Deforestation displaced: trade in forest-risk commodities and the prospects for a global forest transition

We attribute observed forest loss to agricultural and forestry commodities using a simple land-balance model encompassing cropland, pastures, forest plantations and (indirectly) other land uses. The model is implemented at national level, except for Brazil and Indonesia, where it is implemented at subnational level in order to capture geographically divergent drivers of deforestation in the countries that both hold a large share of remaining tropical forests, but also account for a large share of tropical forest loss. The model is based on two main premises: (1) where cropland expands, it first expands into pastures (if there was a gross loss of pasture area) and then into forests (if there was gross forest loss), and (2) where pastures and forest plantation areas expand, they primarily replace forest land. While simple, these premises are consistent with data showing that forests and other native vegetation (such as woodlands and shrublands) are the main sources of new agricultural land in the tropics [17], that the expansion of forest plantations tend to come at the expense of natural forests [18–21], but also that (at least in the tropical Americas) pastures are a significant source of new cropland [17, 22]. We thus argue that the land-balance model captures the predominant deforestation-related land-use transitions across the tropics.

More specifically, the land-balance model attributes forest loss in a given country or subnational region proportionally to the expansion of cropland (net of pasture loss), pasture and forest plantations, capped at total estimated forest loss in the region. Formally, observed forest loss, ${\rm{\Delta }}{F}_{t},$ in a given year (t) is attributed to expanding cropland (${\rm{\Delta }}{F}_{CL,t}$), permanent pastures $\left({\rm{\Delta }}{F}_{PP,t}\right)$ and forest plantations (${\rm{\Delta }}{F}_{FP,t}$) in the following manner:

$$\begin{eqnarray}&&{\rm{\Delta }}{F}_{CL,t}\,=\,{\rm{MIN}}\left[\Space{0ex}{3.5ex}{0ex}\Space{0ex}{3.35ex}{0ex}{\rm{MAX}}\left[{\rm{C}}{\rm{L}}{{\rm{E}}}_{t}-{\rm{G}}{\rm{P}}{{\rm{L}}}_{t};0\right];\right.\left.{\rm{\Delta }}{F}_{t}\cdot \displaystyle \frac{{\rm{MAX}}\left[{\rm{C}}{\rm{L}}{{\rm{E}}}_{Lt}-{\rm{G}}{\rm{P}}{{\rm{L}}}_{t};0\right]}{{\rm{MAX}}\left[{\rm{C}}{\rm{L}}{{\rm{E}}}_{{t}}-{\rm{G}}{\rm{P}}{{\rm{L}}}_{t};0\right]+{\rm{P}}{\rm{P}}{{\rm{E}}}_{{t}}+{\rm{F}}{\rm{P}}{{\rm{E}}}_{t}}\right],\end{eqnarray} \tag{ 1 }$$

$$\begin{eqnarray}&&{\rm{\Delta }}{F}_{PP,t}=\,{\rm{MIN}}\left[{\rm{P}}{\rm{P}}{{\rm{E}}}_{{t}};{\rm{\Delta }}{F}_{t}\cdot \displaystyle \frac{{\rm{P}}{\rm{P}}{{\rm{E}}}_{{t}}}{{\rm{MAX}}\left[{\rm{C}}{\rm{L}}{{\rm{E}}}_{{t}}-{\rm{G}}{\rm{P}}{{\rm{L}}}_{{t}};0\right]+{\rm{P}}{\rm{P}}{{\rm{E}}}_{{t}}+{\rm{F}}{\rm{P}}{{\rm{E}}}_{{t}}}\right],\end{eqnarray} \tag{ 2 }$$

$$\begin{eqnarray}&&{\rm{\Delta }}{F}_{FP,t}=\,{\rm{MIN}}\left[{\rm{F}}{\rm{P}}{{\rm{E}}}_{{t}};{\rm{\Delta }}{F}_{t}\cdot \displaystyle \frac{{\rm{F}}{\rm{P}}{{\rm{E}}}_{{t}}}{{\rm{MAX}}\left[{\rm{C}}{\rm{L}}{{\rm{E}}}_{{t}}-{\rm{G}}{\rm{P}}{{\rm{L}}}_{{t}};0\right]+{\rm{P}}{\rm{P}}{{\rm{E}}}_{{t}}+{\rm{F}}{\rm{P}}{{\rm{E}}}_{{t}}}\right].\end{eqnarray} \tag{ 3 }$$

