Carbon dioxide emission tallies for 210 U.S. coal-fired power plants: A comparison of two accounting methods

Abstract

Annual CO₂ emission tallies for 210 coal-fired power plants during 2009 were more accurately calculated from fuel consumption records reported by the U.S. Energy Information Administration (EIA) than measurements from Continuous Emissions Monitoring Systems (CEMS) reported by the U.S. Environmental Protection Agency. Results from these accounting methods for individual plants vary by ± 10.8%. Although the differences systematically vary with the method used to certify flue-gas flow instruments in CEMS, additional sources of CEMS measurement error remain to be identified. Limitations of the EIA fuel consumption data are also discussed. Consideration of weighing, sample collection, laboratory analysis, emission factor, and stock adjustment errors showed that the minimum error for CO₂ emissions calculated from the fuel consumption data ranged from ± 1.3% to ± 7.2% with a plant average of ± 1.6%. This error might be reduced by 50% if the carbon content of coal delivered to U.S. power plants were reported.

Implications:

Potentially, this study might inform efforts to regulate CO₂ emissions (such as CO₂ performance standards or taxes) and more immediately, the U.S. Greenhouse Gas Reporting Rule where large coal-fired power plants currently use CEMS to measure CO₂ emissions. Moreover, if, as suggested here, the flue-gas flow measurement limits the accuracy of CO₂ emission tallies from CEMS, then the accuracy of other emission tallies from CEMS (such as SO₂, NO_x, and Hg) would be similarly affected. Consequently, improved flue gas flow measurements are needed to increase the reliability of emission measurements from CEMS.

Introduction

Carbon dioxide emissions from coal-fired power plants can be calculated from flue gas composition and volume measurements made by Continuous Emission Monitoring Systems (CEMS) or from fuel consumption and quality measurements using mass balance calculations. Results from CEMS are included in the U.S. Environmental Protection Agency (EPA) Clean Air Markets Division (CAMD) database (U.S. Environmental Protection Agency, 2012a) whereas fuel consumption and quality measurements are included in the U.S. Energy Information Administration (EIA) Form-923 database (U.S. Energy Information Administration, 2012). Comparison of CO₂ emission tallies from these databases (Ackerman and Sundquist, 2008; Borthwick et al., 2011; Huang and Gurney, 2011) shows similar population totals, but significantly different results for individual units and plants, which leads to several questions. Are CO₂ emissions more accurately measured by CEMS, or calculated from fuel consumption? Why do results from these accounting methods differ? Definitive answers to these questions are needed to inform discussion of CO₂ emission trading programs, performance standards, or taxes. Comparison of results from tests of varied duration for units at unnamed plants has suggested that CO₂ emissions measured by CEMS are more accurate than those calculated from fuel consumption (Evans et al., 2009) and, alternatively, that both methods are prone to error (Zimmerman, 2009). Accordingly, this study attempts to answer these questions by using published data to compare CO₂ emissions measured by CEMS with CO₂ emissions calculated from fuel consumption measurements for 210 coal-fired power plants during 2009.

Method

Data

The year-2009 data used in this study are from the CAMD website (U.S. Environmental Protection Agency, 2012a), EIA spreadsheet files (U.S. Energy Information Administration, 2012), and summaries of these data (Emissions & Generation Resource Integrated Database, 2012). Some additional data (Bragg et al., 1998; Keystone, 2002; U.S. Energy Information Administration, 2003; Illinois Geological Survey, 2007), which have been described elsewhere (Quick, 2010), were used with the EIA data to calculate CO₂ emission factors. Methods used to certify flue-gas flow at these plants were from the EPA Emissions Collection and Monitoring Plan System (U.S. Environmental Protection Agency, 2012b). Finally, an unpublished data set that includes the C content of commercial coal shipments was used to estimate the error resulting from the use of the calculated CO₂ emission factors.

