Weather and biotic interactions as determinants of seasonal shifts in abundance measured through nest-box occupancy in the Siberian flying squirrel


Study area and nest-box occupancy

The study area is located in the Kauhava region, western Finland (62° 54′–63° 16′ N, 22° 54′–23°47′ E; ca. 1,300 km2 area; altitude 42 m), where the landscape is mainly characterized by a mosaic of commercially managed coniferous forests, agricultural land and peatland bogs23,49. Some mixed and old-growth forests as well as many clear-cuts and sapling areas are also found within the area. The area is sparsely populated, and settlement mainly consists of one-family houses and farmhouses.

The flying squirrel is dependent on natural cavities, which have become scarce in Finnish managed forests, including our study area40. In the study area, flying squirrels used nest boxes built for Pygmy owls (Glaucidium passerinum) that are set up for research purposes (e.g.23,40,50). This nest-box type resembles cavities made by the great spotted woodpecker (Dendrocopos major) with the thickness of the front wall > 50 mm and the diameter of the entrance-hole of 45 mm. Nest boxes were grouped so that there are 2 boxes 80–100 m apart within a forest site, the sites being at least 0.8–1.0 km apart23,40. The 2 boxes per site were within an average flying squirrel female territory (8 ha51), and the data for these boxes were combined, that is, if one of the boxes was occupied the site was classified as occupied. In other words, the site was used as a sampling unit (on average 364 ± 121 nest boxes in 208 ± 61 sites yearly).

The occupancy of the nest boxes by flying squirrels was checked every spring and autumn in 2002–2018. Sites were often visited more than once in both spring and autumn, and we control for the number of visits in our analysis (on average 4 ± 1 visits per year on a site). Occupancy was determined by the presence of flying squirrel nesting material within the nest box (ball-shaped nest made of lichen, moss and other soft material, distinct from the nests built by any other animal in the area). If the nest was not used, the nest material was lacking or was a flat layer in the bottom of the nest box, often covered with bird nest materials. The flying squirrel occupied about 9% of the available nest-boxes23. The density of nest boxes was low (0.3 boxes per 1 km2) suggesting that nest boxes had only a minor role in the spatial distribution of the flying squirrel population within the area. The nest boxes were in various forest types, but the detection probability in different forest types does not differ substantially in our data23.

The occupancy patterns were expected to reflect the seasonal mortality and dispersal patterns described in the introduction of this study (seasonal models: (i) dispersal model, (ii) summer survival model, and (iii) winter survival model). Individuals do have more than one nest during year in nest-boxes, dreys and natural cavities21. We could not observe individuals if they did not use nest boxes. Natural cavities were, however, rare near the nest boxes40 and the nest boxes were made to resemble natural cavities by using the trunk of spruce (Picea abies) or aspen (Populus tremula). Communal nesting behavior or reproductive success do not differ for flying squirrels living in these nest boxes and natural cavities in Finland21. The lack of cavities means that flying squirrels present in the area had a reason to build nests to nest boxes, because cavities or nest boxes are preferred nesting places over dreys21. Thus, there should not be much individuals not using our nest boxes, although it is clear that such a cases do occur (see “Discussion”). The data includes, for example, cases where the residents died during summer, but we did not detect them, because dispersers recolonised the nest box. In practise, the number of such cases remains low in our data. It would simultaneously require that in an occupied nest box (occupancy rate of available nest boxes was on average 9%) the resident adult dies during summer (adult summer mortality is not high) and a disperser arrives to the site, which likelihood for a specific nest box remains low. Finally, we are unaware of species that might prevent flying squirrels from using the nest boxes, except for the Pygmy owl. In spring, 5 to 10% of nest-box sites (3–6% of nest boxes) were occupied by breeding pygmy owls and in autumn 17% of nest-boxes included food-stores of Pygmy owls50. Pygmy owls do not prey on flying squirrels but may affect the availability of nest boxes. However, one nest box per forest site was available for flying squirrels even in the sites used by a Pygmy owl, thanks to the study design of two nest boxes per site.

