Can we avert an Amazon tipping point? The economic and environmental costs

The IEEM Platform [45, 46] is an economy-wide decision-making framework for evidence-based public policy and investment design based on a recursive dynamic computable general equilibrium (CGE) model [47–49]. In this study, we linked IEEM with spatial LULC and ESM, IEEM + ESM [35, 36], to understand the regional economic repercussions of moving toward an environmental tipping point in which the Amazon experiences irreversible changes to its ecosystems. IEEM models, databases and instructional materials are openly available on the OPEN IEEM Platform website¹¹.

Among CGE models, the value-added of IEEM is: (a) its integration of detailed environmental information through the United Nations System of Environmental-Economic Accounting [50] to represent the economy and the environment in a comprehensive and inter-connected way, all consistent with the System of National Accounts [51]; (b) the indicators IEEM generates are those demanded by policy makers including GDP, employment and poverty, but also metrics of wealth, sustainability, natural capital stocks and ecosystem services supply; and (c) IEEM's environmental modeling modules capture the specific dynamics of natural capital-based sectors while IEEM's linkage with spatial ESM enables estimation of impacts on the future flow of market and non-market ecosystem services.

The core IEEM database is an environmentally extended Social Accounting Matrix constructed for each of our five Amazon focal countries [52] based on each country's System of National Accounts, supply and use tables and integrated economic accounts. Crops, livestock, forestry and employment data for each country are disaggregated by first level administrative divisions using a combination of agricultural census data, LULC maps, and national household and continuous employment surveys. This is an important step for enabling LULC and ESM that accurately reflects subnational (state and department-level in this case) demand for land and the application of spatially targeted policies. IEEM outputs and subsequent LULC and ESM include 21 states and departments in the five countries found within the Amazon biome, plus the aggregated non-Amazonian parts of each country, for a total of 26 regions (see supplementary information (SI) section 1 for detailed data sources for each country).

We develop a baseline projection for each of the focal countries for the period 2020–2050. The economies are projected for this period based on the International Monetary Fund's World Economic Outlook GDP growth rates [53] and the United Nations Department of Economic and Social Affairs population projections [54]. Deforestation is projected to 2050 based on historical trends [55]. Deforestation increases the availability of cleared land for crop and livestock activities with endogenous allocation of land between these two uses. The extraction of natural assets such as petroleum and minerals occurs at the same rate as GDP growth for each country. Equilibrating mechanisms or model closures for government, savings-investment, and balance of payments are detailed in SI section 1.

We first develop a reference business-as-usual (BASE) scenario for the region from 2020 to 2050 from a projection of economic development and land use that follows current trends. We then introduce the tipping point scenario (COMBI) which simulates the impacts of approaching an Amazon tipping point and is comprised of three main dimensions: (a) climate change impacts from increased water and temperature stress; (b) increasing frequency and intensity of drought, and; (c) increasing frequency and intensity of fire. The difference between the COMBI and BASE scenario reveals the economic, natural capital and ecosystem service impacts of approaching a tipping point.

Upon this future trajectory for the region (COMBI), we implement policy actions aimed at averting a tipping point. This policy intervention scenario (POLICY) begins with concerted regional efforts to strongly reduce deforestation over the next 10 years. There are various strategies for achieving this, including command and control mechanisms such as improved monitoring and enforcement of existing laws, market-based incentives such as the G4 Cattle Agreement (suppliers that have deforested since 2010 are excluded from selling to signatory slaughterhouses) [56] and the Soy Moratorium (soybean trader signatories agreed to not purchase soybean on lands deforested after 2006), as well as policies and incentives for introducing more sustainable productive practices (e.g. intensive cattle production systems based on mixed grass-legume pastures [57], integrated crop-livestock systems [58], establishing silvopastoral and agroforestry systems in already degraded areas [30, 31] and sustainable forest management systems) and payment for ecosystem services [1]. Experience in Brazil over the last 15 years has generated numerous lessons that can be applied in the region to reduce deforestation [59–61]. In addition to reducing deforestation, the policy intervention scenario includes investment in agricultural intensification, climate-adapted agricultural systems and improved fire management.

