The Influence of Spatio-Temporal Variation in Food Availability and Nest-Predation Risk on Clutch-Size Decisions

Abstract.

The role of both temporal and spatial variability in nest predation and food availability in influencing birds' decisions about clutch size has not been studied. I examine patterns in nest-predation risk and food availability across a range of severity of wildfire to investigate their relationship to clutch size in the Dark-eyed Junco (Junco hyemalis), which breeds commonly in burned forests. Spatial variation in burn severity led to lower risk of nest predation in patches of intermediate severity in the first two years after the fire, while food availability was inversely related to burn severity during only the first year post-fire. Spatial variation in risk of nest predation explained variation in clutch size only during the first year post-fire when food was limited, consistent with parents investing less in risky nest locations during periods of nutritional stress. Nest-predation risk increased seasonally during both year 1 and year 2 post-fire. Clutch size rose dramatically over the first breeding season post-fire, paralleling a unique seasonal increase in food availability in year 1, consistent with juncos tracking temporal variation in food availability by investing more in eggs. Results are consistent with parents balancing spatio-temporal variation in resource limitation with predation risk in their investment in eggs. Disagreement between existing studies as to the relative importance of food and nest predation further highlights the need for carefully designed experimental approaches that integrate explanations for both temporal and spatial trends in sources of selection likely to shape the evolution of clutch-size decisions.

Resumen.

El papel que la variación temporal y espacial en la depredación de nidos y en la disponibilidad de alimento juega sobre las decisiones relacionadas con la postura de huevos en las aves no ha sido estudiado. En este estudio, examino los patrones en el riesgo de depredación de los nidos y en la disponibilidad de alimento en ambientes expuestos a distintos grados de severidad de fuegos silvestres para investigar su relación con el tamaño de la nidada en Junco hyemalis, una especie que comúnmente cría en bosques quemados. La variación espacial en la severidad de las quemas condujo a un menor riesgo de depredación de nidos en parches expuestos a una severidad intermedia en los dos años posteriores a las quemas. Por otro lado, la disponibilidad de alimento estuvo inversamente relacionada con la severidad de las quemas sólo durante el primer año después de éstas. La variación espacial en el riesgo de depredación de los nidos explicó la variación en el tamaño de la nidada sólo durante el primer año posterior a las quemas, cuando el alimento era limitante. Esto sugiere que los padres invierten menos en sitios de anidación riesgosos durante períodos de estrés nutricional. El riesgo de depredación de los nidos aumentó estacionalmente durante el primer y el segundo año después del fuego. El tamaño de la nidada aumentó de forma dramática a lo largo de la primera época reproductiva después del fuego, en paralelo con un aumento estacional único en la disponibilidad de alimento durante el primer año. Esto es consistente con la hipótesis de que las aves siguen la variación temporal en la disponibilidad de alimento, invirtiendo mas en huevos. Los resultados también concuerdan con la hipótesis de que los padres hacen un balance entre la variación espacial y temporal en la limitación de recursos con el riesgo de depredación al determinar su inversión en los huevos. Las discordancias entre los estudios existentes en cuanto a la importancia relativa del alimento y de la depredación de los nidos resalta además la necesidad de contar con enfoques experimentales diseñados cuidadosamente para integrar las explicaciones para las tendencias temporales y espaciales en las fuentes de selección que pueden moldear las decisiones de postura de huevos.

Introduction

The relative importance of “bottom-up” (resource limitation) and “top-down” (predation) influences in shaping reproductive strategies in birds has been debated for decades (e.g., Martin 1987, Boutin 1990, Newton 1998). Much attention in explaining observed variation in avian clutch size, for example, has been focused on food limitation (Lack 1947, reviewed in Martin 1987). Where food is limited, parents are expected to reduce the number of young they attempt to raise because of a greater risk of mortality from starvation or because of impaired growth that can affect survival (Lack 1947, Martin 1987, Sæther 1994, Martin 1995). Yet nest predation is the primary source of nest failure (Ricklefs 1969, Martin 1993) and may also be a key factor shaping the evolution of clutch size. Greater risk of nest predation is predicted to favor smaller clutch size (Ricklefs 1969, Martin et al. 2006) by saving adults more energy for renesting attempts following failure (Foster 1974, Slatkin 1974). While interspecific comparative analyses have provided important insights into factors associated with clutch-size variation (Martin et al. 2000, 2006), phenotypic responses to variation in food availability and nest predation within a species can shed light on the relative importance of these factors in the expression of clutch size (West-Eberhard 1989, Ghalambor and Martin 2001).

