DOE Scientists Release Comments Citing Inaccuracies and Misrepresentations from Lark Study

We recently reviewed the article published by Lark et al. (2022) in PNAS, detected various problematic assumptions, approaches, data, and results in that study. Based on our findings, we concluded that these authors overestimated GHG emissions of corn ethanol consumption due to the RFS. In response to our comments, Lark et al. have stated that they believe “Taheripour et al.’s conclusions to be unsupported and based upon several misunderstandings and misinterpretations of [our] methods and results.” In what follows, we review the responses provided by Lark et al. and show that our comments did not misinterpret Lark et al.’s findings, rather that our comments are supported by the literature and actual observation and are based on the statements provided by Lark et al. in their original publication.

(More information here)

(DOE’s initial response here)
In this document, we provide responses to the Lark et al. comments one by one. To avoid
confusion, throughout we refer to Lark et al.’s 2022 original publication as Lark et al.(a) and the
response of those authors to our original comments as Lark et al.(b).
Our detailed review of the original paper and the responses by those authors reveals various
major deficiencies, problematic assessments, and misinterpretation of the existing literature in
Lark et al.(a)&(b). In summary, our major findings are:
– Lark et al.(b) admitted that their “results reflect the impacts of corn ethanol demand in general,
regardless of the source of such increases”. Hence, the title of Lark et al. (a) should not be
“Environmental outcomes of the US Renewable Fuel Standard”, as many factors affected the
expansion in demand for corn ethanol.
2
– Table S22 of the Lark et al.(a) SI shows that almost all of the estimated regional transitions
between cropland and pasture land are statistically insignificant. Hence, they build their analyses
on the statistically insignificant assessments of land transitions. Lark et al.(a) have not revealed
their estimated equations and their parameters at the NRI data point level.
– Lark et al.(a)’s treatment of soil organic carbon (SOC) and reporting of its uncertainty appear to
be based on a misunderstanding of the information extracted from other studies, including their
inaccurate use of the carbon response functions (CRFs) derived from Poeplau et al. (2011) and
overestimation of the SOC sequestration potential in CRP lands.
– Our original comment noted that Lark et al.(a) double counted N2O emissions. In their response
they argued otherwise without directly addressing the double counting issue we raised. We
explained again in this document why they did double counting.
– We questioned the inconsistency between their price evaluation period (2006-10) and land use
change assessment period (8 years from 2008 to 2016). In response Lark et al. (b) reveled that
“our estimated price effects would be somewhat larger if we used all years up to 2016”. We
explained that selecting inconsistent time segments to avoid higher price effects is not an
acceptable scientific choice.
– Lark et al.(b) in various instances selected and interpreted the existing literature in favor of
their analyses, while the literature clearly shows otherwise. For example, the authors referred to
price analyses developed by Irwin and Good (2013) to support their incorrect assumption that
DDGS is only a substitute for corn. However, Irwing and Good (2013) concluded that the price
of DDGS reflects the value of corn and soybean meal.
– In our original comments we highlighted that Lark et al.(a) ignored market mediated responses
(e.g., yield improvements). In response Lark et al.(b) argued that: “We carefully consider and
account for both yield increases and DDG offsets”. We showed that they implemented close to
zero yield improvements, ignored demand responses, and incorrectly specified DDGS offsets.
– Lark et al.(b) argued they followed the approached used by Carter et al. and hence their results
are valid. We explained that relying on Carter et al. (2017) is a problematic choice due to various
deficiencies.
– In our original comments we noted that Lark et.(a) missed various important land transitions
that frequently occur within cropland (including but not limited to return of cropland pasture to
crop production) and hence they overestimated the land use impacts of corn ethanol. In response,
Lark et al.(b) admitted they intentionally missed “cropland pasture” and stated that their
approach “purposely avoids the separate problematic category of Cropland-Pasture”. In
response, we explained that, unlike the claim made by Lark et al.(b), cropland pasture is a
standard land category recognized by the FAO, IPCC, and USDA Agricultural Censuses. We
provided detailed information about this land category and its magnitude and changes over time.
We then showed that, while the NRI data implicitly includes this type of land, Lark et al. made
no effort to capture its changes over time. Instead, they incorrectly assigned changes in the CRP
land to ethanol.
3
– Finally, we would like to note that our original comments included several comments that Lark
et al.(b) did not respond to, including our remote sensing reanalysis revealing false land use
change classifications in their original study.
– In conclusion, we find that the Lark et al.(a) paper is more problematic than what we initially
evaluated to be the case.
2. The use of CDL and NRI data
In our original comment, we highlighted the hazards involved in determining land types using
CDL data and explained that the use of this data layer by Lark et al.(a) leads to overestimation of
GHG emissions from ethanol.
In response to our original comment on this topic Lark et al.(b) stated that they used NRI, not
CDL, to estimate the types, amount, and regional location of land conversion and that conversion
of pastureland and CRP land to cropland generates substantial carbon debt. In addition, Lark et
al.(b) stated that we conflated their methods for estimating water quality impacts (Lark et al.(a)
SI ln 696) with their methods for identifying land transitions.
Our responses follow.
i. Using both CDL and NRI data
In our original response, we did not dispute Lark et al.(a)’s use of NRI data to calculate land
conversion at data points but pointed out the issues caused by the use of CDL data to determine
the location and characteristics of converted land at a high-resolution scale. Our original
comment outlined the consequences of this use.
In addition, using both CDL and NRI data is a questionable practice, as these two data sets
follow different definitions, protocols, and approaches. To highlight these differences, we clarify
below a few items related to “pastureland” and “CRP land,” the two land classes examined by
Lark et al.(a).
The NRI data set presents “pastureland” as a class of land under that title with the following
definition:
“A land cover/use category of land managed primarily for the production of introduced forage
plants for livestock grazing. Pastureland cover may consist of a single species in a pure stand, a
grass mixture, or a grass-legume mixture. Management usually consists of cultural treatments:
fertilization, weed control, reseeding, renovation, and control of grazing. For the NRI, [this]
includes land that has a vegetative cover of grasses, legumes, and/or forbs, regardless of
whether or not it is being grazed by livestock.” (USDA, 2020)
This definition clearly suggests that the land category of “pastureland” in NRI data represents
primarily managed land for producing forage plants, while some natural vegetation could be
included as well. The NRI area of this land category has been around 120 million acres since
2012.
4
On the other hand, the CDL data set represents a class of land labeled “grassland/pasture” with
an area around 380 million acres since 2012, more than three times of pastureland in the NRI
data set. This class of land in the CDL data set covers a wide range of land types that are not
included in the NRI data set which basically covers managed land used for forage production by
definition. Mapping an estimated change in “pastureland” obtained from the NRI data at a data
point to the CDL data at the grid cell level relies on problematic method and non-scientific
judgments.
In the case of CRP land, the mapping between the NRI and CDL data sets is even more
problematic, as the latter data set does not identify CRP land, and the mapping process is more
arbitrary.
ii. Issues with CDL data
The comments above should make it clear that we have not conflated the Lark et al.(a) method
in using CDL data to determine the location and characteristics of converted land at a highresolution scale with their methods for estimating water quality impacts. In our original comment
on this topic, we used the following quote from the Lark et al.(a) SI to describe the issues related
to use of CDL data:
“For the period 2008-17, we used the USDA-NASS Cropland Data Layer (CDL) and a look-up
table to convert CDL land cover classes to vegetation types simulated by AgroIBIS.”
In their response, Lark et al.(b) argued that the above quote is taken from their SI, in which they
explained their method of calculating water quality, and we had misrepresented it as their land
transition method. To clarify, we used the above quote simply because it explicitly noted the use
of CDL data, although Lark et al.(a) noted the use of CDL data indirectly in a few other places.
For example, in their main manuscript under the title of Cropland Area Change they noted the
following:
“[T]he high-resolution field data (37) were used only to identify the possible locations and
characteristics of converted land, whereas the data from the NRI were used to estimate the
magnitude of conversion and how much of it could be attributed to the RFS.
Reference 37 in their paper refers to CDL data.
iii.“Pastureland” definition
Although by definition “pastureland” in the NRI data set represents primarily managed land
producing forage for livestock feed, Lark et al.(a) referred to this class of land as natural land
and argued that conversion of this type of land to cropland results in “substantial carbon debt.”
Following are some observations that do not support their claim.
Figure 1 shows land transitions to and from the NRI pastureland category over time and indicates
a land transition of 28 million acres to pastureland, mainly (71%) from cropland, between 1982
and 1997 at the national level. On the other hand, during the same time period, about 38.6
million acres of land left the pastureland category, mainly becoming cropland, CRP land,
rangeland, forest or other land types. In the five years that followed (1997-2002), about 19
5
million acres of land (mainly cropland) moved to pastureland and about the same amount left
this category of land.
The same pattern of land exchange can be seen in the next three periods, 2002-2007, 2007-2012,
and 2012-2017, with around 10 million acres in and 10 million acres out for each 5-year
segment. These large exchanges between pastureland and other types of land, in particular with
cropland, confirm that farmers continuously rotate a portion of their managed land between
cropland and pastureland to produce either primary crops or animal feed plants. The frequent
rotations between cropland and managed pastureland should not be interpreted as conversion of
natural land to cropland, as Lark et al.(a) misstated and misused in their analysis. It is also
important to note that, at the national level, the total area of pastureland increased by 2.1 million
acres between 2007 and 2012 and then decreased by 1.4 million acres in 2012-2017. This means
that due to all drivers of land use (including biofuels) pasture land has increased, not decreased,
between 2007 and 2017.
Figure 1. Land transitions to and from pasture land from 1982 to 2017. Based on USDA (2020).
Moreover, the NRI data set, like any other data set, is subject to various potential errors due to
the sampling process, data collection, remote sensing, and processing data. Studies have
addressed quality of this data set (e.g., Copenhaver et al., 2021); two examples of differences
across various versions of this data set are shown in Table 1. As shown in this table, the 2012
and 2015 NRI data releases provide entirely different pictures of the land transition between
cropland and pastureland.
