Deliberative, draft document - Do not cite, quote, or distribute

Development of Emissions Estimating
Methodologies for Dairy Operations

Draft

Prepared by:

U.S. Environmental Protection Agency

Office of Air and Radiation
Office of Air Quality Planning and Standards
109 T.W. Alexander Drive
Research Triangle Park, N.C. 27709

June 2022

This document is a preliminary draft. It has not been formally released by the U.S. Environmental Protection Agency
(EPA) and should not at this stage be construed to represent Agency policy. It is being circulated for comments on its
technical merit.


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table of Contents

1	INTRODUCTION	1-1

1.1	Confinement Site Descriptions	1-1

1.1.1	CA5B	1-1

1.1.2	IN5B	1-2

1.1.3	NY5B	1-3

1.1.4	WA5B	1-4

1.1.5	WI5B	1-6

1.2	Open Source Site Descriptions	1-6

1.2.1	IN5A	1-7

1.2.2	TX5A	1-8

1.2.3	WA5A	1-9

1.2.4	WI5A	1-10

1.3	Data Sampled	1-11

1.3.1	Particulate Matter	1-11

1.3.2	Animal Husbandry	1-12

1.3.3	Biomaterials Sampling Methods and Schedule	1-12

2	REVISIONS TO DATA SET AND EMISSIONS DATA SUMMARY	2-1

2.1	Revisions to the 2010 Data Set	2-1

2.2	Comparison between the 2010 and Revised Barn Data Sets	2-2

2.2.1	Mechanically Ventilated Barns	2-2

2.2.2	Naturally Ventilated Barns	2-3

2.2.3	Milking Centers	2-4

2.3	Data Completeness Criteria for the Revised Data Set	2-5

2.4	Comparison Between the Revised Data Sets and NAEMS Datasets Used in Peer-

reviewed Published Papers	2-6

2.4.1	Naturally Ventilated Barns	2-7

2.4.2	Open sources	2-8

3	RELATIONSHIPS ESTABLISHED IN LITERATURE	3-1

3.1	NH3 and H2S from Confinement Sources	3-1

3.2	Particulate Matter from Barns	3-4

3.3	NH3 and H2S for Open Sources	3-5

4	SITE COMPARISON, TRENDS, AND ANALYSIS	4-1

4.1	Mechanically Ventilated Dairy Barns (IN5B-B1, IN5B-B2, NY5-B1, WI5B-B1 and

WI5B-B2)	4-2

4.1.1	Emissions data	4-2

4.1.2	Environmental data	4-3

4.1.3	Ambient Data	4-6

4.2	Milking Centers (IN5B-MC and NY5B-MC)	4-7

4.2.1 Emissions Data	4-7

1


-------
Deliberative, draft document - Do not cite, quote, or distribute

4.2.2	Environmental data	4-7

4.2.3	Ambient Data	4-9

4.3	Naturally Ventilated Barns (CA5B-B1, CA5B-B2, WA5B-B2 and WA5B-B4)	4-10

4.3.1	Emissions Data	4-10

4.3.2	Environmental Data	4-11

4.3.3	Ambient Data	4-13

4.4	Open Sources (IN5A, WI5A and TX5A)	4-14

4.4.1	Emissions Data	4-14

4.4.2	Environmental Data	4-15

4.4.3	Ambient Data	4-15

5	DEVELOPMENT AND SELECTION OF MODELS FOR DAILY EMISSIONS	5-1

5.1	Mechanically Ventilated Barns	5-1

5.2	Milking Centers	5-4

5.3	Naturally Ventilated Barns	5-7

5.4	Open Sources	5-9

5.5	Corrals	5-11

6	MODEL COEFFICIENT EVALUATION	6-1

6.1	Mechanically Ventilated Barns Model	6-1

6.1.1	NH3 Model Evaluation	6-1

6.1.2	H2S Model Evaluation	6-3

6.1.3	Particulate Matter Models	6-5

6.2	Milking Centers	6-5

6.2.1	NH3 Model Evaluation	6-5

6.2.2	H2S Model Evaluation	6-6

6.2.3	Particulate Matter Model Evaluation	6-7

6.3	Naturally Ventilated Barn Model	6-8

6.3.1	NH3 Model Evaluation	6-8

6.3.2	H2S Model Evaluation	6-10

6.3.3	PM10 Model Evaluation	6-12

6.3.4	PM25 Model Evaluation	6-14

6.3.5	TSP Model Evaluation	6-16

6.4	Open Sources	6-18

6.4.1	NH3 Model Evaluation	6-19

6.4.2	H2S Model Evaluation	6-20

7	ANNUAL EMISSION ESTIMATES AND MODEL UNCERTAINTY	7-1

8	MODEL APPLICATION AND ADDITIONAL TESTING	8-1

8.1 Model Application Example	8-1

8.1.1	Mechanically Ventilated Barn Example	8-3

8.1.2	Milking Center Example	8-4

8.1.3	Naturally Ventilated Barn Example	8-4

8.1.4	Lagoon Example	8-5

8.1.5	Corral Example	8-6

11


-------
Deliberative, draft document - Do not cite, quote, or distribute

8.1.6 Combining Structures	8-7

8.2	Model Sensitivity Testing	8-7

8.2.1	Sensitivity to Inventory	8-7

8.2.2	Sensitivity to Climate	8-8

8.2.3	Model Limitations	8-16

8.3	Comparison to Literature	8-31

8.3.1	Mechanically Ventilated Barn	8-32

8.3.2	Naturally Ventilated Barn	8-33

8.3.3	Lagoon	8-33

8.3.4	Corral	8-34

8.4	Replication of Independent Measurements	8-34

8.4.1	Lagoon	8-35

8.4.2	Corral	8-35

9	CONCLUSIONS	9-1

10	REFERENCES	10-1

List Of Appendices

Appendix A - PM Sampling
Appendix B - Data Processing
Appendix C - Data Completeness
Appendix D - Summary Statistics
Appendix E - Time Series Plots
Appendix F - Scatter Plots
Appendix G - Modeling Results

in


-------
Deliberative, draft document - Do not cite, quote, or distribute

List of Tables

Table 1-1: Dairy Confinement Sites Monitored Under NAEMS	1-1

Table 1-2: Dairy Open Source Sites Monitored Under NAEMS	1-7

Table 2-1. Percent difference in NH3 summary statistics between the 2010 and revised dataset (at 75%

data completeness)	2-3

Table 2-2. Percent difference in H2S summary statistics between the 2010 and revised dataset (at 75%

data completeness)	2-3

Table 2-3. Percent difference in PM summary statistics between the 2010 and revised dataset (at 75% data

completeness)	2-3

Table 2-4. Percent difference in PM10 summary statistics between the 2010 and revised dataset (at 75%

data completeness)	2-4

Table 2-5. Percent difference in TSP summary statistics between the 2010 and revised dataset (at 75%

data completeness)	2-4

Table 2-6. Percent difference in NH3 summary statistics between the 2010 and revised dataset (at 75%

data completeness)	2-5

Table 2-7. Percent difference in H2S summary statistics between the 2010 and revised dataset (at 75%

data completeness)	2-5

Table 2-8. Percent difference in PM summary statistics between the 2010 and revised dataset (at 75% data

completeness)	2-5

Table 2-9. Comparison of naturally ventilated NH3 emissions in the model dataset to published datasets.2-
7

Table 2-10. Comparison of naturally ventilated H2S emissions in the model dataset to published datasets.

	2-7

Table 2-11. Comparison of lagoon and basin NH3 emissions in the model dataset to published datasets.2-8
Table 2-12. Comparison of corral (TX5A) NH3 emissions in the model dataset to published datasets. ...2-8

Table 4-1: Relationship classification based on R2 values	4-2

Table 4-2. Year mechanically ventilated barns were constructed	4-4

Table 4-3. Mechanically ventilated environmental parameter regression analyses	4-5

Table 4-4. Mechanically ventilated ambient parameter regression analyses	4-6

Table 4-5. Milking center environmental parameter regression analyses	4-9

Table 4-6. Milking center ambient parameters regression analyses	4-10

Table 4-7. Naturally ventilated environmental parameter regression analyses	4-12

Table 4-8. Naturally ventilated ambient parameters regression analyses	4-14

Table 4-9. Open source environmental parameter regression analyses	4-15

Table 4-10. Open source ambient parameters regression analyses	4-16

Table 4-11. Corral ambient parameters regression analyses	4-16

Table 5-1. Parameter combinations tested as mechanically ventilated barn models for NH3 and H2S

emissions	5-1

Table 5-2. Parameter combinations tested as mechanically ventilated barn models for PM10, PM2.5, and

TSP emissions	5-1

Table 5-3. Selected daily models for mechanically ventilated barns	5-3

Table 5-4. Summary of barn construction dates for mechanically ventilated barns	5-3

Table 5-5. Parameter combinations tested as milking center models	5-4

Table 5-6. Selected daily models for milking centers	5-7

Table 5-7. Parameter combinations tested as naturally ventilated barns models	5-8

Table 5-8. Selected daily models for naturally ventilated barns	5-9

Table 5-9. Parameter combinations tested as open source models for NH3 and H2S emissions	5-9

Table 5-10. Selected daily models for lagoons sources	5-10

Table 5-11. Parameter combinations tested as corral models for NH3 and H2S emissions	5-11

Table 5-12. Selected daily models for corrals	5-12

IV


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 6-1. Model coefficients developed using the jackknife approach for NH3 emissions from

mechanically ventilated barns	6-2

Table 6-2. Model fit statistics for the mechanically ventilated barns NH3 jackknife	6-2

Table 6-3. Model coefficients developed using the jackknife approach for H2S emissions from

mechanically ventilated barns	6-4

Table 6-4. Model fit statistics for the mechanically ventilated barns H2S jackknife	6-4

Table 6-5. Model coefficients developed using the jackknife approach for NH3 emissions from milking

centers	6-6

Table 6-6. Model fit statistics for the milking center NH3 jackknife	6-6

Table 6-7. Model coefficients developed using the jackknife approach for H2S emissions from milking

centers	6-7

Table 6-8. Model fit statistics for the milking center H2S jackknife	6-7

Table 6-9. Model coefficients developed using the jackknife approach for NH3 emissions from naturally

ventilated barns	6-9

Table 6-10. Model fit statistics for the naturally ventilated barns NH3 jackknife	6-9

Table 6-11. Model coefficients developed using the jackknife approach for H2S emissions from naturally

ventilated barns	6-11

Table 6-12. Model fit statistics for the naturally ventilated barns H2S jackknife	6-11

Table 6-13. Model coefficients developed using the jackknife approach for PM10 emissions from naturally

ventilated barns	6-13

Table 6-14. Model fit statistics for the naturally ventilated barns PM10 jackknife	6-13

Table 6-15. Model coefficients developed using the jackknife approach for PM2 5 emissions from

naturally ventilated barns	6-15

Table 6-16. Model fit statistics for the naturally ventilated barns PM2 5 jackknife	6-15

Table 6-17. Model coefficients developed using the jackknife approach for TSP emissions from naturally

ventilated barns	6-17

Table 6-18. Model fit statistics for the naturally ventilated barns TSP jackknife	6-17

Table 6-19. Model coefficients developed using the jackknife approach for NH3 emissions from open

sources	6-19

Table 6-20. Model fit statistics for the open sources NH3 jackknife	6-19

Table 6-21. Model coefficients developed using the jackknife approach for H2S emissions from open

sources	6-20

Table 6-22. Model fit statistics for the open source H2S jackknife	6-20

Table 7-1. Back transformation parameters	7-1

Table 7-2. Annual Uncertainty Model Details	7-2

Table 8-1. Daily calculation parameter values	8-3

Table 8-2. Comparison of confinement source NH3 emissions (kg) on January 1, 2021, for different

inventory levels at a theoretical Brown County farm	8-8

Table 8-3. Comparison of confinement source total 2021 NH3 emissions (kg) for different inventory

levels at a theoretical Brown County farm	8-8

Table 8-4. Comparison of lagoon NH3 emissions (kg) for different surface areas for theoretical Brown

County farm	8-8

Table 8-5. Comparison of estimated corral NH3 emissions (kg) for different inventory levels for

theoretical Brown County farm	8-8

Table 8-6. Summary of average daily temperature at the two meteorological sites	8-10

Table 8-7. Summary of average daily relative humidity at the two meteorological sites	8-11

Table 8-8. Summary of average daily wind speeds at the two meteorological sites	8-11

Table 8-9. Total annual emission from a theoretical mechanically ventilated barn in WI and CA	8-12

Table 8-10. Total annual emission from a theoretical milking center in WI and CA	8-13

Table 8-11. Total annual emission from a theoretical milking center in WI and CA	8-15

Table 8-12. Total annual emission from a theoretical lagoon in WI and CA	8-15

v


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 8-13. Total annual emission from a theoretical milking center in WI and CA	8-16

Table 8-8-14. Parameter ranges tested for the dairy models	8-17

Table 8-15. Emission factors for dairy barns from literature	8-31

Table 8-16. Emission factors for dairy lagoons from literature	8-32

Table 8-17. Emission factors for dairy corrals from literature	8-32

Table 8-18. Comparison of resulting mechanically ventilated scrape barn NH3 emission from various

estimation methods	8-32

Table 8-19. Comparison of resulting mechanically ventilated flush barn NH3 emission from various

estimation methods	8-32

Table 8-20. Comparison of resulting naturally ventilated barn NH3 emission from various estimation

methods	8-33

Table 8-21. Comparison of resulting naturally ventilated barn H2S emission from various estimation

methods	8-33

Table 8-22. Comparison of resulting dairy lagoon NH3 emission from various estimation methods	8-34

Table 8-23. Comparison of resulting dairy corral NH3 emission from various estimation methods	8-34

Table 8-24. Model performance evaluation statistics for lagoon NH3 estimates	8-35

Table 8-25. Model performance evaluation statistics for corral NH3 estimates	8-36

List of Figures

Figure 1-1: CA5B farm layout	1-2

Figure 1-2. IN5B farm layout	1-3

Figure 1-3. NY5B farm layout	1-4

Figure 1-4. WA5B farm layout	1-5

Figure 1-5. WI5B farm layout	1-6

Figure 1-6:Aerial view ofIN5A	1-8

Figure 1-7. Aerial view of TX5A	1-9

Figure 1-8. Aerial view of WA5A	1-10

Figure 1-9. Aerial view of WI5A	1-11

Figure 2-1. Ratio of mean predicted emissions for portion of day with valid emissions measurements to
mean predicted emissions for the complete day at the finishing (A) and sow (B) farm. Error

plotted against number of valid 30-minute measurements (from Grant et al., 2013b)	2-6

Figure 6-1. Comparison of variation in coefficients and standard errors for NH3 mechanically ventilated

barn model	6-3

Figure 6-2. Comparison of variation in coefficients and standard errors for H2S mechanically ventilated

barn model	6-5

Figure 6-3. Comparison of variation in coefficients and standard errors for NH3 milking center model. .6-6
Figure 6-4. Comparison of variation in coefficients and standard errors for H2S milking center model.. 6-7
Figure 6-5. Comparison of variation in coefficients and standard errors for NH3 naturally ventilated barn

model	6-10

Figure 6-6. Comparison of variation in coefficients and standard errors for H2S naturally ventilated barn

model	6-12

Figure 6-7. Comparison of variation in coefficients and standard errors for PM10 naturally ventilated barn

model	6-14

Figure 6-8. Comparison of variation in coefficients and standard errors for PM2 5 naturally ventilated barn

model	6-16

Figure 6-9. Comparison of variation in coefficients and standard errors for TSP naturally ventilated barn

model	6-18

Figure 6-10. Comparison of variation in coefficients and standard errors for NH3 open source model.. 6-19

VI


-------
Deliberative, draft document - Do not cite, quote, or distribute

Figure 6-11. Comparison of variation in coefficients and standard errors for H2S open source model. .6-21

Figure 8-1. 2017 Census of Agriculture plot indicating dairy inventory	8-9

Figure 8-2. Comparison on average daily temperatures at test locations in Wisconsin (WI) and California

(CA)	8-10

Figure 8-3. Comparison of average daily relative humidities at test locations in Wisconsin (WI) and

California (CA)	8-10

Figure 8-4. Comparison of average daily wind speeds at test locations in Wisconsin (WI) and California

(CA)	8-11

Figure 8-5. Comparison of daily mechanically ventilated barn emission at test dairy in locations WI and

CA	8-12

Figure 8-6. Comparison of daily milking center emission at test dairy locations in WI and CA	8-13

Figure 8-7. Comparison of daily naturally ventilated barn emission at test dairy locations in WI and CA.8-
14

Figure 8-8. Comparison of daily lagoon emission at test dairy locations in WI and CA	8-15

Figure 8-9. Comparison of daily milking center emission at test dairy locations in WI and CA	8-16

Figure 8-10. Mechanically ventilated barn limitation tests for H2S	8-18

Figure 8-11. Mechanically ventilated barn limitation tests for NH3	8-19

Figure 8-12. Milking center limitation tests for gaseous pollutants	8-20

Figure 8-13. Milking center limitation tests for particulate matter	8-21

Figure 8-14. Maximum values of relative humidity for each temperature at which the PM10 equation

yields negative emissions	8-22

Figure 8-15. Naturally ventilated barn limitation tests for gaseous pollutants	8-23

Figure 8-16. Maximum values of inventory for each wind speed at which the NH3 equation yields

negative emissions	8-23

Figure 8-17. Naturally ventilated barn limitation tests for PM10	8-25

Figure 8-18. Naturally ventilated barn limitation tests for TSP	8-26

Figure 8-19. Naturally ventilated barn limitation tests for PM2 5	8-27

Figure 8-20. Maximum values of wind speed and temperature for each inventory level at which the

particulate matter equations yields negative emissions	8-28

Figure 8-21. Lagoon limitation tests for gaseous pollutants	8-29

Figure 8-22. Corral limitation tests for H2S	8-30

Figure 8-23. Corral limitation tests forNH3	8-30

Figure 8-24. Maximum values of wind speed and relative humidity for each temperature at which the

particulate matter equations yields negative emissions	8-31

Figure 8-25. Scatter plot of the observed lagoon NH3 emissions versus the emission model estimates. 8-35
Figure 8-26. Scatter plot of the observed corral NH3 emissions versus the emission model estimates. ..8-36

vii


-------
Deliberative, draft document - Do not cite, quote, or distribute

GLOSSARY / ACRONYMS

Acronyms	Defintion

-2LogL	negative twice the likelihood

ACI	Akaike information criterion

ACIc	Adjusted Akaike information criterion

ADMs	average daily means

AFO	animal feeding operation

BIC	Schwarz Bayesian Information Criterion

bLS	backward Lagrangian Stochastics

d	day

dsm3	Dry standard cubic meter

EEMs	Emissions estimating methodologies

EPA	Environmental Protection Agency

FANS	Fan Assessment Numeration System

g	gram

g/d	grams/day

H2S	hydrogen sulfide

hd	head - inventory of cows

hPa	hectopascal

kg	kilogram

LAW	live animal weight

LNME	Normalized mean bias of natural log data

m	meter

MB	mean bias

MC	milking center

ME	Mean error

MS	Manure solids

MUN	Milk urea nitrogen

NAEMS	National Air Emissions Monitoring Study

NCEI	National Centers for Environmental Information

NH3	ammonia

NMB	normalized mean bias

NME	normalized mean error

PI	Principal Investigator

PM	particulate matter

PM10	particulate matter with aerodynamic diameters less than 10 micrometers

PM2.5	PM with aerodynamic diameters less than 2.5 micrometers

QAPP	quality assurance project plan

QC	quality control

s	second

SAB	science advisory board

SDS	Separated digested solids

SP	Settling ponds

SS	Solid separation

SS	separated solids

Vlll


-------
Deliberative, draft document - Do not cite, quote, or distribute

Std Dev	Standard deviation

TAN	total ammoniacal nitrogen

TEOM	tapered element oscillating microbalance

TKN	total Kjeldahl nitrogen

TSP	total suspended particulate

USDA	U.S. Department of Agriculture

VOCs	volatile organic compounds

VRPM	vertical radial plume mapping

IX


-------
Deliberative, draft document - Do not cite, quote, or distribute

1 INTRODUCTION

1.1 Confinement Site Descriptions

Five milk production facilities (dairy operations) had barns monitored under NAEMS.
The locations were selected based on site-specific factors including representativeness of facility
age, size, design and management, and herd diet and genetics. Three free stall and two open free
stall dairy facilities were monitored as a part of NAEMS. Table 1-1 summarizes the sites and
their characteristics.

Table 1-1: Dairy Confinement Sites Monitored Under NAEMS

Site

Monitoring
Period

Site Type

Ventilation
type

Number
of barns
measured

Manure
Collection

Manure
Storage4

Bedding
Type5

NY5B

10/24/07 -
10/23/09

Free stall

Mechanically
Ventilated

l3

Scrape

Digester/
SS/SSP

SDS

IN5B

8/24/07 -
8/23/09

Free stall

Mechanically
Ventilated

23

Scrape

Digester/
SS/Lagoon

SDS

WI5B1

9/12/07 -
10/31/09

Free stall

Mechanically
Ventilated

2

Flush

SP/Lagoon

Mattress/
shavings

CA5B

9/26/07 -
2/1/10

Open
free stall2

Naturally
Ventilated

2

Flush

SP/Lagoon

Soil/MS/
Almond shells

WA5B1

9/28/07 -
9/27/09.

Open
free stall2

Naturally
Ventilated

2

Flush

SP/SS/
SSP/Basin

MS

1Barn sites that also have measured area sources.

2Cows are free to walk from open free stall barn into dry lots between the barns.

3Monitored units include the milking center.

4Labeled consistent with the site reports, where: SP = Settling Pond; SS = solid separation; SSP= Solid Storage Pad

5MS = Manure solids; SDS = Separated digested solids

1.1.1 CA5B

In 2010, the California site (CA5B) was a 1,200-cow Holstein dairy farm. The farm has
two naturally ventilated free stall barns, a milking center, and a lagoon and settling ponds (Figure
1-1). The farm also included exercise lots, which were located adjacent to each barn. Lactating
cows were milked two times daily in the centrally located milking center. The on-site heifer
program (i.e., activities to raise their own heifer calves until they can join the milking herd) was
held on the north end of the farm, separated from the study area.

The two naturally ventilated free stall barns, barn 1 (Bl) and barn 2 (B2), were monitored
as part of NAEMS (Zhao, et al., 2010). Each barn had four free stall rows, two on each side of a
central feed lane, housing 600 cows each. Barn 1 had the fresher cows (i.e., cows that recently
gave birth) and served as the breeder barn, while barn 2 had pregnant lactating cows and the hard
breeders (i.e., cows that have a hard time getting pregnant). The cows were generally inside the
barns, particularly on hot days to provide shade.

1-1


-------
Deliberative, draft document - Do not cite, quote, or distribute

The manure handling system included a barn flushing system, three settling ponds and a
lagoon. Manure solids taken from the settling ponds were spread on nearby fields in the spring
and fall.

-Lagoon

¦•Settling ponds

N

Hay shed
' Bam

Barns 1 and 2
were monitored

Eon-

Exercise pen

Freestall bam 1

OFtS

	IZZI	

Exercise pen

Freestall bam 2

Exercise pen

Milking
center





Exercise
per



Maternity/
bam

P

Figure 1-1: CA5B farm layout.

Source: Zhao, et al. (2010)

1.1.2 IN5B

The dairy farm in Indiana (IN5B) had 3,400-head capacity of Holstein cows. The dairy
consisted of two free stall barns, a holding barn, milking parlor, and a dry cow barn (Figure 1-2).
NAEMS gathered measurements from the two freestall barns, barn 1 (B1) and barn 2 (B2), and
the milking center (MC), which consisted of the holding barn (area where cows waited
approximately 45 minutes prior to milking) and milking parlor (Lim, et al., 2010). Each barn

1-2


-------
Deliberative, draft document - Do not cite, quote, or distribute

used a bank of exhaust fans to pull air through the barns. Each barn housed typically housed up
to 1,700 cows, with approximately 3,400 Holstein cows were milked three times a day in the 72-
stall rotary parlor. For the NAEMS, measurements of airflow and emissions focused on the western
half of each of the barns.

The manure was removed from both freestall barns by scraper, while the manure from the
holding barn and milking parlor was flushed. The manure removed from the freestall barn and
milking center are held in a reception pit, and then then directed to a digester that produced
methane gas which was used in generators on the farm. Digester effluent was separated, with the
digested solids moved a storage area and the liquid stored in a two-stage pond/lagoon system.
The liquid was then either irrigated onto or injected into land in the surrounding area. The
separated digested solids were used as bedding in the free stall barns.

4

Milking
parlor

Holding
barn

ir

Freestall barn 1

Dry cow barn

Freestall barn 2

Lagoon stage 2





Digester



Digested
manure
storage

Lagoon stage 1

Figure 1-2. IN5B farm layout.

Source: Lim et al. (2010)

1.1.3 NY5B

The dairy facility monitored in New York (NY5B) had a capacity of 1,000 Holstein cows
and consisted of a mechanically ventilated free stall barn and a milking center, a naturally
ventilated free stall barn, along with housing facilities for dry cows, steers, and calves on the
same site (Figure 1-3). Measurements were collected from the mechanically ventilated 6 row free
stall barn (barn 1 or Bl) and the MC during the study (Bogan, et al., 2010). The MC included a

1-3


-------
Deliberative, draft document - Do not cite, quote, or distribute

double-20 milking parlor, 31 free stalls and four bedded-pack box stalls for special-needs cows.
Cows were brought in for milking three times per day.

The manure was removed from both the B1 and MC by scraper and deposited in a below-
grade gravity flow channel that led to a centralized agitation and pumping station located in the
covered connecting alley between the structures. From the alley, the manure was transferred to
an anaerobic digester. The digester effluent was processed with a screw-press solid-liquid
separator. The separated solids were stockpiled as bedding, land-applied to far-off fields, or sold.
The liquid was pumped to long-term storage that was about 2.3 km away to the northeast.

Dry cow/steer barn

Bunker silos

Figure 1-3. NY5B farm layout.

Source: Bogan, et al. (2010)

1.1.4 WA5B

The dairy facility located in Washington State (WA5B) was a 5,600-head Holstein dairy
farm. The farm buildings included the milking parlor and six naturally ventilated symmetrically-
distributed free stall barns (Figure 1-4). The farm also includes a total of ten corrals/exercise
pens that are distributed around the barns. Two of the free stall barns, barn 2 (B2) and barn 4
(B4), were monitored as part of NAEMS (Ramirez-Dorronsoro, et al., 2010). Barn 2 housed 600
cows in four rows of free stalls and Barn 4 housed 700 cows in six free stall rows.

1-4


-------
Deliberative, draft document - Do not cite, quote, or distribute

Manure from the free stall barns was flushed automatically three times daily and scraped
as needed. The effluent was directed, via pipes, to the waste handling and treatment system that
included a sand separation pit, two primary settling ponds, a manure separation pad (which
includes screen separators and centrifugal solid separators), and a pair of serpentine settling
systems, in which each one had five sequential settling cells. Both serpentine cells then
discharged into a central cell. The liquid effluent from the central cell was directed to the storage
lagoon. The solid effluent from the sand separation pit, depending on the season and
temperature, also was directed to two manure drying ponds, located south of the manure
separator pad. The dried manure was used for bedding and land application, and the liquid was
applied to surrounding fields. The site's lagoon was also monitored as a part of NAEMS (Section
1.2.3).

SSP

tO
~

N

it

Milking center..



/Trim shed



L3

Dry lot

Dry lot I

Bam 1 |

Barn 2

Dry lot

0FIS n !„~
a Dry lot

Barn 3

Barn 4

Dry lot

Dry lot

Bam 5

Barn 6

Dry lot

Dry lot

Dry lot

Dry lot

Lagoon

Wast© handling
and treatment sh
facilities /

Figure 1-4. WA5B farm layout.

Source: Ramirez-Dorronsoro, et al. (2010)

1-5


-------
Deliberative, draft document - Do not cite, quote, or distribute

1.1.5 WI5B

The dairy facility monitored in Wisconsin (WI5B) had a total capacity of 1,700 Hoi stein
dairy cows, and consisted of four free stall barns, a holding barn, and sixth barn that is divided
into the calving pen for 2-year-olds and a hospital barn (Figure 1-5). Two of the free stall barns,
barns 1 (Bl) and 2 (B2), located on the north side of the farm, were monitored as a part of
NAEMS (Cortus et al., 2010). Barn 1 (Bl) had capacity of 275 cows in four rows of free stalls,
and barn 2 (B2) had a capacity of 375 cows housed in five rows of free stalls.

Approximately halfway through the study, the manure removal system was changed in
the barns. Initially, manure was removed by flushing three time per day. The manure flushed
from the parlor, holding pens, and free stall barns was directed to a solid separator. Solids were
directed to pads to wait for land application, while the liquid portion was pumped back into the
vertical tanks to flush the barns. After September 19, 2008, the flush system was replaced with a
tractor scrape system, which was already in use in barns 5 and 6.

Source: Cortus et al. (2010).

1.2 Open Source Site Descriptions

Three dairy lagoons and a dairy corral (TX5A) were monitored under NAEMS (Table
1-2). Sites were selected to capture different stages and manure practices typical of the industry.
The sites selected also represent the broad geographical extent of dairy production to also

1-6


-------
Deliberative, draft document - Do not cite, quote, or distribute

represent different climatological settings for farm and any regional differences in farm
practices.

Dairy lagoon emissions were measured continuously at one farm (IN5A) for one year and
for up to 21 days each season for two years at the two other farms (WA5A and WI5A). The dairy
corral (TX5A) was also monitored for up to 21 days each season for two years.).

Table 1-2: Dairy Open Source Sites Monitored Under NAEMS

Site

Source

Manure

Manure

Monitored

Collection

Storage3

IN5A

Lagoon

Flush

Lagoon

WA5A1

Lagoon

Flush

Lagoon

WI5A1

Lagoon2

Flush

Lagoon

TX5A

Open Corral

Scrape

SB/Lagoon

1	Site that also had barn monitoring sites during NAEMS

2	Lagoon can be single or double stage.

3SB= Settling Basin

1.2.1 IN5A

The Indiana open source site consisted of three barns, a feed storage area, special needs
barn, milking parlor, and an office and tool and repair shops (Figure 1-6). The facility had a
capacity of 2,600 cows (Grant and Boehm, 2010a).

The monitored lagoon received effluent from the parlor and holding area. Manure was
flushed from the holding area and milking parlor every half hour. A small fraction of waste was
held in a slurry tank. The wastewater (flush) from the holding area and milking parlor was
transferred to a settling basin before being transferred to the clay-lined lagoon. The clay-lined
waste lagoon was 85m (280 ft) wide and 116m (380 ft) long, with a surface area of 9,884 m2
(106,400 ft2). Sludge had never been removed from the lagoon (Grant and Boehm, 2010a).

1-7


-------
Deliberative, draft document - Do not cite, quote, or distribute

Manure pit

Figure 1-6:Aerial view of IN5A

Source: Grant and Boehm (2010a)









Manure pit



Slurry tank

Settling
basin

Parlor and
holding

Lagoon under
measurement

1,2.2 TX5A

The Texas dairy (TX5A) consisted of ten corrals, milking parlor, office, hay shed,
commodities barn, calving/fresh cow barn and truck scale (Figure 1-7). The facility had a
capacity of 3,400 Holstein cows (Grant and Boehm, 2010b). Wastewater from the dairy drains to
two earthen sludge/settling basins before entering a retention/treatment structure. Runoff from
the corrals drains to the larger of two retention structures which are connected in series.

Manure was scraped twice a week from the corral surface with some scrapings used as
bedding and the remainder was pushed to the south into ditches, which drained into the runoff
pond. Manure was vacuumed instead of scraped if persistent wet conditions occurred.

1-8


-------
Deliberative, draft document - Do not cite, quote, or distribute

Hospital

Lagoons

Figure 1-7. Aerial view of TX5A

Source: Grant arid Boehm (2010b)

mi

teed

center	'

1 , i g|

Parlor and holding area

Settling basins

Manure

windrows

1,2.3 WA5A

The Washington farm (WA5A) consisted of six barns, a milking parlor, and an office
(Figure 1-8). The facility has a capacity of 4,400 milking cows and 1,200 dry cows in three units
(Grant and Boehm, 2010c). The farm has free stall style barns, with automated flushing that
occurred four times daily. Manure was transferred to an upper settling basin from a sand
separation pit. Liquids were skim separated and then returned as flush to the barns. One lagoon
was actively filled while the other was drying or sludge was being entirely removed. The settled
solids (sludge) were completely removed within a year by front end loader. The settled solids
(sludge) were removed annually by a front-end loader. These remaining solids were then strained
through screens and centrifugal/screw presses, and the liquid transferred to large serpentine
concrete basins for secondary settling. These solids are then dried for bedding. The water
removed from the settled solids is stored in a large, clarified water storage basin for dilution of
barn flush water from the lagoons.

The two upper lagoon/settling basins were measured as part of NAEMS, as well as two
free stall bams described as in Section 1.1.4, Gaseous emissions occur both during lagoon filling
and during sludge removal. The east lagoon was rectangular with dimensions of 183m (600 ft)
by 72 m (235 ft). The west lagoon was five-sided with dimensions of approximately 183 m (600
ft) long and 83m (271 ft) wide with the southwest corner of the lagoon cut off. The east lagoon

1-9


-------
Deliberative, draft document - Do not cite, quote, or distribute

was measured for gaseous emissions. At maximum capacity this lagoon had a liquid depth of 5
m (18 ft), surface area of 13,098 m2 (141,000 ft2) and a volume of 186,300 nr® (2,005,500 ft3).

