Correspondence Gregory P. Asner, Department of Global Ecology, Carnegie Institution, 260 Panama Street, Stanford CA 94305, USA. Tel: 650-462-1047; fax: 650-462-5968. E-mail: firstname.lastname@example.org
New deforestation and selective logging data and climate change projections suggest that biodiversity refugia in humid tropical forests may change more extensively than previously reported. However, the relative impacts from climate change and land use vary by region. In the Amazon, a combination of climate change and land use renders up to 81% of the region susceptible to rapid vegetation change. In the Congo, logging and climate change could negatively affect the biodiversity in 35–74% of the basin. Climate-driven changes may play a smaller role in Asia-Oceania compared to that of Latin America or Africa, but land use renders 60–77% of Asia-Oceania susceptible to major biodiversity changes. By 2100, only 18–45% of the biome will remain intact. The results provide new input on the geography of projected climate change relative to ongoing land-use change to better determine where biological conservation might be most effective in this century.
Although the impacts of climate change on tropical forest carbon stocks hinge upon the direction of precipitation changes (Prentice et al. 2007), the effects on biodiversity are likely negative regardless of the precise nature of climate change. Tropical forest species are specialized to narrower climate ranges than are temperate species (Tewksbury et al. 2008). Tropical species are also often poor dispersers (Van Houtan et al. 2007), but the speeds and distances required for species to move with changing climates across tropical forests are large (Loarie et al. 2009; Wright et al. 2009). Droughts coupled with increased evapotranspiration from rising temperatures can cause forest dieback expressed as the loss of both carbon and tropical species (Phillips et al. 2009). Moreover, there is a significant likelihood of future forest dieback in the Amazon under most climate change projections (Malhi et al. 2009). However, for biodiversity, scenarios of increased precipitation may be equally precarious, as tropical forests already tend to occupy the warmest and the wettest parts of the globe. It is uncertain which species will adapt to novel climates projected to concentrate in tropical forest biomes (Williams et al. 2007).
Projecting spatially explicit maps of future LU change is difficult. Deforestation may spread unexpectedly to areas that are currently pristine, and forests may be allowed to regrow in previously cleared areas. However, deforestation tends to advance along fronts such as the Arc of Deforestation in Brazil (Soares-Filho et al. 2006), and is usually preceded by selective logging (Asner et al. 2006). The global footprint of present-day logging thus serves as an estimate of the footprint of advancing deforestation.
Despite the importance of climate change, deforestation, and logging in the humid tropics, it remains unknown where biodiversity will likely be buffered from the combined effects of these forces. Although most studies focus either on LU or climate change in regions such as Amazonia (Shukla et al. 1990; Cox et al. 2004; Malhi et al. 2009), on deforestation globally (Achard et al. 2007; Hansen et al. 2008), or on climate and LU impacts on specific taxa or guilds (Jetz et al. 2007), we consider the combined effects of LU and climate change on the entire humid tropical forest biome. We integrated recent global deforestation and logging maps derived from 2000 to 2005 satellite imagery of clear-cutting (Hansen et al. 2008) and high-resolution satellite data (Asner et al. 2009) with projected future vegetation changes from 16 different Global Climate Models (GCMs) driving the Lund-Potsdam-Jena Managed Land (LPJmL) Global Dynamic Vegetation Model (Bondeau et al. 2007) to calculate the degree of projected reshuffling among plant functional types by the year 2100. We used plant functional type reorganization as a metric for biodiversity change, since the types of species occupying the canopy have cascading effects on the greater composition and diversity of tropical forests. Because we did not explicitly model future LU change, we ignore interactions between climate and LU such as fire.
We sought to understand the extent to which projected climate impacts and current advancing deforestation and logging co-vary spatially. Considering these impacts together reveals opportunities for land managers and decision-makers. In areas projected to be affected by climate change, reducing pressure from deforestation may increase the resilience of these ecosystems and facilitate the ability of species to move and keep pace with climate (Williams et al. 2005). In contrast, deforested regions likely to be spared from the more severe climate changes may be good candidates for restoration.
