How Do We Do It?
Climate models help us imagine possible climate futures.
We use weather forecasts for short-term planning. Climate projections can be used for long-term planning—but they are not the same as forecasts. There are uncertainties based on model limitations and unknown future human behavior that make long-range forecasting very different from predicting tomorrow's weather.
Climate projections look much further into the future than weather forecasts. They address uncertainties by focusing on the range of future conditions that would likely occur and how it will change in response to changes in the factors that affect it.
This 13-minute video—featuring SNAP's Nancy Fresco and Katie Spellman, a postdoctoral fellow at the UAF International Arctic Research Center—explains the importance and relevance of computer modeling in making sense of climate change.
How do we make these projections? We use climate models.
Climate models are simplified versions of reality that try to explain climatic processes with just the most necessary parts of the system. Their usefulness is evident when we compare observed historical climate and simulated data—the models capture the most important climate patterns.
Climate models use data to calculate how the global climate varies. These data include:
- Initial climate conditions
- “Forcings” such as atmospheric greenhouse gas, solar and volcanic variability
- Ocean and atmosphere variability
- Land surface conditions
- Feedbacks such as the carbon cycle and the water cycle
- The end product is a simulation of future climate. Because the end product is based on statistical probabilities, the data are most reliable when averaged across time or space, such as the projected average of 30 years of winter precipitation for your community, or the likely hottest temperature that might occur on the North Slope.
SNAP evaluates and downscales the best performing models for the North
A Global Climate Model (GCM) focuses on projections of climate change by simulating how Earth’s physical processes respond to increasing greenhouse gas concentrations. Over 20 global climate modeling centers produce more than 60 GCM versions, all attempting to approximate the global climate.
Climate models aren’t perfect, though.
Between now and about 20–30 years from now, current climate change is the best predictor of the rate of change, but year to year variability is the largest source of uncertainty. Although long-term climate change shows clear trends, those can be masked by natural ups and downs in the short term. Climate models do simulate this kind of variability, but they cannot predict it precisely.
Uncertainty is inherent in SNAP’s work. There are many sources of uncertainty in future climate projections, ranging from the uncertainty of predicting economic and social choices to the sparseness of historical climate data with which to calibrate models. Because these sources are so disparate, the uncertainty in projections cannot reliably be captured in a single number or probability.
Best practices for making projections
Use multiple decades
Averages over 20–30 years are more resistant to transient variability in climate models and to natural variations in regional climate. Compare a future (like 2030–2069) to a historical reference frame (like 1970–1999), and keep in mind that the later the historical reference frame, the more climate change is already in it!
Use multiple climate models
Picking one model is not good practice because all the models are at least plausible, if not equally likely. Use several separate models if the full range of possibilities is important to your work, or use the average of multiple models if you are more interested in the most likely outputs. This is especially critical between now and about the 2050s or 2060s.
SNAP offers a five-model average. This is the average model output from the top five models that best replicate historical climate in Alaska and the Arctic regions. The five-model average is best for looking at general climate trends over time, as it smooths over the year-to-year and decade-to-decade variability. Because of various smoothing effects, it's not recommended for exploring climatic extremes or annual variability.
Use multiple scenarios
Given the uncertainty around future human behavior, you should pick at least two scenarios that bracket the likely range, unless you are only interested in looking at the “best case” or “worst case.” RCP 4.5 and RCP 8.5 are good choices. This is especially critical after about 2050.
Look at medium–term and longer–term futures
A comprehensive assessment would consider a historical, a mid-21st century future, and a late–21st century future. The two futures should have a high, low and middle range each, possibly with multiple models and multiple emission scenarios in each future window.
Don’t make your assessment area too small
The more local your assessment, the more likely it is that local factors like elevation, vegetation, downscaling processes , etc contribute to the uncertainty. Larger areas are probably more resistant to this local variation, so a watershed, planning unit, or responses across several of these are perhaps more useful to consider.
For fire projections and post-fire vegetation, look at averages across many model runs. In these particular cases outputs are especially helpful for assessing large–scale long–term shifts, but are not meaningful at a pixel by pixel level.
How will the “real” climate compare to any projection?
The future climate we experience will not look exactly like any of these projections, but it will look like a lot of them. There are a range of future climates we may experience, and best practices are to plan for the likely range of climates, impacts, and associated risks for the time frame and region you’re planning for.
Projections are always improving incrementally. Don’t wait for a better projection—you’ll always be waiting and the costs of waiting will increase.