6
Table 1. Land transition between pasture land and cropland in two different releases of NRI data.
Type of land conversion NRI release
2012
NRI Release
2015 % Difference
Cropland to pasture 2007-2012 4,793.4 7,435.2 55.1
Pasture to cropland 2007-2012 4,585.5 6,368.6 38.9
Sources: USDA, 2012 National Resources Inventory, Summary Report, August 2015. USDA, 2015
National Resources Inventory, Summary Report, September 2018.

iv. Summary
In summary, we understand that Lark et al.(a) used NRI data to determine changes in pastureland
and CRP land at data points. We merely point out the issues inherent in using CDL data at high
resolution and note two crucial facts. First, the CDL data set determines land types with a large
margin of error, and this can lead to overestimation of GHG emissions of ethanol. Second, the
mapping between the estimated changes in “pastureland” and CRP land obtained from the NRI
data at a given data point and the CDL data at the grid cell level can also lead to overestimation
of GHG emissions of ethanol. Finally, we reviewed the NRI land category “pastureland,” which
mainly represents managed land with frequent exchanges with cropland. In other words, farmers
frequently switch back and forth between the two categories of managed land: cropland and
pastureland. When the managed land is used for production of forage and grasses for livestock, it
goes into the category of pastureland, and when it is used for production of primary crops it goes
into the cropland category. These transitions should not be interpreted as the conversion of
natural land to cropland, as Lark et al.(a) appear to have done.
v. Remote sensing analysis of parcels asserted to have converted to cropland
Finally, in our original note we commented on remote sensing analysis of parcels asserted to
have converted to cropland. In their response Lark et al.(b) did not comment on our analysis of
their supporting geodatabase, as presented in their SI (titled US_land_conversion_2008-
16.gdb.zip). As we described in our response, we accessed the layer “ytc” (described as “areas
converted to crop production between 2008 and 2016” with the year of expansion listed in the
polygon) in the ArcGIS software and opened the Lark et al.(a) “cropland expansion” layer into
Google Earth Engine (GEE). Using the LandTrendr spatio-temporal curves of the Normalized
Difference Vegetation Index (NDVI), we found that several of the parcels of land identified by
Lark et al.(a) as “expansion to cropland” may often be short-term (less than 10 years) fallow/idle
lands. In preparing these comments, we processed more fields that were identified by Lark et
al.(a) as having converted to cropland during the study time frame (analysis courtesy of Ken
Copenhaver, see Appendix A). We confirmed our original findings that many of the change
classifications by Lark et al.(a) appear to be incorrect.
7
3. The carbon response and CRP land
i. Use of the carbon response functions (CRFs) derived from Poeplau et al. (2011)
In our original response to Lark et al.(a), we pointed out Lark et al.(a)’s potential overestimation
of SOC changes upon grassland conversion, which one of the Lark co-authors already noted in
his paper (Spawn et al. 2019). Subsequently, Lark et al.(b) provided more detail that had not
been included in Lark et al.(a) or elsewhere:
“Upon closer examination, numerous studies factored into the Poeplau et al.[study]. CRFs
represent conversions of previously cultivated grasslands (ranging from just 4 to 100+ years as
grassland) and many of these and others have also been intensively grazed or hayed.
Furthermore, several of the included studies also represent grassland conversions to no-till
cropping. As such, we believe the grassland CRFs of Poeplau et al. 2011 are, in fact, well suited
to characterize the types of grassland conversions observed recently throughout the US.”
Even with a “closer examination,” the data on SOC and the associated meta-data compiled in
Poeplau et al. may still not be fully understood. For example, Poeplau et al. compiled a total of
176 observations from 45 studies related to grassland to cropland conversion (Table 1 of the
Poeplau article) but 27% of the observations were extracted from two studies—Newton et al.
(1945) and Campbell and Souster (1982). Interestingly, both studies were conducted in Canada,
where mean annual temperatures (MAT) of soil are about 4°C and 3°C, respectively. Since the
MAT is one of the variables considered in a specific CRF for grassland conversion, the CRF
Lark et al.(a) used would be less accurate in higher MAT regions, such as Iowa (9°C), Illinois
(11°C), and Nebraska (10°C) in the U.S. Midwest.
Furthermore, 52 observations, including Newton et al. (1945), were obtained from literature
published before 1980, when the soil sampling and measuring methods used might be different
from recent technologies. Lark et al.(a) would have found it useful to have examined the details
behind the Poeplau et al. study before using the CRFs to characterize the grassland conversions
observed recently throughout the US.
In addition, the grassland CRFs were applied to a soil depth up to 100 cm, which is much deeper
than the 90% of confidence interval of soil depths from 15 to 38 cm covered by the dataset of
grassland to cropland conversion (SI Table 2) in Popelau et al.
As a result, Lark et al.(a)’s application of CRFs may not accurately reflect the SOC changes
associated with grassland conversion in more recent years and in the U.S. Midwest.
ii. Estimation of the SOC sequestration potential in CRP lands
As noted in our original response, GTAP LUC modeling does not consider conversion of
CRP lands, and thus related emissions factor (EF) models, such AEZ-EF and CCLUB, do
not model emissions/sequestrations associated with the conversion. Nonetheless, Lark et
al.(b) continued to stress SOC losses from conversion of CRP lands and stated that “field studies
consistently show that CRP lands recover soil carbon to varying degrees during their contract
period that can then be lost upon recultivation,” citing Spawn-Lee et al. (2021).
8
In that article, the authors state that “Field studies assessing SOC changes after recultivation of
CRP lands consistently report either net emissions or indeterminant change [19, 27–31], with
estimated SOC losses as high as 154 MgCO2e ha−1 when CRP land is converted to a corn-soy
rotation managed with conventional tillage [29]. Conversion to no-till management results in
lower but still substantial GHG costs [19].”
When we looked into the references cited, we found out that the authors’ statement on SOC
potentials in CRP lands would have been skewed by the findings of four studies from the
Kellogg Biological Station (KBS) Long-term Ecological Research (LTER) site in southwest
Michigan (Table 2). Although KBS is an invaluable long-term field experimental site, the site
soil is loamy soil with low SOC levels (~1.2%). The CRP-induced SOC accumulation observed
from that site would not be representative of many CRPs in the US.
Table 2. Literature cited in Spawn-Lee et al. (2021) related to SOC sequestration potentials in CRP
lands.
Literature Title Site
location
[19] Ruan and Robertson G
P 2013
Initial nitrous oxide, carbon dioxide, and methane costs of
converting conservation reserve program grassland to row
crops under no-till vs. conventional tillage
KBSLTER
[27] Reeder J D, Schuman
G E and Bowman R A
1998
Soil C and N changes on conservation reserve program
lands in the central great plains
Wyoming
[28] Piñeiro G, Jobbágy E
G, Baker J, Murray B C
and Jackson R B 2009
Set-asides can be better climate investment than corn
ethanol
Metaanalysis
[29] Gelfand I, Zenone T,
Jasrotia P, Chen J,
Hamilton S K and
Robertson G P 2011
Carbon debt of conservation reserve program (CRP)
grasslands converted to bioenergy production
KBSLTER
[30] Zenone T, Gelfand I,
Chen J, Hamilton S K and
Robertson G P 2013
From set-aside grassland to annual and perennial cellulosic
biofuel crops: effects of land use change on carbon balance
KBSLTER
[31] Abraha M, Gelfand I,
Hamilton S K, Chen J and
Robertson G P 2019
Carbon debt of field-scale conservation reserve program
grasslands converted to annual and perennial bioenergy
crops
KBSLTER
As we pointed out in our first response, the literature/data on SOC of CRP lands is very
sparse. The USDA has recognized the need for more observational data from sampling,
measuring, and monitoring soil carbon on CRP acres and has recently launched CRP Climate
Change Mitigation Assessment Initiative projects where Lark will join as one of collaborators.
We believe that Lark et al. will be able to provide more concrete data on SOC of CRP lands once
more data is collected from that project.
9
iii. Cropland-pasture carbon emission factors (EF) in CCLUB
As Lark et al.(b) explained, the three sets of EFs—Woods Hole, Winrock, and AEZ EFs –simply
assign half the value assumed for the conversion of “grassland/pasture” to the conversion of
cropland-pasture land, which results in net SOC loss. This is different from the approach in
CCLUB.
The authors did not mention some key differences between CCLUB and all other sets of EFs
described in many of our published articles and reports. For instance, the AEZ-EF model (Plevin
2014) adopts a definition of cropland-pasture different from CCLUB’s, and all other sets of EFs
consider only generic croplands instead of different types of croplands. Nor do the other sets
consider the effects of land management practices and assumptions for the 20-30 years that
follow land use change (LUC).
The SOC EFs calculated in CCLUB are based on the model’s inclusion of different tillage types
and the assumption that spatially explicit (U.S. county-level) feedstock yield is either constant or
increasing (Taheripour et al., 2021). Significantly, in that study we focused on converted
croplands used mostly for annual food crops (e.g., corn, soy, and wheat) and aggregated the EFs
from the counties where historical feedstock production data are reported in USDA surveys. To
evaluate and calibrate our modeling results corresponding to specific feedstocks of interest, we
conducted meta-analyses of published literature (Qin et al., 2016; Xu et al., 2019) and calibrated
a CENTURY-based model with long-term experimental data (Kwon et al., 2017).
