Introduction
Government regulations of greenhouse gas (GHG) emissions have
evolved rapidly in the past 5 years, particularly in the US.
The US pledged to decrease its GHG emissions by 26–28 % relative to 2005
levels by 2025 as part of the Paris Agreement negotiated at COP21
. In parallel with this agreement, the US Environmental
Protection Agency (EPA) has finalized CO2 and CH4 emission regulations
for numerous source sectors under the White House Climate Action Plan
. Several US states have also taken aggressive action
on emissions, including Massachusetts and California
, among others.
These policy actions require scientists and government agencies to quantify
regional- and national-scale GHG emissions from specific source sectors. In
this paper, we define a source sector as the total emissions from an
industry, such as CO2 from power plants, CH4 from the oil and natural
gas industries, or CH4 from landfills. This review paper focuses on
existing and evolving capabilities for the US. The US has far
greater resources to estimate emissions relative to many developing
countries. Furthermore, GHG emission regulations in the US are nascent
relative to regulations in Europe e.g.,, and the
monitoring strategies discussed in this review could be developed in parallel
with new regulations.
Many national emission regulations in the US target this sector level (Note: the new
presidential administration that assumed office in January 2017 has announced
its intention to discard several of these regulations.). For
example, the US Clean Power Plan mandates a 32 % decrease in power sector
CO2 emissions by 2030 relative to 2005 levels . The EPA and National
Highway Traffic Safety Administration have also extended and strengthened
CO2 emission standards for cars and light trucks through 2025
. In addition to these measures, the EPA has set several
sector-specific CH4 emission targets. In May of 2016, the EPA issued a rule
that will decrease CH4 emissions from oil and gas operations by 40–45 %
relative to 2012 levels by 2025 . In August of 2014, the US
EPA, US Department of Agriculture (USDA), and US Department of Energy (DOE)
released the Biogas Opportunities Roadmap targeting voluntary
reduction strategies for agriculture . Lastly,
the EPA announced regulations for CH4 emissions from landfills in July 2016
. It is important to note that a number of these national
policies are implemented at the state level. For example, each state has a
different emission reduction target under the Clean Power Plan, and each
state can decide how to meet and monitor progress toward that target
.
We examine sector-specific GHG estimates with an eye toward combining or
assimilating multiple data streams. This review article is part of a special
issue of the European Geophysical Union (EGU) journals that focus on data
assimilation and the use of multiple data streams to understand the carbon
cycle. In this context, we explore opportunities to creatively synthesize
both bottom-up emission inventories and top-down atmospheric inverse
modeling. Most government agencies estimate emissions using bottom-up
inventories, which quantify total emissions by estimating the total amount of some
activity and the average emissions per unit of activity. Other efforts
utilize top-down atmospheric inverse modeling, which measures atmospheric GHG mixing
ratios and use those measurements to infer the level and distribution of
emissions at the Earth's surface. In the future, scientists and government
agencies will likely need to combine these approaches into frameworks that can synergistically leverage the
information content of bottom-up datasets and top-down strategies using
atmospheric GHG data to robustly estimate
sector-specific emissions. This review paper focuses on these opportunities.
These frameworks will need to address two key tasks: estimating the total
quantity of GHG emissions from each source type and detect changes or trends
in emissions from that source type. From the standpoint of inverse modeling,
the former problem is more challenging than estimating total emissions and
requires separating the space–time patterns of one emission source from the
patterns of other sources. In the latter case, we not only need to estimate a
trend in total emissions but also to attribute this trend to trends in
specific source sectors. This challenge is complicated by changes in
technology and changes in the spatial or temporal distribution of individual
source sectors. For example, hydraulic fracturing and horizontal drilling
became widely used in the past decade . These operations
utilize new equipment and operational practices, and the spatial distribution
of drilling across the US has changed during that time; these
emissions are literally a “moving target”.
These challenges are further complicated by GHG fluxes from the biosphere,
particularly in the case of CO2. Biospheric and fossil fuel sources will
be important to disaggregate from one another for sound policy evaluation.
These sources are often colocated and trends in one could be mistaken for
trends in the other. In addition, future changes in biospheric CO2 and
CH4 sources may be natural or human-caused (e.g., land use change,
emissions induced by climate change, biological and/or geological carbon
sequestration). Disentangling these natural and human causes will be
challenging. Note that GHG fluxes from the biosphere and
biological–geological carbon sequestration are beyond the scope of this
review.
In this article, we explore the challenge of estimating sector-specific
emissions from several perspectives. First, we discuss bottom-up inventory
efforts. We then explore top-down strategies to estimate sector-specific
emission and the atmospheric datasets available to make both bottom-up and
top-down estimates. Next, we highlight several new or novel approaches for
estimating sector-specific emissions, and lastly, we close the review with a
synthesis discussion of forward-looking opportunities for combining bottom-up
and top-down strategies.
Bottom-up data
Bottom-up efforts typically use an accounting-type approach to estimate
sector-specific emissions. The first step usually involves collecting
activity data: a map or database of economic activity or behavior that leads
to emissions. Examples include the amount of coal burned by power plants, the
number of passenger cars and miles traveled, and the number of cows by
location. A second step entails estimating a set of emission factors (EFs)
for each activity. EFs could include the CO2 emissions per kilogram of coal
burned or the average CO2 emissions per mile traveled by passenger cars.
The product of these two numbers provides a bottom-up estimate of emissions
for a given source sector. State and national governments in the US use this
strategy to construct official emission estimates
e.g.,. A number of academic and government
efforts have produced bottom-up CO2 and CH4 emission estimates at
local–regional e.g.,,
national
e.g.,,
and global scales
e.g.,. In this
section, we primarily discuss bottom-up data with an eye toward how this
information can be combined with top-down strategies.
A prototypical example
We describe the EPA's estimate of CO2 emissions from coal-fired power plants
as a prototypical example of how government agencies construct bottom-up
inventory estimates. The EPA describes the procedure that it uses to estimate
CO2 emissions in compliance with 2006 IPCC guidelines :
first, the agency estimates activity data – coal use by source sector. The EPA
uses retail statistics from the electricity sector to estimate total
consumption by each type of end user (e.g., residential, commercial).
