Program with abstracts

13:00-13:30 Josefin Ahlkrona, Department of Mathematics

Ice Sheets and the Warming Ocean - Modelling Challenges

Ice sheets, like the Greenland Ice Sheet and the Antarctic Ice Sheet, are important components of our climate system. The warming of the ocean and atmosphere means these ice sheets are losing mass, raising concerns about future sea level rise and ocean circulation changes. To predict future ice mass loss, one must model how ice is moving from the cold interior highlands towards the warm coasts, where it melts or breaks into ice bergs. In this talk we describe the basics of numerical models for computing such ice dynamics, and discuss their computational challenges.

13:30-14:00 Susa Niiranen, Stockholm Resilience Centre

Modeling higher trophic levels in a marine system – a size-structured approach

Changing climate together with other stressors are altering the state of world’s marine ecosystems at an unprecedent rate. However, it is often unclear how combinations of different drivers affect ecosystem function and dynamics. Particularly many uncertainties are related to how food webs, where different species interact with each other in various ways, are responding to future environmental and socio-economical change. In our research we study the response of Baltic Sea food web to changes in multiple drivers using a mechanistic size-structured food-web model POEM2.0 as a research tool. In aquatic ecosystems organism body-size largely defines how they interact with their environment, making it a robust proxy to be used in food web modelling instead of species or functional groups. Body-size can, for example, enable an organism to escape unfavourable conditions, and set limits for metabolic rates. In species interactions, body-size is a key factor in determining whether an organism exits an encounter as a predator, or a “bagged meal”. POEM2.0 model includes three functional groups: piscivore fish, planktivore fish and benthic organisms, each divided into a number of size-classes. In fish groups, life history dynamics are accounted for and the biomass of each size class is a function of somatic growth, recruitment, predation mortality and natural mortality. Prey consumption is calculated using a multi-prey type-II feeding function for all size-classes and groups, and the proportion of different prey eaten by a predator is additionally defined by an assumption of optimum predator-prey mass ratio. Finally, work is ongoing for including eco-evolutionary dynamics for the top-predators in the model.

14:00-14:30 Olof Leimar, Department of Zoology

Reinforcement learning leads to bounded rationality in a public goods game

I analyse a public goods game with continuous actions, using actor-critic reinforcement learning, driven by rewards of investment into a joint project. Players in a group invest into the project, receiving joint benefits and paying individual costs. The benefit is a function of the total investment and the cost depends on a group member’s individual investment and quality. I show that actor-critic learning in repeated rounds of play by the group converges to investments that correspond to a Nash equilibrium of a one-shot game with the perceived net rewards as payoffs. I examine the idea that there could be an evolution of the individually perceived costs of investment, expressed in terms of a perceived individual quality. I show that individuals will evolve to act as if they underestimate their own quality, in effect investing less into the project in order to induce other group members to invest more over the medium term. The reason for this phenomenon, where reinforcement learning leads to a one-shot Nash equilibrium given the perceived rewards in the game, but where Darwinian adaptation favours an underestimation of individual quality, effectively reducing investments, is that reinforcement learning responds to immediate rather than long-term rewards, and is thus a form of bounded rationality. Reinforcement learning could well be biologically realistic, but has little capacity to evaluate medium-term consequences of communicating individual quality to social partners. An evolution of perceived individual qualities can then compensate for the cognitive limitations of learning about the characteristics of social partners.



15:00-15:30 Måns Karlsson, Department of Mathematics

Statistical species identification with uncertainty accounting

Correct identification of species, subspecies, sex and age are all of critical importance to get reliable time series of bird (and other taxa) population changes and migratory timings. Through a Bayesian latent variable approach, we present a way of obtaining aposteriori species probabilities based on measured traits, which generalizes to practically any group of species and any type of trait. One may also choose the degree of conservativeness through a parameter value determining the size of so called indecisive regions, which contain trait measurements that are not distinct enough to determine the species with acceptable confidence. In other words, our model does not only predicts species, it also predicts when the measured traits are not sufficient for reliable species identification. Note that the method presented above is not only applicable to species identification, but also to subspecies, sex or age determination, or a combination thereof. We illustrate usage of the method on a data set of four Acrocephalus species.

15:30-16:00 Tom Britton, Department of Mathematics

Epidemic models with symptomatic and asymptomatic cases: who causes most infections?

In real life epidemics some infected show symptoms and others don't, the latter being referred to as asymptomatic cases. These individuals are still infected, and in certain cases able to spread the disease onward. In the talk we present a simple mathematical epidemic model allowing for these two types of infected. Assuming a major outbreak it is easily derived how many that get infected of the two types. A harder question is to derive which type that causes most of the infections, the latter question depends also on finer details of the model. This is negative news for prevention: it is easier to target symptomatic cases, but unless finer details of the epidemic are known, it is unclear if the symptomatic cases cause most infections.


You are most welcome to attend!