Study Species and Study Area.

The red kite (M. milvus) is an endemic raptor of the western Palearctic (Fig. 1), globally listed as “Near Threatened” (24). The European population, estimated as 25,200 to 33,400 breeding pairs, currently accounting for >95% of the global species population, has decreased by ∼30% over the last three generations (34.5 y), mainly due to declines within its core breeding areas in Germany, France, and Spain (26). Spain (our study area) holds roughly 2,000 breeding pairs and an estimated 50,000 wintering individuals (16), representing one of the world’s main breeding strongholds and wintering areas for red kites (31, 32). Nonetheless, the Spanish red kite wintering and breeding populations have experienced sharp decreases (7 to 19% and 31 to 43%, respectively) over the last two decades (15, 16, 33). Although the species trend in the last decade could be considered globally stable (16, 33), in Spain the red kite is legally considered “Endangered,” meaning that the species probability of survival is low as long as the primary threats to the population remain in place (34).

Red kites are opportunistic raptors that feed on a wide range of food resources, from small prey such as rodents, passerines, and young rabbits to all types of meat remains, including organic waste at dumps and carcasses of any size from large ungulates to reptiles, birds, and rodents (24). This makes the species highly susceptible to different toxic compounds, including those legally administered for pest control and deliberately poisoned baits aimed at illegal killing of predators (26, 35⇓–37). Illegal poison use is currently considered among the main threats for the species (24) and has been suggested to be a major factor behind red kite population declines in southwestern Europe (35). Mortality caused by various toxic substances is considered a factor in delaying the expansion of the Scottish reintroduced population of red kites (22), in contrast with reintroduced populations in England (20). Despite the reported high mortality caused by toxicants, these two populations are expanding, however (20, 22). In contrast, extensive anticoagulant rodenticide use in agrarian landscapes in northwestern Spain has been identified as a major driver of short-term population declines at the regional level (38). For instance, high mortality rates and demographic modeling support illegal poison use as a major cause of red kite population declines in small isolated populations in Doñana National Park and the Balearic Islands (13, 21, 23). The suggested major role of poisoning as a driver of large-scale and long-term declines of this species remains unproven, however.

Red Kite Breeding Population.

Data on the distribution and abundance of breeding red kites in mainland Spain were obtained from two national censuses performed in 1994 and 2014, the longest monitoring period available to date (15, 16). The census methodology is described in detail elsewhere (15, 39). In brief, breeding kites were located within 10 × 10-km UTM (Universal Transverse Mercator) squares in March to July of each monitoring year by trained volunteers through both car surveys at low speed (≤40 km/h) to obtain the individuals observed per km and an active search of breeding territories from vantage observation points or intensive nest searching. Data obtained by both methods were converted to population estimates using the equations shown in refs. 15 and 39.

Data on the presence and abundance of breeding red kites in 1994 were obtained for a total of 2,990 10 × 10-km squares (i.e., ∼300,000 km²; 60% of the country’s area). The species presence was confirmed in 386 squares (i.e., 38,600 km²), with an estimated total breeding population of 3,333 to 4,044 pairs (Fig. 1A) (15). In 2014, the species census covered 1,400 10 × 10-km squares (i.e., 140,000 km²), and the presence of red kites was reported in 554 squares, with an estimated population of 2,312 to 2,440 breeding pairs (Fig. 1B) (16). After correction for potential overestimation of breeding pairs detected in the last census, the Spanish breeding population was estimated as ∼2,000 breeding pairs, representing nearly a 40% decrease in its distribution over the two decades (15, 16). Data on red kite presence from both censuses were combined to obtain a response variable of the change in occupancy of breeding red kites per 10 × 10-km square in the 20-y period considered (n = 274; Fig. 1C and SI Appendix, Table S1). Those squares with breeding pairs in both censuses (i.e., 1994 and 2014) were codified as presences (value of 1; n = 167). Squares with breeding kites in 1994 but not in 2014 were considered as absences (or losses, codified as 0; n = 107). Changes in the abundance of breeding pairs among censuses were codified as a variable with two vectors: “number of succeeds,” the number of breeding pairs estimated per 10 × 10-km square in 2014, and “number of failures,” the number of breeding pairs that disappeared between 1994 and 2014 (40). Those squares with no change or an increase in the abundance of breeding pairs in the considered period were assigned 0 failures.

We excluded data from Baleares Islands and Doñana from the analyses, because these populations are geographically isolated from the rest of the Spanish red kite population and seem to present different habitat preferences (20). Nonetheless, both populations are known to have been heavily impacted by poisoning (13, 21, 23, 41).

Wildlife Poisoning Data and Other Environmental Variables.

To assess the potential influence of wildlife poisoning on the observed changes in distribution and abundance of the breeding population of red kite, we used a large and exhaustive database of confirmed or suspected cases of wildlife poisoning compiled over more than 3 decades in Spain by WWF Spain and Spanish Ornithological Society (SEO/BirdLife) (17). Most of these data were gathered by survey programs coordinated by governments of autonomous regions in Spain, and toxicologic analyses were performed at public and private ecotoxicology laboratories designated by the competent authorities. We considered 12 variables related to the incidence of poisoning in the fauna in general and in raptors and kites in particular (SI Appendix, Table S1). According to the low detectability of wildlife poisoning (estimated in 5 to 15% of the total cases) (17, 27), we also considered the number of poisoned dogs per square kilometer and year as a potentially more accurate index of the actual incidence of wildlife poisoning (19). Considering the red kite trophic ecology (i.e., feeding on small carcasses and meat remains) (24), we also calculated the density of poisoned baits as a proxy for the incidence of poisoning in natural ecosystems. Poison-related variables were calculated in two ways: considering all the episodes registered in the database in the period between censuses (i.e., 1995 to 2013; n = 18,500 animals and 4,175 baits) and considering only those episodes in which the presence of a toxic compound susceptible of poisoning wildlife was confirmed (n = 9,562 animals and 3,257 baits). All these variables were initially considered in the models.

