Summer temperature and precipitation govern bat diversity at northern latitudes in Norway

Study area

This study was conducted during a relatively warm period in six valleys in western Norway around 62° N in July 2013. Owing to its complex topography and position relative to the North Atlantic Current, this area has more geographic regions defined by vegetation than any neighbouring countries in Scandinavia and Finland. Steep mountain slopes have left the bulk of the land area unspoilt by man, and only small patches of farmland and settlements exist. Unlike most of Europe (except for mountainous regions), no large fields are found. Relatively steep valleys were selected, where conditions such as precipitation and oceanic influence would vary relatively little within each valley. Moreover, all areas studied have tall mountains blocking out the sun well before sunset and after sunrise. This factor can influence bat distribution and behaviour at these latitudes (Michaelsen et al. 2011). Available publications from the study area (Gjerde and Fuszara 1995, Olsen 1996, Michaelsen 2003, 2012, Michaelsen et al. 2003, 2004a,b, Michaelsen and van der Kooij 2006, Isaksen et al. 2009) suggest that bats can be found in all available landscapes and habitats, ranging from alpine areas to coastal plains with only bracken vegetation, cliffs, cities, roads (street lamps), cultivated areas, freshwater and marine shores, woodlands (all sorts), coastal islands and more. During summer, most species and the highest densities are found close to freshwater and other water sources with low salinity in the warmest parts of the landscape (Michaelsen 2012). At least some species seem to avoid open habitats (Michaelsen et al. 2011). Most efforts in mapping bats stem from lowland areas, but altitudinal records suggest that at least northern bats occasionally can be found up to about 900 m.a.s.l. and just above the treeline (Gjerde and Fuszara 1995). This would include mean July temperatures as low as approximately 9°C based on the coordinates and altitude reported by Gjerde and Fuszara (1995).

A total of 10 of Norway’s 13 species have been confirmed at about 62° N in the study area (see Gjerde and Fuszara 1995, Michaelsen et al. 2003, 2004a,b, Michaelsen and van der Kooij 2006, Michaelsen 2012). Northern bats, soprano pipistrelles [Pipistrelluspygmaeus (Leach, 1825)], Daubenton’s bat [Myotisdaubentonii (Kuhl, 1817)] and whiskered bats [M. mystacinus (Kuhl, 1817)] are common or fairly common. Brandt’s bat [M. brandtii (Eversmann, 1845)] has been recorded only in northern parts in the study area, but not in any of the valleys investigated in this study. Uncommon or rarely recorded species are noctules [Nyctalus noctula (Schreber, 1774)], parti-coloured bats (Vespertilio murinus Linnaeus, 1758) and brown long-eared bats (Plecotus auritus Linnaeus, 1758). Few records exist of the Nathusius’ pipistrelle [P. nathusii (Keyserling and Blasius, 1839)] and only one recording of serotine (Eptesicus serotinus Schreber, 1774) has been made, all during autumn and only on islands along the coast. The three species found in Norway but not in the study area are the barbastelle [Barbastellabarbastellus (Schreber, 1774)], Natterer’s bat [M. nattereri (Kuhl, 1817)] and the common pipistrelle [P. pipistrellus (Schreber, 1774)], all found at single sites in recent years (Flåten and Røed 2007, Isaksen 2007, van der Kooij et al. 2009).

Automatic recording of bats

Automatic recordings of bat calls were made using the batcorder-system (EcoObs, Nürnberg, Germany) and D500 detectors (Pettersson Elektronik AB, Uppsala, Sweden). Up to seven units from the same manufacturer were deployed each night within the same valley covering the entire temperature gradient selected, ranging from 8 to 14°C mean July temperatures (Table 1). In these valleys, a total of 45 sites had ultrasound loggers deployed, each site with known mean July temperature and mean July precipitation. Where the 45 loggers were deployed, the total variation in mean July precipitation varied from 75 to 150 mm (Table 1). The climate variables were retrieved from shape-files based on 30 years normal (1961–1990) provided by the Norwegian Meteorological Institute. Models behind the shape files are described in detail in Tveito et al. (2000). At present, there are no available shape files showing the variation in precipitation during the 24-h cycle. The precipitation variable used in this study does not discriminate between any such temporal variations, but was still considered as being suitable for the following reason: significant precipitation in this region is, to a great extent, caused by low pressures hitting the coast from southwest, yielding longer durations of rainfall (normally 12 h or longer) (e.g., Hanssen-Bauer et al. 2009). Thus, significant precipitation under current climates is generally not caused by events that would have little effect on bats (e.g., short afternoon showers).