Plant selection

This study narrowly compares CO₂ emissions measured by CEMS (CAMD data) with CO₂ emissions calculated from fuel consumption (EIA data) for 210 coal-fired power plants during 2009. Note that this comparison is made at the plant level rather than the individual unit (boiler) level. Aggregation of emissions data at the plant level avoids complications due to differences in how individual units within a plant are identified and accounted for in the CAMD and EIA data collection efforts, as well as variation where unit fuel consumption reported to the EIA was estimated rather than measured. Plant selection criteria include:

•	The plants are listed in both the CAMD and EIA databases.
•	The CAMD CO2 emission data are from CEMS.
•	The CAMD data included gross electric generation.
•	More than 99% of the fuel burned during 2009 was coal.
•	Combined heat and power plants were ignored.
•	The plant capacity factor was greater than 0.10.

The selected power plants represented about one-third of U.S. coal-fired power plants, but were responsible for more than two-thirds of the electricity generated from coal, and nearly 18% of the 7.3 billion tons (CO₂ equivalent) of U.S. greenhouse gas emissions during 2009 (U.S. Environmental Protection Agency, 2013). Notably, the selected plants exclude large gas-fired units (where CEMS are rarely used) and combined heat and power plants (where the data sets differ in the allocation of heat input for electricity). Limiting the comparison to coal-fired power plants excludes plants that burn biomass (where biogenic CO₂ is inconsistently counted). Finally, data were excluded for two plants (ORIS [plant identification] codes 1382 and 6052) where the apparent parasitic energy consumption (gross generation minus net generation) exceeded 30%, and another two plants (ORIS codes 1832 and 2712) where the apparent parasitic energy consumption was a negative value.

Calculated CO₂ emissions

Annual plant-level CO₂ emissions were calculated from the EIA data (U.S. Energy Information Administration, 2012) in several steps. First, the average heating value of the delivered coal was multiplied by the tons of coal consumed to determine the annual coal heat input (British thermal units [Btu]) for each plant. Next, plant-specific CO₂ emission factors (lb CO₂/million Btu) for 2009 were calculated according to the geographic origin, quality, and tons of delivered coal, as described elsewhere (Quick, 2010). The annual heat input from coal was multiplied by the plant-specific CO₂ emission factor to calculate the CO₂ emissions from coal burned at each plant. Carbon dioxide from noncoal fuel consumption was calculated by multiplying the heat input from auxiliary fuels listed with the EIA Generation and Fuel data (natural gas, jet fuel, distillate petroleum, residual fuel oil, tire-derived fuel, and petroleum coke) by their nominal CO₂ emission factors (117, 160, 161, 174, 190, and 225 lb CO₂ /million Btu, respectively) (U.S. Energy Information Administration, 2011). Finally, the amounts of CO₂ from the coal and the auxiliary fuels were added together to obtain the total CO₂ emissions for each power plant. The auxiliary fuels contributed less than 0.7% of the total CO₂ emissions from any of the plants in this study.

Measured CO₂ emissions

Nearly all of the CAMD CO₂ emission results for the 210 plants originate from flue-gas volume and CO₂ concentration measurements from CEMS. Monthly plant CO₂ emission tallies were downloaded from the CAMD website (U.S. Environmental Protection Agency, 2012a) and combined to obtain the annual plant totals for 2009. Note that many plants had multiple CEMS, generally corresponding to the number of combustion units at the plant. In these instances the monthly data represent the sum of emissions from multiple units. Besides the CO₂ emissions, the gross electric generation and the SO₂ emissions were also downloaded from the CAMD website.

Results and Discussion

Figure 1 shows the difference between annual CO₂ emissions measured by CEMS and annual CO₂ emissions calculated from fuel consumption for 210 coal-fired U.S. power plants during 2009. Note that the differences shown in Figure 1 are normally distributed about a mean value near zero (–0.7%). Consequently, both the CAMD and the EIA emission tallies should provide about the same total emissions for the U.S. coal-fired electric generating fleet. However, consistent with previous work (Ackerman and Sundquist, 2008; Borthwick et al., 2011; Huang and Gurney, 2011), Figure 1 also shows that these methods indicate significantly different CO₂ emission tallies for individual plants. Indeed, Figure 1 shows that these differences vary by ±10.8% (two times the standard deviation) for individual plants. Less clear is whether these differences originate from erroneous EIA measurements, CAMD measurements, or both.

Figure 1. Annual CO₂ emissions measured by CEMS (CAMD data) vary by ±10.8% (two times the standard deviation) from CO₂ emissions calculated from fuel consumption (EIA data) for 210 U.S. coal-fired power plants during 2009. The percent difference was calculated according to: 100 × (CAMD – EIA)/[(CAMD + EIA)/2].