Population-scale experiments examining the effects of food limitation and predation on clutch size in birds have focused predominantly on a single limiting factor at a time, and most experiments have examined only food (Newton 1998). To date, experiments have demonstrated that birds are capable of adjusting clutch size in response to (1) high environmental risk of nest predation during a previous breeding season (Doligez and Clobert 2003, Eggers et al. 2006), (2) increases in perceived risk of nest predation (Eggers et al. 2006), and (3) spatial and temporal variation in food availability (reviewed in Martin 1987). Only two studies have examined how parents simultaneously weigh food availability and nest-predation risk in decisions about clutch size. Preston and Rotenberry (2006) showed that reduction in predation risk had no effect on brood size in the Wrentit (Chamaea fasciata), regardless of food availability, but that food supplementation increased brood size during a year of typical food availability, not during a drought year. In contrast, Zanette et al. (2006) found that female Song Sparrows (Melospiza melodia) increased their clutch size moderately where predation was low and females were given supplemental food. They found that birds treated to both conditions increased the number of eggs they laid more dramatically. To my knowledge, no study has also considered a potential role for temporal variation in food and predation risk to influence clutch-size decisions.

While experimental approaches to the study of reproductive strategies in birds have included reductions in predator abundance (Fontaine and Martin 2006), providing food subsidies to nest predators (Preston and Rotenberry 2006), and simulating a real (Doligez and Clobert 2003) or perceived (Eggers et al. 2006) risk of nest predation, alterations to natural food availability have consistently been additions. However, if natural food availability is already high, reproductive effort is relatively fixed (e.g., Sinervo and Licht 1991, Boggs and Ross 1993), or if food is supplemented after parents have already decided their investment, then additional food may not increase that investment (Martin 1987, 1995). Moreover, it is unclear how such manipulations reflect natural variation in food or predation risk or, more importantly, if these manipulations fail to concurrently alter the cues animals use to assess current and/or future food availability or nest-predation risk. In fact, many authors have suggested this possibility as an explanation when animals fail to respond to experimental manipulations (e.g., Julliard et al. 1997, Fontaine and Martin 2006, Preston and Rotenberry 2006). If clutch size is not simply a constraint imposed by either food or predation but rather a result of a strategic balance between resource limitation and predation risk, then an important approach to understanding how food and predation shape the expression of clutch size is to examine responses to dramatic natural temporal and spatial variation in both food availability and nest predation to which a species is known to have evolved.

Here, I report on the importance of food limitation versus nest predation on the expression of clutch size in the Darkeyed Junco (Junco hyemalis), using wildfire as an agent influencing spatial and temporal variation in both factors. Wildfire is the primary recurring disturbance that has shaped floral and faunal communities throughout western North America (Habeck and Mutch 1973, Agee 1993). It has also shaped the local distributions of species by creating a mosaic of patch types when fire severity and extent vary spatially (Platt and Connell 2003, Turner et al. 2003). The Dark-eyed Junco is a primarily ground-feeding and -nesting songbird that occurs in abundance across habitats burned with varying severity (Smucker et al. 2005), so it is likely to have evolved reproductive strategies for dealing with the dramatic temporal and spatial heterogeneity found in its habitat after a fire. Annual productivity of the junco has been linked with the size of populations of rodents such as chipmunks (Tamias spp.), squirrels (Sciurus spp.), and deer mice (Peromyscus maniculatus), which consume junco eggs and young (Clotfelter et al. 2007).

In the northern Rocky Mountains, deer mouse densities increase dramatically after severe fire, as burned forest represents higher-quality habitat than unburned forest (Zwolak and Foresman 2008). Conversely, unburned forests appear to have rodent diversity higher (Zwolak and Foresman 2007) and abundance of squirrels greater than in forest patches burned at high severity (Stuart-Smith and Hayes 2003). If fire severity structures the distribution of rodents, then nest-predation risk in burned habitats may vary. Variation in fire severity is also predicted to affect the abundance of arthropods by removing the living and decaying biomass that provides food for secondary consumers and detritovores. The vegetation and animal prey available for ground-feeding birds is generally predicted to decrease with increasing fire severity during the first breeding season after a fire before significant vegetation has regrown and to increase as revegetation proceeds. Consequently, if food availability or nest-predation risk vary with fire severity as predicted, inferences can be made regarding the relative importance of food limitation and nest-predation risk in decisions about clutch size.

