Validation of the GISS GCM: A Study of Ocean and Climate Modeling
Global warming is hot issue, pun possibly intended, getting much attention from politicians, scientists, and citizens alike. Current research shows an increase in Earth's global mean temperature in the last century of approximately 0.5°C - 0.7°C (IPCC 1995), particularly since the 1970s (Sagan 1997) and predicts that this rise will continue through the next (Titus and Narayanan 1995). The changes have been primarily attributed to anthropogenic effects, such as the burning of fossil fuels, which has caused increase in carbon dioxide, in turn trapping heat within Earth's atmosphere and warming the planet (IPCC 1995). Further temperature changes may have far-reaching effects, as they are predicted to intensify the strength of the hydrologic cycle and possibly cause increased rates of droughts and floods. Sea level rise due to thermal expansion of the upper layers of the ocean and melting ice sheets may endanger low-lying areas around the globe (Titus and Narayan 1995, Hansen and Lebedoff 1988, Lemonick 2001).
To understand the 'hows and whys' of global warming, we must look at the processes that drive climate, both on a regional and global scale. It is more feasible to study than weather on a long-term basis, because while weather fluctuates on a daily basis, being chaotic and nearly impossible to predict for the long-term, climate remains relatively stable over long periods of time. There are many variables that impact climate, including carbon dioxide and methane levels in the atmosphere, sea surface temperatures, solar radiation, and volcanic aerosols. However, scientists are unsure of the extent of the impact of each of the variables. For example, the oceans play a critical role in climate, but the extent of their impact is still a focus of current research.
Scientists at GISS have developed models, GCMs, or General Circulation Models, that simulate the driving processes of the ocean and atmosphere, and can then make predictions of future climates based on variables such as the carbon dioxide level in the atmosphere. Current models cannot necessarily make accurate predictions of the future, and our study sought to examine the accuracy of a particular model. It is important that the climate models developed make predictions that are reasonably accurate, so that they may be a source of information about the effects of atmospheric changes, such as the increase in carbon dioxide levels, on climate.
The focus of our research was to evaluate a GCM and to identify its strengths and weaknesses. The particular model we are using is an atmospheric general circulation model (AGCM) with three treatments that differ in the way that the ocean is included. In order to understand climate, it is important to realize that the oceans have a profound impact on the atmosphere. In the first atmospheric model, the sea surface temperature is specified; therefore, variations in ocean temperature in response to changes in the atmosphere are prohibited. The "q-flux" treatment allows the model to predict the ocean temperature, and reflects the heat budget for the upper ocean. The ocean currents are prescribed, therefore the outcomes of this treatment reflect changes in heat due to the dynamic balance of insolation, radiation, and evaporation. In other words, this treatment allows heat to be redistributed between the ocean and the atmosphere. The "coupled" treatment of the model is a compilation of both the AGCM and the Oceanic General Circulation Model (OGCM). This treatment allows forcings in the ocean and in the atmosphere to respond to one another. In this way, this treatment is most realistic, most like real-life, but may not be the most accurate, as climate is a chaotic entity that is extremely difficult to predict.
Ultimately, climate is driven by the unequal distribution of energy from the Sun on Earth. This inequality is due to the tilt of the Earth on its axis, as well as the Earth's rotation. Energy from the Sun reaches Earth in the form of radiation, and is then either reflected by the clouds, or land and water on the surface of the Earth, or absorbed and re-radiated. Primarily, energy reaches the Earth in the form of light (short wave radiation), it is then re-radiated at lower frequency, mainly as infrared radiation. Infrared (heat) radiation can be trapped by the gases in the Earth's atmosphere, heating the planet by what is known as the Greenhouse Effect. Energy is redistributed on Earth primarily by convection within the atmosphere and the oceans. These resulting currents drive weather systems, as they result in rising and falling air, changes in humidity, temperature, and precipitation.
The many variables that affect weather and climate have been built into the computer model and they can be analyzed individually using climatologies, images that map quantities of a particular variable on a global scale. The variables that we have chosen to analyze in detail are those related to hydrological and radiative data, including cloud cover, precipitation, specific humidity, incident solar radiation, albedo, and net thermal radiation. It is important to study the radiative variables because of their fundamental influence on climate; solar energy and the Earth's absorption and re-radiation of this energy are ultimately the driving force. Hydrological influences are also important, particularly to humans, because these variables are often hardest predict, but affect local weather and climate quite dramatically.
Our main objectives in this investigation are to determine the accuracy of the model in making predictions, to study the roles of the variables in climate and the relationships between them, and to assess the extent of the impact of the ocean on global climate. It is our hope that our findings will help to determine which aspects of the model need improvement; these improvements should in turn lead to more accurate predictions. If we can help to develop an accurate model of climate, we can use this model to predict climate changes, and to predict and anticipate natural and anthropogenic effects on climate. As James Hansen (1994) said, "A climate model is a tool that lets us experiment with a facsimile of the climate system, helping us think about and analyze climate, in ways which we could not, and would not want to, experiment with the real world."
