I can't remember my CatOS syntax exactly, but:
Look at the details for that spanning tree instance. See wehre your spanning tree events are coming from- which port. Track it down to the port and see what's connected there.
Make sure servers are in portfast, switches and multiple uplinks aren't in portfast mode, etc. DOn't turn off STP. DOnt' enable bptu filter on a per-interface basis.
Is it all one big VLAN, or multiple VLANs per site? Does just one VLAN go down, or the entire site? What is CPU load like on the switch during the outage?