To facilitate presentation and discussion of results, we treat the tipping point scenario and the policy intervention scenario as two single scenarios. In reality, they are comprised of a number of sub-scenario components, each introducing different dimensions that drive towards an environmental fallout in the tipping point case, and different positive policy outcomes in the policy intervention case. Specifically, the tipping point scenario (COMBI) is comprised of continued high rates of deforestation, the ongoing average climate change effect on agricultural yields (YIELD component) and increasing frequency and intensity of drought (DRGHT component) and fire (FIR component).

Underpinning the YIELD component of COMBI, evidence shows that tropical South America has become warmer and drier, and these trends are expected to continue. To model the relationship between climate change and crop productivity, we use historical crop yield data combined with observed climate data on monthly temperature and precipitation from the 30 year period from 1981 to 2010, and future climate projections from 2020 to 2050. The DRGHT component simulates the expectation that extreme drought events will become more frequent and intense with climate change. To quantify the increasing likelihood of extreme droughts and their potential effect on agricultural output in a drought year, we use the Standardized Precipitation-Evapotranspiration Index from 1981 to 2050 for each state/department of our focal countries. The FIR component captures how climate change is expected to result in increased frequency and intensity of fire. Combined with climate change, fire will be a key factor determining what kind of forests exist in the future and whether those forests can sustain critical ecosystem services over the long term [23]. We describe the literature and assumptions underpinning each of these sub-scenario components in the accompanying SI section 2.

We then introduce a set of sub-scenario components that jointly describe a policy intervention scenario (POLICY) designed to avert an Amazon tipping point, comprised of net zero deforestation (NDEFOR), intensified agricultural systems on cleared lands and climate-adapted agriculture (YIELD50), and improved fire management (NFIR). We simulate net zero deforestation as being achieved through intense efforts to improve monitoring and enforcement of deforestation across the region and provide alternative opportunities to support forest-based livelihoods. The annual cost of reaching net zero deforestation is estimated at US$5.5 billion [62, 63]. We distribute this cost across countries based on their individual share of the Amazon biome (67% in Brazil, 15% in Peru, 9% in Bolivia, 8% in Colombia and 1% in Ecuador).

To mitigate the climate change impact on agricultural yields and reduced land supply arising from net zero deforestation, the policy intervention scenario includes investment in agricultural intensification, targeted irrigation and expansion of climate-adapted crop varieties. These measures could mitigate up to 50% of projected productivity losses relative to the yield component of the tipping point scenario in average years, though crops remain vulnerable to the extreme droughts simulated in the drought component of that same scenario. The fire management component of the policy intervention scenario represents comprehensive efforts to control and prevent fires in the face of drier, more fire-prone climatic and biophysical conditions. We assume that effective implementation of fire management could prevent 50%–75% of the projected burned area from burning[21].

To calibrate scenarios in IEEM, beginning with the yield shock, we develop statistical models to relate historic variability in crop productivity to existing spatial gradients in climate and periodic shocks arising from droughts. To quantify regional changes in temperature and dryness, we apply the climate space concept, which uses rainfall and the Maximum Cumulative Water Deficit, an index of dryness, to delineate the climate envelope, which in this case is the range of precipitation and temperature that is most suitable for agricultural production [12, 64]. We use a generalized linear model to relate historic yields to the Maximum Cumulative Water Deficit. Based on these results, we make hypothetical yield projections under mean future climate conditions. This allows us to quantify the upper and lower bounds of expected crop losses as a percentage of baseline yield for each administrative unit, without explicitly modeling the sensitivity of specific crops.

Using a similar approach to that described above, we calculate the Standardized Precipitation Evapotranspiration Index, a standard metric used to monitor droughts [65]. We estimate the historic and future frequency of extreme events based on a threshold of one standard deviation of the Index. We use results as the basis of our YIELD scenario shocks implemented in IEEM. To simulate drought impacts on agriculture in the DRGHT scenario, and in addition to yield losses due to changes in the average climatic conditions and the Maximum Cumulative Water Deficit, we assume a 15% loss in yield during extreme drought events, which are expected to become more frequent in the future [66].

Underpinning the FIR scenario, we use model outputs to quantify the interactions between deforestation, climate change and fire in the driest region of the southern Amazon, which is comprised of about 192 million hectares and is where most fires occur [21]. To account for the effect of climate change on dry season length, we use data from the Amazon Climate Source, a data platform that synthesizes information on observed historical climate [67] and model future climate [68] in the Amazon [69]. We use dry-season length as a spatial mask in a Geographic Information System to distinguish between flammable and non-flammable forests, making the assumption that Amazon forest areas with a dry season length of greater than or equal to 3 months are eligible to burn.

