Transcript Food web analysis with ecopath/ecosim

Introduction to multispecies
modeling using
Ecopath with Ecosim
Software for examining impacts of
trophic interactions and harvesting on
multispecies biomass patterns in
Ecosystems
Approaches to
multispecies/ecosystem modeling
• Static community “types” modeling in relation to
habitat factors
• Empirical assessments of linkage “strength”
(bottom up vs top down effects, etc)
• Single species models with mortality rate crosslinkages (MSVPA,MSFOR)
• Biomass/numbers trophic interaction models
(Ecopath/Ecosim)
• Agent-based models with arbitrary components
and linkages (ATLANTIS)
• Complex, spatial multispecies IBMs
Multispecies fisheries issues
• Almost all fisheries are “multispecies” in
sense that gear takes more than one
species
• Usual to classify multispecies fisheries
interactions as:
– “Technical”—fishing targeted at one or more
species usually takes others as bycatch
– “Ecological”—fishery appropriates production
that could otherwise go to higher trophic level
species, but frees up production of species at
lower trophic levels
Trophic level
Species can be positioned in food webs
by mean trophic level of food consumed
4
Piscivores
3
Planktivores and
benthivores
2
Invertebrates
• Trophic level pattern creates possibility of
“trophic cascades”: decrease at 4 causes
increase at 3 causes decrease at 2, etc.
• But web structure weakens such effects
Mixed technical and trophic interactions
Fishery 1
Trophic level
4
Fishery 2
3
Fishery 3
2
• Fishery 1 creates value tradeoffs among species at
TL 4, and increase in surplus production at TL 3
• Fisheries 2, 3 may “rob” surplus production from
Fishery 1
• Fishery 2 may deplete both the piscivore(s) and
their prey
Trophic level
Trophic ontogeny creates risk of
“cultivation-depensation” effects
4
Piscivores
3
Planktivores and
benthivores
2
Invertebrates
• Piscivores typically increase in trophic level as they grow
• High abundance of adult piscivores causes reduction in
abundance of lower TL species that compete with and
prey upon piscivore juveniles
• Reduction in adult piscivores can lead to reduced
survival of its juveniles, lower equilibrium dominated by
TL 2-3 species (examples: bass-bluegill, cod-herring)
Example fisheries questions
involving food web interactions
• Does the Atlantic menhaden fishery impact production of
striped bass?
• Have Bering Sea fisheries caused collapse of Stellar sea
lions?
• Are seals and sea lions causing collapse of Pacific
salmon populations?
• Will bycatch reduction devices in Gulf of Mexico shrimp
fisheries result in decreased shrimp production due to
increases in shrimp-eating fish that have been fished
down by trawling?
• Is food production limiting to rainbow trout in Grand
Canyon, and do rainbow trout have strong negative
impact on native fish recruitment?
Ecopath with Ecosim (EwE) is the only
general modeling software now available
for dealing with such questions
• Can be parameterized with relatively
simple data
• Can represent complex food web
structure, multiple fisheries with complex
targeting and bycatch impacts
• Incorporates “foraging arena” behavioral
effects in trophic interaction predictions
• Allows representation of trophic ontogeny
and cultivation-depensation risks
EwE main components and data
entry/analysis steps
• ECOPATH—Enter basic biomass and trophic
interaction information, output is set of
biomasses Bi and “flows” Qij from biomass pools
i to j.
