Our knowledge of future climate conditions is based on experiments performed with climate models (also known as atmosphere-ocean general circulation models). This section describes how to judge and utilize the output from climate models in regional-scale vulnerability assessments, and where to access climate model information.
Climate model output can be an integral component to a vulnerability assessment (and the subsequent adaptation decisions). Outputs can provide information regarding how much or how fast the climate could change. However, it is important to establish what types of information climate models DO and DO NOT provide. This section will discuss emerging issues regarding scenario selection, downscaling, and uncertainty in the context of regional-scale vulnerability assessments.
What is a climate model?
A climate model is a mathematical representation of the climate system:
...climate models are used to simulate how...changes in GHG [greenhouse gas] emissions and other climate forcing agents will translate into changes in the climate system. Climate models are computer-based representations of the atmosphere, oceans, cryosphere [ice and snow], land surface, and other components of the climate system. All climate models are fundamentally based on the laws of physics and chemistry that govern the motion and composition of the atmosphere and oceans. (National Research Council, 2010; bracketed phrases have been inserted)
The models' simulations for the 21st century are called projections. These projections are designed to help us understand how the addition of greenhouse gases to the atmosphere might change our climate. Models are intended to be heuristic tools, illuminating how many processes (e.g., the way in which oceans transport heat around the planet; the strength and location of the jet streams) might respond to the addition of these greenhouse gases, and the subsequent warming that occurs. Climate models are not intended to be "prediction machines" that reveal the precise future conditions in a particular location.
It is also important to note that climate models, by themselves, do not yield information about the impacts of climate change. They simply provide simulations for future temperature and precipitation, which can be converted into statistics (e.g., average seasonal temperatures, annual return-frequency for days with temperatures above 100°F) that can be compared to similar statistics representing the current climate. It is essential that users understand the types of climate information that are most important to the performance of their respective transportation networks. Performing a sensitivity analysis (see "Value of Examining Historical Climate" section above) is a way to identify the types of climate information (i.e., which measures of temperature and precipitation, and on what time scale) that correspond to the quality and reliability of service, or to damage and repair of assets.
Whenever a climate model is "run" into the future, a set of assumptions regarding the future trajectory of the planet's greenhouse gas emissions and other climate-forcing agents (e.g., aerosols) is used as inputs to the model. These assumptions chart a path forward for the world's population growth, economic growth, and rates of technological development and transfer of technology. These assumptions, taken together, constitute an emissions scenario.
The Intergovernmental Panel on Climate Change (IPCC) has developed a set of standard emissions scenarios that are used in climate models. These scenarios are typically designated by a set of letters and numbers (e.g., "A1B", "B2") that communicate information about the various assumptions regarding future population, economic growth, and technological development that relate
to the scenarios' respective emissions trajectories.
It is important to note that the likelihood of occurrence of any scenario has not been assessed - there are no probabilities assigned to the individual scenarios. The scenarios simply represent possible future pathways. They do not necessarily "bracket" what future conditions could look like, nor do they constitute what is most likely to occur.
Climate models render the Earth in a series of "pixels" or grid boxes (Figure 5) that are several degrees of longitude each side (roughly 60-200 miles, depending on the geographic location and the particular model). Within each grid box, calculations for meteorological variables are performed, and the flows of mass and energy between grid boxes are tracked.
The resolution of current climate models provides a coarse view of the land surface. For example, details of coastlines and mountain ranges that are smaller than the size of the grid boxes are not directly incorporated into the model's calculations. Similarly, aspects of meteorology and climate that occur on these relatively small spatial scales (e.g., structure of weather fronts, the evolution and properties of clouds) are not always represented well.
Downscaling techniques have emerged as a way to transform the output from climate models to smaller spatial scales, often to grid boxes that are a quarter or an eighth of a degree of latitude and longitude on each side (between about 10-20 miles on a side). There are two general methods of downscaling:
Statistical downscaling: Empirically-observed relationships between observed climate (at high resolution) and modeled climate (at a coarse scale) are applied to future projections. These relationships act as a "transfer function" that allow small-scale information to be added to the patterns generated from the coarse-scale global climate model. Statistical downscaling techniques are based on the assumption that historically observed relationships between larger-scale climate patterns and the smaller-scale patterns (colloquially, these small-scale patterns are sometimes referred to as microclimates) will not change as a result of climate change.