Figure 1-8. Aerial view of WA5A
Source: Grant arid Boehm (2010c)

1.2.4 WI5A

In 2010, the Wisconsin farm (WI5A) had a total of six barns, a milking parlor with holding pen,
and a special needs area (Figure 1-9). The farm had a capacity of 1,700 Holstein cows (Grant and
Boehm, 2010d). Manure from the free stall barns and the milking parlor complex was removed
by flushing three times daily. The manure flushed from the parlor, holding pen, and free stall
barns flows to a solids separator, from which the solids are removed and stacked on a pad until
they were spread on fields. The liquid effluent from the solids separator was pumped back into
vertical tanks for reuse to flush the barns. Once a week, enough water was removed from the
third stage of the three-stage lagoon and added to the flush tanks to make up for water lost in the
recycled flush system. The three-stage lagoon receives effluent from the two free stall barns
measured by the barn component of NAEMS (Section 1.1.5), as well as the other bams and
milking parlor. The lagoons are pumped out into trucks twice yearly. The first and second stages
of the three-stage lagoon system were monitored, as well as two free stall barns as described in
Section 1.1.5.

1-10


-------
Deliberative, draft document - Do not cite, quote, or distribute

The first lagoon had a width of 52 m (170 ft) and length of 82 m (270 ft). At maximum
capacity, the first lagoon had a surface area of 4,264 m2 (45,900 ft2) and a volume of 10,561 m3
(373,000 ft3). The second lagoon had a width of 37 m (120 ft) and length of 79 m (260 ft). At
maximum capacity, the second lagoon had a surface area of 2,898 m2 (31,200 ft2) and a volume
of 6,420 m3 (226,700 ft3). Both lagoons had liquid depths of 3 m (11 ft) and sludge was last
removed from the second lagoon in 2006.

Figure 1-9. Aerial view of WI5A

Source: Grant and Boehm (2010d)

1.3 Data Sampled

N AHMS collected a host of data from the sites. Data collected included gaseous pollutant
samples, particulate matter samples, meteorological data, confinement parameters, and
biomaterial samples. All procedures for barn sites were outlined in the project Quality Assurance
Project Plan (QAPP) (Heber et al., 2008) and open sources were summarized in open source
project QAPP (Grant, 2008), and are summarized in Section 4 of the main report. The following
section outlines any collection specific to the dairy sites.

1.3.1 Particulate Matter

At any one time, the sampled filterable particulate matter (PM) size class was either equal
to or less than a nominal aerodynamic diameter of 10 micrometers (PMio), and 2.5 micrometers
(PM2.5) or total suspended particulate (TSP). Appendix A contains summary tables, which note

1-11


-------
Deliberative, draft document - Do not cite, quote, or distribute

the particulate matter sampling schedules for the confinement sites. Particulate matter emissions
data were not collected specific to the open sources.

1.3.2	Animal Husbandry

In general, the producer provided pen inventories and information about changes to site
operational procedures like bedding, on a weekly basis. For NY5B, the producer also provided
daily milk production.

1.3.3	Biomaterials Sampling Methods and Schedule

All analyses of biomaterials were performed by an independent laboratory (Midwest
Laboratories, Omaha, NE). Samples were collected based on procedures outlined in the QAPP
(Heber, 2008). Specific sampling details for each site are summarized below. There were no
lagoon samples collected for content analysis.

1.3.3.1	CA5B

Manure sampling was conducted approximately bimonthly during the second year of the
study, with samples collected from the reception lane for the flushed manure in B1 and B2. The
samples were analyzed for solids content, total nitrogen, ammoniacal nitrogen, and ash content to
provide data for the nitrogen balance of the barns.

At the same time as manure sampling, samples of feed and fresh bedding (scraped soil
and manure solids blended with almond shells or rice hulls) were taken from each barn. The
samples were analyzed for solids content, total nitrogen, and ash. Sampling was added late in the
study and only cover the second year of the study (Zhao, et al., 2010).

1.3.3.2	IN5B

Manure in the barns was sampled quarterly between 11/26/07 and 1/20/10. For each
collection, at least four samples were collected from each of the two barns and analyzed for
ammoniacal nitrogen, total nitrogen, pH, total solids, and ash (added later in the study). Samples
of feed were also taken quarterly from each barn and analyzed for total nitrogen, total solids, and
ash. Sampling was added late in the study and only cover the second year of the study (Lim, et
al., 2010).

Bedding and milk tank samples were collected semiannually. Bedding samples were
analyzed for total nitrogen and total solids, while the milk tank samples were only analyzed for
total nitrogen.

1.3.3.3	NY5B

1-12


-------
Deliberative, draft document - Do not cite, quote, or distribute

The daily volume of milk shipped (total milk less non-saleable milk) from the farm was
copied manually from the yearly calendar where milk production was recorded daily by farm
staff. Milk production data from B1 included the cows housed in the MC. Additionally, the farm
reported milk urea nitrogen (MUN) and protein content nearly every day.

Bedding (post-digested separated manure solids) was sampled from each pen on
approximately a monthly basis during the study's second year. The samples were analyzed for
pH, solids content, total nitrogen, and ammoniacal nitrogen, and ash content. A single sample of
the feed and water were taken at the end of the study. The feed was analyzed for solids content,
total nitrogen, and ammoniacal nitrogen, and ash content, while the water sample was analyzed
for total nitrogen, and ammoniacal nitrogen, and sulfur content.

Representative manure samples were collected in B1 from each the four pens, and the
two manure alleys between the outside row of free stalls and the adjacent row of the head-to-
head free stalls. Sampling was conducted approximately monthly during the second year. The
samples were analyzed for pH, solids content, total nitrogen, and ammoniacal nitrogen.

1.3.3.4	WA5B

Sampling was conducted approximately bimonthly during the second year of the study.
Samples of feed, bedding, and manure were taken from each barn. Bedding and feed samples
were analyzed for total solids and total nitrogen content. Manure samples were analyzed for pH,
total solids, total nitrogen, and ammonia content. Milk samples were taken from the holding tank
and analyzed for total nitrogen only.

1.3.3.5	WI5B

Manure in the barns was sampled quarterly for the last year of the study. Each collection
was composed of four samples from each of the two barns. Samples were analyzed for
ammoniacal nitrogen, pH, and total solids.

1-13


-------
Deliberative, draft document - Do not cite, quote, or distribute

2 REVISIONS TO DATA SET AND EMISSIONS DATA SUMMARY

The section catalogs the changes made to the dairy dataset prior to model development
(Section 2.1), considers further changes to the data completeness criteria (Section 2.2), and
finally compares the model development dataset to the initial dataset received in 2010 (Section
2.3) and published literature (Section 2.4) to determine the effect of the data revisions.

2.1 Revisions to the 2010 Data Set

As described in Section 4.2 of the main report, the NAEMS monitoring data were
submitted to EPA in 2010, with revisions submitted in 2015. Revisions included modifying the
approach used to determine the inlet concentrations of ammonia (NH3) and hydrogen sulfide
(H2S) to align time used to determine valid concentrations at the barn inlet and outlet, using a 10-
day running average of inlet concentrations rather than interpolation, and invalidating air flow
rates for periods when the ventilation system was not operating. Corrections were submitted for
IN5B, NY5B, WA5B, and WI5B. A revised file for CA5B was not submitted by the NAEMS
principal investigator (PI).

In addition to the revisions submitted by the PI, EPA reviewed the validity of negative
emission values present in the data set. Negative calculated emission values can occur in the
NAEMS data set due to a range of different scenarios as described in the SAB review of the
2012 emissions estimating methodologies (EEMs) developed by EPA (U.S. EPA SAB, 2013).
These different negative emission scenarios include calculation biases for emission values that
were close to the instrument's detection limit, biases due to lack of lag time corrections, or from
outdoor events that increased pollutant concentration outside of the barns. EPA developed a
procedure for removing negative emission values that resulted from elevated background
concentrations. For this procedure, EPA determined the median emission value for each
pollutant., then excluded negative emissions values that fell outside of a range based on
uncertainty range established in the QAPP for each pollutant the. Appendix B describes this
process in more detail. The negative emissions removed accounted for between 2% (NH3 and
TSP) and 26% (PM2.5) of the total number of average daily emission values available for the
pollutant. Appendix B provides a summary of the number of values removed due to this process
by barn for each pollutant.

The 2010 data sets for dairy open sources (lagoons, basins, and corrals) were provided to
EPA by the NAEMS PI. The datasets contain 30-minute NH3 values obtained using the
backward Lagrangian Stochastics (bLS) model and vertical radial plume mapping (VRPM), and
H2S emissions obtained using the bLS model. The extensive data sets also include fields used to
determine the quality and validity of the emissions data. Based on a literature review of papers
published since NAEMS (Grant & Boehm 2020, Grant et al., 2020, Grant & Boehm 2015, Grant

2-1


-------
Deliberative, draft document - Do not cite, quote, or distribute

et al., 2013a), EPA revised the acceptance criteria for the 30-minute data. Overall, the number of
valid 30 minute bLS NH3 values for lagoons increased and H2S decreased. The opposite
occurred for the corral site, TX5A, as the number of bLS measure estimates NH3 and H2S
decreased and increased, respectively. Appendix B summarizes the changes in data acceptance
criteria and the affects it had on the number of 30-minute values available for each site.

Literature (Grant et al., 2013a) also suggested bLS measurements could be adjusted to be
comparable to VRPM results. To prepare the 2012 NAEMS data sets of 30-minute values for use
in calculating daily averages, the bLS NH3 values for sites IN5A and WI5A were adjusted by
multiplying the emissions values by 1.19 (Grant & Boehm 2020) and 1.13 (Grant & Boehm
2020), respectively. After the adjustment, the bLS and VRPM data were used together to
determine which day had more than 24 half hour values to meet the revised 52% completeness
criteria days. In cases where 30-minute emissions flux values were available for both the bLS
model and VRPM, the average of the bLS and VRPM values were used. A practical example of
the calculation is provided in Appendix B. The Table B-23 presents an example calculation for
two days at site IN5A, (one day with both bLS and VRPM data, and one day with only bLS
data).

2.2 Comparison between the 2010 and Revised Barn Data Sets

The influence of the previous described corrections on the revised data sets can be
observed by comparing the summary statistics of all the valid emission values (at 75% data
completeness) between the 2010 dataset, as summarized in the final site reports, and the revised
data set. The following sections summarize the differences between the 2010 data set and revised
data set for each of the barn types for a set of standard summary statics (e.g., mean, standard
deviation, count (N), minimum, maximum, and number less than 0 (N<0)) of the average daily
emissions. For summary tables presented, the percent difference was calculated as the revised
data set minus the 2010 version of the data set, divided by the 2010 version of the data set (e.g.,
% Diff = (Revised - Data2oio)/Data2oio * 100). This calculation yields negative values when
decreases were seen in the revised version of the dataset.

2.2.1 Mechanically Ventilated Barns

In general, the 2010 and revised data set vary less than 10% for the barns at IN2B for
NH3 (Table 2-1) and H2S (Table 2-2), while the data sets for the PM size fractions (Table 2-3)
were not changed. The exceptions are the increase in the number of H2S values less than zero
(N<0) at IN2B (Table 2-2). There was more of a difference in the data sets for NY5B,
particularly with the minimum value of H2S (Table 2-2), which was revised from a very large
negative value (-226 g/d) to a small positive value (34.05 g/d). NY5B was the only site that had
changes to the particulate matter data set (Table 2-3), most notable of which was a decrease in

2-2


-------
Deliberative, draft document - Do not cite, quote, or distribute

the number of negative values for PMio. The WI5B saw some of the biggest differences in NH3
data, largely due to the increase in the number of valid average daily means (ADM) available for
NH3 after the revisions. The WI5B data sets for PM10, PM2.5, and TSP were unchanged.

Table 2-1. Percent difference in NH3 summary statistics between the 2010 and
revised dataset (at 75% data completeness).

Parameter IN5B B1 IN5B B2 NY5B B1 WI5B B1 WI5B B2

Mean

3%

3%

6%

-4%

-3%

Standard Deviation

5%

5%

5%

-11%

-3%

N

0%

0%

-12%

19%

20%

Minimum

-6%

-6%

-1%

25%

-26%

Maximum

4%

9%

7%

-2%

-2%

N<0

0%

0%

0%

0%

0%

Table 2-2. Percent difference in H2S summary statistics between the 2010 and
revised dataset (at 75% data completeness).

Parameter IN5B B1 IN5B B2 NY5B B1

WI5B B1 1 WI5B B2

Mean

1%

-2%

10%

0%

0%

Standard Deviation

0%

1%

3%

0%

-3%

N

2%

4%

-12%

-3%

-2%

Minimum

2%

2%

764%

0%

0%

Maximum

-2%

8%

-3%

4%

-5%

N<0

47%

67%

0%

33%

-88%

Table 2-3. Percent difference in PM summary statistics between the 2010 and
revised dataset (at 75% data completeness).



NY5B Bl,

NY5B Bl,

NY5B Bl,

IN5B,

WI5B,

Parameter

PMj.o

PM2.5

TSP

PM

PM

Mean

5%

2%

2%

No difference

No difference

Standard Deviation

5%

1%

0%

No difference

No difference

N

0%

2%

0%

No difference

No difference

Minimum

0%

0%

0%

No difference

No difference

Maximum

7%

1%

1%

No difference

No difference

N<0

-50%

13%

0%

No difference

No difference

2.2.2 Naturally Ventilated Barns

For the naturally ventilated barns, there were no changes in the CA5B datasets for any
pollutant and no changes in the WA5B datasets for NH3, H2S, or PM2.5. For PM10 (Table 2-4),
both WA5B barns saw an increase in the number of valid ADM, including new maximums more
than 50% larger than in the 2010 data set. The TSP data set (Table 2-5) also changed, most
notably there was an 18% decrease in the number of valid ADM at both barns and an increase in
the minimum value for barn 2.

2-3


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 2-4. Percent difference in PM10 summary statistics between the 2010 and
revised dataset (at 75% data completeness).

Parameter

CA5B B1

CA5B B2

WA5B B1

WA5B B2

Mean

No difference

No difference

20%

12%

Standard Deviation

No difference

No difference

63%

38%

N

No difference

No difference

1%

1%

Minimum

No difference

No difference

0%

0%

Maximum

No difference

No difference

83%

68%

N<0

No difference

No difference

0%

0%

Table 2-5. Percent difference in TSP summary statistics between the 2010 and
revised dataset (at 75% data completeness).

Parameter

CA5B B1

CA5B B2

WA5B B1

WA5B B2

Mean

No difference

No difference

3%

1%

Standard Deviation

No difference

No difference

5%

6%

N

No difference

No difference

-18%

-18%

Minimum

No difference

No difference

522%

0%

Maximum

No difference

No difference

0%

0%

N<0

No difference

No difference

0%

0%

2.2.3 Milking Centers

For the IN5B MC, most changes were minor for NH3 (Table 2-6) and H2S (Table 2-7).
The most notable change is the increase in the number of negative ADM for both gaseous
pollutants due to the changes in emission calculation. There were no measurements of PM10,
PM2.5 or TSP made at the IN5B milking center.

The NY5B MC had minor changes to the NH3 dataset and mostly minor changes to the
H2S data set. One of the largest changes was an increase in the minimum value for H2S (Table
2-7), which was the result of the removal of a large negative ADM. The data sets for the PM size
fractions (Table 2-8) generally saw minor changes. The notable exception is the 33% decrease in
the number of negative values for ADM. This statistic is a little misleading, as there were only
four values, and one of which was dropped during the revision.

2-4


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 2-6. Percent difference in NH3 summary statistics between the 2010 and
revised dataset (at 75% data completeness).

Paramptpr

IMCR

NY5B

Mean

7%

0%

Standard Deviation

8%

0%

N

0%

-7%

Minimum

0%

15%

Maximum

4%

-2%

N<0

8%

0%

Table 2-7. Percent difference in H2S summary statistics between the 2010 and
revised dataset (at 75% data completeness).

Parameter

IN5B

NY5B

Mean

2%

-2%

Standard Deviation

-4%

0%

N

1%

1%

Minimum

0%

764%

Maximum

-12%

-2%

N<0

39%

0%

Table 2-8. Percent difference in NY5B MC PM summary statistics between the
2010 and revised dataset (at 75% data completeness).

Parameter

PMio

PM-..

TSP

Mean

-1%

2%

1%

Standard Deviation

11%

1%

0%

N

8%

0%

0%

Minimum

0%

11%

0%

Maximum

0%

1%

1%

N<0

-33%

0%

0%

2.3 Data Completeness Criteria for the Revised Data Set

The appropriate data completeness criteria to use in a study depends on the size of the
dataset and the accuracy needed. A study by Grant et al. (2013b), in which NH3 emissions were
modeled from swine lagoons based on NAEMS data, investigated data completeness and
associated accuracy. The swine lagoon NH3 emissions dataset had limited data availability at a
data completeness of 75%. Grant et al. (2013b) explored how much the data completeness
criteria could be relaxed but still result in data with acceptable error. The study suggested an
error of ±25% to be acceptable and determined that a daily data completeness of 52% (or 25 out
of 48 30-minute periods) gave less than ±25% error (see Figure 2-1). Using this revised daily
completeness criteria resulted in a substantial increase in the size of the dataset.

Based on Figure 2-1 from the Grant et al. (2013b) study, it can be observed that a daily
completeness criterion of 75% (36 out of 48 30-minute periods) would give an error of

2-5


-------
Deliberative, draft document - Do not cite, quote, or distribute

approximately 10%. If it is assumed that the relationship between data completeness and error
from the Grant et al. (2013b) study is representative of other NAEMS datasets, the effect of
relaxed data completeness criteria can be investigated for other NAEMS sources.

The NAEMS PI provided EPA with additional analysis that examined the effect of
different completeness criteria by comparing the number of valid ADM. EPA reviewed these
data for the barn data site and retained the 75% completeness criterion. For the open source sites,
EPA review found that adjusting the daily data completeness to 52% provided significantly more
data and justified the increase in the error. The full analysis can be found in Appendix C.

Number of measurements	Number of measurements

Figure 2-1. Ratio of mean predicted emissions for portion of day with valid emissions
measurements to mean predicted emissions for the complete day at the finishing (A) and sow (B)
farm. Error plotted against number of valid 30-minute measurements (from Grant et al., 2013b).

2.4 Comparison Between the Revised Data Sets and NAEMS Datasets Used in
Peer-reviewed Published Papers

Where possible, EPA compared the revised dataset developed for this report to values
presented in peer reviewed journals and reports to quantify any differences due to the application
of the revised calculation methods and other adjustments discussed in Section 2.1. Summaries of
the gaseous emissions from naturally ventilated barns can be found in Joo et al. (2015). Lagoon
and basin summaries have been presented in Grant and Boehm (2015), and corrals in Grant et al.
(2020). Summaries of the mechanically ventilated barn data and particulate matter data could not
be found at the time of writing.

A simple comparison of the summary statistics presented in these papers and the
summary statistics of the revised dataset is presented in the following sections. Overall, the
dataset used for model development and presented in the papers are different due to difference in
data screening methods. For NH3 and H2S at naturally ventilated barns, the model development
dataset contains at least twice the number of observations than used in the article due to different
choices in processing the data. Similarly, the revisions to the acceptance criteria for open sources

2-6


-------
Deliberative, draft document - Do not cite, quote, or distribute

noted in Section 2.1 also resulted in difference in differences between the published data set and
the modeling data set. For the open sources, the acceptance criteria used by EPA are the
culmination of several published papers aiming to improve the data quality and go beyond what
was discussed in the compared work. Overall, the comparison highlights that EPA has done
extensive analysis and review of the dairy data sets to obtain a robust data set for model
development.

2.4.1 Naturally Ventilated Barns

Despite no difference between NH3 and H2S in the revised data set and the submitted
2010 data set (Section 2.2.2) for WA5B, the published data has different maximum, minimum,
and average values for both (Table 2-9 and Table 2-10). A closer examination of Joo et al.
(2015) reveals a more extensive outlier removal process, whereby anything outside 1.5 times the
interquartile range were designated as outliers. The article also only reports on data collected in
the second year of the study (November 2008 to October 2009) since there were "more and
longer trouble-free periods" (Joo et al., 2015). The article further truncates the data by focusing
on one-week data sets of continuously collected measurements selected every two months, for a
total of 7 weeks (49 days) of data. The model data set contains at least twice as many days as the
published data set, which quickly explains the differences seen.

Table 2-9. Comparison of naturally ventilated NH3 emissions in the model dataset

to published datasets.

Site

Units

Statistic

Model
Dataset

Published
Studies

Study

WA5B B2

Emissions
(kg day1)

Mean

26.6

14.1

Joo et al. 2015

Minimum

-156.4

10.8

Maximum

96.6

19.7

WA5B B4

Emissions
(kg day 1)

Mean

54.7

19.4

Joo et al. 2015

Minimum

9.0

17.2

Max

170.9

21.2

Table 2-10. Comparison of naturally ventilated H2S emissions in the model

dataset to published datasets.

Site

Units

Statistic

Model
Dataset

Published
Studies

Study

WA5B B2

Emissions
(gday1)

Mean

555.6

397.4

Joo et al. 2015

Minimum

-5,400.9

123.5

Maximum

6,513.6

542.4

WA5B B4

Emissions
(gday1)

Mean

1,130.9

627.7

Joo et al. 2015

Minimum

-11,640.1

0.0

Max

17,960.3

1711.8

2-7


-------
Deliberative, draft document - Do not cite, quote, or distribute

2.4.2 Open sources

Section 2.1 and Appendix B outline how EPA altered the acceptance criteria for the open
sources. The changes were culled from several peer reviewed journal articles (Grant & Boehm
2020, Grant et al., 2020, Grant & Boehm 2015, Grant et al., 2013a) published since the 2010
receipt of the NAEMS data. While each of the articles referenced typically focus on one site,
EPA developed a list of revisions to be applied to each site that represent the state of the science
for the method. As such, the lagoon NH3 values (Table 2-11) differ from the values published in
Grant & Boehm (2020) due to difference in the acceptance criteria.

Table 2-11. Comparison of lagoon and basin NH3 emissions in the model dataset

to published datasets.

Site

Units

Statistic

Model
Dataset

Published
Studies

Study

IN5A

Emissions
(g s1)

Mean

0.23

0.27

Grant &
Boehm
2020

Minimum

-0.14

0.17

Maximum

1.07

0.39

WI5A

Emissions
(gs1)

Mean

0.07

0.22

Grant &
Boehm
2020

Minimum

-0.04

0.07

Maximum

0.91

0.42

Similarly, NH3 emissions from dairy corrals varied from the published work due to
revisions to the acceptance criteria that EPA implemented. These revisions resulted in 6
additional daily average emission values from the Grant publication (Table 2-12). These
additional days shift the average of the daily means higher than in the published work and
increased the variability, as shown by the increase in the standard deviation. As noted previously,
the acceptance criteria used by EPA are an attempt to apply the revisions from several published
papers aiming to improve the data quality and go beyond what was discussed in the compared
work. Overall, the comparison highlights that EPA has done extensive analysis and review of the
dairy sets to obtain a robust data set for model development.

Table 2-12. Comparison of corral (TX5A) NH3 emissions in the model dataset to

published datasets.

s

		:

N

Mean (k^d )

Standard Deviation

Revised

73

755.0

317.5

Grant et al. 2020

67

287.6

144.7

2-8


-------
Deliberative, draft document - Do not cite, quote, or distribute

3 RELATIONSHIPS ESTABLISHED IN LITERATURE

Developing EEMs for dairy AFOs is complex as many variables potentially influence
emissions. Therefore, to be efficient as possible in this study, a focused approach was used. The
focused approach involved developing models based on variables that could potentially have a
major influence on air emissions. This assessment was made based on theoretical considerations
and observations reported by previous studies that have investigated the influence of variables on
emissions from dairy AFOs.

3.1 NH3 and H2S from Confinement Sources

Emissions from barns originate from the nitrogen and sulfur content in urine and manure
deposited in pits or on the floor along with any bedding material present in the barn. The amount
of NH3 and H2S emitted depend on the amount of manure produced and its characteristics, that is
the total ammoniacal nitrogen (TAN) and sulfur content, (Sanchis, Calvet, del Prado, and
Estelles (2019)). Multiple factors influence the generation and release of NH3 and H2S
emissions, such as the type of building and its volume, flooring type, housing density, manure
management, livestock management practices, milk yield, diet, animal behavior, and factors
affecting the microclimate within the buildings (e.g., temperature, humidity, airflow) (Bjerg et
al., 2013, lioiiuonin el a I 2') I (\ Herbut and Angrecka 2014). The following section outlines the
relationship between these specific parameters and emission rates, as well as whether the
parameter, or suitable proxy, is available in the NAEMS data set.

Manure volume is a key factor influencing NH3 and H2S emissions in both mechanically
ventilated and naturally ventilated barns. That is, the more manure and urine there is, the more
precursor material there is for NH3 and H2S emissions. No estimates or measurements on the
amount of manure generated were taken at any of the dairy sites. However, other parameters,
such as inventory and live animal weight (LAW), can be used as proxies for fresh waste
generation as more or larger animals would produce more waste. Both inventory and LAW were
determined daily at each site and were selected for further investigation.

Second to volume, the compositional characteristics—that is nitrogen, ammonia, and
sulfur content of the waste—provides information on the amount of NH3 and H2S than can form
and be emitted by the barn. As noted in Section 2.3, sampling for total ammoniacal nitrogen
content (TAN), total Kjeldahl nitrogen (TKN), and sulfur content occurred for various
components of the barn, including bedding material and the waste collected from the floor.
However, a limited number of samples were taken over the course of the study. Including them
in the regression analysis would limit the number of days available for model development, and
thereby the variability of other factors included in the model. EPA has looked at interpolating the
data between samplings to extend the data to more days, however, this does require assumptions

3-1


-------
Deliberative, draft document - Do not cite, quote, or distribute

about the behavior of nitrogen and sulfur content in the manure between samples. Knowing the
incoming nitrogen and sulfur content of the feed, water, and bedding would inform the
interpolation process, leading to better assumptions as this would indicate the maximum amount
of nitrogen and sulfur introduced into the system, allowing from mass balance checks. However,
data on feed and water and was not provided by the producers. As such, the limited data
available on waste characteristics (i.e., TAN, TKN, sulfur content) were excluded from the
model development dataset.

Manure pH has a strong correlation with both NH3 and H2S emissions (Rotz et al. 2014,
Montes et al., 2009). The ammonia fraction of TAN is partly a function of pH, so pH would
provide an indication of NH3 available in the manure (Montes et al., 2009). For H2S, water with
an acidic pH has an increased concentration of molecular hydrogen sulfide, which increases the
potential for H2S emissions. However, like TAN and TKN measurements, only limited pH data
were collected during NAEMS. As such, the limited data available were excluded from the
model development dataset.

The Sanchis et al. (2019) review overwhelmingly found air temperature in the barn had a
positive relationship with NH3 emissions for both mechanically and naturally ventilated barns.
The higher temperatures increase NH3 losses by decreasing the solubility of NH3 and increasing
the proportion of TAN as NH3 gas (Meisinger and Jokela, 2000). For a similar reason, manure
temperature is highly correlated to NH3 emissions. NAEMS collected barn exhaust temperature
and ambient temperature at all sites and these factors were selected for further investigation.
Ambient temperature was chosen for further investigation, as it is related to barn conditions and
would provide an alternative barn based temperature monitoring for operators.

The studies cited by Sanchis et al. (2019) found, in some cases, the relationship between
temperature was affected by the floor type (e.g., slatted versus solid) and manure handling
system. EPA investigated the type of manure management system (i.e., flush or scrape) for the
mechanical barns for further analysis. A similar analysis was not included for the naturally
ventilated barns, as both sites used flush systems. Bedding type was also considered, however
the study data only indicated in general the type of bedding used in the barns. In the case of
CA5B, the operator used several bedding types as they were available (Zhao et al., 2010) with no
reliable indication of when those changes occur or what the percentage of each bedding type was
on any given day.

Schmithausen et al. (2018) also noted permanent under floor storage of slurry potentially
contributed to higher NH3 emissions. The site description of two mechanically ventilated sites,
IN5B and NY5B, suggest that they utilize a reception pit to hold scraped material as part of their
manure management system. While the NY5B notes the deep reception pit is in the connecting

3-2


-------
Deliberative, draft document - Do not cite, quote, or distribute

alley between the freestall barn and milking center, the location of the pit at IN5B was not
documented. It was noted that the material in the reception pits, at both sites, were transferred to
a digester on a regular basis. Because the material was transferred on a regular basis and was not
long term, a variable to account for under floor storage was not included at this time.

The ventilation rate of mechanically ventilated barns has been shown as having a positive
correlation toNH3 emissions across several studies (Kavolelis, 2003; Philippe, et al., 2011;

Samer et al., 2012). Ventilation rates are typically driven by the temperature inside the barn,
which is affected by the outside temperature. For modeling purposes, this suggests that
temperature, either barn or ambient, might make a good proxy for ventilation rate.

For naturally ventilated barns, the ventilation or air flow through the barn is driven by the
wind. Many studies (Arogo et al., 1999; Bjerg et al. 2013, Wu et al. 2012; Schrade et al., 2012;
and Herbut and Angrecka, 2014) have found a strong correlation between emissions and wind
speed, and occasionally wind direction (I'cidk-r and \ Killer (2<)| I)). However, Saha et al. (2014)
did not find the clear relationship between wind speed and emissions. Saha et al. (2014)
suggested that the effects of wind speed might be masked by other environmental parameters,
such as temperature and relative humidity, or the presence of other buildings and slurry tanks
that might influence wind entering the building. Bjerg et al. (2013) noted that the more important
component to release was air velocity over the manure, which is not necessarily correlated to
wind speed in the barn, as air movement could be affected by numerous things, such as animals
and other obstructions in the barn. For modeling purposes, wind speed was selected for further
study for naturally ventilated barns.

The literature review did not find references showing a correlation between either NH3 or
FhS emission in mechanically ventilated barns and relative humidity. Sanchis et al. (2019)
suggests that there are no significant effects due to the high variability of relative humidity in the
barn environment. However, Sanchis et al. (2019) noted studies of naturally ventilated barns
showed that higher relative air humidity leads to reduced NH3 emission rates. In general, higher
air humidity values are expected to yield reduced NH3 concentrations, since NH3 is highly water-
soluble and would be absorbed by the water vapor in the air and less gaseous NH3 would be
measured. However, this is only true within a certain temperature range and the management
strategies would also affect this relationship. Saha et al. (2014) also noted the effect of relative
humidity might be related to the changes in animal activity and performance in response to heat
stress. Because of the potential relationship between NH3 and moisture, relative humidity was
selected for further study for both mechanically and naturally ventilated barns.

3-3


-------
Deliberative, draft document - Do not cite, quote, or distribute

Animal and management activities, such as feeding and milking, can affect emission rates
(Ngwabie, et al. 2011, Hempel 2016). There was no specific daily information on management
activities recorded by NAEMS.

3.2 Particulate Matter from Barns

The release of PMio, TSP, and PM2.5 (collectively referred to as PM) into the air of dairy
barns is caused by the physical suspension of a range of different materials in the barns including
feed, manure, bedding, and skin or hair (Cambra-Lopez et al., 2011). Accordingly, the EPA
chose live animal weight and inventory as predictor variables, as they are related to the amount
of source material. One study, Garcia et al. (2013), found an inverse relationship between
milking center capacity and PM2.5 concentration on the farm, which was attributed to the larger
dairies being newer and more efficiently operated. This suggests there are different management
practices at newer barn that can affect particulate emissions. Likely making the use of inventory
more nuanced than with other animal types.

Physical suspension of PM from barn surfaces can be caused by air flow, animal activity,
and human activity (Aarnink and Ellen, 2007); however, EPA did not receive barn activity
measurements and could not explore the influence of this variable further. Airflow, or ventilation
rate, was recorded for all barn sources. As mentioned in the previous section, mechanical
ventilation rates are related to ambient and barn temperature, thus meaning that temperature
could be a potential surrogate variable that represents airflow. For naturally ventilated buildings
wind speeds may have an influence on the air flow, which in turn could potentially affect the PM
emissions from the buildings. Accordingly, EPA selected the airflow for further review, as well
as wind speed from naturally ventilated barns. Temperature was selected for both mechanically
ventilated barns, due to the correlation with airflow, and naturally ventilated barns. While Takai
et al. (1998) did not find seasonal variation with PM emission from naturally ventilated barns,
Mostafa et al. (2016) did see greater emissions in summer and lower values in winter. The longer
observation periods of PM during NAEMS showed some seasonality, with the highest values
occurring in the summer.

Physical suspension may also be influenced by moisture conditions and relative humidity
(Cambra-Lopez et al., 2010). A study by Takai et al. (1998) examined PM emissions from a
variety of livestock types including dairy cattle and reported that relative humidity greater than
70% contributed to particles aggregating together and thus reducing emissions. Accordingly, for
dairy barns, the variables ambient relative humidity and barn relative humidity were selected for
further investigation.

3-4


-------
Deliberative, draft document - Do not cite, quote, or distribute

3.3 NH3 and H2S for Open Sources

The release of NH3 and H2S from open sources follows similar mechanics as release from
waste in the barns. That is, the amount of NH3 and H2S emitted will depend on some of the same
factors as the barn, such as the compositional characteristics. With lagoons and basins, the
amount of waste can be characterized by the lagoon surface area in addition to farm level
inventory and live animal weight. For open source model development, EPA used lagoon surface
area to normalize emissions, as it represents the amount of the manure that can exchange gas
with the atmosphere. For corrals, the area of the corrals was selected along with the inventory for
the farm since the emissions measurements covered a wider area. As with barn sources, TAN,
TKN, and sulfide content of the manure has a major influence on dairy open source NH3 and FhS
emissions (see section 3.1 for details). For NAEMS open source sites, there were no
measurements of TAN, TKN, or sulfide at the three sites. As a result, EPA could not investigate
these parameters further.

