Pine marten predation of common goldeneye nests: Effects of cavity age and habitat override any effect of microtine rodent abundance

Abstract According to the alternative prey hypothesis (APH), the temporal synchrony in population fluctuations of microtine rodents and other small herbivores in boreal areas is caused by generalist predators with numerical and functional response to microtines, leading to an increased predation of prey alternative to microtines in the low phase of the microtine population fluctuations. The tree‐climbing pine marten (Martes martes) is a food generalist that includes bird eggs among its alternative prey, also eggs of the cavity‐nesting common goldeneye (Bucephala clangula). We used long‐term data to test whether pine marten predation of goldeneye eggs in nest boxes varied as predicted by the APH. As a measure of microtine abundance at the time of nesting, we applied two measures. First, for goldeneye nests located <40 km from our microtine trapping site, we applied the trapping index of microtine rodents. Second, to also use data from nests located >40 km from our microtine trapping site, and from nests in years when trapping was not conducted, we used two proxies for the microtine abundance: whether boreal owls (Aegolius funereus) nested in any of our boxes <40 km from each goldeneye nest and the average clutch size of these boreal owls. The probability of predation of a goldeneye nest was independent of the microtine trapping index and independent of the proxies for microtine abundance. However, it increased with cavity age, taken as the number of nesting seasons elapsed since the actual nest box was installed, and declined with distance from habitat with forest canopy. The effect of cavity age confirms that the long‐term spatial memory of pine marten is an important factor in the pattern of its predation on nests in tree cavities.