Origin of the CO₂ emission measurement error

To identify the origin of the measurement error indicated by Figure 1, the CAMD and EIA CO₂ emission tallies for each plant were divided by the plant's annual gross electric generation (MWh); the reason for this data transformation is explained in the supplemental material. The resulting CAMD and EIA CO₂ emission rates (lb CO₂/MWh) are compared in Figure 2. The comparison shows both data sets have similar average CO₂ emission rates (less than 1% difference), but the CAMD emission rates show a 30% greater range of variation compared to the EIA emission rates. The CAMD and EIA CO₂ emission rates are directly compared in Figure 3. Note the larger standard error in Figure 3a (where the CAMD rates are the dependent variable) compared to Figure 3b. Measurement error in the dependent variable inflates the standard error (Hutcheon et al., 2010). Also note the diminished slope of the best-fit line in Figure 3b (0.60) where the CAMD rates are the independent variable compared to the slope for the best-fit line in Figure 3a (1.01). The diminished slope is called attenuation bias, and results from measurement error in the independent variable (Hutcheon et al., 2010). These observations show that the difference between the CAMD and EIA CO₂ emission tallies shown in Figure 1 is largely due to CAMD measurement error. Thus, for individual plants the annual CAMD CO₂ emission tallies are less accurate than those calculated from EIA fuel consumption data.

Figure 2. Annual CO₂ emission rates calculated from CAMD and EIA data for 210 U.S. power plants during 2009 show similar mean values (2,062 and 2,074 lb CO₂/MWh, respectively), but more variable CAMD emission rates compared to EIA emission rates (standard deviations of 178 and 137 lb CO₂/MWh, respectively).

Figure 3. Attenuation bias due to CAMD measurement error is shown by comparison of best-fit lines where (a) EIA CO₂ emission rates are used to predict CAMD rates, and (b) CAMD CO₂ emission rates are used to predict EIA rates (data are for 210 U.S. coal-fired power plants during 2009). Although both lines share the same significance and correlation (P < 0.001, R ² = 0.61), note the larger standard error for (a) and that the slope for (b) is closer to zero.

Cause of the CAMD CO₂ measurement error

Consideration of the measurement frequency as well as calibration and certification protocols helps to explain the cause of the CAMD measurement error shown in Figure 3. The annual emission totals for these power plants are based on hourly flue-gas volume and CO₂ concentration measurements. EPA rules (U.S. Environmental Protection Agency, 2009) require daily calibration of instruments in CEMS that measure the flue-gas CO₂ concentration to within 1.0% CO₂ (by volume); this corresponds to a relative precision between ±7 and ±12% for typical flue gas CO₂ concentrations (between 8 and 14% CO₂ by volume). Importantly, the calibration is accomplished using cylinder gas standards, which minimizes the possibility of significant CO₂ concentration measurement bias. Moreover, the frequent (daily) calibration requirement substantially reduces the influence of the limited measurement precision. For example, during 2009 each power plant reported between 3,486 and 76,658 hourly CO₂ measurements (average = 17,442). Assuming daily calibration with a precision of ±10%, the CO₂ concentration measurement errors would cancel during the year and at most would cause a ±0.8% error in the annual emission total (10/√(3,468/24) = 0.8). Although leaks in the sample gas collection system can reduce the measured CO₂ concentration, this potential bias is readily identified during mandatory annual or semi-annual audits (Boze, 2010).

The possibility of significant and persistent flue-gas flow measurement error is more difficult to dismiss. Unlike the CO₂ concentration measurement, the flue-gas flow measurement is not challenged with a calibration standard. Indeed, a huge known volume of flue gas would be required to directly calibrate the flue-gas flow measurement system at a power plant; such standards do not exist. Consequently, the accuracy of the flue-gas flow measurement cannot be assured. Although the flue-gas flow measurement system is not directly calibrated, it must be periodically certified during annual or semi-annual audits to comply with EPA rules. Certification is accomplished by comparing flue-gas flow measurements from the installed equipment with independent flue-gas flow measurements obtained using a EPA-approved method (Table 1) (U.S. Environmental Protection Agency, 2012c). Method 2 was initially developed to certify the performance of CEMS used to measure SO₂ emissions. Unrealistically high SO₂ emission tallies at some power plants were subsequently attributed to a high flue-gas flow bias due to the inability of method 2 to account for velocity decay near the stack wall, as well as nonlaminar, inclined, or swirling flow within the stack (Placet et al., 2000). Accordingly, the EPA modified method 2 for these situations, and special provisional methods were approved for rectangular ducts.

