A month-long initiation into Agent-based Models

By Andrew F Johnson April 11th

Figure 1. Hard core team Blanca, Emilie and Andrew working in Uppsala for a day.

I travelled from the Scripps Institution of Oceanography, where I am currently a post-doc in marine ecology, to the Stockholm Resilience Centre (SRC) to work with Maja, Emilie and Blanca (and after meeting a lot of friendly faces, a fair few others as well).

The aim of my visit was to parameterize an Agent-Based Model (ABM) that Emilie has been building for the last few months. In more “normal speak” I basically came to learn about a method of simulating the actions and interactions of fishers, and help the SRC team make sure the ideas they were using to design their model made sense based on my work with small-scale fisheries in Baja California Sur, Mexico.

We are interested in understanding what factors trigger a phenomena called sequential exploitation. This is a situation that occurs when a harvester (in this case a fisher) harvests a local area until the area is no longer profitable. That harvester then moves on to the next profitable area and harvests there until it is no longer profitable and so on and so forth (Figure 2). This sequential exploitation doesn’t always happen, particularly in well managed systems in which harvesters know that taking too much means there won’t be enough over the long-term. In our Mexican fisheries study system, however, we see considerable overexploitation of fish stocks and therefore, in certain cases sequential exploitation phenomena occur.

To try and understand how and why this movement from local to neighboring resources occurs, Emilie’s model accounts for a number of different factors that we think play a part in the movement of fishers between areas. These factors include things like:

  • How mobile a fisher may be (this could be determined by age, boat technology, wealth)
  • How much local versus neighboring-area fish there are (this could be determined by how productive local waters are or how much they have been fished historically)
  • How much fishers communicate with each other (this may be determined by the mobility of a fisher and the chance of that fisher “bumping” into others)
  • How quickly fish populations replenish after being harvested. (this could be determined local productivity and the species in question)

Figure 2. An early schematic diagram of our model system showing 4 areas each fished by 4 fisher groups, 50% of which can move between areas (those with the red hats). Not the prettiest image, but it’s a starting point.

After a month of playing with numbers, looking at patterns in Mexican fisheries data (Figure 3) and emailing local Mexican scientists and fishers, the model has been finalized and we are working hard to collate and organize the results. These results come from pressing the play button on the ABM and watching how our agent-fishers move between areas with different levels of fish stocks. Although it is still early days, we have had some exciting results and look forward presenting our findings in a paper which should be ready by this summer.

I thoroughly enjoyed my time working with Maja’s group at SRC and was surprised at how collaborative the atmosphere there was. I hope this first collaboration leads to more in the future and I look forward to visiting again.

Figure 3. Some basic analyses of the empirical data informing the model looking at the interplay between squid and finfish.landings in the Gulf of California, Mexico.