Here ${\rm{G}}{\rm{P}}{{\rm{L}}}_{{t}}$ denotes gross pasture loss and ${\rm{C}}{\rm{L}}{{\rm{E}}}_{{t}},$ ${\rm{P}}{\rm{P}}{{\rm{E}}}_{{t}},\,{\rm{F}}{\rm{P}}{{\rm{E}}}_{{t}}$ the expansion of cropland, permanent pastures and forest plantations respectively (i.e. these variables are zero where these land classes are shrinking).

Forest loss attributed to cropland expansion (${\rm{\Delta }}{F}_{CL,t}$), is further allocated to individual crops or crop group aggregates (i), in relative proportion to their expansion in area (${\rm{C}}{\rm{G}}{{\rm{E}}}_{i,t}$) (e.g. if areas planted to soybeans account for half of the total cropland expansion in a country, half of the country's cropland deforestation will be attributed to the country's soybean production⁵ ), according to the following:

$$\begin{eqnarray}&&{\rm{\Delta }}{F}_{CL,i,t}={\rm{\Delta }}{F}_{CL,t}\cdot {\rm{C}}{\rm{G}}{{\rm{E}}}_{i,t}/\,\displaystyle \sum _{i}{\rm{C}}{\rm{G}}{{\rm{E}}}_{i,t}.\end{eqnarray} \tag{ 4 }$$

We attributed and tracked deforestation embodied in production and trade individually for the main forest-risk commodities—cattle meat, soybeans, palm oil and forestry products (from tree plantations)—but group other crops into ten major crop categories, following FAO (the Food and Agriculture Organization of the United Nations): cereals (tracking rice separately, as this is a major crop in tropical regions), other oilseeds, pulses, roots and tubers, vegetables, fruits, tree nuts, fiber crops, and other crops (see table S1 which is available online at stacks.iop.org/ERL/14/055003/mmedia for a full list of all crops included in each category). The latter category mainly includes typical tropical cash crops, such as tea, coffee, cacao and spices, as well as sugar cane.

Based on empirical evidence on time lags between forest clearing and establishment of soy in Brazil [23] and oil palm plantations in Southeast Asia [20], we chose to average changes in area of cropland, crop group, permanent pastures and tree plantations over the three years following the forest loss. Moreover, to account for that, while deforestation is a one-time event, the cleared land typically yields agricultural and forestry commodities over many years (for perennial and short rotation crops with a time-lag before the first harvest), the attributed deforestation was amortized over a number of years' production [24]: i.e. the deforestation attributed to a given land use is spread equally over production from that land in T years following the deforestation event, which means that the total amount of deforestation embodied in production of a given commodity in a given year is calculated as the total deforestation attributed to the land use producing that commodity in the T previous years, divided by T (where T is the amortization time). Here, to get a longer time series, enabling us to analyze decadal trends, we use a relatively short amortization period of 5 years (results for 1 and 10 years amortization period are given in table S2 and figure S2, which shows that overall the difference between 5 and 10 year amortization is small).

The forest loss data used was a spatially-explicit dataset over gross tree cover loss (based on remote-sensing data at approximately 30 m resolution) updated from Hansen et al [3]. This provided annual information on tree cover loss for the period 2001–2014. In these data, forests (pre-forest loss) are defined solely on the physical characteristics of the tree cover (i.e. land cover, rather than land use), based on the (year 2000) canopy cover prior to forest loss [3]; here we applied a minimum threshold of 25% canopy cover to define the forests. (Results for 10% and 75% canopy cover shown in the supplementary materials, see figures S3 and S4.) Forest loss is subsequently defined as a complete loss of tree cover at the pixel scale [3].

