Well, alright, not crap exactly. As I discussed in a previous post, 90's models have so far shown that they are doing a reasonable job of predicting temperature trends over the last 15 years. Plus, methods like hindcasts, which predict the last century so we can compare the result to observations, give us confidence. Unfortunately though, there are still a bunch of uncertainties we haven't quite cracked yet. Think of it as the resolution of a camera (if you want to, I'm not a fascist or anything, you can do whatever you want). An image taken from a distance might give you an idea of what's going on, but it's kind of blurry and the fine details aren't there. Get closer, or improve the resolution, and you can make out faces and the like. Currently we're in the blurry phase of climate modeling...we have an idea of what the future will be like, but the proverbial giraffe we're predicting might turn out to be a horse standing by a lamppost.
This was discussed recently, in slightly more scientific terms, by Stevens and Bony (2013) in a Science perspectives article. The article reads pretty much like an intervention for the climate community. Once the Intro is over, it gets on with step 1: admitting you have a problem. The figure below shows a group of four models from the CMIP5 project, the modelling study the next Intergovernmental Panel on Climate Change (IPCC) report will use. In each case, the discs show a world map after warming has reached 4 degrees warmer than the present. The top shows the variation in solar radiation due to clouds, while the bottom four show rainfall changes
So we are pretty certain it will be more rainy on the equator. Probably. And a change in cloud cover will increase the amount of solar radiation across the US and Europe. Or...it will be less. So clouds and rainfall are two of the things we basically aren't very good at predicting in the future. Not only the amount of change, but whether we'll see an increase or a decrease isn't yet obvious. It isn't just some of the physical parameters we don't understand either. Climate sensitivity is also a problem, the magic ratio that says "x amount of CO2 in the atmosphere leads to y amount of heating". You can alter that ratio in the model, leading to pretty drastically different results.
The recommendation of this paper is that the community as a whole goes "back to basics" and focuses on advancing understanding and improving the numerical representations of various processes, rather than the current trend of adding in other factors like soil moisture and the biosphere. I don't really know why the focus can't be on both areas at once, since it seems plain rude to ignore bits of the natural world. I guess the real take home from this is to not get overly cocky; the dream of those original scientists at the GFDL hasn't come true just yet.
Stevens, B., & Bony, S. (2013). What Are Climate Models Missing? Science, 340 (6136), 1053-1054 DOI: 10.1126/science.1237554
