Expansion of the cattle and soy industries in the Amazon basin has increased deforestation rates and will soon push all-weather highways into the region's core. In the face of this growing pressure, a comprehensive conservation strategy for the Amazon basin should protect its watersheds, the full range of species and ecosystem diversity, and the stability of regional climates. Here we report that protected areas in the Amazon basin--the central feature of prevailing conservation approaches--are an important but insufficient component of this strategy, based on policy-sensitive simulations of future deforestation. By 2050, current trends in agricultural expansion will eliminate a total of 40% of Amazon forests, including at least two-thirds of the forest cover of six major watersheds and 12 ecoregions, releasing 32 +/- 8 Pg of carbon to the atmosphere. One-quarter of the 382 mammalian species examined will lose more than 40% of the forest within their Amazon ranges. Although an expanded and enforced network of protected areas could avoid as much as one-third of this projected forest loss, conservation on private lands is also essential. Expanding market pressures for sound land management and prevention of forest clearing on lands unsuitable for agriculture are critical ingredients of a strategy for comprehensive conservation.

No abstract

Conservation scientists generally agree that many types of protected areas will be needed to protect tropical forests. But little is known of the comparative performance of inhabited and uninhabited reserves in slowing the most extreme form of forest disturbance: conversion to agriculture. We used satellite-based maps of land cover and fire occurrence in the Brazilian Amazon to compare the performance of large (> 10,000 ha) uninhabited (parks) and inhabited (indigenous lands, extractive reserves, and national forests) reserves. Reserves significantly reduced both deforestation and fire. Deforestation was 1.7 (extractive reserves) to 20 (parks) times higher along the outside versus the inside of the reserve perimeters and fire occurrence was 4 (indigenous lands) to 9 (national forests) times higher. No strong difference in the inhibition of deforestation (p = 0. 11) or fire (p = 0.34) was found between parks and indigenous lands. However, uninhabited reserves tended to be located away from areas of high deforestation and burning rates. In contrast, indigenous lands were often created in response to frontier expansion, and many prevented deforestation completely despite high rates of deforestation along their boundaries. The inhibitory effect of indigenous lands on deforestation was strong after centuries of contact with the national society and was not correlated with indigenous population density. Indigenous lands occupy one-fifth of the Brazilian Amazon-five times the area under protection in parks--and are currently the most important barrier to Amazon deforestation. As the protected-area network expands from 36% to 41% of the Brazilian Amazon over the coming years, the greatest challenge will be successful reserve implementation in high-risk areas of frontier expansion as indigenous lands are strengthened. This success will depend on a broad base of political support.

Severe drought in moist tropical forests provokes large carbon emissions by increasing forest flammability and tree mortality, and by suppressing tree growth. The frequency and severity of drought in the tropics may increase through stronger El Niñ o Southern Oscillation (ENSO) episodes, global warming, and rainfall inhibition by land use change. However, little is known about the spatial and temporal patterns of drought in moist tropical forests, and the complex relationships between patterns of drought and forest fire regimes, tree mortality, and productivity. We present a simple geographic information system soil water balance model, called RisQue (Risco de Queimada -Fire Risk) for the Amazon basin that we use to conduct an analysis of these patterns for 1996-2001. RisQue features a map of maximum plant-available soil water (PAW max ) developed using 1565 soil texture profiles and empirical relationships between soil texture and critical soil water parameters. PAW is depleted by monthly evapotranspiration (ET) fields estimated using the Penman-Monteith equation and satellite-derived radiation inputs and recharged by monthly rain fields estimated from 266 meteorological stations. Modeled PAW to 10 m depth (PAW 10 m ) was similar to field measurements made in two Amazon forests. During the severe drought of 2001, PAW 10 m fell to below 25% of PAW max in 31% of the region's forests and fell below 50% PAW max in half of the forests. Field measurements and experimental forest fires indicate that soil moisture depletion below 25% PAW max corresponds to a reduction in leaf area index of approximately 25%, increasing forest flammability. Hence, approximately one-third of Amazon forests became susceptible to fire during the 2001 ENSO period. Field measurements also suggest that the ENSO drought of 2001 reduced carbon storage by approximately 0.2 Pg relative to years without severe soil moisture deficits. RisQue is sensitive to spin-up time, rooting depth, and errors in ET estimates. Improvements in our ability to accurately model soil moisture content of Amazon forests will depend upon better understanding of forest rooting depths, which can extend to beyond 15 m. RisQue provides a tool for early detection of forest fire risk.

The forests of Russia cover a larger area and hold more carbon than the forests of any other nation and thus have the potential for a major role in global warming. Despite a systematic inventory of these forests, however, estimates of total carbon stocks vary, and spatial variations in the stocks within large aggregated units of land are unknown, thus hampering measurement of sources and sinks of carbon. We mapped the distribution of living forest biomass for the year 2000 by developing a relationship between ground measurements of wood volume at 12 sites throughout the Russian Federation and data from the MODIS satellite bidirectional reflectance distribution function (BRDF) product (MOD43B4). Based on the results of regression-tree analyses, we used the MOD43B4 product to assign biomass values to individual 500 m × 500 m cells in areas identified as forest by two satellite-based maps of land cover. According to the analysis, the total living biomass varied between 46 and 67 Pg, largely because of different estimates of forest area. Although optical data are limited in distinguishing differences in biomass in closed canopy forests, the estimates of total living biomass obtained here varied more in response to different definitions of forest than to saturation of the optical sensing of biomass.