• ECOSIM—Predict biomass (and age structure)
changes over time starting from Ecopath base
values
• ECOSPACE—Predict biomass changes over
spatial grid of habitat cells over time, starting
with ECOSIM parameter estimates
Think about biomasses Bi and
flows Qij between them as a web
Catch
• Qij is a rate (biomass
per year)
• Total consumption by a
predator is a sum of
Qs, eg Q24+Q34
• Prey mortality rate on
any link is Mij=Qij/Bi
• Fishing rate is Fi=Ci/Bi
C5
B5
B4
Q34
Q35
B3
Q24
B2
Q13
Q12
B1
ECOPATH MODEL
DEVELOPMENT
• Define a set of biomass “pools” or functional
groups that are of policy interest
• Enter estimates of three of the following four
quantities for each group:
–
–
–
–
Bi-biomass per area
PBi- annual production per biomass
QBi-annual consumption per biomass
EEi-proportion of production explained by modeled
predation and harvest
• Enter estimates of DCij, proportions of species i
in diets of species j (food web structure)
Ecopath “solution”
• Given 3 of the 4 basic inputs for each group i, and
the diet composition matrix DCij, Ecopath solves for
the fourth quantity for each i, and for the trophic flows
Qij
• Qij=QBj x Bj x DCij (Total consumption by j x DCij)
•
•
(Consumption of j  Catch of j)
EEj 
(T otalMortalityof j  BA of j)
EEj is a critical measure of “mass
balance”; EE>1.0 is physically
impossible and indicates some error in
the inputs used to calculate it
Defining biomass pools or “groups”
• There is no “right” way to do this; more
groups (finer splitting) does not imply
“better” model
• Start simple, examine effect of adding
complexity
• Two approaches to defining pools
– “Partition” whole ecosystem biomass
– “Selection” of groups of particular interest
(Fulton: selection approach is “best”)
Estimating production PB
• Definition of P: ΔB = Production - Mortality
• This means Production= ΔB + Mortality
• If the initial B’s are assumed to be near an
“equilibrium” (B’s changing slowly), ΔB=0 and
Production=Mortality
• If we express production and mortality as
specific rates x biomasses, this means
PBxB=ZB, i.e. PB=Z where Z is an estimate of
total instantaneous mortality rate of B
• Over complex multispecies/age “groups”, need
to estimate PB as a weighted average of PB
components PBa, each weighted by its Ba/B.
Estimating Consumption QB
• Gastric evacuation: QB=(gut wt)x(guts/yr)
• Bioenergetics: backcalculate QB from observed
weight changes and estimated efficiencies and
metabolic losses
• Temming method assuming vonB growth
– Q=(dW/dt+3KW)/e
– K=vonBertalanffy growth K
– e=0.5-0.6
-QB=sum over ages of Qage x Nage, divided by biomass
Ecotrophic Efficiency
EE i 
 M ij  Fi
j
PB i
Note here that PBi is usually an input
(total mortality rate+biomass
accumulation rate), while Mij=Qij/Bi and
Fi=Catchi/Bi are calculated from other
inputs PB,QB,DC
What to do when your model does
not “balance”, i.e. EEi>1.0 implies
inconsistency in input values
• Increase estimated biomass Bi
• Increase estimated PBi (total mortality rate
Zi)
• Reduce DCij for one or more predators j
• Reduce QBj for one or more predators j
• Make BAj less negative
Other ECOPATH inputs for more
complex models
•
•
•
•
Immigration and emigration rates
Biomass accumulation rates (dB/dt)
Detritus “fate” (partitioning of available B)
Partitioning of catch into landings, discards
for multiple fishing fleets—>Fik, k=fleet
• Multistanza population dynamics
Nearshore
rockfish
4+
1+
0
Log
Numbers
at age
Multistanza, age-structured life
history representation in
Ecosim and Ecospace
Cascading
bottleneck
effects
Weight
at age
Shift from density
dependent mortality to
density dependent growth
Age (months)
Each stanza (range of ages) can be assigned distinctive:
1) Total mortality rate Z, varying with stanza-specific predation rates
2) Prey and habitat preferences (diet composition, distribution)
3) Behavioral tactics: respond to changes in food availability by
changing growth rate and/or activity and associated predation risk
4) Vulnerability to fishing and bycatch
Simulating time dynamics:
ECOSIM
• Simple biomass pools simulated over time by
integrating the rate equation
dBi/dt=ei∑Qji-∑Qij-Fi-Mi
where changes in Qij are predicted with the
foraging arena equation
Qij=aijvijBiBj/(2vij+aijBj)
• Multistanza populations simulated with agestructured equations for numbers N, body
weights W at age:
Ni,a+1,t+1=Ni,a,te-Zi
Wi,a+1,t+1=ki1aQi/Bi+ki2aWi,a,t
• These equations are all solved on one-month
time steps
Estimating trophic interaction rate
parameters for Qij predictions
• The a’s and v’s of Qij=aijvijBiBj/(2vij+aijBj) are
estimated in two steps:
• First, assume that the maximum possible
consumption rate vijBi is a multiple κij of the
Ecopath base rate MijBi:
vij= κijMij (user must specify κij)
• Next, calculate rate of effective search aij by
solving the foraging arena equation for aij with Qij
set to Ecopath base value and vij set to κijMij.
• This approach forces the dynamic model to
predict the Ecopath base rates whenever all the
Bi are at Ecopath base values.
