In general, emissions estimates in this report were generated by multiplying some activity, such as coal consumption, by an emissions coefficient. The reliability of both the activity data and the emissions coefficients used in this report varies widely. This Appendix discusses the uncertainties associated with the estimates presented in the report.
In general, estimates of carbon dioxide emissions are more reliable than estimates for other gases. While this report does not explicitly calculate uncertainty ranges, it is likely that the estimate of carbon dioxide emissions is accurate to within 5 percent, implying an emissions range from 1.3 billion to 1.5 billion metric tons. To the extent that there is statistical bias in the point estimate, the actual (unobservable) figure is likely to be somewhat higher than the point estimate, because this report is unlikely to have captured all emissions sources.
Estimates of methane emissions are much more uncertain. The level of precision is probably on the order of 30 to 50 percent. Estimates of methane emissions are also likely to understate actual emissions, as a result of the exclusion of sources that are unknown or difficult to quantify.
Nitrous oxide emissions estimates are highly uncertain. Estimates of emissions from nitrogenous fertilizers are accurate to an order of magnitude, making them either the largest source of nitrous oxide emissions or, alternatively, an insignificant source. Coefficients for nitrous oxide emissions from fossil fuel combustion are not available for all sources, and where available, may be unreliable. Additionally, several known sources of nitrous oxide are not measured and therefore are excluded from estimated totals.
Coal. Coal production and consumption data are based on weight--short tons of coal. Coal consumption by regulated electric utilities, including both tonnage and energy content, is universally reported to the EIA and the Federal Energy Regulatory Commission (FERC). Utility coal consumption accounts for about 77 percent of U.S. coal consumption. There are likely to be only minor errors (around 1 percent) in reported utility coal consumption. Industrial, residential, and commercial coal consumption estimates are subject to potentially larger errors, especially in the counting of residential and commercial sector consumption.
The statistical discrepancy for coal production (the difference between reported consumption and reported production less exports, plus imports, plus stock changes) was on the order of 10 million short tons, or about 1.1 percent of consumption, in 1992.
Natural Gas. Most natural gas is sold or transported by regulated local distribution companies. The statistical discrepancy for natural gas is 1 to 3 percent of consumption, with reported consumption always smaller than reported production. This may imply (as discussed in Chapter 2) some systematic source of underreporting of consumption.
Inaccuracies in natural gas volumetric data come from inherent limitations in the accuracy of natural gas metering, as well as the usual problems of misreporting and timing differences. For example, natural gas consumption by electric utilities, as reported by the utilities, differed by about 0.5 percent from natural gas consumption as reported by natural gas sellers in 1993.(Note 1)
Petroleum. U.S. petroleum consumption was estimated on the basis of "petroleum products supplied," which means the volume of petroleum products shipped from primary storage facilities. Since there are only about 200 oil refineries in the United States, coverage of crude oil inputs and refinery outputs is generally complete.
The EIA requires a detailed breakdown and accounting of petroleum products produced by refineries, including refinery fuel. There are several statistical anomalies in EIA petroleum data:
By definition, the source of "unaccounted for" crude oil is unknown. It is likely due to imprecisions in recorded crude oil production, import, and stock change data. In EIA's State Energy Data Report, which presents con sumption estimates, unaccounted for crude oil is included in consumption.
The unfinished oil discrepancy is probably due to the asymmetric treatment of interrefiner sales of unfinished oils. To the buyer, who knows the intended use of the product, it is motor gasoline or distillate fuel. To the seller, it is an unfinished oil. In the State Energy Data Report, the unfinished oil discrepancy is accounted for through an adjustment to "other oils." However, the implication is that total oil consumption figures are more reliable than the exact distribution of consumption across specific petroleum products.
Overall, it is likely that petroleum consumption estimates are accurate to within 5 percent or so.
Nonfuel Use. Data for nonfuel use of petroleum products are more uncertain than those for total use of petroleum products. There are two main methods of estimating nonfuel use:
The main uncertainty in estimating carbon sequestered from nonfuel use is not the amount of product used, but the fate of its carbon. The sequestration percentages used in this report are estimates, originally based on the typical fate of a particular class of products. The actual distribution of nonfuel uses of products is not always known with precision and could vary considerably from the "typical" usage. However, since sequestration through nonfuel use corresponds to only about 5 percent of total emissions, even large variations in the amount sequestered would have a small effect on estimated total emissions.