Table 1. EPA methods used to measure and certify flue-gas flow. Results from the installed measurement system must agree with results from one or more of these methods

Method	Description
2	Assumes laminar flow, no wall effects adjustment
D2H	Assumes laminar flow, default wall effects adjustment
M2H	Assumes laminar flow, measured wall effects adjustment
2G	Assumes inclined flow, no wall effects adjustment
2GH	Assumes inclined flow, measured or default wall effects adjustment
2F	Assumes swirling flow, no wall effects adjustment
2FH	Assumes swirling flow, measured or default wall effects adjustment
2J	Rectangular ducts, assumes laminar flow, measured or default wall effects adjustment
2GJ	Rectangular ducts, assumes inclined flow, measured or default wall effects adjustment
2FJ	Rectangular ducts, assumes swirling flow, measured or default wall effects adjustment

About half of the 210 plants included in this study used more than one of the methods listed in Table 1 to certify their CEMS during 2009. For plants that used only one method, Table 2 shows that those that used method 2 reported significantly higher CO₂ emissions than indicated by plant fuel consumption. As noted earlier, method 2 has long been known to potentially overestimate flue-gas flow (Placet et al., 2000) and, consequently, CO₂ emissions. Apparently, this positive bias persists. Table 2 also shows that plants with CEMS that were certified using methods 2GH and M2H reported significantly lower CO₂ emissions than indicated by fuel consumption, which is consistent with negative bias due to underestimated flue-gas flow. Together, these observations support the idea that the CAMD measurement error is at least partly due to biased flue-gas flow measurements, which systematically vary with the flue-gas flow certification method. However, the flue-gas flow measurement bias accounts for less than half of the larger range of variation of CAMD CO₂ emission rates compared to EIA CO₂ emission rates shown in Figure 2. Consequently, an additional source of CAMD measurement error remains to be identified. Given the limited frequency of CEMS certification (once or twice a year), the precision of the flue gas flow measurements during certification could contribute to this remaining, albeit unidentified, measurement error.

Table 2. The difference between CAMD and EIA CO₂ emissions varied with the method used to certify the CAMD flue-gas flow measurement at 112 power plants during 2009

Method	n ^a	CO2 % difference ^b	^σ %diff. ^c	t-Statistic ^d	P ^e
2	32	+3.0	5.4	−4.5	0.00 ^f
2F	9	+1.6	5.8	−1.3	0.19
D2H	13	−0.2	5.0	−0.4	0.70
2FH	12	−0.8	4.9	0.0	0.96
M2H	17	−3.1	4.2	+1.9	0.06 ^g
2GH	29	−3.0	4.8	+2.5	0.01 ^f
^aNumber of plants.
^bAverage percent difference between CAMD and EIA CO2 emissions, calculated as: 100 × (CAMD – EIA)/[(CAMD + EIA)/2].
^cStandard deviation of the percent difference.
^dTwo-sample, ntotal =210, assuming equal variances.
^e Significance, two-tailed distribution.
^f Method significantly different at 95% confidence.
^g Method significantly different at 90% confidence.

Improving the accuracy of CAMD CO₂ emission tallies

The EIA data are reported by month and obviously lack the temporal resolution provided by CEMS. Indeed, the CAMD data are reported every hour and can be summed by varied time interval (day, night, week, month, and so forth). This temporal resolution is useful for carbon-cycle studies (Gurney et al., 2009) and potentially, calibration of orbital CO₂ emission verification tools (Velazco et al., 2011). However, the excellent temporal resolution provided by CEMS is diminished by their limited accuracy, which is at least partly due to flue-gas flow measurement bias. Where accuracy is important and temporal resolution useful, it should be possible to proportionally adjust the CEMS CO₂ measurements to comport with the annual CO₂ emissions calculated from EIA fuel consumption measurements. This adjustment would improve the accuracy of the CAMD CO₂ emission measurements and retain the temporal resolution provided by CEMS.