Each unit was placed <100 m from a river and along woodland edge, a preferred habitat for many species, and care was taken to find similar structures for all units within the same valley. The selected areas had continuous woodland throughout the selected valleys and beyond, only fragmented by narrow roads or rivers. Bats should not have any problems moving from any site within a valley to the site where the loggers were deployed. At temperatures about 8–9°C, woodland was patchy only in low alpine habitats. Here, most tree species assumed bush-shaped structures except for rowan (see Moen 1999 for further definitions of alpine habitats). In this habitat, loggers were placed at the edge of such bush-habitats. The time of deployment of loggers was from July 10 to 27, a time that would cover the lactation period before the young are weaned at these latitudes (author’s unpublished data). Loggers were deployed for only one night per site. In one valley, Valldalen, recording was performed over a period of two additional nights with temperatures ranging from 8 to 10°C. Data were collected only on nights considered to be suitable for bats (see Fischer et al. 2009 for relevance to the statistical outcome).

On both hardware systems, recording time was set to start at 21:00 h (minimum of 1 h and 31 min before sunset) and end at 05:00 h (minimum of 19 min after sunrise) local time. With the batcorder II-system (v. 2.0), the default settings were used: quality=20, threshold=-27 dB, post-trigger=400 ms, critical frequency=16 kHz. With the D500 units (v. 2.2.5), the following settings were applied: sampling frequency=500, pre-trigger=off, recording length=3 s, high pass filter=yes, t.sense=medium. The input gain was set to 30, trigger level to 28, and allowed for continuous recording (no pause between each recording).

Post processing

Two software systems were used to process the recorded files from the units. bcAdmin (EcoObs, Nürnberg, Germany) was used only to convert raw-files recorded by the batcorder-system to wave files, and SonoChiro (v.3.2.3. with Qt. 4.8.4., compiled 17 October 2013; Biotope, Mèze, France) was applied to all analysis of ultrasound. The northern boreal package was utilized to narrow down the species included in the analysis (see SonoChiro Team 2013). Information following each recording was as follows: identification to group and/or species level with confidence indexes (0–10), quality of recording (0–10), peak frequency (mean of N pulses), pulse interval (mean of N pulses), presence of feeding buzz, presence of social calls and more (SonoChiro Team 2013). A total of 6208 recordings were identified by the software to contain bat calls from one or more species. From these recordings, SonoChiro produced a csv file with 6701 rows, each with an identification of a bat species. The software indicated the possibility that a second species was present in 488 recordings and even a third species in five recordings. Manual analysis (see Skiba 2009, Middleton et al. 2014) was performed on all these recordings, and further recordings suggesting uncommon species (noctules, parti-coloured bat, Nathusius’ pipistrelle and brown long-eared bat), species never found in the study area (the pond bat, Myotis dasycneme (Boye, 1825), and Natterer’s bat) and recordings suggested to be northern bat with mean frequency outside the range of 28–35 kHz. A total of 699 wave files were analysed manually. Manual analysis reduced the number of recordings with more than one species to 96 and no recordings with three species were found. Two recordings could not be identified to genus and one recording had no bat calls. After manual analysis of bat calls, the total number of rows in the dataset were N=6302 and all were included in further analysis. Each of the 45 sites investigated using ultrasound detectors were then assigned with either presence (1) or absence (0) of bats (excluding the northern bat).

Roosts and distribution of captured bats

Maternity roosts of the most common species in the study area were found through radio telemetry, through inspections of roosts following reports from the public or by random observations during fieldwork. Moreover, several bats were captured and banded at their hunting grounds in the study area in June and July from 2004 to 2012, almost exclusively at temperatures ranging from 13–14°C (mean July). In 2014, some effort was made using a Sussex Autobat lure (David Hill/University of Sussex, Brighton, UK) to increase the knowledge of reproductive status and distribution of the sexes at temperatures below 13°C (mean July) at 62° north.