For all scenarios, we assume that future climate will follow the high-emissions scenario outlined by the Intergovernmental Panel on Climate Change [70] which most closely matches current emissions pathways and thus the status quo which is consistent with the tipping point scenario. The high-emissions scenario, referred to as a Representative Concentration Pathway, assumes that atmospheric greenhouse gases will accumulate at the current rate of emissions, causing a climate forcing of 8.5 Watts per square meter.

Annual demands for forest, forest plantation, cropland and grazing areas are generated by IEEM. IEEM demand for land is spatially allocated across a 1 km grid with the LULC change model, which is used to generate baseline and scenario-based projected LULC maps. These maps and future precipitation, temperature, and potential evapotranspiration projections are the main inputs used to model changes in ecosystem services, with most other inputs to those models held constant. We use the Dyna-CLUE [71–74] LULC change modeling framework to spatially allocate LULC change using empirically quantified relationships between land use and location factors, in combination with the dynamic modeling of competition between land use types. Dyna-CLUE calculates suitability maps for each land use type that reflect the probability of each land use class occurring for each grid cell and assigns demands accordingly (see SI section 3 for additional details).

We use four of the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) models [75] to calculate scenario-based, spatially explicit changes in ecosystem services supply, namely: the sediment delivery ratio model to calculate soil erosion mitigation services; the carbon storage model to calculate carbon storage and carbon sequestration potential; the annual water yield model to calculate water regulating services, and; the nutrient delivery ratio model, which is used as a proxy for the water purification potential of landscapes in absorbing nitrogen and phosphorus (see SI section 4 for additional details). We parametrize these four InVEST models with the IEEM + ESM datapackets[76] which render them virtually "plug and play".

Land use and land cover classes subject to change in our modeling are forests, forest plantations, shrubs and other vegetation, cropland, and grazing and pasture areas. LULC change was modeled individually for the Amazon region's 26 different states or departments, depending on the administrative division of each of the countries included (figure 1). Throughout the entire Amazon region, we identified areas subject to change using a 1 km grid.

Zoom In Zoom Out Reset image size

With a detailed understanding of LULC change in the landscape, we assess changes in ecosystem services first as the tipping point impact with respect to a hypothetical future without a tipping point (COMBI vs. hypothetical BASE) and then the policy impact with respect to the tipping point (POLICY vs. COMBI; figure 2). Moving toward a tipping point would result in large losses (>25%) of carbon storage (COMBI, figure 3) in some of Brazil's northeastern states including Alagoas, Pernambuco, Paraíba, Rio Grande do Norte, and Ceará, while Piauí would experience a reduction of 5%–25% in carbon storage. Rio Grande do Sul, Santa Catarina and Paraná would experience a loss of more than 25%. The Colombian and Ecuadorian Amazon would experience a loss of up to 5% along with the Bolivian Amazon, and the Brazilian states of Amazonas, Acre and Roraima. In contrast, policy interventions to avert a tipping point would increase carbon storage (POLICY, figure 3) by more than 25% across most of the region.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

With regard to water regulating services, the tipping point impact (COMBI, figure 3) on these services would be largely muted across the region. In the case of policy intervention (POLICY, figure 3), water regulating services would improve with more water used for ecological functions such as evapotranspiration. This improvement would result in less runoff, thus reducing the impacts of floods, while maintaining better water quality and more water for dry season flows and other important biological and ecosystem functions. Compared to the tipping point, improvements to water regulation through policy intervention would be between 1% and 5% across much of the region.

A tipping point would result in reductions in water quality (COMBI, figure 3), proxied for by nitrogen exports to streams, and would be most acute in Bolivia's south and in Brazil's southeastern states of Minas Gerais, São Paulo and Paraná with a reduction of more than 25% driven by an increase in nitrogen exported to water bodies. A 5%–25% reduction in water quality would be found in the case of Peru's Andes and coast and the Brazilian states of Roraima, Amapá, Pará, Maranhão, Pará, Mato Grosso and Mato Grosso do Sul. Policy intervention would have a strong impact on improving water quality (POLICY, figure 3), equal to or greater than 50% for much of the region. In the case of Minas Gerais, there would be an increase of over 25% and in Paraná, there would be a reduction in water quality of 5%–25% due to increases in nitrogen exported to water bodies.

