The impact of considering land intensification and updated data on biofuels land use change and emissions estimates

GTAP-BIO database version 9

The standard GTAP databases do not include production, consumption, and trade of biofuels. Taheripour et al. [16] introduced the first generation of biofuels (including grain ethanol, sugarcane ethanol, and biodiesel) into the GTAP standard database version 6, which represented the world economy in 2001 [17]. The early versions of the GTAP-BIO model were built on this database and used in several applications and policy analyses [3, 4, 18–21]. The California Air Resources Board (CARB) developed its first set of ILUC values using this database and early versions of the model [22]. The Argonne National Lab also used the results of this model in developing the early versions of the life cycle analyses (LCA) of biofuels [21, 23].

When the standard GTAP database version 7, which represented the world economy in 2004 was released [24], Taheripour and Tyner [25] introduced first- and second-generation biofuels into this database. Several alternative aggregations of this database have been developed and used in various studies to evaluate the economic and land use impacts of biofuel production and polices [26–31]. CARB has used this database to develop its final ILUC values [32, 33], and Argonne National Lab also used the outcomes obtained from this database in its more recent LCA analyses.

The GTAP-BIO 2004 database in comparison to its 2001 version had several advantages including but not limited to: (1) providing data on cropland pasture for the US and Brazil; (2) disaggregating oilseeds into soybeans, rapeseed, palm, and other oilseeds; (3) disaggregating coarse grains into sorghum and other coarse grains; (4) introducing cellulosic crops and corn stover collection as new activities into the database; (5) disaggregating vegetable oil industry into soybean oil, rapeseed oil, palm oil, and other vegetable oils and fats and their corresponding meal products; (6) dividing the standard food industry of GTAP into two distinct food and feed industries; and (7) covering a wide range of biofuels including ethanol produced from grains, ethanol produced from sugar crops, four types of biodiesel produced from soybean oil, rapeseed oil, palm oil, and other oils and fats, three types of cellulosic ethanol produced form corn stover, switchgrass, and miscanthus and three types of drop-in cellulosic biofuels produced from the corn stover, switchgrass, and miscanthus.

The GTAP-BIO 2004 database with all of the above advantages is now out-of-date. During the past decade, the global economy has changed significantly with major consequences for agricultural and energy markets including biofuels. On one hand, demand for agricultural products has increased across the world at different rates due to growths in income and population. Expansion in biofuel production due to public policies has contributed to the expansion in demand for agricultural products in some regions and at the global scale, as well. On the other hand, the agricultural sector has evolved considerably across the world: crop production and its geographical distribution have changed, the mix of crops produced in most countries has changed, crop yields have improved due to technological progress in many regions, crop production has been negatively affected in some regions due to severe climate conditions, and international trade in agricultural products has changed. Major changes occurred in the livestock industry, as well: demand for meat and meat products has shifted from red meat towards white meat, more biofuels by-products and meals were used in animal feed rations, and land intensification has been extended in the livestock industry. The biofuel industry has grown rapidly across the world and, in particular, in US, Brazil, and EU. Biofuel producers now operate more efficiently than before. Unlike the early 2000s, the biofuel industry is now a mature industry which operates without government subsidies. However, they still benefit from biofuel mandates. The 2004 database misses all these changes and many other changes which occurred in the global economy. Therefore, it becomes necessary to update the GTAP-BIO database.

To accomplish this task, following our earlier work in this area [16, 25, 34], we explicitly introduced biofuels into the latest publicly released version (V9) of the standard GTAP database which represents the world economy in 2011 [35]. This means is that all the steps that we followed to introduce biofuels into the 2001 and 2004 databases had to be repeated for the 2011 GTAP database but using 2011 data for all the biofuels components. Thus, production, consumption, trade, prices, and co-products had to be introduced into the 2011 database. The full description of this task is reported in [36]. Here, we explain the main important aspects of this task.

