheytheresultof
overharvestingtheEasternoyster(Crassostreavirginica)?In:Lynch,M.P.,Krome,
E.C.(Eds.),UnderstandingtheEstuary:AdvancesinChesapeakeBayResearch.
ChesapeakeResearchConsortiumPublication129(CBP/TRS24/88),Gloucester
Point,VA,pp.536546.
NSF,[Link]
99-14,NationalScienceFoundation,Ballston,VA.
Odum,E.P.,[Link].Science164,262270.
Odum,H.T.,[Link],[Link],NewYork,331pp.
Pahl-Wostl,C.,[Link]:ChaosandOrderEntwined.
Wiley,NewYork,267pp.
Patten,B.C.,[Link]
Science78,206222.
Patten,B.C.,[Link]:[Link]
Naturalist119,179219.
Patten,B.C.,Auble,G.T.,[Link]
Naturalist117,893922.
Patten,B.C.,Bosserman,R.W.,Finn,J.T.,Cale,W.G.,[Link]
[Link]:Patten,B.C.(Ed.),SystemsAnalysisandSimulationinEcology.
AcademicPress,NewYork,vol.4,pp.457579.
Pauly,D.,Christensen,V.,Dalsgaard,J.,Froese,R.,Torres,F.,[Link]
MarineFoodWebs.Science279(5352),860863.
Pimm,S.L.,[Link],London,219pp.
Pimm,S.L.,Lawton,J.H.,[Link].
Nature268,329331.
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9
Author's personal copy
QuantitativeMethodsforEcologicalNetworkAnalysisandItsApplicationtoCoastalEcosystems 57
Platt,T.C.,Mann,K.H.,Ulanowicz,R.E.(Eds.),[Link]
[Link],Paris,157pp.
Polis,G.,Winemiller,K.,[Link]:IntegrationofPatternsandDynamics.
ChapmanandHall,NewYork,472pp.
Polovina,J.J.,[Link]
Bulletin83(3),457460.
Rutledge,R.W.,Basorre,B.L.,Mulholland,R.J.,[Link]:an
informationtheoryviewpoint.JournalofTheoreticalBiology57,355371.
Sanderson,J.,Harris,L.D.(Eds.),[Link]:ATop-Down
[Link],BocaRaton,FL.
Simon,H.A.,Hawkins,D.,[Link].
Econometrica17,245248.
Szyrmer,J.,Ulanowicz,R.E.,[Link].EcologicalModelling35,
123136.
Tarjan,R.,[Link]
Computing1,146160.
Tribus,M.,McIrvine,E.C.,[Link].ScientificAmerican225,179188.
Ulanowicz,R.E.,[Link].
JournalofTheoreticalBiology85,223245.
Ulanowicz,R.E.,[Link]
Biosciences65,219237.
Ulanowicz,R.E.,[Link]
[Link]:Fasham,M.J.R.(Ed.),FlowsofEnergyandMaterialsinMarine
[Link],London,pp.2347.
Ulanowicz,R.E.,[Link]:EcosystemsPhenomenology.
Springer,NewYork,203pp.
Ulanowicz,R.E.,[Link]:[Link]:Patten,
B.C.,Jorgensen,S.E.(Eds.),ComplexEcology:ThePart-WholeRelationin
[Link],EnglewoodCliffs,NJ,pp.549560.
Ulanowicz,R.E.,[Link],[Link],
NewYork,201pp.
Ulanowicz,R.E.,[Link]:Pimentel,D.,
Westra,L.,Noss,R.F.(Eds.),EcologicalIntegrity:IntegratingEnvironment,
[Link],Washington,DC,pp.99113.
Ulanowicz,R.E.,[Link]
[Link]:Sanderson,J.,Harris,L.D.(Eds.),Seriesin
LandscapeEcology:[Link],BocaRaton,FL(Eds.),pp.113142.
Ulanowicz,R.E.,[Link].
ComputationalBiologyandChemistry28,321339.
Ulanowicz,R.E.,[Link]
220,18861892.
Ulanowicz,R.E.,[Link]:NaturalLifebeyondNewtonandDarwin.
TempletonFoundationPress,WestConshohocken,PA.
Ulanowicz,R.E.,[Link]:Heatdeathorperpetualharmonies?
InternationalJournalofDesignandNatureandEcodynamics42(2),8396.
Ulanowicz,R.E.,Abarca-Arenas,L.G.,[Link]
structureandfunction.EcologicalModelling95,110.
Ulanowicz, R.E., Baird, D., 1999. Nutrient controls on ecosystem dynamics:
The Chesapeake mesohaline community. Journal of Marine Systems 19,
159172.
Ulanowicz,R.E.,Goerner,S.J.,Lietaer,B.,Gomez,R.,[Link]:
resilience,efficiencyandthereturnofinformationtheory.EcologicalComplexity6,
2736.
Ulanowicz,R.E.,Kemp,W.M.,[Link]
Naturalist114,871883.
Ulanowicz,R.E.,Norden,J.S.,[Link].
InternationalJournalofSystemsScience1,429437.
Ulanowicz,R.E.,Puccia,C.J.,[Link].Coenoses5,
716.
Ulanowicz,R.E.,Tuttle,J.H.,[Link]
rehabilitationinChesapeakeBay.Estuaries15,298306.
Ulanowicz,R.E.,Scharler,U.M.,[Link]
networksoftrophicflows.EcologicalModelling210,278286.
Ulanowicz,R.E.,Wolff,W.F.,[Link]:loadeddice?Mathematical
Biosciences103,4568.
Ulanowicz,R.E.,Zickel,M.J.,[Link]:Kleidon,
A.M.,Lorenz,R.D.(Eds.),Non-EquilibriumThermodynamicsandtheProductionof
[Link],Berlin,pp.5667.
Vezina, A.F., Platt, T.C., 1988. Food web dynamics in the ocean. I. Best-estimates
of flow networks using inverse methods. Marine Ecology Progress Series 42,
269287.
Wang,H.V.,Johnson,B.J.,[Link]
[Link]
EcosystemModeling1,5190.
Whipple, S.J., Patten, B.C., 1993. The problem of nontrophic processes in trophic
ecology: a network unfolding solution. Journal of Theoretical Biology 163,
393411.
Woodwell,G.M.,Smith,H.H.(Eds.),[Link]
[Link],NewYork,vol.22,264pp.
Wulff,F.,Ulanowicz,R.E.,[Link]
[Link]:Wulff,F.,Field,J.G.,Mann,K.H.(Eds.),Flow
[Link],Berlin,pp.232256.
Yodzis, P., 1989. Patterns in food webs. Trends in Ecology and Evolution 4 (2),
4950.
Zorach,A.C.,Ulanowicz,R.E.,[Link]:How
manyrolesarethere?Complexity8(3),6876.
Treatise on Estuarine and Coastal Science, 2011, Vol .9, 35-57, DOI: 10.1016/B978-0-12-374711-2.00904-9