We illustrate recent trends in economic teleconnections with the case of corn and sugar cane ethanol in the USA and Brazil, respectively (figure 1). With the US plan to supply 15% of its gasoline consumption through alternative fuels, including ethanol, by the year 2017 (US White House 2007), its subsidization of ethanol made from corn ($0.51 per gallon) and its ongoing tariff on imported Brazilian ethanol ($0.55 per gallon), corn prices and production are expanding (Severinghaus 2005; USDA 2006; Valle 2007; Wald & Barrionuevo 2007). Corn's expansion has displaced US soya bean production areas, pushing soya prices up by reducing the global supply of soya. Meanwhile, the expansion of sugar cane production in southern Brazil has had a similar effect, displacing soya production areas and further increasing soya prices (Valle 2007). Demands for soya are also increasing owing to the emerging Brazilian bio-diesel industry and growing imports of soya by the burgeoning animal ration industries (Fearnside 2001; Brown 2004; Clay 2004; Nepstad et al. 2006a; USDA 2007). China is the leader of a group of countries (referred to here as the ‘emerging meat-eating nations’) whose growing affluent classes are consuming larger amounts of ration-fed poultry, pork and other types of meat (Brown 2004). Soya is the best source of vegetable protein in animal ration (Mosse 1990) and is highly sought after by the animal ration industries (Brookes 2001; FAO 2004). The price of soya today is approaching its highest level in history (Zafalon 2007); as a result, the Brazilian soya harvest (2007–2008) is expected to expand 7% over the previous year (Zafalon 2007).

The drought-induced death of Amazon forest's dominant organisms (its canopy trees) may increase the prospect of further disturbance from fire for years afterwards (Holdsworth & Uhl 1997; Brando et al. 2008). The leaf crowns of canopy trees separate the zone of solar radiation absorption and conversion to sensible heat high above the ground from the dark, humid forest floor far below. Each canopy tree that dies creates a canopy gap through which radiant energy penetrates into the forest, warming the forest interior. Field studies involving experimental forest fires indicate that forest susceptibility to fire is inversely related to the density and average height of the leaf canopy (Ray et al. 2005).

Land-use activities of the Amazon contribute to forest susceptibility to fire by providing ignition sources, by fragmenting the forest, and by thinning the forest through logging. Although ignition from lightening is rare in the central forests of the Amazon, man-made sources of ignition are increasingly abundant. Fires set to burn felled forests in preparation for crops or pasture, or to improve pasture forage, frequently escape beyond their intended boundaries into neighbouring forests. During the severe drought of 1998, approximately 39000km² of standing forest caught fire in the Brazilian Amazon (Alencar et al. 2006), which is twice the area of forest that was clear-cut that year. During the severe drought of 2005 (Aragão et al. 2007), at least 3000km² of standing forest burned in the southwest Amazon (Brown et al. 2006). These low, slow-moving fires can be deceptively destructive, killing from 10 to 50% of trees (above 10cm in diameter; Cochrane & Schulze 1999; Barlow & Peres 2004; Alencar et al. 2006). Forest fire can therefore increase susceptibility to further burning in a positive feedback by killing trees, opening the canopy and increasing solar penetration to the forest floor (Nepstad et al. 1995, 1999, 2001; Cochrane & Schulze 1999; Cochrane et al. 1999; Alencar et al. 2004).

Forest degradation is also fostered by fragmentation and edge effects associated with forest clear-cutting for pasture formation. Tree mortality and forest flammability are higher along forest edges (Laurance et al. 1997; Alencar et al. 2004). Selective logging, which can damage up to 50% of the leaf canopy (Uhl & Vieira 1989), also increases forest susceptibility to fire (Uhl & Kauffman 1990; Holdsworth & Uhl 1997). Selective logging degrades about as much forest each year as does clear-cutting (Nepstad et al. 1999; Asner et al. 2005).