Like barn sources, NH3 and FhS emissions are a function of the pH, specifically the pH at
the surface of the manure, and temperature as both parameters affect the chemistry associated
with the generation and release of the pollutants (Arogo et al., 2006, Rotz et al., 2014). Ambient
temperature, along with turbulence, typically represented by wind speed, affect the diffusion and
dispersion of the released gases from the lagoon surface (Arogo et al., 2006, Sommer et al.,
2013). There were continuous measurements of lagoon temperature, lagoon pH for lagoon/basin
sites, and air temperature and wind speed for all NAEMS open sources. Accordingly, these four
variables were selected for further analysis for lagoon/basin sources and air temperature and
wind speed were selected for corral sources.

Like manure in barns, moisture levels can affect the volatilization of NFb and FhS. In
drier environments, evaporation and volatilization are going to occur more rapidly. In a lagoon,
where waste is held as a slurry, it is likely less of a factor than in a corral where manure is often
mixed into the soil creating a drier environment. Grant et al. (2020) suggested that the vapor
pressure deficit might be a more compelling parameter than relative humidity to represent the
potential for volatilization from the manure and soil mixture present in corrals. The vapor
pressure deficit is the difference between how much moisture the air can hold when saturated
and the actual amount of moisture in the air. Unlike relative humidity, the vapor pressure deficit
is not a function of temperature, which also allows for a more consistent comparison between
days. EPA chose to include both relative humidity and vapor pressure deficit to further
investigate their relationship with emissions from the corral.

The presence of a crust or cover on a lagoon or basin will inhibit the transfer of NH3 to
the atmosphere, reducing emissions. Similarly, frozen lagoon surfaces will also stop emissions

3-5


-------
Deliberative, draft document - Do not cite, quote, or distribute

from the surface of the lagoon. The NAEMS made limited observation of the state of the lagoon
(e.g., color, crust) during the study. The lack of daily observations would limit the number of
days available for EEMs development, as the dataset would be limited to only those days with
lagoon surface observations. Due to the limited nature of the observations available, this variable
was not explored further.

3-6


-------
Deliberative, draft document - Do not cite, quote, or distribute

4 SITE COMPARISON, TRENDS, AND ANALYSIS

Before developing the EEMs, EPA evaluated NAEMS data for each pollutant to identify
patterns and trends in the emissions data using a combination of summary statistics (mean,
standard deviation, number of data values, median, minimum, maximum, coefficient of
variation, and number of data values less than zero) and time series plots. Section 4.1
summarizes the emissions trends from the sites, while Appendix D contains the tables of
summary statistics. Appendix E presents the time series plots of the site-specific emissions,
environmental and production parameters, and manure data collected under NAEMS.

Based on the analysis described in Section 3.0, EPA identified the key environmental and
manure parameters that potentially affect emissions from dairy barns and associated open
sources. Parameters of particular interest included inventory, barn conditions (exhaust
temperature, exhaust relative humidity, and airflow), ambient temperature, ambient relative
humidity, and wind speed.

The next step of the analysis was to look at the key environmental and manure
parameters compared to emissions trends. The exploratory data analysis was conducted to
confirm that the variables were selected based on the following criteria: (1) data analysis in this
study and in the literature suggested that these variables had an influence on emissions; (2) the
variables should be easy to measure; and (3) the variables were already in the daily average
NAEMS data and were available for most days of monitored emissions. This third selection
criterion particularly applies to the manure parameters, such as moisture content and TAN
concentration, which were infrequent due to the intensive collection and analysis methods.
Additional time could be taken to develop an appropriate methodology for interpolating between
the few data points available for these parameters in the dataset. However, these parameters are
difficult to acquire as they require chemical analysis from a laboratory.

The exploratory data analysis was also used to explore whether additional parameters
could be included to explain trends. To further explore the trends between the predictor variables
and emissions and determine whether the parameter should be included in developing an EEM,
EPA prepared scatter plots of emissions versus the process, environmental, and manure
parameters and conducted least squares regression analysis to assess the influence of each
variable on emissions. For the regressions, EPA classified the linear relationships based on the
ranges in Table 4-1.

A summary of this analysis for environmental parameters is discussed in Section 4.2.
Again, Appendix D contains summary statistics, Appendix E contains the relevant time series
plots, and Appendix F contains least squares regression analyses between the identified
parameters and emissions.

4-1


-------
Deliberative, draft document - Do not cite, quote, or distribute
Table 4-1: Relationship classification based on R2 values

Range of R' Relationship strength

R2< 0.001

none

0.001 < R2< 0.2

slight or weak

0.2 < R2 < 0.4

modest

0.4 < R2 < 0.6

moderate

0.6 < R2 < 0.8

moderately strong

R2 > 0.8

strong

4.1 Mechanically Ventilated Dairy Barns (IN5B-B1, IN5B-B2, NY5-B1, WI5B-B1
and WI5B-B2)

4.1.1 Emissions data

Appendix D, Table D-l and D-2 presents the summary statistics for daily average
emissions of NH3 for the mechanically ventilated sites in kilograms per day and grams per day
per head (kg d"1 and g d"1 hd"1), respectively. Based on Table D-l, the emissions appear to vary
across sites. However, when presented in a per head basis, as in Appendix D, Table D-2, the
emissions are consistent across sites with average daily emissions ranging from 31.35 kg d"1 hd"1
at WTB5-B2 to 48.28 kg d"1 hd"1 at IN5B-B1. Appendix E, Figure E-l showed that the emissions
follow a seasonal cycle, with greater emissions typically occurring in the summer and decreasing
to lows in winter months. Emissions from the WI5B site have a more muted seasonal cycle on
the first year, with slightly increased values in the second year of the study. This appears to
correlate to a changing from a flush system to a scrape system in September of 2008. As noted in
Section 3, manure management systems can affect the emissions generated in the barn. Appendix
E, Figure E-l suggests it is worth pursuing modeling options that account for the manure
management system.

The summary statistics for daily average FhS emissions are presented in Appendix D,
Table D-3 and D-4 for g d"1 and mg d"1 hd"1, respectively. Unlike NH3, the per head values in
Table D-4 show emission values 2 to 4 times greater at the WI5B barns than the other sites.
Appendix E, Figure E-2 showed the time series plot for FhS emissions. The plot showed a
seasonal trend in FhS emissions for the IN5B and NY5B site, with emissions trending higher in
warmer months. However, the WI5B barns show a very different trend. The H2S emission for
both barns are quite high and variable for the first half of the plot, and then fall to lower levels.
Like the shift with the NH3 emissions, this change corresponds to the switch to a scrape system
in the barns.

Appendix D, Table D-5 and D-6 presents the summary statistics in g d"1 and mg d"1 hd"1,
respectively, for the daily average emissions of PM10 for the mechanically ventilated sites. There
was variation in emissions between sites, both in the total for the day and when normalized on a

4-2


-------
Deliberative, draft document - Do not cite, quote, or distribute

per head basis. The average daily emissions ranged from 9.73 g d"1 (12.49 mg d"1 hd"1) at INB5-
B1 to 562.91 g d"1 (1,571.90 mg d"1 hd"1) at WI5B-B2. The time series plot (Appendix E, E-3)
showed readings hovering between 0 and 500 g d"1, with greater spikes typically occurring in the
summer months. WI5B does experience maximum values that are twice as high as the other
sites. These peaks occur both in the summer of 2008 and 2009, suggesting the change to a scrape
manure management system did not contribute to the highest emission days. The dataset used for
the exploratory data analysis has several negative values, which were further reviewed during the
data review process described in Section 2.

Like, PMio, the PM2.5 average daily emissions vary substantially across sites. The
average daily emissions summarized in Appendix D, Table D-7, indicate that WI5B emissions
are much greater than the other barns. The emissions across all sites range from 21.18 g d"1 at
IN5B-B1 up to 186.75 g d"1 at WI5B-B2. When accounting for inventory difference (Appendix
D, Table D-8), the WI5B are still more than twice any other mechanically ventilated barn
monitored during NAEMS, with an average value of 662.17 mg d"1 hd"1 at WI5B-B1 compared
to 25.89 mg d"1 hd"1 at IN5B-B1. Appendix E, Figure E-4 showed the temporal variability of the
PM2.5 emissions. The plot for IN5B does show some rather large negative numbers in the
exploratory data analysis, which were further reviewed during the data set review process
described in Section 2. The inclusion of these points is likely reason for the lower average values
at IN5B compared to the other sites. The sparse temporal nature of the daily PM2.5 values, due to
a rotating monitoring schedule for the PM size fractions at the NAEMS sites, makes it hard to
determine if there is a seasonal trend to the data. The number of negative daily averages from the
sites varied greatly. The barns at IN2B had the least negative values with 28 and 29 at B1 and
B2, respectively. The remaining sites had nearly twice as many negative values; NY5B-B1 had
53, while WI5B had 53 and 45 at B1 and B2, respectively.

The daily average TSP emissions followed a similar trend to PM10 and PM2.5. That is
WI5B had average emissions substantially greater than the other two sites (Appendix D, Table
D-9), even after accounting for difference in inventory levels (Appendix D, Table D-10). Like
PM2.5, the sparse temporal nature of the daily TSP values makes it hard to determine if there is a
seasonal trend to the data. The plot of WI5B does suggest some seasonality, with slightly greater
emissions in the summer. However, a similar pattern is not obvious at the other sites. There were
fewer negative daily TSP values, with all sites reporting less than 10 negative values.

4.1.2 Environmental data

The statistical summary of the environmental parameters associated with mechanically
ventilated barns are presented in Appendix D, Table D-l 1. The inventory was varied across the
sites, ranging from an average of 211 head at WI5B-B1 to 864 head at IN5B-B2. Appendix E,

4-3


-------
Deliberative, draft document - Do not cite, quote, or distribute

Figure E-6 showed that the number of cows present over the course of NAEMS was consistent,
with any one barn varying by less than 112 cows over the study duration. Of note, the first-year
inventory data from WI5B appears to be based on average inventory of the barn and not actual
inventory levels. Appendix F, Figures F-l through F-5 show the scatter plots of inventory versus
each pollutant. A summary of the findings is provided in Table 4-2. In general, there is a weak
relationship with inventory across all pollutants, except that NH3 has a moderate positive
relationship. Of note, all the PM size fractions show a weak negative linear relationship with
inventory, as the smaller barns have greater emissions. Further investigation showed the barns
with greater inventory are newer, which is consistent with the finding from the literature review
that newer barns had lower PM emissions. As noted in Section 3.2, the difference between the
newer facilities is likely a management practice applied in the newer construction. It is currently
unknown what leads to the decrease in emissions for larger newer farms. A possibility to
somehow account for this unknown factor is to consider the age of the facility in modeling;
however, the limited range in ages (Table 4-1) makes it difficult to incorporate at this time. EPA
will continue to pursue identifying the physical or chemical property driving this decrease in Pm
emissions in newer barns, and a way to incorporate this into the modeling.

Table 4-2. Year mechanically ventilated barns were constructed

Barn Year Constructed

WI5B B1

1990

WI5B B2

1994

NY5B B1

1998

IN5B B1

2004

IN5B B2

2004

Average animal weight for the IN5B and WI5B barns were reported as a constant value.
For NY5B, the daily value reported only vary by less than 5 kg (576 to 580 kg). This limited
range of daily average animal weight is apparent in the time series (Appendix E, Figure E-7).
The regression analyses in Appendix F, Figures F-6 through F-10, summarized in Table 4-2,
showed only a slight or weak relationship between average animal weight and each pollutant.
Trends in live animal weight (i.e., inventory * average animal weight) do not vary dramatically
over the monitoring period (Appendix E, Figure E-8). The regression analyses in Appendix F,
Figures F-l 1 through F-l5 showed similar relationships as inventory, which is the most variable
component of live animal weight.

Exhaust temperatures were comparable across all the sites, ranging from an average of
10.55°C at WI5B-B2 to 12.89°C at NY5B-B1. The time series in Appendix E, Figure E-9 show
the typical seasonal trend, where temperatures peak in the summer, decrease to minimums
around the new year, and then trend upwards during the spring. The linear regression analyses

4-4


-------
Deliberative, draft document - Do not cite, quote, or distribute

(Appendix F, Figures F-16 through F-20) only shows a weak to modest positive relationship to
temperature. However, the figure for IN5B suggests a nonlinear relationship with temperature,
which might be reducing the overall strength of the correlation. The shift in manure management
system at WI5B affected the strength of the relationship for those barns. For example, R2 reached
0.72 with NH3 emissions while the house was scrape and only 0.21 as scrape for NH3. A
summary of the findings is provided in Table 4-2.

A review of the exhaust relative humidity summary (Appendix D, Table D-l 1), were
comparable across all the sites, ranging from an average of 66.8% at WI5B-B2 to 75.4% at
NY5B-B1. The time series (Appendix E, Figure E-10) show the relative humidity is variable, as
there is a spread in the data for any time of the year. The plots suggest dips in humidity for the
spring, with IN5B also suggesting a dip in the fall. When regressed with the emissions (Figures
F-21 through F-25), there are only slight or weak relationships, which are positive for gaseous
pollutants and negative with particulate matter daily emissions (kg/d).

The measured airflow through the barn was comparable across sites and ranged from 131.
dry standard cubic meter per second (dsmV1) at WI5B-B1 to 210. dsmV1 at IN5B-B1. The time
series (Appendix E, Figure E-l 1) showed a seasonal pattern, as ventilation rates would increase
to maintain barn temperatures during warm months. The regression analyses (Appendix F,
Figures F-26 through F-30) showed weak to modest positive relationships with emissions, which
is supported by literature.

Table 4-3. Mechanically ventilated environmental parameter regression analyses

Pollutant

Parameter

R

Rz

Strength

Figure

nh3

Inventory

0.660

0.435

moderate

Appendix F, F-l

HzS

Inventory

0.002

< 0.001

slight or weak

Appendix F, F-2

PM10

Inventory

-0.292

0.085

slight or weak

Appendix F, F-3

PM2.5

Inventory

-0.319

0.102

slight or weak

Appendix F, F-4

TSP

Inventory

-0.327

0.107

slight or weak

Appendix F, F-5

NH3

Average animal weight

-0.423

0.179

slight or weak

Appendix F, F-6

h2s

Average animal weight

0.114

0.013

slight or weak

Appendix F, F-7

PM10

Average animal weight

0.240

0.058

slight or weak

Appendix F, F-8

PM2.5

Average animal weight

0.384

0.148

slight or weak

Appendix F, F-9

TSP

Average animal weight

0.384

0.147

slight or weak

Appendix F, F-10

NH3

Live animal weight

0.653

0.426

moderate

Appendix F, F-ll

HzS

Live animal weight

0.014

< 0.001

slight or weak

Appendix F, F-12

PM10

Live animal weight

-0.278

0.077

slight or weak

Appendix F, F-13

PM2.5

Live animal weight

-0.283

0.080

slight or weak

Appendix F, F-14

TSP

Live animal weight

-0.307

0.094

slight or weak

Appendix F, F-15

NH3

Exhaust temperature

0.493

0.243

modest

Appendix F, F-16

HzS

Exhaust temperature

0.323

0.104

slight or weak

Appendix F, F-17

PM10

Exhaust temperature

0.410

0.168

slight or weak

Appendix F, F-18

4-5


-------
Deliberative, draft document - Do not cite, quote, or distribute

Pollutant

Parameter

R

R2

Strength

Figure

PM2.5

Exhaust temperature

0.484

0.234

modest

Appendix F, F-19

TSP

Exhaust temperature

0.406

0.165

slight or weak

Appendix F, F-20

NH3

Exhaust relative humidity

0.390

0.152

slight or weak

Appendix F, F-21

HzS

Exhaust relative humidity

0.193

0.037

slight or weak

Appendix F, F-22

PM10

Exhaust relative humidity

-0.269

0.072

slight or weak

Appendix F, F-23

PM2.5

Exhaust relative humidity

-0.414

0.171

slight or weak

Appendix F, F-24

TSP

Exhaust relative humidity

-0.322

0.104

slight or weak

Appendix F, F-25

NH3

Airflow

0.536

0.287

modest

Appendix F, F-26

h2s

Airflow

0.232

0.054

slight or weak

Appendix F, F-27

PM10

Airflow

0.425

0.180

slight or weak

Appendix F, F-28

PM2.5

Airflow

0.449

0.202

modest

Appendix F, F-29

TSP

Airflow

0.376

0.141

slight or weak

Appendix F, F-30

4.1.3 Ambient Data

The statistical summary of the ambient parameters associated with mechanically
ventilated barns are presented in Appendix D, Table D-12. The average daily temperatures were
cooler at WI5B at 7.2°C, compared to 12.2°C at IN5B. The time series in Appendix E, Figure E-
12 show the typical seasonal pattern to temperatures (i.e., maximum in summer and minimums
in winter). Of note, data is missing starting in January 2008 at IN5B. No reason for the data loss
was provided in the final site report. With the inclusion of three sites, there are ample
measurements of emissions at the anticipated temperature range for model development. The
scatter plots of ambient temperature (Appendix F, Figures F-31- F-35), summarized in Table 4-3,
show weak-to-modest positive relationships with emissions. The NH3 plots (Appendix F, Figures
F-31) indicate emissions increased more rapidly with temperature at IN5B than the remaining
sites.

Ambient relative humidity is similar between sites, ranging from an average value of
67.8% at NY5B to 68.4% at WI5B. The time series (Appendix E, Figure E-13) show the values
vary by at least 20% for any given time of the year. Like the exhaust relative humidity, there is
an indication that minimum values are more likely in both spring and fall, though the scatter to
the data makes a seasonal pattern hard to discern. The regression analyses (Appendix F, Figures
F-36 - F-40) indicate slight or weak negative relationships between ambient relative humidity
and emissions, even when looking at sites individually.

Table 4-4. Mechanically ventilated ambient parameter regression analyses

Pollutant

Parameter

n

D2

Strength



nh3

Ambient temperature

0.537

0.289

modest

Appendix F, F-31

HzS

Ambient temperature

0.257

0.066

slight or weak

Appendix F, F-32

PM10

Ambient temperature

0.370

0.137

slight or weak

Appendix F, F-33

PM2.5

Ambient temperature

0.398

0.159

slight or weak

Appendix F, F-34

TSP

Ambient temperature

0.348

0.121

slight or weak

Appendix F, F-35

NH3

Ambient relative humidity

-0.110

0.012

slight or weak

Appendix F, F-36

4-6


-------
Deliberative, draft document - Do not cite, quote, or distribute

Pollutant

Parameter

R

R2

Strength

Figure

h2s

Ambient relative humidity

<0.001

<0.001

slight or weak

Appendix F, F-37

PM10

Ambient relative humidity

-0.129

0.017

slight or weak

Appendix F, F-38

PM2.5

Ambient relative humidity

-0.331

0.109

slight or weak

Appendix F, F-39

TSP

Ambient relative humidity

-0.155

0.024

slight or weak

Appendix F, F-40

4.2 Milking Centers (IN5B-MC and NY5B-MC)

4.2.1	Emissions Data

Appendix D, Table D-13 and Table D-14 presents the summary statistics, in kg d"1 and g
d"1 hd"1, for daily average emissions of NH3 for the MCs monitored during NAEMS. The total
emissions (kg d"1) are relatively similar between the barns, though IN5B has a larger standard
deviation. When scaled for the capacity of the MC (Appendix D, Table D-14), NY5B, at 30.3 g
d"1 hd"1, was nearly double the average emission of 15. 7 g d"1 hd_1at IN5B. The time series plot
of NH3 emissions (Appendix E, Figure E-14) showed some seasonality in the data. The plots for
IN5B suggest greater emissions in the warmer months, particularly in the summers of 2008 and
2009. The data at NY5B does not have as strong of a seasonal pattern as IN5B.

In a reversal of what was seen with the NH3 statistics, IN5B had greater overall H2S
emissions (Appendix D, Table D-15) than NY5B and greater scaled emissions (Appendix D,
Table D-16). Average emissions at IN5B were 1,207 g d"1 (2,148 mg d"1 hd"1) compared to 129g
d"1 (2,681 mg d"1 hd"1). The time series plot of H2S emissions (Appendix E, Figure E-15)
suggests some seasonality to the data, with higher readings in the summer months, which may be
related to ventilation rates, and indirectly related to ambient temperature. The peaks at IN5B
were much greater than NY5B, suggesting an additional difference in the site. Further review
showed that IN5B used a flush system and NY5B used a scrape system for manure removal.

Like the emission shift seen at WI5B, it is possible that the manure management system is
influencing the emission levels.

Particulate matter emissions observations were only taken atNY5B. Appendix, Table D-
17 provides the statistical summary in g d"1 and Appendix D, Table D-18 provide them in mg d"1
hd"1. Appendix E, Figure E-16 shows the time series of PM10 emission estimates. The plot
suggests some seasonality to the data, with higher readings in the summer months, which may
relate to ventilation rates. The time series of PM2.5 emission is in Appendix E, Figure E-17, while
Appendix E, Figure E-18 showed the time series for TSP. The sparse nature of the PM2.5 and
TSP data makes it hard to determine if there is any seasonality to the data.

4.2.2	Environmental data

The statistical summary of the environmental parameters associated with MCs is
presented in Appendix D, Table D-19. Daily inventory number were not reported for the MCs.
The capacity of the milking center was used to represent the inventory levels. This is evident in

4-7


-------
Deliberative, draft document - Do not cite, quote, or distribute

the time series (Appendix E, Figure E-19) and the scatter plots (Appendix F, Figures F-41-F-45).
Average animal weight for the IN5B MC was reported as a constant value. For NY5B, the daily
value reported only vary by less than 5 kg (576 to 580 kg), like the mechanically ventilated barn.
This limited range of daily average animal weight is apparent in the time series (Appendix E,
Figure E-20). The regression analyses in Appendix F, Figures F-46 through F-50, summarized in
Table 4-4, showed only a slight or weak relationship between average animal weight and each
pollutant. Because of the constant inventory and near constant average animal weight, trends in
live animal weight (i.e., capacity * average animal weight) do not vary dramatically over the
monitoring period (Appendix E, Figure E-21). The regression analyses in Appendix F, Figures F-
51 through F-55 showed only slight relationships with emissions. To include size of the
operation in the models as a proxy for volume of manure produced, EPA opted to test models
where the emissions were normalized by the capacity of the MC. The models will yield an
estimate of emissions per head capacity of the MC.

Exhaust temperature was comparable between sites (Appendix E, Figure E-22), with
average daily means of 12.8°C at NY5B and 13.2°C at IN5B. The regression analyses (Appendix
F, Figures F-56 - F-60) showed a weak-to-modest correlation between exhaust temperature and
emissions, like the mechanically ventilated barns. Exhaust relative humidity was also
comparable between sites (Appendix E, Figure E-23), with average daily values of 74.2% and
73.8% at IN5B and NY5B, respectively. Like with mechanically ventilated barns, there is a
tendency for the lowest values to occur in the spring and fall. However, the wide scatter of
values for any time of the year, makes any strong seasonal pattern hard to discern. The regression
analyses (Appendix F, Figures F-61 - F-65), only showed slight-to-weak positive correlation
with emissions.

Airflow rates were much lower at NY5B than IN5B, which is clearly demonstrated in the
time series plot (Appendix E, Figure E-24). Average airflow rates were 39.90 dsm V atNY5B
and 183.33 dsmV1 at IN5B. The MC at IN5B is connected to Barn 1 at the site (see Figure 1-2 in
Section 1), while the MC at NY5B is connected to both Barn 1 and a naturally ventilated barn
(see Figure 1-3 in Section 1). It is possible the connection to the naturally ventilated barn
reduced the ventilation needs at the MC. The regression analyses (Appendix F, Figures F-66 - F-
70) showed only a slight to weak correlation with emissions, except for PMio, which has a
modest correlation.

4-8


-------
Deliberative, draft document - Do not cite, quote, or distribute
Table 4-5. Milking center environmental parameter regression analyses

Pollutant

Parameter

R

R'

Strength

Figure

nh3

Inventory (MC Capacity)

0.279

0.078

slight or weak

Appendix F, F-41

HzS

Inventory (MC Capacity)

0.360

0.130

slight or weak

Appendix F, F-42

PMio

Inventory (MC Capacity)





None

Appendix F, F-43

PM2.5

Inventory (MC Capacity)





None

Appendix F, F-44

TSP

Inventory (MC Capacity)





None

Appendix F, F-45

NH3

Average animal weight

0.279

0.078

slight or weak

Appendix F, F-46

h2s

Average animal weight

0.360

0.130

slight or weak

Appendix F, F-47

PM10

Average animal weight

-0.005

< 0.001

slight or weak

Appendix F, F-48

PM2.5

Average animal weight

-0.161

0.026

slight or weak

Appendix F, F-49

TSP

Average animal weight

0.154

0.024

slight or weak

Appendix F, F-50

NH3

Live animal weight

0.279

0.078

slight or weak

Appendix F, F-51

HzS

Live animal weight

0.360

0.130

slight or weak

Appendix F, F-52

PM10

Live animal weight

-0.005

< 0.001

slight or weak

Appendix F, F-53

PM2.5

Live animal weight

-0.161

0.026

slight or weak

Appendix F, F-54

TSP

Live animal weight

0.154

0.024

slight or weak

Appendix F, F-55

NH3

Exhaust temperature

0.518

0.268

modest

Appendix F, F-56

h2s

Exhaust temperature

0.322

0.104

slight or weak

Appendix F, F-57

PM10

Exhaust temperature

0.550

0.303

modest

Appendix F, F-58

PM2.5

Exhaust temperature

0.401

0.160

slight or weak

Appendix F, F-59

TSP

Exhaust temperature

0.348

0.121

slight or weak

Appendix F, F-60

NH3

Exhaust relative humidity

-0.188

0.035

slight or weak

Appendix F, F-61

h2s

Exhaust relative humidity

-0.378

0.143

slight or weak

Appendix F, F-62

PM10

Exhaust relative humidity

-0.111

0.012

slight or weak

Appendix F, F-63

PM2.5

Exhaust relative humidity

-0.241

0.058

slight or weak

Appendix F, F-64

TSP

Exhaust relative humidity

0.184

0.034

slight or weak

Appendix F, F-65

NH3

Airflow

0.381

0.146

slight or weak

Appendix F, F-66

h2s

Airflow

0.332

0.110

slight or weak

Appendix F, F-67

PM10

Airflow

-0.458

0.210

modest

Appendix F, F-68

PM2.5

Airflow

-0.009

< 0.001

slight or weak

Appendix F, F-69

TSP

Airflow

0.106

0.011

slight or weak

Appendix F, F-70

4.2.3 Ambient Data

The statistical summary of the ambient parameters associated with MCs are presented in
Appendix D, Table D-20. The summary statistics indicate the ambient temperatures are similar
for both sites, with average daily mean of 11.13°C at NY5B and 12.20°C at IN5B. Ambient
temperature trends (Appendix E, Figure E-27) follow seasonal patterns, as expected, and the time
series reiterates the similarity in temperatures at both sites. The regression analyses (Appendix F,
Figures F-71 - F-75) summarized in Table 4-5, showed weak-to-modest positive correlation with
emissions.

Ambient relative humidity was also similar between the sites with average daily mean of
67.81% atNY5B and 67.90% at IN5B. The time series (Appendix E, Figure E-28) showed
variability in average daily humidity values, with the lowest values occurring in the spring. The

4-9


-------
Deliberative, draft document - Do not cite, quote, or distribute

regression analyses (Appendix F, Figures F-76 - F-80), summarized in Table 4-5, showed only a
slight-to-weak correlation with emissions.

Table 4-6. Milking center ambient parameters regression analyses

Pollutant

Parameter

R

R2

Strength

Figure

nh3

Ambient temperature

0.495

0.245

modest

Appendix F, F-71

h2s

Ambient temperature

0.296

0.088

slight or weak

Appendix F, F-72

PM10

Ambient temperature

0.568

0.323

modest

Appendix F, F-73

PM2.5

Ambient temperature

0.399

0.159

slight or weak

Appendix F, F-74

TSP

Ambient temperature

0.348

0.121

slight or weak

Appendix F, F-75

NH3

Ambient relative humidity

-0.043

0.002

slight or weak

Appendix F, F-76

h2s

Ambient relative humidity

0.039

0.002

slight or weak

Appendix F, F-77

PM10

Ambient relative humidity

-0.421

0.178

slight or weak

Appendix F, F-78

PM2.5

Ambient relative humidity

0.043

0.002

slight or weak

Appendix F, F-79

TSP

Ambient relative humidity

0.066

0.004

slight or weak

Appendix F, F-80

4.3 Naturally Ventilated Barns (CA5B-B1, CA5B-B2, WA5B-B2 and WA5B-B4)

4.3.1 Emissions Data

Appendix D, Table D-21 and Table D-22 presents the summary statistics, in kg d"1 and g
d"1 hd"1, for daily average emissions of NH3 for the naturally ventilated sites. The average daily
emission rate is substantially different between the sites, ranging from 2.76 kg d"1 (4.98 g d"1 hd"
at CA5B-B1 to 54.65 kg d"1 (56.51 g d"1 hd"1) at WA5B-B4. The time series plot (Appendix E,
Figure E-29) showed the highest emissions at WA5B occurring in late spring to early summer of
2008. After a break in observations, the emission levels mostly drop to lower levels, though it is
still greater than CA5B. CA5B does have quite a few negative days, 37 at B1 and 42 at B2,
which are contributing to the lower overall average compared to WA5B. These negative numbers
were further reviewed during the data set review process described in Section 2, prior to
inclusion in the model development dataset. Appendix E, Figure E-29 also showed the emissions
are variable across the year with no obvious seasonal pattern.

The summary statistics for daily average FhS emissions are presented in Appendix D,
Table D-23 and D-24 for g d"1 and mg d"1 hd"1, respectively. Unlike the NH3 emissions, the
average of the daily emissions are more comparable across the sites. However, reviewing the
time series plot (Appendix E, Figure E-30) showed more variability at WA5B, including a few
very high values and extreme negative values. There were several negative values at each barn,
ranging from 18 values at CA5B-B2 to 45 values at WA5B-B2. Some of the negative numbers
were quite large, -609.00 g d"1 at CA5B-B2 to -11,640.14 g d"1 at WA5B-B2. These negative
numbers were further reviewed during the dataset review process described in Section 2, prior to
inclusion in the model development dataset. Appendix E, Figure E-30 also showed the emissions
are variable across the year with no obvious seasonal pattern.

4-10


-------
Deliberative, draft document - Do not cite, quote, or distribute

The summary statistics for PMio are presented in Appendix D, Table D-25 and D-26 for g
d"1 and mg d"1 hd"1, respectively. Like NH3, the PM10 emissions vary between the barns, even
when accounting for the differences in inventory. Average daily emissions range from -325.80 g
d"1 (-636.79 mg d"1 hd"1) at CA5B-B1 to 11,391.71 g d"1 (11,794.47 mg d"1 hd"1) at WA5B-B4.
CA5B has quite a few negative days, 372 at B1 and 221 at B2, which are contributing to the
lower overall average compared to WA5B, and the overall negative average for CA5B-B1.

These negative numbers were further reviewed during the dataset review process described in
Section 2, prior to inclusion in the model development dataset. The time series plot (Appendix E,
Figure E-31) showed the frequency of the negatives at CA5B, as well as the extremely high
values seen at WA5B.

PM2.5 was like PM10 in that there is a substantial number of negative daily emission
values at CA5B (Appendix D, D-27, and D-28). Specifically, at Bl, 44 of the 47 values are
negative and 40 of 54 are negative at B2. This results in a negative overall average value for
CA5B barns. The WA5B site has fewer negative values, 0 at WA5B-B2 and 6 at WA5B-B4.
These negative numbers were further reviewed during the dataset review process described in
Section 2, prior to inclusion in the model development dataset. The time series plot (Appendix E,
Figure E-3 2) showed the frequency of the negatives at CA5B, as well as the spread in values
seen in at WA5B. No seasonal pattern was apparent.

Regarding the TSP summary statistics (Appendix D, D-29, and D-30), the two sites have
different daily average values despite fewer negative daily emission values for CA5B than the
other PM size fractions. Average TSP daily emissions ranged from 4,766g d"1 (9113mg d"1 hd"1)
at CA5B-B1 to 47,389g d"1 (49,099mg d"1 hd"1) at WA5B-B4. The time series plot (Appendix E,
Figure E-33) showed a lot of variability in readings, which makes a seasonal pattern hard to
discern.

4.3.2 Environmental Data

The statistical summary of the environmental parameters associated with naturally
ventilated barns are presented in Appendix D, Table D-31. The average inventory for most of the
barns is between 514 at WA5B-B2 to 558 at WA5B-B2. WA5B-B4 is the exception, with an
average inventory almost double the other barn of 963.20 head. The time series (Appendix E,
Figure E-34) showed there is some variability in the inventory at the site, with most only varying
by 100 head from the average. The regression analyses (Appendix F, Figures F-81 - F-85) ,
summarized in Table 4-6, generally showed only slight or weak linear relationship with
emissions, except for NH3, which had a moderate positive linear relationship.

