MEMORANDUM  |  19 September 2006

TO	James DeMocker, EPA/OAR/OPAR

from	Jim Neumann, IEc

	Jim Wilson and Andy Bollman, EH Pechan and Associates

subject	Documentation of Analysis of 1990-2002 Emissions for Select 

Non-EGU Stationary Point Sources



	

A recent white paper prepared by EPA/OAQPS staff suggests that the
emissions projection methodology that the Agency has historically used
for non-EGU stationary and nonpoint sources may overestimate actual
emissions for these sources.  As part of our quality assurance reviews
of the draft Section 812 emissions inventories recently put before the
SAB Council Air Quality Modeling Subcommittee, IEc and Pechan have
compared emissions data from the 2002 NEI for select non-EGU stationary
point source categories with estimates based on projection of the 1990
NEI estimates using EPA's current emissions projection methodology. The
results allow us to evaluate whether actual 2002 emissions from an ex
post perspective (derived from the 2002 NEI) are comparable to estimates
that would have been obtained from an ex ante perspective using the 1990
NEI and the current projection methods.  If the ex ante estimates
correspond well with ex post results, we can conclude that the
projection method provides a reasonably unbiased estimate; if the ex
ante estimate is higher than the ex post estimate, we can conclude that
the projection method may overestimate actual emissions for the
1990-2002 period, as the white paper suggests.  This memorandum
documents the methods and the results of that analysis, and provides a
discussion of the implications of this analysis for the Section 812
emissions inventories.

In summary, the results of our analysis do not conclusively support a
particular projection method or a broad adjustment to one component of
the method.  Instead, the results provide insights on the types of
biases the method may present and the sector-specific research needed to
improve projections.  For some source categories and pollutants, the
projection method appears to provide an unbiased estimate.  For others,
the projection is higher than actual, in a few cases by a very wide
margin.  In still other cases, the projection is lower than actual,
though the differences in those cases appear to be slight.  Furthermore,
within a particular source category, the results are mixed across
pollutants.  The implications of these results are that we recommend
that sulfur emissions from copper smelters be adjusted in the Section
812 without CAAA scenario to reflect the unanticipated steep drop in
activity over the 1990-2002 period.

Methods

The analysis involved four steps: 1) Identify a short list of non-EGU
stationary point source emitting sectors and pollutants for analysis; 2)
Collect information on known regulatory actions at the federal or local
level that went into effect between 1990 and 2002 and that affected
emissions from sources in each of the selected sectors; 3) Identify the
relevant emissions driver used in the Section 812 analysis for the
sector; 4) Calculate an "alternative projection" for 2002 for each
pollutant-sector combination based on the 1990 NEI, "grown" using the
emissions driver identified in Step 3, and adjusted for the impact of
regulatory actions identified in Step 2.

We identified six sectors for this analysis: auto assembly plants;
copper smelters; glass manufacturing; cement manufacturing (kilns);
petroleum refineries; and industrial boilers.  For all but one of these
sectors we looked at national emissions estimates; for industrial
boilers, we looked only at sources in the Midwest RPO, comprising five
states: Illinois, Indiana, Michigan, Ohio, and Wisconsin.  We chose a
limited geographic area for the industrial boilers in part because of
the ready availability of data for that region on regulatory actions
during the 1990-2002 period, and in part to ensure the analysis was
manageable in scope.  The sectors represent a range of pollutants,
geographic distribution of sources, and economic output trends -
nonetheless, they are a sample of convenience and we can make no
conclusive statements about the representativeness of the results for
the full range of non-EGU point sources.  In total, the six sector scope
accounts for approximately 27, 21, 9.6, and 1.3 percent of total SO2,
NOx, VOC and PM2.5 emissions, respectively, in the non-EGU point source
component of the 2002 NEI.

Information on regulatory activity in the 1990 to 2002 period was
collected from readily available sources, mostly information from
Regional Planning Organizations that had also been used to develop the
Section 812 with-CAAA emissions projections for 2010 and 2020.  The
sources and information we accessed is described in detail in Chapter 3
of Emission Projections for the Clean Air Act Second Section 812
Prospective Analysis: Draft Report of June 23, 2006, prepared by IEc and
Pechan for EPA/OAR/OPAR.  In this analysis, however, we accessed these
sources for parallel information on the 1990-2002 period; regulatory
activity in that period is not directly used in the Section 812 reports,
which take the 2002 NEI as the "launching point" for projection of the
with-CAAA scenarios and therefore implicitly incorporate the impact of
regulation through 2002.