Limitations of EIA fuel consumption data

Although they allow for more accurate annual CO₂ emission tallies, limitations of the EIA data should also be considered. Both random and systematic measurement errors are likely. Random error (precision) can be evaluated by considering the instrumental precision of the coal consumption and quality measurements, and is reduced with repeated measurements. Systematic error (bias) is more difficult to identify, but is minimized where the measurement instruments are calibrated using standards having a known mass or composition.

One way EIA verifies the tons of coal consumed is by comparison with the amount of delivered coal and the change in coal stocks. Ideally, the tons of coal consumed must equal the tons of delivered coal minus any change in the amount of coal in the plant stockpile. The amount of delivered coal is measured using calibrated scales, which for commercial purposes are calibrated using standard weights to within ±0.25% (National Institute of Standards and Technology, 2012). Stock changes are calculated from annual or more frequent stockpile surveys. The stockpile survey error is thought to be about ±5% (DeSollar, 2011), which corresponds to a ±7.1% error for the stock change (ending stockpile measurement minus beginning stockpile measurement) according to eq 1:

1

Results from stockpile surveys are reported to the EIA, but are not published. Accordingly, an apparent stock adjustment was estimated for each plant to equal the absolute difference between the delivered tons and the consumed tons during 2009. The apparent stock adjustment varied widely (see supplemental material), but typically was about 5% of a plant's annual coal consumption.

Carbon dioxide emissions were calculated using CO₂ emission factors (lb CO₂/million Btu). The plant-specific emission factors used in this study introduce a ±1.3% error, which is significantly less than the ±2.3% error expected for emission factors specified by coal rank (see supplemental material). The error introduced by emission factors is systematic (Frey and Tran, 1999) and cannot be reduced by repeated measurement. However, because the emission factor error may have either positive or negative expressions, its influence can be reasonably considered by simple error propagation methods.

The precision of the heating value measurement can also be considered. In the laboratory, the heating value is measured using a bomb calorimeter to within ±140 Btu/lb_dry for lignite and subbituminous coal, and ±107 Btu/lb_dry for higher rank coal (ASTM, 2000a). The calorimeter is calibrated using a compound with a known heat of combustion so measurement bias is negligible. The precision of the heating value measurement corresponds to an average ±1.0% relative error for the 210 plants in this study, which varied between ±0.8% and ±1.7% depending on the rank, heating value, and estimated moisture content of the delivered coal.

Finally, error due to the collection and preparation of the analysis sample can be estimated. General-purpose sampling procedures specified in ASTM (2000b) method D-2234 are intended to provide an analysis sample with a composition within ±10% of the dry ash yield of the shipment. Although modern sampling systems can achieve better precision (Rose and Byer, 2012), a ±10% error is assumed here. This corresponds to an average ±1.0% relative error due to sample collection, with results for individual plants ranging from ±0.6% to ±1.4% depending on the ash yield and estimated moisture content of the delivered coal. Assuming a ±1.0% relative error for the heating value determination, a ±1.0% relative error due to sample collection, and that 5% of the consumed coal is accounted for by stock adjustments, eq 2:

2

shows that CO₂ emissions calculated from mass and quality measurements associated with a typical shipment should vary by about ±2.0%. Note that most of the power plants received 10 to several hundred coal shipments during 2009. Consequently, random errors attributed to the coal shipment decline as the number of shipment measurements (n) increases, according to 0.25/√n, 1.0/√n, and 1.0/√n for the delivered tons, sample collection, and heating value, respectively. Substituting these values into eq 2 allows calculation of the minimum error of the annual CO₂ emission tally calculated from the EIA fuel consumption data for each plant according to eq 3:

3

where n is the number of shipments received by the plant during 2009, Y is the percent of the consumed coal accounted for by the apparent stock adjustment (expressed as a decimal fraction), X = 1–Y, Z is the plant-specific heating value measurement error, and W is the plant-specific sample collection error. Figure 4 shows results from eq 3 for the 210 plants included in this study.

Figure 4. The minimum error of the annual CO₂ emission tally calculated from the EIA fuel consumption data varies between ±1.3% and ±7.2% for 210 U.S. coal-fired power plants during 2009.