Average animal mass was provided as a single value and not reported daily. The
summary table (Appendix D, Table D-31) and the time series (Appendix E, Figure E-3 5)

4-11


-------
Deliberative, draft document - Do not cite, quote, or distribute

reiterate the single value. With constant values, the regression analyses (Appendix F, Figures F-
86 - F-90) showed only slight or weak relationship with emissions. Combining inventory and
average weight into live animal weight produces a size variable with trends (Appendix E, Figure
E-36), like inventory. Like the inventory regression analyses, Appendix F, Figures F-91 - F-95
showed a light or weak relationship with all pollutants except NH3, which had a moderate
positive relationship.

Average daily mean exhaust temperatures were slightly higher at CA5B. The means
ranged from 11.41°C at WA5B-B2 to 18.75°C at CA5B-B1. The time series (Appendix E,

Figure E-37) show similar trends and ranges between the sites, with lower values at the WA5B
barns. The regression analyses (Appendix F, Figures F-96 - F-100) indicated modest positive
relationships with NH3 and PM10 emissions and slight or weak relationships with other
pollutants.

The average daily exhaust relative humidity values are also slightly higher at CA5B. The
mean values ranged from 45.16% at WA5B-B4 to 58.49% at CA5B-B1. The time series
(Appendix E, Figure E-38) show the highest levels in the winter and lower values in the summer
at both sites. There is a lack of variability at the WA5B barns around January 2008 which will be
further investigated prior to finalizing the models. The regression analyses (Appendix F, Figures
F-101 - F-105) showed only slight to weak relationships with emissions, which were positive for
the gaseous pollutants and negative for the all the particulate matter size fractions.

Estimated airflows at the naturally ventilated barns were comparable and ranged from
882.65 dsmV1 to 1,151.61 dsmV1 at CA5B. The time series (Appendix E, Figure E-39) show
variability across the year, with slightly enhanced airflow during the summer. However, peak
values can occur at any time of year. The regression analyses (Appendix F, Figures F-106 - F-
110) showed modest positive linear relationship with NH3 and PM2.5 emissions. All other
pollutants had a slight positive relationship with airflow.

Table 4-7. Naturally ventilated environmental parameter regression analyses

Pollutant

	GtGr

R

R2

Strength

Figure

nh3

Inventory

0.660

0.435

moderate

Appendix F, F-81

HzS

Inventory

0.002

< 0.001

slight or weak

Appendix F, F-82

PM10

Inventory

-0.292

0.085

slight or weak

Appendix F, F-83

PM2.5

Inventory

-0.319

0.102

slight or weak

Appendix F, F-84

TSP

Inventory

-0.327

0.107

slight or weak

Appendix F, F-85

NH3

Average animal weight

-0.423

0.179

slight or weak

Appendix F, F-86

h2s

Average animal weight

0.114

0.013

slight or weak

Appendix F, F-87

PM10

Average animal weight

0.240

0.058

slight or weak

Appendix F, F-88

PM2.5

Average animal weight

0.384

0.148

slight or weak

Appendix F, F-89

TSP

Average animal weight

0.384

0.147

slight or weak

Appendix F, F-90

4-12


-------
Deliberative, draft document - Do not cite, quote, or distribute

Pollutant

Parameter

R

R2

Strength

Figure

nh3

Live animal weight

0.653

0.426

moderate

Appendix F, F-91

h2s

Live animal weight

0.014

< 0.001

slight or weak

Appendix F, F-92

PM10

Live animal weight

-0.278

0.077

slight or weak

Appendix F, F-93

PM2.5

Live animal weight

-0.283

0.080

slight or weak

Appendix F, F-94

TSP

Live animal weight

-0.307

0.094

slight or weak

Appendix F, F-95

NH3

Exhaust temperature

0.493

0.243

modest

Appendix F, F-96

h2s

Exhaust temperature

0.323

0.104

slight or weak

Appendix F, F-97

PM10

Exhaust temperature

0.410

0.168

slight or weak

Appendix F, F-98

PM2.5

Exhaust temperature

0.484

0.234

modest

Appendix F, F-99

TSP

Exhaust temperature

0.406

0.165

slight or weak

Appendix F, F-100

NH3

Exhaust relative humidity

0.390

0.152

slight or weak

Appendix F, F-101

HzS

Exhaust relative humidity

0.193

0.037

slight or weak

Appendix F, F-102

PM10

Exhaust relative humidity

-0.269

0.072

slight or weak

Appendix F, F-103

PM2.5

Exhaust relative humidity

-0.414

0.171

slight or weak

Appendix F, F-104

TSP

Exhaust relative humidity

-0.322

0.104

slight or weak

Appendix F, F-105

NH3

Airflow

0.536

0.287

modest

Appendix F, F-106

h2s

Airflow

0.232

0.054

slight or weak

Appendix F, F-107

PM10

Airflow

0.425

0.180

slight or weak

Appendix F, F-108

PM2.5

Airflow

0.449

0.202

modest

Appendix F, F-109

TSP

Airflow

0.376

0.141

slight or weak

Appendix F, F-110

4.3.3 Ambient Data

The statistical summary of the ambient parameters associated with naturally ventilated
barns are presented in Appendix D, Table D-32. Ambient temperatures were generally higher at
CA5B leading to an average of the daily mean of 16.34°C compared to 10.07°C at WA5B. The
time series (Appendix E, Figure E-40) showed the typical seasonal trend. Of note, the
temperatures in summer 2008 were substantially lower than summer 2009. The site report noted
the temperature sensor produced a "noisy signal" from late October 2007 to March of 2008. The
average of the sonic anemometers was used as a substitute after analysis to confirm agreement
with the remaining dates (Ramirez-Dorronsoro et al., 2010). The regression analyses (Appendix
F, Figures F-l 11 - F-l 15), summarized in Table 4-7, showed a modest positive relationship with
temperature and weak positive correlations with all other pollutants.

On average, the ambient relative humidity was lower at WA5B (45.81%) than CA5B
(62.01%). The time series (Appendix E, Figure E-41) showed a muted peak around January 2008
for WA5B, like the exhaust relative humidity for the site. The site report offered no explanation
for the plateau to the values. The regression analyses (Appendix F, Figures F-l 16 - F-120)
showed slight or weak negative relationships with the emission value. The negative relationship
between NH3 emission and relative humidity is consistent with Sanchis et al. (2019).

Wind speeds averaged slightly higher at WA5B (2.59 ms"1) than CA5B (1.97ms"1). The
time series (Appendix E, Figure E-42) showed no distinct seasonal trends, as peak and minimum

4-13


-------
Deliberative, draft document - Do not cite, quote, or distribute

values occurred throughout the year. The regression analyses (Appendix F, Figures F-121 - F-
125) showed a modest positive relationship with NH3 emissions, and weak positive relationships
with all other pollutants.

Table 4-8. Naturally ventilated ambient parameters regression analyses

Pollutant

Parameter

R

R2

Strength

Figure

nh3

Ambient temperature

0.537

0.289

modest

Appendix F, F-lll

h2s

Ambient temperature

0.257

0.066

slight or weak

Appendix F, F-112

PM10

Ambient temperature

0.370

0.137

slight or weak

Appendix F, F-113

PM2.5

Ambient temperature

0.398

0.159

slight or weak

Appendix F, F-114

TSP

Ambient temperature

0.348

0.121

slight or weak

Appendix F, F-115

NH3

Ambient relative humidity

-0.110

0.012

slight or weak

Appendix F, F-116

h2s

Ambient relative humidity

< 0.001

< 0.001

none

Appendix F, F-117

PM10

Ambient relative humidity

-0.129

0.017

slight or weak

Appendix F, F-118

PM2.5

Ambient relative humidity

-0.331

0.109

slight or weak

Appendix F, F-119

TSP

Ambient relative humidity

-0.155

0.024

slight or weak

Appendix F, F-120

NH3

Wind speed

0.537

0.289

modest

Appendix F, F-121

HzS

Wind speed

0.257

0.066

slight or weak

Appendix F, F-122

PM10

Wind speed

0.370

0.137

slight or weak

Appendix F, F-123

PM2.5

Wind speed

0.398

0.159

slight or weak

Appendix F, F-124

TSP

Wind speed

0.348

0.121

slight or weak

Appendix F, F-125

4.4 Open Sources (IN5A, WI5A and TX5A)

4.4.1 Emissions Data

Appendix D, Table D-33 presents the summary statistics for daily average emissions of
NH3 for the open source sites, including corrals. Appendix D, Table D-34 presents the emissions
per square meter of surface area. The emissions from the sites with lagoons, IN5A and WI5A,
were comparable, with emissions ranging from 19.83 kg d"1 (2.01 g d"1 m"2) at IN5A to 11.45 kg
d"1 (1.61 g d"1 m"2) at WI5A. The time series (Appendix E, Figures E-43, and E-45) showed the
observations from IN5 A in the same year and show a seasonal pattern. The observations from
WI5B are more spread out over the two-year monitoring period and showed a subtle seasonal
pattern. The NFb emissions for corrals was higher than for the lagoons on a per day basis with
average emissions of 754.97 kg d"1 (222.1 g d"1 hd"1). However, when normalized for the surface
area, it was slightly greater at 3.12 g d"1 m"2 The time series for the corral site (TX5A) is
available in Appendix E, Figure E-52. There are not many summertime observations, so
seasonality is hard to discern.

Appendix D, Table D-35 presents the summary statistics for daily average emissions of
NH3 for the open source sites, including corrals. Appendix D, Table D-36 presents the emissions
per square meter of surface area. The average FhS emissions from the lagoon sites, showed more
of a difference, with emissions ranging from to 0.42 kg d"1 (0.06 kg d"1 m"2) at WI5A to 9.39 kg
d"1 (0.95 kg d"1 m"2) at IN5A. The time series (Appendix E, Figure E-44, and E-46) showed the

4-14


-------
Deliberative, draft document - Do not cite, quote, or distribute

observations from IN5 A in the same year and show a seasonal pattern. The observations from
WI5B are more spread out over the two-year monitoring period and showed a subtle seasonal
pattern. The H2S emissions for the corral was greater than for the lagoons at 10.69 kg d"1 (3.14 g
d"1 hd"1) but was much less when normalized by area (44.18 mg d"1 m"2). The time series for the
corral site is available in Appendix E, Figure E-53. No seasonal pattern was apparent.

4.4.2 Environmental Data

The statistical summary of the environmental parameters associated with dairy lagoons
are presented in Appendix D, Table D-37. Lagoon temperatures were colder at WI5A, which had
an average daily mean temperature of 18.35°C compared to 21.57°C at IN5A. The time series
(Appendix E, Figure E-47) shows the spare nature of the observations but does suggest the
expected trend of lagoon temperatures following seasonal temperature patterns. The regression
analyses (Appendix F, Figures F-126 - F-127; summarized in Table 4-8) shows moderate
relationships with daily emissions (kg/d).

Lagoon pH was consistent between the sites, with average daily mean values at 7.02 and
7.43 for WI5A and IN5A, respectively. The time series (Appendix E, Figure E-48) shows values
typically between 7.0 and 7.5 for most of the observations. There is a small cluster of readings
for IN5A above 8.0 for Fall 2008. The regression analyses (Appendix F, Figures F-128 - F-129),
summarized in Table 4-8, showed only slight or weak relationships with daily emissions (kg/d).

Table 4-9. Open source environmental parameter regression analyses

r. II .. »

; Pollutant

Parameter

R

R2

Strength

Figure

NHs

Lagoon temperature

0.66

0.436

moderate

Appendix F, F-126

h2s

Lagoon temperature

-0.68

0.462

moderate

Appendix F, F-127

NHs

Lagoon pH

-0.2

0.040

slight or weak

Appendix F, F-128

H2S

Lagoon pH

0.4

0.160

slight or weak

Appendix F, F-129

4.4.3 Ambient Data

The statistical summary of the ambient parameters associated with dairy lagoons are
presented in Appendix D, Table D-38. The average ambient temperature observed during
monitoring periods for WI5A (-3.41°C) was much lower than IN5A (6.25°C). The time series
(Appendix E, Figure E-49) show the expected seasonal trend in temperatures. The regression
analyses (Appendix F, Figures F-130 - F-131), summarized in Table 4-9, show modest and
moderately strong positive relationships with FhS and NH3 daily emissions (kg/d), respectively.

Observed ambient relative humidity were comparable between sites, with average daily
means ranging from 71.53% at WI5A to 72.02% at IN5A. The time series (Appendix E, Figure
E-50) show the relative humidity values vary throughout the year with no seasonal pattern. The

4-15


-------
Deliberative, draft document - Do not cite, quote, or distribute

regression analyses (Appendix F, Figures F-132 - F-133) shows a slight negative relationship
with daily emissions (kg/d) of both NH3 and FhS.

Wind speeds were also comparable between sites and ranged from 3.28 m s"1 at IN5A to
3.45 m s"1 at WI5A. The time series (Appendix E, Figure E-51) average daily wind speeds were
equally variable throughout the year at both sites. The regression analyses (Appendix F, Figures
F-134 - F-135) showed only slight correlation with daily emissions (kg/d), which was negative
for NH3 and positive for FhS.

Table 4-10. Open source ambient parameters regression analyses

Pollutant

Parameter

R

R2

Strength

Figure

NHs

Ambient temperature

0.84

0.706

moderately strong

Appendix F, F-130

h2s

Ambient temperature

0.59

0.348

modest

Appendix F, F-131

NHs

Ambient relative humidity

-0.34

0.116

slight or weak

Appendix F, F-132

H2S

Ambient relative humidity

-0.18

0.032

slight or weak

Appendix F, F-133

NHs

Wind speed

-0.25

0.063

slight or weak

Appendix F, F-134

H2S

Wind speed

0.1

0.010

slight or weak

Appendix F, F-135

The statistical summary of the ambient parameters associated with the monitored dairy
corral are presented in Appendix D, Table D-39. Observations of ambient temperature ranged
from -5.64°C to 27.50°C, and followed expected seasonal trends (Appendix E, Figure E-54). The
regression analyses (Appendix F, Figures F-136 - F-137; summarized in Table 4-10) showed a
slight positive relationship between temperature and emissions.

Average daily ambient relative humidity values ranged from 22.3% to 78.54% over the
study at TX5A. The time series (Appendix E, Figure E-55) do not suggest any seasonal trends.
The regression analyses (Appendix F, Figures F-138-F-139) shows slight positive relationships
with emissions. Average daily wind speeds ranged from 2.35 to 6.79 ms"1 and showed no trends
in the time series (Appendix E, Figure E-56). The time series did show a peak value in late
winter to spring of 2009. The regression analyses (Appendix F, Figures F-140 - F-141) do not
show a relationship between wind speed and emissions.

Water vapor deficit estimates ranged from 2.09 to 26.88 hectopascal (hPa) and showed
some tendency for higher values in the summer and fall (Appendix E, Figure E-57). The
regression analyses (Appendix F, Figures F-142 - F-143) summarized in Table 4-10 indicated a
slight relationship between emissions that was positive for NH3 and negative for FhS.

Table 4-11. Corral ambient parameters regression analyses

Pollutant Parameter

R

R2

Strength

Figure

NHs

Ambient temperature

0.17

0.029

slight or weak

Appendix F, F-136

H2S

Ambient temperature

0.003

< 0.001

slight or weak

Appendix F, F-137

4-16


-------
Deliberative, draft document - Do not cite, quote, or distribute

NHs

Ambient relative humidity

0.17

0.029

slight or weak

Appendix F, F-138

h2s

Ambient relative humidity

0.15

0.023

slight or weak

Appendix F, F-139

NHs

Wind speed

0.002

< 0.001

slight or weak

Appendix F, F-140

H2S

Wind speed

0.003

< 0.001

slight or weak

Appendix F, F-141

NHs

Water vapor deficit

0.32

0.102

slight or weak

Appendix F, F-142

H2S

Water vapor deficit

-0.16

0.026

slight or weak

Appendix F, F-143

4-17


-------
Deliberative, draft document - Do not cite, quote, or distribute

5 DEVELOPMENT AND SELECTION OF MODELS FOR DAILY EMISSIONS
5.1 Mechanically Ventilated Barns

The literature review (Section 3) and exploratory data analysis (Section 4) suggested that
EPA should consider ambient temperature, exhaust temperature, ambient relative humidity,
exhaust relative humidity, manure management system, and inventory in the development of the
emission models for mechanically ventilated barns. Barn airflow, or ventilation rate, can have a
substantial influence on the emission rate of gaseous pollutants, but was not included in the
parameter list as it may not be easily obtained at all farms. Since ventilation rate is essentially
driven by the temperature (i.e., the higher ambient temperature the higher the ventilation rate),
the ambient temperature provides an indication of airflow in the models tested.

The various combinations of these parameters were used in test models. For NH3 and
H2S, 9 different combinations were tested as potential models (Table 5-1). There were 17 models
(Table 5-2) tested for particulate matter emissions, which had variations to predict the emissions
normalized by inventory.

Table 5-1. Parameter combinations tested as mechanically ventilated barn models

for NH3 and H2S emissions.

Model Parameters

MV-G1

Inventory, manure management system (Flush, Scrape)

MV-G2

Inventory, exhaust temperature, Exhaust relative humidity, manure management system
(Flush, Scrape)

MV-G3

Inventory, exhaust temperature, manure management system (Flush, Scrape)

MV-G4

Inventory, exhaust relative humidity, manure management system (Flush, Scrape)

MV-G5

Inventory, ambient relative humidity, ambient temperature, manure management system
(Flush, Scrape)

MV-G6

Inventory, ambient temperature, manure management system (Flush, Scrape)

MV-G7

Inventory, ambient relative humidity, manure management system (Flush, Scrape)

MV-G8

Inventory, ambient temperature, exhaust relative humidity, manure management system
(Flush, Scrape)

MV-G9

Inventory, exhaust temperature, ambient relative humidity, manure management system
(Flush, Scrape)

Table 5-2. Parameter combinations tested as mechanically ventilated barn models

for PM10, PM2.5, and TSP emissions.

_

¦ ;

Parameters

MV-P1

Intercept, inventory

MV-P2

Intercept, inventory, exhaust temperature, exhaust relative humidity

MV-P3

Intercept, inventory, exhaust temperature

MV-P4

Intercept, inventory, exhaust relative humidity

MV-P5

Intercept, inventory, ambient relative humidity, ambient temperature

MV-P6

Intercept, inventory, ambient temperature

5-1


-------
Deliberative, draft document - Do not cite, quote, or distribute

Model

Parameters

MV-P7

Intercept, inventory, ambient relative humidity

MV-P8

Intercept, inventory, ambient temperature, exhaust relative humidity

MV-P9

Intercept, inventory, exhaust temperature, ambient relative humidity

MV-P10

Intercept, exhaust temperature, exhaust relative humidity
(Emissions normalized by inventory)

MV-P11

Intercept, exhaust temperature (emissions normalized by inventory)

MV-P12

Intercept, exhaust relative humidity (emissions normalized by inventory)

MV-P13

Intercept, ambient temperature, ambient relative humidity
(Emissions normalized by inventory)

MV-P14

Intercept, ambient temperature (emissions normalized by inventory)

MV-P15

Intercept, ambient relative humidity (emissions normalized by inventory)

MV-P16

Intercept, ambient temperature, exhaust relative humidity
(Emissions normalized by inventory)

MV-P17

Intercept, ambient relative humidity, exhaust temperature
(Emissions normalized by inventory)

For both NH3 (Appendix G, Table G-3) and H2S (Appendix G, Table G-5), models MV-
G5 and MV-G7 had terms that were not statistically significant (p > 0.05) for both pollutants and
were removed from further consideration. For FhS, model MV-G4 and G9 had insignificant
terms. The model fit (-2 log likelihood, AIC, AICc, and BIC) and evaluation statistics (ME,
NME, MB, NMB) for NH3 (Appendix G, Table G-4) and FhS (Appendix G, Table G-5) indicate
the remaining models had comparable performance, which suggested that using ambient
parameters was as effective as models that included barn specific parameters. As noted in the
Process Overview report, the model selection process also looked at how easily obtainable the
parameters are as not to create an undue burden on the operators. Generally, ambient parameters
were preferred since ambient meteorological data is actively recorded across the country and
representative site data is accessible through the National Centers for Environmental Information
(NCEI) website. To further ease any burden, the EPA plans to provide a tool that automatically
populates relevant ambient parameters for any given location instead of requiring producers to
measure and record environmental parameters either inside or outside of the barn to further
reduce the burden of use on the producer.

Therefore, considering ambient temperature is a suitable proxy for barn airflow as
exhaust temperature and representative ambient temperature data is accessible, the EPA
concluded that a model using ambient temperature and relative humidity would be preferable to
one with exhaust temperature and relative humidity. Of the remaining models that used ambient
parameters (MV-G1, and G6), EPA selected model MV-G6 (including the parameters: inventory,
ambient temperature, and manure management system) for further analysis for both NH3 and
FhS as it had the best normalized mean bias of the remaining models. The final form of these
models is presented in Table 5-3.

5-2


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 5-3. Selected daily models for mechanically ventilated barns.

Pollutant



Units

Equation
Number

NHs, Flush

ln(NH3) = 1.746585 + 1.773832 * Inventory + 0.029586 * AmbT

kg/d

Equation 1

NH3, Scrape

ln(NH3) = 1.864935 + 1.773832 * Inventory + 0.029586 * Ambr

kg/d

Equation 2

H2S, Flush

ln(H2S) = 7.406887 + 0.86173 * Inventory + 0.012786 * AmbT

g/d

Equation 3

H2S, Scrape

ln(H2S) = 6.287004 + 0.86173 * Inventory + 0.012786 * AmbT

g/d

Equation 4

For PMio models (Appendix G, Table G-7), models MVP-1 through MVP-9 include
inventory as a proxy for volume of manure produced. While all model terms were statistically
significant (p > 0.05), coefficients for inventory were negative which suggests that emissions
decrease as inventory increases. The negative coefficients for inventory are also seen in models
MVP-1 through MVP-9 for PM2.5 (Appendix G, Table G-9) and TSP (Appendix G, Table G-l 1).
As noted in Section 3.2, Garcia et al. (2012) found a similar inverse relationship with PM2.5
concentrations and inventory for MCs, which was attributed to the larger dairies being newer and
more efficiently operated. Based on the site reports, the older barns have the lowest average
inventory (Table 5-4), which lines up with Garcia et al. (2012). Still unknown is the management
practice in the newer barns contributing to the reduced emissions and how to account for that
practice in the model. Age of the barn and construction year were discussed as a possible
parameter; however, there is not enough variability in construction year available in the NAEMS
data for model construction.

Table 5-4. Summary of barn construction dates for mechanically ventilated barns.

Barn 1 Year Constructed Average Inventory

IN5B-B1

2004

833

IN5B-B2

2004

864

NY5B-B1

1998

467

WI5B-B1

1990

211

WI5B-B2

1994

355

EPA tested a set of models that normalized emissions by inventory, MVP-10 through
MVP-17, which use the same environmental and barn parameters as models MVP-2 through
MVP-9. The goal was to determine if these models could be predictive based on the other
environmental and ambient parameters alone. The model performance statistics (i.e., ME, NME,
MB, NMB) did increase for these models (Appendix G, Tables G-8, G-10, and G-12), suggesting

5-3


-------
Deliberative, draft document - Do not cite, quote, or distribute

accounting for the difference in newer barns is needed for a successful model. Therefore, EPA is
not selecting a model at this time to allow for more research into the reason newer barns have
lower particulate matter emissions, despite increased animal populations.

5.2 Milking Centers

The literature review (Section 3) and exploratory data analysis (Section 4) suggested that
EPA should consider ambient temperature, exhaust temperature, ambient relative humidity,
exhaust relative humidity, milk production, and inventory in the development of the emission
models for MCs. Barn airflow, or ventilation rate, can have a substantial influence on the
emission rate, but was not included in the parameter list as it may not be easily obtained at all
farms. Since ventilation rate is essentially driven by the temperature (i.e., the higher ambient
temperature the higher the ventilation rate), the ambient temperature provides an indication of
airflow in the models tested. EPA tested 24 combinations of these parameters as potential models
(Table 5-5), including which had variations to predict the emissions normalized by inventory
(MC-25 through MC-32). The models to predict normalized emissions were added to incorporate
a barn size into the model, as the relatively consistent inventory of the MCs could reduce the
significance if inventory was used as a predictive parameter. This is demonstrated with the NH3
modeling results (Appendix G, Table G-13), as inventory is insignificant in models MC-10
through MC-16.

Milk production values were only available for NY5B, and when combined with a static
value for barn inventory, as in models MC-1 through MC-8, inventory was dropped from the
model, making the result equivalent to models MC-17 through MC-24 for all pollutants.
Therefore, the summary presented in this section will focus on models MC-8 through MC-32.
Results for all models is summarized in Appendix G.

Table 5-5. Parameter combinations tested as milking center models.

Model

Parameters

MC-1

Intercept, inventory, milk production, exhaust temperature, exhaust relative humidity

MC-2

Intercept, inventory, milk production, exhaust temperature

MC-3

Intercept, inventory, milk production, exhaust relative humidity

MC-4

Intercept, inventory, milk production, ambient relative humidity, ambient temperature

MC-5

Intercept, inventory, milk production, ambient temperature

MC-6

Intercept, inventory, milk production, ambient relative humidity

MC-7

Intercept, inventory, milk production, ambient temperature, exhaust relative humidity

MC-8

Intercept, inventory, milk production, exhaust temperature, ambient relative humidity

MC-9

Intercept, inventory, exhaust temperature, exhaust relative humidity

MC-10

Intercept, inventory, exhaust temperature

MC-11

Intercept, inventory, exhaust relative humidity

MC-12

Intercept, inventory, ambient relative humidity, ambient temperature

5-4


-------
Deliberative, draft document - Do not cite, quote, or distribute

MC-13

Intercept, inventory, ambient temperature

MC-14

Intercept, inventory, ambient relative humidity

MC-15

Intercept, inventory, ambient temperature, exhaust relative humidity

MC-16

Intercept, inventory, exhaust temperature, ambient relative humidity

MC-17

Intercept, milk production, exhaust temperature, exhaust relative humidity

MC-18

Intercept, milk production, exhaust temperature

MC-19

Intercept, milk production, exhaust relative humidity

MC-20

Intercept, milk production, ambient relative humidity, ambient temperature

MC-21

Intercept, milk production, ambient temperature

MC-22

Intercept, milk production, ambient relative humidity

MC-23

Intercept, milk production, ambient temperature, exhaust relative humidity

MC-24

Intercept, milk production, exhaust temperature, ambient relative humidity

MC-25

Intercept, exhaust temperature, exhaust relative humidity
(Emissions normalized by inventory)

MC-26

Intercept, exhaust temperature (emissions normalized by inventory)

MC-27

Intercept, exhaust relative humidity (emissions normalized by inventory)

MC-28

Intercept, ambient temperature, ambient relative humidity
(Emissions normalized by inventory)

MC-29

Intercept, ambient temperature (emissions normalized by inventory)

MC-30

Intercept, ambient relative humidity (emissions normalized by inventory)

MC-31

Intercept, ambient temperature, exhaust relative humidity
(Emissions normalized by inventory)

MC-32

Intercept, ambient relative humidity, exhaust temperature
(Emissions normalized by inventory)

For NH3 (Appendix G, Table G-13) models MC-1 through MC-24 had terms that were
not statistically significant (p > 0.05). All the models predicting NH3 emissions per head (MC-25
through MC-32) were comprised of significant parameters. The model fit (-2 log likelihood,
AIC, AICc, and BIC) and evaluation statistics (ME, NME, MB, NMB) for these models are
presented in Appendix G, Table G-14. The ambient parameter models performed comparably to
their barn parameter counterparts, suggesting selecting the models with the easier to obtain
ambient parameter would be as effective. Therefore, EPA concluded that a model using ambient
temperature and relative humidity would be preferable to one with exhaust temperature and
relative humidity. Of the remaining models that used ambient parameters (MC-28, MC-29, and
MC-30), the NME and ME are comparable for the models. Model MC-30 has a substantially
lower MB and NMB. However, this model only includes relative humidity and not temperature.
The literature search (Section 3) noted that temperature is strongly linked to NH3 emissions and
should be included in the selected model. The model performance plots (Appendix G, Figures G-
20 & G-24) also show better scatter across the one-to-one (1:1) for models MC-28, MC-29,
indicating better predictive performance than model MC-30. Therefore, EPA selected model
MC-29 (including ambient temperature as the predictive parameter) for further analysis for NH3

5-5


-------
Deliberative, draft document - Do not cite, quote, or distribute

as it had the best NMB of the remaining models. The final form of these models is presented in
Table 5-6.

In addition to the models predicting normalized emissions, models MC-9, MC-10, MC-
11, MC-13, MC-15, MC-18, and MC-21 were comprised of significant parameters for H2S
(Appendix G, Table G-15). Of the seven additional models, all but MC-11 contained either
exhaust temperature or ambient temperature, as well as models MC-25 through MC-32.
Comparing the model fit and evaluation statistics (Appendix G, Table G-16) the ambient
parameter models performed comparably to their barn parameter counterparts, suggesting
models utilizing the easier to obtain ambient parameter would be as effective. Therefore, EPA
concluded that a model using ambient temperature and ambient relative humidity would be
preferable to one with exhaust temperature and relative humidity. Of the remaining models that
used ambient parameters (MC-13, MC-21, MC-28, MC-29, and MC-30), the error statistics
(NME and ME) are lower for models MC-13 and MC-21, while the bias statistics (MB and
NMB) are lower for MC-21 and MC-30, with other models being comparable. The scatter plots
of observed versus predicted (Appendix G, Figures G-26 through G-32) for model MC-21 has
more variability in the scatter across the 1:1 line, indicating a slightly better fit. However, this
model includes milk production, which is only available for one site. For this study, it is
preferred to include multiple sites in the model development dataset to represent variability
across the country. Therefore, EPA selected model MC-29 (including ambient temperature as the
predictive parameter) for further analysis for FhS as it had the best NMB the remaining models
(i.e., MC-13, MC-30). The final form of these models is presented in Table 5-6.

For the particulate matter size fractions, only NY5B reported emissions. With the dataset
dropping to one site with a constant value for MC capacity, the coefficient of inventory in
models MC-9 through MC-16 is estimated at zero and eliminates a size estimate from the model.
The focus for the particulate matter model narrowed to just models MC-17 through MC-32. For
PM10, models MC-17, MC-18, MC-19, MC-20, MC-21, and MC-23 have parameters that are
statistically insignificant (Appendix G, Table G-17). The model fit and evaluation statistics
(Appendix G, Table G-18) for models with ambient parameters performed comparably to their
barn parameter counterparts, suggesting models utilizing the easier to obtain ambient parameter
would be as effective. Of the remaining models that used ambient parameters (MC-28, MC-29,
and MC-30), the NME and ME are slightly lower for Model 28, and the bias parameters are
similar. EPA selected model MC-28 (including ambient temperature and ambient relative
humidity as the predictive parameter) for further analysis for PM10 as it had the best NMB of the
remaining models. The final form of these models is presented in Table 5-6.

As noted in Section 6.4 of the main report, the particulate matter model selection starts
with PM10 due to the greater quantity of emissions data. The PM10 models had between 315 and

5-6


-------
Deliberative, draft document - Do not cite, quote, or distribute

436 records available depending on the completeness of the various predictive parameters. For
PM2.5 and TSP the number of records available ranged between 40 - 44 for PM2.5 and 29 - 40 for
TSP. This is substantially less data that were available for PM10 and does not necessarily cover
the breadth of conditions that the PM10 data does. Therefore, the models generated with these
smaller datasets were examined mainly for consistency with the PM10 results to build confidence
in using the same model form for all the particulate matter species.

Compared to the PM10 models, more of the PM2.5 and TSP models have insignificant
terms. For both PM2.5 (Appendix G, Table G-19) and TSP (Appendix G, Table G-21), only
models MC-26 and MC-29 are comprised of significant parameters. Despite the insignificance of
the parameters for most of the models, the relationships were consistent with the PM10 models
and literature. The model performance statistics for PM2.5 (Appendix G, Table G-20) and the
model performance plots (Appendix G, Figures G-41 through G-48) were consistent, with
slightly lower bias metric for model MC-29. For TSP, the performance metrics (Appendix G,
Table G-22) and plots (Appendix G, Figures G-49 through G-56) were comparable. Therefore,
EPA selected model MC-29 for PM2.5 (including ambient temperature as the predictive
parameter) and model MC-28 (including ambient temperature and ambient relative humidity as
the predictive parameter) for TSP to conduct further evaluation and analysis as an emission
estimation method. The full forms of the models are presented in Table 5-6.

Table 5-6. Selected daily models for milking centers.

Pollutant

Formula

Units

Equation
Number

NHs

ln(NH3) = 2.505637 + 0.046434 * AmbT

g/d/hd

Equation 5

h2s

ln(H2S) = 6.898188 + 0.024053 * AmbT

kg/d/hd

Equation 6

PM10

ln(PM10) = 8.042215 + 0.006791 * AmbT - 0.003552 * AmbRH

g/d/hd

Equation 7

PM2.5

ln(PM2S) = 6.58377 + 0.006698 * Ambr

g/d/hd

Equation 8

TSP

ln(TSP) = 7.457268 + 0.010997 * AmbT - 0.003639 * AmbRH

g/d/hd

Equation 9

5.3 Naturally Ventilated Barns

The literature review (Section 3) and exploratory data analysis (Section 4) suggested that
EPA should consider ambient temperature, ambient relative humidity, exhaust relative humidity,

5-7


-------
Deliberative, draft document - Do not cite, quote, or distribute

wind speed, and inventory in the development of the emission models for naturally ventilated
barns. EPA tested 8 combinations of these parameters as potential models (Table 5-5). Models
predicting emissions normalized by inventory were not pursued at this time. However, based on
the initial results of MCs, normalized inventory models may be considered for the final models.

