Population trends of grassland birds in North America are linked to the prevalence of an agricultural epizootic in Europe

Abstract

Globalization of trade has dramatic socioeconomic effects, and, intuitively, significant ecological effects should follow. However, few quantitative examples exist of the interrelationship of globalization, socioeconomics, and ecological patterns. We present a striking illustration of a cascade in which bovine spongiform encephalopathy (BSE; “mad cow disease”) outbreaks in Europe exerted pressure on global beef markets, subsequently affecting North American hayfields and grassland bird populations. We examined competing models, which linked the prevalence of BSE in five focal countries, volume of beef exports to those countries from North America, and the amount of hayfield harvested and the abundance of grassland birds in North America. We found that (i) imports from North America increased 1 y after BSE outbreaks; (ii) probably because fewer cattle remained, the hay harvest in North America was reduced 2 y after the outbreak; (iii) the reduced hay harvest yielded a positive response in grassland bird populations 3 y after the outbreak.

Native prairies in most of North America have been nearly extirpated over the past two centuries (1). In response, many prairie bird species have adopted agricultural grasslands as surrogate habitat (2, 3). Given the loss of native prairie habitat, grassland agro-ecosystems are of new-found importance to conserving biodiversity (2, 4). Agricultural grasslands in North America are comprised primarily of hayfields that support livestock operations, although alternative uses for hay (e.g., bioenergy) are increasing (5). Grassland birds occupy a range of habitat types, such as managed pasture, alfalfa-dominated crops, and mixed-grass hayfields (e.g., ref. 6).

Despite the importance of agro-ecosystems, hayfields pose unique risks to wildlife attempting to use them (7). For example, organochlorine pesticides applied to hayfields may accumulate to potentially hazardous levels in grassland birds (8). Hayfields frequently are small and highly fragmented, characteristics that are associated with reduced nest success (9, 10). Hay harvest also can change vegetation structure, reducing prey abundance (11). Additionally, hay harvest can induce mortality by killing adults or destroying nests and nestlings through direct contact with harvesting machinery (12, 13). Although not all nests are destroyed by hay harvesting, nests remaining in harvested hayfields have a poor survival probability because mowing encourages parental abandonment and exposes the nests to predators (14). These factors, particularly hay harvest practices and land-use change, have resulted in dramatic population declines of grassland birds in recent decades (15, 16).

How hay harvest influences grassland birds has been well studied. However, the mechanisms that drive hay harvest into conflict with grassland birds are less so. Identifying underlying mechanisms is necessary to generate more complete accounts of such conservation issues and to clarify research problems (17). One factor that often drives hay harvest patterns is the nutritive quality of hay (18). For example, sheep, horses, and dairy cows all require high-grade hay with high protein content and low fiber levels (19, 20), a diet that can be achieved only by consuming hay that is harvested early and often. Some beef cattle operations, on the other hand, can use hay of lower nutritive quality, so that hay can be harvested later and less often (21).

The ultimate driver of hay harvest patterns is the economy of the livestock trade, which determines the size of herds maintained by farmers. Because feeding costs represent the bulk of a livestock farmer's expenses (22), it makes financial sense that forage production should be related to herd size (23, 24). Therefore, anything affecting herd size should, in turn, affect hay harvest and grassland birds. Thus it is important to assess the role of international trade and its potential cascade of effects through agricultural-grassland practices that influence avian populations. Some major elements that might affect global livestock trade are disease, supply and demand, and trade restrictions/agreements (25, 26).

Bovine spongiform encephalopathy (BSE) is a fatal neurological disease in cattle that was first detected in 1985 in the United Kingdom (27) and resulted in an epizootic that peaked in 1993. To restrain the epizootic, a number of control measures have been instituted globally, to varying effectiveness. Culling has been a common control measure (28, 29), which depletes herd size and reduces the supply available to meet the demands of beef and milk markets. The magnitude of such culling measures is not trivial; from 1988 to 1998, ~4% (3.9 million) of the cattle in the United Kingdom were culled (30). Trade restrictions have been established as another control measure, such as the worldwide ban on importing livestock products from the United Kingdom instituted in 1989 (31, 32), which has had a notable economic effect on that country's agricultural sector (33, 34).