Winter food

Birch catkins are the main food for the flying squirrel in winter21, likely, because the birch is the most abundant deciduous tree in Finnish forests. However, alder catkins are preferred over birch catkins21, and recent studies indicate that the availability of alder catkins in the winter and spring preceding reproduction is an important determinant of breeding success22,25. Temperature in summer determines catkin production30, that is, catkins mature during summer, are available for flying squirrels starting in autumn and stay dormant over winter. Thus, in the current analyses temperature measured in summer is related to next winters’ catkin availability. Catkins flower in spring but flying squirrels may extend the period of catkin usage by storing them21.

For birch catkin availability, we used estimates from an annual birch catkin survey conducted by the Natural Resources Institute Finland (www.luke.fi). These data are collected to describe nation-wide pollen conditions in Finland. Catkin production of deciduous trees is spatially auto-correlated at scales of up to a few hundred kilometres in Finland30,51, and we used the estimate for central-western Finland, where our study area is located. The birch catkin data for central-western Finland is collected annually at approximately six different locations from 304 trees within the region. We did not have an estimate for alder catkin production, but following earlier studies22,25,26,52, we used aerial pollen estimates for central-western Finland as a proxy for alder catkin production (https://www.norkko.fi/). Pollen data were collected by the aerobiology unit of the University of Turku from 10 locations in Finland using EU standard methods and Burkard samplers. The data consisted of accumulated sums of average daily counts of airborne pollen in 1 m3 of air during spring30. Thus, winter food data used in this study describes yearly changes in catkin availability in the region.

Weather data

We used mean monthly weather information from the weather station maintained by the Finnish Meteorological Institute in Kauhava53. The weather recording station was in the middle of the study area and at the same altitude as the rest of the area. There is minimal spatial variation in mean monthly weather measures within our flat study area. We counted mean temperature and precipitation from monthly means for the following periods: winter (December–February), spring (April–May), summer (June–August) and autumn (October–November). March and September were excluded, as they could not be unequivocally assigned to a specific season and, thus, to life stages of flying squirrels (spring: reproduction; summer: rising juveniles; autumn: dispersal period; winter: surviving from the elements). Including these months to analysis did not change the current results or conclusions.

During the study period of 2002–2018, the temperature had an increasing trend in winter and autumn, and a negative trend in summer (effect of continuous variable year on temperature: in winter positive relationship r2 = 0.09; in spring positive r2 = 0.01; in summer negative r2 = 0.09; in autumn positive r2 = 0.1). For precipitation, the trends were positive or non-existing (effect of year on precipitation: in winter positive r2 = 0.04, in spring positive r2 = 0.02, in summer positive r2 = 0.07, in autumn r2 = 0).

Predation pressure

Flying squirrels are negatively affected by the presence of the Ural owl in our study area23. Other predators play a lesser role without having major impacts on flying squirrels (the goshawk Accipiter gentilis23), or are not very common in the area (the pine marten Martes martes and the eagle owl Bubo bubo48). The Ural owl prefers mature mixed and spruce-dominated forest54, just like the flying squirrel. Data on Ural owls was collected by surveys on natural cavities and nest boxes and by searching for new nest sites annually in 2002–2018. Long-term studies of birds of prey have been carried out in the Kauhava region (e.g.40,48,49), so the locations of Ural owl nests are known. The density of Ural owls was approximately 2 pairs per 10 km2 (48; M. Hänninen & E. Korpimäki, unpublished data).

Using the data for Ural owl nests located during the field surveys, the predator presence at flying squirrel nest-box sites was described by calculating flat-top bivariate Gaussian kernels around each nest (see23,55). Following our earlier analysis23, we calculated the kernels with a flat top distance of 500 m, SD of 4 and cut off distance of 5 km. The flat-top part represents the area where the impact of the avian predator is strongest, beyond which it declines, following the Gaussian distribution. The height of the kernel (0–1) at flying squirrel nest box was used as a proxy for predation pressure (referred to as Ural owl index). The kernels were calculated using ArcGIS 10.1 software by Esri and R 3.2.555. The Gaussian kernels were used because the location of nests was known, but we do not know the exact hunting area of individuals. The kernels were based, however, on expert knowledge on likely hunting distance of the species56. That is, the hunting effort was assumed to be highest close to the bird’s nest and to remain at a high level within a given distance and then decrease symmetrically in all directions when moving further from the nest.