Each step that transforms the data from native units into more useful units inevitably reduces the precision of the resulting data, because the conversion factors are themselves estimates, which may not precisely match the actual composition of the fuel.
Coal. Coal data are collected by State, coal rank, and weight (short tons). Electric utilities are asked to report both the rank and the energy content of the coal they burn. Since, in principle, utilities need to know the energy content of the fuels they purchase with precision, the energy content data should be fairly accurate. On the other hand, there is no direct information about the rank or energy content of coal distributed outside the utility sector, which accounts for about 23 percent of U.S. coal consumption. The EIA's energy data assume that the average energy content of coal by State (except anthracite) in the utility and nonutility sectors is the same.
The quality of coal can vary considerably within States and within a particular rank, and the quality of coal burned outside the utility sector is subject to a considerable degree of uncertainty. Lignite, for example, is defined as containing 6,300 to 8,300 Btu per pound, a range of about 15 percent. Subbituminous coal, by definition, has a range of 8,300 to 11,500 Btu per pound.(Note 5) Thus, there may be errors of up to 15 percent in the industrial and residential/commercial coal conversion factors. It is likely, however, that the actual (but unobservable) imprecision is much smaller than 15 percent, because there is no reason to expect systematic differences in coal quality between utility and nonutility usage. In any case, residential/commercial and industrial coal consumption accounts for only about 5 percent of total U.S. energy-related carbon emissions. Hence, even large errors would have only a small impact on the ultimate estimates.
Natural Gas. The composition of natural gas also varies considerably. In a recent survey of several thousand gas samples taken from local distribution companies around the United States, the Btu content ranged from 970 to 1,208 Btu per thousand cubic feet.(Note 6) However, 80 percent of the samples fell within a much narrower range of 1,006 to 1,048 Btu per thousand cubic feet. Further, the average and median values of the samples fell within 0.3 percent of the national-level figure as reported in EIA's Natural Gas Annual. This comparison suggests that EIA data on the energy content of natural gas are accurate to within 0.5 percent. This is not surprising, because local distribution companies monitor the energy content of natural gas to ensure adherence to contractual specifications, and they report the average energy content to the EIA.
Petroleum. The energy content of petroleum products varies more by volume than by weight. The density and the energy content of petroleum products are rarely measured by producers or consumers, and hence they are frequently not known with precision. Electric utilities measure the energy content of the residual oil they burn and report it to the EIA. Liquid petroleum gases (propane, butane, and ethane) are pure compounds, and their energy content can be computed directly.
Liquid transportation fuels (jet kerosene, gasoline, and diesel fuel) are complex mixtures of many compounds, whose physical properties can vary considerably. Their energy content is not measured by consumers nor directly defined by product specifications. The EIA estimates the energy content of these fuels on the basis of standard or "typical" values for each product. The standard energy contents for motor gasoline and kerosene-based jet fuel are drawn from a 1968 report produced by the Texas Eastern Transmission Corporation.(Note 7) The energy content of distillate fuel oil is drawn from a Bureau of Mines Standard adopted in January 1950.(Note 8) Jet fuel and diesel samples obtained for this report showed an average energy content that differs from EIA estimates by about 2 percent. Samples of motor gasoline analyzed by the National Institute of Petroleum and Energy Research displayed an average energy content that differs from EIA estimates by less than 0.5 percent.
Coal. There are large variations in the carbon and energy content of coals in different parts of the United States. Lignite may have as little as 12.6 million Btu per ton and contain 36 percent carbon, while anthracite may have as much as 98 percent carbon and an energy content as high as 27 million Btu per ton.(Note 10)
The carbon and heating values of coal are, in general, controlled by two factors:
Most of the gross variation in both energy and carbon content (for example, between lignite and anthracite) is due to variations in nonflammable impurities. Consequently, if the Btu content of coal is estimated accurately, most of the variation in the carbon content is removed.
There is, however, residual uncertainty about the ratio of carbon to hydrogen and sulfur in particular coals. The carbon content of any particular coal sample can be determined by chemical analysis, but characterizing the average carbon content of national coal production creates some uncertainty. For this report, the EIA relied on chemical analyses of several thousand coal samples, sorted by State of origin and coal rank, to compute national weighted average emissions coefficients (in million metric tons of carbon per million Btu) for each coal rank.
Natural Gas. Natural gas also varies in composition, but the range of variation is much smaller than for coal. The emissions coefficient used in this report was based on an analysis of some 6,743 recent samples of U.S. natural gas. While there is some residual uncertainty about the exact carbon content of average U.S. natural gas, it is on the order of 1 percent or less.
