Afforestation to mitigate climate change: impacts on food prices under consideration of albedo effects

Future land-use, carbon sequestration and food price development as affected by afforestation were modelled with the partial equilibrium model MAgPIE (Lotze-Campen et al 2008, Humpenöder et al 2014, 2015, Popp et al 2014). MAgPIE is an agro-economic land-use model that minimises the global costs of agricultural production for a given agricultural demand under a set of economic and biophysical constraints. By this it computes optimal, spatially explicit future land-use patterns in five-year time steps.

Agricultural demand in the model is based on projections of future population and gross domestic product (GDP) of the SSP2 scenario (KC and Lutz 2014, Dellink et al 2015, O'Neill et al 2015). This scenario assumes that global population peaks in 2070 at 9.4 billion people, while per capita GDP continues to increase until 2100. Future demand for calories and livestock share in consumption are derived through a regression model that has been estimated with historical data for calories consumed and GDP development (Bodirsky et al 2015) (see also figures S2 and S3). Feed demand for livestock production results from animal-specific feed baskets (Weindl et al 2010, 2015). Socio-economic parameters, such as the demand, are exogenously fed into the model at the level of ten geo-economic world regions.

The model considers the production of 17 different crop groups and 5 livestock commodities. Bioenergy production was not included in this study. Potential crop yields, carbon densities and water availabilities are derived by the Dynamic Global Vegetation Model LPJmL (Bondeau et al 2007, Fader et al 2010, Waha et al 2012, Müller and Robertson 2014) on a spatial resolution of 0.5°. For the starting year of the model (1995) crop yields were calibrated to match attained country yield levels and regional production areas reported by FAOSTAT. For an efficient, nonlinear modelling under computational constraints, spatial input data were aggregated to 600 clusters with similar crop yields, hydrological conditions and market access (Dietrich et al 2013).

In the model there are several options to respond to future changes in demand or other pressures on the land-use system, such as afforestation. The land-use pattern can react flexibly so that one land-use class can be extended at the expense of others, e.g. cropland can be expanded onto former pasture areas, or afforestation might take place on present-day croplands. The model can also reallocate production to locations that are more productive, domestically within a region or via international trade. Another option implemented is the use of irrigation. Finally, agricultural production can be intensified by endogenous investment decisions in yield-increasing technological change.

Agricultural production and all options to increase production are associated with costs. Factor costs account for costs related to capital, labour and fertilizer use and were derived from the GTAP database (Narayanan and Walmsley 2008). The change from one land-use class to another is subject to regionally differing land conversion costs (Schmitz 2012). Yield increases induced by technological change are endogenous in MAgPIE and are connected to additional investment costs for Research & Development (R&D). These costs were derived through a regression between historical investments and observed yield increases (Dietrich et al 2014). An investment horizon of 30 years and a discount rate of 7% are assumed for all investment decisions. Starting from the present distribution of areas equipped for irrigation (Siebert et al 2007), the model can increase irrigated areas at investment costs for the creation of the infrastructure and costs for operation and maintenance (Bonsch et al 2016). The cost effectiveness of production is also influenced by intraregional transport costs which make production at locations far from markets more expensive.

Food commodities can be traded between the world regions. Two trade pools are implemented in the model. Within the first trade pool, trade flows are fixed to fulfil regional, historically observed self-sufficiency rates calculated from FAOSTAT (2010). For the following time steps, the influence of this first trade pool is reduced depending on the scenario, and food commodities are to a larger share traded according to regional comparative advantages (Schmitz et al 2012) (figure S4).

Afforestation and avoided deforestation are incentivized by a price on CO₂ emissions from the land system. While the CO₂ price renders deforestation and the conversion of pasture to cropland more costly, carbon dioxide removal through afforestation is rewarded and lowers the costs in the objective function of the model. Afforestation is implemented as induced regrowth of natural vegetation. Carbon accumulation in living biomass follows sigmoidal tree growth curves where the upper limit is defined by carbon densities from the LPJmL model. Soil and litter carbon densities are assumed to increase linearly over 20 years, starting from the weighted average carbon density of cropland and pasture (Humpenöder et al 2014, 2015). For this study we assumed a CO₂ price that starts at 30 US$ per tonne of CO₂ in 2020 and increases by 5% each year (similar to Calvin et al 2012 and Kriegler et al 2013).