Second, the EPA adjusts these activity data to account for non-combustion uses,
double-counted emissions, and fuel exports and/or imports. For example, a coal
gasification plant in North Dakota produces synthetic natural gas; this fuel
is added to natural gas activity data and subtracted from the coal activity
data. According to the EPA, “Because this energy of the synthetic natural gas is
already accounted for as natural gas combustion, this amount of energy is
deducted from the industrial coal consumption statistics to avoid double
counting” . Third, the EPA estimates the carbon
content of the coal. The EPA uses Energy Information Administration (EIA)
estimates of carbon content by coal rank and state of origin
. The EPA then computes the weighted average carbon content of
coal by state of origin and estimates the end use of coal produced in each
state (e.g., electricity, industry). The agency uses this procedure to
estimate the average carbon content (and EF) for each end use sector in the
US .
IPCC guidelines also require a reference approach: an additional verification
or consistency check against fuel production, imports, and exports
. The new draft inventory then goes through expert review
undertaken by a panel of technical experts. The EPA revises its inventory
estimate based upon this review and distributes the subsequent draft for
public comment. At the conclusion of that process, the EPA issues its finalized
inventory estimate.
The approach outlined above is similar to many government inventories. More
recently, a number of academic efforts have developed very different
approaches that leverage novel data streams (e.g., satellite images of lights at
night) or that use gridded activity data, and these efforts are described in
detail in the next section.
Recent bottom-up efforts
In the past 10 years, inventory efforts have moved from coarse estimates
that rely heavily on proxy activity data to spatially resolved estimates that
use specific activity data and EFs that are tailored to the heterogeneities
in each emission source.
A number of recent CO2 inventories incorporate more comprehensive activity
data or detailed EFs than previously available. At the regional scale,
and develop on-road CO2 emission
estimates for Indianapolis and Massachusetts, respectively. Emissions in the
latter study are within 8.5 % of Federal Highway Administration fuel
consumption statistics but differ from the commonly used, global-scale EDGAR
inventory by 22.8 % . The authors explain that
many global-scale efforts use road density as a proxy for vehicle emissions
but argue that the relationship between road density and emissions is not
constant. Two subsequent studies estimate
on-road CO2 emissions for the entire US at spatial resolutions
down to 1 km2. 's
emission estimates differ from EDGAR by 20–80 % at the municipal level,
though the two inventories produce nearly identical national totals.
At the national scale, the VULCAN inventory is the most
comprehensive academic effort to date. This inventory includes CO2
emissions by sector at high spatial and temporal resolutions – 10 km × 10 km – sub-daily for the year 2002. Furthermore, VULCAN uses more detailed
activity data than the EPA's national inventory. At the global scale, the EDGAR
anthropogenic emission inventory (available for 1970–2010) has moved from a
1∘×1∘ lat–long resolution to
0.1∘×0.1∘ . In a separate
effort, estimate CO2 emissions for 80 countries for the
years 1950–2006, with a particular focus on estimating the seasonal cycle of
CO2 emissions.
A number of studies incorporate more detailed activity data and EFs to
estimate anthropogenic CH4 emissions at both regional and national scales.
At the regional scale, and estimate oil
and gas CH4 emissions from California for 2010 and the Barnett Shale
region for 2013, respectively. Both studies find emissions that greatly
exceed the EPA's estimates. A relatively small fraction of emitters account for
the majority of oil and gas emissions, and argue that
rigorous EFs capture this skewed distribution more effectively than those
used by the EPA. In addition to these oil and gas inventories,
compile field studies of CH4 emissions from agriculture (e.g., cattle,
sheep, and manure management). The authors explain that current emission
inventories use EFs from lab-based experiments, not field observations. These
field observations imply much higher EFs that result in larger emissions, which are more
in line with existing top-down estimates. At the national scale,
create a gridded version of the EPA's CH4 inventory
(0.1 × 0.1 lat–long, monthly resolution for 2012).
point out that the spatial distribution of their
estimate is different from EDGAR, particularly for the oil and gas
industries. Oil and gas emissions in EDGAR correlate with population density,
while emissions in are concentrated in drilling
basins.
A number of additional studies also employ novel inventory methodology or
novel proxy datasets. For example, develop ODIAC (Open-source
Data Inventory of Anthropogenic CO2), a global gridded CO2
inventory constructed using a database of CO2 point sources and satellite
images of lights at night. and develop a
data assimilation framework known as FFDAS (Fossil Fuel Data Assimilation
System). The authors use datasets like population density, carbon intensity
of energy, and satellite images of lights at night, and they report national
emission totals. use a very different approach from any of
the studies above. The authors build a CO2 estimate based upon economic
imports and exports and explore the idea of carbon “leakage”, the carbon
emitted by one country to manufacture products that are then imported by
another country. These studies do not provide emission estimates for each
individual source sector, but ODIAC and FFDAS do incorporate novel datasets
to separate point sources (e.g., power plants) from non-point emissions.
The EPA's GHG Reporting Program (GHGRP) represents an important advancement in
government inventory efforts. The EPA announced the GHGRP in 2009 and emission
reporting began in 2010 . The GHGRP requires all entities that
emit over 25 000 metric tons of CO2 equivalents to report their emissions
to a national registry . This reporting threshold is
equivalent to the GHG emissions of 3439 homes or 5263 cars
. The agricultural sector is excluded from this threshold and
is not required to report its emissions. Despite these omissions, the EPA
estimates that 85–90 % of US GHG emissions are covered under the GHGRP.
Other recent studies, however, argue that the GHGRP is less complete than
estimated by the EPA for two reasons e.g.,.
First, the emissions that are excluded from the GHGRP are sometimes larger
than estimated by the EPA, and second, the EFs used in the GHGRP are smaller than
actual emissions from some source sectors like oil and natural gas.
Recent, direct measurements that support bottom-up efforts
Inventory development requires two different types of data: activity data and
data that can be used to develop EFs. Activity data can come from economic,
census, and remote sensing datasets, among other possible data sources. These
datasets differ from those used to develop EFs. The IPCC provides a database
of EF estimates but encourages countries to take measurements of emitters or
emitting processes to develop tailored, country-specific EFs
. A number of observation strategies can directly support the
development and evaluation of country-specific EFs. We discuss a number of
recent efforts here as well as the advantages and challenges of using these
datasets.
One observation strategy is to measure GHG mixing ratios near an emitter or a
group of emitters. These observations, by factor of their targeted spatial
scale, can be directly used to evaluate a single source type and develop
corresponding EFs. For example, a number of studies report on direct GHG
measurements from individual facilities. These include direct stack
measurements of power plant CO2 emissions e.g.,
and numerous recent studies of CH4 emissions from oil and gas operations:
measurements of emissions from pneumatic controllers ,
compressor stations , transmission and storage
systems , and abandoned wells . In
addition, several site-level studies target agricultural emissions.
and review several measurement
strategies, and specifically review field studies of CH4
emissions from manure.
