A Global Experiment UnderWay

HABITAT LOSS AND FRAGMENTATION ARE THE principal drivers of biodiversity loss, notably in the tropics (7). One of the most immediate results of fragmentation is the loss of top predators. As early as 12 years ago, evidence was provided for top-down effects of predator removal in tropical forest mammalian communities (2). Now, in their report "Ecological meltdown in predator-free forest fragments" (30 Nov., p. 1923), John Terborgh and colleagues have demonstrated that these effects percolate down to taxonomically diverse groups, affecting plant-herbivore dynamics in forest islands that resulted from dam construction in Venezuela. In the accompanying Perspective "Dammed experiments!" (30 Nov., p. 1847), Jared Diamond reminds us that these represent valuable "natural" experiments. Alternatives are more carefully controlled experiments, like the Biological Dynamics of Forest Fragments Project in the Brazilian Amazon, but they are also more costly and difficult to implement beyond a few areas. Although these localized natural experiments are useful, a "natural" fragmentation experiment on a grand scale is possible because most biologically rich tropical forests retain less than 30% of their original extent (3). The Major Tropical Wilderness Areas of the Amazon, Congo, and New Guinea (4) could serve as controls. The impediment to the full use of the global experiment is deriving baseline biological and land use information over large areas, but we believe this limitation can now be overcome. Over the last century, many now-fragmented landscapes were surveyed before fragmentation, and these data from museum collections are starting to become accessible in electronic format. With some incremental investment in transferring collection information to databases, specimen accessibility could expand manifold (5). Species modeling efforts can complement such needs (6). Furthermore, emerging programs to undertake biological surveys and long-term ecological monitoring at unprecedented scales (7) are in the final stages of design or already producing needed biological information. Other data that have become more acces-