Petroleum Products. Crude oil is refined into a wide range of petroleum products, each presenting a different set of uncertainties. In general, the carbon content of petroleum products increases with increasing density. Uncertainties in emissions coefficients arise primarily from picking the wrong density for a fuel, or from mismatching the carbon and energy content of a particular fuel. The emissions factors for liquefied petroleum gas (LPG) and motor gasoline are probably accurate to within 1 to 2 percent. Coefficients for jet fuel and diesel fuel are probably accurate to within 2 to 4 percent, with much of the uncertainty centered in the standard heat contents used. The estimate for residual fuel is more uncertain but probably accurate within 3 to 5 percent, as there are remaining uncertainties about the exact density and carbon content of the fuel.
The uncertainty for some minor petroleum products remains large, in some cases because it has proven difficult to identify exactly how reporters define particular product categories. Products with large remaining uncertainties include petrochemical feedstocks (density and portion of aromatics), lubricants, and waxes and polishes. The uncertainty of the emissions coefficients for these products is probably on the order of 10 percent. Because these products share a large nonfuel use component, their impact on the total carbon emissions figure is muted. Still gas is a highly variable byproduct of the refining process, which is then described as a petroleum product. Thus, the estimated emissions coefficient for still gas may vary by as much as 40 percent.
Unmetered Gas Consumption. Since the estimate for unmetered gas consumption is actually a balancing item, the uncertainty of the estimate is very large, on the order of 100 percent. Fortunately, this is only a small source of carbon emissions.
Flare Gas. Estimates of emissions from flare gas are subject to uncertainty from two sources: estimates of the volume of gas flared, and the application of an appropriate emissions coefficient. Estimates of gas flared are based on State-reported volumes of gas "vented or flared" and a State-by-State estimate of the share flared. States may define "flared" gas differently. This suggests that estimates may be upwardly biased by the inclusion of low carbon emitting gases in the statistics, but the degree of bias is unknown.
The emissions coefficient applied to flare gas represents the average coefficient for natural gas samples with heat contents between 1,100 and 1,127 Btu per standard cubic foot. The EIA estimates the heat content of "wet" gas at 1,110 Btu per standard cubic feet.(Note 11) Anecdotal evidence suggests that most flared gas is flared at gas processing facilities, where the wet gas energy content would be representative. However, if flared gas is mostly "rich" associated gas with a heat content between 1,300 and 1,400 Btu per standard cubic feet, the current coefficient seriously biases estimates downward. Alternatively, it is possible that flare gas from treatment plants is "off spec" gas with a large content of hydrogen sulfide or inert gas and, hence, an emissions coefficient lower than the one actually used.
A second source of uncertainty, common to all the industrial estimates, is the use of stoichiometric computations to estimate emissions. This method calculates an emissions factor based on a chemical reaction known to have taken place. It assumes, in effect, that the product produced (cement, lime, soda ash) is 100 percent pure, and that no raw materials are wasted in its production. In practice, impurities in the output would tend to reduce emissions below the stoichiometric estimate, while "wastage" of raw materials would tend to raise emissions above the estimate.
Where no systematic measurements have been made, estimation methods rely on a limited set of data applied to a large and diverse group of emitters. However, as additional data comes available each year, uncertainty in emissions estimates declines or, at a minimum, is more clearly delineated. In this year's report, additional information on emissions from coal mines, oil and gas operations, and landfills has been incorporated into the estimates, significantly reducing uncertainty levels.
Emissions from ventilation systems in the Nation's gassiest mines are measured on a quarterly basis by the Mine Safety and Health Administration (MSHA). The Bureau of Mines (BOM) has developed a database from MSHA reports for all mines emitting more than 100,000 cubic feet of gas per day. Data on emissions from this source are available for 1985, 1988, 1990, and 1993. Although the measurements themselves should be reasonably accurate, each measurement represents a point in time. Variations in methane emissions across time (e.g., resulting from changes in operating practices) suggest an uncertainty in the range of 10 to 40 percent.
Estimates of emissions from the ventilation systems of nongassy mines are scaled to emissions estimates for 1988 developed by the U.S. Environmental Protection Agency (EPA).(Note 13) The EPA estimates emissions from nongassy mines at 2 percent of total emissions from the ventilation systems of underground coal mines. Thus, an error as high as 100 percent for this source would add less than 1 percent to the total estimate.