Table 5-7. Parameter combinations tested as naturally ventilated barns models.

Model

Parameters

NV-1

Intercept, inventory

NV-2

Intercept, inventory, ambient temperature, ambient relative humidity, wind speed

NV-3

Intercept, inventory, ambient temperature

NV-4

Intercept, inventory, ambient relative humidity

NV-5

Intercept, inventory, wind speed

NV-6

Intercept, inventory, ambient temperature, ambient relative humidity

NV-7

Intercept, inventory, ambient relative humidity, wind speed

NV-8

Intercept, inventory ambient temperature, wind speed

For the gaseous species, models NV-3 and NV-8 had terms that were not statistically
significant (p > 0.05) for NH3 (Appendix G, Table G-24), and models NV-2, NV-3, NV-4, NV-
6, NV-7, and NV-8 had insignificant terms for H2S (Appendix G, Table G-26). The model fit (-2
log likelihood, AIC, AICc, and BIC) and evaluation statistics (ME, NME, MB, NMB) for these
models are presented in Appendix G, Table G-25, and Table G-27 for NH3 and H2S,
respectively. For both pollutants, the statistics for the models were comparable. Therefore, EPA
selected model NV-5 (including as the predictive parameters: inventory and wind speed) for
further analysis for NH3 and FhS as it had the best NMB of the remaining models. The final form
of these models is presented in Table 5-8.

For PM10, all models were comprised of statistically significant parameters (Appendix G,
Table G-28). The model fit and evaluation statistics (Appendix G, Table G-29) suggested
comparable performance across all models, with model NV-2 having slightly better error
metrics. EPA selected model NV-2 (including the predictive parameters: inventory, ambient
temperature, ambient relative humidity, and wind speed) for further analysis. The final form of
the model is presented in Table 5-8.

As noted in Section 6.4 of the main report and with the MC model selection, the
particulate matter model selection starts with the PM10 due to the greater quantity of emissions
data. For naturally ventilated barns, the PM10 models had between 1,457 and 1,469 records
available depending on the completeness of the various predictive parameters. For PM2.5 and
TSP, the number of records available was 93 for PM2.5 and 205 for TSP. The PM2.5 models
(Appendix G, Table G-30) all have insignificant parameters. The relationship generally follows
the expected trend from literature (e.g., negative relationship with relative humidity). However,

5-8


-------
Deliberative, draft document - Do not cite, quote, or distribute

inventory has a negative coefficient in each model. For TSP (Appendix G, Table G-32), all
models are comprised entirely of significant parameters and the predictive parameters have the
same relationships as with PMio. Model NV-2 had reasonable performance for both PM2.5
(Appendix G, Table G-31) and TSP (Appendix G, Table G-33) and would be consistent with the
PM10 formulation that was developed from a much larger dataset. Therefore, EPA selected
model NV-2 (including the predictive parameters: inventory, ambient temperature, ambient
relative humidity, and wind speed) for further analysis. The final form of the models for PM2.5
and TSP are presented in Table 5-8.

Table 5-8. Selected daily models for naturally ventilated barns.

Pollutant

Formula

Units

Equation
Number

NHs

ln(NH3) = 0.188357 + 3.451939 * Inventory + 0.048153
* WindSpeed

g/d

Equation 10

h2s

ln(H2S) = 6.541057 + 0.587702 * Inventory + 0.062678
* WindSpeed

kg/d

Equation 11

PM10

ln(PM10) = 7.64258 + 1.525009 * Inventory + 0.011864 * AmbT
— 0.01521 * AmbRH + 0.173698 * WindSpeed

g/d

Equation 12

PM2.5

ln(PM2.n) = 7.068797 — 0.220453 * Inventory + 0.01121 * AmbT
— 0.003808 * AmbRH + 0.218968 * WindSpeed

g/d

Equation 13

TSP

ln(TSP) = 7.868847 + 2.953893 * Inventory + 0.034508 * AmbT
— 0.033997 * AmbRH + 0.248191 * WindSpeed

g/d

Equation 14

5.4 Open Sources

The literature review (Section 3) and exploratory data analysis (Section 4) suggested that
EPA should consider lagoon pH, lagoon temperature, ambient temperature, and wind speed in
the development of the emission models for open sources. EPA tested 15 combinations of these
parameters as potential models (Table 5-9). Models were developed to predict daily emissions
per meter squared (m2) of surface area of the open source.

Table 5-9. Parameter combinations tested as open source models for NH3 and H2S

emissions.

Model



Parameters



LB-1

Lagoon pH, lagoon temperature

LB-2

Lagoon pH

5-9


-------
Deliberative, draft document - Do not cite, quote, or distribute

LB-3

Lagoon temperature

LB-4

Ambient temperature, wind speed

LB-5

Ambient temperature

LB-6

Wind speed

LB-7

Lagoon pH, lagoon temperature, ambient temperature, wind speed

LB-8

Lagoon pH, lagoon temperature, ambient temperature

LB-9

Lagoon pH, lagoon temperature, wind speed

LB-10

Lagoon pH, ambient temperature, wind speed

LB-11

Lagoon temperature, ambient temperature, wind speed

LB-12

Lagoon pH, ambient temperature

LB-13

Lagoon pH, wind speed

LB-14

Lagoon temperature, ambient temperature

LB-15

Lagoon temperature, wind speed

For NH3, of the 15 models tested, only LB-3, LB-5, LB-6, and LB-15 were comprised of
significant parameters (Appendix G, Table G-34). The model fit (-2 log likelihood, AIC, AICc,
and BIC) and evaluation statistics (ME, NME, MB, NMB) for these models are presented in
Appendix G, Table G-35, and were consistent across the models with significant terms. This
suggests that models with ambient temperature (model LB-5) perform as well as models with
lagoon specific parameters (LB-3 and LB-15). Therefore, EPA selected model NV-5 (including
ambient temperature as the predictive parameter) for further analysis for NH3. The final form of
this model is presented in Table 5-10.

For FhS, of the 15 models tested, only LB-3, LB-5, and LB-6 were comprised entirely of
significant parameters (Appendix G, Table G-36). The model fit and evaluation statistics
(Appendix G, Table G-37), and were consistent across the models with significant terms. This
suggests that models with ambient temperature (model LB-5) perform as well as models with
lagoon specific parameters (LB-3). Therefore, EPA selected model NV-5 (including ambient
temperature as the predictive parameter) for further analysis for FhS. The final form of this
model is presented in Table 5-10.

Table 5-10. Selected daily models for lagoons sources.







Equation

NHs

ln(NH3) = 1.396734 + 0.027201 * AmbT

kg/d m2

Equation 15

h2s

ln(H2S) = 1.189272 + 0.010557 * AmbT

kg/d m2

Equation 16

5-10


-------
Deliberative, draft document - Do not cite, quote, or distribute

5.5 Corrals

The literature review (Section 3) and exploratory data analysis (Section 4) suggested that
EPA should consider ambient temperature, ambient relative humidity, water vapor deficit, and
wind speed in the development of the emission models for corrals. EPA tested 15 combinations
of these parameters as potential models (Table 5-11). Models were developed to predict daily
emissions per meter squared (g/d-m2) of surface area of the corral, as well as emissions per m2
per 1,000 head (g/d-m2-l,000 hd), to account for the stock density of the corral. In total, 30
models were tested to account for the 15 different parameter combinations and two forms of the
emissions.

Table 5-11. Parameter combinations tested as corral models for NH3 and H2S

emissions.

Model

CR-la

Emissions

g/d-m2

Daramatarc

Ambient temperature, ambient relative humidity, wind speed, water vapor deficit

CR-2a

g/d-m2

Ambient temperature, ambient relative humidity, water vapor deficit

CR-3a

g/d-m2

Ambient temperature, ambient relative humidity, wind speed

CR-4a

g/d-m2

Ambient relative humidity, wind speed, water vapor deficit

CR-5a

g/d-m2

Ambient temperature, wind speed, water vapor deficit

CR-6a

g/d-m2

Ambient temperature, ambient relative humidity

CR-7a

g/d-m2

Ambient temperature, water vapor deficit

CR-8a

g/d-m2

Ambient relative humidity, water vapor deficit

CR-9a

g/d-m2

Ambient temperature, wind speed

CR-lOa

g/d-m2

Ambient relative humidity, wind speed

CR-lla

g/d-m2

Wind speed, water vapor deficit

CR-12a

g/d-m2

Ambient temperature

CR-13a

g/d-m2

Ambient relative humidity

CR-14a

g/d-m2

Water vapor deficit

CR-15a

g/d-m2

Wind speed

CR-lb

g/d-m2-l,000 hd

Ambient temperature, ambient relative humidity, wind speed, water vapor deficit

CR-2b

g/d-m2-l,000 hd

Ambient temperature, ambient relative humidity, water vapor deficit

CR-3b

g/d-m2-l,000 hd

Ambient temperature, ambient relative humidity, wind speed

CR-4b

g/d-m2-l,000 hd

Ambient relative humidity, wind speed, water vapor deficit

CR-5b

g/d-m2-l,000 hd

Ambient temperature, wind speed, water vapor deficit

CR-6b

g/d-m2-l,000 hd

Ambient temperature, ambient relative humidity

CR-7b

g/d-m2-l,000 hd

Ambient temperature, water vapor deficit

CR-8b

g/d-m2-l,000 hd

Ambient relative humidity, water vapor deficit

CR-9b

g/d-m2-l,000 hd

Ambient temperature, wind speed

CR-lOb

g/d-m2-l,000 hd

Ambient relative humidity, wind speed

CR-llb

g/d-m2-l,000 hd

Wind speed, water vapor deficit

CR-12b

g/d-m2-l,000 hd

Ambient temperature

CR-13b

g/d-m2-l,000 hd

Ambient relative humidity

CR-14b

g/d-m2-l,000 hd

Water vapor deficit

CR-15b

g/d-m2-l,000 hd

Wind speed

5-11


-------
Deliberative, draft document - Do not cite, quote, or distribute

Models CR-3a, CR-4a, CR-6a, CR-8a, CR-12a, CR-13a, CR-14a, CR-4b, CR-6b, CR-8b,
CR-12b, CR-13b, CR-14b, CR-15b were comprised of significant parameters for NH3 (Appendix
G, Table G-38). The model fit (-2 log likelihood, AIC, AICc, and BIC) and evaluation statistics
(ME, NME, MB, NMB) for these models are presented in Appendix G, Table G-39, and were
consistent across all the models. The models predicting the emissions in g/d-m2-1,000 hd have
lower mean bias and mean error values than their counterpart predicting emissions as g/d-m2.
EPA selected model CR-3b (including the predictive parameters: ambient temperature, ambient
relative humidity, and wind speed) for further analysis for NH3. The final form of this model is
presented in Table 5-12.

For H2S, only model CR-13a was comprised entirely of statistically significant
parameters (Appendix G, Table G-40). Like NH3, the model fit and evaluation statistics
(Appendix G, Table G-41) for the version of the model predicting emissions as g/d-m2-l,000 hd
(i.e., CR-13b) has slightly lower mean bias and mean error values. EPA selected model CR-13b
(including the predictive parameter ambient relative humidity) for further analysis for corral H2S
emissions. The final form of this model is presented in Table 5-12.

Table 5-12. Selected daily models for corrals.

Pollutant

Formula

Equation
Units Number ;

NHs

ln(NH3) = 1.053805 + 0.004993 * AmbT + 0.0031 * AmbRH
+ 0.017832 * WindSpeed

g/d-m2-
1,000 hd

Equation 17

h2s

ln(H2S) = 2.404792 + 0.007177 * AmbRH

g/d-m2-
1,000 hd

Equation 18

5-12


-------
Deliberative, draft document - Do not cite, quote, or distribute

6 MODEL COEFFICIENT EVALUATION

To ensure reliable prediction of the emissions, the model coefficients were evaluated with
the jackknife method (Christensen et al., 2016; Leeden et al., 2008), which examined the
cumulative effect on coefficient estimates of multiple "minus-one" runs. The jackknife approach
called for removing one of the independent sample units from the dataset. For NAEMS, the
individual barns at each site and the lagoons are the mutually exclusive independent sample
units. EPA then determined the associated parameter estimates for the selected model based on
this dataset. This was repeated for each of the sample units. These results were then compared to
the model coefficients based on the full dataset (full model). For each jackknife model, the ME,
NME, MB, and NMB were calculated, based on the equations outlined in Section 6 of the main
report, to facilitate comparison.

EPA also prepared plots showing the variation in coefficients and standard errors for the
selected models and compared to each of the jackknife models. EPA interpreted these plots
similar to Tukey confidence interval plots in that if the result for the jackknife model overlapped
the results for the full model (i.e., the area highlighted in gray on the figures), then the model
coefficients are not inconsistent with one another. If the omission of one monitoring unit (e.g., a
barn or lagoon) resulted in a coefficient that was outside ± 1 standard error of the full model, the
sample unit was reviewed to determine if a specific characteristic of that unit (e.g., animal
placement strategy, manure handling system) might have caused the inconsistency. If the
difference could not be ascribed to an operational characteristic of the unit, the data were
reviewed for outliers that could be removed from analysis, and other potential remediation
measures considered.

6.1 Mechanically Ventilated Barns Model

6.1.1 NH3 Model Evaluation

Table 6-1 and Figure 6-1 show the variation in coefficients and standard errors for the
selected model ("None") and each of the jackknife models. The model coefficients from the
jackknife approach were comparable across the withheld sets (Table 6-1) and remained
significant (p-value <0.05) across all models. The plots in Figure 6-1 show that the results for all
jackknife models overlap the full model estimate ± 1 standard error, except for ambient
temperature. In comparison to the full model, that is where the barn removed is "None", the
maximum percent differences for parameter estimates across the three models were 7%, 23%,
3%, and 4% for inventory, ambient temperature, intercept for the flush barns, and intercept for
scrape barns, respectively. Across all models, the difference in NME and NMB (Table 6-2) in
comparison to the selected model were minor. For NME the values differed by less than 8%. For
NMB the values varied by less than 34%. The largest difference was seen when WI5B B1 was
withheld from the dataset, which decreased the NME and NMB by 8% and 34%, respectively.

6-1


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 6-1. Model coefficients developed using the jackknife approach for NH3
emissions from mechanically ventilated barns.

Barn out

Effect

Estimate

Standard Error

p-value

NONE

Inventory

1.773832

0.06477

<.0001

Ambient Temperature

0.029586

0.00088

<.0001

Flush

1.746585

0.03789

<.0001

Scrape

1.864935

0.04253

<.0001

IN5BB1

Inventory

1.736301

0.07221

<.0001

Ambient Temperature

0.024312

0.00093

<.0001

Flush

1.793836

0.03772

<.0001

Scrape

1.925841

0.04232

<.0001

IN5BB2

Inventory

1.898712

0.07457

<.0001

Ambient Temperature

0.024229

0.00091

<.0001

Flush

1.748491

0.03749

<.0001

Scrape

1.869675

0.0425

<.0001

NY5BB1

Inventory

1.824003

0.06932

<.0001

Ambient Temperature

0.030506

0.00095

<.0001

Flush

1.72461

0.03966

<.0001

Scrape

1.798078

0.04787

<.0001

WI5BB1

Inventory

1.722238

0.07977

<.0001

Ambient Temperature

0.036382

0.00101

<.0001

Flush

1.693687

0.05244

<.0001

Scrape

1.832478

0.05634

<.0001

WI5BB2

Inventory

1.703501

0.07134

<.0001

Ambient Temperature

0.032999

0.00105

<.0001

Flush

1.765095

0.04896

<.0001

Scrape

1.891018

0.05005

<.0001

Table 6-2. Model fit statistics for the mechanically ventilated barns NH3 jackknife.

Barn out

n

LNMEa (%)

NMEb(%) I MEb(kgday') MBb(kgday') IMMBb (%)

Corr

NONE

2192

7.322

24.573

5.959

-0.583

-2.404

0.917

IN5BB1

1771

7.213

25.072

5.003

-0.542

-2.717

0.911

IN5BB2

1762

7.148

25.329

5.042

-0.472

-2.372

0.905

NY5BB1

1846

7.403

24.716

6.115

-0.701

-2.835

0.924

WI5BB1

1676

6.866

22.488

6.538

-0.459

-1.579

0.918

WI5BB2

1713

7.212

23.375

6.523

-0.547

-1.961

0.919

a Based on transformed data (i.e., ln(NH3)).
b Based on back-transformed data.

6-2


-------
Deliberative, draft document - Do not cite, quote, or distribute

Inventory

-f- ^ - f 3

0.040
0.038 -
0.036
0.034 -
0.032 -
0.030 -
0.028 -
0.026 -
0.024 -
0.022
0.020

Ambient Temperature

WI5BB2 WI5BB1 NY5BB1 IN5BB2 IN5BB1

Flush

Scrape

J..OD "

1 80 -















1.75 -



























1.70 -
1.65 -
1.60 -













1.50 -



2.000
1.950
1.900
1.850
1.800
1.750
1.700
1.650
1.600

WI5BB2 WI5BB1 NY5BB1 IN5BB2 IN5BB1

Figure 6-1. Comparison of variation in coefficients and standard errors for NH3 mechanically
ventilated barn model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected NH3 mechanically ventilated model coefficient ("None", gray band for ± SE) for each
model parameter.

6.1.2 H2S Model Evaluation

The variation in coefficients and standard errors for the selected model ("None") and
each of the H2S jackknife models is shown in Table 6-3 and Figure 6-2. The model coefficients
from the jackknife approach were comparable across the withheld sets (Table 6-3) and remained
significant (p-value <0.05) across all models. The plots in Figure 6-2 show that the results for all
jackknife models overlap the full model estimate ± 1 standard error, except for WI5B B1 for
ambient temperature. In comparison to the full model, where the barn removed is "None", the
maximum percent differences for parameter estimates across the three models were 14%, 26%,
2%, and 1% for inventory, ambient temperature, intercept for the flush barns, and intercept for
scrape barns, respectively. Across all models, the difference in NME and NMB (Table 6-4) in
comparison to the selected model were minor for NME (< 8%) and more substantial for NMB
(<32%).

6-3


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 6-3. Model coefficients developed using the jackknife approach for H2S
emissions from mechanically ventilated barns.

Barn out

Effect

Estimate

Standard Error

p-value

NONE

Inventory

0.86173

0.08664

<.0001

Ambient Temperature

0.012786

0.00127

<.0001

Flush

7.406887

0.05129

<.0001

Scrape

6.287004

0.05691

<.0001

IN5BB1

Inventory

0.974345

0.08989

<.0001

Ambient Temperature

0.010264

0.00134

<.0001

Flush

7.389176

0.04755

<.0001

Scrape

6.282462

0.053

<.0001

IN5BB2

Inventory

0.73697

0.09126

<.0001

Ambient Temperature

0.010959

0.00124

<.0001

Flush

7.453061

0.04624

<.0001

Scrape

6.355244

0.0521

<.0001

NY5BB1

Inventory

0.915728

0.09384

<.0001

Ambient Temperature

0.012973

0.00147

<.0001

Flush

7.389581

0.05383

<.0001

Scrape

6.222805

0.06537

<.0001

WI5BB1

Inventory

0.897494

0.11836

<.0001

Ambient Temperature

0.016059

0.00149

<.0001

Flush

7.285544

0.07955

<.0001

Scrape

6.224063

0.08308

<.0001

WI5BB2

Inventory

0.817846

0.10259

<.0001

Ambient Temperature

0.014378

0.00148

<.0001

Flush

7.495271

0.07179

<.0001

Scrape

6.313356

0.07154

<.0001

Table 6-4. Model fit statistics for the mechanically ventilated barns H2S jackknife.

Barn out

n

LNMEa (%)

NMEb(%) I ME1'(g day ') MBb(gday') NMBb (%)

Corr

NONE

2454

4.46

64.308

553.14

-38.66

-4.495

0.58

IN5BB1

1993

4.088

61.644

533.71

-34.72

-4.01

0.592

IN5BB2

1954

3.911

59.42

464

-25.36

-3.248

0.677

NY5BB1

1992

4.736

65.587

615.71

-39.17

-4.173

0.565

WI5BB1

1920

4.696

66.693

561.9

-47.91

-5.686

0.543

WI5BB2

1957

4.653

64.785

564.15

-51.6

-5.925

0.582

a Based on transformed data (i.e., ln(H2S)).
b Based on back-transformed data.

6-4


-------
Deliberative, draft document - Do not cite, quote, or distribute

1.10 -|

1.00 -
0.90 -
0.80 -
0.70 -
0.60 -
0.50 -
0.40 -

7.60
7.50 -
7.40 -
7.30 -
7.20 -
7.10 -
7.00 -

Figure 6-2. Comparison of variation in coefficients and standard errors for H2S mechanically
ventilated barn model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected H2S mechanically ventilated barn model coefficient ("None", gray band for ± SE) for
each model parameter.

6.1.3 Particulate Matter Models

Particulate matter models were not selected at this time.

6.2 Milking Centers
6.2.1 NH3 Model Evaluation

Table 6-5 and Figure 6-3 show the variation in coefficients and standard errors for the
selected model ("None") and each of the jackknife models. The model coefficients from the
jackknife approach were comparable across the withheld sets (Table 6-5) and remained
significant (p-value <0.05) across all models. The plots in Figure 6-3 show that the results for all
jackknife models do not overlap the full model estimate ± 1 standard error. The standard error
was very small for the full model, where the Barn removed is "None", which prevented the
overlap. In comparison to the full model, the maximum percent differences for parameter
estimates across the two models were 29% and 44% for the intercept and ambient temperature,
respectively. Across all models, the difference in NME and NMB (Table 6-6) in comparison to
the selected model were substantial for NME and NMB, with values differing by up to 44% and
104%, respectively. Upon further review, it was determined that the MCs utilize different
manure handling techniques. Specifically, IN5B used a flush system while NY5B used a scrape
system. Additional models using this distinction will be tested for the final report.

Inventory

BB2 WI5BB1 NY5BB1 IN5BB2 IN5BB1 NONE

0.020
0.018 -
0.016 -
0.014
0.012 -
0.010 -
0.008

Ambient Temperature

L

n:

WI5BB2 W15BB1 NY5BB1 IN5BB2 IN5BB1

Flush

i	£.

BB2 WI5BB1 NY5BB1 IN5BB2 IN5BB1 NONE

6.450 1
6.400 -
6.350 -
6.300 -
6.250 -
6.200 -
6.150 -
6.100 -
6.050 -
6.000 -

Scrape

WI5BB2 WI5BB1

6-5


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 6-5. Model coefficients developed using the jackknife approach for NH3

emissions from milking centers.

Site out

Effect

Estimate

Standard Error

p-value

NONE

Intercept

2.505637

0.10119

<.0001

Ambient Temperature

0.046434

0.00335

<.0001

IN5BMC

Intercept

3.155214

0.06261

<.0001

Ambient Temperature

0.026195

0.00297

<.0001

NY5BMC

Intercept

1.783938

0.09766

<.0001

Ambient Temperature

0.064815

0.0051

<.0001

Table 6-6. Model fit statistics for the milking center NH3 jackknife.

c •* *
bltG OUt



INMpi 10/\

NMEb (%)

MEb (kg day"1)

MBb (kg day *)

NMBb (%) I Corr

NONE

713

18.245

54.184

12.63

3.017

12.941

0.364

IN5BMC

376

8.032

30.564

9.232

1.475

4.884

0.264

NY5BMC

337

16.728

43.666

6.819

-0.088

-0.561

0.706

a Based on transformed data (i.e., ln(NH3)).
b Based on back-transformed data.

intercept

Ambient Temperature

3.50

3.00

2.50 -

2.00

1.50

1.00

8.50

0.00

Figure 6-3. Comparison of variation in coefficients and standard errors for NH3 milking center
model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected NH3 for milking center model coefficient ("None", gray band for ± SE) for each model
parameter.

6.2.2 H2S Model Evaluation

Table 6-7 and Figure 6-4 show the variation in coefficients and standard errors for the
selected H2S MC model ("None") and each of the jackknife models. The model coefficients from
the jackknife approach were comparable across the withheld sets (Table 6-7) and remained
significant (p-value <0.05) across all models. The plots in Figure 6-4 show that the results for all
jackknife models do not overlap the full model estimate ± 1 standard error, except the intercept
for the IN5B withheld model. Like the NH3 model, the standard error was very small for the full
model, where the Barn removed is "None", which prevented the overlap. In comparison to the
full model, the maximum percent differences for parameter estimates across the two models were
4% and 120% for the intercept and ambient temperature, respectively. Across all models, the

6-6


-------
Deliberative, draft document - Do not cite, quote, or distribute

difference in NME and NMB (Table 6-8) in comparison to the selected model were substantial
for NME and NMB, with values differing by less than 32% and 79%, respectively. As with the
NH3 models, adding a parameter for manure management system may account for the variability
between sites. Additional models using this distinction will be tested for the final report.

Table 6-7. Model coefficients developed using the jackknife approach for H2S

emissions from milking centers.

Site out

Effect

Estimate

Standard Error

p-value

NONE

Intercept

6.898188

0.07052

<.0001

Ambient Temperature

0.024053

0.00361

<.0001

IN5BMC

Intercept

6.99747

0.05042

<.0001

Ambient Temperature

0.006415

0.0025

0.011

NY5BMC

Intercept

6.621331

0.13313

<.0001

Ambient Temperature

0.052894

0.00711

<.0001

Table 6-8. Model fit statistics for the milking center H2S jackknife.

Site out

n

LNMEa (%)

NMEb (%)

MEb (g day"1)

MBb (g day"1)

NMBb (%)

Corr

NONE

926

6.611

90.97

1204.3

-113.5

-8.571

0.347

IN5BMC

540

4.099

61.55

413.65

-12.28

-1.827

0.466

NY5BMC

386

8.707

84.8

1895.8

-284.9

-12.74

0.448

a Based on transformed data (i.e., ln(H2S)).
b Based on back-transformed data.

1 ntercept

Ambient Temperature

7.00







6.90 ¦







0.06 -











0.05 ¦



6.80





0.04 -



6.60 ¦





0.03 -



6.50









6.40 ¦



0.02 -
0.01 ¦



6.30 •
6.20





•

NY5BMC IN5BMC NONE



NY5BMC IN5BMC NONE

Figure 6-4. Comparison of variation in coefficients and standard errors for H2S milking center

model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected H2S milking center model coefficient ("None", gray band for ± SE) for each model
parameter.

6.2.3 Particulate Matter Model Evaluation

For the MC particulate matter models, we did not complete jackknife analysis because
there was only one site in the dataset. We also did not pursue a model evaluation using a k-fold
cross validation technique based on previous SAB comments (SAB, 2013) recommending
against using this method to select data for temporally correlated data. Future EPA efforts will

6-7


-------
Deliberative, draft document - Do not cite, quote, or distribute

investigate obtaining additional data that would allow for further model testing and evaluation
and an improved emission model.

6.3 Naturally Ventilated Barn Model

A theme across all the results presented below is withholding WA4B B4 from the data set
produces the largest differences across the models. This is likely due to WA4B B4 having an
average daily inventory almost twice the other three barns included in NAEMS. Removing this
barn greatly reduced the variability of inventory values in the data set that the model must
capture.

6.3.1 NH3 Model Evaluation

Table 6-9 and Figure 6-5 show the variation in coefficients and standard errors for the
selected NH3 naturally ventilated barn model ("None") and each of the jackknife models. The
model coefficients from the jackknife approach had some differences, most notable in the models
with WA5B barns withheld (Table 6-9). For the models where WA4B B2 and B4 were withheld,
one or both parameters were insignificant (p-value >0.05). The plots in Figure 6-5 show that the
coefficients for these models also fall outside the full model estimate ± 1 standard error, except
for wind speed. In comparison to the full model, where the barn removed is "None", the
maximum percent differences for parameter estimates across the models were 2292%, 235%, and
23% for the intercept, inventory, and wind speed, respectively. These largest differences all
occurred for the model where WA5B B4 was removed. Across all models, the difference in
NME and NMB (Table 6-10) in comparison to the selected model were the largest when WA5B
B4 was withheld from the dataset, which increased the NME by 32% and decreased NMB by
174%). This is likely due to the reduced variability in inventory values caused by withholding
WA4B B4.

6-8


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 6-9. Model coefficients developed using the jackknife approach for NH3
emissions from naturally ventilated barns.

Site out

Effect

Estimate

Standard Error

p-valuea

NONE

Intercept

0.188357

0.2678

0.484

Inventory

3.451939

0.4106

<.0001

Wind Speed

0.048153

0.01837

0.009

CA5BB1

Intercept

0.734625

0.34491

0.0385

Inventory

2.885717

0.49667

<.0001

Wind Speed

0.043071

0.01873

0.022

CA5BB2

Intercept

0.730143

0.31533

0.0253

Inventory

2.985909

0.45768

<.0001

Wind Speed

0.040555

0.01847

0.0288

WA5BB2

Intercept

-0.84424

0.13064

<.0001

Inventory

4.709923

0.19931

<.0001

Wind Speed

0.019312

0.02201

0.3808

WA5BB4

Intercept

4.505901

1.29423

0.0009

Inventory

-4.658465

2.41694

0.0582

Wind Speed

0.037293

0.02361

0.1149

aBold indicates insignificant p-values (i.e., > 0.05)

Table 6-10. Model fit statistics for the naturally ventilated barns NH3 jackknife.

Site out

11

LNMEa (%)

NME (%)





NIVIB (%)



NONE

605

27.084

75.233

12.818

0.828

4.862

0.636

CA5BB1

431

27.885

72.445

16.265

0.754

3.36

0.601

CA5BB2

396

25.139

69.96

16.995

1.728

7.114

0.599

WA5BB2

482

20.19

51.412

7.179

-0.504

-3.611

0.793

WA5BB4

506

32.404

98.929

9.575

-0.249

-2.571

0.207

a Based on transformed data (i.e., ln(NH3)).
b Based on back-transformed data.

6-9


-------
Deliberative, draft document - Do not cite, quote, or distribute

Intercept

7.00
6.00 -
5.00 -

ii

4.00 -
3.00 -
2.00 -

D.00 -		!

-1.00 -	»

-2.00 -I	,	1	1	1	

WA5BB4	WA5BB2	CA5BB2	CA5BB1	NONE

Wind Speed

0.07
0.06 -

0.04 -	II	II

0.03 -
0.02 ¦

0.01 ¦

0.0C' -I	1	1	1	1	

WA5BB4	WA5BB2	CA5BB2	CA5BB1	NONE

Figure 6-5. Comparison of variation in coefficients and standard errors for NH3 naturally ventilated
barn model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected NH3 naturally ventilated barn model coefficient ("None", gray band for ± SE) for each
model parameter.

6.3.2 H2S Model Evaluation

Table 6-11 and Figure 6-6 show the variation in coefficients and standard errors for the
selected H2S naturally ventilated barn model ("None") and each of the jackknife models. The
model coefficients from the jackknife approach had some differences, most notable the
coefficient for inventory switched to negative in the model with WA5B B4 withheld (Table
6-11) and was insignificant (p-value >0.05). For the models where CA4B B1 and B2 were
withheld, the coefficient from wind speed became insignificant. The plots in Figure 6-6 show
that the coefficients for the model where WA5B B4 was withheld fall outside the full model
estimate ± 1 standard error, except for wind speed. In comparison to the full model, where the
barn removed is "None", the maximum percent differences for parameter estimates across the
models occurred when WA5B was withheld and were 12%, 307%, and 75% for the intercept,
inventory, and wind speed, respectively. Across all models, the difference in NME and NMB
(Table 6-12) in comparison to the selected model were the largest when WA5B B4 was withheld
from the dataset, which increased the NME by 17% and decreased NMB by 92%. Withholding
WA4B B4 from the dataset reduced variability in inventory, which changed the significance of
inventory as a predictive parameter and lowered the bias seen in the model.

6-10


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 6-11. Model coefficients developed using the jackknife approach for H2S
emissions from naturally ventilated barns.

Site out

Effect

Estimate

Standard Error

p-valuea

NONE

Intercept

6.541057

0.14434

<.0001

Inventory

0.587702

0.21921

0.008

Wind Speed

0.062678

0.02193

0.0044

CA5BB1

Intercept

6.593149

0.17451

<.0001

Inventory

0.661236

0.24717

0.0083

Wind Speed

0.036373

0.02762

0.1886

CA5BB2

Intercept

6.557214

0.18007

<.0001

Inventory

0.6616

0.24813

0.0085

Wind Speed

0.03755

0.03114

0.2288

WA5BB2

Intercept

6.559682

0.14376

<.0001

Inventory

0.520217

0.21815

0.0182

Wind Speed

0.075574

0.02381

0.0016

WA5BB4

Intercept

7.344257

0.58948

<.0001

Inventory

-1.214405

1.08122

0.2645

Wind Speed

0.109848

0.01931

<.0001

aBold indicates insignificant p-values (i.e., > 0.05)

Table 6-12. Model fit statistics for the naturally ventilated barns H2S jackknife.

Site out

n

LNMEa (%)

NMEb (%)

MEb (g day"1) | MB^kgday1)

NMBb (%)

Corr

NONE

647

6.461

77.092

677.49

-29.02

-3.302

0.33

CA5BB1

449

6.937

80.862

807.4

-34.82

-3.487

0.326

CA5BB2

380

7.784

89.878

915.9

-39.6

-3.886

0.32

WA5BB2

550

5.832

69.934

603.45

-36.4

-4.218

0.371

WA5BB4

562

5.662

69.734

490.88

-1.791

-0.254

0.249

a Based on transformed data (i.e., ln(H2S)).
b Based on back-transformed data.

