Well, I think I got it. My assumption is, that the hunter is able to follow the prey step by step. With that assumption I can create a path by tracking the movement of the prey, because if the prey was able to walk a certain path the hunter will be too.
I know that this approach has it's shortcoming. If the prey has abilities the hunter has not, like jumping over a barrier etc. But there are two reason the assumption is acceptable. First, it is not realistic, that a hunter is able to follow a prey around the whole map. Once the distance is too great, the hunter has failed and will pickup an other task. The second point is, that if the hunter is not overcoming a barrier a quick A* could help him, if it is not helping, he will abort the hunt.
The idea is, to start with a A* (Figure 1) and create an initial path to the prey. Now, while folling the path, the hunter will expand the path by checking, if the prey leaves a certain area of the last position in the path (Figure2). Once he has left it, the hunters path will be expanded with a new node(Figure 3).
In Figure 4 you can see a path expansion of 4 nodes. An optimization would be to check, if the prey re-enters an area of any node but the last one. In this case the hunter can take a shortcut and the according nodes can be disposed.
How good it will work depends on the choosen radius. I think I will give it a try :)