On-road measurements provide a picture of emissions that is one spatial scale
larger than direct facility observations. This strategy usually entails
measuring trace gas mixing ratios from a ground-based vehicle either on
public roads e.g., or private roads in partnership with
the facility owner e.g.,. Existing studies often
target oil and gas facilities
e.g.,
and mobile CO2 emissions e.g.,. In
the case of oil and gas emissions, explain that mobile
measurements capture an integrated plume that includes all leaks from a given
facility but rarely indicate which components caused those leaks.
The use of facility-level and on-road observations entails a number of
challenges. For example, facility-level observations provide the most insight
into detailed emission processes from specific source sectors but can miss
emission events or processes. Observations of oil and gas facilities provide
a prime example; scientists may not know about some leaks and therefore may
not measure them, other leaks may be in inaccessible locations
e.g.,, and the largest leaks often come from
ephemeral equipment failures at a small number of facilities that are
difficult to identify e.g.,.
Cost also limits facility-level, continuous emission monitoring; it is
typically only used for large point sources like power plants
.
These observation strategies also require extrapolation to produce state- or
national-scale EF estimates. The relationship between activity data and
emissions can be complex and spatially variable, making it difficult to
extrapolate facility or on-road measurements. For example, CH4 emissions
from oil and gas are likely dominated by a small number of malfunctioning
facilities. As a result, it is difficult to develop robust, national-scale
EFs from a modestly sized sample of facilities .
Furthermore, explain that these leaks do not correlate
with production and can vary greatly in time. Different oil and gas drilling
basins also have different overall leakage rates – from 0.3 % in
Pennsylvania's Marcellus Shale region to 8.9 % in Utah's Uintah Basin
e.g.,. These factors
make it challenging to create consistent, generalizable EFs that can
translate activity data into emissions.
These considerations also apply to other source sectors beyond the oil and
gas industries. For example, grazing and manure management practices differ
by region, and manure and landfill CH4 emissions also differ by climate
, all of which make extrapolation more challenging.
EDGAR and the EPA inventory estimates for different US fossil fuel source sectors ,
including several versions of each inventory. CO2 estimates are consistent between the EPA and EDGAR and among inventory
versions. CH4 estimates, however, vary widely, which is an indication of uncertainty in CH4 emissions. All of the estimates
are for 2005 except for EDGAR FT2000, which is for 2000. Note that EDGAR includes CO2 from heating in its electricity
estimate while the EPA does not. As a result, the EDGAR CO2 estimate is higher than the EPA's estimate.
Impact of recent advances
Inventory estimates of sector-specific CO2 emissions from the US are
likely relatively accurate at the national scale but have substantial
uncertainties at the local and state levels. , for
example, compare smokestack versus fuel-based CO2 estimates for US power
plants and find a mean absolute difference of 16.6 % but only a 1.4 % total
difference at the national scale. Furthermore, find biases
of 100 % or more at the urban scale in CO2 emission estimates for mobile
sources. However, they estimate a US national total that is broadly
consistent with other inventories like VULCAN.
By contrast, sector-specific CH4 emissions are more challenging to
estimate and existing inventories for the US are highly uncertain at state
and national scales. For example, several top-down studies indicate that the
California state inventory is likely too low by a factor of 1.2 to 1.9
, and several top-down studies estimate
emissions for oil and gas drilling regions of Utah and Colorado that are up
to 3 times bottom-up estimates e.g.,.
Overall, total US CH4 emissions are likely ∼50 % larger than
estimated by EDGAR or the US EPA . Figure compares several inventory estimates of sector-specific CO2
and CH4 emissions. Existing CO2 inventory estimates are broadly
consistent, while CH4 estimates vary between inventories and among
inventory versions.
CH4 inventories are so uncertain, in part, because of the complexity of
many anthropogenic CH4 source sectors. For example, emission factors for
oil and gas operations are difficult to estimate because a small number of
emitters often account for a large fraction of emissions
e.g., and
because there are so many points along the natural gas production,
processing, transmission, and distribution cycle that leak methane
e.g.,.
Much of the uncertainty in CH4 inventories stems from difficulties
developing accurate EFs. explains, “... measurements for
generating emission factors are expensive, which limits sample sizes and
representativeness. Many EPA EFs have wide uncertainty bounds. And there are
reasons to suspect sampling bias in EFs, as sampling has occurred at
self-selected cooperating facilities.” For example, the EPA's EFs for natural gas
pipelines are based on a limited number of samples from a 1996 EPA and Gas
Research Institute study; these EFs have uncertainties of ±65 %
. Beyond the oil and gas industry, also argue
that many EFs for agriculture are too low. These estimates are based upon a small
number of pilot or lab experiments that were not explicitly designed for GHG
inventory development.
Top-down, inverse modeling strategies
In this section, we discuss inverse modeling strategies – strategies that
leverage observations of atmospheric GHG mixing ratios to infer emissions at
the Earth's surface. We specifically focus on strategies that attempt to
parse the contribution of specific source sectors. The first part of this
discussion (Sect. –) focuses on efforts
at local, urban, and regional scales. These studies do not provide direct
state- or national-level estimates but could be combined or extrapolated to
quantify emissions at larger spatial scales. Many studies in this category
target source sectors that do not overlap spatially, at least at the spatial
scale of interest. The second part of this discussion (Sect. –) explores inverse modeling efforts that
directly quantify sector-specific emissions at the state and national levels.
These efforts use observation networks that are sensitive to emissions across
broad geographic regions. These efforts must also devise strategies to
disentangle emissions from multiple, spatially overlapping source sectors.
Local-scale inverse modeling
Local-scale inverse modeling can best attribute emissions when the study
region has a single, dominant source type. An estimate of total emissions for
the region thus provides insight into the source sector of interest.
Studies that fall within this category often employ one of a few different
strategies to estimate emissions. For example, many efforts use a simple
box-modeling approach to estimate emissions
e.g.,,
while others use an atmospheric transport model to relate GHG observations to
emissions e.g.,. Studies that use the former
strategy typically estimate emissions in a few steps: first, they make GHG
measurements upwind and downwind of the region of interest. Second, they use the
difference between these measurements, the rate of flow through the “box”
(i.e., wind speed adjusted by pressure), and the volume of the box (i.e., the
area of the box and the mixing height of the atmosphere) to calculate total
emissions in the box. Most studies that use box modeling estimate a total
flux for the region of interest, a number that is not spatially resolved.
