Spatially explicit IBMs of fish populations by Geir Huse Department of Fisheries and Marine Biology, University of Bergen, Norway Lecture II, NORFA course

Talk outline 1 Case study:Vertical migration in a mesopelagic fish 2 Modelling adaptive predator-prey interactions by IBMs: Background Problems Brief review of previous studies 3 Case study:Predator-prey interactions between cod and capelin in the Barents Sea

Vertical migration in mesopelagic fish (Strand, Huse & Giske 2002) Maurolicus muelleri Short lived, small mesopelagic fish inhabiting the world’s oceans and Norwegian fjords

Background Artificial evolution is commonly applied in the field of artificial life to generate behavioural strategiesartificial life Objective To investigate whether artificial evolution is a suitable tool for studying biological organisms Why Maurolicus?

Environment data are taken from a Norwegian fjord Seasonal zooplankton and temperature data Seasonal and vertically resolved light Visually foraging predators Simulates 1 day per month with 5 min. time step The model: Environment

A i (mass,energy level,age,depth) S i (energy allocation,spawn,vertical movement) Entire life span: larvae, juvenile, adult Visual feeding Emergent fitness Simulations: With deterministic larval survival With stochastic larval survival The model: IBM

Maurolicus movement

Reduction in genetic variability Stabilisation of character values Evolutionary dynamics Year 1 Year 250 Year 500

Results Spawning strategy: The spawning season was longer in the stochastic than in the deterministic simulation

Feeding dynamics: Juveniles fill up in the morning Maintain a high stomach fullness throughout the day Adults are less risk prone and have a low feeding rate

Comparing predictions with observations January Non symmetric ascent and descentsymmetric

The model predictions fit well with observations on Maurolicus muelleri in Norwegian fjords Findings of a asymmetry in DVM pattern caused by diel differences in stomach fullness is predicted and supported by field observations The results suggest that models using emergent fitness can be a fruitful tool in studying biology Conclusions

Predator -prey interactions: Background Predators impact on prey dynamics in two important ways: By killing them By scaring them

The killing.. Predator-prey interactions by Lotka-Volterra Prey dynamics Predator dynamics

The scaring.. The scaring of prey is emphasized within behavioural ecology Effects of scaring Prey move away from favoured feeding patches due to fear of the predatorfear of the predator Prey spends more time on vigilance Both fear and killing can be studied using IBMs

Modelling predator-prey interactions Predator-prey interactions are part of a more general type of problems referred to as games: the profitability of a strategy depends on what others are doing. Typical games include: intraspecific, interspecific, predator prey interactions Fitness landscape is continuously changing..

Various predator-prey models Iwasa 1982: Fish and zooplanktonFish and zooplankton Van Baalen & Sabelis 1993: Coevolution of patch selection Hugie & Dill 1994: Fish and fish – game theoryFish and fish – game theory McCaughley & al. 1996: Movement and age structure Brown & al. 1999: mountain lion and mule deer

Questions How about large scale predator-prey interactions? How about adaptation over long time scales: season evolution

Predator-prey interactions: the cod-capelin case in the Barents Sea

The Barents Sea ecosystem as a case study A relatively simple ecosystem There is great inter-annual variability => understanding is important to facilitate prediction Resource management => A good system for model development

The Barents Sea

Redrawn from Giske & al Parathemisto Phytoplankton Calanus Polar cod Capelin Cod Herring Krill Barents Sea food web (simplified) Man Seal

Previous capelin model studies: Reed & Balchen 1982 : Comfort maximisationComfort maximisation Giske & al. 1992: A conceptual modelA conceptual model Fiksen & al. 1995: Dynamic programmingDynamic programming Huse & Giske 1998: IBM with ANN and GAIBM with ANN and GA Huse 1998: IBM full life life cycleIBM full life life cycle Huse 2001: Adapted random walkAdapted random walk In prep: Adaptive predator and prey behaviour

Daily time steps Barents Sea oceanography 60x60 squares with 20 km resolution TemperatureTemperature and zooplankton fields Prey feeding on zooplankton is density dependent Probability of finding prey is proportional to prey density – one feeding attempt per day Predator mortality is constant - but not aways.. The model:

Attribute and strategy vectors of super individuals:super individuals A s (mass,internal number,position) S s (W 11,W 12,..,W nm,spawning position) Constant number of individuals represented Fitness: maximisation of net reproductive rate R 0 : internal number x fecundity Fecundity = a*M b IBM:

Movement: Movement for predator and prey is decided at the end of each day using an ANN:

Capelin spawning migration model February obs ARW ”Verbal” model

Fixed predator mortality A population dynamics B Spatial dynamics C ANN weights Seasonal movement Consumption

Fixed predator mortality Spatial correlation between predators, prey and prey resource

HeterogeneousHeterogeneous predator mortality A Population dynamics B Spatial dynamics C ANN weights Seasonal movement Consumption

The “ecology of fear..” Refuge against cod predation Capelin may escape cod by outcooling it..

The “cost of fear..” Simulations with fixed and heterogeneous predation risk (non-adaptive predator) The cost of fear can be estimated as the reduction in growth attributed to the displacement away from the most desirable feeding grounds

The “cost of fear..” Prey weight with variable predation rate: –95.2 g. Prey weight with fixed predation risk: –104.8 g. Statistically significant difference (ANOVA, p<0.0001).

Conclusions Studying predator-prey interactions in models is fun! But it is still hard to know what to make of it.. Many of the predictions for smaller spatial scales are upheld while others are not Large scale movements of predators is an important feature that needs to be addressed (Lima 2002)

Resulting depth distributions

The ING-concept

Temperature

Super individuals (Scheffer et al. 1995) A super individual represents many identical individuals and in this case the number of such identical siblings (n s ) thus becomes an attribute of the super individual: A s = (α1 s,α2 s,α3 s,…,αm s,x s,y s,z s,n s,t)

-..life as it could be (Langton 1989) -Alife focusses on the general laws of life -Life on earth is but a special case -Perform studies in hardware and wetware but mainly software Artificial life (Alife)

Zooplankton and temperature profiles

ING model predictions and optimal solutions Pp= local predator abundance Zb = local zooplankton abundance

Temperature