Although greater food limitation and higher nest predation risk favor females laying smaller clutches (Lack 1947) the expression of clutch size should vary with fire severity if food limitation or nest predation is a more important influence in clutch-size decisions. Consequently, I examine in the junco's food availability and nest-predation risk by season, year, and fire severity and then examine how the expression of clutch size varies with these limiting factors. I assume the effects to be additive and expect lowest clutch sizes to be associated with high nest-predation risk and low food availability and the largest clutches to be laid by females breeding where nest-predation risk is low and food availability is high. I predicted females to lay clutches of intermediate size under conditions of low predation—low food and high predation—high food.

Methods

Study Area

This study was conducted from 24 April to 10 August 2004 and 2005 within the boundaries of the 2850-ha Black Mountain fire of August 2003 near Missoula, Montana. The study area is a moderate-elevation site dominated by mixed-conifer forests of ponderosa pine (Pinus ponderosa), lodgepole pine (Pinus contorta), and western larch (Larix occidentalis). I established a 300-ha subplot to use as a focal study area because of its narrow elevation range (∼1280–1341 m) and having patches that burned at different severities.

I defined the severity at which forest patches burned with a modified version of the Composite Burn Index (Key and Benson 2001). Patches burned at low severity had litter and duff moderately consumed and light charring on trees restricted to their lowest 2 m. Intermediate-severity patches were those with deep charring of litter and duff, densities of early-seral plants (e.g., fireweed, Epilobium angustifolium) greater than than in low-severity and unburned patches, some pre-fire herbs and shrubs persisting, and a few green tree crowns remaining. High-severity forest patches were characterized by largely consumed litter and duff layers, carbonized and mineralized soil, and significant portions of the overstory consumed, including most fine branches (terminal stems) in crowns.

Nest Predation and Clutch Size

Each junco territory was monitored for the entire breeding season and searched for all nesting attempts; nests were located by behavioral cues. Nest searching and territory mapping were initiated several weeks prior to the earliest known nesting attempt in the region in order to ensure first nesting attempts were located. To avoid observer bias, technicians alternated searching for nests individually and in small teams and held effort constant across burn severities. Nests were visited every day to determine the exact days of clutch completion, hatching, and fledging and at least every 3 or 4 days when these events were not imminent. Age of nests at the time they were discovered was calculated by forward- or backward-dating of nests in relation to known dates of nest building, laying, or hatching. During each visit, nests were noted as active, successful, or failed. Active nests were those with an attending parent. Empty nests were considered successful if at least one young was observed as a fledgling near the nest site. Nests were considered failed if all eggs failed to hatch, all nestlings were found dead in the nest, or if the nest was found empty prior to day 8 of the nestling period and no fledglings were located. Typically, nestlings fledge approximately 12 days after hatching (Nolan et al. 2002) but can fledge early if startled. Depredation was considered the cause of nest failure if eggs or nestlings were found in the process of being eaten or removed from the nest by predators, if the nest was disturbed or torn apart, or if broken eggs or dead nestlings remaining showed damage or injuries consistent with a predator attack. Nest failures associated with desertion by parents, starvation, brood parasitism, exposure associated with extreme weather, or failures whose cause was unclear were excluded from analysis. Assigning predation as the cause of failure based on evidence left at the nest (e.g., broken eggshells, destroyed nest) generally introduces more bias in predation estimates than that obtained from video monitoring (Etterson and Stanley 2008) but it is also more practical and less expensive for large-scale demographic studies.

Characteristics of nest sites were measured within 2 weeks of the completion of a nesting attempt. At this time, the severity at which forest patches surrounding nests had burned was estimated within a circle of 50-m radius centered at the nest site. Because nest concealment may affect nest microclimate or influence the perceived risk of nest predation, nest concealment was quantified by an estimate of the percentage of each nest that was visible from a distance of 1 m from each of the four cardinal directions and from directly above. All five measurements were averaged, yielding a single index of concealment for each nest (King et al. 1998). Clutch size was recorded at all nests found prior to hatching because no partial nest losses were observed over the course of the study. Juncos continued to renest after both failed and successful attempts either until they raised two broods (pairs renested up to five times) or until mid-July, when nest initiations ceased.