A tipping point scenario would have spatially diverse impacts on erosion mitigation services (COMBI, figure 3) across the region, with a particularly acute loss across Bolivia (more than 25% loss), along the western slope of the Andes in Peru, and in the Colombian Amazon (5%–25% loss). Paraná in Brazil would experience a more than 25% loss of erosion mitigation services, followed by Minas Gerais (5%–25%), Pará, Roraima, Rondônia, and Acre (up to 5% loss of services). Colombia's non-Amazon region would experience a loss of up to 5%. Policies to avert a tipping point would have a dramatic impact on improving erosion mitigation services (POLICY, figure 3) by more than 50% across much of the region. Bolivia's Amazon region is an exception where a 25% loss would be experienced.

The cumulative regional GDP impact of approaching an Amazon tipping point would be an economic loss of over US$256.6 billion (2019 base year; table 1). In Brazil, the loss would be on the order of US$184.1 billion, followed by US$35.3 billion, US$17.6 billion, US$11.4 billion, and US$8.2 billion in Colombia, Bolivia, Ecuador, and Peru, respectively. In terms of wealth, the cumulative regional impact would be a loss of US$81.55 billion; on a national basis, wealth forgone in Brazil would be US$55.2 billion, followed by US$16.9 billion, US$5.4 billion, US$3.98 billion, US$2.3 billion and US$1.7 billion in Colombia, Bolivia, Ecuador, and Peru, respectively.

Through the lens of GDP, policies to avert a tipping point, including net zero deforestation, would mitigate some but not all of these economic impacts. For the entire region, the net effect would be a US$148.9 billion reduction in GDP. Economic losses would be reduced by more than half in the case of Brazil with a US$81.2 billion loss to GDP, a US$31.2 billion loss in Colombia, a US$16.8 billion loss in Ecuador, US$14.6 billion in Bolivia, and US$5.1 billion in Peru.

Strategies to avert a tipping point, including net zero deforestation, would not only mitigate negative economic impacts, but would generate positive gains for inter-generational wealth; the impact of a net zero deforestation policy would be particularly strong. For the region, policy intervention would generate an additional US$339.3 billion in cumulative wealth. For Brazil, this would add US$162.8 billion in wealth, followed by Colombia, Ecuador, Peru, and Bolivia, with additional wealth of US$98.2 billion, US$47.3 billion, US$20.95 billion, and US$10.1 billion, respectively.

The GDP trajectory by country, as a difference from business-as-usual (BASE), is strongly influenced by the climate change projections and derived indicators used to estimate agricultural productivity and drought and fire-related shocks throughout the analytical period (figure 4). The variability of these factors resulted in irregular impacts through time where some years would register markedly worse impacts than others due to year-to-year variation in the severity of drought and fire. For example, in the case of Bolivia, policies to avert a tipping point, when considered from the perspective of GDP, would appear to generate economic benefits as measured by GDP in the long-term near the final years of the analytical period.

Zoom In Zoom Out Reset image size

In terms of the tipping point scenario's effects on GDP, Brazil, Colombia and Peru's economic trajectory would follow a relatively steep decline throughout the analytical period (figure 4). With regard to the tipping point impact on wealth, Peru would suffer a rather precipitous drop in wealth at the end of the period due to a fire that destroys a large tract of forest; this occurs due to the probabilistic nature of how we model fire (see SI 2). This important loss of natural capital stocks would go undetected from the perspective of GDP.

In contrast, in the policy intervention scenario, wealth would trend upward for most countries (figure 4). Bolivia, however, would experience a boost to wealth in the initial years with a tendency toward the baseline scenario levels of wealth in later years. The smoother trend in the wealth impacts of the policy scenario compared with the tipping point scenario is explained by the gradual implementation of investments to improve agricultural yields.

A tipping point would push at least 171 000 more people into poverty across the region. While the policy intervention scenario would mitigate some of the increase in poverty, it would still result in 159 600 more poor people; this number of poor represents approximately 0.05% of the current combined population of the five focal countries.