Climate forecasting and climate change projections are currently made with general circulation models (GCMs) that integrate the complex interactions of atmosphere-land-ocean-ice systems to simulate the Earth's climate and make projections of its future change on timescales from decades to centuries. This report reviews the current state of GCMs, including their approaches, limitations, uncertainties, and performance. The report also describes various downscaling techniques used to scale GCM output from coarser spatial and temporal resolution to finer resolution.
The Earth's climate is a highly complex and nonlinear system, and it is difficult to fully understand and simulate how climate is changing in space and time. With the advent of high-speed computers and a better understanding of the climate system, numerical modeling of atmospheric and oceanic circulations has allowed climate forecasting and climate change projections to become increasingly skillful. Owing to the complex and chaotic nature of the global climate system, there are limitations in the degree to which GCMs are able to reproduce the Earth's climate. It is essential that decision-makers be aware of these limitations and the general strengths and weaknesses of GCMs, prior to using the products derived from them.
To review the current state of GCMs and downscaling techniques and recommend best practices in their use
The project team described the state of the art in GCMs and discussed the sources of uncertainty in projections based on them. The team explained how GCMs are processed to reduce their inherent uncertainties, correct for known biases, and apply to finer spatial and temporal scales. They also made some recommendations on best practices in the use of GCM models.
The "skill" of a GCM is a term used to measure the ability of a model to reproduce observed climate. Skill is assessed by comparing output from GCMs from the 20th century with observations from the 20th century. The more skillful the GCMs are in reproducing/simulating the 20th century, the more confidence there is in future climate projections. Based on the many assessments that have been made of GCM skill, some conclusions can be reached:
- GCMs are more skillful at predicting temperature than precipitation.
- GCMs are more skillful at predicting mean climate conditions than predicting climate extremes or variability.
- There is more skill in longer averaging periods.
- The projections from GCMs are most relevant when considering their long-range implications, such as for the second half of the 21st century.
GCMs are limited in their ability to reproduce actual climate observations due to several intractable sources of uncertainty, which include the magnitude of future greenhouse gas (GHG) emissions, computational limitations, an incomplete understanding of atmospheric processes, and the inherently chaotic nature of the atmosphere. Due to the coarse spatial resolution of GCMs and their low skill in representing daily and monthly rainfall and temperature, GCM results typically require additional computational steps before being used in impact studies or for adaptation planning. This "post-processing" typical consists of bias correction to reduce known systematic errors and downscaling to produce higher spatial and temporal resolution output. Temporal downscaling is usually required because GCMs do not produce realistic daily climate data, especially in the case of precipitation. Spatial downscaling is required because the course resolution of the GCMs precludes them from representing topography, land use, and land cover features that influence local climate. There are two general categories of downscaling approaches: statistical downscaling, which involves the use of empirical relationships between coarse scale GCM output and higher resolution observations, and dynamical downscaling, which relies on higher resolution regional climate models (RCMs) that use GCM output as the input or boundary conditions to simulates the climate over a smaller region. Each category has associated strengths and weaknesses, although typically statistical downscaling performs as well or better than dynamical downscaling and is easier and more economical to use.
Climate models can provide substantial information about the possible climate changes that can be expected in the years ahead. However, for that information to be useful for an impacts assessment or for adaptation planning, careful consideration must be given to what information is gathered and how it is used. In the use of GCM data, an ensemble approach is recommended. This means that multiple runs from multiple GCMs should be used to best characterize the uncertainty associated with the models and the climate system itself. Downscaling is an appropriate technique to scale the GCM output to finer spatial and temporal resolution, but it should be used only when there is consensus on the direction of change in the GCMs, e.g., the models tend to project an increase or decrease in precipitation. The selection of downscaling methodology should be based on the particular application in terms of the variables of interest, timeframe, and spatial resolution required; the existence of previous studies; and the availability of historical observation data.