model decision-making part 2

Movement Decisions

As an update to my post entitled model decision-making (17/01/2012, here), i've changed my mind yet again (!) on the best way for my agents to move around.

Food is by far the most logical incentive, but then 'cost' became an arbitrary factor that counted for nothing real and only worked in making the model use more factors for no apparent reason. The only way I could see around this was to incorporate food and mortality into the cost factor - as was originally intended when I chose to centre my ABM around ideas of least-cost modelling.

so now, my agents make movement choices based on:

1. road cost
2. habitat cost
3. pheromone cost

Food is still available in the cell, depending on the habitat at that location, but a propensity to avoid cells recently inhabited (via the amount of pheromone available) means that agents are directed at cells that are likely to have greater food availability.

Basing movement decisions on habitat and pheromone implicitly include aspects of food, cover, mortality, mating opportunities etc rather than having to specifically include each separate detail in the model.

This makes more sense to me, and reduces the complicatedness of the model. This has become a real issue to me as improving the model tend to naturally lead to an ever-increasing array of parameters and choices and extra code.

Where a road exists, this overwrites the habitat cost.

Feedback and Improvements

I've also had a lot of good feedback from Bart and Beccie, out in Belize, and I hope to incorporate their ideas and suggestions for improvements in subsequent drafts of the model. At the moment, I have simplified some of the habitats (by combining some types of forest together) and simplified the roads to 3 types:

1. large, regularly used tarmac roads
2. small, infrequently used non-tarmac roads that allow cars to pass
3. small trails and tracks inside and outside cockscomb including some pedestrian tracks that have no car use

Interactions Between Agents

At the moment of writing (june 2012) I'm having trouble deciding how best males and females should interact and how best to attract females and males together for appropriate mating opportunities. I seem to have 4 distinct types of interactions that update how the agents view pheromones from other individuals:

1. own strength - so how strongly individuals interact with their own pheromones
2. avoidance of males - how strongly females interact with male pheromones (generally they should try to avoid males so as not to be harassed, although this changes when they are in heat and ready to mate, but could potentially be more severe an avoidance if she is pregnant or has cubs?)
3. mating attraction - how strongly females should be attracted to male pheromones when they are in heat and ready to mate
4. male mating attraction - how strongly males should be attracted to females when they are in heat and ready to mate

Additional issues arise when males sense a female is ready to mate, as they are likely to stop actively hunting and looking for food and may become quite thin in their pursuit of a mating opportunity. Males also cover more distance per day when searching for a female in heat.

I'm currently running some trial simulations looking for the effects of these pheromone parameters on the territory sizes of males and females, and in the number of individuals that die and are given birth to. It seems the interaction between these various parameters may be causing some unexpected results in my data.

Territory Sizes

Lastly, territory sizes are become a real concern to me. Territories are based on the number of cells that have an individuals pheromone located in them. Two separate parameters dictate how pheromone survive and propagate - pMax (the maximum amount of pheromone deposited in a cell when visited by an agent) and pDegrade (the rate at which the pheromone degrades and eventually disappears). 

My initial trials used pMax of 100 and pDegrade of 0.98, which gave stable territory sizes of around 70-100 cells per individual. Of course, it wasn't until I realised that this was a tenth of the size that territories needed to be that I ran into a problem!

This needs some more work and investigation, and at the moment any combination of settings seems to either generate territory sizes that are still too small, or that steadily increase throughout the simulation (presumably territories would stabilise at some point but too much time is required for them to stabilise for the parameters settings to be effective).

Running the model for 5000 ticks (or around 4 years) currently takes around 15-20 minutes in batch mode and I would rather not wait any longer before proceeding with the intricacies of the model and starting to collect some data.... although its looking increasingly likely that this may be the case.