The physical forest loss definition employed by Hansen et al [3] includes loss of both natural forests and the harvesting of planted forests. Thus, we only present results for a subset of 156 primarily tropical and subtropical countries (figure S1), as this is where agriculture is primarily expanding into native vegetation [17]. We further excluded forest loss within tree plantations in Malaysia and Indonesia, where the Hansen et al [3] data is known to include rotational clearing [25, 26] (further details in supplementary methods).

National level data on cropland and permanent pasture areas, as well as harvested area for the 12 crops/crop groups, in 2000–2014, were taken from FAOSTAT [27]. We assume that pasture expansion into forests is primarily for extensive cattle grazing for meat production, in line with the approach of Opio et al [28]. To estimate gross expansion of cropland and pasture, we add estimates of gross losses of grasslands (assumed to approximate pasture loss) and cropland from remote sensing data [29] to the net changes in area.

Forest plantation area data at the national level were also from FAO [4]. An annual time series was created by interpolation of data available in 5 year intervals (2000, 2005, 2010, and 2015). For forest plantation area we only assess net area changes, as there were no data on gross loss for this land-use class.

Given that the approach for attributing forest loss to agricultural and forestry commodities employed here (equations (1)–(4)) is non-spatial, it cannot differentiate between expansion of land uses directly into forests and expansion that pushes another land use into forests (often termed indirect land-use change). However, we take two steps that help identifying the direct deforestation drivers. First, by using estimates of gross cropland and pasture expansion in equation (1), we aim to capture the (in tropical America) common land use transition of cropland expanding into pastures, indirectly pushing cattle ranchers into forest frontiers, allocating forest loss to expanding pastures where this is occurring. Second, while indirect effects may occur at all spatial scales, the mixing of direct and indirect drivers is likely more prominent the larger the spatial unit at which equations (1)–(4) is evaluated (as spatial information is aggregated and lost). We therefore run the land-balance model at subnational level for Brazil and Indonesia, which together accounted for 40% of tropical forest loss in 2001–2014 [3].

For the subnational analysis of Brazil and Indonesia data on agricultural and forest plantation area were collated at microregion (n = 558) and province (n = 34) level, respectively. For Brazil, agricultural and forest plantation statistics were primarily taken from the Brazilian Institute of Geography and Statistics (IBGE) [30, 31] and the Brazilian Tree Industry (IBA and ABRAF) [32]. For Indonesia, data were taken from the Ministry of Agriculture [33] and the Ministry of Forestry [34].

Our land-balance model does not cover some other land uses that also involve tropical deforestation, such as mining [35], expansion of urban settlements [36], and infrastructure [37]. However, the direct contribution of these land uses to deforestation is in most instances small, their major impact being indirect, through the opening up of forest to colonization or pushing other land uses into forests. We aim to capture such indirect land-use changes in our model, by using assessments of gross expansion of cropland and pastures. More importantly, however, our model does not capture forest clearing for timber without successive establishment of cropland, pastures or tree plantations. While timber extraction can be an important driver of deforestation and forest degradation in some countries, the lack of a clearly delineated land use following forest loss makes this driver difficult to quantify [14]. Our model may also insufficiently capture the role of small-scale and subsistence farming in deforestation, due to non-commercial production being excluded from many countries' official agricultural statistics [38], something we test for in our analysis.

After allocating deforestation to crops, cattle meat and forestry products, we used a physical trade model to assess the country-to-country trade flows of the embodied deforestation (for details on the physical trade model itself, see [8, 39]). The model is based on bilateral trade data as reported to FAOSTAT [27] for ∼400 primary and processed agricultural and forestry products, as well as on production data for 130 crop commodities, 7 primary livestock products and industrial roundwood. Trade flows were considered for most countries in the world (figure S1).