Improvements in GTAP-BIO model

Birur et al. [1] used an improved version of the GTAP-E model [41] and developed the first version of the GTAP-BIO model to analyze the impacts of biofuel production on energy and agricultural markets and to study the market. This early model version was able to trace market-mediated responses due to biofuel production. Responses such as but not limited to: (1) increases in crop prices due to expansion in feedstock demand for biofuel production; (2) reductions in crop demands in non-biofuel uses such as food and feed; (3) changes in the global trade of crops and other agricultural products; (4) expansion in crop supplies across the world; (5) substitution between biofuels and fossil fuels; (6) crop switching as relative prices changed; and (7) competition for limited resources. However, the model was not able to accurately quantify these impacts and was missing several other important market-mediated responses due to several limitations.

The first version of the model did not include biofuel by-products such as Distiller’s Dried Grains with Soluble (DDGS) and oilseed meals. Hence, the model was missing the impacts of biofuel production on the livestock industry and animal feed rations. Therefore, it provided misleading results on livestock demand for crops, leading to overestimation of biofuel impacts on demand for crops and land use changes. In addition, the first model did not consider the fact that productivity of new land likely would be lower than the existing cropland. Furthermore, the first model did not include any yield response to higher crop prices. More importantly, it was incapable to trace changes in physical land. Over the past decade, many modifications were introduced to GTAP-BIO to improve its performance and eliminate its initial deficiencies. Golub and Hertel [42] explained some of the early modifications. Here, we briefly outline them and introduce some newer modifications.

Taheripour et al. [3, 4] introduced biofuel by-products in the model and defined a module to take into account substitution between biofuels by-products (such as DDGS and oilseed meals) and feed crops in livestock feed rations. Hertel et al. [20] improved the model to distinguish between productivities of the new and existing croplands. They developed a new land supply system to trace changes in physical land. In addition, they defined a module to better take care of crop yield responses to changes in crop prices and production costs. The impacts of these modifications on the outcomes of the model were substantial, basically leading to lower induced land use changes compared with the initial model.

The three main modifications made by Hertel et al. [20] were significant contributions. However, these authors established their modifications based on some limited real-world observations. First, they assumed that the productivity of new land is about 2/3 of the productivity of existing cropland everywhere across the world. Second, they assumed that the land transformation elasticity among forest, pasture, and cropland equals to 0.2 across the world, and also used a uniform land transformation elasticity of 0.5 to govern allocation of cropland across alternative crops everywhere around the world. Finally, they assumed that crop yield response with respect to changes in profitability of crop production is uniform across regions and crops. They also assumed that crop harvest frequency remains fixed, meaning no expansion in multiple cropping and no conversion of idled cropland to crop production. Many of these limitations were removed over time.

Tyner et al. [23] partially removed the last issue mentioned above by introducing cropland pasture into the model for only US and Brazil, where data were available. Cropland pasture is a particular marginal cropland that usually is used as pasture land but moves to cropland when more cropland is needed. The model developed by these authors and the subsequent work continued to ignore multiple cropping and assumed idled cropland will remain idle.

Taheripour et al. [5] used a biophysical model (TEM) and estimated a set of extensification parameters which represent productivity of new cropland versus the existing land by region at the spatial resolution of Agro-Ecological Zone. Using a tuning process, Taheripour and Tyner [29] developed a set of land transformation elasticities by region according to recent real-world observations on land use changes across the world. These land transformations elasticities govern land allocation across land cover categories and distribute cropland among crops.

Recently, Taheripour et al. [43] introduced several more important improvements: First, they altered the land use module of the model to take into account intensification in cropland due to multiple cropping and/or returning idled cropland to crop production. They defined a new set of regional intensification parameters and determined their magnitudes according to observed land use changes across the world in recent years. They also altered the assumption that the elasticity of yield improvement with respect to changes in profitability of crops is uniform across regions. Instead, they defined regional yield responses and tuned their magnitudes according to observed regional changes in crop yields.

