This page's content is no longer actively maintained, but the material has been kept on-line for historical purposes.
The page may contain broken links or outdated information, and parts may not function in current web browsers.

Research Results

Evaluation of the GISS GCM ModelE


The global mean temperature on the Earth has steadily been rising since the late 1970's (Sagan, 1997) and is projected to continue rising throughout the next century (Titus, 1995). IPCC (the Intergovernmental Panel on Climate Change) states that during this century, the global mean temperature has risen by 0.6°C, and could continue to rise at a slightly faster rate, 0.2°C (0.36°F) to 0.3°C (0.54°F) this decade (IPCC 2000). This could eventually lead to rapid climatic change during the next decade or half century (IPCC, 2000). The increase in greenhouse gases in the atmosphere, a result of the increased burning of fossil fuels, is leading to global warming (IPCC, 2000). This in itself may not seem significant, but one must consider that global temperature increases of only 3°C (5.4°F) could cause a significant change in the Earth's energy balance, substantially affecting climate (Lemonick, 2001). For example, during the last ice age, the ocean temperature was only 2.3°C (4.1°F) lower than it currently is, and 6.5°C (11.7°F) lower than current land temperatures (Gross, 1996).

An example of one consequence of this rise in temperature is a rise in the sea level of the oceans. As temperature increases, the water in the upper layer of the ocean expands, and occupies more space due to thermal expansion. One projection by the U.S. Environmental Protection Agency states that by 2050 sea level would have risen 15 cm (5.9 in), and by 2100, the sea level would be 34 cm (13.4 in) above current levels (Titus, 1995). Consequently, many coastal cities such as New York could become severely flooded during peak storm periods, and permanently flooded in low lying coastal areas, e.g. lower Manhattan (Gornitz, 2000).

How are these predictions made? To what extent are they reliable? To answer these questions, scientists have been developing a tool that would allow us to understand more about the complex intricacies of our climate system, and could eventually help simulate future climate change. This tool is called the General Circulation Model, or GCM. These models are important because they enable us to better understand the factors responsible for our current climate, their impacts and interrelationships, and their effect on our climate.

One must be aware that when the term "climate" is used, it refers to weather averaged over long periods of time, e.g. over decades, centuries, even millennia. Weather is defined as the state of the atmosphere at any given time. Some examples of this are rainfall, humidity, and temperature fluctuations, which can change instantaneously. Weather in itself is very 'noisy' -- e.g. very chaotic. Climate, on the other hand, has been relatively stable over the past 10,000 years (Burroughs, 1997). It is influenced by various forcings. A climate forcing is a change imposed on the climate system that alters the balance and exchange of energy between the Earth and outer space. For our purposes in climate modeling, however, a forcing is specifically an external change that alters the energy balance. An example of a forcing is greenhouse gases, such as CO2, O3, and CH4, which alter atmospheric conditions, leading to warmer temperatures as a result.

There are many hypotheses in the scientific community based on different models regarding when changes in the climate will occur and how extreme they will be. For example, how will temperatures increases? What impact will this have on precipitation? Sea level? An example of one such prediction is the IPCC's business-as-usual scenario. IPCC forecasts an increase in the global mean temperature between 0.2°C (0.36°F) and 0.5°C (0.9°F) per decade. Based on this scenario, the global mean temperature would rise approximately 0.3°C (0.5°F) by 2010, 1.5°C (2.7°F) by 2050, and 3°C (5.4°F) by 2100 from current levels (IPCC, 2000). Agriculture could also be affected. Because of the increased temperature, countries in the middle to high latitudes (e.g. the United States and Canada) may see significant gains in crop yields, while poorer countries would become poorer (Sagan, 1996).

Science Background

Not only does our atmosphere substantially influence climate, but the oceans have a significant role in climate as well. For example, ocean currents, such as the Gulf Stream, transfer heat from the tropics to the mid-latitudes (Gross, 1996). In addition, ocean currents make climate in certain places (e.g. the British Isles) warmer than other places of similar latitude (Gross, 1996). The winds of the Earth, produced by differences in atmospheric air pressure, transfer heat throughout the Earth. A warm air mass transfers heat from the warm oceans to land, while a cold air mass transfers heat from the land to the ocean, thereby warming it.