The products are tracked to the country where they are physically consumed as food or in industrial processes. Crop products used as livestock feed are traced through indirect trade in traded animal products. The model also accounts for multi-country supply chains and processing; for example, it includes trade in bread and pasta. Trade of highly-processed products, such as frozen pizzas, is excluded as this data is not available in FAOSTAT. To identify the production-to-consumption trade links, processed products were converted into physical units of primary commodity equivalents. To avoid double counting, we perform this conversion based on the carbon content of the products. This implies that each carbon molecule originating from a given primary product will get the same weight. We cross-checked the resulting conversion factors with available information on typical conversion efficiencies for main agricultural products [40]. The trade data in primary equivalents were integrated with national level production data for the primary commodities. This enabled us to establish clear links between countries of production and countries of consumption, based on the crucial underlying assumption that exports and consumption are met proportionally by imports and domestic production. For a mathematical formulation of the trade model we refer to Kastner et al [39].

The physical units of the trade flows were then converted into embodied cropland (i.e. land use; based on crop yield data from FAOSTAT [27]) and embodied deforestation (based on the model described in section 2.1) based on an assumption of proportionality (i.e. if X% of the production of soy in country A is consumed in country B, then X% of soy cropland and deforestation in country A is embodied in exports to country B).

We categorized all countries into four classes⁶ intended to represent the main stages of a forest transition [5], following Hosonuma et al [42]: countries exhibiting low deforestation rates are classified as pre- or post-transition depending on whether forest cover is high or low (or if net reforestation is occurring); countries with high deforestation rates are classified as early-transition if gross deforestation is increasing and remaining forest cover is not too low, and late-transition otherwise. We tested the decision algorithm for robustness and found that it was mainly sensitive to the threshold for high/low deforestation rate. After cross-checking the classification with literature [7, 43], we decided to use a slightly lower threshold than Hosonuma et al [42] (see figure 2(a) for an exact description of the final classification algorithm). We also manually adjusted the classification for a few post-transition countries that were not classified as such⁷ and excluded countries with less than 5% forest cover.

Zoom In Zoom Out Reset image size

For this classification, data on forest cover (2015) and net forest cover change (2005–2015) were taken from the FAO Global Forest Resources Assessment [4]. Trends in gross forest loss were estimated as annual averages (2005–2014) from a simple OLS regression on updated country-level forest loss data from Hansen et al [3]. The number of countries in each stage, by region, and their main characteristics are summarized in table 1 and the full classification is shown in figure 2(b).

Attributed forest loss was also dominated by a few commodities, with more than 40% of the embodied deforestation associated with expanding pastures for cattle meat production (2.2 Mha yr⁻¹). Other commodities/commodity groups found to be associated with a large share of deforestation were forestry products (0.8 Mha yr⁻¹), palm oil (0.4 Mha yr⁻¹), other cereals (0.4 Mha yr⁻¹) and soybeans (0.4 Mha yr⁻¹), together accounting approximately for another 40% of total embodied deforestation.

However, the commodities associated with forest loss vary greatly between countries and continents (figure 3): in Latin America, cattle meat accounted for more than 60% of the embodied deforestation, whereas in Asia-Pacific palm oil and forestry products each accounted for around a third of the embodied deforestation. In Africa, cattle meat contributed just over a quarter, and the remainder was a diverse mix of other cereals, roots and tubers, pulses and other oilseeds. But there is also large country variation within these broad regional trends: for example, in Brazil, cattle meat dominates (72%) the embodied deforestation, whereas in Argentina and Paraguay, it accounted for just under half, and soybeans accounted for around a third.