Lark et al. dismissed the “cropland pasture” category by stating that it is “defined by economists
as land that variously cycles between cultivation and perenniality.” Cropland-pasture is a
standard land category and is defined by USDA, FAO, and IPCC (see Taheripour et al., 2021, for
more details). All three define “cropland-pasture” as “temporary pasture and meadows.” It is true
that this land type (like CRP) has not been well documented for its carbon inventories. It should
be noted, however, that if we want land classes to match carbon inventories, we would need
many, more-detailed land classes instead of aggregating land classes currently available from
land cover datasets (e.g., NLCD) for economic and emission modeling. Thus, researchers at the
Argonne National Laboratory (ANL) have evaluated several scientific approaches (e.g.,
changing the frequency of switches between cropland and pasture phases that influences SOC
levels) to model the current EFs for cropland-pasture conversion.
iv. Additional comments on CRP land and carbon sequestration
In our original comments we noted that:
“Lark et al. applied the CRFs that were based on conversion of “native” or undisturbed
grassland to cropland. CRP land is likely to be less rich in soil carbon stocks than native or
undisturbed grassland because it has been under vegetation cover for only a limited number of
years.”
In response to this comment Lark et al.(b) argued that:
“[T]he minimum is 10-15 years (the length of a CRP contract) and this is often far exceeded if
the land was enrolled for more than one contract cycle (e.g., 36% of all CRP land between 2013
10
and 2016 was enrolled for at least 2 contract cycles; Bigelow et al. 2020). Field studies
consistently show that CRP lands recover soil carbon to varying degrees during their contract
period that can then be lost upon recultivation, and, while direct measurements of CRP pre- and
post-conversion are notably lacking and much needed, when conducted, they have found that
emissions can be comparable to those observed following “natural” grassland conversions (see
discussion and references in Spawn-Lee et al. 2021).
We question Lark et al. above characterization with the below specific points.
• According to the USDA:
“The CRP is a Federal program established under the Food Security Act of 1985 to
assist private landowners to convert highly erodible cropland to vegetative cover for 10
years. For NRI, only acres that have been enrolled in CRP general sign-up are included
in the CRP land cover/use category. It does not include acres enrolled in CRP continuous
sign-up.” Source: USDA (2020).
Therefore, the NRI data set provides an incomplete picture from the CRP land and
represents only general signups in this program.
• Regarding the claim that “36% of all CRP land between 2013 and 2016 was enrolled for
at least 2 contract cycles” we did not find such a finding or claim in Bigelow et al.
(2020). Instead, the authors of this reference reported that “36 percent (2.76 million
acres) of the acreage in expiring CRP contracts during 2013-16 was reenrolled in the
CRP”. That is, Bigelow et al. (2020) referred to the expiring CRP contacts in 2013-
2016, not all CRP land in that time period. Their results shows that 2.76 million acers
of the expired CRP lands between 2013-2016 reenrolled again in the program (see Table
2 of the appendix of Bigelow et al., 2020). It seems that Lark et al. (b) misinterpreted the
findings of the original authors.
• The extent to which CRP land sequesters carbon in a short time period is unknown and
highly uncertain. The fact is that it could take decades, not years, to restore the carbon
content of any disturbed land.
• Lark et al.(a) assigned a large emission factor to their over-estimated land conversion
attributed to RFS, which resulted in significantly overestimated carbon implications for
ethanol production. Figure 2 compares the emission factors implied by Lark et al.(a) with
the emission factors provided by other sources. This figure shows that the emission
factors assigned by Lark et al.(a) to each hectare of converted land (pastureland/CRP) are
larger than those of other sources. For example, they are 80% higher than the emission
factors for pastureland imbedded in the AEZ-EF model. Note that the pastureland
emission factor of AEZ-EF mainly represents natural grass, while the land in Lark et
al.(a) is mainly CRP land and primarily managed pastureland. It is also important to note
that the emissions factors of the AEZ-EF model have not been updated with the recent
IPCC tables, which provide lower emissions factors. The overestimated emission factors
and overestimated land conversion in Lark et al.(a) led to overestimated ILUC emissions
for corn ethanol.
11
Figure 2. Pastureland emission factors among different data sources at the national level.
4. The issue of double-counting N2O emissions
Lark et al. did not directly address the double counting issue we raised in our earlier comments.
We reiterate the issue here again. As we stated in our original comments, the GREET corn
ethanol LCA uses the USDA statistics for a given year for corn yield and N fertilizer application
to derive national average N fertilizer use per bushel of corn of that year. That is, by using the
USDA annual statistics, we include all corn acreage (existing and new) in a given year. If
additional corn acreage results from RFS and uses additional N fertilizer, that is reflected in the
USDA annual statistics.
Most corn ethanol LCA studies account for N2O emissions from nitrogen fertilizers in cornfield
by using emission factors, which assume linear/non-linear relationships between N2O emissions
and nitrogen inputs. This is the same approach employed in LCA studies and in Lark et al.(a).
However, in their calculation of additional N2O emissions they did not note that if there is any
change in nitrogen applied to corn, the farming emissions would have already included the GHG
impact of such a change, which is especially true for those LCA studies on total U.S. ethanol
volume.
Thus, when Lark et al.(a) added their N2O emissions to EPA’s results and the GREET results for
both CARB and Argonne in Figure 3 of their study, they accounted for the N2O emissions that
GREET and EPA had already accounted for in the remaining bars of the figure 3. In particular,
for the EPA RIA such N2O emissions were included as “domestic farm inputs and fertilizer
N2O” (domestic land use change for other GHG changes). By adding them to the GHG emissions
from land use change again, we maintain that N2O emissions, as they are presented in Figure 3,
are indeed double counted in Lark et al.(a).
12
5. Inconsistencies in Lark et al.(a) results
In our original comments, we referred to various puzzling inconsistencies in Lark et al.(a) results
at the county level. Lark et al.(b) responded that their spatial results are correct, expected, and
match with ecological theory.
First, it is important to note that we neither generated nor interpreted the county-level results,
which were provided by Lark et al.(a) in their supporting materials, but rather used them to make
some cross-checks. Second, checking the results of Lark et al.(a) at the county level helps verify
consistency in results. When the result of modeling at an aggregated level (e.g., county level)
shows unexpected results, that could suggest more issues with using such results for detailed
analysis (e.g., spatial resolution). We would be happy to review the results of Lark et al.(a) at the
spatial resolution level for annual changes (not the average for the eight years of study period) in
cropland, pasture land, CRP land, wheat, corn, and soy and any other crops.
Finally, the Lark et al.(b) response to our original comment suggests that they did not understand
our comments, so we will expand on our original comments to help. To take one example, the
Lark et al.(a) results at the county level show that in one specific county (fips code 26147), the
area of cropland increases by about 130 ha, which generates a carbon savings (not carbon
emissions) of 18.65 Gg CO2e. It is not clear what ecological theory or detailed spatial resolution
could justify this and other similar observations. In another specific county (fips code 30085), the
area of corn increases by 4 ha, while cropland area increases by 9261 ha. Again, it is not clear
what land transformation elasticities and what county characteristics generated this and similar
observations at the county level. Note that Lark et al.(a) estimated land transformation functions
at the NRI data point level using county-specific characteristics. They did not use spatial data in
this estimation process. The general assertion provided by Lark et al.(b) that their results are
consistent and match theory should be supported by the release of their estimated land
transformation functions, their parameters, and the projected changes in land cover items and
crops.
6. Attribution of ethanol volume to the RFS2
In our original comments, we noted that Lark et al.(a) did not isolate the effects of RFS on
ethanol growth from other policies and market forces and simply assigned the difference
between the targets of RFS1 and RFS2 as the contribution of RFS to ethanol consumption.
In response to this comment, Lark et al.(b) noted that they followed Carter et al. (2017) in
determining the effect of RFS on ethanol consumption and then argued the following:
“Nevertheless, our results reflect the impacts of increased corn ethanol demand in general,
regardless of the source of such increases.”
In response to Lark et al.(b), we provide the following additional comments:
• We believe that, by mistake, the authors misinterpreted the Carter et al. (2017) argument
when they said: “Carter, Rausser, and Smith (2017) argue that ethanol prices were not low
enough to have incentivized additional ethanol use if the RFS2 had not passed.” Indeed,
13
Carter et al. (2017) did argue that in the absence of RFS2 ethanol production was not
profitable.
• However, the Carter et al. (2017) argument that ethanol was not profitable in the absence of
RFS2 is simply inaccurate. Indeed, ethanol production was profitable prior to RFS2. Figure 3
indicates the profitability of ethanol prior to RFS2. This figure from Tyner and Taheripour
(2008) shows the breakeven line for ethanol production (including 12% return on equity) and
the actual profitability of ethanol industry year by year from 2000 to 2008. The figure shows
that, unlike the claim made by Carter et al. (2017), ethanol production was profitable prior to
the approval of RFS2.
Figure 3. Breakeven corn and ethanol prices compared with actual profitability observations. Red
diamonds represent actual observations. The break-even line includes 11% return on equity.
Source: Tyner and Taheripour (2008).
• In Lark et al.(b) the authors recognize that “Nevertheless, our results reflect the impacts of
increased corn ethanol demand in general, regardless of the source of such increases.” With
this new confirmation, the Lark et al.(a) results do not represent solely the RFS impacts but
the impacts of increased corn ethanol demand from a broader range of effects, including the
RFS effects. Hence, a more appropriate title, instead of the original title, should have been
used for the PNAS paper
14
7. Displacement of Distiller’s Grains and Yield improvements
i. DDGS displacement
In our original comments, we pointed out that distiller’s dried grains with solubles (DDGS)
displaces corn and soybean meal in least-cost animal rations, which has important land use
implications since some acres producing soybean meal for feed are replaced by DDGS.
However, in their response, Lark et al.(b) claimed that DDGS only displaces corn:
“Regarding distiller’s grains, Irwin and Good (2013) show that the price of DDGS follows the
price of corn very closely, which implies that they are close substitutes (although they have some
nutrient differences).”
It appears that Lark e al.(b) misinterpreted the work and findings of Irwin and Good (2013), who
analyzed and explained three elements: corn and DDGS prices, soy meal and DDGS prices, and
the ratio of DDGS price to corn price. Using these figures and a few simple regression analyses,
Irwin and Good (2013) concluded the following:
“We show that dried solubles (DDGS) prices are primarily explained by corn and soybean meal
prices, reflecting the value of energy and protein content of the DDGS. Not surprisingly, our
simple model still leaves substantial variation in DDGS prices to be explained.”