Other studies in this category use a more involved approach: they model
atmospheric GHG mixing ratios using an emission inventory and an atmospheric
transport model. Subsequently, these studies scale the inventory using a
single scaling factor (β) to better match modeled mixing ratios against
measured mixing ratios:
yk=∑j=1ms⋅mthk,j(xja)+ϵkxja=βxjb.
In these equations, yk is an atmospheric GHG observation at a given time
and location k. It is one of n total observations (k=1…n). The
variable xj denotes the emissions from a model grid box j at a
specific location and time, and the function hk,j() is an atmospheric
transport model that relates the surface emissions from grid box j to
observation yk. The variable ms denotes the total number of model grid
boxes in space, and mt denotes the number of time periods. In one study,
this emission estimate varies both spatially and temporally
, and in another study, the emission estimate varies spatially but
was constant in time (mt=1; ). The superscripts a and
b denote an emission inventory and final emission estimate, respectively.
In addition, the variable ϵk denotes the cumulative error in the
model and measurement (e.g., error in the estimated transport, in the
measurement, and in the estimated emissions, among other errors). The
objective of this approach is to scale an inventory estimate using a single
scaling factor (β) so that modeled atmospheric mixing ratios on the
right-hand side of Eq. () reproduce the n-observed
atmospheric mixing ratios (yk where k=1…n).
These local-scale efforts can target sources with very large emissions or
very uncertain emissions. For example, many existing studies target emissions
from cities. Cities account for 70 % of global fossil fuel CO2
emissions;
thus, insight into urban emissions provides insight into a large fraction of
total anthropogenic GHG emissions . Note that studies in this
category generally do not discriminate among different urban source sectors
but can provide insight into the contribution of urban CO2 sources versus
power plant CO2 sources (which often occur well outside city limits).
Existing efforts estimate CO2 emissions for Indianapolis, Indiana
; Sacramento, California ; and Salt Lake
City, Utah , as well as CH4 emissions from Boston,
Massachusetts , and Indianapolis .
and use the approach in Eq. (), while the other studies implement box models.
Other studies in this category target oil and natural gas industry emissions.
Existing studies use aircraft observations to estimate CH4 emissions from
Utah's Uintah drilling basin ; southwest Pennsylvania
; Colorado's Denver–Julesburg Basin
; the Barnett Shale in Texas
; and the Haynesville, Fayetteville, and
Marcellus Shale regions (in Texas, Arkansas, and Pennsylvania, respectively)
(). In addition to these aircraft-based studies, one study
uses the SCIAMACHY instrument on the Envisat satellite to estimate CH4
emissions from the Eagle Ford and Bakken Shale regions in Texas and North
Dakota, respectively . Several of these studies find
leakage rates that greatly exceed the EPA's estimated emission factors
e.g.,, while other studies
estimate leakage rates that are comparable to the EPA's numbers
e.g.,. Differences in drilling technology
and practices from one basin to another may account for these contrasting
results e.g.,.
These local-scale inverse modeling studies confer a number of advantages
relative to other top-down strategies. These strategies capture emissions
from all facilities in a given region, including those with anomalously high
emissions. In the past, the EPA has had difficulty designing facility-level
measurements that adequately sample these anomalous emitters (Sect. ). An additional advantage of these strategies is their ease
of implementation relative to those discussed in subsequent sections (Sect. –). Box modeling requires an estimate of
air flow into and out of the box, but this approach does not require a full
atmospheric transport model. Furthermore, the strategies discussed in this
section are not as computationally intensive as many of the state- and
national-scale strategies discussed later in Sect. .
These strategies also bring a number of challenges. Nearly all of the oil and
gas studies listed above use data from a single measurement campaign and
provide a temporal snapshot of emissions. GHG emission reduction policies
make it necessary to monitor trends, a goal that requires sustained
monitoring. In addition, a locality or region must have one dominant source
sector or have spatially (or temporally) nonoverlapping source sectors in
order to attribute emissions using this strategy
e.g.,. For example,
estimate oil and gas emissions from drilling regions that also contain
livestock, landfills, and wastewater treatment facilities, all of which
produce CH4 emissions. The authors subtract an inventory estimate of these
non-hydrocarbon CH4 sources from their estimated emission total, and they
attribute the remaining emissions to oil and gas activities. The authors
point out that these non-oil and gas source sectors are small contributors
relative to oil and gas operations (8.5–19 % of the CH4 emission total
in each region), and uncertainties in these other source sectors would likely
have a small impact on their oil and gas emission estimate.
Complex environmental conditions and the associated atmospheric transport
errors can also pose a challenge for local-scale inverse modeling
strategies, particularly for box models. A simple box modeling setup can be
difficult to apply when atmospheric advection, vertical mixing, or upwind
“clean air” measurements are highly heterogeneous across the box. For
example, report that their CO2 budget for Sacramento,
estimated using a box model, is uncertain by a factor of 2 due to
uncertainties in estimated wind speed and upwind clean air mixing ratios.
Furthermore, estimate CH4 emissions for the Barnett
Shale that vary from 4.4×104 to 10.9×104 kg h-1, depending on the flight. However, the authors explain that two of
the eight flights occurred during nonideal meteorological conditions, and
the range of estimates narrows to 6.1×104 to 8.8×104 kg h-1 when those flights are excluded from the analysis. Atmospheric
transport models can simulate more complex atmospheric transport patterns
relative to box models but still have difficulty modeling local- or
urban-scale phenomena, including small-scale turbulent eddies, air flow
through street canyons, and vertical mixing in a human-built landscape
e.g.,. These modeling challenges also apply to the
state- and national-scale strategies discussed in Sect. –. New innovations in atmospheric monitoring and
instrumentation may reduce some of these uncertainties.
, for example, explain that lidar instruments can
measure atmospheric mixing height, and lidar deployment could therefore
improve certain aspects of atmospheric modeling, particularly at local and
regional scales. In addition, several studies develop high-resolution
meteorological simulations, in part to better resolve atmospheric GHG
transport in urban environments
e.g.,.