Although traditional tropical export crops—such as rubber, sugar, coffee, cacao—overall contributed little (<5%) to deforestation embodied in production, they did make a larger contribution in a handful of countries. An analysis of expansion of different cash crops showed that this was mainly due to rubber in Asian countries (Thailand, Vietnam, Indonesia), coffee and cacao in some African countries (Liberia, Uganda, Congo, Ethiopia, Cameroon, Madagascar), and coffee in some Latin American countries (Honduras, Ecuador, Peru) (table S4).

While deforestation was mainly driven by domestic demand, in total 26% of the embodied deforestation was exported. Again, there were large variations between countries and commodities (figures 4 and S8). Overall, in countries with—on average—larger farm [45] and forest loss patch [44] sizes, deforestation embodied in production was exported to a larger extent (table S3).

Zoom In Zoom Out Reset image size

In total, the share of deforestation attributed to exports was greatest for crops (40%), with some—palm oil, soybeans, tree nuts and other crops—primarily destined for export (63%–77%). Palm oil and soybeans each accounted for just over a fifth of total deforestation embodied in trade. Deforestation for cattle meat, on the other hand, was primarily for domestic demand, with only 11% for exports in total. However, as such a large share of deforestation was attributed to expanding pastures, cattle meat still constituted 18% of the total deforestation embodied in exports.

Deforestation embodied in palm oil, soybeans and other crops were primarily sourced from early-transition countries and a few late- and post-transition countries. Cattle meat and soybeans were particularly exported from some late-transition countries with decreasing, but still high, deforestation rates (mainly Brazil and Argentina, but also a few other Latin American countries). There was also some deforestation attributed to export from a few post-transition countries (embodied in palm oil from Malaysia in particular, and e.g. forestry products from Vietnam) that maintained high gross forest loss rates even though the net forest area is stable or increasing.

For the early-transition countries, a third of embodied deforestation was exported, but most of those exports (0.5 out of 0.6 Mha yr⁻¹) originated from just two countries: Indonesia, with an export share of 48%, and Paraguay, with an export share of 65%. In most other early-transition countries embodied deforestation primarily served domestic demand. For late- and post-transition countries, around a quarter of the total embodied deforestation was exported, though again there was a lot of variation between countries (between 0% and 78% for late-transition countries, and between 0% and 90% for post-transition countries). Amongst the pre-transition countries, only Papua New Guinea (with 24%–50% range between years), exported more than 35% of its embodied deforestation. On the whole, a substantial share of total deforestation embodied in production was exported across the different forest-transition groups and, while there was no significant difference in export shares between the stages, there were large variations between the countries within them, with much of deforestation embodied in exports originating from comparatively few high-deforestation countries where foreign demand was particularly important.

Similarly, although the total share of deforestation attributed to cropland being exported (40%) was twice the share of total cropland area embodied in exports (20%) (figure S6), this difference was mainly driven by a handful of countries: on the one hand, the high share of deforestation attributed to export can largely be explained by Indonesia, Brazil and Argentina, which together account for almost half (44%) of deforestation embodied in crop production, and export between 49% and 76% of this (table S5). On the other hand, the comparatively lower share of cropland attributed to exports is dominated by India, accounting for a quarter of total embodied cropland while only exporting 8% of this (table S6). Thus, there was no general tendency for countries to export a greater share of embodied deforestation than of embodied cropland (figure S10), but our results show that tropical countries with high deforestation rates dominate the exports of embodied deforestation.