These model improvements were targeted towards the first-generation biofuels. Taheripour and Tyner [44] developed a special version of the model (called GTAP-BIO-ADV) to examine the economic and land use impacts of the second-generation biofuels. Unlike other versions of the GTAP-BIO model which put all crops in one nest in the land supply tree, the GTAP-BIO-ADV model uses a different land supply tree which puts cropland pasture and dedicated crops (such as miscanthus and switchgrass) in one nest and all other crops in another nest and allows the land to move between the two nests. They used this setup to avoid conversion of food crops to dedicated energy crops to make greater use of cropland pasture (a representative for marginal land) to produce dedicated energy crops. The GTAP-BIO-ADV model was developed prior to the tuning process described above and only includes those model modifications that were available when the model was developed in 2011.

This paper brings all the modifications explained above less than one umbrella and generates a comprehensive model to have the first- and second-generation biofuels in one model. We also match the model with the 2011 GTAP-BIO database introduced in the data section. Then, we examined the land use impacts and the biofuel pathways outlined in the next sections. Henceforth, we refer to this model as GTAP-BIO-ADV11.

The modeling framework used in this paper is based on the latest model introduced by Taheripour et al. [43] which includes all the modification made in the GTAP-BIO model over time including intensification in cropland due to multiple cropping and returning idled cropland to crop production. To do simulations for the second-generation biofuels, we alter the land supply tree of this model according to the land supply tree of the GTAP-BIO-ADV model. The top left and right panels of Fig. 1 represent the land supply trees of the latest version of the GTAP-BIO and GTAP-BIO-ADV models, respectively. The bottom panel of this figure shows the mix of these two panels which we used in this paper. As shown in the bottom panel, the land supply tree of the new model uses two nests to govern changes in land cover and two nests to manage allocation of cropland among crops, including miscanthus and switchgrass. At the lowest level of this tree, available land is allocated between forest and a mix of cropland–pasture. The second level allocates the mix of cropland–pasture to cropland and pasture. Then, at the third level, cropland is divided between the traditional crops (first nest of cropland) and dedicated crops including cropland pasture (second nest of cropland). Finally, at the top level, the first category of land is allocated among the traditional crops, and the second category between miscanthus, switchgrass, and cropland pasture.

The land transformation elasticities used with this specification match the tuned elasticities reported by Taheripour and Tyner [29] for the land cover and allocation of cropland among the traditional crops. For the cropland nest including miscanthus, switchgrass, and cropland pasture, following Taheripour and Tyner [44], we used a relatively large land transformation elasticity to support the idea of producing dedicated crops on marginal cropland and to avoid a major competition between the traditional crops and dedicated energy crops. For the nest between the first and second groups of cropland, we use the same tuned land transformation elasticities which we used in land allocation among the first group of crops (i.e., traditional crops). With this assignment, the new model replicates the results of the old model for the first-generation biofuels.

The modeling framework developed by Taheripour et al. [43] takes into account intensification in cropland due to multiple cropping and/or conversion of unused cropland. These authors introduced a new land intensification factor into the model and tuned it according to the actual recent historical observations. The modeling framework used in this paper adopts the approach developed by these authors. However, it required changes to introduce land intensification in the new model which uses a different land supply structure.

Taheripour et al. [36] used several relationships to introduce land intensification (due to multiple cropping and or conversion of unused land to cropland) and endogenously determine the size of afs by region. Among all modifications, they used to accomplish this task, they introduced a parameter, called intensification factor and denoted by γ _r, which represents the magnitude of intensification by region. This parameter varies between 0 and 1 (i.e. 0 ≤ γ _r ≤ 1). When \(\gamma_{\text{r}} = 1,\) there is no land intensification. In this case, any expansion in harvested area leads to an expansion in cropland which comes from conversion of forest and/or pasture. On the other hand, when \(\gamma_{\text{r}} = 0,\) it shows that an expansion in harvested area will not expand cropland. In this case, the additional harvested area comes from multiple cropping and/or converting unused cropland to crop production. Taheripour et al. [43] determined the regional values for this parameter, according to recent observed trends in land intensification across the world. Figure 2 represents the regional values of this parameter.

