Everybody knows what distinguishes "weather" from "climate." Weather is unpredictable: It's what ruined the picnic last weekend. Climate is the general scheme of things, subject to variation and occasional extremes, but always drifting back toward a norm—which around here means mild, overcast, wind from the west, chance of showers.
It was more than a century and a half ago that scientists began to realize that, on a long enough time scale, climate is as variable as weather: that Earth's climate has been constantly changing ever since the planet cooled enough to allow liquid water to collect on it. (The evaporation and recondensation of water, absorbing heat at one point on its surface and re-releasing at another, drives the whole weather process.)
Most of the evidence for global warming has emerged from careful analyses of world weather records stretching back a few centuries: squeezing out the "noise" of random variation in the stats to let the long-term trends emerge, and then extending the trend discovered into the future. But recently more sophisticated techniques for modeling climate 50 or 100 years hence have begun to emerge.
Climate modeling is a game played on supercomputers. There's no arcane mathematics required, just unimaginably vast number-crunching capacity. Climate modelers first break up the map of the region under study into rectangular blocks. Each block is assigned a set of numbers representing climate data—some fixed, like elevation, others variable, like temperature and air pressure. Then the computer calculates the impact of each number in each block on all the numbers in all the adjacent blocks. When the results are calculated, the computer updates all the variables in each block accord-ingly and repeats the process, this time based on conditions an hour (or six minutes) into the predicted "future." And repeats it again. And again. And again.
Every climate model involves trade-offs. If, to increase precision, you want to recalculate the state of the system 10 times an hour instead of once and still get a result within your lifetime, something has to give: usually the geographical "resolution" of your model. Early models resolved the earth's surface into blocks eight by 10 degrees on a side. In our latitudes, an eight-by-10 block is about 600 miles east to west by 400 miles north to south. At that resolution, the wildly varied relief of the northwest US comes out one big dull bump: not much use for predicting snowpacks and stream flows. And it still takes weeks or months to crank out those crude results.
The most sophisticated models are now down to four- or five-degree resolution: still not ideal, but four times better than 10 years ago and good enough to get an idea how changing temperatures of air and sea will impact the precipitation patterns that drive the annual round of life in the Northwest.
The modelers don't just plug in today's values, set the software running, and hope that the results will reflect reality. They generally start the model running with known data from 10, 20, or 50 years in the past. If the numbers it spits out as it chugs past the present match up pretty well with present conditions, they have better reason to trust what it will say about the future.