All nests
Nests < 40 km from trapping site in (1972 -2020) trapping years (1977-78 and 1981-2020 S3 Model selection based on Akaike´s information criterion (AIC) to determine which variables influenced the probability of predation of a goldeneye nest situated < 40 km from the microtine rodent trapping site in a year when microtine rodents were trapped (n = 156), corrected for the random effect of nest box ID (n = 62) and temporal autocorrelation of year (n = 34).MicRod (range 0.10 -8.04) denotes the microtine rodent spring trapping index (see text for details).Cav (range 1 -15) denotes the time since the actual nest box was installed (see text for details).Dist (range -100 -100) denotes the shortest distance from the nest box to the nearest edge between forest and an open habitat (see text for details).The GLMM models are sorted by their degree of support from the data, as indicated by the AICc weight (w).Models S4 Model selection based on Akaike´s information criterion (AIC) to determine which variables influenced the probability of predation of a goldeneye nest situated < 40 km from the microtine rodent trapping site in a year when microtine rodents had also been trapped in the previous year (n = 154), corrected for the random effect of nest box ID (n = 60) and temporal autocorrelation of year (n = 33).MicRodCh (range -3.53 -5.03) denotes the change in the microtine rodent trapping index from the previous spring to the current spring (see text for details).Cav (range 1 -15) denotes the time since the actual nest box was installed (see text for details).Dist (range -100 -100) denotes the shortest distance from the nest box to the nearest edge between forest and an open habitat (see text for details).The GLMM models are sorted by their degree of support from the data, as indicated by the AICc weight (w  S5 Model selection based on Akaike´s information criterion for small sample size (AICc) to determine which variables influenced the probability of predation of a goldeneye nest (n = 156) situated < 40 km from the microtine rodent trapping site in a year when microtine rodents were trapped (1977-78 and 1981-2020), corrected for the random effect of nest box ID (n = 62) and temporal autocorrelation of year (n = 34).OwlBr (1 or 0) denotes whether boreal owl nests were found breeding or not in boxes < 40 km from the actual goldeneye nest in the actual year (see text for details).Cav (range 1 -15) denotes the time since the actual nest box was installed, with 1 denoting the first nesting season the box was available (see text).Dist (range -100 -100) denotes the shortest distance from the nest box to the nearest edge between forest and an open habitat, with zero for boxes situated at the edge between forest and open habitat, a negative value for boxes situated within a forest stand, and a positive value for boxes situated in a tree in a clear-cut or a bog (see text for details).The GLMM models are sorted by their degree of support from the data, as indicated by the AICc weight (w).S6 Model selection based on Akaike´s information criterion for small sample size (AICc) to determine which variables influenced the probability of predation of a goldeneye nest (n = 156) situated < 40 km from the microtine rodent trapping site in a year when microtine rodents were trapped (1977-78 and 1981-2020), corrected for the random effect of nest box ID (n = 62) and temporal autocorrelation of year (n = 34).OwlBrCh (-1 or 0 or 1) denotes the change from the previous year in whether boreal owl nests were found breeding or not in boxes < 40 km from the actual goldeneye nest (see text for details).Cav (range 1 -15) denotes the time since the actual nest box was installed, with 1 denoting the first nesting season the box was available (see text).Dist (range -100 -100) denotes the shortest distance from the nest box to the nearest edge between forest and an open habitat, with zero for boxes situated at the edge between forest and open habitat, a negative value for boxes situated within a forest stand, and a positive value for boxes situated in a tree in a clear-cut or a bog (see text for details).The GLMM models are sorted by their degree of support from the data, as indicated by the AICc weight (w  S7 Model selection based on Akaike´s information criterion for small sample size (AICc) to determine which variables influenced the probability of predation of a goldeneye nest (n = 132) situated < 40 km from the microtine rodent trapping site in a year when microtine rodents were trapped (1977-78 and 1981-2020), corrected for the random effect of nest box ID (n = 60) and temporal autocorrelation of year (n = 31).Clutch (range 0.0 -6.5) denotes the average clutch size of boreal owl nests situated < 40 km from the actual goldeneye nest (see text for details).Cav (range 1 -15) denotes the time since the actual nest box was installed, with 1 denoting the first nesting season the box was available (see text).Dist (range -100 -100) denotes the shortest distance from the nest box to the nearest edge between forest and an open habitat, with zero for boxes situated at the edge between forest and open habitat, a negative value for boxes situated within a forest stand, and a positive value for boxes situated in a tree in a clearcut or a bog (see text for details).The GLMM models are sorted by their degree of support from the data, as indicated by the AICc weight (w  S8 Model selection based on Akaike´s information criterion for small sample size (AICc) to determine which variables influenced the probability of predation of a goldeneye nest (n = 106) situated < 40 km from the microtine rodent trapping site in a year when microtine rodents were trapped (1977-78 and 1981-2020), corrected for the random effect of nest box ID (n = 55) and temporal autocorrelation of year (n = 27).ClutchCh (range -6.5 -6.5) denotes the change in average clutch size of boreal owl nests situated < 40 km from the actual goldeneye nest (see text for details).Cav (range 1 -15) denotes the time since the actual nest box was installed, with 1 denoting the first nesting season the box was available (see text).Dist (range -100 -100) denotes the shortest distance from the nest box to the nearest edge between forest and an open habitat, with zero for boxes situated at the edge between forest and open habitat, a negative value for boxes situated within a forest stand, and a positive value for boxes situated in a tree in a clear-cut or a bog (see text for details).The GLMM models are sorted by their degree of support from the data, as indicated by the AICc weight (w).S9 Model selection based on Akaike´s information criterion for small sample size (AICc) to determine which variables influenced the probability of predation of a goldeneye nest (n = 509), corrected for the random effect of nest box ID (n = 181) and temporal autocorrelation of year (n = 47).OwlBr (1 or 0) denotes whether boreal owl nests were found breeding or not in boxes < 40 km from the actual goldeneye nest in the actual year (see text for details).Cav (range 1 -15) denotes the time since the actual nest box was installed, with 1 denoting the first nesting season the box was available (see text).Dist (range -100 -100) denotes the shortest distance from the nest box to the nearest edge between forest and an open habitat, with zero for boxes situated at the edge between forest and open habitat, a negative value for boxes situated within a forest stand, and a positive value for boxes situated in a tree in a clear-cut or a bog (see text for details).The GLMM models are sorted by their degree of support from the data, as indicated by the AICc weight (w  S10 Model selection based on Akaike´s information criterion for small sample size (AICc) to determine which variables influenced the probability of predation of a goldeneye nest (n = 508), corrected for the random effect of nest box ID (n = 181) and temporal autocorrelation of year (n = 47).OwlBrCh (-1 or 0 or 1) denotes the change from the previous year in whether boreal owl nests were found breeding or not in boxes < 40 km from the actual goldeneye nest (see text for details).Cav (range 1 -15) denotes the time since the actual nest box was installed, with 1 denoting the first nesting season the box was available (see text).Dist (range -100 -100) denotes the shortest distance from the nest box to the nearest edge between forest and an open habitat, with zero for boxes situated at the edge between forest and open habitat, a negative value for boxes situated within a forest stand, and a positive value for boxes situated in a tree in a clearcut or a bog (see text for details).The GLMM models are sorted by their degree of support from the data, as indicated by the AICc weight (w  S11 Model selection based on Akaike´s information criterion for small sample size (AICc) to determine which variables influenced the probability of predation of a goldeneye nest (n = 445), corrected for the random effect of nest box ID (n = 176) and temporal autocorrelation of year (n = 44).Clutch (range 0.0 -7.0) denotes the average clutch size of boreal owl nests situated < 40 km from the actual goldeneye nest (see text for details).Cav (range 1 -15) denotes the time since the actual nest box was installed, with 1 denoting the first nesting season the box was available (see text).Dist (range -100 -100) denotes the shortest distance from the nest box to the nearest edge between forest and an open habitat, with zero for boxes situated at the edge between forest and open habitat, a negative value for boxes situated within a forest stand, and a positive value for boxes situated in a tree in a clear-cut or a bog (see text for details).The GLMM models are sorted by their degree of support from the data, as indicated by the AICc weight (w  S12 Model selection based on Akaike´s information criterion for small sample size (AICc) to determine which variables influenced the probability of predation of a goldeneye nest (n = 383), corrected for the random effect of nest box ID (n = 167) and temporal autocorrelation of year (n = 40).ClutchCh (range -6.5 -7.0) denotes the change in average clutch size of boreal owl nests situated < 40 km from the actual goldeneye nest (see text for details).Cav (range 1 -15) denotes the time since the actual nest box was installed, with 1 denoting the first nesting season the box was available (see text).Dist (range -100 -100) denotes the shortest distance from the nest box to the nearest edge between forest and an open habitat, with zero for boxes situated at the edge between forest and open habitat, a negative value for boxes situated within a forest stand, and a positive value for boxes situated in a tree in a clear-cut or a bog (see text for details).The GLMM models are sorted by their degree of support from the data, as indicated by the AICc weight (w