The emissions drivers used in this analysis are the same drivers used in
the Section 812 analysis; most are based on historical data obtained
from DOE that are consistent with the AEO 2005 projections, either of
fuel consumption (e.g., for industrial boilers and cement kilns) or
economic activity.  These are described in detail in Attachment A to
this memo, which also includes the SCC scope of the emissions analysis
by sector.  In general, in this exercise, economic output-based drivers
were used for auto assembly plants, copper smelters, glass
manufacturing, and cement manufacturing; fuel-consumption based output
drivers were used for industrial boilers; and a mix of fuel-consumption
based (for process heaters) and product throughput based factors (for
sources such as fugitive emissions) are used for petroleum refiners. 
Note that, when a fuel-based factor is used, there is differentiation by
fuel type (e.g., natural gas versus coal), but there is no
differentiation within fuel type to indicate such qualities as sulfur
content.

The results presented below include, for each pollutant sector
combination, two NEI reference points and two estimated values.  The
reference points are the reported 1990 and 2002 NEI emissions.  The 2002
NEI estimate is also the Section 812 with-CAAA scenario for that year;
it is effectively our ex post value.  The first estimated value is what
we call the "2002 Alternative Projection".  It is projected from the
1990 NEI with adjustment for growth and identified federal and local
regulatory activity.  This first estimated value is effectively our ex
ante value.  Differences between this estimated ex ante value and the
2002 NEI ex post estimate provide the essential results of the
reasonableness test for the current emissions estimation method.  The
second estimated value is the 2002 without CAAA projection incorporated
as the baseline case for the 812 study.  The without-CAAA scenario is
projected from the 1990 NEI by applying growth factors, but without
adjustment for regulatory activity; it represents a counterfactual
estimate, as if regulatory stringency were frozen at 1990 levels but
emissions activity were allowed to grow.  The estimate is included here
as a reference point, to assess the incremental impact of regulation
during 1990-2002, and to assess implications of the results for the 812
study.  

Results and discussion

The results of the analysis are summarized in Exhibit 1 at the end of
this memo.  As noted above, the results are mixed, but generally tend to
support the contention that the method overstates emissions.  For two
pollutant-sector combinations, it appears the ex ante estimate, based on
projection from the 1990 NEI, closely simulates the ex post value from
the 2002 NEI, within five percent (e.g., NOx emissions for industrial
boilers and glass manufacturing).  For most categories, however, the
method overstates the ex post value, in some cases substantially (e.g.,
SO2 emissions from copper smelters and industrial boilers, PM emissions
from cement manufacturing, VOC emissions from refineries).  For two
categories, the method actually underestimates the ex post value,
although not by a large margin, between five and ten percent (e.g., NOx 
emissions from cement and glass manufacturing).  We have examined the
reasons for these results by sector, reported below.

When interpreting the results it is important to keep in mind several
factors:

New stack tests or the release of new emission factors by EPA can
produce emission changes in situations where no emission control device
has been added, or no operational changes have occurred.  In some cases,
noted below, we believe updated emissions factors are a main reason
behind poor performance of the ex ante approach.

The base year emissions are less accurately determined or determined
differently  in the 1990 NEI than in the 2002 NEI.  Sometimes, the 1990
NEI estimate is actually from an earlier year that has been carried
forward to 1990 using an economic growth projection.  

Reductions in emissions rates might not be reflected if controls that
might have been applied in the interim period are missed.

It is very difficult in the ex ante approach to predict the impact of
fuel switching to avoid exceeding permit limits or triggering PSD or new
source review provisions on a local level.  In some cases, noted below,
it appears that fuel switching may be an explanation for overestimates
of sulfur dioxide emissions projections.

As pointed out in the OAQPS white paper, capital turnover –
retirements and/or replacements - often leads to better emissions
efficiency.  In several cases noted below, we believe this is an
important omitted factor in the ex ante estimates.  Note also that we
did not attempt to account for NSPS or New Source Review in our
methods—capital turnover may, in fact, be the result of an effort to
respond to NSPS/NSR requirements.

It is difficult to ensure that all relevant state and local regulations
are included in our analysis.  While it is very difficult to assess the
impact of missing information in our retrospective analysis, this may
not be as large an issue for the  prospective application of the ex ante
method in the Section 812 prospective analysis, because the Regional
Planning Organizations have a strong incentive to accurately capture the
impact of all prospective regulation in their member states as part of
the regional haze planning process.  As noted above, RPOs are the main
source of information in the Second Prospective analysis for
"on-the-books" state and local regulations.