Habitat data

The areas of different land use classes within a buffer of 200 m were calculated for each nest box in ArcGIS and R. The buffer corresponds roughly to the estimated home-range size of female flying squirrels50. Thus, the selected spatial scale captured habitat composition at the level central for reproductive success. Landscape maps were based on SLICE dataset57, two forest classifications from 1997 and 2009 (METLA, https://www.maanmittauslaitos.fi/en/opendata), and Landsat images (https://landsat.usgs.gov/), so that yearly changes in forest cover (e.g. clear-cutting of forest) were taken into account. For a detailed description of map processing, see40. We compared which forest composition best describes the squirrel presence and selected the one best fitted to the data based on an Akaike Information Criterion (AIC). That is, model combinations with different forest types and age classes were tested and the one with lowest AIC-values was selected to final models being best fitted for the analysis. The habitat best explaining flying squirrel occurrence included all mature and old spruce and mixed coniferous–deciduous forests. Pure pine forests, which are not preferred by the species21, were excluded.

The available habitat data ended in the year 2015 because we had no information for changes in the forest cover after 2015. We updated the habitat data until 2018 with the values for 2015, but in the end decided to use only the habitat data until 2015 and omitted it from the final models, because it had no effect (see “Results”). Thus, we gained full power to analyse the effects of weather, winter food and predator pressure on flying squirrel occupancy patterns.

Analyses—dispersal model, summer survival model, and winter survival model

We built three binary models (using GLIMMIX in SAS 9.4. software) with nest-box occupancy in different seasons as a response variable. In each model, the nest-box site was a repeated factor (using generalized estimation equations, GLIMMIX SAS) and the year and average number of nest-box visits per year were continuous explanatory variables. To simplify the models, we used an AIC comparison to select the weather variables that were best fitted to the model (AIC < 2 were included in final models; interactions between precipitation and temperature were included to the analysis). Similarly, either the alder or birch index was selected for final models based on the smaller AIC value. Consequently, the variables selected for the final models were not strongly correlated (variance inflation factor < 4). To be able to compare model estimates of different explanatory variables, we standardized the continuous variables as (x − μ)/σ, where x is a raw data value, μ is the mean, and σ is the standard deviation.

In the dispersal model (model i, see introduction), nest boxes used only in autumn (coded with value 2; potential natal dispersal cases) were compared to boxes used only in spring or both in spring and summer (coded with value 1). The final model was: value 2 vs value 1 = Alder pollen during the previous winter + the Ural owl index + summer rain + summer temperature + year + number of nest-box visits; site as repeated factor. In other words, the preliminary AIC analysis indicated that summer rain and temperature were the best-fitted weather variables for this model.

In the summer survival model (ii), spring only boxes (value 1) were compared to boxes used in both spring and autumn (oversummer occupancy; value 2). This model describes the apparent summer mortality of resident adults. The final model was: 2 vs 1 = birch catkins in the previous winter + the Ural owl index + summer rain + winter temperature (previous year) + year + number of nest-box visits; site as repeated factor. In this model, interaction term between summer rain and temperature (see predictions in aims) was not significant and dropped from the final model during preliminary AIC analysis of weather variables.

In the overwinter survival model (iii), we analysed occupancy status the following spring (i.e. after the winter; 2 = yes or 1 = no). Interaction terms between occupancy status (autumn only, i.e. dispersers vs spring and autumn, i.e. residents) and food or Ural owl index were included in the model. This was done to compare whether the responses differed between apparent dispersers (autumn only) and resident adults (spring and autumn). The overwinter survival model was run without spring-only boxes (the result was the same if we included spring-only boxes), because these were not expected to be occupied in the next winter. The final model was 2 vs 1 = occupancy status from the previous year (class variable: autumn only 1, both spring and autumn 2; from now on called season) + birch catkins during the winter + the Ural owl index + summer temperature (previous year) + winter temperature (current winter) + winter rain (current winter) + autumn rain (previous autumn) + season*Ural owl index + season*birch catkins + year + number of nest-box visits; site as repeated factor.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *