NeuralGCM combines physics-based modeling and a neural network trained on NASA precipitation observations to simulate global precipitation more accurately than other methods, particularly for capturing the daily precipitation cycle and extreme events.