Clearly this statement does not align with the Lark et al.(b) interpretation. Irwin and Good (2013)
did not state that corn and DDGS are close substitutes. Detailed animal feed ration formula in
fact shows substitution of both corn and soymeal by DGS (see Benavides et al. 2020). Indeed,
the findings of Irwin and Good (2013) support the fact that the price of DDGS reflects the value
of energy and the protein content of this by-product.
Lark et al.(b), go on to state the following:
“Taheripour et al. use the GREET model to argue that the net loss is about half an acre because
DDGS displace some soybean meal which saves land because soybeans are lower yielding than
corn. This point is not relevant to our modeling because our LUC modeling estimates how
farmers respond to price changes, i.e., rather than making a mechanical adjustment in acreage
based on ethanol production, farmers make planting decisions based on prices.”
It seems that Lark et al. did not understand our original response, given their assumption that
DDGS only replaces corn. In fact, the substitution of DDGS for corn plus soybean meal has been
widely documented by Weightman et al. (2011), Buenavista et al. (2021) and very recently by
Haque et al. (2022). The later specifically state: “DDGS substitutes [for] soybean meal (SBM),
di-calcium phosphate, and corn in swine diets, providing lysine, phosphorus, and energy. In
DDGS, lysine is very restrictive to 0.7%, whereas phosphorus is relatively high (0.71%).” In
fact, any least-cost ration feed formulation program will substitute DDGS for soybean meal as
well. Therefore, utilizing soybean meal substitution ratios is not “mechanical” but reflective of
livestock use of DDGS. Lark et al.(b)’s statement that DDGS and corn “have some nutrient
differences” exactly explains why DDGS does not simply substitute for corn only, as assumed in
their modeling. Overlooking these more complex substitution effects in their modeling casts
doubts on Lark at al.(a)’s analysis. In fact, soybean yield is much lower than corn yield per acre,
15
ignoring of soymeal substitution by DDGS would result in significantly less LUC offsetting
effects by DDGS.
ii. Yield improvements
In our original comments, we noted that Lark et al.(a) failed to take yield improvement into
account in their modeling approach, and we made some calculations to address the importance of
yield improvement. Lark et al.(b) responded to our comment on yield improvement with the
following:
“Our model estimates what land use would have been if demand for corn had been 1.3b bushels
lower than it was under the RFS2. Taheripour et al. note that corn yield has increased since
2007, but these improvements may have persisted absent the RFS2, so it should not be assumed
that all of the change in corn yield before and after 2007 is attributed to RFS2. Moreover, our
price analysis accounts for the possibility that price increases cause yield increases. If yields
tend to increase when prices increase, then the resulting supply increase would mitigate the
price effects. Our goal was not to estimate the amount by which cropland increased after RFS2,
but rather the difference between observed cropland use and what would have happened if corn
demand were 1.3b bushels lower.”
In what follows, we present additional comments regarding the above response.
a) The claim that “Taheripour et al. note that corn yield has increased since 2007, but these
improvements may have persisted absent the RFS2” represents the views of Lark et al. on
yield improvement. While various papers have shown that yield does respond to higher
prices (Houck and Gallagher, 1976; Lyons and Thompson, 1981; Choi and Helmberger,
1993; Huang and Khanna, 2010; Weersink et al., 2010; Berry and Schlenker, 2011; Yu et al.,
2012; Goodwin et al., 2012; Haile et al., 2016; Miao et al., 2016; Kim et al., 2018; Rosas et
al., 2019), Lark et al.(b) apparently ignored this important fact by arguing that yield
“improvements may have persisted absent the RFS2”, especially when Lark et at. asserted
significant corn price increase from the RFS.
b) Lark et al.(b) noted that “it should not be assumed that all of the change in corn yield before
and after 2007 is attributed to RFS2.” This is unquestionably the case, and no one has
asserted otherwise. However, an economic model that is supposed to represent the effect of a
policy (RFS) for 30 years should take into account the impacts of potential yield
improvements due to higher crop prices, in particular the sharp price increases projected by
Lark et al.(a) (i.e., the 31%, 19%, and 20% price increases for corn, soybeans and wheat,
respectively), and one should expect to observe some yield improvements in the Lark et al.(a)
results. However, it appears that Lark et al.(a), although stating that “We carefully consider
and account for both yield increases and DDG offsets,” assigned only very small yield
improvements, not significantly different from zero, in their analyses.
c) Third, the Lark et al.(a) modeling approach failed to capture the demand side responses to
higher crop prices. They assert that, due to RFS2, the prices of corn, soybeans, and wheat
would increase by 31%, 19%, and 20%, respectively. It is logical to question what the
16
impacts of these large price increases would be on demand for these commodities. In fact, the
Lark et al.(a) results show nearly no demand response.
iii. An assessment of yield improvement and demand response in Lark et al.(a)
In what follows we provide some analyses to highlight the importance of the missing factors
noted above in the Lark et al.(a) work, singly and in combination. Above in (a) we noted the
effect of inaccurately specifying the effects of substituting DDGS for corn and soybeans meals,
but here we follow Lark et al.’s assumption of 1/3 of corn as the only credit for DDGS to
highlight other issues imbedded in the Lark et al.(a) results. Following Lark et al.(a), we also
assume 2.8 gallons of ethanol per bushel of corn rather than the average conversion rate of corn
to ethanol of about 2.9 gallons per bushel of corn.
Table 3 show the observed yields for corn, soybeans, and wheat in 2007 and 2015. As shown in
this table, corn, soybean, and wheat yields increased between 2007 and 2015 by 11.8%, 15.0%,
and 8.3% respectively. These are significant yield improvements. Of course, not all of these yield
improvements should be assigned to the RFS. The critical questions are:
• What portion of the observed yield increases should be assigned to the RFS?
• What yield impairments are included in the Lark et al.(a) results?
Table 3. Observed yields in 2007 and 2015.
*
Description Corn Soybeans Wheat
Observed yields in 2007 (tonnes/ha) 9.46 2.81 2.70
Observed yields in 2015 (tonnes/ha) 10.57 3.23 2.93
Percent change in yields 2007-15 11.8 15.0 8.3
* The observed yields in 2007 and 2015 are approximately on the long-run yield trend lines.
To assess the extent to which Lark et al.(a) have taken into account yield improvements and
demand response—the two important market-mediated responses that affect land use (Hertel et
al., 2010)—we developed the following analyses.
Table 4 shows potential yield responses to the assumed price increases by Lark et al.(a) for corn,
soybeans, and wheat. For example, given the price increases of 31%, 19%, and 20% for corn,
soybeans, and wheat claimed by Lark et al.(a), a very small yield to price response of YDEL=
0.05 leads to yield improvements of 1.55%, 0.95% and 1% for corn, soybeans, and wheat. The
corresponding yield improvements for YDEL=0.25 are 7.75%, 4.75%, and 5% for corn,
soybeans, and wheat, as shown in Table 4. These are significantly lower than the observed yield
improvements for 2007-2015. This means that even a YDEL=0.25 in combination with the
assumed large price increases by Lark et al.(a) do not correspond to the observed yield
improvements in 2007-2015.
17
Table 4. Potential yield improvements for alternative yield to price responses with assumed
increases in crop prices by Lark et al.(a)
Description Corn Soybeans Wheat
Price increases assumed by Lark et al.(a) (%) 31 19 20
Percent change
in yield under
various YDEL
assumptions
Percent change in yield with YDEL=0.00 0.00 0.00 0.00
Percent change in yield with YDEL=0.05 1.55 0.95 1.00
Percent change in yield with YDEL=0.10 3.10 1.90 2.00
Percent change in yield with YDEL=0.15 4.65 2.85 3.00
Percent change in yield with YDEL=0.20 6.20 3.80 4.00
Percent change in yield with YDEL=0.25 7.75 4.75 5.00
Table 5 shows the required expansion in corn area to produce 5.5 billion gallons (Bgal) of corn
ethanol, targeted by Lark et al.(a), using the following assumptions:
• No yield improvement in corn
• Conversion rate of 2.8 gallons of ethanol per bushel of corn
• A one-to-one displacement between DDGS and corn with 1/3 credit in land use for
DDGS
• No change in demand for corn, i.e., zero demand elasticity for corn.
We refer to the required expansion in corn area using these assumptions as the max-corn area.
As shown in Table 5, the max-corn area would be about 3.516 million ha (Mha). The expansion
in corn area estimated by Lark et al.(a) is 2.8 Mha. The difference between these two values is
about 0.716 Mha, as shown in Table 5. In what follows we examine the yield improvements that
would close this gap.
Table 5. Max-corn area requirement.
Description Assumptions/Calculated
Results
Increase in corn ethanol production by Lark et al.(a) (Bgal) 5.5
Corn to ethanol conversion rate (gallons/bushel) 2.8
Corn needed (bushels) 1,964,285,714
Pounds per tonne 2204.62
Pounds per bushel corn 56
Corn needed (tonnes) 49,895,220
Corn yield in 2007 (tonnes/ha) 9.46
Yield improvement (% change) 0
Required expansion in corn area (ha) 5,275,181
Required expansion in corn area adjusted for DDGS (ha) 3,516,788
Max-corn area (ha) 3,516,788
Estimated corn area expansion by Lark et al.(a) (ha) 2,800,000
18
Description Assumptions/Calculated
Results
Difference between max-corn area and Lark et al. results (ha) 716,788
Table 6 shows land saving due to yield improvements and/or demand response in various
scenarios. The top section of this table shows data for production, consumption (domestic and
net exports), area, and yield for corn, soybeans, and wheat in 2007.