6-11


-------
Deliberative, draft document - Do not cite, quote, or distribute

Intercept

Inventory





1.00 -



8.00 -







1 i . I



I 1 1 1







-0.50 -

wa:

BB4 WA5BB2 CA5BB2 CA5BB1 NONE

7.00 -











6.50 ¦



	I	I	*		I

-1.50 -





























Wind Speed

0.14
0.12 -















0.08 -





















0.06 -
0 04 -













0.02 -
0.00 -

















WA5BB4	WA5BB2	CA5BB2	CA5BB1	NONE

Figure 6-6. Comparison of variation in coefficients and standard errors for H2S naturally ventilated
barn model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected H2S naturally ventilated barns model coefficient ("None", gray band for ± SE) for each
model parameter.

6.3.3 PM10 Model Evaluation

Table 6-13 and Figure 6-7 show the variation in coefficients and standard errors for the
selected PM10 naturally ventilated barn model ("None") and each of the jackknife models. The
model coefficients from the jackknife approach had some differences, most notably the
coefficient for inventory switched to negative in the model with WA5B B4 withheld (Table
6-13) and became insignificant. For the models where WA4B4 was withheld, the coefficient for
ambient temperature also became insignificant (p-value >0.05). The plots in Figure 6-7 show that
the coefficients for the model where WA5B B4 fall outside the full model estimate ± 1 standard
error, except for ambient relative humidity. In comparison to the full model, where the barn
removed is "None", the maximum percent differences for parameter estimates across the three
models were 15%, 138%, 80%, 24%, and 20% for the intercept, inventory, ambient temperature,
ambient relative humidity, and wind speed, respectively. Across all models, the difference in
NME and NMB (Table 6-14) in comparison to the selected model were the largest when WA5B
B4 was withheld from the dataset, which increased the NME by 16% and decreased NMB by
37%.

6-12


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 6-13. Model coefficients developed using the jackknife approach for PM10
emissions from naturally ventilated barns.

Site out

Effect

Estimate

Standard Error

p-valuea

NONE

Intercept

7.64258

0.16783

<.0001

Inventory

1.525009

0.14917

<.0001

Ambient Temperature

0.011864

0.00333

0.0004

Ambient Relative Humidity

-0.01521

0.00154

<.0001

Wind Speed

0.173698

0.01064

<.0001

CA5BB1

Intercept

7.695149

0.18357

<.0001

Inventory

1.399494

0.16322

<.0001

Ambient Temperature

0.018588

0.00384

<.0001

Ambient Relative Humidity

-0.01564

0.00178

<.0001

Wind Speed

0.181527

0.0118

<.0001

CA5BB2

Intercept

7.726456

0.19289

<.0001

Inventory

1.420078

0.16427

<.0001

Ambient Temperature

0.014917

0.00397

0.0002

Ambient Relative Humidity

-0.015634

0.00196

<.0001

Wind Speed

0.175816

0.01265

<.0001

WA5BB2

Intercept

6.831711

0.24796

<.0001

Inventory

2.045075

0.17514

<.0001

Ambient Temperature

0.020629

0.00419

<.0001

Ambient Relative Humidity

-0.0115

0.00199

<.0001

Wind Speed

0.192966

0.01355

<.0001

WA5BB4

Intercept

8.81874

0.46389

<.0001

Inventory

-0.576586

0.90282

0.5241

Ambient Temperature

0.002425

0.00354

0.494

Ambient Relative Humidity

-0.012854

0.00154

<.0001

Wind Speed

0.138497

0.01071

<.0001

aBold indicates insignificant p-values (i.e., > 0.05)

Table 6-14. Model fit statistics for the naturally ventilated barns PM10 jackknife.

Site out



LNMEa (%)



MEb (g day"1)

MBb (kg day1)

NMBb (%) I Corr

CA5BB1

1214

5.102

79.404

4772.9

-701.9

-11.68

0.372

CA5BB2

1088

5.412

81.443

5265.9

-688.8

-10.65

0.358

NONE

1457

4.896

82.575

4195.9

-668.8

-13.16

0.374

WA5BB2

1024

4.537

76.692

3944.7

-926.4

-18.01

0.462

WA5BB4

1045

4.156

95.397

2384

-277.5

-11.1

0.208

a Based on transformed data (i.e., In(PMio)).
b Based on back-transformed data.

6-13


-------
Deliberative, draft document - Do not cite, quote, or distribute

Intercept

Inventory

9.50
9.00
8.50
8.00
7.50
7.00
6.50
6.00

-4-

WA5BB4	WA5BB2	CA5BB2	CA5

Ambient Temperature

2.50
2.00
1.50
1.00
0.50
0.00
-0.50
-1.00
-1.50
-2.00

Ambient Relative Humidity

0.03
0.02
0.02

o.oi
o.oi

0.00
-0.01

WA5BB4	WA5BB2	CA5BB2	CA5BB1

WA5BB4	WA5BB2	CA5BB2	CA5BB1	NONE

Wind Speed



Figure 6-7. Comparison of variation in coefficients and standard errors for PM10 naturally ventilated
barn model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected PMio naturally ventilated barns model coefficient ("None", gray band for ± SE) for each
model parameter.

6.3.4 PM2.5 Model Evaluation

The analysis for the PM2.5 naturally ventilated barns was a departure from the other
evaluations, more of the models have coefficients that vary and are insignificant (Table 6-15).
When compared to the full model, the coefficients vary up to 125%, 4,370%, 406%, 21,410%,
and 25% for the intercept, inventory, ambient temperature, ambient relative humidity, and wind
speed, respectively, and the large differences are not limited to the model with WA5B B4
withheld. Table 6-15 and Figure 6-8 show the variation in coefficients and standard errors for the
selected PM2.5 naturally ventilated barn model ("None") and each of the jackknife models. The
plots in Figure 6-8 show that most of the coefficients for the models overlapped the full model

6-14


-------
Deliberative, draft document - Do not cite, quote, or distribute

estimate ± 1 standard error. The models for the WA5B barn both fell outside for the intercept and
inventory, and the WA5B B1 model fell outside for ambient relative humidity. The difference in
NME and NMB (Table 6-16) across the models with a barn withheld compared to the selected
model changed by as much as 40% for NME and 1,566% for NMB.

Table 6-15. Model coefficients developed using the jackknife approach for PM2.5
emissions from naturally ventilated barns.

Site out

Effect

Estimate

Standard Error

p-valuea

NONE

Intercept

7.068797

1.15954

<.0001

Inventory

-0.220453

0.75959

0.7753

Ambient Temperature

0.01121

0.02585

0.6681

Ambient Relative Humidity

-0.003808

0.01023

0.7125

Wind Speed

0.218968

0.0563

0.0002

CA5BB1

Intercept

6.922323

1.15234

<.0001

Inventory

-0.432386

0.76218

0.579

Ambient Temperature

0.015697

0.02584

0.5493

Ambient Relative Humidity

0.001448

0.01082

0.8946

Wind Speed

0.232037

0.05911

0.0002

CA5BB2

Intercept

5.999344

0.97451

<.0001

Inventory

-0.637279

0.60064

0.3062

Ambient Temperature

0.056741

0.02418

0.0293

Ambient Relative Humidity

0.012843

0.00944

0.1876

Wind Speed

0.237943

0.06181

0.0002

WA5BB2

Intercept

-1.742952

1.50484

0.2592

Inventory

4.220142

0.79698

<.0001

Ambient Temperature

0.135315

0.02619

<.0001

Ambient Relative Humidity

0.049877

0.01071

0.0001

Wind Speed

0.221498

0.0743

0.0044

WA5BB4

Intercept

13.01778

2.71873

0.0035

Inventory

-9.854431

5.35402

0.1099

Ambient Temperature

-0.005191

0.0234

0.8255

Ambient Relative Humidity

-0.012329

0.00844

0.1545

Wind Speed

0.163688

0.02852

<.0001

aBold indicates insignificant p-values (i.e., > 0.05)

Table 6-16. Model fit statistics for the naturally ventilated barns PM2.5 jackknife.

Site out



LNMEa (%)

NMEb (%)

MEb (g day *)

MBb (kg day*)

NMBb (%)

Corr

CA5BB1

89

8.295

59.345

1154

9.362

0.481

0.651

CA5BB2

78

6.288

37.718

820.71

50.306

2.312

0.821

NONE

93

8.789

62.65

1167

-19.48

-1.046

0.665

WA5BB2

56

5.461

48.197

625.08

198.89

15.335

0.901

WA5BB4

56

5.877

54.701

1018.8

-91.41

-4.908

0.718

a Based on transformed data (i.e., ln(NH3)).

b Based on back-transformed data.

6-15


-------
Deliberative, draft document - Do not cite, quote, or distribute

Intercept

Inventory

JT		4

WA5BB4 WA5BB2 CA5BB2 CA5BB1 NONE

5.00 -
O.OO ¦

—i—



























20.00 -



WA5BB4 WA5BB2 CA5BB2

Ambient Temperature

WA5BB4 WA5BB2 CA5BB2 CA5BB1 NONE

0.070
0.060
0.050
0.040
0.030
0.020
0.010
0.000
-0.010
-0.020
-0.030

Ambient Relative Humidity

.1..

Wind Speed

0.350
0.300 -
0.750 -
0.200 -
0.150 -
0.100 -
0.050 -
0.000 -

WA5BB4 WA5BB2 CA5BB2 CA5BB1	NONE

Figure 6-8. Comparison of variation in coefficients and standard errors for PM2.5 naturally ventilated
barn model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected PM2.5 naturally ventilated barn model coefficient ("None", gray band for ± SE) for each
model parameter.

6.3.5 TSP Model Evaluation

Table 6-17 and Figure 6-9 show the variation in coefficients and standard errors for the
selected TSP naturally ventilated barn model ("None") and each of the jackknife models. The
model coefficients from the jackknife approach were comparable across the withheld sets (Table
6-17) and remained significant (p-value <0.05) across all models, except for ambient temperature
in the model where WA5BB4 was removed. The plots in Figure 6-9 show that all the coefficients
overlap the full model estimate ± 1 standard error, except for inventory for the model where
WA5BB4 was removed. In comparison to the full model, that is where the barn removed is
"None", the maximum percent differences for parameter estimates across the three models were

6-16


-------
Deliberative, draft document - Do not cite, quote, or distribute

17%, 141%, 56%), 25%o, and 18%> for the intercept, inventory, ambient temperature, ambient
relative humidity, and wind speed, respectively. Across all models, the difference in NME and
NMB (Table 6-18) in comparison to the selected model changed by as much as 16%> for NME
and 160% for NMB.

Table 6-17. Model coefficients developed using the jackknife approach for TSP
emissions from naturally ventilated barns.

Site out

Effect

Estimate

Standard Error

p-valuea

NONE

Intercept

7.868847

0.58294

<.0001

Inventory

2.953893

0.48928

<.0001

Ambient Temperature

0.034508

0.01069

0.0021

Ambient Relative Humidity

-0.033997

0.00508

<.0001

Wind Speed

0.248191

0.04211

<.0001

CA5BB1

Intercept

7.667585

0.48937

<.0001

Inventory

2.477977

0.44054

<.0001

Ambient Temperature

0.048926

0.01002

<.0001

Ambient Relative Humidity

-0.026332

0.00445

<.0001

Wind Speed

0.294612

0.03075

<.0001

CA5BB2

Intercept

7.786063

0.68673

<.0001

Inventory

2.998098

0.56151

<.0001

Ambient Temperature

0.034621

0.01325

0.0127

Ambient Relative Humidity

-0.032651

0.00638

<.0001

Wind Speed

0.238451

0.05294

<.0001

WA5BB2

Intercept

6.616785

0.81649

<.0001

Inventory

3.762081

0.52641

<.0001

Ambient Temperature

0.048947

0.01322

0.0005

Ambient Relative Humidity

-0.026808

0.00659

0.0001

Wind Speed

0.235277

0.04912

<.0001

WA5BB4

Intercept

6.558937

1.4622

<.0001

Inventory

7.12147

2.73945

0.0131

Ambient Temperature

0.0151

0.01245

0.2317

Ambient Relative Humidity

-0.042411

0.0058

<.0001

Wind Speed

0.203451

0.05134

0.0001

aBold indicates insignificant p-values (i.e., > 0.05)

Table 6-18. Model fit statistics for the naturally ventilated barns TSP jackknife.

Site out

n

LNMEa (%)

NMEb (%)

MEb (g day"1)

MBb (kg day"1)

NMBb (%)

Corr

CA5BB1

135

4.902

44.574

9954.9

-1381

-6.185

0.875

CA5BB2

146

6.598

55.473

10927

-932.6

-4.734

0.799

NONE

205

6.07

52.783

8639.5

-492.6

-3.009

0.807

WA5BB2

167

5.659

49.037

7695.7

-297.8

-1.898

0.821

WA5BB4

167

6.446

57.093

5315

12.023

0.129

0.666

a Based on transformed data (i.e., In(TSP)).

b Based on back-transformed data.

6-17


-------
Deliberative, draft document - Do not cite, quote, or distribute

Intercept

Inventory

9.00
8.50 -
8.00 -
7.50 -







T





~

?



7.00 -









6.50 ¦
6.00 -













5.50 -
5.00 -
4.50 -













4.00 -



Ambient Temperature

0.07

0.06	-

0.05	-

0.04	-

0.03	-

0.02	-

0.01	-

0.00	-¦

Wind Speed

0.350 i
0.300 -
0.250
0.200
0.150 -
0.100 -
0.050 -
0.000

12.00 i
10.00
8.00 -
6.00
4.00
2.00
0.00

	

WA5BB4 WA5BB2 CA5BB2

Ambient Relative Humidity

0.000
-0.010 -
-0.020 -
-0.030 -
-0.040
-0.050
-0.060

	f-

~4

WA5BB4 WA5BB2 CA5BB2 CASBB1

Figure 6-9. Comparison of variation in coefficients and standard errors for TSP naturally ventilated
barn model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected TSP naturally ventilated barn model coefficient ("None", gray band for ± SE) for each
model parameter.

6.4 Open Sources

For the corral models, we did not complete jackknife analysis because there was only one
site in the dataset. We also did not pursue a model evaluation using a k-fold cross validation
technique based on previous SAB comments (SAB, 2013) recommending against using this
method to select data for temporally correlated data. Future EPA efforts will look into obtaining
additional data that would allow for further model testing and evaluation and an improved
emission model.

6-18


-------
Deliberative, draft document - Do not cite, quote, or distribute

6.4.1 NH3 Model Evaluation

Table 6-19 and Figure 6-10 show the variation in coefficients and standard errors for the
selected NH3 open source model ("None") and each of the jackknife models. The model
coefficients from the jackknife approach were comparable across the withheld sets (Table 6-19)
and remained significant (p-value <0.05) across all models. The plots in Figure 6-10 show that
the results for all jackknife models do not overlap the full model estimate ± 1 standard error,
except the model where IN5A was withheld for ambient temperature. In comparison to the full
model, the maximum percent differences for parameter estimates across the two models were
13% and 24% for the intercept and ambient temperature, respectively. Across all models, the
difference in NME and NMB (Table 6-20) in comparison to the selected model were substantial
for NME and NMB, with values differing by up to 38% and 77%, respectively.

Table 6-19. Model coefficients developed using the jackknife approach for NH3

emissions from open sources.

Site out

Effect

Estimate

Standard Error

p-value

NONE

Intercept

1.396734

0.0248

<.0001

Ambient Temperature

0.027201

0.00195

<.0001

IN5A

Intercept

1.576653

0.06521

<.0001

Ambient Temperature

0.033848

0.00616

<.0001

WI5A

Intercept

1.323888

0.01843

<.0001

Ambient Temperature

0.031531

0.00152

<.0001

Table 6-20. Model fit statistics for the open sources NH3 jackknife.

Site out

n

LNMEa (%)

NMEb (%)

MEb (g day"1)

MBb (kg day"1)

NMBb (%)

Corr

IN5A

28

12.225

53.586

0.865

-0.048

-2.958

0.84

NONE

157

9.709

38.766

0.712

-0.034

-1.859

0.821

WI5A

129

8.159

31.915

0.601

-0.008

-0.433

0.887

a Based on transformed data (i.e., ln(NH3)).
b Based on back-transformed data.



Intercept

	 0.045

Ambient Temperature







1.60 -

j 0.040 ¦





1.50 -

0.030 ¦

i









1.30 ¦

fpf 0.025 ¦
i 0.020 -







1.20 -





1.10 -

0.010 ¦
0.005 -
0 000









WI5A IN5A NONE

W15A IN5A NONE

Figure 6-10. Comparison of variation in coefficients and standard errors for NH3 open source model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected NH3 open source model coefficient ("None", gray band for ± SE) for each model
parameter.

6-19


-------
Deliberative, draft document - Do not cite, quote, or distribute

6.4.2 H2S Model Evaluation

Table 6-21 and Figure 6-11 show the variation in coefficients and standard errors for the
selected H2S open source model ("None") and each of the jackknife models. The model
coefficients from the jackknife approach were comparable across the withheld sets (Table 6-21)
and remained significant (p-value <0.05) across all models. The plots in Figure 6-11 show that
the results for all jackknife models do not overlap the full model estimate ± 1 standard error,
except the model where IN5A was withheld. In comparison to the full model, the maximum
percent differences for parameter estimates across the two models were 7% and 68% for the
intercept and ambient temperature, respectively. Across all models, the difference in NME and
NMB (Table 6-22) in comparison to the selected model were substantial for NME and NMB,
with values differing by up to 20% and 98%, respectively.

Table 6-21. Model coefficients developed using the jackknife approach for H2S

emissions from open sources.

Site out

Effect

Estimate

Standard Error

p-value

NONE

Intercept

1.189272

0.03163

<.0001



Ambient Temperature

0.010557

0.0022

<.0001

IN5A

Intercept

1.109037

0.01639

<.0001



Ambient Temperature

0.003382

0.00127

0.0203

WA5A

Intercept

1.189558

0.03019

<.0001



Ambient Temperature

0.011581

0.00218

<.0001

WI5A

Intercept

1.226774

0.04029

<.0001



Ambient Temperature

0.009725

0.00256

0.0005

Table 6-22. Model fit statistics for the open source H2S jackknife.

Site out

n

LNMEa (%)

NMEb (%)

MEb (g day1)

MBb (kg day1)

NMBb (%)

Corr

NONE

70

9.258

63.688

0.499

-0.011

-1.403

0.587

IN5A

13

1.475

76.161

0.052

0

-0.032

0.782

WA5A

69

8.922

61.188

0.484

-0.01

-1.321

0.615

WI5A

58

9.575

58.078

0.542

-0.009

-0.914

0.525

a Based on transformed data (i.e., ln(H2S)).
b Based on back-transformed data.

6-20


-------
Deliberative, draft document - Do not cite, quote, or distribute

Intercept

Ambient Temperature







0.014











1



1.20 -



I T

0.010



	 -	-



t t

<







i

0.006





1.10 -



	T	

1.05 -





f





0.002
0.000





WI5A WA5A IN5A NONE

WI5A WA5A IN5A NONE

Figure 6-11. Comparison of variation in coefficients and standard errors for H2S open source model.

Variation in coefficients and standard errors (black closed circle and ± SE bar) for each jackknife model
with the selected H2S open source model coefficient ("None", gray band for ± SE) for each model
parameter.

6-21


-------
Deliberative, draft document - Do not cite, quote, or distribute

7 ANNUAL EMISSION ESTIMATES AND MODEL UNCERTAINTY

To estimate annual pollutant emissions, the results of the daily emission models are
summed over the number of operating days per year. This approach requires values for the
necessary ambient and barn parameters. For an actual emissions estimate, the daily estimates are
based on meteorology from nearby monitors and barn occupancy and weight records for the year
from the producer. For farms with multiple barns, annual emissions are determined for individual
barns and summed across barns to calculate total annual farm-scale emissions.

As noted in Section 6 of the main report, the model results are transformed values of the
emissions. To convert to the native emission units (e.g., kg or g), the back transformation
equation (Equation from Section 6 of the main All Sector report) is applied using the values of Et
and C provided in Table 7-1 for each emission model. Section 8 contains an example of this
calculation.

Table 7-1. Back transformation

parameters

Animal Type

Pollutant

Et

0

Resulting units

Mechanically Ventilated barn

nh3

1.03966

3

kg/d

Mechanically Ventilated barn

HZS

1.11434

628

g/d

Mechanically Ventilated barn

PMio

a

a

a

Mechanically Ventilated barn

PM2.s

a

a

a

Mechanically Ventilated barn

TSP

a

a

a

Milking Center

nh3

1.21693

3

g/d/hd

Milking Center

h2s

1.30119

628

kg/d/hd

Milking Center

PMio

1.0057

2200

g/d/hd

Milking Center

PM2.5

1.00796

680

g/d/hd

Milking Center

TSP

1.0311

978

g/d/hd

Naturally Ventilated barn

nh3

1.46499

3

g/d

Naturally Ventilated barn

HZS

1.23366

628

kg/d

Naturally Ventilated barn

PMio

1.27211

2200

g/d

Naturally Ventilated barn

PM2.s

1.33005

680

g/d

Naturally Ventilated barn

TSP

1.25126

978

g/d

Lagoon/basin

nh3

1.0079

3

kg/d m2

Lagoon/basin

h2s

1.03006

3

kg/d m2

Corral

nh3

1.0066

3

g/d-m2-l,000 hd

Corral

h2s

1.00007

3

g/d-m2-l,000 hd

a Annual models were not calculated to allow time to optimize the daily models.

EPA also developed an estimate of uncertainty for total annual emissions, characterized
by the random error in the model prediction using an approach similar to the Monte Carlo
analysis. Under this approach, EPA developed the statistical properties of predicted annual
emissions by replicating annual sums of daily emissions. EPA ran these simulations for several
different intervals of a predictor variable that fell within the observed range. For example,
naturally ventilated barn inventory ranged from 500 to 600 head. The simulations were then run

7-1


-------
Deliberative, draft document - Do not cite, quote, or distribute

for inventory intervals of 5 head (e.g., 500, 505, 510). Table 7-2 lists the predictor variable and
the number of intervals used for the annual uncertainty simulations for each model.

Table 7-2. Annual Uncertainty Model Details

Source Type

Pollutant

Simulation
Variable

Number of
Simulations

k

Emission
Units

Mechanically ventilated
barn - Flush

h2s

Inventory

10,000

3,457,126

g/d

Mechanically ventilated
barn - Scrape

h2s

Inventory

10,000

3,453,490

g/d

Mechanically ventilated
barn - Flush

nh3

Inventory

10,000

35,180

kg/d

Mechanically ventilated
barn - Scrape

nh3

Inventory

10,000

35,258

kg/d

Mechanically ventilated barn

PMio

a







Mechanically ventilated barn

PM2.s

a







Mechanically ventilated barn

TSP

a







Milking Center

H2S

Ambient
temperature

10,000

9,392,217

g/d-1,000
hd

Milking Center

nh3

Ambient
temperature

10,000

55,494

kg/d-
1,000 hd

Milking Center

PM10

Ambient
temperature

10,000

1,082,872

g/d-1,000
hd

Milking Center

PM2.5

Ambient
temperature

10,000

498,298

g/d-1,000
hd

Milking Center

TSP

Ambient
temperature

10,000

1,557,418

g/d-1,000
hd

Naturally ventilated barn

H2S

Inventory

10,000

4,963,976

g/d

Naturally ventilated barn

nh3

Inventory

10,000

73,495.7

kg/d

Naturally ventilated barn

PM10

Inventory

10,000

59,332,385

g/d

Naturally ventilated barn

PM2.5

Inventory

10,000

5,181,114

g/d

Naturally ventilated barn

TSP

Inventory

10,000

83,299,795

g/d

Lagoon/basin

H2S

Ambient
temperature

10,000

2,606.3

g/d m2

Lagoon/basin

nh3

Ambient
temperature

10,000

4,114.1

g/d m2

Corral

h2s

Ambient relative
humidity

10,000

18,479.4

mg/d-m2-
1,000 hd

Corral

nh3

Ambient
temperature

10,000

1,278.5

g/d-m2-
1,000 hd

a Annual models were not calculated to allow time to optimize the daily models.

Simulations were run 10,000 times for each day for each interval to create an average
uncertainty associated with the annual emissions from a single barn. EPA added a random
residual to each day of the simulation to replicate the variability that would be seen in a real-

7-2


-------
Deliberative, draft document - Do not cite, quote, or distribute

world application of the model. For each of the intervals run, EPA calculated standard statistics
(i.e., minimum, median, mean, maximum, range) and used these to calculate the uncertainty for a
single source via:

/	Range	\

Single source uncertainty = 0.5 x 	-	-	 x 100

\Meaian annual emission)

Equation 19

EPA then plotted this single barn uncertainty against its associated annual emissions.
This plot was then fit with a curve to model annual percent uncertainty for a single source (i.e.,
barn, lagoon, basin). For all uncertainty models, the curve took the form of:

Is

Uncertainty (%) =

Annual Emissions	Equation 20

Where:

& is a constant, listed in Table 7-2, and

Annual Emissions are the total sum from the daily models.

EPA has not calculated particulate matter annual uncertainty models for the mechanically
ventilated barns in order to allow more time to optimize the models. EPA will include the annual
uncertainty models in the final report.

Multiplying this percentage by the annual emissions calculated for the source provides
the resulting uncertainty in the native emission units (e.g., kg or g), demonstrated in Equation 21.

Percent uncertainty x Annual emissions
Resulting Uncertainty = 	—		Equatjon n

To propagate the uncertainty across all sources at a farm, EPA combined the estimates of
absolute uncertainty for each source according to:

Total farm uncertainty = J(UB1)2 + —I- (UBi)2 + (UL1)2 + —I- (ULj)2

Equation 22
Where:

Total farm uncertainty = total uncertainty for the total emissions from all farm sources.
UBi = the resulting uncertainty for barns, with i representing the total number of barns on
the farm,

ULj = the resulting uncertainty for manure sheds, with j representing the total number of
open sources on the farm.

EPA notes that the uncertainty framework described above reflects the random
uncertainty (error) in the prediction of daily emissions calculated using the emission models,
which includes the random uncertainty in the measurements used to develop the equation. This

7-3


-------
Deliberative, draft document - Do not cite, quote, or distribute

framework does not, however, consider systematic error (e.g., bias) in either NAEMS
measurements or the emission model. Section 8 provides an example of how the daily, annual,
and annual uncertainty calculations are completed.

7-4


-------
Deliberative, draft document - Do not cite, quote, or distribute

8 MODEL APPLICATION AND ADDITIONAL TESTING

Key to the development of any model is the demonstration of the use and practical
examples of how the model behaves and replicates independent data. This section provides a
series of example calculations to demonstrate the application of the models (Section 8.1), the
sensitivity of the models to their inputs (Section 8.2), a comparison of the models developed to
literature (section 8.3), and a test of model performance against an independent data set (Section
8.4). Finally, this section wraps up with a discussion of data limitations that could be driving
sensitivity or performance issues.

8.1 Model Application Example

The following sections demonstrate how the daily emission models from Section 5 and
the annual uncertainty from Section 7 are used to calculate emissions for an example farm for
each structure type. Details about the use of the emission models to demonstrate compliance with
Clean Air Act (CAA) permitting thresholds will be addressed in a forthcoming implementation
document. This example is provided to walk through a calculation to demonstrate how the
system of equations is intended to work.

In Section 6.4 of the main report, the data were log-transformed prior to developing the
models, the results of the models will need to be back-transformed per Equation 7 to represent
emissions in units of grams or kilograms.

Ybp = e^*Et-C

Where:

Ybp is the back transformed predicted emissions;
yp is the model predicted (log transformed) emissions;

Et is the average residual between model-predicted and observed (or measured)
emissions on the natural log scale; and

C is a constant added to the data prior to the log transformation.

To complete the back transformation, users need two parameters that are specific to each
model: 1) Eu the residual between model-predicted and observed (or measured) emissions on the
natural log scale; and 2) C, which is a constant added to the data prior to the log transformation.
The values for Et and C for the dairy models are provided in Table 7-1.

Once the emission models are finalized, EPA will work with stakeholders to develop a
tool to facilitate the calculation of barn and open source emissions. For transparency and to help
stakeholders better understand the process of calculating emissions, this section will walk
through example calculations to estimate NH3 emissions from a mechanically ventilated barn,
milking center, naturally ventilated barn, and lagoon.

8-1


-------
Deliberative, draft document - Do not cite, quote, or distribute

The examples in this section use a fictional farm located in Brown County, Wisconsin on
January 1, 2021. Wisconsin was chosen as it is a top five milk producing state according to the
USD A Economic Research Service data

(https://www.ers.usda.gov/webdocs/DataFiles/48685/milkcowsandprod.xlsx?v=9708). The
ambient weather data used in each equation can be obtained for free from several sources
including the National Centers for Environmental Information (NCEI;

https://www.ncdc.noaa.gov/cdo-web/). NCEI stores hourly and daily ambient data from various
monitors located across the country that can be used for emission estimation. The Green Bay
International Airport, WI site (WBAN: 14898), a Local Climatological Data (LCD) Station
located in Brown County was selected as to represent the meteorological information for a
theoretical farm for testing. Its data file provides the daily average values of the key
meteorological parameters needed for calculations.

The naturally ventilated barn and corral models presented in this report use wind speed in
the model calculations. The height at which wind speed is measured influences the observation
as friction with the surface will affect the observation. That means, the closer to the ground the
measurement is made, the more friction will act to slow the speed. NAEMS winds were
monitored at a height of approximately 2.5 meters at open sources and site specific heights at
barn sources, while the National Weather Service (NWS) sites archived at NCEI are typically
monitored at 10m. Therefore, the difference in measurement heights between NAEMS and NWS
requires an adjustment to the wind. The relationship between wind speed and height is well
established and can be written as:

Where Vr is the wind velocity at a height of 10 m (Zr) and V is the wind velocity height at 2.5 m
(Z), and m is the friction coefficient, which is a function of atmospheric stability and the
underlying surface roughness. The value of m can vary, ranging from 0 to 1, with lower values
over low roughness surfaces (water) and higher values for rougher terrain (e.g., rolling terrain or
urban settings) (Arya, 1999). To adjust the 10m NWS wind measurement to a height comparable
to the study data used to develop the model, the equation can be rewritten, resulting in

EPA is determining the best value of m to use for corrals and naturally ventilated barns. For the
purposes of the example calculations, we will use the average daily wind speed from the NWS
site.

Equation 23

Equation 23

8-2


-------
Deliberative, draft document - Do not cite, quote, or distribute

In addition to weather information, the models also use the number of cows present in the
barn. For this fictitious farm, we assume the barn has a capacity of 500 cows. The equations use
thousands of cows, so this value will be divided by 1,000 for use in the emission models. A
summary of the input values for the example calculations is provided in Table 8-1.

Table 8-1. Daily calculation parameter values

Parameter Value

Daily Average Ambient Temperature (°C)

-9.4

Daily Average Relative Humidity (%)

86

Average Wind Speed (ms"1)

2.55

Inventory (thousand head)

0.50

8.1.1 Mechanically Ventilated Barn Example

For this example, we will assume the barn uses a scrape manure management system,
which would use Equation 1, in Section 5.1, to calculate the log transformed values as follows:

ln(NH3) = 1.86494 + 1.773832 * Inventory + 0.029586 * AmbT

( 500 \

In(NH3) = 1.86494 + 0.1.773832 * (^-^J + 0.029586 * -9.4

In(NH3) = 1.86494 + 0.8869 - 0.2781
In (NH3) = 2.4737

To back transform the results to NH3 in kg, use Equation 7, from the main report. For a flush
managed mechanically ventilated barn, Et is 1.03966 and C is 3.

NH3 = e2-4731 X 1.0 3 9 6 6 - 3

This comes to 9.34 kg NH3 for the day. This process is repeated for each day, then the daily
emissions are added together to get an annual estimate of emissions. After considering the values
for each day in 2021, the total annual emission for the barn was calculated at 7,108 kg. To
calculate the uncertainty associated with this estimate, use Equation 17 with the value of k from
Table 7-1. This results in an annual uncertainty of:

,	35,180

Uncertainty (%) = 	= 4.95%

'	7,108.31

This translates to an uncertainty of ± 351kg. Thus, the final annual estimate for this barn is
7,108kg ± 352 kg. This calculation would be repeated for any other mechanically ventilated barn
on the site.

8-3


-------
Deliberative, draft document - Do not cite, quote, or distribute

8.1.2	Milking Center Example

For this example, we will use Equation 5, in Section 5.2, to calculate the log transformed
values as follows:

ln(NH3) = 2.505637 + 0.046434 * AmbT
In(NH3) = 2.505637 + 0.04643 * -9.4
In(NH3) = 2.505637 - 0.4368
In (NH3) = 2.0692

To back transform the results to NH3 in kg, use Equation 7, from the main report. For a milking
center, Et is 1.03966 and C is 3.

NH3(	—	) = e20692 x 1.2169 - 3

d ¦ 1,000 head

kg

NH3(-——£-	-) = 6.64

d ¦ 1,000 head

This comes to 6.64 kg NH3/d-l,000 head, which we can multiply by the 0.5 thousand head to get
3.32 kg NH3 for the day. This process is repeated for each day, then the daily emissions are
added together to get an annual estimate of emissions. After considering the values for each day
in 2021, the total annual emissions for the milking center were calculated at 4,161.53 kg. To
calculate the uncertainty associated with this estimate, use Equation 17 with the value of k from
Table 7-1. This results in an annual uncertainty of:

. . 55,494

Uncertainty (%) = 	= 13.33%

'	4,161.53

This translates to an uncertainty of ± 555 kg. Thus, the final annual estimate for this milking
center is 4,161.53 kg ± 554.94 kg.

