We examine the evidence for the possibility that 21st-century climate change may cause a large-scale ''dieback'' or degradation of Amazonian rainforest. We employ a new framework for evaluating the rainfall regime of tropical forests and from this deduce precipitation-based boundaries for current forest viability. We then examine climate simulations by 19 global climate models (GCMs) in this context and find that most tend to underestimate current rainfall. GCMs also vary greatly in their projections of future climate change in Amazonia. We attempt to take into account the differences between GCM-simulated and observed rainfall regimes in the 20th century. Our analysis suggests that dry-season water stress is likely to increase in E. Amazonia over the 21st century, but the region tends toward a climate more appropriate to seasonal forest than to savanna. These seasonal forests may be resilient to seasonal drought but are likely to face intensified water stress caused by higher temperatures and to be vulnerable to fires, which are at present naturally rare in much of Amazonia. The spread of fire ignition associated with advancing deforestation, logging, and fragmentation may act as nucleation points that trigger the transition of these seasonal forests into fire-dominated, low biomass forests. Conversely, deliberate limitation of deforestation and fire may be an effective intervention to maintain Amazonian forest resilience in the face of imposed 21st-century climate change. Such intervention may be enough to navigate E. Amazonia away from a possible ''tipping point,'' beyond which extensive rainforest would become unsustainable.carbon dioxide ͉ drought ͉ fire ͉ tropical forests ͉ adaptation

No abstract

captures the maximum possible range of changes in surface temperature and precipitation for three continental-scale regions. We find that, of the CMIP5 GCMs with 6-hourly fields available, three simulate the key regional aspects of climate sufficiently poorly that we consider the projections from those models 'implausible' (MIROC-ESM, MIROC-ESM-CHEM, and IPSL-CM5B-LR). From the remaining models, we demonstrate a selection methodology which avoids the poorest models by including them in the set only if their exclusion would significantly reduce the range of projections sampled. The result of this process is a set of models suitable for using to generate downscaled climate change information for a consistent multi-regional assessment of climate change impacts and adaptation.

We study the influence of station network density on the distributions and trends in indices of areaaverage daily precipitation and temperature in the E-OBS high resolution gridded dataset of daily climate over Europe, which was produced with the primary purpose of Regional Climate Model evaluation. Area averages can only be determined with reasonable accuracy from a sufficiently large number of stations within a grid-box. However, the station network on which E-OBS is based comprises only 2,316 stations, spread unevenly across approximately 18,000 0.22°grid-boxes. Consequently, grid-box data in E-OBS are derived through interpolation of stations up to 500 km distant, with the distance of stations that contribute significantly to any grid-box value increasing in areas with lower station density. Since more dispersed stations have less shared variance, the resultant interpolated values are likely to be over-smoothed, and extreme daily values even more so. We perform an experiment over five E-OBS grid boxes for precipitation and temperature that have a sufficiently dense local station network to enable a reasonable estimate of the area-average. We then create a series of randomly selected station subnetworks ranging in size from four to all stations within the E-OBS interpolation search radii. For each sub-network realisation, we estimate the grid-box average applying the same interpolation methodology as used for E-OBS, and then evaluate the effect of network density on the distribution of daily values, as well as trends in extremes indices. The results show that when fewer stations have been used for the interpolation, both precipitation and temperature are over-smoothed, leading to a strong tendency for interpolated daily values to be reduced relative to the ''true'' areaaverage. The smoothing is greatest for higher percentiles, and therefore has a disproportionate effect on extremes and any derived extremes indices. For many regions of the E-OBS dataset, the station density is sufficiently low to expect this smoothing effect to be significant and this should be borne in mind by any users of the E-OBS dataset.

No abstract