The second section of Table 6, Scenario 1, repeats the calculations of Table 5 with varying yield
improvements due to higher crop prices. This scenario considers a tiny yield-to-price-response of
YDEL=0. 032. This small value of YDEL in combination with assumed increases in crop prices
by Lark et al.(a) generates yield increases of 0.99%, 0.61%, 0.64% for corn, soybeans, and
wheat, respectively. In this scenario, the required area for producing 5.5 Bgal corn ethanol drops
slightly, from 3.516 Mha to 3.482 Mha. However, these small yield improvements generate
lower demands for land for corn, soybeans, and wheat production compared to the status quo in
2007. The total land saving due to the assumed yield improvements in this scenario is about
0.623 Mha, as shown. In this scenario, the net increase in demand for corn area would be 2.8
Mha, as seen in the last line of Scenario 1. This is identical to the projection made by Lark et
al.(a) for the expansion in corn area due to RFS. Indeed, the results of Scenario 1 reveal that Lark
et al.(a) applied very small yield improvements.
The small yield improvements imbedded in Lark et al.(a) (i.e., 0.99%, 0.61%, 0.64% for corn,
soybeans, and wheat, respectively) may be compared with the observed yield improvements of
11.8% for corn, 15% for soybeans and 8.3% for wheat for the period 2007-2015 shown in Table
3. The comparison indicates that Lark et al.(a)’s too-small yield improvements led to an
overestimated need for additional land for corn production.
The second scenario presented in Table 6 repeats the first scenario but with a higher yield-toprice response of YDEL=0.175. This is not a high yield response, given the existing estimated
value for this parameter. However, it is large enough to compensate for the expansion in corn
land by savings in land due to yield improvements. This value of YDEL in combination with
assumed increases in crop prices by Lark et al.(a) generates yield increases of 5.43%, 3.33%,
3.5% for corn, soybeans, and wheat, respectively. These yield improvements are significantly
lower than the observed yield increases in the period 2007-2015 presented in Table 3. With
YDEL=0.175, the required area for producing 5.5 Bgal corn ethanol drops to 3.335 Mha, and the
land saving due to yield improvements is about the same, as shown in Table 6. Scenario 2 clearly
shows that with a better assessment of yield improvements, even the questionable modeling
framework of Lark et al.(a) projects no expansion in demand for cropland due to the RFS.
The third scenario presented in Table 6, unlike the first two, introduces demand response into the
picture. In this scenario, it is assumed that the domestic and foreign users of U.S. corn, soybeans,
and wheat are responding to higher crop prices with small price elasticities of 0.05 and 0.1 for
the domestic and foreign crop users respectively. The demand response, even with the assumed
small elasticities, generates land savings of 1.126 Mha.
19
This land saving result also suggests that Lark et al.(a) overlooked an important market-mediated
response and so overestimated the land use implications of ethanol. Note that we limited our
analyses of yield and demand responses to corn, soybean, and wheat to match the narrow
viewpoint of Lark et al.(a). Extending our analyses to other crops certainly suggests more room
for land use savings due to market-mediated responses.
In conclusion, Lark et al.(a) significantly overestimated the land use implications of ethanol
production because of the three factors explained above: miscalculated replacement of corn and
soymeal by DDGS, an assumed yield improvement close to zero, and overlooked reduction in
demand due to higher crop prices.
It is important to also note that Lark et al.(a) dismissed the large share that U.S. agriculture has
of commodity markets at the global scale. The next section examines the consequences of this
important omission.
Table 6. Effects of yield improvements and demand responses on corn ethanol land use changes.
Description Corn Soybeans Wheat
Status
quo in
2007
Yield (tonnes/ha) 9.46 2.81 2.70
Harvested area (ha) 35,013,780 25,959,240 20,638,784
Production (tonnes) 331,177,280 72,859,180 55,820,360
Net export (tonnes) 56,680,022 29,564,479 30,601,278
Used for ethanol (tonnes) 77,448,268
Domestic use 197,048,990 43,294,701 25,219,082
Price increases based on Lark et al. (% change) 31 19 20
Scenario
1
Percent change in yield with YDEL=0.032 0.99 0.61 0.64
Land needed to satisfy crop production of 2017 with higher yield (ha) 34,669,855 25,802,362 20,507,536
Land saving by crop due to higher yields compared with status quo (ha) -343,925 -156,878 -131,248
Total land use saving compared with status quo (ha) -632,052
Corn area requirement for 5.5 Bgal ethanol with higher yield (ha) 3,482,244
Net land area needed for 5.5 Bgal ethanol with higher yield (ha) 2,850,192
Scenario
2
Percent change in yield with YDEL=0.175 5.43 3.33 3.50
Land needed to satisfy crop production of 2017 with higher yield (ha) 33,211,637 25,123,686 19,940,700
Land saving by crop due to higher yields compared with status quo (ha)
-1,802,143 -835,554 -698,084
Total land use saving compared with status quo (ha)
-3,335,781
Corn area requirement for 5.5 Bgal ethanol with higher yield (ha) 3,335,780
Net land area needed for 5.5 Bgal ethanol with higher yield (ha) 0
Scenario
3
Price elasticity of domestic demand 0.05 0.05 0.05
Price elasticity of foreign demand 0.1 0.1 0.1
Savings in domestic demand due to higher prices (tonnes) -3,054,259 -411,300 -252,191
Savings in foreign demand due to higher prices (tonnes) -1,757,081 -561,725 -612,026
Total saving in demand (tonnes) -4,811,340 -973,025 -864,216
20
8. Estimation of price effects
In our original comments, we questioned the approached used by Lark et al.(a) to assess the price
impacts of RFS and its implications for land use change. In our original comments, we first
highlighted the work done by Filip et al. (2019) and noted that their findings are in sharp contrast
with the results of Lark et al.(a). In our original comments, we noted that Filip et al. (2019)
concluded that “price series data do not support strong statements about biofuels uniformly
serving as main leading source of high food prices and consequently the food shortages.”
Then we noted that “Lark et al. evaluated the price impact of 5.5 Bgal by using the observed
prices for the 2006-2010 period to evaluate the price impacts while their assignment of 5.5
billion gallons to the RFSs is for the first eight years between 2008 and 2016. During 2006-2010
crop prices increased significantly, but these prices dropped in the following years and came
back to much lower levels.” Following this statement, we provided some statistics to highlight
the odds with the Lark et al.(a) approach.
In response to our comments, Lark et al.(b) repeated their approach, said that the statistics we
provided are irrelevant, and stated that their “price effects modeling is valid and consistent with
existing literature.”
We believe that Lark et al.(b) misunderstood or misinterpreted our critique. We understand their
approach. They used the observed prices and their estimated BAU prices for 2006-2010 to assess
the price impacts on corn, soybeans, and wheat (see description of Table S1 in SI of Lark et
al.(a)), and they selected the eight years between 2008 and 2016 to calculate land use changes
due to the RFS. Our critique highlighted the mismatch between the time period of 2006-2010 and
the eight years between 2008 and 2016, as clearly noted in our original comments. For the sake
of clarity, we restate our critique in what follows.
Lark et al.(a) used the actual observations and their estimated BAU prices for the period 2006-
2010 to calculate the price impacts of RFS (5.5 Bgal ethanol) on corn, soybeans, and wheat
prices. They then used the results from that analysis to assess the land use impacts of the
additional demand for ethanol (5.5 BGal) between 2008 and 2016.
What justifies the selections of these time segments? What is the validity of using calculated
price impacts obtained for 2006-2010 to evaluate the induced land use changes by 5.5 Bgal of
ethanol during the years between 2008 and 2016? We believe that these are straightforward
questions. It is puzzling to use the price impacts of 2006-10 for the eight years between 2008 and
2106, when prices followed different patterns in these two periods.
Saving in land due to saving in demand by crop (ha) -482,498 -335,524 -308,724
Total saving in demand for land due to demand response (ha) -1,126,746
Land use saving compared with status quo in scenario 2 (ha) -3,335,781
Total land use saving (ha) -4,462,526
Corn area requirement for 5.5 Bgal ethanol in scenario 2 (ha) 3,335,780
Net change in land area (ha)
-1,126,746
21
In our original comments, we provided some statistics to show that the crop prices and their
changes are very different between 2006 and 2010 and between 2008 and 2016. At that time, we
used the FAO data to show that prices followed different patterns in these two periods. Here, we
use the data imbedded in Figure 1 of Lark et al.(a) to show the same issue.
Figure 4 replicates Figure 1 of Lark et al.(a) and Table 7 shows annual percent changes in corn,
soybeans, and wheat using the Lark et al. data (perhaps with some insignificant approximation,
as we read the data for the Figure 1 of Lark et al.(a)). The last two columns of Table 7 show the
average annual percent changes in crop prices for the five years 2006 to 2010 and the eight years
2008 to 2015. As shown in this table, the averages of the two periods are very different. The
averages for the first period are quite large: 33.5% for corn, 23.7% for soybeans, and 24.9% for
wheat. The averages for the second period are tiny or negative: 0.8% for corn, -1.9% for
soybeans, and -6.2% for wheat.
These comparisons clearly show that the prices followed different patterns in 2006-2010 and
2008-2016. That being the case, using estimated price impacts of RFS for the first time period
and applying them to the second is problematic.
Figure 4. Replication of observed prices presented in Figure 1 of Lark et al.(a).
Table 7. Annual % changes in observed corn, soybeans, and wheat prices from 2006-2016 and their
averages for 2006-2010 and 2008-2015. Obtained from Lark et al.(a) Figure 1 with some
approximations.
Year 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Average of
annual %
changes
2006-
10
2008-
15
Corn 59.1 51.4 -29.2 -6.7 92.9 -2.2 13.6 -36.7 -20.0 -5.3 33.5 0.8
Soybeans 33.6 67.7 -30.0 2.2 45.2 0.7 10.3 -3.3 -32.4 -8.2 23.7 -1.9
Wheat 20.8 108.6 -50.4 -16.7 62.0 -16.0 13.2 1.3 -25.6 -17.2 24.9 -6.2
22
The last part of the Lark et al.(b)’s response to our critique indeed discloses why Lark et al.(a)
selected the time period of 2006-10 and not the second time period for the calculation of price
impacts. Lark et al.(b) stated that: “As we explain in the supplementary text (SI ln 963), our
estimated price effects would be somewhat larger if we used all years up to 2016. This fact is
visible in Figure 1 in the main text”.
