To predict the actual snow, Kapnick runs the model starting from July 1 of every year for which data is available. She and her team do that multiple times and get 10-12 possible futures. When those separate possibilities start to converge, they can start making predictions. It’s like running a simulation game over and over again until you start seeing a pattern in how the simulations turn out. The water in the air and the soil, the salinity of the ocean, the temperature of the sea surface, the temperature at every atmospheric layer, the directionality and speed of the wind—they all go in, and snowfall prediction comes out.