This figure highlights different CO2 observation networks and how the spatial coverage of those networks
differ. These networks include tower and regular aircraft sampling sites (a); several recent, intensive aircraft
campaigns (b); the GOSAT satellite (c); and the OCO-2 satellite (d). Note that the dots in each
panel are not equivalent. An in situ monitoring site in panel (a) often provides continuous or daily data, while
each dot in panels (c) (GOSAT) and (d) (OCO-2) indicates the location of a single observation. Public
towers and public aircraft sites are operated by NOAA, DOE, Environment Canada, and partners, and the sites shown
are current through 2016. Private towers are operated by Earth Networks, and the locations here are current through 2012.
Most tower and aircraft sites also include CH4 observations, as does GOSAT.
Observations that support local-scale inverse modeling
Many recent, local-scale observation efforts focus on urban monitoring and on
oil and gas basins. Existing urban, atmospheric measurement networks include
Salt Lake City, Utah ; Los Angeles, California
; Oakland, California , the Bay Area
Air Quality Management District ; and Indianapolis,
Indiana . Recent local-scale
aircraft campaigns include the INFLUX campaign focused on the Indianapolis
metro region , the SENEX and SONGNEX campaigns focused on
multiple oil and gas drilling basins , and the
Barnett Coordinated Campaign (; Fig. ).
In addition to these urban and oil and gas studies,
and use spectroscopic CO2 and CH4 observations,
respectively, to identify emissions from resource extraction in the Four
Corners region of the western US.
The observational strategies described above are relatively diverse. These
efforts include a combination of aircraft and stationary sites (e.g.,
telecommunication towers or building rooftops). Some of these campaigns
provide a 1- or 2-day snapshot in time (e.g, most oil and gas studies),
while other campaigns involve sustained measurements over 1 year or more
(e.g., urban observation networks like LA Megacities and the Indianapolis
INFLUX project).
State- and national-scale inverse modeling
The top-down strategies discussed in this section provide sector-specific GHG
emission estimates across larger regions, regions that typically have
several overlapping source sectors. Furthermore, these strategies make
spatially variable adjustments to existing inventories, unlike the strategies
outlined in Sect. . The three strategies discussed in this
section use both GHG observations and inventories to attribute
sector-specific emissions. Each approach, however, uses a different mix; the
first approach relies most heavily on existing inventories, while the last
relies most on GHG observations.
Overall, these strategies have been relatively successful at attributing
CH4 emissions, but promising strategies for CO2 are nascent. Biospheric
CO2 fluxes are large relative to anthropogenic CO2 emissions at diel to
monthly timescales, particularly during the growing season, and the
spatiotemporal distribution of these fluxes is highly uncertain
e.g.,. These factors have limited the success of
CO2-focused efforts.
The first strategy discussed here scales the individual source sectors in a
bottom-up inventory. This setup is often similar to a multiple linear
regression:
xja=∑i=1pβixj,ib,
where i denotes an individual source sector from a bottom-up inventory, and
p indicates the total number of source sectors in the inverse model. The
observational constraint (yk) in this approach is the same as in Eq. (). This setup also assumes that the initial emission estimate
(xj,ib, where j=1…ms⋅mt and i=1…p) is defined at each
ms spatial location, at each mt time period, and for each p
source sector. In one study, this initial emission estimate is spatially
but not temporally resolved (e.g., mt=1; ), while in
another study, it is resolved in both space and time . The
p unknown scaling factors (βi, where i=1…p) adjust the magnitude
of different source sectors in the bottom-up inventory; these factors are
estimated by the inverse model. As a result of this setup, the estimated
emissions (xja) will always be a linear combination of
source-specific emission patterns in an existing bottom-up inventory.
Studies that use this approach often estimate the scaling factors (βi)
using Bayesian statistics; these frameworks can weigh uncertainty in the
measurements (yk) and in the atmospheric model (hk,j) against
uncertainty in the initial or prior guess for the scaling factors (typically
one; e.g., ).
A handful of studies leverage this approach to attribute emissions
of CH4. For example, and use
atmospheric measurements from tall towers to estimate emissions from
individual source sectors in California. Both studies find higher CH4
emissions from agriculture relative to the EDGAR emission inventory.
This scaling factor approach brings several strengths and weaknesses. An
advantage of this approach is that it not only provides an estimate of total
emissions but also the contributions of individual source sectors. The
approach can be relatively easy to implement from a statistical perspective.
With that said, one still needs to run an atmospheric transport model and
must have an estimate of background or upwind, clean air mixing ratios.
A notable challenge of this strategy is that it requires accurate knowledge
of the spatial distribution of each source sector. The estimated emissions
will always be a linear combination of source-specific emission patterns
from an existing inventory, and errors in the spatial distribution of these
inventories will propagate into errors in sector-specific attribution.
Furthermore, the atmospheric GHG observations (yk) must be sensitive to
differences in the space–time patterns among different source sectors. Worded
differently, each of the p source sectors must have differing
spatiotemporal patterns, and each sector must explain substantial variability
of the observations (yk). If the former condition does not hold, then some of
the p source sectors will be collinear; colinearity can lead to unphysical
scaling factors (βi) and unrealistically large uncertainty estimates
e.g.,. If the latter condition does not hold, then
the scaling factors may be poorly constrained by the data, resulting in
uncertain or unrealistic sector-specific estimates. To account for these
challenges, only report source-specific estimates when they
obtain scaling factors that are statistically significantly different from
zero.
A second common inverse modeling strategy scales an emission inventory at
the model grid level to better reproduce the atmospheric observations
(yk). All of the strategies discussed previously scale the spatial
patterns in an existing inventory. By contrast, this strategy scales the
emission level at each location in the model domain, and the resulting
estimate can have spatial patterns that are different from any inventory.
These estimates have the following general form:
xja=βjxjb.
Note that xjb and xja are the total emissions from model grid
box j, not the emissions by sector. Hence, the scaling factors (βj
where j=1…ms⋅mt) adjust total emissions, and all of the ms⋅mt factors are typically estimated simultaneously. Several studies
estimate scaling factors that vary spatially but are the same at each time
step e.g.,. One study allows the
scaling factors to vary in both space and time . This
approach is also Bayesian in nature: the modeler sets an initial guess for
the scaling factors (typically unity) and an uncertainty in that initial
guess; this information guides the estimate for the scaling factors,
particularly when these factors are under-constrained by the available
observations (yk; e.g., ).
This strategy does not support attribution in and of itself, but several
studies adapt this approach for that purpose. These studies attribute
emissions in each model grid cell (j) using the attribution in a bottom-up
inventory. For example, let's say that an inventory estimates that 60 % of
the emissions in a given grid cell are from oil and gas and 40 % are from
cattle and manure. The inverse modeling estimate will attribute emissions in
that grid box in the same proportion:
xj,ia=βjxj,ib.