Afforestation is considered to be most effective in the tropical zone because the combined effect of carbon sequestration and albedo change are assumed to lead to a net cooling, while for the boreal and temperate zones the effect is presumably much lower. To assess the food price effects of afforestation under differing levels of ability to decrease global temperatures, we considered three scenarios where afforestation was limited to certain latitudinal zones. Within these areas the decision to afforest was based on its cost-effectiveness under a CO₂ price on land-use emissions. The effect of albedo was not included directly in the model, but scenarios with different influence on albedo-induced radiative forcing were assessed. In the first scenario afforestation was not restricted at all (unrestricted aff), in the second not allowed in the boreal zone north of 50°N (no boreal aff), and finally it was limited to the tropical zone between 20°S and 20°N (only tropical aff). The definition of tropical and boreal zones thereby follows Bala et al (2007). These afforestation scenarios were compared to a scenario of avoided deforestation, where terrestrial CO₂ emissions were also priced but no afforestation was considered, and to a business as usual (BAU) case without any emissions pricing (see also table 1).

While limiting large-scale afforestation to the tropics seems plausible from a climate mitigation perspective, it could still result in severe food price hikes in tropical regions. Enhanced international trade of agricultural commodities could be one option to buffer these price increases in tropical regions. For the only tropical aff case we therefore assessed how more liberalised trade influenced food prices (only tropical aff tradelib). In this scenario, trade departed more quickly from historical agricultural trade patterns towards more international trade based on comparative advantages. While in our default setting the influence of historical trade patterns decreased by 0.5% per year, in this scenario it was reduced by 1% per year (see also figure S4).

For these scenarios we calculated Laspeyres food price indices that comprise vegetable and livestock products. The Laspeyres formula weights prices according to base year quantities and is also the common approach used, for instance, by The World Bank (2015) to calculate its consumer price index. Food prices derived from MAgPIE reflect the marginal costs of food production (shadow prices), i.e. the costs that would arise for the production of one additional commodity unit. They are formed as a consequence of altered demand and production costs and therefore show the relative long-term commodity price development. Food prices in the BAU scenario are driven by the increasing demand for food from a growing and wealthier population. In the avoided deforestation scenario food prices additionally reflect the pricing of land-use-change emissions, the thus reduced attractiveness to reduce the area of forest or convert pastures to cropland, and the increased need to invest into yield-increasing technology. Food prices in the afforestation scenarios are the result of all these factors and an additional reward on carbon uptake through afforestation which leads to decreasing agricultural areas.

The growing demand for food (figure S2) leads to an expansion of croplands in the BAU scenario. Globally, cropland area increases by 360 million hectares (Mha) until 2100, leading to a reduction of the area of pasture by 275 Mha and of forests and other natural vegetation by about 85 Mha. The introduction of a price on CO₂ emissions from land-use change stops the net conversion of forest to agricultural areas on a global level. In the avoided deforestation scenario, cropland expands by 77 Mha, with most of the change happening in Africa (40 Mha) at the expense of pasture (17 Mha) and forest (13 Mha), while in Europe there is some regrowth of forests (14 Mha).

In the afforestation scenarios the CO₂ price provides an incentive for afforestation so that forest area increases substantially in all regions where this option was given considering the latitudinal restrictions. Under unrestricted afforestation, more than 2500 million ha are newly afforested globally between 2010 and 2100, which is equivalent to an increase of global forest area by more than 60%. The largest areas of afforestation in absolute terms are in Africa (630 Mha) and Latin America (600 Mha), but afforestation is also substantial in all other regions when compared to their total land areas (figure 1). While in most regions afforestation leads mainly to a reduction in pasture, in Europe and Pacific Asia more croplands are converted to grow forests.

