Study Declares Agriculture In Deforested Amazon Will Fail

Another driver of environmental changes in the Amazon is the change in atmospheric composition, which may cause changes in the global climate. Most global climate models predict that greenhouse gas accumulation and associated increases in the radiative forcing of the atmosphere will cause a substantial (more than 20%) decline in rainfall in eastern Amazonia by the end of the century, with the steepest decline occurring during the dry season (Malhi et al 2008).

In addition to the radiative effect of CO₂ as a greenhouse gas, atmospheric CO₂ has a physiological effect on vegetation canopy processes; higher partial pressure of CO₂ in the atmosphere often stimulates canopy photosynthesis and decreases stomatal conductance, increasing the water-use efficiency of plants, in particular of C3 plants (Sellers et al 1996).

Here, we focus on the three major services provided by the Amazon ecosystems: climate regulation, carbon storage, and agriculture production. Our study evaluates how local climate patterns are modified under different deforestation scenarios, and the role of radiative and physiological effects of CO₂ on these ecosystem services. In doing so, we aim to assess the resilience of the primary forests and productivities of soybean and pastures in the Amazon under scenarios of deforestation and increased CO₂ concentration.

We evaluate the carbon storage of the primary forests and the productivity of soybean and pasture in the Amazon under several scenarios of regional deforestation and increased CO₂ using a simplified model that represents the interactions between climate and land use. We analyze four different scenarios for 2050, considering: (1) radiative effects of increased CO₂, (2) radiative and physiological effects of increased CO₂, (3) effect of changes in land use on the regional climate and (4) radiative and physiological effects of increased CO₂ plus the effect of changes in land use on climate. In all cases, the 2050 climate is the average of the period 2041–2060.

The primary forest, soybean and pasture productivity models were implemented using Dinamica EGO, an environmental modeling platform for the design of analytical and space–time models (Soares-Filho et al 2013)). Figure 2 shows the basic structure of the model developed.

Primary forest productivity is simulated using the CARLUC model (carbon and land-use change) designed by Hirsch et al (2004). During each monthly time step, the model assumes that wood, leaf, and root carbon pools increase by an overall amount equal to the Net Primary Productivity (NPP), given by:

This formulation is based on the 3-PG model by Landsberg and Waring (1997). NPP is driven by photosynthetically active radiation (PAR, moles of photons m⁻² month⁻¹), and modified by four dimensionless functions representing vapor pressure deficit (f_VPD, 0–1); temperature (f_Temp, 0–1); soil water (f_SW, 0–1); and fraction of absorbed photosynthetically active radiation (f_APAR, 0–1) (Hirsch et al 2004). The carbon-use efficiency (cue, ratio of NPP to Gross Primary Productivity) and quantum efficiency (qe, mol-C mol-PAR⁻¹) parameters convert photons to net carbon stored (Hirsch et al 2004).

Soybean daily dry mass (DM) production is determined by the intensity of radiation and average temperature according to Costa et al (2009). Carbon assimilation is simulated using the concept of light-use efficiency (Monteith 1977). The physiological process is based on two specific parameters: thermal time to flowering and to seed maturation (Costa et al 2009). Total assimilation is allocated to different plant parts, depending on the stage of development (Costa et al 2009). Yield is estimated based on the percentage of dry matter allocated to reproductive organs as a function of growth stage (Costa et al 2009). The simulation is completed when the crop reaches physiological maturity (Costa et al 2009). The model that describes the dynamics of soybean daily dry matter accumulation is as follows:

Pasture dry mass accumulation is calculated as a dynamic system consisting of live (green) and dead tissues according to McCall and Bishop-Hurley (2003). Live tissue enters the system as a result of photosynthesis (McCall and Bishop-Hurley 2003). If not consumed, live tissue eventually senesces and flows into the dead pool (McCall and Bishop-Hurley 2003):

Senescence is proportional to the amount of live green mass (DM). The base senescence rate (σ_t) varies seasonally, assuming greater values in the post-reproductive period of grasses (McCall and Bishop-Hurley 2003). Senescence rate is also determined as a function of the available water content (f_SE) (McCall and Bishop-Hurley 2003). At low levels of available soil water, senescence increases above base levels (McCall and Bishop-Hurley 2003).

In all three models, temperature affects net carbon assimilation penalizing it when it is outside the range of optimum temperature. Optimum temperature range for primary forest is from 25 to 29 ° C, for pasture is from 30 to 35 ° C and for soybeans is from 28 to 32 ° C.

Validation of the productivity models is presented in the online supplementary material (available at stacks.iop.org/ERL/8/024021/mmedia).