This statement is surprising and merits some attentions for two reasons. The first reason is that
the inclusion of a longer time period (up to 2016) led to a larger price impact of ethanol which is
inconsistent with the observation that corn prices fell after 2014 even though ethanol production
was higher than in the 2006-2010 period. The second reason is the fact that Lark et al.(a) did not
use the findings from the study that included the data up to 2016 for their analysis. There is more
justification for using estimates based on the updated data set than for applying estimates from
an earlier time period. The fact that Lark et al.(a) did not use the updated estimates may suggest
that they were not convinced about the validity of their model and its findings when extended to
include the later time period.
The statement of Lark et al.(b) mentioned above seems to confirm that the method used by Lark
et al.(a) generates somewhat larger price increases if they assess the price impacts for the correct
period, from 2008 to 2016, and to avoid estimating larger price impacts, the authors selected the
time period of 2006-2010, which is unrelated to their selected time period for the land use
assessment. Choosing an unrelated time period to avoid estimating larger price impacts is
not a scientific approach and leads to flawed results.
At the beginning of their response to our comments on this topic, Lark et al.(b) stated that
“Our price effects modeling is valid and consistent with existing literature. As explained in our
study’s main and supplementary texts, our modeling approach is the same as in Carter, Rausser,
and Smith (2017), which contains all the details of the model specification and identification
strategy.”
We make several comments related to this statement below.
a) Carter et al. (2017) provided an assessment for the price impact of corn ethanol and their
modeling approach focused only on corn. There are a large number of papers that have
examined the effects of corn ethanol production on crop prices and obtained a wide range
of different outcomes. Lark et al.(a) should have justified their reliance on only one study
for their price impact and analyzed the implications of using price effects from other
studies on their findings.
b) We contend that using the findings on price effects from a commodity-by-commodity
reduced form model in Carter et al. (2017) to analyze land use change is not appropriate
when multiple crops are competing for land because it does not explicitly consider
substitution among crops on the same land. Corn, soybeans, and wheat are produced
using a common factor of production: land. Farmers allocate their land to maximize their
profits. Using a single commodity inventory model and estimating separate equations for
these commodities is not a valid approach when three commodities compete for land. A
change in inventory for one of these commodities would affect farmers’ land allocation
and therefore crop prices.
23
c) Carter et al. (2017) evaluated the effect of the RFS on corn price in the context of a
closed economy, essentially missing the fact that the U.S. share of the global trade of
corn is not small. However, a closed economy approach is not a proper method to use for
highly traded commodities such as corn, soybeans, and wheat. In particular, the share of
U.S. in the global market for soybeans is significantly large. As can easily be calculated
from the first section of Table 6, about 17%, 41%, and 53% of U.S. corn, soybeans, and
wheat, respectively, are exported to other countries. The size of U.S. soybeans exports in
the global commodities market is large (See Taheripour and Tyner (2018 for details). It is
problematic to follow Carter et al. (2017) and use a closed economy modeling approach
for the three highly traded commodities of corn, soybeans, and wheat.
d) Finally, Carter et al. (2017) compared their estimated BAU prices with actual
observations to assess the impact of RFS on corn ethanol. This is a problematic
comparison. The Carter et al. (2017) BAU prices are obtained from their estimated model
for the goals of RFS1. The same model results for actual observations should be
compared with BAU, not the actual observations. Model results with RFS1 and RFS2
should have been compared, ceteris paribus, holding all other modeling assumptions the
same, instead of comparing model results in the BAU to actual observations. Figure 5
highlights the correct and problematic comparisons. The problematic comparison of BAU
to observed data assigns the model estimation error to the difference between actual
observation and BAU projection in each year. Depending on the sign of the error, this
comparison could lead to over- or underestimation.
Figure 5. Estimated price impact: Correct and problematic comparison.
24
9. Ignoring land category of cropland pasture in modeling land transition
In our original comments we noted “Lark et al. did not recognize cropland pasture as a subcategory of cropland in their analyses and perhaps treated this type of land as pasture land or
fallow land. This misidentification and the method used by the authors to assess land return is
likely to have artificially led to the additional demand for active cropland being met largely by
CRP land and not by cropland pasture.” Then we explained the reasons why ignoring the land
category of cropland pasture leads to overestimation of land use emissions.
In response to this comment, Lark et al.(b) admitted that they intentionally overlooked this type
of land, stating that “Our modeling of land transitions purposely avoids the separate problematic
category of Cropland-Pasture.” They then provided their reasons for disregarding this important
category of land. In what follows we respond to the Lark et al.(b) claims.
i. Unlike the claim made by Lark et al.(b), cropland pasture is not a “problematic category” of
land. The land category cropland-pasture is a standard category of land defined and
recognized by the Food and Agricultural Organization (FAO) of the United Nations and the
Intergovernmental Panel on Climate Change (IPCC) as “temporary pasture and meadows.”
By definition, it is a sub-category of cropland, representing a portion of the existing cropland
which has not been harvested and is temporally used as pasture. The U.S. Agricultural
Censuses (UACs) have identified this category of land as “cropland pasture”, For details, see
Taheripour et al. (2021).
ii. Lark et al.(b) claimed that “Cropland-pasture has been identified as an enigmatic land
classification that obfuscates valid estimation of LUC, and its use has contributed to
systematically downward-biased estimates of emissions from corn ethanol-induced LUC
(Malins et al. 2020; Spawn-Lee et al. 2021).” We have several responses to this statement.
• Taheripour et al. (2021) have responded to this claim, originally made by Malins et al.
(2020). We refer the readers to our 2021 paper and its supporting documents to evaluate
the credibility of the claim made by Malins et al. (2020). Here we provide some
additional analyses to shed more lights on this claim.
• Cropland pasture is not “an enigmatic land classification.” It is a clearly defined and
well-known land category, and its area has been recorded and reported in the UACs.
Table 5 shows the areas of cropland and cropland pasture since 1959. The area of
cropland pasture has fluctuated between 25 and 36 Mha from 1950 to 2002, constituting
about 14% to 20% of total cropland. The area of this land category has declined in the
last three UACs, to 14.5 Mha in 2007, 5.2 Mha in 2012, and 5.6 Mha in 2017.
Historically, the area of cropland pasture has decreased when commodity markets were
strong and vice versa. The area of this land category has been more than the total area of
Illinois and Iowa in many years. The area of cropland pasture has been always larger
than the area of CRP land, except in recent years. Lark et al.(b) labeled millions of
hectares of land classified under the land category of cropland pasture as “enigmatic” in
contradiction with the definitions, identifications, and data provided by the FAO, IPCC,
25
and USDA. In what follows we explain why the area of cropland pasture has declined in
the last three UACs.
• The claim that taking into account cropland pasture “obfuscates valid estimation of LUC,
and its use has contributed to systematically downward-biased estimates of emissions
from corn ethanol-induced LUC” needs careful attention. As mentioned above,
historically the area of cropland pasture has declined when commodity markets were
strong and vice versa. Of course, returning cropland pasture (which has been without
crop production for a short period) will not cause significant land use emissions
compared with the conversion of forest or natural pasture. While the existing
observations indicate that area of pasture has declined after 2002, there is no evidence of
deforestation, so it is wise to take into account reduction in this category of land in
assessing LUC due to biofuels. When actual observations do not support deforestation in
the U.S. but do support return of cropland pasture to crop production why one should
ignore actual observations. Taking into account reduction in cropland pasture reduces the
estimation of ILUC. Valid estimation of land use changes and their corresponding
emissions are a result of account actual observation rather than assumptions.
iii. Lark et al.(b) noted that the UACs do not provide annual data and that the definition of
cropland pasture has changed in recent UACs, and quoted a report provided the USDA
experts (Bigelow and Borchers, 2017) to confirm that the definition of cropland pasture
changed in the 2007 and 2012 UACs. We reply to these claims in what follows.
• The fact that UACs provide data every five years, i.e., we do not have annual data on
cropland pasture area, is a limitation but not a reason to ignore this land category and its
role in both managing the supply side of the crop markets and buffering unnecessary land
transitions. In fact, the UAC results provide the required frameworks for developing
many annual surveys conducted by the USDA on land use and land use changes.
Providing a good assessment for annual land transition from active cropland to cropland
pasture and back again is a valuable and nontrivial task. One cannot ignore the existing 5-
year information on this land category provided by the UACs.
• Changes in the definitions of data items happen frequently in censuses and annual
surveys, even in the NRI data. Inconsistencies across different versions of a database are
also common. In Table 1, we highlighted two inconsistencies between two versions of the
NRI data. Here we have another one. The NRI release of 2012 reported that 351.7 million
acres of cropland remained in this category of land between 2007 and 2012. The NRI
release of 2017 reported 348.1 million acres of cropland remain in the cropland category
for the same time period. The difference between these two numbers is 3.6 million acres,
which is not a small area. Should we recommend not using the NRI data because of this
and many other inconsistencies that may suggest some changes in the definitions,
approaches, or data processed? Of course not. For the same reasons, we cannot disregard
information on cropland pasture.
• Lark et al.(b) referred to Bigelow and Borchers (2017) to note that the cropland pasture
definition has changed in recent censuses. It is true that the definition of cropland pasture
26
in these censuses changed, and Bigelow and Borchers (2017) noted that fact. But that’s
not the whole story. As described by Bigelow and Borchers (2017), only a portion of the
observed reduction in area of cropland pasture was due to the change in definition of this
land category. Taheripour et al. (2021) noted that the area of cropland pasture has
declined in recent censuses due to the change in the definition, return of cropland pasture
to cropland, and conversion of cropland pasture to pasture land. Table 8 shows reductions
in cropland pasture and increases in pasture land in the 2007 and 2012 censuses.