Auto Assembly Plants

The results for auto assembly plants in Exhibit 1 suggest that the
projection method is not capturing several developments in this industry
in the 1990 to 2002 period.  Our research suggests that there were no
significant regulatory changes between 1990 and 2002 affecting VOC
emissions for this industry, as indicated by the fact that the 2002
alternative projection and 2002 Without CAAA estimates are identical. 
We may have missed some State/local regulations that went into effect
during the period, for example in non-attainment areas.  More likely,
the much lower 2002 NEI estimate appears to reflect capital turnover -
there were facility closures in some locations, and the new facilities
emit at lower NSPS rates. 

It is also interesting to note that the growth factor applied in the
without CAAA and alternative projection are based on projected changes
in the Transportation Equipment sector (NAICS 336), which produces
higher growth factors than the change in U.S. car and truck production
from 1990 to 2002.  Changing the growth factor basis in this case would
produce a lower alternative projection, but would not approximate the
negative emissions trend observed from the 1990 to 2002 NEI.

Glass Manufacturing

The glass manufacturing results show a very close match between the 2002
NEI and the alternative projection, suggesting the projection method
works well for this industry.  The 2002 alternative projection applies a
40 percent NOx control efficiency (80 percent rule effectiveness) in
areas where the CAAA require RACT implementation (to glass furnaces). 
The growth indicator for this sector is glass and glass products output.
 Growth factors vary by region, ranging from 0.99 (mid-Atlantic) to 1.7
(Mountain States).

Copper Smelters

The copper smelter industry results highlight a major issue with
virtually any projection method - it can be very difficult to predict a
major industry collapse.  During the 1990-2002 period, all but three
U.S. copper smelters shut down.  The key industry issue in the time
period is international competition in copper production.  As a result,
sulfur dioxide emissions dropped by almost 90 percent during this
period.  An additional factor contributing to the steep emissions drop
is that one of the smelters still operating underwent a major
modification during the mid-1990s that dramatically lowered SO2
emissions; the impact of this upgrade is not incorporated in the
alternative projection at this point, owing to uncertainty over whether
the upgrade was motivated by the CAAA regulation.

Another interesting insight is that, while application of a
retrospective activity factor ought to have captured this substantial
dropoff in copper smelter activity, in this case the Section 812 study
growth factor applied is too broad to have picked up this phenomenon. 
The growth factor applied in the analysis is for Other Primary Metals
sector, which includes all non-ferrous metals and foundries; copper
smelters were clearly an exception to the otherwise increasing activity
trend in the broader primary metals sector during this period.

Cement Manufacturing

The cement manufacturing results in Exhibit 1 indicate that the
projection method can yield very different outcomes across pollutants
for the same industry.  The results for NOx emissions indicate fairly
good agreement between the 2002 NEI and the alternative projection - the
alternative projection, in fact, underestimates emissions growth of the
1990-2002 period.  

For SO2, however, the alternative projection fails to capture a
substantial reduction in emissions over the 1990-2002 period.  There
were SO2 emission limits imposed for cement kilns during this period,
but very few add-on control devices were installed because the nature of
the manufacturing process reduces SO2 emissions, so few individual
plants violated the limits.  The apparent SO2 reductions from 1990 to
2002 may actually be an indication of better emission measurement
methods, suggesting the 1990 NEI is biased high, or from switching to
lower sulfur coal prompted by non-CAAA motivations, for example, the
rapid adoption of low-cost, low-sulfur Powder River basin coal over this
period.  

The results for particulate matter, however, indicate a complete failure
of the method.  There are only three possible reasons for this large
discrepancy: inaccurate activity indices, overestimated emissions rates,
or omitted regulations.  We can probably eliminate the activity factor
explanation, because that would have affected the apparently accurate
NOx emissions estimates as well.  We may have failed to identify
regulations during this period, but any regulation that had such a
dramatic effect would be difficult to miss, suggesting this could be a
contributing factor at best.  We conclude that the large discrepancy in
the PM estimates results from misestimated emissions in the 1990
inventory.  It is also interesting to note that almost 70 percent of the
PM emissions in 1990 are in New York State; one hypothesis is that New
York may have either used a very high emissions factor in 1990, passed a
regulation that we failed to identify, or both.

Overall, the results for this industry suggest to us that the best
strategy for improving prospective emissions estimates is to carefully
examine emissions factors, which are pollutant specific, rather than
attempt to make adjustments to activity factors.  In this industry,
attempting to adjust activity factors might improve estimates for one
pollutant while reducing accuracy for another pollutant.  