Food Availability

Because the Dark-eyed Junco forages on the ground (Nolan et al. 2002), I focused sampling on the forest floor. Adults eat both seeds and arthropods, but nestlings depend almost entirely on arthropod prey provided by parents (Nolan et al. 2002). In addition, because the availability of calcium, carotenoids, and other nutrients found in arthropod bodies can limit egg production (Graveland et al. 1994, Bortolotti, et al. 2003), arthropod availability is likely an important factor in clutch-size decisions. In order to estimate how food availability varied with burn severity and within a breeding season, I sampled terrestrial arthropods during three periods evenly spaced across the breeding season: (1) during early May when females were initiating their first clutch, (2) during late June, the peak of the breeding season after typically the wettest month of the summer, and (3) during late July when clutches were no longer initiated. I randomly selected points within the study area at which to collect arthropods in pitfall traps, excluding any points that fell on rock outcrops, where it was impossible to bury the traps, or within 50 m of a road or trail, where the traps could be seen or disturbed by people. During each period I continued sampling random points until I located 23 points each in areas categorized as burned at low, moderate, and high severity. At each point I installed a crosswise array of five traps at 1-m intervals. Each trap consisted of a 10-ounce white plastic cup buried in the soil so that the 8-cm opening was level with the ground surface. I filled each trap halfway with a killing solution of unscented castile soap and water and ran the traps for five days each on the same Julian dates: (1) initiation of first clutches (days 121–125), (2) mid-season (days 163–167), and (3) end of breeding season (days 213–217). At the end of each period of sampling, the contents of all five traps at each point were pooled, fine-sieved, and transferred to Whirl-Pak bags containing 95% ethanol. I collected samples from arrays in the same order and time in which they were placed. Arthropods >3 mm in length and from families known to be prey of juncos (Nolan et al. 2002) were oven dried (60 °C for 4 hr) and weighed.

Breeding Density

Because population density can affect habitat quality (Fretwell 1972) and has the potential to influence clutch size through resource competition (Both 1998), it is important to eliminate density dependence as an explanation for clutch-size variation across severities. I examined territory size and territory density as indices to assess whether habitats of different burn severities differed in junco density. I mapped territories by spot-mapping counter-vocalizing territorial males (Ralph et al. 1993). The locations of territorial males engaged in individual singing and counter-singing, including aggressive interactions toward other males, were marked on a high-resolution aerial photo of the study area (1:700). To improve accuracy of mapped locations, individual nests and conspicuous landmarks throughout the study area were located with hand-held GPS units and plotted on georectified hand-held maps. Observations taken from the day of the first nest of the year to that of the last active nest of each year were used to estimate territory area, defined as the minimum convex polygon (MCP; Odum and Kuenzler 1955) enclosing the locations at which territorial disputes, singing, and counter-singing took place.

Statistical Analyses

I developed a set of a priori candidate models that reflected my assessment of likely causes of variation in the probability predation of junco nests in burned forest. In seasonal environments nest success may vary by season or even within a season because of variation in food availability or nest-predation risk (Nilsson 1989, Hochachka 1990). Moreover, seasonal changes in nest-predation risk may be related to nest concealment or conditions within habitats differing in fire severity. I devised candidate models incorporating the following variables: (1) Julian date of nest initiation, (2) nest concealment, (3) severity of fire effects surrounding the nest, and (4) a year effect. On the basis of combinations of these variables, I evaluated a candidate set of 16 a priori models that I believed could reasonably explain variation in nest predation. Because of the sample size of nests available for analysis and the relatively large number of models evaluated, I did not consider models with interaction terms. Using the output from PROC GENMOD, I evaluated the degree of support for each model with goodness-of-fit tests (Hosmer and Lemeshow 1989) and a second-order information criterion (AIC_c; Hurvich and Tsai 1993) that includes a bias adjustment for small samples. I ran a goodness-of-fit test of the global model (the model containing all variables) to determine whether this model provided an adequate fit to the data.

I selected the best model by judging the degree of support as indicated by ΔAIC_c and normalized Akaike weights. Models with Δ AIC_c ≤ 2 were considered to have substantial support, whereas models with Δ AIC_c ≥ 4 were considered to have little to no empirical support (Burnham and Anderson 2001). Estimates of daily nest-predation probability by burn severity and year post-fire were compared with chi-square goodness-of-fit tests. I performed partial regressions of predicted nest-predation probability and its confidence limits as a function of year post-fire and Julian nest-initiation date for presentation by using the ESTIMATE function and back-transformed values from the logit scale for presentation (proportion =e^estimate/[1 + e^estimate]). I used α = 0.05 as the level of statistical significance.