BSE clearly has exerted a major pressure on global beef markets and farmers; therefore this dynamic presents a model system to investigate cascading effects of globalization on agricultural practices and, subsequently, wildlife conservation. We sought to examine the role of the global BSE epizootic in altering agricultural practices in North America, specifically patterns of hay harvested for livestock forage, and any subsequent effects on North American grassland birds. Recognizing that such cascading effects may not be manifested immediately, we evaluated potential relationships over several time lags.

Results

D-separation (D-sep) tests (Table 1) (35) show that model A (Fig. 1) is clearly rejected by the data; the probability of a causal process hypothesized by this model actually generating the data we compiled is <0.05. However, models B, C, and D (Fig. 1) are plausible enough to have generated the data we compiled. To discriminate further among the three plausible models, a likelihood ratio comparison (Table 2) (36) illustrates that model C is favored consistently over models B and D.

Table 1.

Basis sets for tests of conditional independence illustrated by the models shown in Fig. 1

	r (probability)
Basis set	Model A	Model B	Model C	Model D
(1,4) | {2,3}	0.225 (0.421)	0.110 (0.711)	0.095 (0.748)	0.201 (0.467)
(1,5) | {4}	0.763 (0.0003)	0.457 (0.084)	0.004 (0.989)	0.319 (0.230)
(2,5) | {1,3,4}	0.158 (0.591)	0.138 (0.654)	0.018 (0.953)	0.092 (0.756)
(3,5) | {1,2,4}	0.156 (0.594)	0.275 (0.361)	0.565 (0.038)	0.483 (0.074)
C	20.39	8.52	7.23	10.22
pr (C)	0.009	0.385	0.512	0.250

Variable notation is such that 1 = outbreak of bovine spongiform encephalopathy, 2 = imports of cattle, 3 = exports of cattle, 4 = hay production, and 5 = bird populations as derived from the Breeding Bird Survey. In each basis set, variables enclosed in parentheses are implied to be independent, conditional upon the variables enclosed in brackets (‘causal parents’). How well the data fit the model is tested with Fisher's C statistic, with eight degrees of freedom for each model (a significant C statistic indicates inadequate fit).

Fig. 1.

Socioeconomic cascades between BSE epizootics, international trade, agriculture, and population trends of grassland birds represented as four alternative directed acyclic graphs (hypotheses). In each model, further variables are noted as I = imports of cattle, E = exports of cattle, Hay = hay production, and Birds = bird populations as derived from the Breeding Bird Survey. Subscripts to variables represent the number of years since a BSE outbreak (in year t).

Table 2.

Likelihood ratios of the four models in Fig. 1 expressed as probability of the model above and below the diagonal

	Model A	Model B	Model C	Model D
Model A	—	0.02	0.02	0.04
Model B	43.01	—	0.76	1.54
Model C	57.29	1.33	—	2.05
Model D	27.98	0.65	0.49	—

Except in one instance, the individual claims of conditional independence in model C are all well supported. The only claim that is highly improbable is that bird populations are conditionally dependent on cattle exports from North America in the previous 2 y. Thus, we are assured at least that grassland bird populations are the end point in a cascade wherein (i) BSE outbreaks in Europe result in increased cattle exports from North America the next year (Fig. 2A), (ii) increased cattle exports are followed by a decrease in the hay harvest in North America the next year (Fig. 2B), and (iii) a positive response in grassland bird populations then is detected in the year after hay harvest reduction (Fig. 2C).

Fig. 2.