Petroleum Refining

The results for petroleum refining indicate that the projection method
is capable of capturing the effect of regulation, but that the effective
application of static emissions rates in the base case neglects capital
turnover.  During the period 1990 to 2002, 47 refineries shut down. 
Many of these were smaller facilities, and the remaining refineries
generally increased capacity to compensate.  Production increased
relatively slowly, as evidenced by the 2002 Without CAAA estimate being
only modestly higher than the 1990 NEI estimate.  Note that NOx
emissions at these facilities are driven by fuel use, while VOC
emissions are driven by throughput estimates.  It appears that during
the 1990-2002 period fuel intensity per unit of production declined, an
observation consistent with the explicit modeling of energy efficiency
as a result of capital turnover in the AEO modeling on which the
activity estimates are based.

At the same time, CAAA regulations were put in place for this industry. 
The alternative projection includes the effect of petroleum refinery
MACT standards on VOC emissions from process vents and fugitives.  NOx
controls are applied to process heaters in 1-hour ozone nonattainment
areas and the Northeast OTR.  Note that the estimates presented in
Exhibit 1 include emissions from refinery processes, but exclude tankage
and boilers.  The effect of these regulations is to reduce the
alternative projection relative to the 2002 Without CAAA estimate, for
both NOx and VOC emissions.  

The 2002 NEI estimates, however, are substantially lower than the
alternative estimate for both pollutants.  We believe that two factors
are behind this result.  First, consolidation of the industry likely led
to the better environmental performance - the remaining refineries were
likely be more efficient, and therefore have lower emissions rates. 
Second, capacity expansions at these plants may have been made with less
polluting technologies, perhaps in response to regulation or anticipated
regulation at the local level that we failed to identify.  The major
effect on emissions rates of these factors was not captured by the
projection method.

Industrial Boilers

Industrial boiler activity dropped in the Midwest RPO during the
1990-2002 period, as evidenced by the 2002 without CAAA estimate for NOx
being lower than the 1990 estimate.  It appears that, in addition to
reductions in activity, some switching to lower sulfur fuels also
occurred, leading to a steeper decline in sulfur emissions than activity
factors alone would suggest.  The activity factors reflect this shift -
this region has declines in industrial boiler coal and residual oil use
during the period with increases in some other fuels - and therefore the
alternative estimate captures some of this effect.  But not all of the
effect on SO2 and PM emissions is reflected.  

The alternative projection estimates for NOx are actually very close to
the 2002 NEI estimate, reflecting the application of NOx RACT to boilers
in 1-hour ozone nonattainment areas during this period.  However, we
applied no SO2 or PM controls for the 1990 to 2002 period.

Recommendations for Modifying the Section 812 Emissions Inventories

In light of our findings, we recommend the following adjustments be made
to the Section 812 emissions inventories:

Re-estimate copper smelter sulfur dioxide emissions for the Without CAAA
case.  As currently configured, the Without CAAA case for this sector
does not reflect the very steep drop-off in activity that occurred in
this industry.  Without adjustment, the large differences in the 2002
Without CAAA and 2002 NEI emissions for this sector (over 300,000 tons)
would be attributed to the CAAA, which would significantly overestimate
both costs and benefits of the Amendments.  We suggest the 2002 Without
CAAA be re-estimated by examining the potential emissions at the one
smelter that installed controls during the 1990-2002 period, resulting
in a national estimate that would be somewhat higher than the national
2002 NEI that is the basis of the With CAAA scenario.  We further
recommend that the re-estimated 2002 Without CAAA value then be used to
project the 2010 and 2020 Without CAAA scenario for this sector.  We
also suggest that an alternative approach be considered for post-2002
projections that make use of a more specific (copper) growth indicator
for this sector. 

Re-examine direct PM emissions estimates from the 1990 NEI for cement
manufacturing and industrial boilers.  The results suggest that the 1990
NEI might be employing very outdated emissions estimates for 1990 for
these sectors; with better information, it might be possible to improve
the accuracy of these retrospective estimates.  Other sectors might also
need to be examined for the same effect.

Examine the causes of fuel switching at cement manufacturers and Midwest
industrial boilers.  The phenomenon of increased penetration of
low-cost, low-sulfur PRB coal is well-established. The rate of fuel
switching in some parts of the country, notably the Midwest and likely
including New York, where there is a high concentration of cement
manufacturing activity and emissions, could be expected to be quite
high, and this effect ought not be attributed to the CAAA.