Table 8. Historical data on area of cropland, cropland pasture, and pasture land in US Agricultural
Censuses since 1959.
Census Cropland Cropland
pasture
Share of cropland pasture in
cropland Pasture land
1959 181.3 26.5 14.6 188.7
1964 175.7 23.2 13.2 198.4
1969 185.7 35.7 19.2 183.8
1974 178.1 33.5 18.8 181.6
1978 183.7 29.6 16.1 175.4
1982 180.2 26.3 14.6 169.3
1987 179.4 26.3 14.7 166.1
1992 176.2 27.0 15.3 166.3
1997 174.5 26.1 15.0 160.6
2002 175.7 24.4 13.9 160.0
2007 164.5 14.5 8.8 165.4
2012 157.7 5.2 3.3 168.1
2017 160.4 5.6 3.5 162.2
iv. In our original comments, we noted that modeling land transitions due to ethanol are more
complicated than the simplified approach taken by Lark et al.(b). We noted that ignoring land
transitions among the sub categories of cropland, including cropland pasture, can artificially
in a biased way move the land transition to the pasture and CRP land categories, which are
the only options in Lark et al.(a). As noted above, Lark et al.(b) said that they intentionally
disregarded the land category of cropland pasture because the UACs do not provide annual
data for this land category. In what follows we show that Lark et al.(a) failed to properly use
the NRI data to identify land transitions within cropland and among the subcategories of this
type of land. Proper analyses of land transitions within cropland and between cropland and
pastureland with NRI data could help to capture correctly induced land use changes due to
ethanol.
• As we noted in our original comments, Lark et al.(a) picked two land transitions. The first
one is between CRP land and cropland. As we noted before, the observed transitions from
CRP to cropland were not induced directly by the RFS nor by ethanol consumption. The
observed reductions in the CRP land are not even limited to the transition from this
27
category of land to cropland. A large portion of the observed reductions in CRP land is
due to transitions from this category to pasture land. The observed reductions in CRP
land were mainly induced by budget cut approved by Congress, not farmer’s decision. In
addition to yield improvements, farmers have uncultivated/unused land available to
produce more crops in response to more demand for ethanol. As shown by the NRI data,
a large portion (37%) of the observed transitions from CRP to other land categories
between 2002 to 2017 was transition to pasture land. Even a portion of the CRP land
moved to cropland may have remained uncultivated. Conversion of CRP due to RFS or
ethanol consumption is a speculative assumption.
• Lark et al.(a) picked land transitions between pasture and cropland as well. They did not
specify the transitions between these land categories in a statistically valid manner. Table
S22 of the Lark et al.(a) SI shows that almost all of the estimated regional transitions
between cropland and pasture land are statistically insignificant.
• Figure 6 represents three land categories of NRI data including cropland, pasture land,
and CRP land, with their sub-categories. The NRI data include other land categories that
are not presented in this figure. Figure 6 also shows potential land transitions between the
main three categories and within each category with red arrows. Blue arrows indicate the
two land transitions that Lark et al.(a) chose to study.
• As shown in Figure 6, Lark et al.(a) ignored a number of important transitions, including
but not limited to transitions within cultivated cropland, shown on the top and left side of
the figure. For accurate results, it is crucial to correctly specify the annual transitions
within all row crops and between row crops and other subcategories of cultivated
cropland, such as fallow land, not planted land, and “pasture in cropland.” Pasture in
cropland refers to pasture in rotation with row crops. Disregarding these transitions leads
to an overestimation of ILUC for biofuels in general.
• Figure 6 also shows that transitions between subcategories of cropland and pasture land
can also be assessed, such as transitions between managed pasture and row crops. A
transition between these two types of land might not generate emissions, while a
transition between natural pasture and cropland might generate some emissions. Rather
than distinguishing between these two types of transitions, Lark et al.(a) bundled the two,
which results in larger emissions.
• While the NRI data set do not explicitly represent the land category of “cropland pasture”
defined by the UACs, it implicitly covers that type of land either in the main pasture land
category or in the sub-category of pasture in cropland. Rather than tracing the annual
transitions between and among these lands, Lark et al.(a) estimated overall transitions
between cropland and pasture land and produced statistically insignificant estimated land
transitions.
28
Figure 6. Three NRI land categories, their sub-components, and connections.
10. Conclusion
In a recent publication Lark et al.(a) examined “Environmental Outcomes of the US Renewable
Fuel Standard” by assembling a set of loosely connected problematic empirical methods. In a
detailed review, we find that this paper suffers from various deficiencies and provided flawed
analyses and problematic assessments.
In our comments, we noted that the authors failed to isolate the impact of the RFS on crop prices
and land use changes from other drivers of ethanol production and consumption. Lark et al.(b)
admitted this important fact and explicitly stated that their “results reflect the impacts of corn
ethanol demand in general, regardless of the source of such increases”. It seems appropriate to
revise the title of the original paper to reflect its contents.
The modeling approach used by Lark et al.(a) consists of assessing the price impacts of the RFS
on three commodities (corn, soybeans, and wheat) for the period of 2006 to 2010 plus an
assessment of land use changes for 8 years from 2008 to 2016. The authors, with no scientific
justification, incorrectly used the calculated price effects for 2006-2010 to calculate land use
29
changes from 2008 to 2016, while crop prices followed entirely different pattens in these two
time periods of time.
To evaluate the price impacts mentioned above, the authors followed Carter et al. (2017) who
assessed the impacts of the RFA on the corn price in a closed economy setup using an
econometric approach. In our review we discussed that the Carter et al. (2017) approach suffers
from various deficiencies. In addition, we noted that estimating three separate equations for corn,
soybeans and wheat prices based on Carter et al. (2017) misses the interactions between the
supply sides of these commodities through the market for land and that leads to flawed results.
To evaluate land use changes Lark et al.(a) concentrated on land conversion between pasture and
cropland and between CRP land and cropland. Table S22 of the Lark et al.(a) SI shows that
almost all of the estimated regional transitions between cropland and pasture land are statistically
insignificant. Hence, they build their analyses on the statistically insignificant assessments of
land transitions between cropland and pasture land. For land transition between cropland and
CRP land we discussed that the CRP lands were returned to crop production at the end of CRP
contracts, not induced by RFS. The reduction in area of CRP land were occurred due to budget
cut not RFS.
In our review, we showed that the use of CDL data in determining the location of converted land
and their characteristics at the grid cell level can lead to overestimation of GHG emissions of
ethanol. Furthermore, we discussed that the mapping between the estimated changes in
“pastureland” and CRP land obtained from the NRI data at a given data point and the CDL data
at the grid cell level can also lead to overestimation of GHG emissions of ethanol. We also used
the NRI data and showed that historically land transitions between pastureland and cropland
occurred between managed pasture and cropland. However, Lark et al.(a) incorrectly interpreted
the land transition between these two land categories as transition between natural pasture to
cropland with unjustifiable land use emissions.
We showed that Lark et al. selected the existing literature in various instances in favor of their
analyses, while the broad literature clearly shows otherwise.
Our review and analyses confirm that Lark et al. dismissed two important market-mediated
effects of biofuel production: Yield improvements and demand responses. We showed that
incorporating these effects, even for small fractions of the observed responses in 2007-2016,
leads to a small fraction of the estimated land use changes by Lark et al.(a).
Lark et al.(a)’s treatment of soil organic carbon (SOC) and reporting of its uncertainty appear to
be based on a misunderstanding of the information extracted from other studies, including their
inaccurate use of the carbon response functions (CRFs) derived from Poeplau et al. (2011) and
overestimation of the SOC sequestration potential in CRP lands.
In our review we outlined why Lark et al.(a) double counted N2O emissions. We also explained
with details why these authors overestimated the carbon content of CRP land.
In our original comments we noted that Lark et. (a) missed various important land transitions that
frequently occur within cropland (including but not limited to return of cropland pasture to crop
production) and hence they overestimated the land use impacts of corn ethanol. In response, Lark
30
et al.(b) admitted they intentionally ignored “cropland pasture” and stated that their approach
“purposely avoids the separate problematic category of Cropland-Pasture”. In response, we
explained that, unlike the claim made by Larke et al.(b), cropland pasture is a standard land
category recognized by the FAO, IPCC, and USDA Agricultural Censuses. We provided detailed
information about this land category and its magnitude and changes over time. We then showed
that, while the NRI data implicitly includes this type of land, Lark et al. made no effort to capture
its changes over time. Instead, they incorrectly assigned changes in the CRP land to ethanol.
In conclusion, we find that the Lark et al.(a) paper is more problematic than what we initially
evaluated to be the case.
11. References
Abraha, M., Gelfand, I., Hamilton, S.K., Chen, J. and Robertson, G.P., 2019. Carbon debt of
field-scale conservation reserve program grasslands converted to annual and perennial bioenergy
crops. Environmental Research Letters, 14(2), p.024019.
Benavides, P.T., H. Cai, M. Wang, and N. Bajjalieh, 2020, “Life-cycle analysis of soybean meal,
distiller-dried grains with solubles, and synthetic amino acid-based animal feeds for swine and
poultry production,” Animal Feed Science and Technology 268 (2020): 114607.
Bigelow, D. and Borchers, A., 2017. Major uses of land in the United States, 2012 (No. 1476-
2017-4340).
Bigelow, D., Claassen, R., Hellerstein, D., Breneman, V., Williams, R. and You, C., 2020. The
Fate of Land in Expiring Conservation Reserve Program Contracts, 2013-16 (No. 1476-2020-
047).
Berry, S. and Schlenker, W., 2011. Technical report for the ICCT: empirical evidence on crop
yield elasticities. Weather, pp.1-18.
Buenavista, R.M.E., Siliveru, K. and Zheng, Y., 2021. Utilization of distiller’s dried grains with
solubles: A review. Journal of Agriculture and Food Research 5, p. 100195.