Relationships between BSE epizootics, international trade, agriculture, and population trends of grassland birds. (A) Number of cases of BSE in five focal countries (in year t) and the next year's (t + 1) tonnage of cattle exported to those countries from North America. (B) Tonnage of cattle exported from North America (in year t + 1) to five focal countries, in which BSE broke out in year t, and the tonnage of hay produced in North America the following year (year t + 2). (C) Tonnage of hay produced in North America 2 y after BSE outbreaks in five focal countries (year t + 2) and the next year's (t + 3) summed population index for 20 North American grassland bird species as determined through the Breeding Bird Survey, 1966–2007. Regression lines were fitted as second-order polynomial to illustrate trends better.

Post hoc linear regressions for each species (Table 3) revealed that 17 of the 20 species assessed showed statistically significant negative relationships between abundance and hay production the previous year. Collectively, this pattern is illustrated (Fig. 2C) by the strong relationship (F_1,39 = 38.55, P < 0.05) between the abundance of all grassland birds and hay production. The three species most strongly correlated to hay production were Grasshopper Sparrow (Ammodramus savannarum), Sedge Wren (Cistothorus platensis), and Eastern Meadowlark (Sturnella magna).

Table 3.

Results of a linear regression between annual hay production in North America and the next year's population index for 20 North American grassland bird species as determined through the Breeding Bird Survey, 1966–2007

Common name	Scientific name	F1,39	P
Sharp-tailed Grouse	Tympanuchus phasianellus	3.73	0.06
Northern Harrier *	Circus cyaneus	23.37	0.00002
Ferruginous Hawk *	Buteo regalis	11.98	0.001
Upland Sandpiper	Bartramia longicauda	3.38	0.07
Long-billed Curlew	Numenius americanus	8.14	0.007
Short-eared Owl *	Asio flammeus	12.92	0.0009
Horned Lark *	Eremophila alpestris	29.65	0.000003
Sedge Wren *	Cistothorus platensis	33.58	0.000001
Sprague's Pipit *	Anthus spragueii	29.13	0.000004
Vesper Sparrow *	Pooecetes gramineus	26.72	0.000007
Lark Bunting *	Calamospiza melanocorys	19.91	0.00007
Savannah Sparrow *	Passerculus sandwichensis	14.91	0.0004
Grasshopper Sparrow *	Ammodramus savannarum	37.41	0.000000
Baird's Sparrow *	Ammodramus bairdii	24.37	0.00002
Le Conte's Sparrow	Ammodramus leconteii	1.77	0.19
McCown's Longspur	Calcarius mccownii	6.89	0.01
Chestnut-collared Longspur *	Calcarius ornatus	26.22	0.000009
Bobolink *	Dolichonyx oryzivorus	31.71	0.000002
Eastern Meadowlark *	Sturnella magna	32.93	0.000001
Western Meadowlark *	Sturnella neglecta	26.22	0.000009
All birds		38.55	0.000000

Because the time periods did not differ, degrees of freedom for the F statistic were the same for each test.

*A statistically significant relationship exists for this species when α is set to 0.003 through Bonferroni correction.

The fact that 85% of the species-level regressions were statistically significant indicates a low likelihood of spurious correlations from multiple comparisons. Nonetheless, even if we apply a Bonferroni correction to reduce spuriousness (which lowers α to 0.003), 15 of the 20 species-level comparisons remain statistically significant (Table 3).

Discussion

Our results illustrate that grassland bird populations in North America are influenced by the following process: (i) BSE outbreaks in Europe in any given year result in an increase in cattle imports the following year, thus reducing the herd sizes that need to be maintained by North American farmers; (ii) this reduction in herd size creates a reduction in North American hay harvest 2 y after the outbreak; and (iii) this reduction in hay harvest ultimately yields a positive response in grassland bird populations in the year after hay harvest reduction. A surprising aspect of this cascade is that each stage seems to lag behind the other by 1 y, so that grassland bird populations in North America exhibit population benefits a full 3 y after a BSE outbreak in Europe. These results provide one of the few examples showing that disease dynamics and policy decisions in one part of the globe can have cascading consequences for biodiversity in another.