Zoom In Zoom Out Reset image size

The restriction to no boreal afforestation reduces the afforested area by about 13% globally, but hardly changes the amount of land conversion in tropical regions. In the only tropical afforestation scenario, in contrast, the area of forest establishment is cut by half (table 1). While it remains at comparable levels in the tropical regions Africa and Pacific Asia it is lower in Latin America (435 Mha), because areas in the south (>20°S) were not considered for afforestation (figure S9).

While in the BAU scenario investments into yield-increasing R&D are rather modest, the introduction of a price on CO₂ emissions prevents further agricultural expansion and necessitates higher yields in the avoided deforestation scenario. In the afforestation scenarios, pasture and cropland area decrease globally, which results in even more substantial yield increases needed to fulfil food demands (table 1). Throughout the afforestation scenarios, the highest rates of yield-increasing technological change are seen in 2020, when the pricing policy on land-use emissions is implemented. These rates are, especially in the tropical regions, substantially higher than those observed in the recent past (Fischer et al 2014). Until the end of the century average annual technological change rates range between 0.44% in BAU and 1.34% in the unrestricted afforestation scenario. Large regional differences are observed, with yields being about 5.5 times as high in Africa at the end of the century in the unrestricted case compared to 2010, but less than double within Europe in the same scenario (see figure S8 for regional yield development).

Afforestation leads to considerable carbon sequestration. While in the BAU case more than 90 Gt of CO₂ are released as a result of land-use change, up to 860 Gt CO₂ are sequestered in the case of unrestricted afforestation between 2010 and 2100. The pricing of CO₂ emissions from land-use change in the avoided deforestation scenario results in no net release of carbon from the land-use system.

Restricting afforestation to non-boreal and tropical regions reduces the area and therefore the amount of carbon sequestered (figure 2(a)). For the no boreal scenario carbon removal is 8%, and the afforested area about 13% lower globally compared to the unrestricted scenario. In the only tropical afforestation scenario, terrestrial carbon uptake is about 40% lower than in the unrestricted scenario, while afforestation area is reduced by about 50%. The stronger reduction of afforestation area relative to CDR is as a result of higher carbon accumulation rates in temperate and tropical forests compared to boreal regions.

Zoom In Zoom Out Reset image size

The increasing food demand from a growing population with an increased per capita demand for meat products does not lead to very significant changes in food prices. In the BAU scenario, without any pricing of emissions from the land-use system, food prices are projected to stay rather constant, or to decrease slightly to about 10% lower than in 2010 (figure 2(b)), caused by a decline in demand towards the end of the century (figure S2). The exponentially increasing CO₂ price on land-use-changes emission in the avoided deforestation scenario prevents the conversion of pasture and forest to cropland. Increasing land scarcity and the necessary investment costs for research and development increase prices at maximum by about 40% on global average in this case.

Afforestation leads to competition for land between carbon sequestration and agricultural production and results in substantial food price increases. Under unrestricted afforestation food prices increase by about 80% up to 2050 and are on average more than four times higher in 2100 than in 2010. Excluding boreal regions from afforestation reduces this effect only by about 9% in 2100. However, when afforestation is limited to the zone of highest cooling effectiveness—the tropics—the food price impact is significantly reduced. In the only tropical afforestation scenario, food prices peak in 2075 having increased by about 100%, followed by a decline in prices due to decreasing demand for food at times of high agricultural yields and a slowdown of forest expansion. Especially in the unrestricted and no boreal scenarios, the additional land-use competition through afforestation influences prices much more strongly than the mere effect of emission pricing in the avoided deforestation scenario. Food prices are also sensitive to the CO₂ price. Lower CO₂ prices lead to lower carbon sequestration, but also reduce food prices (figure S9).