The biome extent was defined at http://globalmonitoring.sdstate.edu/, which is based on the humid tropical forest portion of the map produced by Myers et al. (2000). Forests were mapped by compiling the 500 m resolution Vegetation Continuous Field (VCF) tree canopy data (VCF Collection 4, Version 3) into the following categories: 0–50% and 50–100% forest cover as of 2005. We also calculated gross deforestation between 2000 and 2005 from the data provided by Hansen et al. (2008), which they calibrated to higher resolution satellite data. We defined the overall footprint of recent deforestation by classifying these gross rates of forest loss at a threshold level of 1% or more per pixel from 2000 to 2005 (Figure 1).
The global extent of selective logging operations, defined here as a combination of recently harvested forests and new timber concessions, was estimated using multiple data sources (Asner et al. 2009). For Brazil and Peru, which represented 41% of our total global logging area (Table 1), the geographic extent of logging was mapped for the years 2001–2003 using the Carnegie Landsat Analysis System (CLAS) (Asner et al. 2005; Asner et al. 2006; Oliveira et al. 2007). Throughout the Congo basin of central Africa, the extent of logging operations and concessions was recently mapped (Laporte et al. 2007), representing 14% of our global logging footprint. Direct mapping of logged areas and concessions in Papua New Guinea in 2002 was also recently completed (Shearman et al. 2008), representing 5% of the global total. In addition, we ran CLAS on Borneo for imagery collected in 2002, which provided an additional 8% of the global logging map. Together, direct satellite mapping of selective timber harvest covered 68% of the logging results. For other regions of the humid tropical forest biome, we estimated the general extent of logging activities using Google Earth™ (GE) imagery, guided by statistics provided by the United Nations Food and Agricultural Organization (FAO) State of the World's Forests assessment (FAO 2007) as described in detail by Asner et al. (2009). Combined, the logging area dataset represents the period 2000–2005, and is temporally compatible with the deforestation map. The map of logging extent is shown by region in Figure 1.
Table 1. Partitioning of data sources used to estimate the logging footprint in humid tropical forests. Values are in km2 and percentage of total global dataset
Country or region
Papua New Guinea
Other (Google Earth)
We ran the LPJmL Dynamic Vegetation Model (Sitch et al. 2003; Bondeau et al. 2007) on 16 GCMs from the World Climate Research Programme's CMIP3 data driven by the moderate-high SRES A2 greenhouse gas emission scenario. The climate data have been bi-linearly interpolated to 1-degree resolution and corrected for bias in 1961–1990 with an extended CRU data set (Oesterle et al. 2003) (additive correction for air temperature and cloud cover, multiplicative for precipitation). We considered average results for recent (1971–2000) and future (2069–2098) time periods. Within the analysis area, there were seven plant functional types with nonzero fractions: tropical broadleaf evergreen trees, tropical broadleaf drought deciduous (“raingreen”) trees, temperate needleleaf evergreen trees, temperate broadleaf evergreen trees, temperate broadleaf deciduous (“summergreen”) trees, C3 grass, and C4 grass. Bare soil was also modeled.
As a proxy for vegetation reshuffling from climate change, we calculated the Euclidean distance (ED) between the current (x) and future (y) vectors of plant functional type fractions such that for each grid cell and each GCM (Figure S1). We summarized the variation among GCMs by mean, median, lower, and upper quantiles (Figure 2). Most studies projecting severe Amazon dieback used the HadCM3 GCM (Cox et al. 2004). While Amazon dieback is a general feature across larger suites of GCMs (Malhi et al. 2009), the extent and magnitude is generally less severe than that projected by the HadCM3 GCM. Nonetheless, the general patterns of vegetation reshuffling are consistent across a range of summary statistics. We chose to report the median results from all models since the median is less influenced by extremely severe GCMs results such as from HadCM3. Next, we thresholded the ED statistic by quantiles into low (0–33%), medium (33–66%), and high (66–100%) climate impacts (Figure 3), which allowed us to explore refugia from the perspective of both moderate and severe climate change.
Refugia from the climate and LU combination
To map climate effects on vegetation reshuffling with logging and deforestation, we used an inverse distance weighting algorithm to interpolate the 1° projections across a finer 20-km grid. We overlaid these projections with 20-km grids of recent deforestation and logging. We calculated the land area of potential refugia without recent deforestation, logging, and two levels of projected climate impacts. For refugia from moderate climate impacts, we included all areas within the humid tropics less than the 66% global quantile of increasing vegetation reshuffling (Figure 4, left column). For refugia from severe climate impacts, we included all areas less that the 33% global quantile (Figure 4, right column). Our estimates of refugia areas are conservative because we include areas that were deforested before 2000 (e.g., much of West Africa) and areas with naturally low forest cover (e.g,. much of southern China) as potential refugia. The MODIS VCF data does not distinguish between historic deforestation and naturally low vegetation cover.
