How accurate is that ten-day forecast?"Most TV meteorologists make ten-day forecasts without telling viewers that after the first two or three days, it's probably a crapshoot," says Jack Williams, USA Today's founding weather editor. But long-term forecasts aren't entirely worthless. Scientists can roughly predict daily highs and lows more than a week in advance, and forecasts are more reliable in winter than summer. Saying whether it will rain, however, is dicey until four or five days out, says Shaun Tanner, senior meteorologist at the Weather Underground, an online weather portal. Both suggest a simple strategy to test the reliability of extended forecasts. Let's say you want clear weather for a backpacking trip. Start checking the forecast ten days out, and keep tracking it every day. If it changes dramatically, it's not reliable; if it stays roughly the same, you can trust it. That's because forecasts are tested by running them through a complicated computer model hundreds of times—a technique called ensemble forecasting. During each run they alter one data point (like the position of the jet stream) to gauge how it affects the final result. "If all of these runs converge on the same solution," says Williams, "it means the weather pattern is stable and the forecast is reliable."