As shown in Fig. 2, in China and India, the parameter of land intensification equals 0, indicating that in these two countries, an expansion in harvested area does not lead to an expansion in cropland. On the other hand, in some countries/regions, the parameter of land intensification is close to 1, for instance Japan and East Asia. In these regions, any expansion in harvested area will equal an identical expansion in cropland with no intensification. Finally, in some countries/regions, the land intensification parameter is in between 0 and 1, say in Brazil and sub-Saharan Africa. In these regions, a portion of expansion in harvested area comes from land intensification and a portion from expansion in cropland. We use these values in our new model with one exception. For the case of Malaysia–Indonesia region, while the intensification parameter is less than 1, we assumed no intensification in this region, because it is the main source of palm oil and multiple cropping for palm tree is meaningless.

Following the existing literature [45, 46] which confirms yield improvement due to higher crop prices, Taheripour et al. [43] developed a set of regional elasticities which show yield to price response (known as YDEL) by region. Figure 3 represents these regional yield elasticities. Unlike the earlier version of the GTAP-BIO model which commonly assumed YDEL = 0.25, as shown in Fig. 2, the size of this elasticity varies between 0.175 and 0.325. Several regions including South America, East Asia, and Oceania have the lowest yield response, while Brazil has the highest rate.

Land use changes

In addition to changes in land cover, expansion in corn ethanol generates changes in the mix of cropland. In particular, it transfers some cropland pasture to the traditional crops. For the expansion in corn ethanol from 2011 to 15 BGs, about 129 thousand hectares of cropland pasture will be converted to the traditional crops, as shown in the first panel of Table 1. This is about 0.12 hectares per 1000 gallons of ethanol. For the case of corn ethanol, deforestation covers 32% of the land requirement and the rest (68%) is due to conversion of pasture to cropland.

An expansion in soybean biodiesel produced in the US by 0.5 BGs increases the global harvested area by about 64.5 thousand hectares, but only 56% of this expansion transfers to new cropland due to intensification. Therefore, global cropland increases by 36.1 thousand hectares. The index of land requirement for 1000 gallon of soybean biodiesel is about 0.07 hectares. Ignoring the land intensification, this index jumps to 0.13 hectares per 1000 gallons of soybean biodiesel. These indices are similar to their corresponding values for the cases of corn ethanol. For this pathway, the rate of conversion from cropland pasture to traditional crops is about 0.13 hectares per 1000 gallons of biodiesel, very similar to the corresponding rate for corn ethanol.

We now turn to induced land use changes for cellulosic biofuels produced from dedicated energy crops such as miscanthus or switchgrass. The narrative of induced land use changes for these biofuels is entirely different from the description of induced land use changes for the first-generation biofuels producing biofuels (say ethanol) from traditional crops (say corn) generates market-mediated responses such as reduction in consumption of crops in non-biofuel uses, switching among crops, expansion in biofuels by-products (which can be used in livestock feed rations instead of crops), and yield improvement. These market-mediated responses reduce the land use impacts of producing biofuels from traditional crops as described by Hertel et al. [20]. However, producing cellulosic biofuels from energy crops such as miscanthus or switchgrass may not generate these market-mediated responses.

For example, consider producing bio-gasoline from miscanthus, which we examine in this paper. This pathway produces no animal feed by-product. Therefore, an expansion in this biofuel does not lead to a reduction in livestock demand for crops. Miscanthus is not used in other industries. Hence, we cannot divert its current uses to biofuel production. Thus, miscanthus should be produced for every drop of bio-gasoline. For example, if we plan to produce 1 BGs of miscanthus bio-gasoline, then we need about 775 thousand hectares of land (with a conversion rate of 66.1 gallons per metric ton of miscanthus and 19.5 metric tons of miscanthus per hectare as we assumed in developing the GTAP-BIO database). Now, the question is: From where will the required land for miscanthus production come?

It is frequently argued that dedicated energy crops should not compete with the traditional food crops. This means no or little conversion from the traditional food-feed crops to cellulosic energy crops. It is also commonly believed that cellulosic energy crops should be produced on low-quality “marginal land”. Beside this widespread belief, the definition and availability of “marginal land” are subject to debate [49]. If the low-quality marginal land is entirely unused, then producing cellulosic crops on these lands may not significantly affect competition for land. In this case, unused land will be converted to miscanthus as needed to meet the feedstock demand for the stipulated expansion in cellulosic biofuel.