The water of the ocean is capable of holding heat from insolation (incoming radiation from the sun) and the atmosphere for long periods of time due to its high specific heat. Specific heat is a measure of the amount of energy a substance can hold, and of pure substances, water has the greatest specific heat. As solar radiation comes in from the atmosphere, part of that energy is transferred to the ocean, which heats the surface of the ocean, causing the water at the surface to evaporate, forming water vapor. This process releases latent heat (heat present in a body but not detectable) from the ocean. Another type of heat involved in the regulation of climate is sensible heat (i.e the heat transferred from a warm object to a cold object.) Together, the sensible heat flux and the latent heat flux (heat flux meaning the flow of heat) contribute to the regulation of Earth's climate.

The Study

Research at the NASA Goddard Institute for Space Studies in New York City is focused on developing a full coupled ocean model that is reliable. In order to do this, the Atmospheric GCM and the Ocean GCM (both described in greater detail in the Methodology) must both be reliable enough in order to simulate climate based on the interactions between both the oceans and the atmosphere. Coupled models are essential to understanding climate because they enable all the known forcings and interrelationships within the climate system to change.

The General Circulation Model, which is a computer simulation written in Fortran (a computing language used for mathematical applications,) employs physics principles and formulas in an attempt to simulate the processes affecting the Earth's climate. The model is three-dimensional, divides the atmosphere into layers, and is based on the transfer of mass, momentum, and energy. Mass, relative to the model, is defined as water (ocean currents), air, and salt present in seawater. The model assumes there is a budget for each of these factors in order to carry out the calculations required to successfully simulate climate.

The GISS GCM has been recently updated. One change in the model is the coding, which has been improved so that the model runs faster. In addition, the equations that govern the transfer of mass, momentum, and energy within the climate model were modified. An important part of model development is validation studies and research is concerned with contribution to a range of validation studies. Therefore, it is necessary, in order to have confidence in the conclusions of the model (e.g. floods, droughts, changes in agricultural production), to compare the model output for recent times to current conditions. In order to validate the model, many variables must be analyzed to see if the model is doing an acceptable job, and is producing realistic output. It is imperative that the data that the models are producing is as accurate as possible.

There are certain questions that will be addressed in this investigation. They are:

  • How do the model simulations compare with the observations? For what variables are they consistently on target? Consistently different?
  • In what regions do we find the greatest discrepancies? How can discrepancies in one variable be explained by disparity in another? What is the statistical significance of these discrepancies?
  • How do these discrepancies change when considering differences between two extremities (January and July)?

Each version of the GISS GCM differs in its treatment of the ocean. Each gives its own "snapshot" of climate at any given time. To validate the models, first the model output (which is a control run based on 1981 levels of trace gases) must be compared to available observations (from the 1980's). This is done in a qualitative manner. Then, statistical t-tests and difference maps (a plot of the difference between the observation and the model) are used to confirm the observations mathematically. Although the Earth is being looked at as a whole, emphasis is placed on regions which seem to be problematic for the models, such as the Amazon Rainforest, the South Pacific Ocean, the El Niño region, the African Rainforest, the Tibetan Plateau, and Indonesia. My area of specialty was the Tibetan Plateau. Also, the analysis is limited to January and July, because the climate changes the most between these two months, enabling us to focus on extremities.

In order to validate the models, a set of observed data is required. To obtain a global view of climate, we need a variety of measurements. Unfortunately, total confidence can never be placed in this observed data, due to the limitations of any given observation procedure. Each variable that was analyzed in this paper was compared to the output produced by the SST Model. The variables analyzed were: surface air temperature, absorbed solar radiation, precipitation, total cloud cover, and cloud top pressure. Surface air temperature information was obtained from the OORT Project, which used radiosonides to obtain the data. Precipitation data came from Legates, which is a land-based data collection operation. Total cloud cover and cloud top pressure data came from ISCCP (the International Satellite Cloud Climatology Project). Data for comparison of absorbed solar radiation came from ERBE (the Earth Radiation Budget Experiment), which is another source of satellite data.

Each version of the model produces its own output. This output was produced for a certain period of time (five years), a small but sufficient amount of observations to reduce variability. The model divides the world into blocks of 4 degrees of latitude by 5 degrees of longitude. The grid has a resolution of 72 by 36, or 3312 grid boxes in total. The three-dimensional atmosphere within the model is divided into twelve "layers," two in the stratosphere, and ten in the troposphere. This enables the model to also assess atmospheric factors such as storm formation more accurately.

