Models: Climate, Hydrologic, Vegetation and More
To understand the Earth’s complex climate, researchers employ powerful computer simulations called climate models. Think of climate models as science-powered video games that allow us to glimpse probable future climate conditions, such as temperature and precipitation in the 2080s. Keep this in mind: climate models make projections not predictions. Climate models can’t predict exactly what will happen in the future; instead, they estimate a range of physically and probabilistically realistic futures that allow us to plan ahead and hedge our bets against likely climate scenarios and impacts. Climate models come in two flavors: global climate models and regional climate models. Let’s start with the better-known global climate models (GCMs).
Global Climate Models (GCMs)
As their name suggests Global Climate Models (GCMs) model the whole Earth. They do this by dividing the globe into large three-dimensional, box-like cells that include the atmosphere to the space just above the ground. Inside these cells atmospheric dynamics are simulated within the computer. As with a video game or other computer simulation, the cells have a resolution that’s dependent on how much information can be packed into a given cell. For GCMs, the more cells you have to divide the world into, the more detailed and higher resolution that simulated world will be. (Incidentally, this isn’t just a metaphor; “resolution” is the term of art in the climate-modeling world.)
Yet, while engineers have consistently built higher and higher resolution into GCMs—thank you, Moore’s law—this resolution still isn’t fine enough to assess regional climate impacts in any great detail. Consider an important Pacific Northwest feature: the Cascade Mountains. Under the coarse resolutions of many GCMs, the Cascades are replaced by a kind of featureless plain that gently tilts upward toward Wyoming. This lack of detail produces a problem for climate researchers. The Cascades create a major rain shadow that makes the eastern half of the Pacific Northwest significantly drier than its western counterpart. That rain shadow needs to be accounted for. Detail needs to be put back on the map. Luckily there are two solutions: regional climate models and a process called downscaling.
Regional Climate Models (RCMs)
Regional Climate Models (RCMs), as their name suggests, model climate for specific regions. RCM simulations tend to be more detailed than global climate model (GCM) simulations. Because of this, RCMs are able to represent important regional features—such as mountain ranges and corresponding rain shadows—that might otherwise get missed in lower resolution GCMs. However, RCMs have several major drawbacks to GCMs. One important drawback is also their strongest selling point: their high level of detail.
RCMs create higher resolution grid cells than most GCMs. And while that’s great, high-resolution simulations also demand a lot of computing power. The result: RCMs are often only run once for any given experiment. (A single simulation is called a run.) That’s a problem because climate model simulations are at their heart experiments. As with any experiment, you want to run your experiment again and again (and then again some more) to make sure the results you initially received weren’t just some kind of fluke.
CIRC researchers have gotten around RCMs’ computing/single-run problem by employing distributed computing that uses idle computing time on volunteers’ computers to run additional simulations. (CIRC researchers accomplished this through a project called weather@home, which itself is part of larger distributed computing effort called climateprediction.net. Feel free to sign up here and join the effort. For more on this, see the CIRCulator article “Improving Regional Climate Models with Citizen Science.” ) A drawback of distributed computing is that it isn’t available to everyone.
RCMs have another drawback when compared to GCMs: they need to be “nested” inside GCMs to account for the larger, global processes outside their cells. This too demands computing power. The reason, in part, is due to the fact that researchers prefer to use more than one type of GCM in what’s called an ensemble. Like running multiple simulations of a single RCM, running GCM simulations in an ensemble is essentially like running your experiment over and over and tends to produce better results.
For all these reasons, CIRC’s research has tended to employ GCMs not RCMs, to model local climate. So how, you might ask, do we overcome GCMs’ lower resolution problem? Enter downscaling.
Downscaling takes the coarse (low-resolution) data from GCMs and turns it into high-resolution data that accounts for local landscape features and local climate. The type of downscaling CIRC researchers tend to employ is called statistical downscaling. Statistical downscaling works by, in effect, “training” (another term of art) GCM data to statistically fit with local historical climate data. Getting our downscaling right accounts for a big part of what we do at CIRC.
The GCMs we use in our statistical downscaling are not created in-house. Instead, they come out of the fifth phase of the Coupled Model Intercomparison Project (CMIP5), an international effort to provide consistent output from the world’s GCMs. CMIP5 acts as a kind of giant library for the many GCMs, their simulations, and resulting data of the Earth’s climate made available by the dozens of modeling groups around the world.
CMIP5 model data in hand, CIRC researchers then perform our own vetting process, selecting the models and their data that depict the Pacific Northwest’s climate the most accurately. We then downscale the data resulting from the CMIP5 GCM runs to statistically fit our region and its many geographic quirks. For instance, when CIRC participated in the PUMA project, we downscaled GCM data for the specific watersheds used by the Portland Water Bureau and Seattle Public Utilities, the Pacific Northwest’s two largest municipal water providers. This helped the two utilities understand how big picture climate changes projected by GCMs would play out locally and affect their millions of customers. (In another project, we went a step further, downscaling GCM data for the entire continental United States. For more on how this works, see the story “Back to the Futures” in Northwest Climate Magazine. Or, go to our Publications.) Which brings us to the subject of other computer models.
Other Computer Models
With downscaling data in hand, we feed those data into hydrologic and vegetation computer models. These computer simulations provide us the ability to simulate local climate impacts (e.g., determining the likelihood of wildfires or flooding in a region given the trends in warming and precipitation gleaned from the GCMs). CIRC researchers use a similar, though substantially different, set of computer models to simulate future changes along our coast. (For more info, see Coastal Hazards & Extremes.)
While our work with GCMs entails simulating climate to the year 2100, CIRC researchers also focus on what’s happening to our climate now. Several of our Climate Tools—including Climate Engine—employ near real-time simulations of the climate that give our region’s farmers and other important decision makers essential information about drought and other impacts that happen in the here and now.
Li, Sihan, Philip W. Mote, David E. Rupp, Dean Vickers, Roberto Mera, and Myles Allen. “Evaluation of a Regional Climate Modeling Effort for the Western United States Using a Superensemble from Weather@home.” Journal of Climate 28, no. 19 (2015): 7470–7488.
Rupp, David E., John T. Abatzoglou, Katherine. C. Hegewisch, Philip. W. Mote. “Evaluation of CMIP5 20th century climate simulations for the Pacific Northwest USA,” Journal of Geophysical Research: Atmospheres 188 (2013). doi:10.1002/jgrd.50843.