However, if the low-quality marginal land is used by livestock producers as grazing land (e.g., cropland pasture in the US), then producing energy crops on cropland pasture directly and indirectly affects the livestock industry, and that generates some consequences. In this case, the livestock industry demands more feed crops, uses more processed feed, and/or converts natural forest to pasture in response to converting cropland pasture to miscanthus.

Now, consider the induced land use changes for the third experiment which extends production of the US bio-gasoline from miscanthus by 1 BGs. As shown in the bottom panel of Table 1, the anticipated expansion in miscanthus bio-gasoline increases the global harvested area by 156.4 thousand hectares. However, due to intensification, the global cropland area grows only by 79.7 thousand hectares. Therefore, the index of land requirement for 1000 gallons of miscanthus bio-gasoline is about 0.08 hectares in the presence of land intensification. Ignoring intensification, the index of land requirement increases to 0.16 hectares per 1000 gallons of bio-gasoline. These land requirement indices are not very different from the corresponding figures for corn ethanol. However, three is a major difference between corn ethanol and miscanthus bio-gasoline when we compare their impacts on cropland pasture.

As shown in Table 1, an expansion in US miscanthus bio-gasoline by 1 BG converts 1077.6 thousand hectares of cropland pasture to cropland. This is about 1.08 hectares per 1000 gallons of miscanthus bio-gasoline. This figure is approximately 9 times higher than the corresponding figure for corn ethanol. This difference is because producing miscanthus bio-gasoline does not create the market-mediated responses which corn ethanol generates. The change in cropland pasture area (i.e., 1077.6 thousand hectare) is higher than the direct land requirement for producing 1 BG of miscanthus bio-gasoline (i.e., 763 thousand hectares). When the livestock industry gives up cropland pasture at a large scale, it uses more feed crops and/or processed feed items, and that generates some land use changes including more conversion of cropland pasture to traditional crops. Furthermore, a large conversion of cropland pasture to miscanthus increases the rental value of pasture land (a substitute for cropland pasture) significantly, and that generates some incentives for a mild deforestation in the US, as shown in the lowest panel of Table 1. In the third experiment, the price of miscanthus increases by 53% and the livestock price index (excluding non-ruminant) goes up by about 0.5% which is 5 times higher than the corresponding figure for the forestry sector. Pasture rent grows at about 5% across US AEZs, while the corresponding rate for forest is less than 1%. For the case of corn ethanol, which induces mild conversion of cropland pasture forest and pasture rents grow similarly at rates less than 1% across AEZs in the US. Finally, it is important to note that the tuned land transformation elasticity for forest to agricultural land in the US is small, according to recent observations [29]. In conclusion, while producing miscanthus bio-gasoline slightly increases demand for cropland, it induces major shifts in marginal land (say cropland pasture) to miscanthus production.

Land use emissions

First, consider induced land use emissions for the first-generation biofuels including corn ethanol and soybean biodiesel for four alternative modeling and database cases: (1) 2004 database with no intensification; (2) 2004 database with intensification; (3) 2011 with no intensification; and (4) 2011 with intensification. The emission results for the first three cases (i.e., cases 1, 2, 3) are taken from Taheripour et al.  [43]. The last case represents the results of the simulations conducted in this paper.

Figure 4 shows the results for corn ethanol. With intensification in cropland, an expansion in US ethanol from its 2011 level to 15 BGs generates 12 g CO₂e/MJ emissions. The corresponding simulation with no intensification generates 23.3 g CO₂e/MJ emissions. This means that the new model which takes into account intensification in cropland and uses tuned regional YDEL parameters generates significantly lower emissions, approximately by half. The corresponding cases obtained from the 2004 databases represent the same pattern, but demonstrate lower emissions rates. An expansion in corn ethanol from its 2004 level to 15 BGs generates 8.7 g CO₂e/MJ emissions with intensification and 13.4 g CO₂e/MJ with no intensification.