All variables in this equation are as defined earlier. As a result of this
setup, the total emissions in any one model grid box may differ from the
inventory. However, the relative magnitude of the source sectors in any one
grid box will be the same as in the bottom-up inventory.
and leverage this strategy to estimate
CH4 emissions for California using aircraft and tower-based observations,
respectively. Like and , they also find
higher emissions from agriculture relative to EDGAR. and
further apply this strategy to attribute emissions at
continental scales; these studies use Envisat/SCIAMACHY and the GOSAT
satellite, respectively, to estimate sector-specific CH4 emissions across
North America. Both studies estimate larger emissions from agriculture
relative to the EPA and EDGAR inventories. estimate oil
and gas emissions that are a factor of 2 larger than EDGAR, while
find that these emissions are broadly consistent with
EDGAR.
This strategy has a number of advantages and weaknesses relative to other
approaches. The strategy can be used to estimate emissions at grid scale, and
the resulting emission estimate will not be a linear combination of
existing inventory estimates. However, it assumes that the inventory has
correctly estimated the relative magnitude of each emission source in each
model grid box. Errors in this relative magnitude will produce errors in the
sector-specific attribution.
Third, a number of studies leverage a strategy known as geostatistical
inverse modeling (GIM) to estimate GHG fluxes generally
e.g., and anthropogenic
emissions specifically
. This approach
attributes patterns in the emissions to individual anthropogenic source
sectors when possible. However, it will leave emissions as unattributable
when those emissions do not match the space–time patterns in any bottom-up
inventory or when the information content of the atmospheric observations is
insufficient for attribution:
xja=∑i=1pβixj,ib+ξj.
The elements xj,ib can be individual source sectors from a bottom-up
inventory (similar to Eq. ). The inverse model will then map the
emissions onto those patterns to the extent possible. The inverse model will
further add (or subtract) emissions at the model grid scale to better
reproduce the atmospheric observations (yk). These emissions are denoted
by ξj (where j=1…ms⋅mt), and a GIM typically labels the
emissions in ξj as unattributable. Furthermore, existing studies allow
xj,ib and ξj to vary both spatially and temporally with j, in
contrast to the studies described earlier in this section. Note that existing
GIM studies fix the coefficients (βi) in both space and time. In
reality, the relationship between xj,ib and GHG emissions may vary
spatially and temporally by grid box j. Two recent GIM studies experiment
with allowing the coefficients to vary by region or biome in the context of
anthropogenic and biospheric fluxes.
Several studies leverage this strategy in the context of both anthropogenic
CH4 and CO2 emissions. use a GIM and in situ
atmospheric measurements to estimate sector-specific CH4 emissions in the
US; like , they find higher emissions from the agriculture
and oil and gas sectors relative to inventory estimates.
also use this strategy to separate CH4 emission patterns due to wetlands
from anthropogenic emissions and to evaluate bottom-up estimates of the
former emission category. Two studies implement
a GIM-based framework to identify anthropogenic CO2 emission patterns
using in situ and satellite CO2 observations. They
investigate whether the atmospheric signal resulting from anthropogenic
CO2 emissions can be reliably identified given the confounding signal from
biospheric CO2 fluxes. They find that in situ and remote sensing CO2
networks can only identify anthropogenic emissions in a few regions during a
few months of the year.
The GIM approach makes more conservative assumptions relative to other source
attribution strategies discussed in this section. A GIM will only attribute
emissions to patterns in a bottom-up inventory when that inventory matches
patterns in the atmospheric GHG observations. In , for
example, the GIM maps 60 % of total US CH4 emissions onto patterns in the
EDGAR inventory but indicates that 40 % of the total emissions are
unattributable to the patterns in any bottom-up dataset. By contrast, the
other approaches discussed above will attribute 100 % of the emissions. In
GIM studies like , the unattributable emissions indicate
shortfalls in either the GHG observation network or available bottom-up data.
In the former case, existing atmospheric observations do not provide enough
information to reliably estimate sector-specific emission patterns. For
example, the information content of the atmospheric observations in
is insufficient to uniquely constrain emissions from coal
mining, and those emissions are included in ξj instead of ∑i=1pβixj,ib. In the latter case, the unattributable emissions in
ξj indicate inaccuracies in the spatial distribution of available
inventory estimates. Many existing inventories do not have well-developed
activity data for the oil and gas industry, and the unattributable emissions
in provide information about shortfalls in these activity
datasets.
modify the existing GIM framework to better isolate
anthropogenic CO2 emissions. The authors exploit differences in the
spatiotemporal properties of biospheric versus fossil fuel fluxes to do this
attribution. Specifically, the authors argue that biospheric fluxes have
smooth spatiotemporal patterns, and fossil fuel emissions do not have smooth
patterns. The authors then partition ξj into two components (smooth and
non-smooth) and attribute these emissions to the biosphere and fossil fuels,
respectively. The study examines emissions in January when biospheric fluxes
are smaller than in other months.
In summary, this section discusses statistical innovations that help isolate
individual emission sources. In addition to these innovations, accurate
models of atmospheric transport also play a crucial rule. A number of studies
indicate the deleterious influence of transport errors. For example,
argue that atmospheric transport errors hinder the
detection of fossil fuel emission patterns across the US. The
authors also argue that biospheric fluxes mask fossil fuel patterns to a
similar degree. Numerous additional studies examine the effects of transport
errors on CO2 modeling, though not in the context of fossil fuel emissions
e.g.,.
Several efforts could reduce these transport modeling errors. Like
urban-scale studies (Sect. ), national inverse modeling
studies have also begun moving toward high-resolution meteorology
simulations. These studies simulate atmospheric GHG transport at high
resolution over the US and Canada and utilize coarser resolutions elsewhere
to save on computational costs. For example, national-scale studies using the
Weather Research and Forecasting (WRF) model model GHG transport at resolutions of up
to 8–10 km , and studies using the
GEOS-Chem model simulate CH4 transport at resolutions of up to ∼50km
e.g.,. In addition to these efforts, NASA's
Atmospheric Carbon and Transport – America campaign (ACT–America, Fig. a) aims to diagnose and reduce atmospheric transport errors
. The campaign includes new tower sites and 5 years of
aircraft flights across the eastern US. Many flights will travel through
frontal systems and extratropical cyclones to better characterize and
evaluate atmospheric transport errors.