To evaluate the productivity of primary forests, soybeans and pastures, we conduct five sets of simulations that represent the present climate and climate change due to changes in atmospheric composition and Amazon deforestation, as follows:

(a)	Control run: to estimate the current productivity of agricultural crops and primary forests, we used the climate database developed by Sheffield et al (2006) for the period between 1971 and 2000. This database is constructed by combining a suite of global observation-based datasets, disaggregated to 3-hourly time intervals using the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis. The variables used are precipitation, air temperature, downward shortwave radiation, surface pressure and specific humidity. A comparison of the simulated productivity values against the observations is presented in the online supplementary material (available at stacks.iop.org/ERL/8/024021/mmedia).
(b)	Radiative effects of CO2: these simulations consider only climate predictions for the IPCC A2 scenario for the 2041–2060 period. This scenario, published in 2000 and initially considered pessimistic, has become the most realistic CO2 scenario for the period 2001–2010 (Van der Werf et al 2009). As the IPCC AR4 report shows, there is much less climate difference for the period 2020–2050 between emissions scenarios than between climate models for the same scenario. To avoid individual model biases, we used the climate anomalies simulated by seven AR4 IPCC models and added these to the climatology used in the control run. The seven models employed are (1) the NCAR CCSM3 (National Center for Atmospheric Research, USA); (2) CNRM CM3 (Centre National de Recherchés Météorologiques, France); (3) GISS ER (NASA/Goddard Institute for Space Studies, USA); (4) INM CM3.0 (Institute for Numerical Mathematics, Russia); (5) IPSL CM4 (Institute Pierre Simon Laplace, France); (6) MRI CGCM2.3.2 (Meteorological Research Institute, Japan) and (7) MIROC3.2 (Center for Climate System Research, National Institute for Environmental Studies, and Frontier Research Center for Global Change, Japan). The average of the climate anomalies from these seven climate models is likely to be more representative than the climate anomaly of any individual model.
(c)	Radiative and physiological effects of CO2: in addition to future climate conditions as described in (b), this set of simulations also considers the physiological effect of elevated CO2 concentration on carbon assimilation by primary forests and agricultural crops. For primary forests, Lloyd and Farquhar (2008) found that, for a 170 ppm increase in CO2 concentration, there was a 30% increase in the assimilation of carbon by tropical forests. For simplicity, we assumed that the response is linear (0.18% ppm−1). For crops, Tubiello et al (2000) found that, for an increase of 350 ppm in the CO2 concentration, there was a crop yield increase of 25% in C3 crops, and 10% in C4 crops. Again, assuming that this increase is linear, we used 0.0714% ppm−1 for soybean (C3 crop) and 0.029% ppm−1 for the C4 pastures that dominate in Amazonia. For the A2 scenario, the IPCC (2007) predicts 559 ppm for 2050. For the control simulation, we use 380 ppm.
(d)	Effect of land use on the regional climate: we considered three land-use scenarios. (i) First, the control scenario, which is based on the 2002 deforestation map (figure 3(a)), and is just slightly different from the end-of-deforestation (EOD) land-use scenario from (Nepstad et al 2009). The end-of-deforestation scenario is plausible given the reversal of Amazon deforestation trend that occurred after 2004 (an accumulated decline by 2011 of 68% from the historical 1996–2005 baseline of 19 600 km2 per year). However, there is significant pressure to expand agricultural production in Brazil to meet domestic and global demands. Brazil’s powerful agricultural sector hopes to double agricultural and livestock output by 2020. The Brazilian government’s Growth Acceleration Plan, for example, is a heavily capitalized, inter-ministerial program that has few environmental safeguards and will increase the profitability of deforestation-dependent activities by lowering the costs of transportation, storage, and energy (Nepstad et al 2011). Thus the profitability of deforestation is rising, and could remain high for many years or decades given the global outlook for continued growth in agricultural commodity prices (Grantham 2011). As a result, high rates of return to agriculture will put more pressure on the Brazilian government to soften environmental laws, such as the recent revision of the Brazilian Forest Code. In light of these events, a reversal of the trend toward decreasing deforestation in Brazil appears plausible (Soares-Filho et al 2012). To include these opposing trends, we included two other deforestation scenarios, the business as usual and the governance by 2050 from Soares-Filho et al (2006) (figure 3), described below. (ii) The business-as-usual scenario for 2050 (BAU) assumes that: (1) recent deforestation trends will continue; (2) highways currently scheduled for paving will be paved; (3) compliance with legislation requiring forest reserves on private land will remain low; and (4) new protected areas (PAs) will not be created or not enforced. The BAU scenario assumes that as much as 40% of the forests inside of PAs are subject to deforestation, climbing to 85% outside. (iii) The governance scenario for 2050 (GOV), assumes that Brazilian environmental legislation is implemented across the Amazon basin through the refinement and multiplication of current experiments in frontier governance. These experiments include enforcement of mandatory forest reserves on private properties through a satellite-based licensing system, agro-ecological zoning of land use, and the expansion of the PA network (Amazon Region Protected Areas Program), which has already occurred (Soares-Filho et al 2010). Their final product includes annual maps of simulated future deforestation under user-defined scenarios of highway paving, protected area networks, protected area effectiveness, deforestation rates and legal deforestation constraints. These three land-use scenarios, regardless of their likelihood, cover a wide range of deforestation extents for 2050, thus allowing us to assess the effects of basinwide land-use changes on climate and the modeled ecosystem services. For analyzing the effects from climate feedbacks only, we then assume that, in the BAU and GOV scenarios, all deforested cells are either occupied by soybean crops or by pasture, totaling then five land-use scenarios. To convert land-use change to anomalies in climate, we use the semi-empirical climate model of Zeng and Neelin (1999), who demonstrate that the anomaly in precipitation (P’, in mm d−1) after deforestation is proportional to the anomaly in the reflected surface radiation (Sr’, in W m−2), or the incoming surface radiation multiplied by the anomaly in albedo (α’). Yanagi (2006) calculated empirical coefficients for the Zeng and Neelin model for trimester time scales (equations (4)–(7)): The surface albedo in each land-use scenario is calculated as a weighted average of the different types of land cover (13% for the forest, and 11% for bare land). For pastures and soybeans, albedo depends on LAI, reaching a maximum of 20% for pastures and 26% for soybeans (Costa et al 2007). Finally, we use equations (1)–(3) to calculate productivity of primary forests and agriculture and compare simulations outputs for the year 2050 with those of the control run. We also perform simulations with the land-use scenarios but without the climate model, i.e., climate feedbacks are not included.
(e)	Radiative and physiological effects of CO2 plus the effects of changes in land use: to evaluate the combined effect of all factors on the primary forests, soy and pastures, our model adds projections of climate change calculated by different IPCC AR4 models and the physiological effect of CO2 (item c) to the climate change induced by land-use change (item d). The simulated yields for the 2041–2060 period are compared to those of the control run.