Results shown in Figure 4 are considered minimum values due to unrecognized systematic measurement error (bias) associated with stockpile measurements, sample collection protocols, and other sources. Bias in stockpile surveys may have multiple causes, such as the difference between the survey datum and the true elevation of the ground surface under the stockpile. Like flue-gas flow monitoring systems, coal sample collection systems are not directly calibrated. Instead, these systems are certified by comparison with results from an independent sample collection method (ASTM, 2010), where the frequency of certification is specified by commercial contracts, rather than government regulations. Other sources of bias not considered in this study include additional CO₂ from acid-gas sorbents and avoided CO₂ where unburned C is found in ash. However, as discussed in the supplemental material, ignoring CO₂ from acid-gas sorbents introduces a small negative bias (0 to –3.2% range, average –0.4%) that is trivial compared to the ±10.8% variation shown in Figure 1. The amount of CO₂ lost to unburned C introduces a positive bias, but this error is also too small (estimated +0.3% to +4.4% range) to account for this variation.

Improving the accuracy of EIA CO₂ emission tallies

The EIA data collection (U.S. Energy Information Administration, 2012) was not designed to measure CO₂ emissions from combustion. Not surprisingly, there are limitations where these data are used for this purpose, as well as opportunities for improvement. For example, additional data collection might be undertaken to account for measurement bias due to CO₂ emissions from acid gas sorbents as well as unburned C in ash. Likewise, consistent reporting of coal distributions to U.S. power plants from terminals and transfer stations would reduce the stock adjustment error for plants that receive these distributions. Perhaps the most significant limitation of the EIA data is the lack of C assays for coal shipments. This requires the use of an emission factor to calculate CO₂ emissions. If the C content of each shipment were measured, then the terms in eq 3 related to the emission factor (1.3²) and the heating value measurement ([Z/√n]²) could be replaced with a single term based on the ±2.5% C _dry random error (ASTM, 2000c) associated with the C assay. As a result, the average measurement error of the EIA CO₂ emission tallies for individual plants shown in Figure 4 would decrease from ±1.6% to ±1.0%, which represents a potential 50% improvement in accuracy. Moreover, the C assay could be made on coal samples already collected for the heating value analyses. Assuming a C analysis and reporting cost of $200 for each of the ˜23,300 coal shipments listed with the year-2009 EIA data, this improved accuracy would have cost about $4.7 million, which is relatively inexpensive (Sanchez et al., 2012).

Conclusion

The attenuation bias and inflated standard error shown in Figure 3 indicate that the annual CO₂ emission tallies for 210 coal-fired power plants calculated from EIA fuel consumption data are more accurate than those measured by CEMS. The limited accuracy of emission tallies based on CEMS measurements is at least partly due to biased flue-gas flow measurements, which systematically vary with the flue-gas flow certification method. Although the EIA data allow for more accurate CO₂ emission tallies, propagation of weighing, sample collection, stock adjustment, emission factor, and laboratory errors showed that the cumulative minimum error for these tallies ranged from ±1.3% to ±7.2% with a plant average of ±1.6%. This error might be reduced by 50% if the carbon content of coal delivered to U.S. power plants were reported.

This study has implications for the U.S. Greenhouse Gas Reporting Program where large coal-fired power plants currently report annual CO₂ emissions measured using CEMS. The results presented here indicate that more accurate emission tallies can be calculated from data reported to the EIA-923 power plant operations survey. Moreover, efforts to improve the accuracy of CO₂emission tallies from CEMS are warranted if these tallies are used to design and implement regulations intended to reduce greenhouse gas emissions.

Supplemental Material

Supplemental data for this article can be accessed on the publisher's website http://dx.doi.org/10.1080/10962247.2013.833146.

Acknowledgment

Art Diem (EPA) reviewed an early draft of the paper and generously provided some of the data used in this study. Correspondence with Kevin Gurney (Arizona State University) and Richard Winschel (CONSOL Energy, Inc.) as well as Channele Wirman and Rebecca Peterson (EIA) informed this work. Reviews by David Tabet (Utah Geological Survey) improved the paper. Finally, the thoughtful comments and suggestions from my anonymous reviewers are acknowledged and appreciated.