The first version of the GCM is called the Atmospheric GCM (AGCM), or more commonly referred to as the "SST" Model. SST stands for "Sea Surface Temperature," meaning that this model uses the observed, fixed SST to calculate other factors, such as evaporation and heat flux, in order to simulate climate. In general, the SST is obtained from direct observation, is inputted into the model, and is used in creating the model's depiction of climate.

The second version of the GCM is the Mixed Layer, or Qflux model. This model is based primarily on the heat budget for the upper layers of the ocean. It is based on incoming radiation from the atmosphere and the transfer of that energy through currents. Although the effect of these currents is held constant, this model is designed to calculate more factors with less information. This model must take into account factors such as heat transfer from the atmosphere to the ocean.

The third version of the GCM is called the Coupled Model. This model eventually will have the greatest predictive power, because it includes the most physics, and runs with the least prescriptions. This model is our best representation of our understanding of forcings that contribute to the Earth's climate today. It allows the SST, atmospheric conditions, and the effect of currents to change. It is also called the "dynamic" model because it is like our dynamic Earth -- many of feedback mechanisms to do with the ocean react freely and chaotically in the climatic system.

In order to do qualitative comparisons, graphical plots called climatologies were created using an in-house Fortran program for each variable of interest. It uses a color scheme to create plots superimposed over world regions. In analyzing these plots, the first step is a check to assess the reasonableness of the plots, given what is already known about the various interrelationships within climate. Secondly, regions where the model data is similar to the observation data are compared and noted. Finally, regions where the model data differs significantly from observed data are compared and noted.

In order to do quantitative comparisons, two methods were used. One method, difference maps, takes the difference between the observed data and the model data for each version. This enables us to see regions of significant error in the models, but only for factors in which it is certain that the observations are reliable. The other method is the Student's T-test, which is a statistical test used to assess the significance of differences between the model and observation. This test takes into account chance variability, and shows us whether differences shown are purely due to reasonably expected interannual variability or something more significant, such as an error in the model data or observed data.


Upon preliminary inspection of the model's simulation of surface air temperature it was noted that temperatures were much cooler over the Tibetan Plateau region than the rest of the surrounding region. (The Tibetan Plateau region, for our purposes, is defined to be the region between 14°N and 46°N latitude and 50°E and 125°E longitude.) As you can see in Figure 1, below, there was a large discrepancy in temperature that led us to investigate this region in more depth. It was decided to limit the analysis to July due to both time constraints and the fact that we are more interested in how the model handles extremities. The following variables were compared in an in-depth study of this region: precipitation, absorbed solar radiation, total cloud cover, cloud top pressure, and surface air temperature (Surf_Temp). It is expected that the low temperatures over this region can be explained by the model's deficiencies in simulating other variables.


Figure 1. Box in Surf_Temp plot shows the observed discrepancy that prompted further investigation of this region. Region also boxed out on the Primary Grid.

Surface Air Temperature

East of the Himalaya Mountains, the model underestimates the Surf_Temp within a range of 3°C and 30°C. Moving eastward, however, it is found that the difference between the model and the observations decreases. West of the Himalayas, towards India, Pakistan, Afghanistan, and Iran, the model overestimates the surface air temperature. Surface air temperature is directly influenced by insolation (incoming sunlight) reaching the surface. One would expect that the absorbed solar radiation at the surface should be less as well. This was confirmed by the difference plot below (See Fig. 3). For the most part, where solar radiation was overestimated, the surface air temperature was overestimated, and where solar radiation was underestimated, the temperature was underestimated.


Figure 2. Difference Map (SST-OORT) Blue region shows model underestimates the temperature over the Tibetan Plateau, and overestimates it to the west of the Himalayas.


Moving towards the west away from the Himalayas, the model tends to underrepresent precipitation in India by 8 to 16 mm/day. Over the Tibetan Plateau region itself, the model estimates too much precipitation (3 mm to 6 mm). In addition, in the Bay of Bengal, the model also predicts too much precipitation. Over the South China Sea, precipitation in the model is less than the observed. The increased precipitation over Mongolia and the Gobi Desert could be linked to the exaggerated amount of low clouds in the region. (See Figure 3.) This is because most of the rainmaking clouds are low clouds.


Figure 3. Difference Map (SST-LEGATES) Blue region shows model underestimates the precipitation over parts of India, while slightly overestimating precipitation over Mongolia.