To assess the response of primary forests and agricultural systems, we compared simulated productivity of the primary forests, crops and pastures under scenarios of climate and deforestation to those simulated under current conditions. The statistical significance of differences was evaluated using the test t of Student. When the output mean from the modeled scenario was not different from the control (current climate) at 5% level of significance, the system is considered resilient.

4.1. Resilience of carbon storage

Three important results from our analysis stand out:

First, in nearly every scenario considered, both carbon storage and agriculture yield decrease in Amazonia in the first half of the 21st century. Loss of carbon storage from the primary forests is a consequence of a relatively drier and warmer Amazon climate in the first half of the 21st century, although this effect is partially compensated by the physiological effects of rising CO₂. Moreover, expansion of agriculture land (scenarios GOV and BAU) introduces climate feedbacks that reduce rainfall in all the seasons, thereby affecting the yield in all land uses. Considering all the effects, individually and combined, carbon storage decreases in every modeled scenario, while pasture and soybean yields are resilient only in scenarios in which there is no expansion of agriculture land and subsequent climate feedback.

Second, our results indicate that trading carbon storage in Amazon ecosystems for alleged increases in agriculture output could be a no-win scenario (lose–lose situation). Carbon storage in 2050 decreases from 145 Mg-C ha⁻¹ (IPCC A2 + CO₂P) to 105–112 Mg-C ha⁻¹ under the governance scenarios (a reduction of 23–27%), and 64–74 Mg-C ha⁻¹ in the business-as-usual scenarios (a reduction of 49–56%). Similarly, pasture productivity in 2050 decreases from 16.3 Mg-dry mass ha⁻¹ yr⁻¹ (end-of-deforestation scenario), to 11.4 Mg-dry mass ha⁻¹ yr⁻¹ in the GOV scenario (a reduction of 30%), and to 10.8 Mg-dry mass ha⁻¹ yr⁻¹ in the BAU scenario (a reduction of 34%), a consequence of the climate feedbacks. Similarly, soybean yield in 2050 decreases from 2.5 Mg-grains ha⁻¹ (no additional deforestation scenario), to 1.9 Mg-grains ha⁻¹ in the GOV scenario (a reduction of 24%), and to 1.8 Mg-grains ha⁻¹ in the BAU scenario (a reduction of 28%).