Statistics

All statistics were performed using R v. 3.0.2 (R Core Team 2013). In the first step, a full generalised linear model (GLM) containing both mean July temperatures and mean July precipitation, including an interaction term and a quadratic term for precipitation, was fitted as predictors to the data. The response was a binary variable indicating either the presence of bats (1) or the absence (0) at each site where detectors were deployed. The northern bat was excluded (see the Discussion section for exclusion arguments). Model simplification (backward selection) was performed by dropping one variable at a time with the ANOVA-command and Chi-tests. The second-order Akaike information criterion (AICc) (Burnham and Anderson 2002) was used as inspection criterion (with the MuMIn package) (Barton 2013) and Area under an ROC curve (AUC) to evaluate the performance of each fitted model (R330 package) (Lee and Robertson 2012). Nagelkerke’s R² (fmsb package) (Nakazawa 2013) was presented as an estimate of the explained deviance of fitted GLM.

In addition to the GLM, a generalized linear mixed-effects model (GLMER with the lme4 library) (Bates et al. 2013) was fitted to the data. Here, temperature was the only fixed effect and precipitation was fitted as a random effect. Both conditional and marginal RGLMM2 were presented to give an estimate of explained deviance of fixed effects (temperature) and the GLMER as a whole. A full description of the approach is found in Nakagawa and Schielzeth (2013). Model selection here was limited to comparing the 0-model (intercept and random factor only) and one full model with both fixed effect (temperature) and random effect (precipitation) using ANOVA and Chi-test.

With regard to the sampling design, signs of spatial autocorrelation were tested on Pearson residuals in the final model objects using Mantel tests provided in the ade4 package (Dray and Dufour 2007).

Recorded species and genera

Bats species other than the northern bat were recorded at sites with mean July temperatures as low as 10°C (Table 2A). This record included the soprano pipistrelle and Myotis bats (possibly several species). As expected, more than one species were recorded in all the seven 13–14°C sites. Around and above the treeline in sub-alpine and low alpine areas at 8 and 9°C, only the northern bat was recorded in 10 out of the 13 sites. Northern bat maternity roosts were found at three locations in areas with mean July temperatures as low as 10 and 11°C (Table 2B). Maternity roosts and lactating females of other species were found in sites with temperatures as low as 12°C (the whiskered bat) and 13°C (the soprano pipistrelle and the Daubenton’s bat) (Table 2B). Neither maternity roosts nor lactating females of the brown long-eared bat were found in the valleys included in this study, but lactating females were captured elsewhere at these latitudes in areas with mean July temperatures above 12°C (author’s unpublished data).

Models and deviance explained

A GLM fitted to the data suggests that mean July temperatures have a significant positive effect and mean July precipitation a significant negative effect on bat diversity (Figure 1). The AICc, AUC and R² values for fitted models with significant terms are listed in Table 3, and parameter estimates, errors and p-values of the final model in Table 4A. The GLM, including both temperature and precipitation without an interaction or quadratic term, was the most parsimonious model and explained an estimated 79% of the variation in the dataset. AUC suggests a very good performance of the final model (Table 3).

Figure 1

Predicted values (lines) of finding other species than the northern bat in temperature (A) and precipitation (B) gradients (x-axis) based on the GLM-model. Each tick represents one of the 45 sites where ultrasound loggers were deployed. The figure predicting the effect of temperature is based on a separate GLM-model fit including temperature as a single factor with average precipitation values. The figure presenting precipitation shows predicted values only in the temperature range where variation in diversity was found during this study (from 10 to 12°C).

In the GLMER, where mean July temperatures were fitted as a fixed effect and precipitation as a random effect, dropping temperature from the model yielded a significant increase in ΔAICc (32.87167) as expected and suggested by the GLM. Marginal RGLMM2 suggests that temperatures alone as a fixed effect explained about 66% of the variation in this dataset, with the model itself (conditional RGLMM2) about 91%. Table 4B shows the parameter estimates of the mixed model.

There was no significant effect of spatial autocorrelation when tested on residuals of fitted models in neither the GLM (p=0.4566) nor the GLMER (p=0.8847).