Estimation of the critical κij
vulnerability multipliers
• These determine basic model stability and diversity
properties
• Κij >1.0, and depends on initial predator biomass Bj
relative to “natural” maximum as well as on foraging
arena behaviors.
• For species j that have been depressed severely by
fishing (low Ecopath Bj), must set κij very high (>10) to
predict recovery
• Very low κij (near 1.0) often work best for links involving
juvenile predators
• Can sometimes estimate by fitting to time series data
• Play same role as recruitment compensation ratios, but
in reverse (low κij implies high compensation ratio); often
difficult to estimate for same reasons (lack of contrast in
historical data)
Time forcing in Ecosim
• The basic rate equation dBi/dt=ei∑Qji-∑Qij-Fi-Mi
is parameterized so as to predict no change
(dB/dt=0) unless Ecopath biomass accumulation
rates are nonzero.
• Dynamic change can be caused by forcing
variation over time in
1. Fishing mortality rates Fi
2. Primary production rates PBk
3. Trophic interaction parameters a,v
• Patterns of forcing variation can be imposed by:
1. Sketching changes in the user interface, or
2. Using Excel .csv time series data files
Dynamic response in B’s is
dominated by assumptions about the
vulnerability multipliers κij
• High κij values (e.g. 10) imply weaker
limitation of predation effects through
foraging arena behaviors, stronger
interaction effects
• Low κij imply less variation in Mij with
changes in predator abundance, and also
stronger “compensatory” changes in
predator per-capita consumption rates
∑Qij/Bj
Adding realism to a,v predictions in
the foraging arena equation
• Foraging time adjustment: a’s and v’s can be varied so
as to simulate changes in feeding time aimed at keeping
food intake and growth constant (causes destabilizing
type II response)
• Risk-sensitive foraging (indirect trait mediated effects):
foraging time can be made inversely related to predation
risk.
• Handling time limitations: type II effects caused by
high handling times per prey
• Prey switching: rates of effective search aij can be
made to vary positively with prey biomasses Bi to
simulate processes like search image formation and
choice of alternative foraging modes by predators
• Trophic mediation: effects of other Bs on vulnerability
Main ECOSIM capabilities
• Simple time scenarios with sketched
forcing patterns
• Fishing efforts varied dynamically with
bioeconomic response submodel
• Fitting to time series data in .CSV files by
varying κij (“stock assessment” methods)
• Optimization of fishing efforts by fleet and
over time using search algorithms
• Stochastic simulations with randomly
varying ECOPATH parameters (sensitivity
analysis)
Evaluation of spatial policies:
ECOSPACE
• Consists of a map grid of Ecosim models linked
through dispersal/migration rates and movement
of fishing effort
• Each map grid cell has user-defined habitat
type, each group uses one or more types (has
high predation rate, low feeding rate in habitat
types that are not suitable)
• Mainly useful for evaluation of spatial closure
(MPA) policies; NOT a next step or improvement
on Ecosim for examining dynamic change
The Ecospace basemap
• Map layers include habitat type, relative primary
productivity, MPA type, relative fishing cost
• Can generate from global bathymetry and
primary productivity database, by web link to
UBC Sea Around Us Project server
• More often, simply define map area, resolution
and enter habitat type data using the Ecospace
user interface manually or using copy/paste from
data in Excel csv files
Key Ecospace inputs for each
biomass group
• Which habitats does it use?
• How much does it move, and how much
does it suffer when in unsuitable cells?
• What is its seasonal migration pattern?
• How do its movement rates respond to
local variations in predation risk and food
availability?
Ecospace solution options
•
•
Ecospace time predictions involve solving very
large system of differential/difference
equations, m x n x p of them (m=rows,
n=columns, p=groups), plus spatial effort
predictions
Three solution options:
1. Simple biomass equations (fast)
2. Biomass equations+aggregate age-structure
dynamics for multistanza populations (slower)
3. IBM packet solutions for multistanza populations
(much slower)
(Avoid option 1 for models with multistanza
groups)
Linking Ecospace with
hydrodynamic models
• Ecospace can read and use some spatial and
temporal outputs from physical models
• These outputs include
– Advection fields (surface current patterns)
– Total nutrient concentration fields
– Salinity fields
• NB: the idea here is that physics affects biology
but not the reverse, so can first run a physical
model then force the biology calculations with its
results