Effects of boreal owl nest occurrence and clutch size as proxies for microtine rodent abundance: goldeneye nests within 40 km from the microtine rodent trapping site
In order to test the APH also with data from goldeneye nests situated > 40 km from the microtine rodent trapping site, and with data from nests in years when microtine trapping was not conducted (1972-76 and 1979-80), we used boreal owl breeding occurrence and average boreal owl clutch size < 40 km of each goldeneye nest as proxies for microtine rodent abundance.
To see how representative these proxies were, we substituted microtine abundance with these proxies for goldeneye nests situated < 40 km from the trapping site in the years when trapping of microtine rodents was conducted.When the proxy for microtine abundance was boreal owl breeding occurrence, the most parsimonious model included only the interaction between cavity age and distance to forest edge (Table S5).This model had the second highest weight (ER = 0.93, Table S5).One of the models with boreal owl breeding occurrence included had the highest weight (Table S5).In this model, the effect of boreal owl breeding occurrence was not significant, while the effect of the interaction between cavity age and distance to forest edge was significant (Table S13a).The most parsimonious among the models with boreal owl breeding occurrence included had the fourth highest weight (ER = 0.39, Table S5).In this model, the effect of boreal owl breeding occurrence was not significant, while the effect of the interaction between cavity age and distance to forest edge was significant (Table S14a).The nonsignificant effect of boreal owl breeding occurrence was positive, which is opposite to what was predicted.Thus, the effect of boreal owl breeding occurrence was similar to the effect of microtine index (cf.Table 1b).
When the proxy for microtine abundance was year-to-year change in boreal owl breeding occurrence, the most parsimonious model, which also was the model with highest AICc weight, included only the interaction between cavity age and distance to forest edge (Table S6).The highest-ranked model among the models with year-to-year change in boreal owl breeding occurrence included had the second highest weight (ER = 0.36, Table S6).In this model, the effect of change in boreal owl breeding occurrence was not significant, while the effect of the interaction between cavity age and distance to forest edge was significant (Table S13b).The most parsimonious among the models with year-to-year change in boreal owl breeding occurrence included had the fifth highest weight (ER = 0.15, Table S6).In this model, the effect of change in boreal owl breeding occurrence was not significant, while the effect of the interaction between cavity age and distance to forest edge was significant (Table S14b).The effect of yearto-year change in boreal owl breeding occurrence was positive, which is opposite to what was predicted.Thus, the effect of year-to-year change in boreal owl breeding occurrence was similar to the effect of year-to-year change in microtine index (cf.Table 2b).
When the proxy for microtine abundance was boreal owl clutch size, the most parsimonious model, which was the model with the third highest AICc weight, included only the interaction between cavity age and distance to forest edge (Table S7).One of the models with boreal owl clutch size included had the highest weight (Table S7).In this model, the effects of the two interactions involving boreal owl clutch size were marginally non-significant (Table S14c).The most parsimonious among models with boreal owl clutch size included had the fourth highest weight (Table S7).In this model, the effect of the interaction between boreal owl clutch size and distance to forest edge was significant (Table S14c).The positive interaction between boreal owl clutch size and distance to forest edge means that the probability of predation of a goldeneye nest as function of boreal owl clutch size increased more rapidly with increasing distance from forest interior into open habitats.The positive effects of boreal owl clutch size were opposite to what was predicted.Thus, the effect of boreal owl clutch size was similar to the effect of microtine index (cf.Table 1b).
Finally, when the proxy for microtine abundance was year-to-year change in boreal owl clutch size, the most parsimonious model, which was the model with second highest AICc weight, included only intercept (Table S8).The highest-ranked among the models with annual change in boreal owl clutch size included was the fourth highestranked (Table S8).In this model, the effect of year-to-year change in boreal owl clutch size was not significant (Table S13d).The most parsimonious among models with annual change in boreal owl clutch size included was the sixth highest-ranked (Table S8).In this model, the effect of year-to-year change in boreal owl clutch size was not significant (Table S14d).The positive effect of annual change in boreal owl clutch size in both models is opposite to what was predicted.Thus, the effect of annual change in boreal owl clutch size was similar to the effect of annual change in microtine index (cf.S8, n = 106 nests, 55 boxes, 27 years, DAIC = 1.16).DAIC denotes the difference in AIC from the highest-ranked model among those that included the variable representing the proxy for microtine abundance.All models included in this analysis are described in Tables S5-S8.Generalized linear mixed-effect models with log link function, binomial distribution, and Laplace approximation to the likelihood.
Explanatory variables are standardized.For the ordered factor "Change owl breeding", "L" denotes the linear term, evaluating a straight-line relationship, while "Q" denotes the quadratic term, capturing non-linear patterns.