These results indicate that the 2011 database generates higher emissions for corn ethanol compared with the 2004 databases, regardless of modeling approach. However, the new model which takes into account intensification in cropland and uses tuned regional YDEL values projects lower emissions, regardless of the implemented database. The 2011 database generates more emissions for corn due to several factors including but not limited to: (1) less availability of cropland pasture in the US in 2011; (2) less flexibility in domestic use of corn in 2011; (3) less flexibility in US corn exports in 2011; (4) smaller US corn yield in 2011; (5) more reductions in US crop exports (in particular soybean and wheat) in 2011; (6) larger DDGS trade share in 2011; (7) smaller capital share in corn ethanol cost structure; and (8) finally, the marginal land use impacts of ethanol in 2011 are much larger than 2004, because the base level of ethanol in 2011 is much larger than 2004.

Figure 5 shows the results for soybean biodiesel. In the presence of intensification in cropland, an expansion in the US soybean biodiesel by 0.5 BGs generates 18 g CO₂e/MJ emissions. The corresponding simulation with no intensification generates 25.5 g CO₂e/MJ emissions. This means that, similar to the cases for corn ethanol, the new model which takes into account intensification in cropland and uses tuned regional YDEL parameters generates significantly lower emissions. The corresponding cases obtained from the 2004 databases represent the same pattern. An expansion in the US soybean biodiesel by 0.5 BGs generates 17 g CO₂e/MJ emissions with intensification and 21.6 g CO₂e/MJ with no intensification. Furthermore, producing soybean biodiesel in the US encourages expansion in vegetable oils produced in some other countries including more production of palm oil in Malaysia and Indonesia on peat land, which entails extremely high emissions. This is one reason why land use change emissions induced by US soybean biodiesel production are generally higher than those induced by US corn ethanol production.

Unlike the case of corn ethanol, these results indicate that the 2011 database generates slightly higher emissions for soybean biodiesel compared with the 2004 databases, regardless of modeling approach. This observation is due to several factors including but not limited to: (1) conversion of a larger portion of US soybean exports to domestic use in 2011 which reduces the size of land conversion in US; (2) Brazil, Canada, and other countries produce more soybeans in 2011; (3) significantly larger oilseed yields across the world (except for US) generates weaker land conversion outside the US; (4) larger availability of oilseed meals in 2011 which contributes to a higher share of pasture in 2011; and larger share of palm oil in total vegetable oils in 2011.

We now turn to induced land use emissions for miscanthus bio-gasoline. Two alternative cases are examined to highlight the role of soil carbon sequestration gained from production of miscanthus on marginal land. First, we assume that producing miscanthus on cropland pasture does not improve soil carbon sequestration. Then, following the literature [48, 49]³, we take into account the fact that producing miscanthus on marginal land improves the soil carbon content. The existing literature confirms that producing miscanthus on marginal land improves its soil carbon content.

For the first case, an expansion in US miscanthus bio-gasoline by 1 BGs generates about 27 g CO₂e/MJ emissions. Compared with corn ethanol and soybean biodiesel, this figure is large. As mentioned before, an expansion in US miscanthus bio-gasoline by 1 BGs transfers about 1117.6 thousand hectares of cropland pasture to miscanthus production and other tradition crops. Only about 70% of this conversion goes to miscanthus. Hence, if we ignore the carbon saving from miscanthus production, then producing bio-gasoline from miscanthus generates more emissions than corn ethanol. For the second case, as shown in Fig. 6, the emissions score for miscanthus to bio-gasoline drops to about −6 g CO₂e/MJ. This figure is in line with the results reported by Wang et al. [50]. These authors used induced land use results obtained from an earlier version of the GTAP model and emissions factors from the CCLUB calculated that producing ethanol from miscanthus generates negative land use emissions by −7 g CO₂e/MJ. On the other hand, Dwivedi et al. [45], who used farm and firm level data in combination with some limited field experiments, reported that converting miscanthus to ethanol generates about −34 to −59 g CO₂e/MJ land use emissions. These results underscore the fact that for the case of cellulosic biofuels, the magnitude of induced land use emissions varies significantly by the method of calculating land use changes and largely depends on the assigned emission factor to the converted marginal land.