Tree mortality induced by drought, fire, fragmentation and selective logging is the initial step of a process of forest degradation that is reinforced by positive feedbacks that can, potentially, convert forest ecosystems into fire-prone ‘brush’ vegetation (figure 2). Although we are unaware of systematic field surveys of repeatedly burned forests to document the types and extent of this severe forest degradation, our field observations in northeastern Mato Grosso, southeastern and eastern Pará states and near Santarém suggest that repeated disturbance can transform Amazon forests into low-stature ecosystems dominated by invasive, coppicing tree species (such as Solanum crinitum and Vismia guianensis), grasses (including Imperata brasiliensis, Paspalum spp.), bamboo (Guandu spp.) and ferns (such as Pteridium aquilinum; Ray et al. 2005; Alencar et al. 2006; D. C. Nepstad 2004–2007, unpublished data). Although still rare in occurrence, forests invaded by grasses may be the most susceptible to conversion to brush owing to the large amount of fuel produced by many graminoids (Uhl & Kaufmann 1990) and owing to their inhibitory effect on tree regeneration (Nepstad et al. 1996). Grasses invade forests through either seed dispersal by animals or wind, or deliberate seeding by cattle ranchers seeking to capitalize on the burning of the forest reserves on their properties (D. C. Nepstad 2007, unpublished data). The potential of grasses to displace extensive areas of forest is best exemplified in Southeast Asia, where a single grass species (Imperata cylindrica) now dominates soils once covered by moist tropical forest across approximately 300000km² (MacDonald 2004).

A more likely near-term shift in Amazon climate may be associated with changes in sea surface temperature that are usually associated with Amazon drought. Rainfall tends to decline in the Amazon when sea surface temperatures rise along the Pacific coast of northern South America through El Niño episodes (Marengo et al. in press). The warming of the sea surface between western Africa and the Gulf of Mexico, the Northern Tropical Atlantic Anomaly, is also associated with drought in the Amazon Basin (Marengo et al. in press). Some climatologists believe that these anomalies will become more frequent as greenhouse gases accumulate further in the atmosphere (Trenberth & Hoar 1997; Timmermann et al. 1999; Hansen et al. 2006).

Deforestation alone may also provoke reductions in rainfall (Nobre et al. 1991; Silva Dias et al. 2002; Da Silva et al. in press). It now appears that clearing beyond approximately 30% of the forests of the region may trigger a decline in rainfall (Sampaio et al. 2007; Da Silva et al. in press) and that projected future trends in deforestation may inhibit rainfall more during El Niño episodes than during non-El Niño years (Da Silva et al. in press).

We provide a preliminary map and quantitative estimate of the potential impacts of a near-term forest dieback by assuming ‘business-as-usual’ patterns of future deforestation as estimated by Soares-Filho et al. (2006), by assuming that Amazon climatic conditions of the January 1996 through December 2005 decade (Nepstad et al. 2004) are repeated in the future, that a tree mortality threshold is exceeded when PAW falls below 30% of its maximum value to a depth of 10m (Nepstad et al. 2004, 2007a), and that logging will expand across the Amazon as described in the rent-based economic model of Merry et al. (in press) (figure 3). We do not invoke fire in this estimate, although it is probable that a substantial portion of the forests that are damaged by drought and/or logging, and additional forests that are not damaged, will experience understorey fire. Given these assumptions, 31% of the Amazon closed-canopy forest formation will be deforested and 24% will be damaged by drought or logging by the year 2030. If we assume that rainfall declines 10% in the future, then an additional 4% of the forests will be damaged by drought. If we assume that carbon release from deforestation is as described in Soares-Filho et al. (2006), that selective logging reduces forest carbon stocks by 15% (Asner et al. 2005), that drought damage causes a 10% reduction in forest biomass (Nepstad et al. 2007a) and that fire affects 20% of the forests damaged by drought or logging, releasing an additional 20% of forest carbon to the atmosphere, then 15–26Pg of carbon contained in Amazon forest trees will be released to the atmosphere in less than three decades in a large-scale dieback without invoking the influence of global warming. This is equivalent to 1.5–2.6 years of current, worldwide, average, human-induced carbon emissions (Canadell et al. 2007).