For 85% of the species we assessed, we detected a particularly strong negative relationship between hay production and grassland bird populations in the following year. Grasshopper Sparrow, Sedge Wren, and Eastern Meadowlark showed especially prominent declines with increased hay production; this finding is not surprising, because Sedge Wrens reach higher densities in idled haylands than in areas under production (37), and both Grasshopper Sparrow and Eastern Meadowlark are sensitive to the frequency and timing of hay harvest (38, 39). The negative relationship between hay production and bird populations illustrates that these species would be good indicators of environmental conditions, because they exhibit measurable sensitivity to an environmental perturbation (e.g., hay harvest). Knowledge of such population sensitivity and of at least one socioeconomic cascade that drives habitat availability should allow more detailed and effective species-level recovery strategies in the United States and Canada.

Only three species did not exhibit a negative relationship between hay harvest and population trend: Upland Sandpiper (Bartramia longicauda), Sharp-tailed Grouse (Tympanuchus phasianellus), and Le Conte's Sparrow (Ammodramus leconteii). In part, perhaps, this absence of a negative relationship may be explained by both Upland Sandpiper and Le Conte's Sparrow also commonly populating other habitat types. Upland Sandpipers nest primarily in grazed pastures and ungrazed grasslands, and significantly fewer nest in hay and crop fields (40). Le Conte's Sparrows also are known to use emergent wetlands (41) and mesic grasslands (42). It is unclear why Sharp-tailed Grouse did not exhibit a negative relationship between hay harvest and population trend, and this result could be an area for future research.

Despite the evidence presented here, we acknowledge that these patterns are correlative and do not necessarily imply causation. Indeed, our models ignore many potentially important details; for instance, hay production can be affected dramatically by precipitation, temperature, and fertilization (43). Nonetheless, the patterns we detected indicate that socioeconomic factors can have a quantifiable effect on wildlife, sometimes in regions far removed from the source. Such knowledge will allow us to improve management recommendations and, particularly, to predict more accurately future periods of anthropogenic adversity to bird populations. Such predictive capacity will improve our ability to manage wildlife in agro-ecosystems by revealing processes that might frustrate certain management approaches, such as the failure of some agro-environment schemes to protect biodiversity (44).

The socioeconomic cascade we document here can have important additional side effects. For example, increasing global demand for beef (45, 46) can trigger overproduction, which can be exacerbated by economic subsidies (47). Maintaining the resultant food reserves can create inequitable access to food at a global scale, increasing the incidence of hunger (48). Increased stocks of cattle also bear ecological costs, such as soil and air pollution, deforestation, and desertification (49, 50).

Studies such as ours make it possible to evaluate more accurately the effects of socioeconomic changes on wildlife populations and to draft management plans that better mitigate the consequences of human activities. Understanding how socioeconomic cascades may affect wildlife is essential for economically efficient policymaking. As an example, the Canada-European Union Comprehensive Economic and Trade Agreement (CETA), for which negotiations began in 2009 and will end in 2011, will provide Canadian agriculturalists better access to European markets (51). Such bi- and multilateral trade agreements generate substantial controversy (52), particularly in terms of the unanticipated environmental impact of global trade (53). The study we present here is a small but important example of how some policies, such as CETA, might be drafted with the anticipation of certain impacts.