Emissions from degasification systems may be the single largest source of uncertainty in estimates of emissions from coal mines. The estimation method scales emissions to estimates of emissions for 1988 developed by the EPA. Measurements of emissions were limited to a few mines, with the remainder of the estimate based on known emissions from ventilation systems and estimated recovery efficiencies of degasification systems. The recovery efficiencies have an uncertainty in the neighborhood of 20 percent.
Reliable measurements of emissions from surface mines are available for only five sites. Thus, estimates for this report were based on an emissions range supplied by the Intergovernmental Panel on Climate Change (IPCC).(Note 14) The range of emissions suggested by the IPCC implies an uncertainty range of plus or minus 75 percent. However, estimates of emissions extrapolated from the five measured sites suggest an uncertainty level of less than 10 percent.(Note 15) Assuming the larger uncertainty level would add only about 10 percent to the overall uncertainty of estimates of emissions from coal mines, because the volume of emissions from surface mines is relatively insignificant.
Emissions from post-mining activities are also estimated on the basis of an emissions range supplied by the IPCC. This emissions range implies an uncertainty in the area of plus or minus 60 percent. However, the magnitude of emissions from this source is similar to that of emissions from surface mines, thus also contributing about 10 percent to the overall uncertainty of coal mine emissions estimates.
This report's estimate of emissions from production, transmission, and distribution is based on emissions factors developed from a small sample of model facilities, as recommended by the IPCC. Facility selection was based on accessibility for data collection and operator interviews, size, process type, location, and age. In order to develop time-series emissions estimates, emissions factors for a single year (1990) were scaled to some known activity data, such as pipeline miles. Because newly added pipeline miles are typically plastic pipe with low emission rates, this scaling mechanism may bias the estimates upward. Recently, a more extensive, systematic analysis of the oil and gas production and distribution system was completed by Kirchgessner et al.(Note 16) A comparison of their estimates and those that appear in this report indicates a potential underestimate of emissions in excess of 100 percent.
Estimates of emissions from venting are also uncertain. The EIA maintains statistics on gas vented or flared as reported on a State-by-State basis, but no distinction is made between venting and flaring in those statistics. Gas flared releases carbon dioxide rather than methane. This report estimates the national share vented on the basis of the estimated share vented for each State.
An additional uncertainty associated with estimates of methane vented is methane vented at "stripper wells." Associated natural gas production at oil wells producing less than 10 barrels per day may be at pressures and volumes too low to be of commercial value. The gas may be vented or merely evaporate from storage tanks. Such emissions are not captured in any data series. The magnitude of the emissions is impossible to estimate, but "stripper wells" comprise 14 percent of U.S. oil production.
Methane emissions from mobile combustion may be larger than the estimate offered in this report, but it is unlikely that they are significantly smaller. Emissions factors for mobile transportation assume a well-maintained fleet. A fleet of inadequately maintained vehicles may have as much as 10 times the level of emissions of a fleet of well-maintained or new vehicles. Although much of the U.S. fleet is well-maintained, a portion is old and/or poorly maintained.
Emissions for many of the 105 mostly large landfills were estimated for 1992, based on volumes of gas recovered multiplied by gas recovery efficiency.(Note 17) Gas recovery efficiency was estimated, with an associated uncertainty of plus or minus 25 percent. For years other than 1992, emissions from this source were estimated using a model of landfill waste emissions that is benchmarked to the 1992 data. The model parameters include a low yield and high yield scenario that imply an uncertainty of 35 percent.
Emissions from all other landfills were also estimated using an emissions model. This model's input parameters could vary by 30 percent from the mean. A crucial input into the model is amount of waste in place. The amount of waste in place was calculated from estimates of waste landfilled annually between 1960 and 1994 and using a regression equation to backcast waste flows from 1940 to 1960. The range of published estimates for years in which multiple sources were available suggests an uncertainty in the neighborhood of plus or minus 33 percent, while the error associated with the regression equation probably adds another 2 to 10 percent additional uncertainty.(Note 18)
The ratio of waste in place in the 105 landfills relative to that in all other landfills was assumed to remain constant over time. This may be misleading, since the total number of landfills has been declining, with greater shares of waste believed to be directed toward larger landfills. Because those landfills with measured emissions for 1992 are likely to have higher-than-average emissions per ton of waste, estimates may be biased upward in earlier years.