Observations that have been used to attribute emissions at state and national scales
The observations discussed in this section do not provide a direct constraint
on an individual source sector but have been used by existing regional- and
national-scale inverse modeling studies (Sect. ) to support
sector-specific attribution. These observations are typically distributed
across a broad geographic region. They are therefore sensitive to emissions
over a large area and can constrain larger regions, albeit with less detail
than the local approaches discussed in Sect. .
Observations in this category include air samples collected atop
telecommunication towers and from aircraft: the NOAA tall-tower observation
network , regular NOAA aircraft monitoring
, the Environment and Climate Change Canada tower
monitoring network , the California Greenhouse Gas Research
Monitoring Network e.g.,,
and a privately funded tower network operated by Earth Networks (Fig. ). Most of the inverse modeling studies discussed in the previous
section (Sect. ) use these in situ observation networks to
estimate sector-specific emissions
.
The current tower network is sensitive to emissions from some source sectors
but not to others. Many of the NOAA tall towers and regular aircraft sites
are in or near the Great Plains. As a result, the network has sensitivity to
agricultural emissions and to several oil and gas basins but has little
sensitivity to emissions from east coast population centers. Earth Networks,
by contrast, has focused its efforts on the east coast proximal to large
population centers. The state of California has a dense network of
publicly operated towers. In contrast to these regions, the network is sparse
across the western US outside of California and northern Colorado. On the one
hand, the population in these regions is sparse and some emission sectors are
likely to be small (e.g., vehicle emissions). On the other hand, large
resource extraction regions are beyond reach of the long-term monitoring
network, regions like the Powder River basin coal mining region of Wyoming or
the Bakken oil and gas basin in Montana and North Dakota.
NOAA's regular aircraft monitoring network complements these tower-based
sites. The flights measure GHG mixing ratios across a vertical atmospheric
profile. These datasets can help evaluate vertical mixing and transport in
atmospheric transport models, and observations from the middle and upper
troposphere can be used to quantify background clean air mixing ratios, a
necessity for the inverse modeling studies described in Sect. . A downside is that NOAA's aircraft profiles are usually
limited in frequency to one or two times per month, unlike towers, which often
have continuous observations. Scientists at NOAA have also invented a
technology known as AirCore that can observe vertical atmospheric GHG
profiles from a weather balloon . This technology could
become a key component of the long term monitoring network in the future.
A number of intensive aircraft campaigns provide observations across entire
state or multi-state regions (Fig. ). These include the 2010
CalNex campaign , the 2013 SEAC4RS campaign
, and the ACT-America campaign (2015–2019; ). A few existing studies use these observations to attribute
state-wide emissions. For example, use CalNex data to
attribute state-wide CH4 emissions from California.
Several satellites make total column observations of CO2 and CH4 (e.g.,
AIRS, TES, IASI, Envisat/SCIAMACHY, GOSAT, OCO-2, and GHGSat).
describe a number of these satellites in detail, and
provide a thorough overview of CH4-observing satellites.
Several of these satellites (Envisat/SCIAMACHY, GOSAT, OCO-2, and GHGSat)
observe in the shortwave infrared. Relative to other satellites, these four
are more sensitive to GHG mixing ratios in the lower troposphere and, hence,
to emissions at the surface e.g.,. Only a
handful of existing studies use these datasets to attribute sector-specific
emissions in the US, and these studies focus on CH4, not CO2
e.g.,.
For example, use GOSAT observations to estimate
sector-specific CH4 emissions in North America and find results that are
broadly consistent with emission estimates derived from the US tall-tower
and aircraft monitoring network . ,
however, explain that GOSAT observations are too sparse to constrain CH4
emissions from California outside of the Los Angeles Basin.
Synthesis discussion
In this section, we synthesize progress to date on estimating sector-specific
CO2 and CH4 emissions at state and national scale. We also discuss
forward-looking opportunities to improve sector-specific GHG emission
estimates, with a particular focus on opportunities to integrate bottom-up
and top-down strategies.
Recent innovations in both bottom-up and top-down efforts have advanced
scientists' abilities to identify emissions from specific source sectors.
Several efforts have produced high-resolution, sector-specific inventory
products that are based on more accurate, detailed activity data and EFs.
These products are largely driven by research in academia and by the Joint
Research Centre in Europe. New inverse modeling strategies can incorporate
these inventory estimates in more rigorous ways that are not limited to the
spatial patterns in the inventory. In addition, more extensive observations
are available to support these inverse modeling efforts, observations that
span a number of spatial scales. For example, numerous intensive measurement
campaigns in the past 5 years have focused on large GHG-emitting regions,
particularly cities and oil and gas production basins. The national US in
situ network and remote sensing GHG observations have also expanded in the
last decade, though the US in situ network expansion is smaller than the
level required for robust evaluation of a wide array of GHG source sectors.
Despite these advances in bottom-up inventories, top-down strategies, and
measurement density, the scientific community has only been able to use
inverse modeling and atmospheric data to improve sector-specific emission
estimates in a relatively small number of cases. To date, the community has
had more success integrating top-down and bottom-up estimates for CH4 than
for CO2; the atmospheric signal from biospheric CO2 fluxes often
obscures the signal from fossil fuel emissions, except in some urban
environments. National CH4 inventory estimates are often uncertain by a
factor of 2–3 at the sector level, while CO2 inventories typically agree
to within 5 % (Fig. ). Arguably, the community has been able to
use top-down inverse modeling to improve these inventories when they arguably
stood to benefit most.
Specifically, the community has been most successful with top-down,
sector-specific attribution in two types of scenarios: intensive measurement
campaigns paired with local-scale inverse modeling and opportunistic cases.
In the former case, the community has put substantial resources into
intensive, local-scale measurement campaigns for a few specific source
sectors. Measurements from each affected locality or region provide a puzzle
piece, and the community has begun to assemble a cohesive, national-scale
picture by amalgamating these individual pieces. The community has employed
this strategy in the case of CH4 emissions from oil and gas operations
(e.g., the SENEX, SONGNEX, Barnett Coordinated Campaign) and, to a
lesser degree, in the case of urban CO2 emissions (including recent
measurement efforts in Los Angeles, Salt Lake City, Boston, Indianapolis, and
Oakland). These campaigns typically provide a snapshot of current emissions
and would need to be repeated in the future to estimate how emissions vary
over time.