Our research was a comparison of a 5-year climatology of the GCM with three ocean treatments with observational data, with the aim of understanding how well hydrological and radiative processes are being simulated by each ocean treatment. The data to be analyzed are simulated data generated by the three treatments of the General Circulation Model (GCM) (atmospheric, q-flux, and coupled), as well as data from observations. The three treatments of the AGCM are distinct in their treatment of the ocean. In the atmospheric model, there is a fixed Sea Surface Temperature (SST). Because 70% of the Earth's surface is ocean, having a fixed Sea Surface Temperature will tend to make outputs of this version of the model closest to the observation data. However, this model is only useful if the SST is known. It is not useful for making future predictions, and does not allow for changes in the energy exchanges between the ocean and the atmosphere. In the Q-flux model, a fixed amount of energy is exchanged through ocean currents. This mixed ocean layer treatment allows the model to predict the ocean temperature; changes in the ocean heat content occur due to changes in evaporation, radiation, etc. The coupled model is a combined AGCM (Atmospheric General Circulation Model) and OGCM (Oceanic General Circulation model). In this treatment, variables are allowed to vary in response to one another, for example, SST in response to atmospheric forcings. While the coupled model is the most physically realistic, because it allows variables to change one another as they do in real life, it is also the least accurate, as only the starting values are prescribed. However, in order to make accurate predictions of future climate, we will need to use a simulation like the coupled model.
Observational data were gathered from satellites, airplanes, radiosondes, land and sea based stations, and were interpolated for areas between stations. Because neither satellites nor surface based stations can make comprehensive global measurements, it was important that we compile data from a number of sources. Satellites, however, are considered to be the best data source (Jones and Wigley 1990). Data were collected for many variables including radiative variables (such as temperature, incident solar radiation, albedo, and re-radiation) and hydrological variables (such as cloud cover, snow cover, precipitation, and barometric pressure) for the summer and winter seasons (in January and July). The months of January and July were selected because they represent seasonal patterns and extremes. Some variables, such as surface air temperature, can be measured directly. Others must be resolved from other values. Where possible, data were collected at three different pressure intervals within the atmosphere: at the surface, at 850mb, and at 500mb (approximately 5.5 km). The observation data used were for a 5-year period in the 1980s.
The first phase of our research was to qualitatively compare the simulated data generated by the three treatments of the model with observed data from different information sources. In order to accomplish this task, global climatologies (averages over a significant numebr of years) were produced . These climatolgies were visualized using the NMAPSx program, which uses a color scheme to manufacture plots superimposed over a map of the world. Climatologies were produced for each variable (e.g. surface air temp, SLP, OLR, etc.), FOR both the observation data and the simulated data for each version of the model. These visualizations use colors to indicate differences in the values for the particular variable, and allow for simpler qualitative analysis of global patterns.
We analyzed the data qualitatively by visually comparing the observed data image to the image produced by the SST model. We first assessed whether the images were globally realistic for each variable, and whether gross features were indicated (e.g. showing the Intertropical Convergence Zone (ITCZ)) and in their expected location. Next, we looked for global patterns, and identified world regions of great consistency and areas where the model and observation data sets diverged. After examining each variable individually, we compared our findings for different variables, and formed an integrated summary to select regions for more in depth study, looking for patterns such as regions of great accuracy or inaccuracy across all variables. We then related these findings to unique characteristics (i.e. landforms, meteorological aspects, ocean currents) of the regions.
To quantitatively compare the sets, we used the Nmapsx utility to create difference maps for a number of variables. We created difference maps for the variables of surface air temperature, total cloud cover, sea level pressure, outgoing thermal radiation, and absorbed solar radiation. We then used the maps to highlight the differences between the models and observations, and to assess whether the differences were due simply to interannual variability within the data, or due to inaccuracies in the model. Using these maps, WE focused on the regions of the Amazon in South America, the North American Continent, and the Tibetan Plateau in Asia, and tried to find connections between the variables that might explain why there were differences between the model and observations. My focus was on the region of the continental (48 states) and coastal United States.
To compare the versions of the model to one another, we utilized difference maps along with student's T-tests to determine if the differences between the Q-flux and coupled models and the SST model were statistically significant. Significance tests indicate whether the differences are indeed due to problems in the model, rather than due simply to interrannual variability.
Figure 1 shows the global total cloud coverage in the observational data and in the SST treatment of the model for July. After comparing the SST model to the observation data (figure 1), we notice that gross features such as the ITCZ and Mid-Latitude Storm Tracks are represented as expected, e.g. we notice bands of cloud cover just north of the equator in July, and south of the equator in January. The Southeast Asian monsoons are indicated as areas of intense cloud cover, while the Sahara desert region contains few clouds. In the Southern Hemisphere, cloud cover is more uniformly distributed in a band because there is less land. While global patterns were well represented, as shown in the above figure, the model tended to be less intense than observations for many variables. We also noted expected agreement between variables on a global scale. For instance, in more arid regions (areas with low specific humidity and precipitation), we saw greater outgoing thermal radiation, and higher sea level pressures. Areas of intense low cloud cover also tended to have more precipitation and cooler surface air temperatures as anticipated.