Consider assessing the impact of consolidation in petroleum refining on
emissions rates in the Without CAAA case.  The emissions projection
techniques applied to this sector appear to have missed the effects of
capital turnover that may be associated with consolidation of petroleum
refining.  One possibly fruitful research objective would be to assess
whether emissions rates in 1990 might have been higher than industry
averages at those plants that closed during the 1990's.  If they were,
consolidation would have led to reductions in the overall emissions
intensity of production in the industry.  This would be an interesting
finding, although in the end it may have little impact on the emissions
estimates we use.  For example, while some studies have concluded that
consolidation was the results of economic forces unrelated to the CAAA,
the role of the CAAA in petroleum refining consolidation might never be
conclusively determined; it is therefore not clear whether a true
Without CAAA scenario might have resulted in at least some of these
potentially less efficient plants staying in operation.  At minimum, it
will be important to ensure that in this sector the difference in
emissions between the 2002 Without CAAA and the 2002 With CAAA emissions
estimates, which will be counted as part of the benefits of the CAAA, is
fully costed and also attributed to the impact of the CAAA. Exhibit 1: 
Results of Emissions Simulation Exercise

Sector	Reference Point or Estimate	Emissions (tons per year)



SO2	NOx	VOC	PM2.5	PM10

Auto Assembly Plants	1990 NEI

	42,200



	2002 NEI (With CAAA)

	34,150



	2002 Alternative Projection

	61,250



	2002 Without CAAA

	61,250



Glass Manufacturing	1990 Emissions

58,507





2002 NEI (with CAAA)

61,348





2002 Alternative Projection

58,674





2002 Without CAAA

70,454



	Copper Smelters	1990 Emissions	323,390





	2002 NEI (with CAAA)	35,000





	2002 Alternative Projection	335,824*





	2002 Without CAAA	335,824





Cement Manufacturing	1990 Emissions	173,203	135,475

44,862	164,230

	2002 NEI (with CAAA)	118,510	154,900

1,630	6,125

	2002 Alternative Projection	199,700	140,275

52,550	192,320

	2002 Without CAAA	199,700	154,250

52,550	192,320

Petroleum Refineries	1990 Emissions

179,039	199,450



	2002 NEI (with CAAA)

128,250	104,103



	2002 Alternative Projection

165,371	175,040



	2002 Without CAAA

181,091	212,963



Industrial Boilers	1990 Emissions	707,190	185,455

16,270

	 (Midwest RPO states:	2002 NEI (with CAAA)	242,275	125,150

3,980

	  IL, IN, MI,OH, WI)	2002 Alternative Projection	439,450	130,212

12,120



2002 Without CAAA	439,450	161,520

12,120

	NOTES: The memo text explains the derivation of the four reference
points and estimates.  All estimates are national (48-state) in scope,
except where noted. 	Shaded cells indicate comparisons where growth
projections are based on AEO fuel use rather than throughput estimates. 
See text for explanation of copper smelter alternative projection.

 The document, "Improving EPA Emissions Forecasting For Regulatory
Impact Analyses" describes an interim approach for forecasting emissions
to future years to be used in regulatory analyses, and is available at:
www.epa.gov/ttn/ecas/articles.html

 Note that different methods and data are used in the Section 812 study
than in the PM NAAQS RIA for emissions projections.  The RIA is based on
REMI/EGAS output, which has more detailed industry sector output data,
while the 812 study is based on AEO 2005, which is more highly
aggregated but based on integrated population, fuels use, and economic
output modeling.

 Note that, because we use retrospective emissions driver data, the
estimate is not a true ex ante perspective value.

 Note that for this analysis we did not consider the impact of the
interim or final hazardous waste MACT rule on this sector.  That rule
took effect after 2002, but some facilities may have complied early. 
Many cement kilns burn hazardous waste with other fuels, and therefore
would be subject to the rule.  While the rule's focus is on HAPs, there
are some co-control benefits for PM.

 Note that our recommendations focus on adjustments necessary to avoid
biasing the assessment of benefits and costs when we ultimately compare
the With CAAA and Without CAAA results.  In cases where an estimation
bias exists in both scenarios, the need for an adjustment is less
urgent.  This differs from the perspective in the OAQPS White Paper,
where accuracy in the absolute level of emissions is critically
important to projecting the likelihood of PM NAAQS attainment.  Also,
evaluating the utility of any potential adjustment to the 812
inventories should consider both the relative significance of the
source/pollutant component within the overall 812 inventories as well as
the severity of any inconsistencies such adjustments may lead to with
respect to the AEO energy driver data and other inputs.

	

				  PAGE  2 	

 

																			  PAGE  10 

Confidential / Not for Release  |   draft date

