Looking at the accumulation distribution,I am not entirely clear as to how that could be simulated by ABM methods. What bothers me is that if I follow the rules / methodology you outlined, I should have a lot more points in the output that accumulations provide.

Moreover, in the submission you are expecting only coordinates, but not the statuses. So if that’s the case, are you thinking to compare the accumulation coordinates with all the agents locations, regardless of their “alive” status? Also, is nearest-neighboring metric calculated with / without replacement ?

And finally, moving of agents in the geologic environment is governed by lots of things, including permeability, and could be simulated provided geomodel of sorts. In either way, seismic “environment” values do not really tie to permeability directly. Hence I am not sure that seismic volume here is a correct “driver” of the fluid particles.

I was wondering about nearest-neighbor too. The equation also seems wrong (addressed in another post). It looks like you can concentrate all your points around a single answer point and do well?

That is correct; there will be more agents at the end of the simulation than accumulations. The goal of this challenge is to predict accumulations AND mimic the behavior (in a simplistic way) of fluids in the subsurface. The journey is just as important as the destination.

Please provide a summed file of the location of both living and dead agents.

To your final comment, yes, many factors dictate movement in the subsurface. Traditional basin modeling methods try to account for all these factors to predict migration, but this realism comes at a cost. With current modeling methodologies, a team will spend months building and tweaking a model that will be so computationally expensive that it can only be run a few times before the decision-makers need an answer. This challenge will produce a simpler tool that allows all geoscientists to quickly experiment with scenarios on the data they are already using to make interpretations.

This is an observation on the scoring. Comparing points by nearest x, nearest y, and nearest z will not necessarily yield the nearest point. Here’s a 2d example. Our point P is (2,2). P1 is (3,10), P2 is (12,4) and P3 is (5,5). The nearest x is from P1, the nearest y is from P2, but the closest point is P3.

This method of scoring is computationally easy and runs fast. For a discussion of finding the nearest point, a considerably harder problem, see Knuth, vol 3.