Drivers of climate impacts
To better understand which changes in plant functional types drive the overall regional results from Figure 4, we mapped the median absolute fractional change in each plant type (Figure 5). ED patterns are driven primarily by changes in three plant functional types. They generally involve decreases in evergreen trees, increases in drought deciduous trees, and variable changes in C4 grasses. This increase in drought deciduous trees at the expense of evergreen trees is a consistent feature across LPJ simulations from all 16 GCM projections despite varying predictions for precipitation change. This response results from a change in the water balance caused by consistent increases in temperature in the global humid tropical forest biome (IPCC 2007).
Among the 16 GCMs each coupled to the LPJ vegetation model, there are consistent spatial patterns of relatively severe climate-change effects on humid tropical forests concentrated in the northeast of the South American tropics and along the edges of the African tropics (Figures 2 and 3). Climate impacts are relatively less severe in the Asia-Oceania tropics. Broadleaf evergreen trees, broadleaf drought deciduous trees, and C4 grasses are the dominant plant functional types across the biome. Climate impacts are primarily expressed by a reshuffling among these three types, but particularly with transitions from broadleaf evergreen to broadleaf drought deciduous vegetation (Figure 5). See supplementary online information for further details on the GCM and LPJ projections (Figures S2 and S3).
In the Amazon-Guyana Shield region, deforestation is systematically removing forests and biodiversity to the southeast, while logging extends the direct human impact on biodiversity more than twice as far north and westward (Figure 6). The footprint of deforestation and logging now covers 29% of Amazonia. At the continental scale, about 24% and 15% of the humid tropical forest biome in South and Central America, respectively, are now directly affected by clearing and logging (Table 2).
Table 2. Percentage of the humid tropical forest biome with current major land use (LU) as expressed by deforestation and selective logging, predicted major climate change causing a reshuffling of vegetation types (C), a combination of climate change and land use (C+LU), and neither climate change nor land use (N)
C + LU
The most severe climate-driven pressures on the Amazonian biota are projected to strike 37% of basin from the northeast and southwest, with an additional 50% of the region undergoing moderate climate-driven changes (Figure 6). On a continental scale that also includes Mesoamerica and the Brazilian Atlantic Forest, climate change may negatively impact 63–66% of humid tropical forests (Table 2). Combined, climate change and LU renders the Amazon basin susceptible to forthcoming biodiversity changes in 57–81% of the region, depending upon the severity of the latest climate change projections and the sensitivity of the biota to those projected changes. At the continental scale, only 16% and 9% of Central and South American humid tropical forests, respectively, have low projected climate change and current LU pressure (Table 2). See supplementary online information for information on country-level statistics of LU and projected climate change (Table S1).
Projected changes in the Congo basin evolve mostly from selective logging and climate change (Figure 6b). Logging operations cover vast areas of Gabon and the borderlands connecting Cameroon, Congo, and the Central Africa Republic. There is now extensive ongoing timber extraction in the heart of the humid forests in Democratic Republic of Congo (DRC). These logging areas occur at the geographic centroid of the predicted zone of low climate impact, yet a region of projected climate-driven change does stretch across a vast area of the southern Congo basin in DRC. In all, logging and climate change could negatively affect the biodiversity in 35–74% of the basin. At a continental scale, about 71% of all humid Afrotropical forests may be destined for climate and LU impacts under business-as-usual regimes (Table 2).
Humid tropical forests of Asia and Oceania originally covered an area approaching that of Amazonia, but both deforestation and logging have been and continue to be dominant forces of change throughout the region (Figure 6c). Climate-driven changes in biome composition may play a smaller role here compared to that of Latin America or Africa, but LU still renders up to 77% of the region susceptible to major biodiversity losses (Table 2). Submontane to montane portions of Indonesia and New Guinea offer the best opportunities for refugia in Asia and Oceania.