There is some uncertainty associated with the average size of cattle, which could affect the animals' energy requirements. Cattle sizes have been changing rapidly over the past decade in response to market forces. This report uses slaughter weights as a proxy for average animal size, a method that may be imperfect. The slaughter sizes vary over time by approximately 33 percent.
Emissions are tied to the amount of waste an animal produces. The amount of waste produced is a function of size and diet. Thus, changes in animal sizes, which are difficult to monitor, create additional uncertainty.
As discussed above, slaughter weights have been used as an imperfect tool for capturing changes in animal size. This proxy measure varies by 30 percent over time. Uncertainty in estimates of animal populations is on the order of 5 percent or less.
There are large uncertainties associated with the estimate of methane emissions from wetland rice cultivation. Emissions estimates are based on several studies of rice paddies in the United States, which provide daily emissions rate ranges. Studies have shown large seasonal and time-of-day variations in methane flux. Many variables affect methane production in rice fields, including soil temperature, redox potential, and acidity; substrate and nutrient availability; addition of chemical and/or organic fertilizers; rate of methane oxidation; and rice plant variety. The wide range of emissions provided by different researchers suggests an uncertainty of several hundred percent.(Note 19)
The emissions factors were derived from studies of "conventional" combustion facilities (i.e., those equipped with burners and grate combustion, with flame temperatures well beyond 1,000oC). Other types of facilities are used in the United States, adding to the uncertainty of estimates presented in this report.
The emissions estimates presented in Chapter 6 are taken from data in National Air Pollutant Emission Trends, 1900-1993, a report published by the EPA Office of Air Quality Planning and Standards. Although true values of criteria pollutant emissions are not known, the EPA states that, "beginning with the 1900 to 1992 report, EPA set the primary goal of preparing emission trends that would also represent the best available estimates of emissions."(Note 23) The EPA also explains that one of the difficulties they experience in providing emissions estimates is balancing consistency of estimation methods with completeness and accuracy of data. Therefore, as new methods and data become available, emissions estimates may be revised. As an example, the estimates for 1992 emissions were increased by 10 percent for carbon monoxide in the latest Trends report, due in part to an improved method for estimating emissions from highway vehicles. Estimates for nitrogen oxides and nonmethane volatile organic compounds were revised by less than 2 percent.(Note 24)
Obtaining accurate estimates for greenhouse gas emissions and sequestration related to land use is a difficult endeavor. Consumption of resources that sell in markets is often easily measured, as are corresponding emissions rates. Not so for nonmarket resources. These estimates are usually calculated by scientists from small sample measurements extrapolated to much larger areas, with corresponding error. For example, a great deal of data is collected annually on the quantity of wood extracted from U.S. forests, but the data are invariably expressed in terms of the small portion (less than 50 percent) of each tree that can be converted to marketable wood products. The rest is ignored. Researchers cope with this problem by creating ratios of marketable to unmarketable tree volume and applying them to commercial data, but such ratios can only serve as rough approximations.
Carbon dioxide is by far the most important greenhouse gas. Chapter 7 of this report reflects the disproportionate importance of forestland in the carbon budget. Most of the estimates for forest carbon sequestration and emissions came from the U.S. Department of Agriculture (USDA) Forest Service scientist Richard Birdsey. His estimation methods were inclusive of all forest carbon, including all above- and below-ground portions of all live and dead trees; all soil carbon; all carbon contained in dead organic matter on the forest floor; and all carbon in understory vegetation. Mr. Birdsey obtained his data primarily from statewide USDA forest inventories, a nationwide biomass study prepared by the USDA Forest Service, and a special report containing estimates of the proportion of tree volume that is below ground. These sources are subject to sampling errors, estimation errors, and errors in converting data from one reporting unit to another. There is also error in determining the carbon content of wood. Finally, many estimates from forest ecosystem studies were used to predict carbon quantities in other ecosystems or large geographic areas, with corresponding potential for error.
The Birdsey estimate for forest carbon sequestration is a net 106 million metric tons from all U.S. forests in 1987. This is based on an estimated 461 million metric tons of sequestration overall, reduced by 355 million metric tons for timber harvest, land clearing, and fuelwood use. Of the available estimates, this is the lowest and hence most conservative. The EPA estimated net carbon flux at 114 million metric tons in 1990,(Note 25) and Birdsey himself tentatively raised the figure, based on the latest survey of forest areas in the United States conducted by the USDA Forest Service, to 130 million metric tons in 1992.(Note 26) The consensus is on a net sequestration rate in the range of 110 to 130 million metric tons for the early 1990s.