To account for other factors able to influence the observed changes in the red kite breeding population, we considered several environmental factors previously identified as important to determine habitat suitability for the species in Spain. Seoane et al. (18) highlighted the importance of different land uses (e.g., cropland, forest, pasture), topographic variables (e.g., elevation, slope), and climatic variables (e.g., temperature) on the distribution and abundance of red kites. Among these variables considered important for the red kite in 1994, we incorporated into our analyses those susceptible of meaningful and noticeable changes in 1994 to 2014. We used CORINE Land Cover information from 1990 and 2012 (42) to calculate the percent change in crop, forest, and pasture lands, as well as urban surfaces, per 10 × 10-km square among the considered red kite censuses (SI Appendix, Table S1). We assumed that the remaining variables considered important for explaining habitat suitability for the red kite (e.g., climate, topography) had not changed noticeably in the period analyzed.

We considered the number of breeding pairs of red kite estimated per 10 × 10-km square in 1994 and the geometric mean of the breeding pairs estimated in 1994 and 2014 as a way to better determine red kite abundance during the considered period (SI Appendix, Table S1). Squares with a higher breeding population could be more resilient to species threats such as poisoning and land use change and thereby have a higher probability of remaining occupied. Red kite population sizes were used also to control the potential effect of population size on the number of poisoned kites recorded per square. The species population size was included in the predictive models in two ways, as an independent explanatory variable and as a divisor of the number of poisoned red kites in each square, to obtain an explanatory variable of poisoned kites per breeding pair (SI Appendix, Table S1).

Analysis.

We used GLMs with binomial error distributions and “logit” links to explain the observed changes in the distribution and abundance of breeding red kites between 1994 and 2014. From the significantly correlated explanatory variables (i.e., those with a Spearman correlation P < 0.05), we included in the models the one with the greater ecological significance or, if not easily discriminated, the variable with best results in a univariate GLM (e.g., lower corrected Akaike information criterion [AICc]) (43).

We used multimodel selection to choose the best models (43). We randomly selected 70% of the data to train the models and the remaining 30% to test the results. We used the training dataset to generate a set of models with all possible combinations of the considered explanatory variables. We tested both linear and quadratic responses, as well as interactions between pairs of variables. The resulting models were ranked by lower AICc and higher relative weight (ω_m), which indicates the probability that a model was the best one among all the candidate models evaluated. From these ranked models, we selected a set of best models comprising those whose weights added to 0.95 (Σω_m = 0.95). From the set of best models, we depleted the redundant ones, that is, those including the same variables of other more explicative models (i.e., with lower AICc and higher ω_m) but with some additional variable that complicates the model without improving its predictive ability. Once filtered, model weights were recalculated to sum to 1. The models resulting from this selection process were combined into a single model considered the final best model (43, 44). The coefficients of the explanatory variables included in the final best model were calculated by weighting each variable coefficient by the weight of the model in which the variable was included.

The relative contribution of each variable to the final best model (ω_i) was calculated by summing the weights of the models (ω_m) within the set of best models (Σω_m = 0.95) that included the considered variable. The most important variables were those with the highest relative weight (ω_i = Σω_m). We expected that our approach of selecting noncorrelated variables to include in the models would minimize multicollineality, thereby minimizing the issues associated with pondered models (44). Nonetheless, to obtain additional information on the performance of the explanatory variables, we also calculated Z values and P values using the Wald test (SI Appendix, Table S3) for each variable included in the final best model. All statistical analyses were performed in R (45).

Global Assessment of Poisoning as a Species Threat.

We searched the IUCN Red List (29) for animal species threatened by poisoning at global scale according to the IUCN criteria for the inclusion of threats. We considered a species to be threatened by poisoning when threat category “5.1.2: Unintentional Effects (species is not the target)”, “5.1.3: Persecution/Control,” or “9.3.3: Herbicides and Pesticides” was recorded in the IUCN’s assessment of the species. We restricted our search to only global assessments of animal species (i.e., kingdom Animalia). We also considered these figures for vertebrate species (i.e., phylum Chordata) and for raptors (i.e., orders Accipitriformes, Cathartiformes, Falconiformes, and Strigiformes) and carnivores (i.e., order Carnivora), as predators are either frequently prosecuted through illegal poisoning or affected by secondary poisoning (e.g., consuming poisoned prey, such as rodents) (1, 4⇓⇓⇓–8, 27, 35⇓⇓–38, 46, 47). The main results of this search are described in detail in SI Appendix, Table S4.

Data Availability.

Data on red kite populations are available in the census publications (15, 16). Additional details on this, as well as access to the ANTÍDOTO database on wildlife poisoning, are available on request from WWF Spain or SEO/BirdLife.