Food prices in different regions are affected differently by the modelled afforestation scenarios (figure 3). Unrestricted afforestation leads to the highest prices of all scenarios over the century within all regions, with the highest values occurring in Pacific Asia (PAS: 630) and Latin America (LAM: 640). In the Former Soviet Union (FSU) the increase is lowest, with prices three times higher in 2100.

Zoom In Zoom Out Reset image size

Excluding the boreal zone from afforestation leads to lower food commodity prices than unrestricted afforestation, especially in regions that are partly in the boreal zone. In Europe (EUR) and FSU estimated food prices in 2100 are then about 30% lower. FSU turns into a net exporter of crops, EUR into a net exporter of livestock products towards the end of the century (figures S6 and S7), which also influences food prices in other regions. In Africa (AFR) and LAM, prices are 7% lower in the no boreal than in the unrestricted scenario in 2100, even though afforested area differs by less than 1% (see also figure 1).

Limiting afforestation to the tropical zone results in a food price index much closer to the BAU scenario, and much lower than for unrestricted and no boreal afforestation, but in tropical regions the price increases are still substantial. In Pacific Asia the food price index is highest in 2100 with a value of 400, while in Latin America the maximum index level of 219 is reached in 2070. The influence on temperate and boreal regions is much lower. In EUR, NAM and FSU the price indices are at maximum increased by 35% to 40% compared to the BAU case. In this scenario, food price increases are in all regions lower than the assumed increase in GDP (figure S5).

More liberalised trade helps to buffer food price increases driven by tropical afforestation. We compared the food prices of the only tropical scenario to a scenario where the deviation from historical trade patterns was twice as fast (figure 4). In this only tropical aff tradelib scenario the overall, interregional trade volume increases faster (see also figures S6 and S7). Latin America turns from an exporter of food commodities into a net importer towards the end of the century. Africa further increases its imports of livestock products, which are mostly supplied by North America. In 2075, the year in which prices are highest globally, food prices are reduced by more than 25% in Latin America and Africa (figure 4). In Pacific Asia, where food prices are highest in 2100, the price index changes from 400 to 219. Subsequent price increases in Europe are negligible. While trade liberalisation has a strong influence on prices, it does not decrease afforested area (1275 Mha) or the sequestered amount of carbon (552 Gt).

Zoom In Zoom Out Reset image size

Our results show that large-scale afforestation can lead to significant carbon sequestration in the land-use sector, but can also lead to strongly rising food prices. In our study, these food price increases were the consequence of a large-scale transformation of the land-use sector, where food has to be produced on a much smaller overall agricultural area. In the scenario of unrestricted afforestation, cropland area is reduced by almost half to a global value of about 800 Mha in 2100, and pasture shrinks by more than 50% to about 1465 Mha, values that were last observed at around the year 1900 (Klein Goldewijk et al 2011). This decline in agricultural areas is enabled by significant investments into yield-increasing technological change and comes along with a pronounced increase in food prices. Avoided deforestation alone does not drastically spike food prices, which is in line with an earlier study by Schneider et al (2011). The finding that afforestation drives up food prices is also the result of a previous study by Calvin et al (2014), in which afforestation was also incentivized by a price on emissions from land-use, and resulted in increasing wheat prices. In contrast to this study, we report a combined food price index for meat and food-crop products for different afforestation scenarios. We also compare food prices under afforestation to a scenario where emissions from land use are priced, which leads to avoided deforestation. This comparison shows that most of the price increase can in fact be attributed to afforestation, while the emissions pricing alone is of lesser importance.