Clutch sizes of resident multi-brooded passerine species commonly show a mid-season peak, with smaller clutches laid both early and late in the season (Klomp 1970, Crick et al. 1993). Rarely are clutch sizes distributed normally, but the residuals from ANCOVAs commonly are, and parametric tests are robust to violations of normality in response variables (Winer et al. 1991). Ordinal logistic regression (OLR) is recommended for such cases in which the response variable is an ordered discrete number with an upper bound (Thompson et al. 1998). I tested for differences in clutch size among burn severities, within each breeding season, and between the first two years after the fire by using both PROC LOGISTIC (OLR) and PROC GLM and included a quadratic function of nest-initiation date to test for a nonlinear effect of clutch size and nest-initiation date. Partial regressions of predicted values and errors of (1) clutch size and (2) nest-predation date as a function of burn severity, Julian nest-initiation date, and other significant variables were generated for presentation with ESTIMATE on the basis of PROC GLM models. I examined variation in arthropod abundance by using ANOVA with burn severity and year as fixed factors and period of sampling as a covariate. Clutch-size and arthropod data met assumptions of normality (all Kolmogorov—Smirnov tests: P > 0.09)

Coordinates associated with locations of territorial males were entered into a geographic information system so the area of each territory could be calculated, by the MCP method (Odum and Kuenzler 1955). To determine the minimum number of observations necessary to delineate a territory accurately, I first analyzed a subset of 15 territories for which 113 mapped song perches were available. I graphed polygon area versus number of song perches for a randomly subsampled cluster of song perches and found an asymptote at 30 perches. Consequently, I based territory-size estimates on MCPs drawn around 30 locations and compared territory size by fire severity and year with ANOVA. Territory-density estimates were computed as the total number of territories in each category of fire severity divided by the total area of that category within the study area as computed from classification of burned-area reflectance (Clark and Bobbe 2006), a preliminary classification of satellite imagery that shows a fire's effects on vegetation reflectance. Predicted values of nest-predation risk are presented with 95% CIs, while other values are reported as means (± SE).

Results

Nest Predation

Average precipitation during the study was similar to the long-term averages (Fig. 1). The date of the first nest initiation detected was similar in both years of the study (2004: Julian day 125; 2005: Julian day 123). Only first nesting attempts (initiated before 1 June of each year) considered, date of nest initiation in patches of different burn severity did not differ (burn severity: F_2,58 = 1.8, P = 0.17; year: F_1,58 = 2.1, P = 0.15). A total of 168 nests (low fire severity = 49, intermediate = 59, high = 60) was found, representing 1745 exposure days. An additional 15 nests whose failures were associated with desertion (3), brood parasitism (10), or an unknown cause (2) were excluded from analyses (low severity: n = 4; intermediate severity: n = 6, high severity: n = 5). Most nests (68%) were located during nest building, egg laying, or incubation. I observed six instances of depredation: two by American red squirrels (Tamiasciurus hudsonicus), two by yellow pine chipmunks (Tamias amoenus), and two by gopher snakes (Pituophis catenifer) in which the predators were removing eggs or nestlings from nests. No known avian nest predators (Corvidae) were observed in the study area.

Only one of the models in the candidate set received substantial support (Table 1) and contained variables for fire severity and Julian nest-initiation date. The fact that the top model contained the parameters of the next two best models, along with its high Akaike weight (w_i = 0.57), strongly suggests this model is the best. In this model, and as predicted, daily nest-predation probability was significantly lower in patches burned at intermediate severity than in the other two categories and was not significantly different between patches burned at low and high severity (Fig. 2A, D). Daily nest-predation probability increased as the breeding season progressed in both years, but the two years of the study did not differ in this variable (Julian date: χ²₁ = 4.1, P = 0.04; year: χ²₁ = 2.9, P = 0.09; Fig. 2A, D). There was no significant difference between 2004 and 2005 in the slope of the seasonal increase in nest predation (χ²₁ = 1.0, P = 0.31).

Food Availability

As predicted, variation in fire severity led to variation in availability of the junco's food during the first breeding season post-fire. During that first year, biomass of terrestrial arthropods was inversely related to burn severity in May (Julian day 125) and June (Julian day 167), but in the second year biomass was similar across severities throughout the junco's breeding season (severity: F_2,373 = 3.7, P = 0.03; date: F_2,373 = 22.0, P < 0.001; year: F_1,373 = 40.9, P < 0.001, date × year: F_2,373 = 6.4, P < 0.01; severity × year: F_2,373 = 1.4, P = 0.26, Fig. 2B, E). In the first year post-fire, arthropod biomass was low in early May when first nests were initiated but rose dramatically to peak in June. Differences among burn severities in terrestrial-arthropod biomass were no longer detectable in late July (Julian day 213) when clutches were no longer being initiated. Arthropod biomass was generally higher during the second year post-fire, especially during April, but arthropod biomass was not clearly related to burn severity (Fig. 2).