The geography of projected climate and LU change casts a sobering shadow over the future of biodiversity in humid tropical forests. At the global level, only about 20% of the biome will likely remain beyond the footprints of these reorganizing forces. Our knowledge of deforestation and logging varies by region, with the most detailed information available for Amazonia and the least detail in parts of Asia due mainly to the complexity of LU patterns and cloud cover (Achard et al. 2007; Asner et al. 2009). However, we know that deforestation characteristically spreads across the landscape from advancing fronts, some large in scale and others much smaller. But perhaps it is most surprising that projected climate impacts are also unevenly distributed across the tropics, and that these spatial patterns are consistent among the 16 GCMs considered here.
Although this consistency suggests that these spatial patterns are robust, we note that variation among the 16 GCMs represents only a subset of uncertainty in projected climate impacts, which also includes emission levels and species responses to climate. The single emission scenario we used was considered moderate-high, but it has been exceeded by actual emissions (Raupach et al. 2007). Nonetheless, we ignore possible interactions among LU and the climate system that may feedback on emission levels. For example, drying forests may facilitate burning thereby accelerating LU change and emissions (Gullison et al. 2007).
Dynamic vegetation models offer a more mechanistic approach to estimating species responses to climate change than do static niche modeling approaches (Thuiller et al. 2006). LPJ consistently projects increases in drought deciduous species despite variability in GCM predictions of future precipitation. Under decreasing or unchanged precipitation, this result is intuitive, as increasing temperatures will drive increased evapotranspiration and water stress. Other studies predicted similar increases in evapotranspiration across a range of GCM projections for the Amazon basin (Malhi et al. 2009). In models that project increasing precipitation, such as from the NASA Goddard Institute for Space Studies GCM, the LPJ-projected response of increases in the geographic extent of drought deciduous species may seem counterintuitive, but vegetation projections are difficult in novel warmer and wetter climates for which analogous communities are globally rare (Williams et al. 2007). Even if LPJ incorrectly simulates vegetation in these novel climates, it is likely that tropical species, with their specialized climate tolerances and poor dispersal abilities, will be impacted by these changes.
These broad patterns of LU and climate change provide critically needed insight into the geography of strategies that could be implemented to protect tropical biodiversity in the coming century. Regions of high climate impact, such as in NE Amazonia, are prime targets for reduced LU pressure to allow species to migrate along an east-west corridor stretching between the Guyana Shield and Andes-Amazon countries. The connectivity between high- and low-impact climate zones in this region may be central to maintaining biodiversity. Positive outcomes could also be reached by reducing logging and deforestation pressure in the SE Amazon, where the forecast for climate-driven change is less acute, and thus protection could function to preserve biodiversity within reserve boundaries.
In the Congo, support for reduced-impact logging could play an important role in the heart of the basin where climate change plays a weaker role. This would provide a core haven for a greater regional biodiversity that will otherwise be frayed by climate change and human activities along the perimeter of the Congo basin. Today, there is only a small, yet rapidly shrinking area of Asia-Oceania resting outside of major LU pressures. Nonetheless, of all the regions in the humid tropics, Oceania in general holds what may be the most promise for sustaining biodiversity, given that climate-change impacts on the biota are projected to be weaker in this region. Reduction of losses incurred by logging and deforestation could have a proportionally larger positive effect here than in other tropical regions.
The combined effects of climate change and spreading LU represents one of the greatest global change experiments on Earth today. The geographic footprints of climate-driven vegetation change, deforestation, and logging are not spatially aligned, presenting both opportunities and challenges in efforts to preserve biodiversity. Reducing pressure from LU change in areas projected to be severely impacted by climate change may bolster climate resilience and facilitate migration. International mechanisms to reduce emissions such as the United Nations program to Reduce Emissions from Deforestation and Degradation (REDD) may decrease LU impacts on biodiversity by limiting deforestation. Restoration in areas with little remaining forest, and where climate impacts are projected to be minimal, may be an effective way to implement REDD by providing the dual benefits of offsetting emissions and restoring refugia. And in general, the effects of projected climate change on biodiversity should be considered in the application of these policies to determine where both carbon sequestration and biological conservation might be most effective.
We thank D. Lindenmayer and two reviewers for critical comments on the manuscript. The Carnegie Landsat Analysis System (http://claslite.ciw.edu), Global Spectranomics project (http://spectranomics.ciw.edu), and this study are supported by the John D. and Catherine T. MacArthur Foundation and the Gordon and Betty Moore Foundation.