Limiting afforestation to the tropics—where it is most effective in decreasing global temperatures –substantially reduces the impact on food prices. Earlier studies with earth system models showed that afforestation in the tropics, through the combined effect of carbon sequestration and albedo change, leads to a net cooling, while planting trees in the boreal zone might even increase global temperatures (Bala et al 2007, Bathiany et al 2010, Arora and Montenegro 2011). While this simplified, latitudinal dependence seems to hold true in general, exceptions are possible under specific site conditions. Since historical boreal and tropical deforestation took place on the most productive lands with above-average carbon stocks and below-average snow cover, a reforestation of some boreal areas might also decrease global temperatures (Pongratz et al 2011). And an afforestation of tropical and subtropical desert areas could result in net warming because of the prevalence of the albedo effect (Keller et al 2014). Desert areas with high albedo, however, were not considered for afforestation in our study. Rather, afforestation was restricted to agricultural areas in certain latitudinal zones, excluding boreal and temperate zones where afforestation might not show a global cooling effect. Integrating the albedo-induced radiative forcing effect of afforestation directly in the model, as has been done by Jones et al (2015), should be considered for future model applications.

While limiting afforestation to the tropics reduced food prices globally, food price indices remained higher in tropical regions. These increased price levels in the tropics could be buffered by a more liberalised trade policy, with an ensuing shift of agricultural production to non-tropical regions. However, this interregional reallocation would also increase the import dependency of some tropical regions and might hamper the development of the agricultural sector within these regions.

Before afforestation can be considered as a serious means to mitigate climate change, deforestation has to come to an end. In our study this happened as soon as there was a price on CO₂ emissions from deforestation. At the moment, however, no such policy is in place on a global level and much of the carbon stored in tropical forests is released into the atmosphere. Gross carbon emissions from tropical regions were estimated to be around 0.81 GtC yr⁻¹ between 2000 and 2005 (Harris et al 2012), with yearly emissions of deforestation from the Amazon basin alone accounting for 0.18 GtC between 2000 and 2010 (Song et al 2015). The current trend is opposite to what we described in our only tropical afforestation scenario. Between 1993 and 2012 tropical forests lost aboveground biomass carbon (−0.21 GtC yr⁻¹), while boreal and temperate forests gained it by about the same amount (+0.18 GtC yr⁻¹) (Liu et al 2015). However, Brazil—the country with the greatest absolute forest area reduction—has recently reduced its deforestation curve through conservation policies and stricter law enforcement on the ground (Assunção et al 2015, Tollefson 2015, FAO 2015b). China has initiated a large afforestation programme, with plans to increase afforested area by 40 Mha by 2020, a measure which was found not only to sequester carbon but also to decrease local land-surface temperatures (Peng et al 2014). And in December 2015, ten African countries launched AFR100, and initiative to restore 100 Mha of degraded and deforested land by 2030—partly as a climate change mitigation measure (WRI 2015). These developments are just few of many that indicate that global afforestation efforts now have better prospects for success.

Continuous yield increases and substantial investment into yield-increasing R&D would be needed to fulfil the food demands of a growing population, especially when agriculture competes with afforestation. The high price on CO₂ emissions, and hence the strong incentive to free up agricultural land for afforestation, initiates continuous yield-increasing technological change in our study, with values well above those observed historically. In contrast to other partial equilibrium land-use models (e.g. GLOBIOM: Kraxner et al 2013, GCAM: Calvin et al 2014), technological change is endogenously derived within MAgPIE (Dietrich et al 2014, Von Lampe et al 2014), and yields tend to increase stronger in response to additional pressures on the land-use system (Lotze-Campen et al 2014, Nelson et al 2014, Delincé et al 2015). During recent decades, yields of main staple crops increased linearly at average rates of 1% (wheat, rice, soybean) and 1.5% (maize), while the relative annual rate of increase constantly dropped (Fischer et al 2014). Increased investment into R&D would be needed to make afforestation a realistic option, but when research spending increased in recent years this was largely driven by the development in single countries like China and India. Almost every third OECD country actually had a negative trend in public agricultural R&D spending. And in the developing world, especially in Sub-Saharan Africa, where in our afforestation scenarios yields more than tripled between 2010 and 2100, public spending on agricultural R&D amounted to only about 1.6 billion US$ or 5% of global agricultural R&D spending in 2008, and almost half the African countries had a negative trend in their budgets (Beintema et al 2012). This trend of low R&D spending would certainly have to turn around in order to achieve the yields projected in our model.