8.1.3	Naturally Ventilated Barn Example

For this example, we will use Equation 10, in Section 5.3, to calculate the log
transformed values as follows:

ln(NH3) = 0.188357 + 3.451939 * Inventory + 0.048153 * WindSpeed

ln(NH3) = 0.188357 + 3.451939 * Inventory + 0.048153 * WindSpeed

( 500 \

In(NH3) = 0.188357 + 3.451939 * 	 + 0.048153 * 2.55

3	Vl,000/

In (NH3) = 0.188357 + 1.7260 + 0.1228

8-4


-------
Deliberative, draft document - Do not cite, quote, or distribute

In (NH3) = 2.0371

To back transform the results to NH3 in kg, use Equation 7, from the main report. For a naturally
ventilated barn, Et is 1.03966 and C is 3.

NH3 = e2 0371 X 1.46 4 9 9 - 3

This comes to 8.23 kg NH3 for the day. This process is repeated for each day, then the daily
emissions are added together to get an annual estimate of emissions. After considering the values
for each day in 2021, the total annual emissions for the barn were calculated at 3,462.82 kg. To
calculate the uncertainty associated with this estimate, use Equation 17 with the value of k from
Table 7-1. This results in an annual uncertainty of:

73,495.70

Uncertainty (%) =	=21. %

o,4oz.oz

This translates to an uncertainty of ± 734.96 kg. Thus, the final annual estimate for this barn is
6,192.70 kg ± 351.80 kg. This calculation would be repeated for any other naturally ventilated
barn on the site.

8.1.4 Lagoon Example

For this example, we will use Equation 15, in Section 5.4, to calculate the log
transformed values as follows:

ln{NH3) = 1.396734 + 0.027201 * AmbT

ln(NH3) = 1.396734 + 0.027201 * -9.4

ln(NH3) = 1.396734 - 0.2557

In (NH3) = 1.1410

To back transform the results to NH3 in kg, use Equation 7, from the main report. For a lagoon,
Et is 1.0079 and C is 3.

NH3 = e11410 x 1.0079 - 3

This comes to 0.1548g NFb/d m2 This is multiplied by the surface area of the lagoon to estimate
emissions for the whole lagoon. For this example, we will assume the lagoon is 10,000 m2,
which would result in emissions of 1,547 kg NH3 for the day.

This process is repeated for each day, then the daily emissions are added together to get
an annual estimate of emissions. After considering the values for each day in 2021, the total
annual emissions for the lagoon were calculated at 8,961.21 kg. To calculate the uncertainty

8-5


-------
Deliberative, draft document - Do not cite, quote, or distribute

associated with this estimate, use Equation 17 with the value of k from Table 7-1. This results in
an annual uncertainty of:

4,114.1

Uncertainty (%) = 	= 0.46%

y	8,961.21

This translates to an uncertainty of ± 41.14 kg. Thus, the final annual estimate for this lagoon is
8,961.21 kg ±41.14 kg. This calculation would be repeated for any other lagoon on the site.

8.1.5 Corral Example

For this example, we will use Equation 17, in Section 5.5, to calculate the log
transformed values as follows:

ln(NH3) = 1.053805 + 0.004993 * AmbT + 0.0031 * AmbRH + 0.017832 * WindSpeed

n(NH3) = 1.053805 + 0.004993 * -9.4 + 0.0031 * 86 + 0.017832 * 2.55
ln(NH3) = 1.053805 - 0.0469 + 0.266 + 0.0455
In (NH3) = 1.3189

To back transform the results to NH3 in kg, use Equation 7, from the main report. For a corral, Et
is 1.0066 and C is 3.

NH3 = e13189 x 1.0066 - 3

This comes to 0.07641 g NFb/d m2 1,000 head. This is multiplied by the surface area of the
corral and inventory to estimate emissions for the whole corral. For this example, we will assume
the surface area of the corral is 100,000 m2 and the farm population is 3,400 head, which would
result in emissions of 260 kg NH3 for the day.

This process is repeated for each day, then the daily emissions are added together to get
an annual estimate of emissions. After considering the values for each day in 2021, the total
annual emissions for the corral were calculated to be 124,562.33 kg. To calculate the uncertainty
associated with this estimate, use Equation 17 with the value of k from Table 7-1. This results in
an annual uncertainty of:

1,278.5

Uncertainty (%) =	= 0.01%

Iz4,b6z.33

This translates to an uncertainty of ± 12.79 kg. Thus, the final annual estimate for this corral is
124,562.33 kg± 12.79 kg.

8-6


-------
Deliberative, draft document - Do not cite, quote, or distribute

8.1.6 Combining Structures

To calculate total farm emissions, the emissions from each unit are added. As an
example, consider a farm with a 500 head mechanically ventilated barn, 500 head naturally
ventilated barn, milking center with a 500 head capacity at any given time, and 10,000 m2
lagoon. That is, the same emissions as the examples in sections 8.1.1 through 8.1.4. The annual
farm emission estimate from four sources is:

Farm Total Emissions = 7,108.31 + 4,161.53 + 6,192.70 + 2,439.20

Farm Total Emissions = 19,901.74 kg NH3

To estimate the total farm uncertainty, use Equation 41:

Total Farm Uncertainty & barn 1 U ham 2 Umilking center Uiagoon

Total Farm Uncertainty = ^(351.80)2 + (554.94)2 + (734.96)2 + (41.41)2
Total Farm Uncertainty = 986.71 kg

The final annual NH3 estimate for the farm is 19,901.74 ± 986.71 kg. Once the emission models
are finalized, EPA will work with stakeholder to develop a tool to facilitate the calculation of
barn and open source emissions.

8.2 Model Sensitivity Testing

To further test the models, EPA varied the model parameters to ensure the model results
would vary based on these key parameters. Two different tests were conducted: 1) the number of
cows was increased while the meteorological parameters were held constant, and 2) inventory
was held constant while the meteorological parameters were replaced with the values for a
warmer climate.

8.2.1 Sensitivity to Inventory

To test the sensitivity of the confinement sources to inventory, the initial placement was
doubled to 1,000 cows. Using the same meteorology from Section 8.1, the emissions for the
dairy barns on January 1, 2020, is summarized in Table 8-2. For mechanically ventilated barns
and milking centers, doubling the inventory at least doubled the NH3 emissions for the same
meteorological conditions. For naturally ventilated barns, doubling the inventory resulted in a
sevenfold increase in NFb emissions. The large increase in the naturally ventilated barn
emissions is further discussed in Section 8.2.3.3. These same ratios are seen when considering a
year's worth of meteorology (Table 8-3).

8-7


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 8-2. Comparison of confinement source NH3 emissions (kg) on January 1,
2021, for different inventory levels at a theoretical Brown County farm.

Source Type

500 head

1 000 head

Mechanically Ventilated

9.34

26.91

Milking center

3.32

6.62

Naturally ventilated

8.23

62.49

Table 8-3. Comparison of confinement source total 2021 NH3 emissions (kg) for
different inventory levels at a theoretical Brown County farm.

Source Type

500 head 1 1,000 head

Mechanically Ventilated

7,108

18,820

Milking center

4,162

8,323

Naturally ventilated

3,463

24,511

For lagoons, doubling the surface area of the lagoon doubles both the daily and annual
NH3 emissions (Table 8-4). For corrals, doubling the inventory present doubles both the daily
and annual NH3 emissions (Table 8-5). The observed relationships suggest the models are
sensitive to the size parameters, while scaling appropriately.

Table 8-4. Comparison of lagoon NH3 emissions (kg) for different surface areas

for theoretical Brown County farm.

NH, Emissions (kg) 10,000 m' 20,000 m'

Daily (1/1/2021)

1.51

3.02

Annual (2021)

8,961

17,922

Table 8-5. Comparison of estimated corral NH3 emissions (kg) for different
inventory levels for theoretical Brown County farm.

NH3 Emissions (kg)

3,400 head

6,800 head

Daily (1/1/2021)

259.48

518.96

Annual (2021)

124,562

249,125

8.2.2 Sensitivity to Climate

To further test model sensitivity, specifically that climate differences were producing
different emission results, EPA calculated the emissions for the same farm in two distinctly
different climate regions. The first was the theoretical farm in Brown County, Wisconsin from
the previous examples (Section 8.1). The NH3 emission for these same theoretical barns were
calculated using meteorological data from Livermore Municipal Airport in Alameda County,
California. These locations were chosen based on 2017 Census of agriculture data indicating
areas of high dairy inventory (Figure 8-1). USDA Economic Research Service data (available at:
https://www.ers.usda.gov/webdocs/DataFiles/48685/milkcowsandprod.xlsx?v=9708) also notes
California and Wisconsin are the top two dairy producing states in the country, further affirming
the reasonableness of the testing locations.

8-8


-------
Deliberative, draft document - Do not cite, quote, or distribute

(CA). Source: https://www.nass.usda.gov/Publicatioris/AgCensus/2017/Online Resources/Ag Atlas Maps/17-
M209g.php

For the test sites, the temperatures from the Wi sconsin (WI) site were generally less than
the California (CA) site (Figure 8-2). On average, the temperatures in Wisconsin were 7°C less
than those in California (Table 8-6), with difference between individual monthly averages
varying from 1.6 to 20.8°C lower, except for July when Wisconsin edged 0.6°C higher. With
respect to relative humidity, the California and Wisconsin sites experienced a similar range of
daily average relative humidities throughout the year (Figure 8-3 and Table 8-7). Wisconsin
edged a little higher July through October, leading to an overall average 1.6% higher. Average
daily wind speeds (Figure 8-4 and Table 8-7) were generally lower in California, with monthly
average barely higher June through August. The following sections provide a summary of the
calculations using the California meteorological data compared to the previous examples.

8-9


-------
Deliberative, draft document - Do not cite, quote, or distribute

Temperature (°C)

40.0

I

-20.0	!¦

-30.0

s\\	rtN-	a"v a"v ^."V -VV	aN1

j9>	j& jS?	jfr J?	Kjy

¦S? •/ ^ S? 4?J? ^ 4? / / JS?

	Wl 	CA

Figure 8-2. Comparison on average daily temperatures at test locations in Wisconsin (Wl) and
California (CA).

Table 8-6. Summary of average daily temperature at the two meteorological sites.

Site

Statistic

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Overall

W!

Min

-12.8

-21.7

-2.8

1.7

6.1

13.9

16.1

17.8

13.3

3.9

-6.1

-14.4

-21.7

Max

2.2

4.4

12.8

18,9

25.0

28.3

27.2

25.0

22.8

21.7

12.8

11.1

28.3

Average

-4.8

-9.5

3.4

9.1

14.0

21.9

21.5

21.9

17.3

12.6

2.2

-2.0

9.0

CA

Min

4.4

8.3

6.1

11.1

13.9

15.0

18.9

18.9

17.8

12.2

10.0

0.0

0.0

Max

19.4

15.6

18.9

20.0

25.6

30.0

30.6

28.3

28.9

23.9

16.7

12,2

30.6

Average

10.3

11.3

11.5

15.1

18.4

21.3

23.5

23.4

22.4

16.9

13.0

8.0

16.3

Relative Humidity (%)

	Wl 	CA

Figure 8-3. Comparison of average daily relative humidities at test locations in Wisconsin (Wl) and
California (CA).

8-10


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 8-7. Summary of average daily relative humidity at the two meteorological

sites.

Site

Statistic

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Overall

Wl

Min

54.0

51.0

47.0

43.0

39.0

42.0

63.0

66.0

60.0

52.0

50.0

0.0

0.0

Max

91.0

79.0

87.0

88.0

85.0

92.0

91.0

90.0

87.0

91.0

81.0

86.0

92.0

Average

75.9

66.4

64.2

63.7

63.1

64.9

72.1

76.1

72.3

75.8

66.5

69.4

69.2

CA

Min

35.0

35.9

39.4

38.6

49.2

42.7

58.1

51.0

42.3

53.0

31.8

28.0

28.0

Max

95.3

92.0

94.4

93.5

82.0

86.7

82.1

73.0

86.4

90.7

93.9

86.3

95.3

Average

68.3

66.2

73.0

70.3

67.5

69.9

67.3

62.3

70.3

67.6

67.5

64.6

67.8

Average wind Speed (ms-1)

12.00

^_WI 	CA

Figure 8-4. Comparison of average daily wind speeds at test locations in Wisconsin (Wl) and
California (CA).

Table 8-8. Summary of average daily wind speeds at the two meteorological sites.

Site

Statistic

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Overall

Wl

Min

1.4

0.5

1.5

2.2

1.8

1.5

0.9

1.1

0.9

1.6

1.9

0.0

0.0

Max

7.4

7.6

8.4

6.9

7.2

5.4

5.1

4.5

5.4

7.2

6.9

10.3

10.3

Average

3.6

3.8

4.5

4.1

3.6

3.5

2.9

2.5

3.0

3.4

3.9

4.0

3.6

CA

Min

0.6

1.2

1.2

1.7

1.9

1.3

1.7

1.6

1.2

1.1

0.4

0.0

0.0

Max

7.0

4.6

5.1

4.9

6.2

6.4

5.6

5.0

4.8

6.4

3.9

6.2

7.0

Average

2.2

2.4

2.6

3.4

3.7

4.2

3.7

3.2

2.6

2.7

1.6

2.5

2.9

8.2.2.1 Mechanically Ventilated Barn

When the daily calculations are performed for the entire year for a mechanically
ventilated dairy barn with 500 cows, the California site typically has higher daily emissions for
both NH3 and H2S and for either manure management system than the Wisconsin site (Figure
8-5). Table 8-9 contains the estimated annual emissions for the different combinations of
pollutant and manure management system. For the mechanically ventilated scrape barn from the
example in Section 8.1.1, the total annual NH3 emissions estimate for the farm using
meteorological data from California was 8,689 kg— a 1,581 kg increase from the same

8-11


-------
Deliberative, draft document - Do not cite, quote, or distribute

mechanically ventilated barn with meteorological data from Wisconsin. A similar trend is seen
across the other pollutant and manure management system combinations. This is consistent with
the trend of lower temperatures yielding lower emissions seen during the data exploration in
Section 4. Overall, this suggests that the emission models can account for regional temperature
differences in the results for mechanically ventilated barns.

Estimated Daily NH3 Flush MV (kg)

35.0

Estimated Daily NH3 Scrape MV (kg)

40.0





0.0

rft n.Q ~C> rft ,ft

& 4? / /V

	Wl 	CA

5.0 ^

0.0

*ft ytft ."ft -"ft aO i"ft  „ft ,ft „ft rft -ft rft rft -ft

jf- jf 
-------
Deliberative, draft document - Do not cite, quote, or distribute

across the other pollutants, with increases ranging from 38% to 152%. This is consistent with the
trend of lower temperatures yielding lower emissions seen during the data exploration in Section
4. Overall, this suggests that the emission models can account for regional temperature
differences in the results for milking centers.

Estimated Daily NH3 MC (kg)

Estimated Daily H2S MC (kg)



aN	aS	a\	aN aN a\ aN a\ a\

-Ov Jy rSr rSy	.v	jy rSy _ov „

	Wl 	CA

Estimated Daily TSP MC (kg)



-Wl 	CA

Estimated Daily PM2.5 MC (kg)

-0.04

a*V n\ a\ a\ aS a% n*V n% aV a\ oV aN

o?	/I?	O? ^?v VT?V ^	v^V v^V

-v\s ^ ^ ^ ^	^ <*\s ,?\% ss\N ^

Figure 8-6. Comparison of daily milking center emission at test dairy locations in Wl and CA.
Table 8-10. Total annual emission from a theoretical milking center in Wl and CA.



Wl Emissions

CA Emissions

Pollutant

(kg per year)

(kg per year)

nh3

4,162

5,479

h2s

189

474

PMio

74

112

PM2.5

18

24

TSP

185

427

8-13


-------
Deliberative, draft document - Do not cite, quote, or distribute

8.2.2.3 Naturally Ventilated Barn

A naturally ventilated dairy barn with 500 cows in California typically has lower daily
emissions than the same barn in Wisconsin (Figure 8-7) for gaseous pollutants and PM2.5. Table
8-11 has the estimated annual emissions of the pollutants studied. The differences in the annual
gaseous pollutants are minor, as the models are based on average daily wind speed which is only
slightly different between the sites. Table 8-11 shows a larger difference with the PM2.5 annual
emissions, and the plot shows several large spikes when using the Wisconsin meteorological
data. Looking into the data, these data points are associated with days with high average daily
wind speeds and suggests some limitation in the model performance for these instances. This is
discussed further in Section 8.2.3.3. For PM10 and TSP, the spikes in emissions are generally due
to higher wind speeds combined with lower relative humidities to mitigate the emission. These
relationships are explored more in section 8.2.3.3

Estimated Daily NH3 NV (kg)

Estimated Daily H2S NV (kg)



4.0
2.0
0.0

-0)" <$* <$" -O" V -Cr Jbf .O" -Ci"	-O" Jb*

^	n? 4? ^ / J? > ^

0.4
0.2
0.0

A »	A >	rt >	yv>	A >	A*	AT	A y	A *	AT

jy	&

& ^ ^ ^ J? /• /V

Estimated Daily PM10 NV (kg)

Estimated Daily PM2.5 NV (kg)

16.0
14.0
12.0
10.0
8.0
6.0
4.0
2.0
0.0
-zo

^ J* ^ ^ ^	^

•S? 4* 4r	^ ^ ^



40000.0
35000.0
30000.0
25000.0
20000.0
15000.0
10000.0
5000.0
0.0 —

jLl.

AT AT AT PIT AT AT AT AT	AT AT A >	AT

a0 
-------
Deliberative, draft document - Do not cite, quote, or distribute
Table 8-11. Total annual emission from a theoretical milking center in Wl and CA.



Wl Emissions

CA Emissions

Pollutant

(kg per year)

(kg per year)

nh3

3,463

3,274

h2s

297

275

PM10

777

962

PM2.5

89,168

23,113

TSP

112

369

8.2.2.4 Lagoon

Repeating the daily calculations for the dairy lagoon using the California meteorological
data typically has higher daily emission values than when using the Wisconsin meteorological
data (Figure 8-8). Table 8-12 has the estimated annual emissions of each pollutant studied and
shows a roughly 40% increase for both pollutants using the warmer temperatures from
California. This is consistent with the trend of warmer temperatures yielding greater emissions
seen during the data exploration in Section 4 and noted in the literature review in Section 3.
Overall, this suggests that the emission models are capable of accounting for the different
growing regions in the lagoon results.

Estimated Daily NH3 (kg)	Estimated Daily H2S (kg)

70.0	20.0

	Wl 	CA		Wl 	CA

Figure 8-8. Comparison of daily lagoon emission at test dairy locations in Wl and CA.

Table 8-12. Total annual emission from a theoretical lagoon in Wl and CA.

Pollutant

Wl Emission
(kg per year)

CA Emission
(kg per year)

nh3

8,961

12,525

h2s

2,734.2

3,748.8

8.2.2.5 Corral

The emission estimates for a corral using the meteorological data from California, are
slightly lower than calculations with the Wisconsin meteorological data (Figure 8-9). Table 8-13
has the estimated annual emissions of each pollutant and shows the total annual NH3 emissions
estimate for the theoretical California corral was 124,261 kg, which is a 302 kg decrease from
the same theoretical corral in Wisconsin. The FhS model only shows a minor difference between

8-15


-------
Deliberative, draft document - Do not cite, quote, or distribute

the emissions for the two climates. This generally limited sensitivity is discussed more in Section

8.2.3.5.

Estimated Daily NH3 (kg)

600.0

Estimated Daily H2S(kg)

70

z

2.0

100.0 *

0.0 |	J	1	1	1	1	1	1	1	1	J	J	1

# 0"? & # Q-P A® 0-v°

.0? a0 a0  oP 0? of <1? o? <•$> &

	Wl 	CA

LO

0.0 1	1	1	1	1	1	1	1	1	1	1	1	1

5# qP o"P # # o? Qp 0-P
A'1 ^ ¦$? Jy a0 -? A-P

•r -T ^ ^ ^ ^ ^ ^

	Wl 	CA

Figure 8-9. Comparison of daily milking center emission at test dairy locations in Wl and CA.
Table 8-13. Total annual emission from a theoretical milking center in Wl and CA.

Pollutant

Wl Emission
(kg per year)

CA Emission
(kg per year)

nh3

124,562

124,261

h2s

1,902.7

1,789.7

8.2.3 Model Limitations

As noted in the 2013 SAB review (US EPA SAB, 2013), extrapolating to conditions
beyond those represented in the model development dataset could produce unrealistic results. To
test the limitations of the model, EPA conducted a series of emission calculations over a range of
conditions that could be seen at a farm in the US. These emission calculations tested one
parameter at a time, with the selected parameter varied by a constant value through the range.
For example, ambient temperature was increased by 1°C from the minimum value in the model
development dataset up to the maximum value. While one parameter was tested, the remaining
parameters were held constant at the average value seen in the model development dataset. The
resulting emission values were reviewed and plotted to determine if the model resulted in
unrealistic emission values, such as negative emissions or rapid increases in emission rates.

The dairy equations included some combination of inventory, ambient temperature,
ambient relative humidity, and wind speed. The ranges of ambient parameters are based on the
NAEMS dataset. The number of cows in a single barn or milking center are based on barn
capacity numbers provided by consent agreement participants. The range values tested for each
parameter are in Table 8-14.Table 8-14

This analysis does not account for interaction between multiple terms within an equation,
which could further affect the results. For example, a dairy barn with higher ambient
temperatures would be able to cover a larger range of inventory per barn before producing
negative NH3 emissions. Conversely, a barn with lower ambient temperatures would cover a

8-16


-------
Deliberative, draft document - Do not cite, quote, or distribute

smaller range of inventory before producing negative NH3 emission values. However, the
analysis does provide a general range where the model produces reasonable results.

To further explore any limitations in the models, emissions were calculated for all
combinations across the range of values specified in Table 8-14. A list of all the combinations of
the three inputs was created using the R statistical software. R was then used to calculate the
emissions using the method shown in section 8.1. The results were then filtered down to only the
results that produced negative values to generate the plots for each pollutant. The following
sections outline the analysis for each of the selected models.

Table 8-8-14. Parameter ranges tested for the dairy models.

Parameter

Upper limit

Lower limit

Average Value

Increment

Ambient temperature (°C)

32.0

-23

10.0

0.8

Ambient relative humidity (%)

93

24

68.1

1

Wind speed (ms"1)

11.2

0.00

2.3

0.15

Inventory (head)

5,000

0

1,000

70

8.2.3.1 Mechanically Ventilated Barn

The initial analysis for mechanically ventilated barns is presented in Figure 8-10 and
Figure 8-11. Neither the FhS (Figure 8-10) nor NH3 (Figure 8-11) models produce negative
emissions under average conditions. Additional analysis of the 5,110 combinations of conditions
tested produced negative values. The models also produce a rapid increase in emissions when
estimating barns with inventories greater than 2,000 head. The largest barn in the NAEMS had
an average daily population of 833, which would account for the unrealistic behavior with
extreme inventory numbers. Based on the consent agreement participant data, more than 90% of
the participating barns fall below a capacity of 2,000 head. This suggests the model would still
be appropriate for the bulk of the participants. EPA will explore models that predict emissions
normalized by inventory, as these models will produce a linear relationship between inventory
and emissions (with other factors constant), regardless of the size of the operation.

8-17


-------
Deliberative, draft document - Do not cite, quote, or distribute

H2S-Flush emissions versus inventory

H2S-Flush emissions versusTemperature





g













	5













	4	













	3













	2_













1













	O-









2	3

Inventory (thousand head)

-10	0	10	20

Ambient temperature (*C)

H2S-Scrape emissions versus inventory

H2S-Scrape emissions versusTemperature

_§#> 25

2	3

Inventory (thousand head)





1.60
1.40













1 20













l.CC













O.SOj
" 0,60
0.40





































0 20













	O.OO









-30 -20

-10	0	10	20

Ambient temperature (*C)

Figure 8-10. Mechanically ventilated barn limitation tests for H2S.

Visualization of the results for H2S - Flush (top row) and H2S - Scrape (bottom row) tests of inventory
(left) and ambient temperature (right).

8-18


-------
Deliberative, draft document - Do not cite, quote, or distribute

N H3-Flush emissions versus inventory

NH3-Flush emissions versus Temperature

2	3	4

Inventory (thousand head)





80.00













70 00



























50.00













40.00



























2C.CC













10.00













	0;00-









-20	-10

0	10	20

Ambient temperature (*C)

NH3-Scrape emissions versus inventory

NH3-Scrape emissions versus Temperature





90.00
80.00





















































5C.CC













40.00









































10.00













0.00









12	3	4

Inventory (thousand head)

-30	-20

-10	0	10	20

Ambient temperature (*C)

Figure 8-11. Mechanically ventilated bam limitation tests for NH3.

Visualization of the results for NH3 - Flush (top row) and NH3 - Scrape (bottom row) tests of inventory
(left) and ambient temperature (right).

8.2.3.2 Milking Center

The milking centers analysis for gaseous pollutants is presented in Figure 8-12 and
particulate matter is presented in Figure 8-13. Neither the H2S nor NH? (Figure 8-12) models
produce negative emissions under average conditions. The relationship of emissions to
increasing temperature is fairly linear through the expected conditions and does not display any
extreme behavior that would suggest extrapolation issues.

8-19


-------
Deliberative, draft document - Do not cite, quote, or distribute

Figure 8-12. Milking center limitation tests for gaseous pollutants.

Visualization of the results for H2S (left) and NH3 (right) tests of ambient temperature.

The PMio and PM2.5 models (Figure 8-13) do produce negative emission values less than
-11°C and -18.2°C for PM10 and PM2.5 models, respectively, at average relative humidity levels.
Additional analysis of 5,390 combinations of temperature and relative humidity values shows the
PM10 model (Figure 8-14) will produce negative emission estimates when temperatures fall
below zero in an increasingly drier environment. That is, the lower the temperature, the lower the
relative humidity needed to produce a negative emissions value. For example, the equation for
PM10 will produce negative emissions at any level of relative humidity when ambient
temperature falls just below zero. Similarly, at -21.4°C, the equation can produce negative
number when relative humidity is less than or equal to -60%.

8-20


-------
Deliberative, draft document - Do not cite, quote, or distribute

PM10- emissions versustemperature

Ambient temperature (°C)

TSP- emissions versus temperature

-20	-10	0	10	20	30

Ambient temperature (eC)

PM25- emissions versustemperature

PM10- emissions versus relative humidity

1000.0

2

I 500.0
£ 400.0

0.0

0	20	40	60	80

Ambient reiative humidity (%)

TSP- emissions versus relative humidity

900.00

"g 500.00
o 500.00
J?

— 400.00
E 300.00

0.00

0	20	40	60	80

Ambient relative humidity (%)

Figure 8-13. Milking center limitation tests for particulate matter.

Visualization of the results for PM10 (top row), TSP (center row), and PM2.5 (bottom row) tests of
ambient temperature (left) and ambient relative humidity(right).

8-21


-------
Deliberative, draft document - Do not cite, quote, or distribute

Figure 8-14. Maximum values of relative humidity for each temperature at which the PM10
equation yields negative emissions.

8.2.3.3 Naturally Ventilated Barn

The naturally ventilated barn analysis for gaseous pollutants is presented in Figure 8-15.
Analysis for PMio, PM2.5, and TSP are presented in Figure 8-17, Figure 8-19, and Figure 8-18,
respectively, and particulate matter is presented in Figure 8-13. The FhS (Figure 8-12) model
does not produce negative emissions under average conditions with varying inventory. The NH3
model will produce negative emission for very small inventories (i.e., less than 70 head) under
average conditions. Further testing of 5,548 combinations of wind speed and inventory show at
very low wind speeds (< 1 ms"1), an inventory as large as 140 cows will produce negative
emissions. As wind speed increases, the corresponding inventory needed to produce a negative
number also decreases. These thresholds are demonstrated in Figure 8-16. The sensitivity
analysis testing shows rapid increases in NH3 and FhS emissions at high inventories. EPA will
explore models that predict emissions normalized by inventory, as these models will produce a
linear relationship between inventory and emissions (with other factors constant), regardless of
the size of the operation.

8-22


-------
Deliberative, draft document - Do not cite, quote, or distribute

Dairy - NV H2S emissions versus inventory

Dairy - NV H2S emissions versus wind speed

Inventory (thousand head)

0.00	2.00

4.00	6.00	8.00

Wind Speed (ms-l)

10.00 12.00

Dairy - NV NH3 emissions versus inventory

Dairy - NV NH3 emissions versus wind speed

§3000000

Inventory (thousand head)

100.0
90.0
80.0
70.0
5" 60.0

i

O 50.0
w 40.0
30.0
20.0
10.0
0.0

0.00	ZOO

4.00	6.00	5

Wind Speed (ms-l)

00 10.00 12.00

Figure 8-15. Naturally ventilated barn limitation tests for gaseous pollutants.

Visualization of the results for H2S (top row) and NH , (bottom row) tests of inventory (left) and wind
speed (right).

NV-NH3

0.16
=5" 0.14

ft)

0.12

"O

2 o.i

i/i

o 0.08

X. 0.06
tj 0.04

OJ

C 0.02

-Max of Inventory

rNm-3--3-int>or-.ooCTio

Wind speed (ms1)

Figure 8-16. Maximum values of inventory for each wind speed at which the NHs equation yields
negative emissions.

8-23


-------
Deliberative, draft document - Do not cite, quote, or distribute

Though it is hard to see on the figures, the PMio and TSP models (Figure 8-17, and
Figure 8-18) produce negative values under average conditions for very small inventory levels.
Further analysis of 29,903,720 combinations of inventory, ambient temperature, ambient relative
humidity, and wind speed show that the models will produce negative values for progressively
lower temperatures and winds speeds for increasing temperatures (Figure 8-20). For example,
with the PMio model (top graph, Figure 8-20) for an empty barn, the model will produce a
negative emission value for temperatures less than 32°C and wind speed less than 9 ms"1. As
inventory increases to 1,050 head, negative emissions only occur at temperatures below -30°C
and wind speeds less than 1 ms"1. The sensitivity analysis testing shows rapid increases in PMio
and TSP emissions at high inventories. EPA will explore models that predict emissions
normalized by inventory, as these models will produce a linear relationship between inventory
and emissions (with other factors constant), regardless of the size of the operation.

The PM2.5 model (Figure 8-19) did not produce negative values under average conditions.
However, looking across the combinations of inventory, ambient temperature, ambient relative
humidity, and wind speed, the PM2.5 model produces negative emission estimates at low wind
speeds and temperatures combined with low inventory levels (Figure 8-20). As inventory levels
increase, the negative emission estimates can occur at higher values of temperature and wind
speed. This is due to the negative relationship between PM2.5 and inventory in the model, which
will need to be further explored. One option is to explore models that predict emissions
normalized by inventory, as these models will produce a positive linear relationship between
inventory and emissions (with other factors constant), regardless of the size of the operation.

8-24


-------
Deliberative, draft document - Do not cite, quote, or distribute

Dairy - NV PM10 emissions versus inventory

Dairy - NV PM10emissions versusTemperature

inventory (thousand head)

-30	-20	-10

0	10	20

Ambient temperature (*C)

30	40

Dairy - NV PM10 emissions versus relative
humidity

Dairy - NV PM10 emissions versus wind speed

Ambient relative humidity (%)

0.00 2.00	4.00	6.00	8.00 10.00 12.00

Wind Speed (ms-1)

Figure 8-17. Naturally ventilated barn limitation tests for PM™.

Visualization of the results for PMm tests of inventory (top left), ambient temperature (top right),
relative humidity (bottom left), and wind speed (bottom right).

8-25


-------
Deliberative, draft document - Do not cite, quote, or distribute

Dairy - NV TSP emissions versus inventory

Dairy - NV TSP emissions versus Temperature

1	2	3	4	5

inventory {thousand head)





30.0	;













25.0



























15.0













10 f) m













	5_Q_j













	04>—









-30	-20

0	10	20

Ambient temperature (*C)

Dairy - NV TSP emissions versus relative humidity

Dairy - NV TSP emissions versus wind speed

20	40	60	80

Ambient relative humidity (%)

0.00	2.00	4.00	6.00 8.00 10.00 12.00

Wind Speed (ms-1)

Figure 8-18. Naturally ventilated barn limitation tests for TSP.

Visualization of the results for TSP tests of inventory (top left), ambient temperature (top right), relative
humidity (bottom left), and wind speed (bottom right).

8-26


-------
Deliberative, draft document - Do not cite, quote, or distribute

Dairy - NV PM25 emissions versus inventory

Dairy - NV PM25 emissions versusTemperature





1.8
1.6



























12























































0.4













0,2-













	aoJ









2	3

Inventory (thousand head)

0	10

Ambient temperature (*C)

Dairy - NV PM25 emissions versus relative
humidity

Dairy - NV PM25 emissions versus wind speed

20	40	SO

Ambient relative humidity (%)

4.00	6.00	8.00

Wind Speed (ms-l)

10.00 12.00

Figure 8-19. Naturally ventilated barn limitation tests for PM2.5.

Visualization of the results for PM2.5 tests of inventory (top left), ambient temperature (top right),
relative humidity (bottom left), and wind speed (bottom right).

8-27


-------
Deliberative, draft document - Do not cite, quote, or distribute

NV- PM10

Max of temperature ( C)	Max of wind speed (nrts-1)

NV - TSP

Max of tempa-aturefC)	Max of wind speed (ms-1)

u 20

-30

0 70 140 210 280 350 420 490 560 630 700 770 840 910 980
Inventory (head)

* * f -V? ^ & & & <$>	& of?