Materials and Methods

To examine how the populations of North American grassland bird species changed in relation to hay harvest at the continental scale, international trade, and global disease prevalence (see below), we used an index of relative abundance (% annual population change) derived from the North American Breeding Bird Survey [BBS (54)] from 1966 to 2007. The BBS is an annual roadside survey conducted across North America on more than 4,000 routes that are each 39.4 km long, along which observers record all birds detected at 800-m intervals. The roadside data collection makes BBS data particularly well suited to monitor bird populations in agricultural areas (55). We focused on population trends in 20 of the 28 species classified as “grassland breeding” (54). We excluded Ring-necked Pheasant (Phasianus colchicus) from analysis because it is a frequently hunted nonnative gamebird in North America, and its populations are heavily regulated by factors other than agriculture. Also, because we sought to keep our analyses at a multinational/continental scale, we omitted seven species because they did not occur in both the United Sates and Canada (all occurred on fewer than 14 BBS routes in Canada). The seven species we did not assess because of such single-country endemism were Greater Prairie Chicken (Tympanuchus cupido), Mountain Plover (Charadrius montanus), Barn Owl (Tyto alba), Burrowing Owl (Athene cunicularia), Cassin's Sparrow (Aimophila cassinii), Henslow's Sparrow (Ammodramus henslowii), and Dickcissel (Spiza americana).

We obtained national annual hay production data for 1966–2007 from Statistics Canada and from the National Agricultural Statistics Service of the US Department of Agriculture. Hay production levels for each year are estimated (in tons) by these organizations based on several data sources, especially mandatory surveys and site visits. Although these data are liable to suffer from some error, such as nonreporting and missampling, we did not expect these errors to be variable across years.

The Food and Agriculture Organization (FAO) of the United Nations was our source for data on international importation and exportation of beef cattle (tonnage) for the period 1966–2007. The FAO monitors international trade in more than 600 commodities, and these data are processed and disseminated according to the standard International Merchandise Trade Statistics Methodology. Data are supplied to the FAO by national authorities.

We obtained global data on the occurrence of all incidences of BSE from 1989 to 2007 from the World Organization of Animal Health (OIE). These data are expressed as the number of indigenous cases in each country per million bovines aged over 24 mo. Many countries did not institute surveillance until the late 1990s, and other countries generally have had relatively low incidence rates that seemed to begin at the turn of the century. Therefore, we focused on the five countries with the highest prevalence of BSE and the longest uninterrupted history of monitoring (the United Kingdom, Ireland, France, Switzerland, and Portugal).

We used D-separation (D-sep) tests (35) to examine the plausibility of four directed acyclic graphs (i.e., models; Fig. 1), which varied only in the lag time between each step; each can be viewed as a testable explanation for the data distribution. We chose the D-sep method over classical path analysis because of strong nonlinear relationships (Results) and the repeated-measures of each study country over all years; D-sep tests are robust to such nonlinearity and temporal hierarchy.

Each model in Fig. 1 expressed BSE as influencing both imports of cattle from North America and exports of cattle to Europe, which in turn affect the amount of forage needed to maintain livestock in North America; the amount of hay harvested then affects the grassland bird population index. If the ordering of the variables in Fig. 1 is correct, this effect will be reflected within the numerous independent elements of the D-sep test. A significant (α = 0.05) difference between observed and expected independencies yields rejection of the model.

We first identified a basis set (Table 1) (56), which is a list of all variable pairs not directly connected in Fig. 1. We then identified all variable combinations that are directly connected [i.e., the “causal parents” (56)], in Fig. 1 (see also Table 1). Each claim of independence between variables then was evaluated using Spearman's partial correlation coefficient (56). We then combined these tests of independence in analyzing each full model in Fig. 1, using Fisher's C statistic, which estimates the probability of each D-separation claim in every model. A large C value is associated with statistical significance and thus provides strength of evidence against a model. Finally, to compare model plausibility, we constructed a likelihood ratio matrix (Table 2) of the four models in Fig. 1. We executed all D-sep tests in MS-DOS using the program DGRAPH.exe (35).

To describe the dependent relationship between hay production and grassland bird populations in the best model, we used linear regression to make post hoc comparisons for each study species. For each of these tests, we set α = 0.05; for intermodel comparison, we applied a Bonferroni correction to set α = 0.003.