Winter forecast, part II: NOAA's predicts a warm winter for the Central U.S.
Let's follow up on yesterday's discussion about the long range forecast for the coming United States winter. Those of you outside the U.S. will probably be more interested in what the International Research Institute for Climate Prediction has to say for your country, and I encourage you to check out their excellent web site for their seasonal forecasts.
The official National Oceanic and Atmospheric Administration (NOAA) 90-day forecast for the upcoming winter, issued on November 20 by their Climate Prediction Center (CPC), calls for above average temperatures across the Central U.S. and Alaska. The remainder of the country has equal chances of above or below average temperatures. A dryer than average winter is expected over much of the Southern U.S., including the drought-stricken Southeast U.S.
Figure 1. Temperature forecast for the upcoming winter--December, January, and February 2009--made by NOAA's Climate Prediction Center. No areas of the country ar forecast to have an above-average chance of being colder than normal, but the Central U.S. has up to a 50% chance of having above-average temperatures.
How are the NOAA winter forecasts made?
NOAA uses several tools to make their forecasts. One key tool is their Climate Forecast System (CFS) model. This model includes a version of the GFS forecast model that we use for everyday weather and hurricane track forecasts. The CFS model also includes an ocean model that interacts with the atmospheric model. These models solve mathematical equations of fluid flow using a supercomputer for the entire globe, on a 100-km grid. NOAA also uses statistical models, which look at past winters and see how they depended on quantities such as sea surface temperature anomalies. Temperature trends are important, too--if it has been warmer than average the last ten years, it's a good idea to forecast a warmer than average winter.
Figure 2. Skill of the official 90-day forecasts issued 0.5 months in advance by NOAA's Climate Prediction Center. Note that the average skill over the past ten years is not very high (9 on a scale of 0 to 100), and has remained flat, indicating that our skill in making long-range forecasts has not improved.
How good are the NOAA winter forecasts?
NOAA rates its forecasts using the Heidke skill score, which is a measure of how well a forecast did relative to a randomly selected forecast. A score of 0 means that the forecast did no better than what would be expected by chance. A score of 100 depicts a "perfect" forecast, and a score of -50 depicts the "worst possible" forecast. For 90-day temperature forecasts issued 0.5 months in advance, NOAA has averaged a 9 out of 100 on the Heidke scale since 1995 (Figure 2). So, while there is some skill in forecasting what winter temperatures will be like, this skill is not much better than flipping a coin. Depressingly, Heidke skill scores for three-month precipitation forecasts are even worse, averaging just a one on a scale of 1 to 100 over the past 15 years.
Let's look at some examples. Last's year's winter temperature forecast issued in mid-November did poorly (Figure 3), failing to forecast that the U.S. would have equal areas with both above and below average temperatures. The 90-day forecast done in mid-November of 2005 for the winter of 2005-2006 was awesome, with a Heidke skill score of 45. However, the 90-day forecast done in mid-November of 2006 for the winter of 2006-2007 had virtually no skill, with a Heidke skill score of one.
Figure 3. Temperature forecast for Dec 2007-Feb 2008 issued by NOAA's Climate Prediction Center on November 15, 2007 (top). They predicted Equal Chances (EC) of either above or below-average temperatures for the Northwestern U.S. (white colored areas), and a 30-60% chance of above average temperatures over most of the remainder of the country. In reality, the U.S. experienced an average winter, with approximately equal areas of the country receiving above and below average temperatures (bottom). Image credit: National Climatic Data Center.
Why do seasonal forecasts do so poorly? Primarily, it's because the long-term weather patterns are chaotic and fundamentally unpredictable. To a lesser degree, we are limited by our imperfect physical understanding of what controls the climate, and our imperfect computer models we use to simulate the climate. As computer power continues to increase and our models include better representations of the weather and climate at finer grid sizes, I anticipate that seasonal forecasts will improve. However, given that long-range forecasts have not improved since 1995 despite a large increase in computer power, I doubt that this improvement will be more than 10-20% over the next thirty years.
Seasonal forecast models vs. climate models
A common complaint one hears about global warming predictions made by climate models is, "How can we trust the predictions of these climate modes, when they so such a lousy job with seasonal forecasts?" It's a good question, and there is no doubt that seasonal forecasts have pretty marginal skill. However, there is a fundamental difference between making a seasonal forecast and making a 100-year climate forecast. A seasonal or a short-term weather forecast is what mathematicians call an "initial value" problem. One starts with a set of initial meteorological and oceanographic values that specify the initial state of the planet's weather, then solve the equations of fluid flow to arrive at the state of the atmosphere a few days, weeks, or months into the future. This forecast is highly sensitive to any imperfections one has in the initial conditions. Since there are large regions of the atmosphere and ocean we don't sample, it's guaranteed that the prediction will suffer significantly from imperfect initial conditions. Furthermore, the chaotic and turbulent nature of the atmosphere leads to many "bumps" in the weather pattern over time scales of days, weeks, and months. The nature of turbulence makes it impossible to accurately forecast these "bumps" that are superimposed on the mean state of the climate.
A 100-year climate forecast, on the other hand, is what mathematicians call a "boundary value" problem. Given an initial and final set of factors (called "forcings") that influence the climate, one runs a climate model 100 years into the future. The final state of the climate will depend on the strength of the forcings supplied. This type of model is not very sensitive to initial conditions, and is not trying to forecast the "bumps" of chaotic, turbulent atmospheric motion superimposed on the mean climate. Rather, one is trying to forecast the mean climate. As computer power increases and our physical understanding of how the climate works grows, these type of models will continue to significantly improve. While climate models do fail to properly simulate important aspects of our past climate, such as the Arctic warming of the 1930s, and the observed 0.1°C global temperature increase that occurs at the peak of the 11-year solar sunspot cycle, they have been very successful at simulating things like the global cooling triggered by the 1992 Mt. Pinatubo eruption, and the observed pattern of greatest global warming in the Arctic. I believe that climate models are already significantly more reliable than seasonal forecast models, and should continue to improve steadily in coming years.
Support the Portlight Christmas for Gulf Coast Kids Honor Walk
Saturday is the portlight.org Christmas for Gulf Coast Kids Honor Walk. This is a fundraiser to buy gifts for the kids along the Gulf Coast who might not have much in their stockings this year because of the ravages of Hurricane Ike. Our own StormJunkie will be walking up the Arthur Ravenel Jr. Bridge in Charleston, SC, and will be taking his webcam along. Tune in to the webcam site at 2:30 pm EST to follow the walk, and participate in a live chat. Sponsorships of any amount, small or large, are appreciated! The cam will go active about an hour before the walk. It should be a cold but beautiful day.