White is in trouble, as Black is ahead in material and about ready to play Qxg2# (or mate on f1 instead). So be counterintuitive and think about offense; examine how White’s pieces bear on Black’s king. Both bishops are trained on the king’s position, along with the queen and—don’t overlook this—the rook on f1. In fact the only two pieces coordinated against the same square are the rook and a2 bishop. That doesn’t mean the other pieces aren’t doing important work; it just should stimulate your thinking and cause you to realize that White is pretty close to being able to play Bxf7#, with the king unable to flee because the bishop on h8 cuts off its dark escape squares. Why does this matter? Because what prevents Bxf7# is Black’s queen on f2. The queen is pinned to the mating square.
So: attack it. The one piece White has free for the purpose—the only piece not participating in the mating threat that makes the idea go—is his queen. He plays it to g3 with check, which is imperative to keep Black busy and prevent him from delivering mate. To the casual eye Qg3+ may look dangerous for White, but the danger is illusory. Black can’t play QxQ without losing the game on White’s next move. Black has no choice but to interpose his knight on g6 and let his queen be lost a move later.
It’s easy to miss the win for White here by overlooking the power of the rook on f1 and the mating threat it creates with the bishop on a2. The rook looks impotent because it's pinned (by the Black rook on e1), but that pin wouldn't prevent the rook from providing valuable cover to White's bishop if the bishop were to land on f7. You can only be sure to see these possibilities by considering how each of your pieces bears on the enemy king—and without dismissing possibilities, either, just because they are obstructed. The obstructions may turn out to make fine targets; they may be pinned to the squares your attacking pieces would like to reach.