The yield increases triggered by afforestation could also alter agricultural N₂O and CH₄ emissions, a dynamic that was not in the focus of this study. Intensification could both increase or decrease N₂O emissions from soils, depending on whether intensification is reached through higher inputs (e.g. fertilizer) or better agronomic practices (Bodirsky and Müller 2014, Lassaletta et al 2014). CH₄ emissions from the livestock sector would likely be decreased by intensification due to a more efficient feed conversion (Herrero et al 2013).

The food-price increases presented in this study have to be seen in the context of a general increase in wealth. For this study we assumed the GDP development of the SSP2 scenario (Dellink et al 2015), which is steadily increasing for all model regions, and is also the basis for the increased per capita demand for food products. In most regions the rates of GDP increase are higher or in the same range as the price increases due to afforestation, so that share of expenditure for food would stay constant or decrease for a representative agent (see figure S5). Still, increases in wealth would not necessarily be distributed evenly among the population, so that the change in prices reported here could still have drastic impacts on the poorer parts of society. This is especially true for people whose share of expenditure on food is currently quite high, such as the poorest people in some African and Asian countries who currently expend above 70% of their available income on food (FAO 2015a).

A number of factors influence the formation of food prices, and our study focuses on the more long-term drivers. In the coming decades, a growing global population is expected to increase the demand for food, in particular for livestock products (Alexandratos and Bruinsma 2012, Bodirsky et al 2015). This, together with a likely elevated demand for bioenergy, will increase the total demand for agricultural products. These long-term trends are overlain by a number of more short-term factors affecting prices, such as weather variability, financial speculation or restrictive export policies in response to increasing prices (Mueller et al 2011). Lagi et al (2015), for instance, were able to replicate the FAO food price index between 2004 and 2012 with a dynamic model, where the underlying upward trend was due to an increasing demand for ethanol production, while the short-term peaks were caused by speculation. Our model is designed to capture the medium-term to long-term drivers of food price formation, and reveals the relative difference between afforestation scenarios and a world without forest-based climate mitigation. It does not consider specific policies and drivers on local or short time scale.

Food demand was provided exogenously to the model as a function of per capita income and population. Since price hikes in the afforestation scenarios were quite high with respect to the BAU case, it could be expected that the consumption of agricultural products declines, in spite of relatively low demand-to-price elasticises of food products, especially in high income countries (Hertel 2011, Muhammad et al 2011). Also for this reason, MAgPIE represented the upper range of food price estimates when climate change effects were assessed in a model intercomparison (Nelson et al 2014). However, we also assumed that currently developing regions become relatively wealthy towards the end of the century when food prices are projected to be at the highest level, which would result in lower shares of income expenditure on food and low demand elasticities.

Afforestation at the scale as described in this study would imply macro-economic effects that should be subject to further research, for instance within a general equilibrium framework. The MAgPIE model is a partial equilibrium model of the agricultural sector, impacts of afforestation on other sectors of the economy such as labour, capital and carbon markets were therefore not part of this study. We would expect that increasing food prices also increase the income of net food sellers, and reduce the incomes of net buyers as non-food expenditures are reduced, which could in consequence change the demand for food (Dorward 2012). Afforestation might also create new jobs in the short term for the planting of trees, but these jobs would vanish once the forests are established. Rent-seeking behaviour and opportunities to invest in land under a policy rewarding carbon removal could substantially shift production input factors from other sectors. Furthermore, our analysis of trade was focused on the agricultural sector. For the only tropical afforestation scenario we assessed how trade liberalisation would influence regional food prices. We have, for instance, not considered how the consequential change in trade flows (e.g. increased imports of livestock products to Africa) would have to be compensated by trade flows in other sectors to avoid trade deficits, or how trade liberalisation would affect economies in general. Finally, the creation of an international market for carbon credits could create a substantial flow of money from CO₂-emitting countries to those actively sequestering carbon through afforestation. These revenues could be used to finance, among other things, the import of food.