In sum, an increase in agriculture area of 47% (2.201/1.496, these values refers to deforested area in GOV2050 scenario and in the EOD scenario, respectively, see figure 3 and table 2) under the 2050 GOV scenario is offset by a decrease of 24–30% in agriculture yield, resulting in a net change of pasture output of only 3% (1.47⋅0.70 = 1.03, the value 1.47 refers to the 47% expansion of the deforested area and the 0.70 refers to 30% decrease in agricultural productivity). If the entire new deforested land were occupied by soybeans (a somewhat unrealistic scenario) the net change of soybean output would be 12% (1.47⋅0.76 = 1.12, the value 1.47 refers to the 47% expansion of the deforested area and the 0.76 refers to 24% decrease in agricultural productivity) in the GOV scenario. This would be the best situation in terms of total soybean output for the region. In a more realistic soybean expansion scenario, in which soybean area expands by 10% (Gouvello et al 2010), and the remaining deforested area is occupied by pasturelands, total soybean output would decrease by 16% (1.10⋅0.76 = 0.84, the value 1.17 refers to the 10% expansion of the deforested area and the 0.76 refers to 24% decrease in agricultural productivity) under the GOV scenario, again a consequence of regional climate change due to deforestation.

In the 2050 BAU scenario, an increase in agriculture area of 142% (figure 3) is offset by a decrease of 33–34% in agriculture yield (table 3), leaving a net change of pasture output of 60% (2.42⋅0.66 = 1.60, the value 2.42 refers to the 142% expansion of the deforested area and the 0.66 refers to 34% decrease in agricultural productivity). In the more realistic soybean expansion scenario (10% soybean expansion and the remaining deforested area occupied by pasturelands), total soybean output would decrease by 26% (1.10⋅0.67 = 0.74, the value 1.10 refers to the 10% expansion of the deforested area and the 0.67 refers to 33% decrease in agricultural productivity).

To summarize our second point, agriculture expansion in Amazonia may impose a large loss of several ecosystem services. One of the services quantified here, carbon storage, would decrease from 23% to 56% depending on the land-use scenario. Other losses in ecosystem services, although not quantified here, will most likely ensue, including biodiversity loss and spread of infectious diseases (Foley et al 2007). Yet the expected increase in agriculture output that would compensate (and justify) such losses may not occur. In our simulations, pasture production may increase by only 3%, and soybean production may decrease by 16% in the governance land-use scenario. In BAU scenario (142% increase in Amazon agriculture area), pasture output may increase by 60%, and soybean output may actually decrease by 26%. In all cases, a no-win situation is realized: the loss of ecosystem services is associated with a loss in soybean productivity and total Amazon soybean production. In turn, pasture total output would be much lower than that expected from pasture expansion. This unexpected no-win scenario arises as a consequence of the climate feedbacks introduced from changes in land use, which were usually ignored in previous studies.

This leads to our third conclusion: large-scale expansion of agriculture in Amazonia may be self-defeating. This is particularly worrisome for eastern Pará and northern Maranhão, where local precipitation appears to depend strongly on forests, and changes in land cover would drastically affect the local climate, maybe, to a point that agriculture becomes unviable.

Our results are subject to a number of caveats. First, we used simple linear models that, although representing the most relevant processes involved, may miss second order processes or feedbacks. Nevertheless, our climate feedback results are comparable, both in the direction and magnitude of the response, to the results of more sophisticated global climate models used by Sampaio et al (2007) and Costa and Pires (2010), who investigated the effects of the same land-cover change scenarios on the climate of Amazonia.

Second, model bias can never be eliminated, regardless of model complexity. We can only overcome this limitation by using multiple model ensembles, which lies beyond the scope of this study. On the other hand, we avoided single model bias in the future climate scenarios by using future climate results from seven different IPCC models.

Third, uncertainties in the future scenarios cannot be eliminated, either in predicted climate (IPCC A2) or land-use trajectories. A steep decline in deforestation rates in the Brazilian Amazon in recent years demonstrates that land-use trajectories can change drastically in a short period of time (Soares-Filho et al 2012). However, future efforts to reduce deforestation will need to address the increasing global demand for food production (especially protein from cattle and soy) that will build up pressure to expand the agriculture frontier, especially in southern and eastern Amazon (Lapola et al 2011).

Fourth, the scenarios used here do not consider possible agriculture land abandonment, and subsequent forest regrowth, and thus these effects are not included in the carbon storage calculations. If these effects were included, the results would be intermediate between the control and each deforested scenario.

As a final word, large-scale agriculture expansion in Amazonia may introduce climate feedbacks that would reduce precipitation, leading agriculture expansion in Amazonia to become self-defeating: the results of this study suggest that the more agriculture expands, the less productive it becomes. This would be a no-win situation, in which we all lose. Therefore, agriculture expansion in Brazil should prioritize land already converted and must be accompanied with comprehensive forest conservation to counteract externalities of this process.