Figure 2 is a difference map; showing the difference between the SST model and observations of total cloud cover for July in the continental United States. We examined the SST model vs. the observations for the summer season in the region of the continental and coastal United States (lower 48 states), the most dramatic differences between the simulated model data and the observation data are seen in the variable of cloud cover. It seems that the misrepresentations of total cloud cover in the model tend to cause simulations of other variables to be off target as well. For example, in the Gulf of Mexico, the total cloud cover is overestimated by 8.2% to nearly 40% (Figure 2). Consequently, precipitation is also overestimated, by .5 mm to over 10 mm. In addition, too much cloud cover would lead to lower net solar radiation; in a large portion of the Gulf, it is underestimated by over 100 W/m2, because intuitively if there are more clouds, less sunlight can penetrate to the surface.
Model images of the Pacific Coast region show highly underestimated cloud cover, by approximately 25% to 75%. This leads to a higher net solar radiation than expected (by 12 W/m2 to almost 130 W/m2), because solar radiation is not reflecting off of clouds, and therefore more radiation can get through into the atmosphere. More radiation being absorbed, and then re-radiated, would lead to higher than expected surface air temperatures, shown in model images as much as 16.2°C higher than the observation data (Figure 3). There is no significant difference, however, in modeled vs. observed precipitation, perhaps because July is a dry month for this region.
An important finding in our in-depth study of United States climate is the dramatic impact that clouds can play in climate. One of our objectives was to investigate the interrelationships of the climate variables, and how they affect model functioning. Simulating cloud cover in the model is a very complicated task, and this example clearly demonstrates its importance. However, since there are multiple feedbacks in the climate system concerning clouds, it can be very difficult to track down the particular cause of model errors in a particular region. In the case of North America, we observed that an inaccuracy in cloud cover seemed to lead the model astray in simulating many other variables, and therefore misrepresenting many aspects of climate in the United States. Although this was brought out using simulated (model) data, were able to see the effects of cloud cover on many other variables including thermal radiation, solar radiation, and surface air temperature. Previous versions of the SST model also seemed to have difficulty accurately depicting cloud cover.
Our other major objective was to assess the accuracy of the model in making predictions. We found that on a global scale, the SST model is fairly consistent in modeling gross climate features. This was expected, as discussed earlier, because in this treatment of the model, the sea surface temperature is input directly. Since sea surface encompasses 70% of the Earth's surface, the model should model climate quite accurately. We did notice that the SST model tends to exaggerate some features, such as the Southeast Asian monsoons, or the Sahara Desert, in terms of the intensity of many variables, making these features more extreme in the model than what is actually observed. Therefore, if the model developers make slight changes in the way these regions are represented, the model may represent them more accurately.
Two concerns that arose in our research involve uncertainties in the observations used, and uncertainty in the workings of the model. Although we used observation data from a number of sources, incorporating both surface based and satellite based instrumentation, some of the data was interpolated data, as there are not weather stations or satellites that can cover all parts of the world. We are unsure of how accurate the observations are, particularly in remote ocean regions, or in areas that are sparsely populated, and for some variables the data from different sources seemed to conflict. This became a limitation when assessing the differences between the simulated model data and the observation data. It is important to have confidence in the observations before deciding that differences are due to misrepresentations by the model. Also, since we ourselves did not write the program for the treatments of the model, we may draw conclusions based on assumptions that may not be reflected in the model's programming.
The importance of creating a globally valid fixed-SST model is that it is the next step in developing a model that will accurately predict climate change. Our research objective was to identify areas, whether regions or variables, that the model does not simulate accurately so that they may improved. Questions remain about what is causing clouds to be misrepresented by the model, and why the model intensity seems to be exaggerated in many variables. Once these questions can be answered, and updates can be made to the SST model to make it simulate climate even more accurately, we can again compare the SST model to the coupled model. A future version of the coupled model can be used to predict the impacts of future climate change, a topic that could influence our lives dramatically in the future.
Gornitz, V., S. Couch, and E.K. Hartig, 2000. Climate change and the coast: Impacts in the New York City metropolitan region. Eos, 81, S73.
Hansen, James, 1994. Public Understanding of Global Climate Change. Comments at Symposium in Honor of Carl Sagan's 60th Birthday. Cornell University, Ithaca, NY. 14 October 1994.
Hansen, J, and S. Lebedoff 1988. Global Surface Air Temperatures: Update through 1987. Geophysics Res. Lett: 15, 323-326.
Intergovernmental Panel on Climate Change, 1996. Climate Change 1995: The Science of Climate Change. New York: Cambridge UP.
Jones, Philip D. and Tom M. L. Wigley, 1990. Global Warming Trends. Scientific American.
Lemonick, Michael D., 2001. Life in the Greenhouse.
Sagan, Carl 1997. Billions and Billions — Thoughts on Life and Death at the Brink of the Millennium. New York: Ballantine.
Titus, James G. and Vijay Narayanan, 1995. The Probability of Sea Level Rise. U.S. Environmental Protection Agency.