Nice extension it shows how the situation can appear quite complex. And in your example its not even clear where the ice level is.
But at the end of the day the issue is when is the ice going to break.

Since we are talking about ice you can see that the same concept applies to the melting in the arctic. The models assumed a gradual strain/stress relationship with steady melting. The system in reality had fracture points.

It seems to make more sense to instead assume that the system has stress/strain fracture points then figure out when they would be triggered. For melting ice it makes sense given the currents etc that ice less than a meter thick would fragment so the models should have gone non-linear at that point.

Another example is the melting of the Greenland ice cap. It makes sense to assume once say enough meltwater is available to lift 10% of the cap the system is susceptible to collapse although to my knowledge no model include this mechanical collapse situation even though we have seen it in action.

Looking at oil supply the system its obvious that now that supplies are tight the system is strained and breakdown events become increasingly probable.

And of course any "event" will lead to a lot of other feedback loops getting triggered.

I used to work on something called schocastic resonance.

http://en.wikipedia.org/wiki/Stochastic_resonance

Cracking under strain is probably a stochastic resonance condition.
What this means is that strained systems are really undergoing vibrations which drive any resonators in the system. This driven coupling leads to feedback then highly non-linear ( explosive) behavior.

Thus stressed complex systems are full of both vibration modes and oscillators that are driven by the vibrations and subject to feedback.

DO NOT TAUNT HAPPY FUN BALL

http://en.wikipedia.org/wiki/Happy_Fun_Ball