PREDICTING CLOUD POWER WITH COMPUTER MODELS
The main tools used by climate scientists to predict future climate change are computer models. These models work by integrating data on various atmospheric variables, such as atmospheric levels of greenhouse gases, cloud conditions and many other variables, for a particular time period. Then, based on computed interactions between these variables, the models predict the resulting climate for the represented time period.
Because it is impossible to represent all atmospheric conditions over every inch of the planet, most climate models present a simplified version of the Earth; they represent the Earth by dividing it into grid boxes or "grid cells" -- usually each about the size of the state of Delaware. Each of these cells is represented as a single, unbroken uniform area. This means that conditions across each cell--including cloud conditions--are only approximated, generalized or averaged; sub-regions within cells that deviate from such cell-wide approximations, generalizations or averages are not directly represented.
So even though a Delaware-sized cell would be big enough to hold thousands of clouds, a climate model would not represent each of these clouds. Instead, it would treat the entire cell as a single "box" with an "average" cloudiness (a fraction between 0 and 1) representing how much of its total area is covered by clouds. The cell's other cloud characteristics would also be represented statistically in terms of their cell-wide averages.
Climate models identify numbers representing each cell's cloud characteristics by using simplified models of clouds. One recently developed approach involves embedding a cloud model in each cell of a climate model, says David Randall, director of the Center for Multiscale Modeling of Atmospheric Processes (CMMAP)--a Science and Technology Center that is funded by the National Science Foundation and is devoted to improving the representations of cloud processes in climate models. In other words, this approach involves integrating cloud data from small cloud models representing each "cell" into a larger climate change model that simulates climate for the entire area covered by the cells.
It is difficult to create accurate climate models and cloud models. Obstacles include the many enigmatic interactions between cloud processes operating at various scales (see The Enigma of Cloud). Unfortunately, inaccuracies incorporated into representations of cloud processes operating at any particular scale cascade throughout models, influencing predictions of processes operating at other scales.
Other modeling challenges are created by the sheer number of particles and droplets in clouds. After all, even a small cloud that has a volume of only one cubic kilometer may contain more than 1017 droplets! It is currently impossible to accurately track such large numbers of particles. Another challenge is representing the many and varied types of cloud and precipitation particles carried by clouds. "Just think of all the different shapes of snowflakes; how these shapes are represented in models can have important effects," says Hugh Morrison, a scientist at NCAR.
This means that it is currently impossible to simulate the ever-changing structure of clouds in a completely realistic way. Nevertheless, by averaging cloud characteristics over grid cells, models try to produce approximate simulations of present and future conditions in the most realistic ways possible.
Scientists need to work with a number of different models to predict climate because models that are currently available all have particular strengths and weaknesses. How do scientists evaluate models and determine which aspects of climate they each predict best?
By comparing the model's predictions of future conditions to observations, as they become available. This is done by using the climate model to simulate future conditions. The more that simulated climate conditions would resemble actual climate conditions as they develop, the more the model would be trusted to generate realistic, long-term predictions of those climate conditions.
By comparing the model's simulations of past and present conditions to recorded observations. This is done by using the climate model to recreate climate conditions for a period--such as the last 100 years to the present--for which records exist. The more the simulated climate conditions would resemble actual conditions, the more the model would be trusted to generate realistic, long-term predictions of future conditions.
"No model is perfect," says James Hurrell, a senior scientist at NCAR. "They all have shortcomings." Therefore, there is uncertainty in even the best climate models--partly because of the difficulties of representing clouds in models.
Scientists are continually improving climate models by identifying discrepancies between predictions or observed conditions and simulations made by climate models, and then working to eliminate these discrepancies. "We are constantly looking for what is wrong with climate models (not what is right with them) in order to determine how we can improve them," says Randall. They do so by applying the laws of physics into models, and by generating and inputting into models additional relevant data on various climatic characteristics, including cloud conditions.
THE COMMUNITY EARTH SYSTEM MODEL
One of the most important climate change models currently used by scientists is the ever-evolving Community Earth System Model (CESM), which allows researchers to conduct fundamental research into the Earth's past, present and future climate states. (The CESM is managed by NCAR with funding from NSF and the U.S. Department of Energy.)
The latest version of the CESM, which was released in June 2010, represents a major improvement over its predecessors. It is currently contributing to an ambitious set of climate experiments that will be featured in the fifth assessment report of the IPCC slated for release in 2013.
The CESM is continually improved by what NCAR's James Hurrell, chair of the CESM scientific committee, describes as "a collective decision." Anyone can join any of the 12 working groups that develop and apply the model, says Hurrell. Through the working group structure, ideas by community scientists are suggested, tested and compared to the current model so that their strengths and weaknesses can be assessed. Then, all the CESM working groups together decide whether to incorporate suggested changes into the CESM."
THE FUTURE OF CLIMATE MODELLING
Climate models cannot include as much detail as weather models, and climate projections are made on regional to global scales rather than on local scales--covering far larger spatial scales than do weather models. Also, models simulate global climate over years, decades or millennia--covering far longer periods than do weather models.
Currently, the shorter the time period or the smaller geographic area covered by a prediction generated by large-scale climate models, the less reliable it is. "But climate scientists get up every morning and come to work to try to improve the climate models," says Randall. As part of this effort, CMMAP is currently working to create models whose grid cells are small enough to represent individual clouds.
Also, NSF is currently advancing the development of higher-resolution models by sponsoring--together with the U.S. Departments of Energy and Agriculture--a $49 million joint research program to produce high-resolution models for predicting climate change and its resulting impacts. Called Decadal and Regional Climate Prediction Using Earth System Models (EaSM), the program is designed to produce models that are significantly more powerful than existing models. They will be able to generate predictions of climate change and associated impacts over shorter time periods and over smaller geographic areas than current models are capable of.
But despite the need for higher-resolution models with smaller grid cells, "the need for lower-resolution models will continue into the foreseeable future," says Morrison. Why? Because some applications might require models to simulate long time periods, like thousands of years, that are currently impossible to simulate with the higher-resolution models.