Clutch Size

Junco clutches varied from 1 to 5 eggs with a mode of 4. There was a significant interaction between year and nest-initiation date in explaining clutch size (F_1,94 = 23.6, P < 0.001), so each year was analyzed separately. Clutch size increased with nestinitiation date during the first breeding season post-fire and was greater at intermediate fire severity than at low or high severity (date: F_1,33 = 14.5, P < 0.001; severity: F_2,33 = 5.9, P < 0.01; date × severity: F_1,33 = 0.7, P = 0.20, Fig. 2C). Conversely, clutch size decreased with nest-initiation date through the second year post-fire but did not differ among severities (date: F_1,67 = 5.3, P = 0.02; severity: F_2,67 = 0.6, P = 0.44; date × severity: F_1,67 = 0.7, P = 0.41, Fig. 2F). Quadratic terms modeling clutch size as a modal function of nest-initiation date were not significant in either year and so were excluded from models used to estimate clutch size as a function of burn severity and nest-initiation date (Fig. 2). OLR models met proportional-odds assumptions (all models P > 0.17) and showed qualitatively similar results, containing no quadratic effects of nest-initiation date on clutch size in either year.

Density

Territory size was unrelated to burn severity and year post-fire (severity: F_2,104 = 1.9, P = 0.15; year: F_1,104 = 3.7, P = 0.05; severity × year: F_2,104 = 1.7, P = 0.18). Territory density in all three categories of fire severity was similar in the first (low: 0.68 ± 0.08 ha^-1; intermediate: 0.65 ± 0.05 ha^-1; high: 0.68 ± 0.06 ha^-1) and second years (low: 0.73 ± 0.07 ha^-1; intermediate: 0.76 ± 0.04 ha^-1; high: 0.69 ± 0.05 ha^-1) post-fire.

Discussion

Resource limitation and nest-predation risk represent two fundamentally important sources of selection known to shape reproductive decisions across taxa (Lima 1998), yet how individuals weigh variation in these factors in deciding how many eggs to lay has remained unclear. I found that in the Dark-eyed Juncos these decisions were consistent with parents adjusting clutch size to spatial and temporal variation in food availability and nest-predation risk.

Spatial variation in burn severity led to variation in nest-predation risk during both of the first two years post-fire, with nests in patches burned at intermediate severity experiencing consistently lower predation rates (Fig. 2). Conversely, fire severity shaped food availability only during the first year post-fire, when food availability was inversely related to fire severity. During this year, plots with high food availability and high-nest predation risk (low severity), and plots with low food availability and high nest-predation risk (high severity) did not differ in clutch size. In comparison, females nesting in plots with low nest-predation risk and intermediate food availability (intermediate severity) increased their clutch size. When food availability rose to very high levels across all burn severities (year 2), clutch size was unrelated to spatial variation in nest predation.

Under additive models of clutch size, the increase in parental investment in eggs in high-food, low-predation environments (relative to low-food, high-predation habitats) is equal to the sum of the increases in clutch size associated with females laying in habitats with either an increased food supply or a reduced predation risk. Synergistic models predict a multiplicative relationship. Both the additive and synergistic models predict the largest clutch sizes will be associated with the case with the lowest predation rate and most abundant food (habitat burned at intermediate severity, year 2) and that lowest clutch sizes will be found in the most food-poor, high-risk habitats (burned at high severity, year 1), a result that was not observed. Parental assessment of spatial variation in nest-predation risk influenced the juncos' clutch size during the food-poor first year post-fire—parents laid more eggs in the safest habitat. Yet the lack of this same response during the generally food-rich second year post-fire suggests that food availability may mediate females' responses to perceived predation risk. This hypothesis predicts that parents invest less in risky nests when food is rare and the cost of laying an additional egg is high and that they hedge their bets in space against predation risk when they are under nutritional stress.

Additional variables, including timing of laying, population density, and the female's age or social status, have been postulated to explain spatial variation in clutch size (Kluijver 1951, Perrins 1965, 1991, McCleery and Clobert 1990). With the exception of the female's age and social status I accounted for these other confounding variables in the analyses, and habitat preference by high-quality females leading them to settle preferentially in specific habitats does not explain seasonal patterns of clutch-size variation. Although adults' mortality during the breeding season is often considered to be relatively low (Ricklefs 1973, Sillett and Holmes 2002), increased mortality of adults is predicted to favor increased reproductive effort, which can manifest itself as an increase in clutch size (Martin 1988). Spatial and temporal variation in the risk of predation to incubating females were not assessed in this study but could help explain observed patterns in clutch size if females are able to assess their risk of predation accurately in time and space and adjust clutch size accordingly.

