There is no specificity in the SAFO which restricts it to high altitude operations.
With respect to stalls which may occur prematurely due to ice contamination, we typically see a nose up control input immediately following the pitch down generated by the stall. Some of this may be due to the flight crew failing to identify the stall. This, in turn, may be because a stall due to ice contamination can take place well before artificial stall warning and may generate wing pressure distributions that result in unique, often violent aircraft behavior not previously experienced in clean stalls.
At Roselawn, the DFDR data shows an attempt to raise the nose back to a more normal attitude from the pitch down attitude following the first roll event. This attempt raised the nose back above the angle of attack which happened to be critical at that moment due to the existing ice contamination. The resulting second roll inverted the airplane; at this point, the crew had little chance of recovery. Had they unloaded the wing and accelerated, managing the angle of attack carefully, they most likely would have recovered. Indeed, another ATR72 experienced a buffet while approaching South Bend a couple of hours following the Roselawn accident. That crew identified the ice accretion aft of the boots, and, from the comfort of an airplane still under control, made some very conservative decisions regarding configuration and speed which led to a normal descent into warmer air.
There is certainly a strong argument for procedures aimed at minimizing altitude loss close to the ground. Even then, retaining control of the airplane is obviously paramount. I recall the first accident that I ever investigated; this was a Fairchild Metro II at Washington Dulles. The captain had managed to flame out both engines due to ice ingestion during a night approach to runway 1R. As this occurred in final, he had little option but to execute a forced landing (other than mashing both start buttons, which we were pretty sure would have worked but which did not cross his mind at the time). He picked out a clear farm field and set up a power off approach into it. At some point he became aware that he was too slow and still high. I'd have to look in my notes, if I could figure out which box they were in, but he may have encountered the stall warning. At that point, he did something which I have always wondered whether I would have done. He shoved the stick forward, regained some margin above the stall, and immediately pulled back into a flare a couple of feet above the ground. The resulting landing left the airplane unsalvageable but the occupants essentially unhurt.
The argument that we have been trying to advocate for so many years is this: approaches to stalls are trained with a clean wing, in a steady deceleration, with a steady G, at an altitude well within the performance margins of the airplane. The aircraft behavior prior to and during this stall is entirely predictable and is so modeled by the simulator. Procedures for recovery which allow minimum altitude loss to overshadow the more important reduction in angle of attack lead to incorrect responses in nearly every type of stall situation other than the one demonstrated in the sim.