Animal Versus Wind Dispersal and the Robustness of Tree Species to Deforestation Daniel Montoya, 1 * Miguel A. Zavala, 1,2 Miguel A. Rodríguez, 1 Drew W. Purves 3 Studies suggest that populations of different species do not decline equally after habitat loss. However, empirical tests have been confined to fine spatiotemporal scales and have rarely included plants. Using data from 89,365 forest survey plots covering peninsular Spain, we explored, for each of 34 common tree species, the relationship between probability of occurrence and the local cover of remaining forest. Twenty-four species showed a significant negative response to forest loss, so that decreased forest cover had a negative effect on tree diversity, but the responses of individual species were highly variable. Animal-dispersed species were less vulnerable to forest loss, with six showing positive responses to decreased forest cover. The results imply that plant-animal interactions help prevent the collapse of forest communities that suffer habitat destruction. H abitatdestructionisoftencitedasthesingle greatest cause of global biodiversity loss (1). These anthropogenic changes trigger biological responses that sometimes end in a biotic collapse, a problem that has led ecologists to face the question of how much habitat is enough for species to persist (2). The dominant theoretical framework for understanding the ef- fects of habitat loss is metapopulation theory, which focuses on the dynamic balance of local extinctions and colonizations that characterize fragmented populations at regional scales (3). According to this theory, regional habitat loss not only removes biodiversity held in the lost habitat but also reduces the occurrence of species within the remaining habitat (4). This idea has become a central tenet of conservation practice, causing a shiftinfocusfromthelocaltothelandscapescale. For example, it is the source of the current em- phasis on the maintenance and creation of habitat corridors to foster dispersal among patches (5, 6). However, empirical tests of this prediction have been restricted to short-lived animal species [espe- cially butterflies and birds (7, 8)], short spatial scales (9), and short time scales, over which ob- servations are likely to be dominated by short- term responses that may or may not be indicative of the long-term impacts of habitat loss. We analyzed the relationship between local forest cover and the occurrence of 34 canopy- dominant tree species [28 native to the study region and 6 exotic (table S1)] in 89,365 survey sites distributed across peninsular Spain (10) (Fig. 1). The data set was extracted from the Spanish Second National Forest Inventory (IFN2), which placed a 25-m-radius circular sample plot ineach1×1kmgridcellthatitclassifiedasbeing forested [occupied by woody vegetation (11)].For each plot q, we calculated a local forest cover H q , defined as the fraction of the nearest eight grid cells to q that were also classified as forested in IFN2 (using a larger neighborhood degraded the statistical significance of some effects documented here but had no qualitative effect on conclusions). Because the Iberian Peninsula has chronically suf- fered from forest destruction and conversion into agricultural and degraded states ( 12), H q is a mea- sure of net forest loss from prehistory to the present. Thus, we interpret the species responses to H q ob- servedintheIFN2surveyasresponsestoforestloss. We used logistic regression to quantify, for each species j, the probability of occurrence of j in plot q as a function of H q . For comparison among species we used the fitted logistic curves to calculate, for each species j, a scalar W j , de- fined as the natural log of the ratio of the prob- ability of occurrence at 0% local forest cover to the probability of occurrence at 75% cover. Neg- ative W j implies that species j shows a negative response to decreased forest cover and vice versa. We used error propagation to calculate a conserv- ative (upper) estimate of the confidence interval for W j . The results presented below are robust considering either native and exotic species com- bined or native species only [supporting online material (SOM)]. Of the 34 species, 24 showed a statistically significant negative response to decreased forest cover [negative W j value with confidence intervals not including zero (Fig. 2A)]. This is consistent with the decrease in average tree species richness with decreased forest cover observed in the IFN2 data (Fig. 3) and in previous studies (13). The observed relationship between species richness in this case was approximately linear over most of the range in H q , which was captured well by the logistic regressions (Fig. 3). However, rich- ness was lower than expected for H q ≥ 80% and H q = 0. Such abrupt changes could reflect the effects of spatial configuration (that is, fragmen- tation) when habitat cover goes from nearly con- tinuous to fragmented (with the first appearance of edges) and falls to very low levels (14), al- though threshold responses can also result from some forms of animal-mediated dispersal (15). Among species there was large and statistical- ly significant variation in W j . For species with statistically significant negative W j (those with confidenceintervalsnotincludingzero), W j ranged from –0.03 to –1.53, which corresponds to a pro- portional reduction in probability of occurrence, for the 75 to 0% scenario, of 3 to 78%. Moreover, there were six species with statistically significant posi- tive responses to reductions in forest cover (Fig. 2A). These species were more likely to be found in plots surrounded by nonforested land. If this magnitude of interspecific variation in response to forest loss proves to be typical, it will be critical to identify measurable species traits that predict it. Although we did not attempt an ex- haustive search of such traits, we did examine the importance of two traits related to dispersal (seed size and animal- versus wind-mediated seed dis- 1 Departamento de Ecología, Universidad de Alcalá, 28871 Alcalá de Henares, Madrid, Spain. 2 Centro de Investigación Forestal, Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria, Carretera de la Coruña km 7, 5, 28040 Madrid, Spain. 3 Microsoft Research Cambridge, 7 J. J. Thomson Avenue, Cambridge CB3 0FB, UK. *To whom correspondence should be addressed. E-mail: daniel.montoya@alu.uah.es Fig. 1. Distribution of survey sites in peninsu- lar Spain. IFN2 consisted of 89,365 circular sampl- ing sites (radius = 25 m) distributed across penin- sular Spain (average den- sity approximately one per square kilometer). Survey sites were placed in continuous forest lo- cations, so their distribu- tion matches that of the remaining forest. 13 JUNE 2008 VOL 320 SCIENCE www.sciencemag.org 1502 REPORTS on June 13, 2008 www.sciencemag.org Downloaded from