I had an idea. Actually several of them that you could use. They were inspired by nature.

Context: Much about how things in nature are (movement, reproduction, appearance, structure, etc.) can be predicted in nature using initial conditions (a manifestation of chaotic* systems). If the initial conditions are known, simple calculations can provide predictions for future growth. It explains why trees grow in groves, ferns grow in clumps, and generally why species generate clumps of growth.

Idea: Using this, features of plants and animals could be predicted, as well as whether they will produce offspring (using distance between agents, local environment, individual properties and health of each agent, as well as other possible factors)

Interesting results: The player would have to avoid things like overhunting, overfishing, or irreversible destruction of habitat or risk completely removing or crippling entire species. While this may prove to inhibit some things in gameplay, it would make it much more interesting. Also, species could be cleared to make way for other species strategically for more hardcore players.

Second idea:

Context: AI (neural networks specifically) are fairly simple in principle, but they have another advantage: they can support many interacting processes which run independently of each other. Simple does not mean easy to understand or implement on standard devices, but there are many who have already done it.

Idea(2): Give each animal a pre-trained** process to run on, and it will react with things around it and, based on things observed (ex. hearing a predator, finding food, etc.) and experienced (ex. lions don't attack at night, the player can attack from a distance) said agent will change its behavior over time. Have a small independent process (which could be bundled with others for efficiency reasons) for each animal.

Possible Results: More realistic-behaving animals with less effort, both computational and programming.

I don't know if you would want to try either of these, but they are very interesting, and also intuitive. They explain why different biomes occur, why people in different areas think differently, and why self-sustaining reactions occur, along with a lot of other things.

One thing though: you may want to have a on/off setting for natural selection, as it may prove annoying in some types of gameplay or specialized maps.

Also just a video on graphics efficiency i found really interesting (unrelated to my ideas): https://www.youtube.com/watch?v=hYMZsMMlubg