I realize the problems that cave radios have had over the last thirty or so years, especially propagation by induction.
I wonder if something completely different would work. Consider the concept of a daisy-chain mesh Wi-Fi network at 900 MHz. If the first node was placed at the entrance, and then more nodes placed as the cavers go deeper, maybe they could get enough range from node to node to get short messages in or out. If the air passageways were good, and if high performance dipole antennas were used, would this work?
I'm guessing that only the slowest data rates could be supported, too slow for voice. Short coded messages of a few characters might work.
I'm looking at a per-node cost of about $50.