Table S1
Number of goldeneye nests distributed on number of nest boxes used._____________________________________________________________________

Table S2
Number of goldeneye nests distributed on number of years._____________________________________________________________________

Table
). Models with DAIC ≤ 2.0 from the best model are shown in bold.

Table
). Models with DAICc ≤ 2.0 from the best model are shown in bold.

Table
Models with DAICc ≤ 2.0 from the best model are shown in bold.______________________________________________________________________

Table
). Models with DAICc ≤ 2.0 from the best model are shown in bold.

Table
). Models with DAICc ≤ 2.0 from the best model are shown in bold.

TABLE S13
Parameter estimates from the highest-ranked GLMM model among models that included the proxy for microtine rodent abundance (boreal owl breeding index and boreal owl clutch size) for the probability of predation of a goldeneye nest situated < 40 km from the microtine rodent trapping site in a year when microtine rodents were trapped, corrected for the random effect of nest box ID and temporal Change in boreal owl breeding index from previous year (TableS6, n = 156 nests, 62 boxes, 34 years, DAIC = 2.01).cBorealowlclutchsize(TableS7,n = 132 nests, 60 boxes, 31 years, DAIC = -0.74).dChange in boreal owl clutch size from previous year (TableS8, n = 106 nests, 55 boxes, 27 years, DAIC = 2.17).DAIC denotes the difference in AIC from the most parsimonious model.All models included in this analysis are described in TablesS5-S8.Generalized linear mixed-effect models with log link function, binomial distribution, and Laplace approximation to the likelihood.Explanatory variables are standardized.For the ordered factor "Change owl breeding", "L" denotes the linear term, evaluating a straight-line relationship, while "Q" denotes the quadratic term, capturing non-linear patterns.

TABLE S14
Parameter estimates from the most parsimonious GLMM model among models that included the proxy for microtine rodent abundance (boreal owl breeding index and boreal owl clutch size) for the probability of predation of a goldeneye nest situated < 40 km from the microtine rodent trapping site in a year when microtine rodents were trapped, corrected for the random effect of nest box ID and temporal autocorrelation of year.Boreal owl breeding index (TableS5, n = 156 nests, 62 boxes, 34 years, DAIC = 1.89).bChange in boreal owl breeding index from previous year (TableS6, n = 156 nests, 62 boxes, 34 years, DAIC = 1.84).cBorealowl clutch size (TableS7, n = 132 nests, 60 boxes, 31 years, DAIC = 1.99).d Change in boreal owl clutch size from previous year (Table