Other cases of successful source attribution have been largely opportunistic.
In certain cases, the community had the right atmospheric measurements and
spatially distinct source sectors to attribute emissions at large spatial
scales. For example, find large CH4 emissions in Texas
and Oklahoma that do not fit the spatial distribution of cows, and CH4
measurements in that region correlate with measurements of higher-order
alkanes. The authors conclude that a large fraction of those emissions are
likely due to oil and gas operations. reach similar
conclusions using satellite observations from GOSAT.
Numerous future opportunities would improve scientists' ability to merge
bottom-up inventories, inverse modeling, and atmospheric GHG data for better
GHG source attribution:
Combine the strengths of existing datasets
Many inverse modeling studies to date use only in situ or satellite GHG data
to estimate emissions. CH4 inverse modeling studies for North America
provide a good example. use in situ observations from long-term monitoring stations, use remote sensing observations
from Envisat/SCIAMACHY, and use remote sensing
observations from GOSAT. Future studies may be able to attribute emissions
more effectively by leveraging the strengths of all available in situ and
remote sensing datasets. Different datasets often bring complementary
strengths for this attribution: remote sensing datasets have broad spatial
coverage and in situ datasets have complete temporal coverage and greater
sensitivity to surface emissions, among other strengths. A number of
challenges may have prevented the synthesis of multiple datasets in past
efforts: large datasets entail a number of computational challenges, the data
are not always accessible, and the observations can have different
information content or error characteristics that are challenging to balance
in a single framework. Future efforts that can combine these disparate
datasets likely stand the best chance of attributing emissions to specific
source sectors.
Expand several existing measurement strategies
Expanded GHG measurements would also advance efforts to attribute emissions
to specific source sectors. As discussed earlier, some of the most successful
top-down efforts to attribute emissions have been intensive aircraft
campaigns. These campaigns are more flexible than the long-term monitoring
network and can easily target source sectors of interest by flying in
specific regions, in flight patterns that encapsulate the source of interest,
and by flying at certain times of year that have fewer competing biogenic
sources. An expansion of these campaigns would enable scientists to target
specific source sectors, including CO2 emissions from large power plants,
CH4 from agriculture, and CH4 from coal mines, among other source
sectors. These aircraft campaigns could then be used to estimate
regional-scale EFs. Existing aircraft campaigns, for example, target CH4
leakage rates for a range of different oil and gas drilling basins (see
Sect. –). The long-term in situ
atmospheric network and GHG monitoring satellites could be used to
intelligently extrapolate and gap-fill these regional EFs at larger spatial
scales and to identify broad trends over time.
In addition, successful cases of sector-specific attribution have usually
involved observations that span multiple spatial and temporal scales. This
strategy allows scientists to bridge between the regional scale that
atmospheric observations are best able to constrain and the facility-level
scale where inventories are strongest. For example, atmospheric observations
can be used to identify regional differences between top-down and bottom-up
estimates. Subsequent facility-level and on-road measurements can indicate
why those regional differences occurred and how to improve EFs in a way that
will bring inventories into agreement with top-down estimates. This
measurement strategy can be expensive and requires extensive coordination,
but several studies employ it successfully in the case of oil and gas CH4
emissions e.g.,. Bottom-up and
top-down estimates of these emissions disagree at regional and national
spatial scales e.g.,. Subsequent facility
and on-road measurements elucidate why: a small number of malfunctioning
facilities account for a large percentage of emissions. EFs that account for
this skewed distribution are more consistent with regional top-down estimates
e.g.,.
Effective source attribution will also likely require the use of secondary
tracers. Measurements of some secondary tracers, like ethane, have expanded
markedly in the past several years with advances in instrumentation. With
that said, measurements of tracers like radiocarbon are only available for
some of the long-term US monitoring sites.
Improve inverse modeling strategies with an eye toward secondary
tracers
The inverse modeling community has yet to develop inverse modeling strategies
that can fully leverage observations of secondary tracers. This task is not
straightforward and would likely require the development of new strategies.
These strategies would need to quantify heterogeneities in the ethane content
of natural gas or the disequilibrium effect in the case of radiocarbon.
Furthermore, these strategies may need to relate the primary and secondary
tracers in a single statistical framework and account for uncertainties in
that relationship. Observations of these secondary tracers have historically
been very sparse, so few existing studies focus on designing statistical
inverse modeling frameworks to fully exploit these tracers.
Develop detailed activity data as part of bottom-up efforts
Top-down efforts, like those outlined above, can help in developing
regional-scale EFs for different source sectors. These studies can be
particularly helpful when EFs are challenging to determine at facility scale.
For example, direct measurements of oil and gas facilities are difficult to
design because a small number of leaks account for the majority of emissions,
and these large emitters may be difficult to find and/or representatively
sample (see Sect. ).
In contrast to EFs, activity data can only come from bottom-up inventory
efforts. In fact, top-down efforts depend upon reliable activity data for
attributing emissions (Sect. and ). Efforts
to improve these activity datasets would markedly improve source attribution.
In many cases, these activity data exist but are not publicly available or
are not available in gridded form. cite local fuel sales
or electric utility bills as examples. CH4 emissions from oil and gas
provide an additional example. Oil and gas wells generally report production
figures to state regulatory agencies, but this reporting varies by state,
does not have a consistent format, and can be difficult to find (e.g.,
http://pmc.ucsc.edu/~brodsky/wellindex.html). The inaccessibility of
accurate activity data for oil and gas operations is a barrier to source
attribution in recent national-scale CH4 inverse modeling studies
. represents an important
step forward in this area; the authors develop gridded versions of the EPA's
activity data. These activity data are key to connecting inverse modeling
results with bottom-up estimates of specific source sectors. Future bottom-up
efforts should particularly focus on the development and public release of
gridded activity data.
In synthesis, future improvements in bottom-up inventories and top-down
strategies would likely complement one another and translate into more
reliable, sector-specific emission estimates; scientists will likely need to
combine both strategies to robustly estimate GHG emissions from individual
sources. Improved activity data would lead to gridded inventory estimates
with more accurate spatial and temporal patterns. Top-down frameworks could
then harness these patterns, along with more extensive, future GHG
observations, to estimate regional-scale EFs for specific source sectors.
National-scale observations of secondary tracers like radiocarbon and ethane
would further strengthen these top-down efforts for applicable source
sectors. This coordinated, combined approach offers the most promising
opportunity to evaluate state and national GHG emission reduction policies
in the US.