Temperature also has the potential to influence decisions about clutch size. Nest concealment is an important determinant of nest microclimate (Lloyd and Martin 2004), and birds commonly orient nests to allow direct insolation of overly cool nest sites (e.g., Austin 1974) or to avoid overheating of nests by direct sunlight (e.g., Burton 2006). In the system I studied, variation in nest microclimate is unrelated to clutch size (Robertson 2009), suggesting that variation in nest microclimate is less important than variation in food availability and nest-predation risk in clutch-size decisions. Yet ambient temperature may influence clutch size at broad geographic scales; reduced egg viability in warm climates favors smaller clutches (Cooper et al. 2005). Local variation in ambient temperature associated with vegetation structure (e.g., canopy cover) in burned landscapes could be sufficient to influence clutch-size decisions. This hypothesis predicts a lower clutch size in severely burned patches that allow the greatest percentage of sunlight to reach the forest floor where juncos nest. Yet females may easily offset such costs by locating nests in more concealed microsites, making more direct tests of this hypothesis necessary. Finally, variability in the availability of environmental calcium can play a role in the evolution of clutch size (reviewed in Patten 2007), potentially constraining clutch-size decisions when local food sources provide too little.

Responses of clutch size to temporal variation in food availability and nest-predation risk provide additional insight into how these sources of selection can influence decisions about clutch size. Specifically, clutch size should be positively related to increases in food availability but negatively related to increases in nest-predation risk. The likelihood of predation of a junco nest increased through the breeding season in both years of the study and across all burn severities (Fig. 2). In contrast, during the first breeding season post-fire, the availability of the junco's food increased dramatically through the season from initially low levels but throughout the second year post-fire it remained at a relatively high and consistent level. The dramatic seasonal increase in clutch size observed during the first year post-fire paralleled the seasonal increase in food availability and spanned the entire range of known clutch sizes in this species (1–5, Nolan et al. 2002), consistent with juncos tracking food availability by investing more in eggs. Such a seasonal clutch-size increase has not previously been documented in birds and suggests an important role for food limitation as a determinant of parents' decisions about clutch size.

The seasonal decline in clutch size observed during the second year post-fire, when food was abundant, was consistent with predicted responses by parents to the seasonal increase in nest predation-rate but not with predicted responses to temporal variation in food availability. This result suggests that when food availability is above some threshold, fitness is maximized by laying fewer eggs during riskier periods of the breeding season. Seasonal declines in clutch size have been documented in several species of birds (e.g., Elridge and Krapu 1988, Hochachka 1990, Crick et al. 1993) and have been hypothesized to result from (1) individuals timing their breeding to match a seasonal deterioration in food supplies that can reduce the offspring's survival and (2) later breeding by individuals in poorer condition (Brinkoff and Cavé 1997). Evidence from this study is inconsistent with both explanations. Food did not decline seasonally, and breeding pairs did not differ systematically in their nest-initiation date. Seasonal increases in nest-predation rates have recently been documented in other passerines and have been attributed to increases in the abundance and activity levels of important nest predators (Grant et al. 2005, Peak 2007). Until recently (e.g., Natarajan and McCulloch 1999, Dinsmore et al. 2002), however, researchers had only limited tools for estimating temporal variability in the time-dependent probability of nest failure and predation, so seasonal declines in nest success may be far more common than is generally thought. In passerines breeding in the Temperate Zone, the bird's age, breeding condition, and migratory behavior have been shown to be unimportant to seasonal declines in clutch size (Askenmo and Unger 1986, Gil-Delagado et al. 2005, Dhondt et al. 2002), but alternative mechanisms such as parasite loading could also play a role in shaping seasonal variation in clutch size (Møller 1994).

In contrast to the results of Zanette et al. (2006) I found neither additive nor synergistic effects of food availability and nest-predation risk on avian clutch size, but both these factors were related to clutch-size variation. Zanette et al. (2006) also observed a negative correlation between clutch number and clutch size and suggested that there are cumulative energetic, nutritional, or physiological costs to continual renesting following depredation, resulting in a successive decline in clutch size. This mechanism fails to explain both seasonal and spatial patterns of clutch-size variation I observed. In further contrast to my study, results of Preston and Rotenberry (2006) suggest that nest-predation risk is unimportant in decisions about clutch size. Although predator-satiation experiments like the one carried out by Preston and Rotenberry (2006) may reduce nest predation, only removal of predators is likely to concurrently alter the visual and auditory cues birds use to assess the environmental risk of nest predation. Indeed, those authors concede that parents were likely unable to detect reduced nest-predation risk because food-supplemented predators remained present and active throughout the study site. When exposed to large-scale removal of predators, juncos breeding in riparian forest patches elsewhere in their breeding range do not increase their clutch size (Fontaine and Martin 2006), consistent with results this study would predict under typical levels of food availability. My estimates of the probability of predation of a junco nest varied from 0.3% to 6.2% per day, corresponding to the full range of daily nest-predation rates observed in long-term studies of this species (based on a 25-day nesting period, Clotfelter et al. 2007). It is unlikely, then, that variation in nest-predation risk in my study area was too small to elicit a response in clutch size.