Looking at the total consumption-related deforestation, i.e. the imports plus the domestic consumption of embodied deforestation (calculated as the total production in the country minus any exports), the largest consumer countries were primarily those with much domestic deforestation. Brazil was—by far—the largest consumer of embodied deforestation, irrespective if total or per capita consumption were considered (figure S12). The rest of Latin America also had high levels of consumption. In Asia-Pacific, the largest consumer countries were Indonesia (primarily domestic), China (imports only) and India, although from a per-capita perspective only Indonesia lies in the higher end, while China and India have amongst the lowest per-capita consumption of embodied deforestation globally (1 m² capita⁻¹ yr⁻¹). For Africa as a whole, the consumption of embodied deforestation was not negligible: in total, it was on par with that of Latin America excluding Brazil (1 Mha yr⁻¹), and per capita (9 m² capita⁻¹ yr⁻¹) it was only surpassed by the Latin American countries and Indonesia. Russia, the European countries as well as the rest of Asia-Pacific had similar levels of per-capita consumption of embodied deforestation (6 m² capita⁻¹ yr⁻¹), while the US consumed less (2 m² capita⁻¹ yr⁻¹).

While the amount of deforestation embodied in production and trade fluctuated over the 2005–2013 period (figures S5 and S6), it showed a slightly increasing temporal trend. The share of embodied deforestation attributed to exports particularly increased for early-transition countries overall (though with large variations for individual countries), and while the total amounts of embodied deforestation decreased in the second half of the time period, the exported amounts remained fairly constant. Embodied cropland did not see a corresponding increase in exported share, potentially indicating an increasing role of foreign demand for forest-risk commodities relative to other commodities.

The vast majority (79%) of the exported deforestation ends up being consumed in countries that are increasing their forest cover (post-transition countries). Late-transition countries consumed another 8% of the internationally-traded deforestation. The largest export flows went from early- and late-transition countries to post-transition countries (figure 5), and the exported deforestation from pre-transition countries was primarily consumed in post-transition countries in Europe (figure S12(a)).

Zoom In Zoom Out Reset image size

A handful of countries accounted for a large part of the import and consumption of embodied deforestation. Of the top 10 importing countries—accounting for half of the imports—8 were post-transition countries, with much of the imports consumed in Europe and Asia-Pacific (table S7). China, India, Russia and the US were the four individual countries with the most imported deforestation, together accounting for about a third of the total imports.

With post-transition countries consuming most of the deforestation embodied in trade, the gains in forest cover in these countries have been partly offset by deforestation elsewhere (figure 6(a)). In many cases, deforestation embodied in imports actually exceed or is comparable to the net domestic forest-area in the importing countries. For example, between 2010 and 2013 the United Kingdom imported 31 kha yr⁻¹ embodied deforestation, exceeding the 17 kha yr⁻¹ net increase in forests inside the country. For Malaysia, the 22 kha yr⁻¹ imported deforestation also exceeded the 14 kha yr⁻¹ domestic net reforestation (which also conceals substantial gross loss of primary forests still happening in the country). For India the imports of deforestation offset 58% of the 178 kha yr⁻¹ reforestation gains.

Zoom In Zoom Out Reset image size

In total, for all post-transition countries, a third of the net reforestation gains were offset by imports of embodied deforestation from elsewhere. If one looks specifically at the post-transition countries that have pledged to reforest cleared or degraded forest land under the Bonn Challenge, the share of current reforestation offset by imports of embodied deforestation was even slightly higher (36%), and while absolute deforestation embodied in imports was still small relative to committed reforestation targets, so was 2010–2015 net forest gain in these countries (table S8). This suggests that forest transitions in many cases have in part been enabled by not only importing land-demanding products from abroad, but also by displacing some of the deforestation, and concomitant environmental impacts, to other countries.

For the 20 late-transition countries that decreased their gross forest loss 2005–2014, half (10) increased their imports of embodied deforestation (figure 6(b)), while the other half decreased their deforestation imports. Among the 10 countries which were increasing their imports, there were a few countries where imports of embodied deforestation increased more than the reduction in deforestation (e.g. Zimbabwe, for which the trend in gross forest loss was 0.5 kha yr⁻¹ while imports increased by 2.7 kha yr⁻¹), in total offsetting 24% of reduced deforestation. This offset decreases to 11% if the 10 countries that reduced their imports of embodied deforestation are included.