Second is the change in landholder behaviour that is already evident in places in Mato Grosso and that could rapidly become widespread when sound land stewardship and compliance with environmental legislation are viewed by a growing number of producers as the necessary conditions for participating in commodity markets and for obtaining access to credit and financing. The soya growers of Mato Grosso are in the midst of such a change today as they enter their second year of a moratorium imposed by the Associação Brasileira das Indústrias de Óleos Vegetais (The Brazilian Vegetable Oil Industry Associaton). This 2-year moratorium on the purchase of soya planted on soils recently cleared from Amazon forest was stimulated by a campaign against Amazon soya initiated by the environmental organization Greenpeace (Greenpeace 2006). Soya producers and their organizations are currently developing criteria systems that would certify their farms as environmentally sound (Arini 2007). It is in the context of market pressures and certification systems that landholders are agreeing to adopt fire prevention techniques such as fire breaks along their forest borders, to seek compliance with the private forest reserve requirements of federal legislation and to conserve their riparian zones.

Third is the prospect of restricting the advance of Amazon cattle ranching. Approximately one-quarter of the previously forested lands in the Brazilian Amazon are in some stage of abandonment (Houghton et al. 2000), and most of these are degraded cattle pastures. The replacement of carbon- and species-rich forests with cattle production systems that support less than one animal unit per hectare presents very high societal costs and low societal benefits, thereby violating Brazil's own constitution, which states that the ‘social function’ of the land is to provide maximum economic and ecological benefits to the society (Benatti 2003; Ankersen & Ruppert 2006). One way to reduce deforestation driven by cattle ranchers, many of whom are using pastures to help stake claims to land parcels (Alston et al. 1999; Alencar et al. 2004), would be to prohibit the clearing of lands that are marginally suitable for mechanized agriculture (except where smallholder agriculture systems are appropriate). In an initial attempt to map the areas that are the most suitable for highly productive, mechanized agriculture, we employed the Shuttle Radar Topography Mission (Farr et al. 2007) dataset to map slope and height above river channels (as a measure of potential for inundation), and the RADAMBRASIL and FAO soils maps (described in Nepstad et al. 2004) to identify soils that typically contain rocks, impeding layers, and poor drainage (ultisols, lithosols, sands and hydromorphic soils; figure 4). The resulting map correctly identifies large areas of arable land where the majority of Amazon soya production is located, in northern Mato Grosso and in the Santa Cruz region of Bolivia. It also indicates potential for agro-industrial expansion in southern Peru, in the Madre de Dios region. These regions have strong dry seasons and may be suitable for soya bean expansion and, perhaps, other crops that require a dry period. A large block of arable, forested land with very little rainfall seasonality is also found in northern Colombia, northwestern Amazonas state and Northern Peru. This area may be the target of oil palm expansion, which needs little rainfall seasonality. The total area of the Amazon that is ‘suitable’ for industrial agriculture (with no edaphic or climatic restrictions) is 33%, while the area that is ‘very suitable’ (with only one restriction) is 22%. Constraining cattle pasture expansion in the Amazon would not necessarily restrict the growth rate of the Amazon cattle herd. Cattle production systems that incorporate silage and agricultural residue into the diet of the herd can achieve grazing densities of more than six animals per hectare (Landers et al. 2005; M. Reis et al. 2007, personal communication), and will be favoured if land prices escalate through growing demands for Amazon soil.

The prospect of restricting the expansion of agricultural and livestock production in the Amazon to those lands of high suitability becomes plausible in the light of a fourth important trend. Most parks and biological reserves in the Amazon have been created far from the agricultural frontier, where the opportunity costs of conservation are low and where, unfortunately, these protected areas exert little influence on deforestation (Nepstad et al. 2006b). But this historical trend may be changing. From early 2004 to 2006, 23Mha of protected areas and national forest districts were created in the Brazilian Amazon, including large areas of protected land in the pathway of the expanding agricultural frontier (Campos & Nepstad 2006). This historic achievement was made possible by a participatory, multiple-stakeholder, regional planning process that has emerged along the BR-163 (Santarém-Cuiabá) highway, which is slated for paving, and by the smallholder farmers’ movement of the Transamazon highway region, which successfully initiated the ‘Terra do Meio’ conservation mosaic (Campos & Nepstad 2006). Through the combination of (i) regional, participatory planning processes that lead to effective land-use zoning systems, (ii) recent government demonstrations of political will in establishing protected areas expressed through the Amazon region protected area programme and in cracking down on corrupt government employees, and (iii) growing market demands for higher environmental and social performance of Amazon farmers and ranchers, it is conceivable that deforestation could be restricted to the one-quarter of the region that is suitable for modern mechanized production (figure 4).