Another potential explanation for studies' divergence in results is that most experimental manipulations of food availability have been additions. This paucity of food-reduction experiments is almost certainly due to the practical difficulty of reducing the natural availability of food supplies over an area sufficient to encompass an adequate sample of individuals. In my study, nest-predation risk explained clutch-size variation only when food availability was at its lowest, a result which would be undetectable through food supplementation. Indeed, individuals may weigh the prospects for successful reproduction quite differently when experiencing a hyperabundance of food than when they are struggling to feed themselves. Inferences about the relative importance of food availability and nest-predation risk will depend not only on the interaction of these factors (Martin 1995, Zanette et al. 2006) but on the degree to which experimental manipulations reflect the range of environmental variability populations commonly experience. It should also be considered that my study focused entirely on arthropod availability because of its importance to nestlings' energetics and growth, but seed availability may vary considerably with fire severity and revegetation and may have an important influence on parental investment in young. While co-variation of the putative selection pressure (i.e., nest predation and food availability) with the predicted response in clutch size increases the strength of inference, the correlative nature of results in this study cannot allow definitive conclusions about the importance of food limitation vs. predation. In my study, estimates of food availability and nest-predation risk in this study were averaged across categories of fire severity, while additional variation in food availability and nest predation risk could be significant at smaller spatial scales. Nonetheless, variation in clutch size with fire severity followed patterns predicted from habitat-specific and seasonal variation in risk of nest predation and of food limitation resulting from the effects of post-fire succession and the differential effect of severity in limiting successional processes.

While I focused on parental responses to assessment of the environmental risk of nest predation, the species responsible for most nest predation in this area, as well as variation in their abundance and efficiency with fire severity, remain unclear. Corvids were absent from this system, but gopher snakes, red squirrels, and yellow pine chipmunks were observed in the act of depredating junco nests. Research measuring differences in the relative abundance of these species across burn severities is still lacking. Published data suggest that densities of rodents that are known nest predators in this region respond to wildfire differently (Stuart-Smith and Hayes 2003, Zwolak and Foresman 2007, 2008), with some species most abundant in severely burned patches (deer mouse) and others most abundant in unburned patches (Sciurus spp.). It is possible, then, for variation in the spatial distribution of different nest-predator species to explain the relative safety of patches burned at intermediate severity.

The ability to adjust clutch size in response to changing prospects for adults and young is contingent on reliable cues that can indicate the current or future state of the environment (Dodson 1989). Results of this study are consistent with parent birds weighing spatial and temporal variation in food availability and nest predation in clutch-size decisions. Such plasticity suggests a sophisticated ability to monitor ongoing risks and opportunities and weigh the relative benefits of different parental investment decisions. To date, all three empirical studies simultaneously investigating the importance of food availability and nest predation to such decisions (Preston and Rotenberry 2006, Zanette et al. 2006, this study) differ in their fundamental conclusions. While this divergence may simply illustrate diversity among species in reproductive strategies, it further highlights the need for carefully designed field experiments to tease apart the importance of, and functional interaction between, nest-predation pressure and resource availability in shaping parental-care strategies. To this end, it will be essential to integrate explanations for both temporal and spatial trends in sources of selection likely to shape the evolution of clutch size.

Acknowledgments

This paper received the benefit of review comments from numerous individuals, including R. Hutto, T. Martin, T. J. Fontaine, R. Fletcher, B. Schwartz, D. Barton, and three anonymous reviewers. I am grateful to M. Hethcoat, B. Crees, A. Lawrence, and K. Asakawa for tireless help in recording field data. I thank the Lolo National Forest for logistical support in conducting this study. Funding was provided by the Lolo National Forest, Forest Service, U.S. Department of Agriculture. This study was conducted in conformity with guiding principles in the care and use of animals and incompliance of the laws of the United States of America.

Literature Cited