Dynamic downscaling: The outputs from a coarse-scale, global climate model are used as inputs for a climate model that operates in a smaller spatial domain (typically sub-continental). This smaller-scale model works much like a global climate model - equations describing the physics and chemistry of the atmosphere are solved in each of the grid boxes on the smaller domain, with the solutions forced to be roughly consistent with the large-scale characteristics of climate that have been simulated in the global model.
Statistical downscaling techniques are computationally less intensive than dynamic methods. Although there are many algorithms for applying statistical downscaling, some of the more popular methods (e.g., bias correction statistical downscaling, constructed analogs) often yield results that are similar to one another.
The use of dynamic methods is an active area of research and the types of information that can be gained by dynamic downscaling are still being established and debated. Given the computational requirements, experiments using many combinations of global models and regional models are time-consuming and expensive; hence, these investigations are likely to continue for the coming years.
It is important to note that downscaling rarely improves the information about climate change that is derived from the coarse-scale, global models. Although downscaling may reveal small-scale patterns of interest (e.g., larger amounts of rainfall on windward slopes of mountains, relative to nearby flatter terrain), the difference in the changes for the future may be less significant (e.g., the change in precipitation for the mountainous areas and the flatter areas may be quite similar). Similarly, downscaling the output from a suite of coarse-scale models will not necessarily result in a tighter range of projections for a particular area.
Which models, scenarios, and spatial scales are right for you?
Unfortunately, the answer to this question in not simple or straightforward. A few important points to keep in mind:
Although there may not be a well established set of "best practices" for selecting models, scenarios, and the spatial scale of future climate information, knowledge of the sensitivity of the transportation system can be used to narrow some of these choices.
Projection of daily-scale extremes
Some of the most important impacts of weather and climate on transportation assets and services occur during relatively short-lived events, including severe storms, floods, and heat waves. However, model abilities to simulate the daily-scale statistics of climate in a particular region are limited. Quantitative information derived from daily-scale statistics should be scrutinized and used with care. For example, changes in the number of extremely hot days occurring in future decades can span a wide range - projections for the end of the 21st century in parts of California exhibit increases in the frequency of intense hot days occurring in a year ranging from a doubling to increases by a factor of 8 or more. For heat waves lasting 5 days or more, the increases in frequency cover an even wider range, with several models projecting 20- and 30-fold increases (Cayan et al., 2009). Although all models and scenarios project a warmer future and an increase in the frequency and intensity of heat waves across California, there is significant uncertainty as to how this warming translates into daily-scale changes in weather at specific locations. Similarly, for changes in precipitation, many models project a future with heavier downpours for many regions in the U.S.; however, the magnitude of the changes can vary widely both across scenarios, and even across models using the same scenario.
Although there is no way to "eliminate" the uncertainty associated with projections of daily-scale extremes, the mere fact that the model results involve uncertainty about the future does not need to be a barrier to making useful conclusions about daily-scale extremes. Two potential techniques include:
A variety of research institutions across the world conduct model experiments, using high-end computing resources. The output of these experiments are made available online (see Text Box on Online Resources) for download.
One of the most commonly used sets of model experiments is the Coupled Model Intercomparison Project (CMIP). The CMIP models act as a benchmark for climate research, and form the core of the assessment of future climate in the IPCC reports. These models that are part of CMIP meet a certain set of requirements regarding their technical specifications, their performance, and the types of experiments to which they've been subjected.
Although it is possible to access the data from any of these climate models, manipulating the data can be a challenge. The data formats (often Network Common Data Format, NetCDF) are not easily read-able by typical desktop software (e.g., Excel) or Geographic Information System (GIS) software packages. In addition, these data files are often extremely large and contain information on a wide range of variables (e.g., temperature, wind speed and direction, heights of pressure surfaces) for the entire globe. However, many of these variables would likely be extraneous when conducting a vulnerability assessment for transportation assets in a specific location.