Campbell, C.A. and Souster, W., 1982. Loss of organic matter and potentially mineralizable
nitrogen from Saskatchewan soils due to cropping. Canadian Journal of Soil Science 62 (4), pp.
651-656.
Carter, C.A., Rausser, G.C. and Smith, A., 2017. Commodity storage and the market effects of
biofuel policies. American Journal of Agricultural Economics 99 (4), pp. 1027-1055.
Choi, J.S. and Helmberger, P.G., 1993. How sensitive are crop yields to price changes and farm
programs?. Journal of Agricultural and Applied Economics, 25(1), pp.237-244.
Copenhaver, K., Hamada, Y., Mueller, S. and Dunn, J.B., 2021. Examining the characteristics of
the cropland data layer in the context of estimating land cover change. ISPRS International
Journal of Geo-Information,10 (5), p. 281.
31
Filip, O., Janda, K., Kristoufek, L. and Zilberman, D., 2019. Food versus fuel: An updated and
expanded evidence. Energy Economics, 82, pp.152-166.
Gelfand, I., Zenone, T., Jasrotia, P., Chen, J., Hamilton, S.K. and Robertson, G.P., 2011. Carbon
debt of Conservation Reserve Program (CRP) grasslands converted to bioenergy production.
Proceedings of the National Academy of Sciences 108 (33), pp. 13864-13869.
Goodwin, B.K., Marra, M.C., Piggott, N.E. and Mueller, S., 2012. Is yield endogenous to price?
An empirical evaluation of inter-and intra-seasonal corn yield response (No. 323-2016-11813).
Haque, M.A., Liu, Z., Demilade, A. and Kumar, N.M., 2022. Assessing the Environmental
Footprint of Distiller-Dried Grains with Soluble Diet as a Substitute for Standard Corn–Soybean
for Swine Production in the United States of America. Sustainability 14 (3), p. 1161.
Haile, M.G., Kalkuhl, M. and Braun, J.V., 2016. Worldwide acreage and yield response to
international price change and volatility: a dynamic panel data analysis for wheat, rice, corn, and
soybeans. In Food price volatility and its implications for food security and policy (pp. 139-165).
Springer, Cham.
Houck, J.P. and Gallagher, P.W., 1976. The price responsiveness of US corn yields. American
Journal of Agricultural Economics, 58(4), pp.731-734.
Huang, H. and Khanna, M., 2010. An econometric analysis of US crop yield and cropland
acreage: implications for the impact of climate change (No. 320-2016-10264).Irwin, S. and
Good, D., 2013. Understanding the pricing of distillers’ grain solubles. farmdoc daily 3.
Irwin, S. and Good, D., 2013. Understanding the pricing of distillers’ grain solubles. farmdoc
daily, 3.
Kim, S., Kim, C., Han, S.H., Lee, S.T. and Son, Y., 2018. A multi-site approach toward
assessing the effect of thinning on soil carbon contents across temperate pine, oak, and larch
forests. Forest ecology and management, 424, pp.62-70.
Kwon, H., Ugarte, C.M., Ogle, S.M., Williams, S.A. and Wander, M.M., 2017. Use of inverse
modeling to evaluate CENTURY-predictions for soil carbon sequestration in US rain-fed corn
production systems. PloS one 12 (2), p. e0172861.
Lark, T.J., Hendricks, N.P., Smith, A., Pates, N., Spawn-Lee, S.A., Bougie, M., Booth, E.G.,
Kucharik, C.J. and Gibbs, H.K., 2022a. Environmental outcomes of the US Renewable Fuel
Standard. Proceedings of the National Academy of Sciences 119 (9), p. e2101084119.
Lark, T.J., Hendricks, N.P., Smith, A., Pates, N., Spawn-Lee, S.A., Bougie, M., Booth, E.G.,
Kucharik, C.J. and Gibbs, H.K., 2022b. Reply to Taheripour et al.: Comments on
“Environmental Outcomes of the US Renewable Fuel Standard.” Available from
https://asmith.ucdavis.edu/news/environmental-outcomes-us-renewable-fuel-standard-reply.
Lyons, D.C. and Thompson, R.L., 1981. The effect of distortions in relative prices on corn
productivity and exports: a cross-country study. Journal of Rural Development/NongchonGyeongje, 4(1071-2019-993), pp.83-102.
32
Malins, C., Plevin, R. and Edwards, R., 2020. How robust are reductions in modeled estimates
from GTAP-BIO of the indirect land use change induced by conventional biofuels?. Journal of
Cleaner Production, 258, p.120716.
Miao, R., Khanna, M. and Huang, H., 2016. Responsiveness of crop yield and acreage to prices
and climate. American Journal of Agricultural Economics, 98(1), pp.191-211.
Newton, J.D., Wyatt, F.A. and Brown, A.L., 1945. Effects of cultivation and cropping on the
chemical composition of some western Canada prairie province soils. Part III. Scientific
Agriculture 25 (11), pp. 718-737.
Piñeiro, G., Jobbágy, E.G., Baker, J., Murray, B.C. and Jackson, R.B., 2009. Set‐asides can be
better climate investment than corn ethanol. Ecological Applications 19 (2), pp. 277-282.
Plevin, R.J., Gibbs, H.K., Duffy, J., Yui, S. and Yeh, S., 2014. Agro-ecological zone emission
factor (AEZ-EF) model (v47) (No. 1236-2019-175).
Poeplau, C., Don, A., Vesterdal, L., Leifeld, J., Van Wesemael, B.A.S., Schumacher, J. and
Gensior, A., 2011. Temporal dynamics of soil organic carbon after land‐use change in the
temperate zone–carbon response functions as a model approach. Global Change Biology 17 (7),
pp. 2415-2427.
Reeder, J.D., Schuman, G.E. and Bowman, R.A., 1998. Soil C and N changes on conservation
reserve program lands in the Central Great Plains. Soil and Tillage Research 47 (3-4), pp. 339-
349.
Rosas, F., Lence, S.H. and Hayes, D.J., 2019. Crop yield responses to prices: a Bayesian
approach to blend experimental and market data. European Review of Agricultural Economics,
46(4), pp.551-577.
Ruan, L. and Philip Robertson, G., 2013. Initial nitrous oxide, carbon dioxide, and methane costs
of converting conservation reserve program grassland to row crops under no‐till vs. conventional
tillage. Global Change Biology 19 (8), pp. 2478-2489.
Spawn, S.A., Lark, T.J. and Gibbs, H.K., 2019. Carbon emissions from cropland expansion in
the United States. Environmental Research Letters, 14(4), p.045009.
Spawn-Lee, S.A., Lark, T.J., Gibbs, H.K., Houghton, R.A., Kucharik, C.J., Malins, C., Pelton,
R.E. and Robertson, G.P., 2021. Comment on “Carbon intensity of corn ethanol in the United
States: state of the science.” Environmental Research Letters, 16 (11), p. 118001.
Qin, Z., Dunn, J.B., Kwon, H., Mueller, S. and Wander, M.M., 2016. Soil carbon sequestration
and land use change associated with biofuel production: empirical evidence. Gcb Bioenergy 8
(1), pp. 66-80.
Taheripour, F., Mueller, S. and Kwon, H., 2021. Response to “how robust are reductions in
modeled estimates from GTAP-BIO of the indirect land use change induced by conventional
biofuels?” Journal of Cleaner Production 310, p.127431.
33
Tyner, W. and Taheripour, F., 2008. Biofuels, policy options, and their implications: analyses
using partial and general equilibrium approaches. Journal of Agricultural & Food Industrial
Organization 6 (2).
U.S. Department of Agriculture. 2015. Summary Report: 2012 National Resources Inventory,
Natural Resources Conservation Service, Washington, DC, and Center for Survey Statistics and
Methodology, Iowa State University, Ames, Iowa.
http://www.nrcs.usda.gov/technical/nri/12summary.
U.S. Department of Agriculture. 2018. Summary Report: 2015 National Resources Inventory,
Natural Resources Conservation Service, Washington, DC, and Center for Survey Statistics and
Methodology, Iowa State University, Ames, Iowa.
http://www.nrcs.usda.gov/technical/nri/15summary.
U.S. Department of Agriculture. 2020. Summary Report: 2017 National Resources Inventory,
Natural Resources Conservation Service, Washington, DC, and Center for Survey Statistics and
Methodology, Iowa State University, Ames, Iowa.
https://www.nrcs.usda.gov/wps/portal/nrcs/main/national/technical/nra/nri/results/.
Weersink, A., Cabas, J.H. and Olale, E., 2010. Acreage response to weather, yield, and price.
Canadian Journal of Agricultural Economics/Revue canadienne d’agroeconomie, 58(1), pp.57-
72.
Weightman, R.M., Cottrill, B.R., Wiltshire, J.J.J., Kindred, D.R. and Sylvester‐Bradley, R.,
2011. Opportunities for avoidance of land‐use change through substitution of soya bean meal and
cereals in European livestock diets with bioethanol coproducts. Gcb Bioenergy 3 (2), pp. 158-
170.
Xu, H., Sieverding, H., Kwon, H., Clay, D., Stewart, C., Johnson, J.M., Qin, Z., Karlen, D.L. and
Wang, M., 2019. A global meta‐analysis of soil organic carbon response to corn stover removal.
Gcb Bioenergy., 11 (10), pp. 1215-1233.
Yu, B., Liu, F. and You, L., 2012. Dynamic agricultural supply response under economic
transformation: a case study of Henan, China. American Journal of Agricultural Economics,
94(2), pp.370-376.
Zenone, T., Gelfand, I., Chen, J., Hamilton, S.K. and Robertson, G.P., 2013. From set-aside
grassland to annual and perennial cellulosic biofuel crops: Effects of land use change on carbon
balance. Agricultural and Forest Meteorology 182, pp. 1-12.
34
Appendix A: Analysis of the Lark et al.(a) Cropland Expansion Layer
35