Computing fortification games in a tri-level Stackelberg Markov chains approach