Inventory (thousand head)

NV-

40
30

u 20

•3

E 10

+J

ro

£ 0

Q.

I -10

-20

PM2.5

—< Max temperature (aC)

Max wind gjeed (ms-1)

4.5
4

3.5
3

2.5
2

1.5
1

0.5
0

PHSQOOpOOOOQOQOOpOOO
¦R in m	j-\ IjD F. on CJi O -r-i in m ^ k/1 i.Q ^ 00 CTi O

Inventory (head)

Figure 8-20. Maximum values of wind speed and temperature for each inventory level at which the
particulate matter equations yields negative emissions.

Visualizations of the results for PMio (top), TSP (middle) and PM2.s (bottom).

8-28


-------
Deliberative, draft document - Do not cite, quote, or distribute

8.2.3.4 Lagoon

The lagoon analysis for gaseous pollutants is presented in Figure 8-21. Both NH3 and
H2S will produce negative emission values when temperatures dip below -11.8°C. EPA will
evaluate whether the model should include a "floor", that is past a certain temperature it is
assumed the lagoon is frozen and is producing minimal emissions. The relationship between
temperature and emissions is positive with no large changes in emission sensitivity.

Dairy-lagoon H2S emissions versusTemperature	Dairy - lagoon NH3 emissions versus Temperature

Ambient temperature (#C)	Ambient temperature (*C)

Figure 8-21. Lagoon limitation tests for gaseous pollutants.

Visualization of the results for tests of ambient temperature for H2S (left) and NH3 (right).
8.2.3.5 Corral

The corral analyses for H2S and NH3 are presented in Figure 8-22 and Figure 8-23,
respectively. Neither the FhS nor the NH3 model produce negative emissions under average
conditions. However, analyzing 397, 936 combinations of temperature, relative humidity, and
wind speed, found that the NH3 model will produce negative emission estimates at low
temperatures (<7.8°C) combined with low relative humidities (<46%) and low wind speeds (<3.9
ms"1). Figure 8-24 show that as temperature increases, there is a smaller range of relative
humidity and wind speeds that produce negative emissions. Otherwise, the relationships between
emissions and predictors do not show any rapid changes in emission sensitivity that are causes of
concern.

8-29


-------
Deliberative, draft document - Do not cite, quote, or distribute

20.0
18.0

15 0

Dairy - Corral H2S emissions versus relative
humidity



























E 14.0













12.0

£

10.0

J?

C 8.0

















































1 60















4.0

2.0



























nn













0 20 40 60 80
Ambient relative humidity (%)

100

Figure 8-22. Corral limitation tests for H2S.

Visualization of the results for tests of relative humidity for H2S.

Dairy - Corral NH3 emissions versus Temperature

	 1;4G—|	

Dairy - Corral NH3 emissions versus relative
humidity

Dairy - Corral NH3 emissions versus wind speed

-30 -20

0 10 20

Ambient temperature (*C)

Ambient relative humidity (%)

0.00 2.00

4.00	6.00	8.00 10.00 12.00

Wind Speed (ms-1)

Figure 8-23. Corral limitation tests for NH3.

Visualization of the results for NH3 tests of ambient temperature (left), relative humidity (center), and wind speed (right).

8-30


-------
Deliberative, draft document - Do not cite, quote, or distribute

Corral - NH3

-Max of RealtVe humidity (96)
-Max of wind speed

UO 00 CTl

s a ¦ S

Ambient Temperature (*C)

Figure 8-24. Maximum values of wind speed and relative humidity for each temperature at which
the particulate matter equations yields negative emissions.

8.3 Comparison to Literature

To further validate the EEMs developed under this effort, EPA compared the results for
the emission models to the emissions calculated using emission factors found in literature. EPA
scanned the literature for a variety of emission factors for this comparison. EPA selected a
variety of recent factors not derived from the NAEMS for comparison, which are summarized
separately for barns, lagoons, and corrals in Table 8-15, Table 8-16, and Table 8-17,
respectively. There were no emission factors identified for milking centers during the literature
review. For the mechanically ventilated barns, the original units provided in Teye, F.K and
Hautala, M. (2010) were g m"2 hr"1, which were converted to kg hd"1 yr"1 based on the reported
floor area of 774 m2 and inventory of 65 head. For naturally ventilated barns, values were
converted based on 500 kg AU"1, and an average weight of 635 kg per head, based on the
NAEMS farms. For the lagoon and corral sources, surface areas in hectare were converted using
the standard factor of 10000 m2/ha. These converted emission factors were then applied to the
theoretical farm sources from the previous example calculations. The following sections
summarize the results for each source type.

Table

3-15. Emission factors for dairy barns from literature

Source

Farm Source

Pollutant

mg
sec1 hd1

Hg
sec1 hd1

kg
hd1 d1

gm2
hr

kg hd 1
yr"1

Teye, F.K and
Hautala, M. (2010)

Mechanically
ventilated barn

nh3







0.12a

12.52

Huang (2017)

Naturally
ventilated barn

nh3

0.98a







30.91

Leytem, et al.
(2012)

Naturally
ventilated barn

nh3





0.08a



29.20

Huang (2017)

Naturally
ventilated barn

h2s



18.5a





0.58

Sas reported in source.

8-31


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 8-16. Emission factors for dairy lagoons from literature.

Source

Farm Source

Pollutant

kg/ha-d

g/m2-d

kg/m2-yr

Leytem, A.B., et al. (2011)

Lagoon3

NH3



2.0b

0.73

Leytem, A.B., et al. (2018)

Lagoon

NH3

43a,c



1.57

a Identified in the study as a wastewater pond
bas reported in source.

crate reported for lagoon associated with a freestall barn (location ID D4)

Table 8-17. Emission factors for dairy corrals from literature.

Source

Farm Source

Pollutant

g/hd-d

kg/hd-d

Leytem, A.B., et al. (2011)

Corral

nh3



0.13a

Moore, K.D., (2014)

Corral

NHb

134.2a

0.134

Bonifacio, H.F., et al. (2015)

Corral

nh3

155a

0.155

aas reported in source.

8.3.1 Mechanically Ventilated Barn

Comparisons were made for an inventory of 500 cows and 1,000 cows for both a cold
weather location (Wisconsin) and a warm weather location (California). The results for
comparing the calculations for NH3 emissions for mechanically ventilated scrape barns are
presented in Table 8-18, and flush barn in Table 8-19. For both inventory levels, the emission
factor from Teye and Hautala (2010) produces an estimate that falls just below the estimate
produced by the emission models developed in this report. For the flush barns, the estimates
based on Teye and Hautala (2010) fall between the estimate for the smaller barn (500 head) and
just below the model estimates for the larger barn (1,000). For both manure management types,
the models developed in the text represent an increase from previously published literature.

Table 8-18. Comparison of resulting mechanically ventilated scrape barn NH3
emission from various estimation methods.

Meteorology
site

Inventory
(hd)

NH3 Emissions (kg yr"1)

EPA 2022 models

Teye and Hautala (2010)

Wl

500

7,098

6,259

CA

500

8,689

6,259

Wl

1000

18,794

12,517

CA

1000

22,657

12,517

Table 8-19. Comparison of resulting mechanically ventilated flush barn NH3
emission from various estimation methods.

Meteorology
site

Inventory
(hd)

NHb Emissions (kg yr1)

EPA 2022 models

Teye and Hautala (2010)

Wl

500

6,183

6,259

CA

500

7,597

6,259

Wl

1000

16,574

12,517

CA

1000

20,006

12,517

8-32


-------
Deliberative, draft document - Do not cite, quote, or distribute

8.3.2 Naturally Ventilated Barn

Like the mechanically ventilated examples, comparisons were made for an inventory of
500 cows and 1,000 cows for both a cold weather location (WI) and a warm weather location
(CA). The results for NH3 are presented in Table 8-20. For the smaller barn (500 head), the
estimates for both the cold and warm meteorological conditions fall well below the estimates
generated by the factors from literature. The estimates for the larger barn (1,000) the models
presented in this work are closer to the estimates provided by emission factors from literature.
This reiterates the results from the sensitivity analysis, where the emission estimates from the
models increase rapidly with size.

For FhS (Table 8-21), the estimates based on the models developed in this report are
slightly greater for the smaller barn in a cold climate compared to literature. The large inventory
examples and the 500 head barn in a warm climate are slightly lower than estimates based on
literature.

Table 8-20. Comparison of resulting naturally ventilated barn NH3 emission from

various estimation methods.

Meteorology
site

Inventory
(hd)

NH3 Emissions (kg yr"1)

EPA 2022 models

Huang(2017)

Leytem, et al. (2012)

WI

500

4,194

15,453

14,600

CA

500

3,816

15,453

14,600

WI

1,000

28,137

30,905

29,200

CA

1,000

26,050

30,905

29,200

Table 8-21. Comparison of resulting naturally ventilated barn H2S emission from

various estimation methods.

Meteorology
site

Inventory
(hd)

H2S Emissions (kg yr"1)

EPA 2022 models

Huang (2017)

WI

500

310

292

CA

500

289

292

WI

1,000

477

583

CA

1,000

447

583

8.3.3 Lagoon

For lagoons, comparisons were made for both a cold weather location (WI) and a warm
weather location (CA) assuming a surface area of 10,000 m2 The NH3 results in Table 8-22
show the models developed in this report generate an estimate that falls between the factors from
literature.

8-33


-------
Deliberative, draft document - Do not cite, quote, or distribute

Table 8-22. Comparison of resulting dairy lagoon NH3 emission from various

estimation methods.

Meteorology
site

Surface
Area (m2)

NH3 Emissions (kg yr-1)

EPA 2022
models

Leytem, A.B.,
et al. (2011)

Leytem, A.B.,
et al. (2018)

WI

10,000

8,961

7,300

15,695

CA

10,000

12,525

7,300

15,695

8.3.4 Corral

For corrals, the comparison was made for both cold (WI) and warm (CA) meteorological
scenarios. Calculations were also made for a small farm (500 head) and a larger farm (1,000
head), assuming a surface area of 10,000 m2 for each farm for the method developed in this
report. The summary for NH3 in Table 8-22 shows the estimates based on the EPA 2022 draft
methods are comparable to the estimates based on emission factors from literature.

Table 8-23. Comparison of resulting dairy corral NH3 emission from various

estimation methods.

Meteorology
site

Inventory
(hd)

Surface
Area (m2)

NH3 Emissions (kg yr-1)

EPA 2022
models

Leytem, A.B.,
et al. (2011)

Moore, K.D.,
et al. (2014)

Bonifacio, H.F.,
et al. (2015)

WI

500

10,000

23,975

23,725

28,288

24,492

CA

500

10,000

22,551

23,725

28,288

24,492

WI

1000

10,000

47,949

47,450

56,575

48,983

CA

1000

10,000

45,101

47,450

56,575

48,983

8.4 Replication of Independent Measurements

A final test of the developed emission models is to compare the predicted emissions to
observed values from an independent study. For this test, EPA was able to obtain some of the
data from the Harper, et al. (2009) study of lagoons in Wisconsin. The data available are for NH3
emissions for two of the three sites, for fall and summer monitoring periods. EPA was also able
to obtain data from the Leytem et al. (2013) study, where an open-freestall production facility
was monitored in southern Idaho. Measurements were collected for both the open-freestall area
and the wastewater ponds. The data from the Idaho open-freestall area was used to test the corral
model and data from the Wisconsin lagoons and the Idaho wastewater pond data was used to test
the lagoon model.

The data provided included the necessary information to estimate emissions using the
developed emission models. These estimates were then compared to the observed values, when
available, using the same model performance statistics noted in Section 6 of the main report.
Scatter plots were also developed to present the ordered pairs with observations on the x-axis and
the model predicted values on y-axis. These plots are useful for indicating trends of either over-,

8-34


-------
Deliberative, draft document - Do not cite, quote, or distribute

or under-prediction across the range of values. The plots include the 1:1 line (solid line) and the
1:0.5 and 1:2 lines (dashed lines). Points that fall on the 1:1 line were predicted correctly, and
points that fall between the 1:0.5 and 1:2 are within a factor of two observations. Good model
performance would be indicated by scatter contained within a factor of two of the 1:1 line, that is
between the 1:0.5 and 1:2 lines. Looking for scatter confined to within a factor of two of the
observation has been used as a model performance metric in air quality modeling by EPA for
some time (Chang & Hanna, 2004) and continues to be included in EPA's Atmospheric Model
Evaluation Tool (Appel, et al. 2011), which is the current model evaluation platform. The
following sections summarize the result for each source type.

8.4.1 Lagoon

The model performance statistics (Table 8-24) indicate an under-prediction of emissions
at both sites. Figure 8-25 shows that the largest under-predictions occur for observations greater
than 10 g d"1 1000 hd"1, as indicated by the drop below the 1:1.05 line on the plot for the Idaho
site. This suggests the current formulation of the model underestimates the highest emissions.

Table 8-24. Model performance evaluation statistics for lagoon NH3 estimates.

Site

n

LNMEa
(%)

NMEb
(%)

MEb
(g d 11000 hd"1)

MBb
(g d 11000 hd"1)

NMBb
(%)

Corr.

ID

2

3

26.177

69.196

4.800

-4.681

-67.47

0.497

Wl

3

20.271

48.388

3.209

-3.209

-48.39

0.999

a Based on transformed data (i.e., In(NHs)).
b Based on back-transformed data.



Dairy NH3 Model Correlation



Dairy NH3 Model Correlation

20

.



20-

..



¦5" ,





¦5" ,





i

/



i

/



1





w











X

~











J'



1

....



1

\



1





1





1 5

/o



1 5

/ 0



5

O

o\
0

0 0
00

0

0



5





0





0

0





0 5 10 15 20
Observed Normalized Normalized NH3 (g/d/1000hd)

O Pred_NH3



0 5 10 15 20
Observed Normalized Normalized NH3 (g/d/1000hd)

O Pred_NH3

Figure 8-25. Scatter plot of the observed lagoon NH3 emissions versus the emission model
estimates.

Results from the Idaho site (left) and Wisconsin site (right).

8.4.2 Corral

The model performance statistics (Table 8-25) show an under-prediction of emissions
from the corral. The plot of observed versus estimated emissions (Figure 8-26) show there are

8-35


-------
Deliberative, draft document - Do not cite, quote, or distribute

slight overpredictions at low emission levels, as the points fall above the 1:1 line, and an
underprediction at higher observed emission levels. As with the lagoon model, this suggests an
underprediction of highest emission values in the model.

Table 8-25. Model performance evaluation statistics for corral NH3 estimates.

Site

n

LNMEa (%)

NMEb (%)

MEb
(g d11000 hd"1)

MBb
(g d11000 hd"1)

NMBb (%)

Corr.

Wl

18

17.371

70.689

1.316

-0.574

-30.84

-0.351

a Based on transformed data (i.e., In(NHs)).
b Based on back-transformed data.

Dairy Corral Normalized NH3 Model Correlation

0 -¦	j

0	2	4	6	8

Observed Normalized NH3 (g/d/1 OOOhd)

O NH3_preci_norm

Figure 8-26. Scatter plot of the observed corral NH3 emissions versus the emission model
estimates.

8-36


-------
Deliberative, draft document - Do not cite, quote, or distribute

9 CONCLUSIONS

Consistent with the Air Compliance Agreement with the AFO industry, EPA has
developed emission estimation methods for NH3, H2S, PM10, PM2.5, and TSP for confinement
and open sources associated with dairy operations. These draft statistical models focus on
parameters that have been identified in published peer-reviewed journals as having empirical
relationships with emissions. These relationships were evaluated within the NAEMS dataset
before selecting parameters for emission model development. EPA also considered which
variables could be measured or obtained with minimal effort.

The inventory was identified as a key parameter and is used in all the models as a proxy
for the volume of manure generated. Temperature and relative humidity parameters were also
identified as important variables for emission rates in the barn emission models. Relative
humidity parameters proved to be key for particulate matter prediction, as the higher moisture
levels keep barn materials from entraining into the air with mechanical disruptions. Confinement
parameters specific to the barn, like exhaust temperature, showed promise as predictive
parameters. However, these parameters are not routinely measured at farms and would therefore
represent an increased burden to operators should they be required for emissions estimation. As
such, all of the draft dairy emission models put forward for potential future use in this document
use parameters that are already routinely collected as part of the standard farm operation (e.g.,
inventory) or are ambient meteorological parameters, which are freely available from public
sources such as National Center for Environmental Information (NCEI,
http s://gis.ncdc.noaa. gov/map s/).

Overall, the method used to develop the emission models allows for the incorporation of
additional emissions and monitoring datasets from other studies, should they become available to
EPA after the release of the emission models. Revised emission models for any individual farm
type could be issued once significant additional data becomes available. Similarly, if monitoring
options for barn parameters become more widespread as automation options grow, future
evaluations could assess whether emission models should be developed to include these
parameters.

EPA recognizes the scientific and community desire for process-based models. The data
collected during NAEMS, and the emission models developed here lay the groundwork for
developing these more process-related emission estimates. EPA supports the future development
of process-based models which account for the entire animal feeding process. While the interim
statistical models allow estimation of emissions from barns and open sources at dairy operations
across the U.S., process-based models would allow producers to estimate the impacts of different
management practices to reduce air emissions, helping to incentivize change.

9-1


-------
Deliberative, draft document - Do not cite, quote, or distribute
10 REFERENCES

Aarnink, A.J.A., Ellen, H.H., 2007. Processes and factors affecting dust emissions from livestock
production. In: Proceedings of DustConf 2007. How to Improve Air Quality.

International Conference, Maastricht, The Netherlands

Appel, K.W., Gilliam, R.C., Davis, N., Zubrow, A., and Howard, S.C. (2011). Overview of the
Atmospheric Model Evaluation Tool (AMET) vl.l for evaluating meteorological and air
quality models, Environ. Modell. Softw., 26(4), 434-443.

Arogo, J., Zhang, R., Riskowski, G. L., Christianson, L. L., & Day, D. L. (1999). Mass transfer
coefficient of ammonia in liquid swine manure and aqueous solution. Journal of
Agricultural Engineering Research, 73(1), 77e86.

Arogo, lactone, Philip W. Westerman, Albert J. Heber, Wayne P. Robarge, and John J. Classen.
Ammonia Emissions from Animal Feeding Operations. Animal Agriculture and the
Environment, National Center for Manure & Animal Waste Management White Papers.
St. Joseph, MI: ASABE, 2006. doi:https://doi.org/10.13031/2013.20247.

Arya, S.P. (1999). Air pollution meteorology and dispersion. Oxford University Press. Blunden,
J., & Aneja, V.P. (2008). Characterizing ammonia and hydrogen sulfide emissions from a
swine waste treatment lagoon in North Carolina. Atmospheric Environment 42, 3277-
3290.

B. Bjerg, T. Norton, T. Banhazi, G. Zhang, T. Bartzanas, P. Liberati, G. Cascone, I.-B. Lee, and
A. Marucci. "Modelling of ammonia emissions from naturally ventilated livestock
buildings. Part 1: Ammonia release modelling" Biosystems engineering 116, no. 3
(2013): 232-245. doi: 10.1016/j.biosystemseng.2013.08.001

Bogan, B.W., A. Chandrasekar, S. McGlynn, C.A. Gooch, and A.J. Heber. 2010. National Air

Emissions Monitoring Study: Data from Dairy Freestall Barn and Milking Center in New
York- Site NY5B. Final Report. Purdue University, West Lafayette, IN, July 2

Bonifacio, Henry & Rotz, C.A. & Leytem, A.B. & Waldrip, Heidi & R.W.Todd. (2015).

Process-based modeling of ammonia and nitrous oxide emissions from open-lot beef and
dairy facilities. Transactions of the ASABE (American Society of Agricultural and
Biological Engineers). 58. 827-846.

Bougouin, A., Leytem, A., Dijkstra, J., Dungan, R. S., & Kebreab, E. (2016). Nutritional and
Environmental Effects on Ammonia Emissions from Dairy Cattle Housing: A Meta-
Analysis. Journal of environmental quality, 45(4), 1123-1132.

Cambra-Lopez, M., Aarnink, A.J.A., Zhao, Y., Calvet, S., & Torres, A.G. (2010). Airborne
particulate matter from livestock production systems: A review of an air pollution
problem. Environmental Pollution 158, 1-17.

Cambra-Lopez, M., Torres, A. G., Aarnink, A. J. A., & Ogink, N. W. M. (2011). Source analysis
of fine and coarse particulate matter from livestock houses. Atmospheric Environment,
45(3), 694-707. doi:10.1016/j.atmosenv.2010.10.018

Chang, J., Hanna, S. Air quality model performance evaluation. Meteorol Atmos Phys 87, 167-
196 (2004). https://doi.org/10.1007/s00703-003-0070-7

10-1


-------
Deliberative, draft document - Do not cite, quote, or distribute

Christensen J.R., Nash, M.S., Chaloud D., and Pitchford A. 2016. Spatial Distribution of Small
Water Body Types Across Indiana Ecoregions. Ecohydrology. 9,122-137. Available at:
https ://cfpub. epa.gov/si/si_public_record_report. cfm?dirEntryId=3 08263&Lab=NERL

Cortus, E.L., L. D. Jacobson, B.P. Hetchler, and A.J. Heber. 2010. National Air Emissions

Monitoring Study: Emissions Data from Two Freestall Dairy Barns in Wisconsin - Site
WI5B. Final Report. Purdue University, West Lafayette, IN, July 2.

Dandan Huang & Huiqing Guo (2017) Diurnal and seasonal variations of odor and gas emissions
from a naturally ventilated free-stall dairy barn on the Canadian prairies, Journal of the
Air & Waste Management Association, 67:10, 1092-1105,

DOI:l 0.1080/10962247.2017.1329172

Feidler, A.M., Miiller, H.J., 2011. Emissions of ammonia and methane from a livestock building
natural cross ventilation. Meteorologische Zeitschrift 20 (1), 059e065.

Garcia, J., Bennett, D. H., Tancredi, D., Schenker, M. B., Mitchell, D., & Mitloehner, F. M.
(2013). A survey of particulate matter on California dairy farms. Journal of
environmental quality, 42(1), 40-47.

Grant, R.H. 2008, Quality Assurance Project Plan for the National Air Emissions Monitoring
Study (Open Source Emissions Component), Ver. 4, Purdue Applied Meteorology
Laboratory, Department of Agronomy, Purdue University

Grant, R.H. and M.T. Boehm. (2010a). National Air Emissions Monitoring Study: Data from the
eastern US milk production facility IN5A. Final Report to the Agricultural Air Research
Council. Purdue University, West Lafayette, IN, June 4.

Grant, R.H. and M.T. Boehm. (2010b). National Air Emissions Monitoring Study: Data from the
southwest milk producer facility TX5 A. Final Report to the Agricultural Air Research
Council. Purdue University, West Lafayette, IN, July 2.

Grant, R.H. and M.T. Boehm. (2010c.) National Air Emissions Monitoring Study: Data from the
northwestern US milk production facility WA5A. Final Report to the Agricultural Air
Research Council. Purdue University, West Lafayette, IN, July 2.

Grant, R.H. and M.T. Boehm. (2010d). National Air Emissions Monitoring Study: Data from the
midwestern US milk production facility WI5A. Final Report to the Agricultural Air
Research Council. Purdue University, West Lafayette, IN, July 2.

Grant, R. H., Boehm, M. T., & Lawrence, A. F. (2013). Comparison of a backward-Lagrangian
stochastic and vertical radial plume mapping methods for estimating animal waste lagoon
emissions. Agricultural and Forest Meteorology, 180, 236-248.
doi:10.1016/j.agrformet.2013.06.013

Grant, R.H., Boehm, M.T., Lawrence, A.F., & Heber, A.J. (2013b). Ammonia emissions from
anaerobic treatment lagoons at sow and finishing farms in Oklahoma. Agricultural and
Forest Meteorology 180, 203-210.

Grant, R.H. and Boehm, M.T (2015). Manure ammonia and hydrogen sulfide emissions from a
western storage basin. Journal of environmental quality, 45, 127-136.

Grant, R. H., & Boehm, M. T. (2020). Ammonia Emissions from Differing Manure Storage
Facilities at Two Midwestern Free-Stall Dairies. Atmosphere, 11(10), 1108.

10-2


-------
Deliberative, draft document - Do not cite, quote, or distribute

Grant, R. H., Boehm, M. T., & Hagevoort, G. R. (2020). Ammonia Emissions from a Western
Open-Lot Dairy. Atmosphere, 11(9), 913.

Harper, L.A. , T.K. Flesch, J.M. Powell, W.K. Coblentz, W.E. Jokela, N.P. Martin, Ammonia
emissions from dairy production in Wisconsin, Journal of Dairy Science, Volume 92,
Issue 5, 2009, Pages 2326-2337, ISSN 0022-0302, https://doi.org/10.3168/jds.2008-1753.

Heber A.J., Ni, J-Q., Ramirez, J.C., Schrock, W., & Elkins, J. (2008). Quality assurance project
plan for the National Air Emissions Monitoring Study (barn component). Purdue
University.

Herbut, P., & Angrecka, S. (2014). Ammonia Concentrations in a Free-Stall Dairy Barn. Annals
of Animal Science, 14(1), 153-166. doi:10.2478/aoas-2013-0065

Joo, H., Ndegwa, P.M., Wang, X., Heber, A.J., Ni, J.Q., Cortus, E.L., Ramirez-Dorronsoro, J.C.,
Bogan, B.W. and Chai, L., 2015. Ammonia and hydrogen sulfide concentrations and
emissions for naturally ventilated freestall dairy barns. Transactions of the
ASABE, 58(5), pp. 1321-1331.

Kavolelis, B. (2003). Influence ventilation rate on ammonia concentration and emission in
animal house. Polish Journal of Environmental Studies, 12(6), 709-712.

Leeden, Rien & Meijer, Erik & Busing, Frank. (2008). Resampling Multilevel Models.
10.1007/978-0-387-73186-5 11.

Leytem, A. B., Dungan, R. S., Bjorneberg, D. L., & Koehn, A. C. (2011). Emissions of

ammonia, methane, carbon dioxide, and nitrous oxide from dairy cattle housing and
manure management systems. Journal of environmental quality, 40(5), 1383-1394.
https://doi.org/10.2134/jeq2009.0515

Leytem, A. B., Dungan, R. S., Bjorneberg, D. L., & Koehn, A. C. (2013). Greenhouse gas and
ammonia emissions from an open-freestall dairy in southern Idaho. Journal of
Environmental Quality, 42, 10-20.

Leytem, A. B., Bjorneberg, D. L., A1 Rotz, C., Moraes, L. E., Kebreab, E., & Dungan, R. S.

(2018). Ammonia emissions from dairy lagoons in the western US. Transactions of the
ASABE, 61(3), 1001-1015.

Lim, T.T., Y. Jin, J.H. Ha, and A.J. Heber. 2010. National Air Emissions Monitoring Study:
Emissions Data from Two Freestall Barns and a Milking Center at a Dairy Farm in
Indiana - Site IN5B. Final Report. Purdue University, West Lafayette, IN, July 30.

Maasikmets, Marek, Erik Teinemaa, Allan Kaasik, Veljo Kimmel, Measurement and analysis of
ammonia, hydrogen sulphide and odour emissions from the cattle farming in Estonia,
Biosystems Engineering, Volume 139, 2015, Pages 48-59, ISSN 1537-5110,
https://doi.Org/10.1016/j.biosystemseng.2015.08.002.

Meisinger, J. J., & Jokela, W. E. (2000). Ammonia volatilization from dairy and poultry

manure. Proc Managing nutrients and pathogens from animal agriculture. Camp Hill, PA,
28-30.

Montes, F., C.A. Rotz, and H. Chaoui. 2009. Process modeling of ammonia volatilization from
ammonium solution and manure surfaces. Trans. ASABE 52(5): 1707-1719.

10-3


-------
Deliberative, draft document - Do not cite, quote, or distribute

Moore, K. D.; Young, E.; Gurell, C.; Wojcik, M. D.; Martin, R. S.; Bingham, G. E.; Pfeiffer, R.
L.; Prueger, J. H.; and Hatfield, J. L., "Ammonia Measurements and Emissions from a
California Dairy Using Point and Remote Sensors" (2014). Space Dynamics Lab
Publications. Paper 95.

Mostafa, E., Nannen, C., Henseler, J., Diekmann, B., Gates, R., & Buescher, W. (2016). Physical
properties of particulate matter from animal houses-empirical studies to improve
emission modelling. Environ Sci Pollut Res Int, 23(12), 12253-12263.
doi: 10.1007/sl 1356-016-6424-8

Ngwabie, N. M., K. H. Jeppsson, G. Gustafsson, and S. Nimmermark. "Effects of Animal

Activity and Air Temperature on Methane and Ammonia Emissions from a Naturally
Ventilated Building for Dairy Cows." Atmospheric Environment 45, no. 37 (Dec 2011):
6760-68. https://doi.Org/10.1016/j.atmosenv.2011.08.027.

Philippe, F. X., Cabaraux, J. F., & Nicks, B. (2011). Ammonia emissions from pig houses:

Influencing factors and mitigation techniques. Agriculture, Ecosystems & Environment,
141(3e4),245e260. http://doi.Org/10.1016/j.agee.2011.03.012.

Samer, M., C. Ammon, C. Loebsin, M. Fiedler, W. Berg, P. Sanftleben, and R. Brunsch.
"Moisture balance and tracer gas technique for ventilation rates measurement and
greenhouse gases and ammonia emissions quantification in naturally ventilated
buildings." Building and Environment 50 (2012): 10-20.

Ramirez-Dorronsoro, J.C., H.S. Joo, P. Ndegwa, and A.J. Heber. 2010. National Air Emissions
Monitoring Study: Data from Two Dairy Freestall Barns in Washington WA5B, Final
Report. Purdue University, West Lafayette, IN, July 30.

Rotz, C Alan, Dawn S Chianese, Felipe Montes, Sasha Hafner, and Henry F Bonifacio. "Dairy
Gas Emissions Model." (2014).

Saha, C. K., C. Ammon, W. Berg, M. Fiedler, C. Loebsin, P. Sanftleben, R. Brunsch, and T.

Amon. "Seasonal and diel variations of ammonia and methane emissions from a naturally
ventilated dairy building and the associated factors influencing emissions." Science of the
total environment 468 (2014): 53-62.

Sanchis, E., Calvet, S., del Prado, A., & Estelles, F. (2019). A meta-analysis of environmental
factor effects on ammonia emissions from dairy cattle houses. Biosystems Engineering,
178, 176-183. doi: 10.1016/j.biosystemseng.2018.11.017

Schmithausen, A. J., Schiefler, I., Trimborn, M., Gerlach, K., Sudekum, K. H., Pries, M., &

Buscher, W. (2018). Quantification of Methane and Ammonia Emissions in a Naturally
Ventilated Barn by Using Defined Criteria to Calculate Emission Rates. Animals, 8(5),
75. doi:ARTN 7510.3390/ani8050075

Schrade, S., Zeyer, K., Gygax, L., Emmenegger, L., Hartung, E., & Keck, M. (2012). Ammonia
emissions and emission factors of naturally ventilated dairy housing with solid floors and
an outdoor exercise area in Switzerland. Atmospheric Environment, 47, 183-194.
doi: 10.1016/j.atmosenv. 2011.11.015

Shi, Zhifang, Xiaoqin Sun, Yao Lu, Lei Xi, and Xin Zhao. "Emissions of ammonia and hydrogen
sulfide from typical dairy barns in central China and major factors influencing the

10-4


-------
Deliberative, draft document - Do not cite, quote, or distribute

emissions." Scientific Reports 9, no. 1 (2019): 1-12. https://doi.org/10.1038/s41598-019-
50269-y

Sommer, S.G., Christensen, M.L., Schmidt, T., Jensen, L.S., 2013. Animal Manure Recycling:
Treatment and Management. John Wiley Sons Ltd, West Sussex, UK.

Takai H., Pedersen S., Johnsen J.O., Metz J.H.M., Koerkamp P.W.G.G., Uenk G.H., Phillips

V.R., Holden M.R., Sneath R.W., Short J.L., White R.P., Hartung J., Seedorf J., Schroder
M., Linkert K.H., Wathes C.M. 1998. Concentrations and emissions of airborne dust in
livestock buildings in Northern Europe. J. Agr. Eng. Res., 70: 59-77. doi: 10.1006/
jaer. 1997.0280

Teye, F. K., & Hautala, M. (2010). A comparative assessment of four methods for estimating

ammonia emissions at microclimatic locations in a dairy building. International journal of
biometeorology, 54(1), 63-74. https://doi.org/10.1007/s00484-009-0255-y

U.S. EPA Science Advisory Board (SAB), 2013. SAB Review of Emissions-Estimating

Methodologies for Broiler Animal Feeding Operations and for Lagoons and Basins at
Swine and Dairy Animal Feeding Operations, EPA-SAB-13-003.

Wu, W., Zhang, G., & Kai, P. (2012). Ammonia and methane emissions from two naturally
ventilated dairy cattle buildings and the influence of climatic factors on ammonia
emissions. Atmospheric Environment, 61, 232-243. doi:10.1016/j.atmosenv.2012.07.050

Zhao, Y., F.M. Mitloehner, L. Chai, J.C. Ramirez, K. Wang, J. Ni, C. Diehl, E.L. Cortus, T.T.
Lim, B.W. Bogan, I. Kilic, and A.J. Heber. 2010. National Air Emissions Monitoring
Study: Data from Two Dairy Freestall Barns in California - Site CA5B. Final Report.
Purdue University, West Lafayette, IN, July 30.

10-5


-------