Absorbed Solar Radiation

Generally over the Himalayas and up through Mongolia and the Gobi Desert, the model underestimates the amount of solar radiation absorbed. Over the Tibetan Plateau, the amount of absorbed solar radiation is significantly underestimated (see Figure 4). However, just above India, near the Himalayas, the model significantly overestimates absorbed solar radiation.

Why is the absorbed solar radiation not simulated well by the model? To answer this question, a difference plot of cloud top pressure was examined. Cloud top pressure is the measure of the height of a cloud. The greater the cloud top pressure, the lower the height of the cloud. This is relevant because of the fact that low clouds (such as nimbostratus) are responsible for reflecting insolation back to the atmosphere, thereby reducing the amount of solar radiation reaching the surface. Figures 5 and 6 demonstrate the model is exaggerating the production of clouds over the Tibetan Plateau region. Figure 5 suggests these are low clouds.


Figure 4. Difference Map (SST-ERBE) Blue region shows model underestimates the amount of absorbed solar radiation received at the surface of the Tibetan Plateau Region.

Cloud Top Pressure and Total Cloud Cover

The model underpredicts, to a large extent, cloud top pressure over much of the region to the south and west of the Himalayas within a range of 140 mb to 480 mb. In these areas, the quantity of high clouds is increased in the model. However, moving over the Himalayas to the Tibetan Plateau itself, cloud top pressure is overestimated. In addition, the model severely exaggerates cloud top pressure in the area directly over the plateau. Over the Bengal Sea and the South China Sea, the model consistently underpredicts cloud top pressure (depicting an excessive area of low clouds.) Moving westward back over land, however, the model begins to exaggerate this variable. See figures 5 and 6.



Figures 5 and 6 are difference maps. Figure 5 shows the total cloud cover is generally underpredicted in the model over the region, while figure 6 shows overestimated cloud top pressures over the plateau, meaning the model is overproducing low clouds.


The new version of the GISS GCM, ModelE is doing quite well in the general sense. It tends to simulate general patterns, such as storm tracks and the Intertropical Convergence Zone (ITCZ) quite well. In order to identify problems within the model, an in-depth study of each of these problematic regions must be completed. This enables the model programmers to make further revisions of the model, until it is the best that it possibly can be. Hence, this is an ongoing process.

From the analysis of the Tibetan Plateau, we have concluded that the model may have problems with moist convection and orography. The model's resolution is too coarse to deal with the drastic relief; it does not do well with mountains. The low surface temperature is linked to other factors such as too many low clouds. It seems that the clouds produced by the adiabatic cooling processes is being overestimated by the model. The manner that the model represents cloud cover should be adjusted. More research is necessary to fix this problem. Another issue was the model's handling of the Southeast Asian monsoons. More study of moisture transport within the model should be completed so that the model correctly simulates this climatic phenomenon.

In addition, this study was limited to the month of July. However, there are seasonal variations in climate. In order to assess the accuracy of the model for the other extreme month as well, a similar analysis must be completed for January. Once this is completed, we can have more confidence in our conclusions, and we will also be able to further speculate as to possible ways to improve the model as a whole.

The overall goal of improving the SST model is to make it reliable enough to be used as part of the full Coupled Model. The SST model is the most primitive model; it is the base for the other two models, the Q-Flux and the coupled model. The coupled model is going to be used to make predictions, so it has to be the best it can possibly be.

In order to better understand why the model has difficulty modeling successfully in this region, the following should be done:

  • Complete an analysis for January on this region
  • Do hypothesis tests to assess the mathematical significance of these results -- can they be attributed to inter-annual variability?
  • Use larger sample sizes (at least 10 years)


  • 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, J, and S. Lebedoff 1988. "Global Surface Air Temperatures: Update through 1987". Geophysics Res. Lett. 15, 323-326.
  • IPCC: Regional Climate Information-Evaluation and Projections, IPCC Working Group I Third Assessment Report, Final Draft Report, Chapter 10 (2001)
  • 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.
  • Intergovernmental Panel on Climate Change, 2000 Climate Change 2001: The Scientific Basis. UK: Cambridge University Press.
  • Lemonick, Michael D., 2001. "Life in the Greenhouse" Newsweek, April 9, 2001
  • Burroughs, William James. Chapter 5, Does the Weather Really Matter: The Social Impacts of Climate Change. Cambridge University Press, 1997.
  • Gross, M. Grant and Gross, Elizabeth. Oceanography – A View of Earth. 7th ed. New Jersey; Prentice Hall, 1996